From 1a4a43250ed2837e633d1a8614a978eb6eedd511 Mon Sep 17 00:00:00 2001 From: Simon Pilgrim Date: Sat, 27 Apr 2019 20:20:02 +0000 Subject: [PATCH] [X86][AVX] Add additional SSE/AVX expandload and compressstore targets llvm-svn: 359390 --- llvm/test/CodeGen/X86/masked_compressstore.ll | 5931 ++++++++++++++++++------- llvm/test/CodeGen/X86/masked_expandload.ll | 5784 +++++++++++++++++------- 2 files changed, 8447 insertions(+), 3268 deletions(-) diff --git a/llvm/test/CodeGen/X86/masked_compressstore.ll b/llvm/test/CodeGen/X86/masked_compressstore.ll index 835dbde..3542bc9 100644 --- a/llvm/test/CodeGen/X86/masked_compressstore.ll +++ b/llvm/test/CodeGen/X86/masked_compressstore.ll @@ -1,245 +1,820 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=haswell| FileCheck %s --check-prefixes=ALL,HSW -; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=skylake-avx512 | FileCheck %s --check-prefixes=ALL,SKX -; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=knl | FileCheck %s --check-prefixes=ALL,KNL +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=sse2 | FileCheck %s --check-prefixes=SSE,SSE2 +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=sse4.2 | FileCheck %s --check-prefixes=SSE,SSE42 +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=avx | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX1 +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=avx2 | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX2 +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=avx512f | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=avx512f,avx512bw,avx512vl | FileCheck %s --check-prefixes=AVX,AVX512,AVX512VLBW ; ; vXf64 ; define void @compressstore_v8f64_v8i1(double* %base, <8 x double> %V, <8 x i1> %mask) { -; HSW-LABEL: compressstore_v8f64_v8i1: -; HSW: # %bb.0: -; HSW-NEXT: vpextrb $0, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB0_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vmovlpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB0_2: # %else -; HSW-NEXT: vpextrb $2, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB0_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vmovhpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB0_4: # %else2 -; HSW-NEXT: vpextrb $4, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB0_6 -; HSW-NEXT: # %bb.5: # %cond.store4 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm3 -; HSW-NEXT: vmovlpd %xmm3, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB0_6: # %else5 -; HSW-NEXT: vpextrb $6, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB0_8 -; HSW-NEXT: # %bb.7: # %cond.store7 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm0 -; HSW-NEXT: vmovhpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB0_8: # %else8 -; HSW-NEXT: vpextrb $8, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB0_10 -; HSW-NEXT: # %bb.9: # %cond.store10 -; HSW-NEXT: vmovlpd %xmm1, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB0_10: # %else11 -; HSW-NEXT: vpextrb $10, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB0_12 -; HSW-NEXT: # %bb.11: # %cond.store13 -; HSW-NEXT: vmovhpd %xmm1, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB0_12: # %else14 -; HSW-NEXT: vpextrb $12, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB0_14 -; HSW-NEXT: # %bb.13: # %cond.store16 -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm0 -; HSW-NEXT: vmovlpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB0_14: # %else17 -; HSW-NEXT: vpextrb $14, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB0_16 -; HSW-NEXT: # %bb.15: # %cond.store19 -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm0 -; HSW-NEXT: vmovhpd %xmm0, (%rdi) -; HSW-NEXT: .LBB0_16: # %else20 -; HSW-NEXT: vzeroupper -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v8f64_v8i1: -; SKX: # %bb.0: -; SKX-NEXT: vpsllw $15, %xmm1, %xmm1 -; SKX-NEXT: vpmovw2m %xmm1, %k1 -; SKX-NEXT: vcompresspd %zmm0, (%rdi) {%k1} -; SKX-NEXT: vzeroupper -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v8f64_v8i1: -; KNL: # %bb.0: -; KNL-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero -; KNL-NEXT: vpsllq $63, %zmm1, %zmm1 -; KNL-NEXT: vptestmq %zmm1, %zmm1, %k1 -; KNL-NEXT: vcompresspd %zmm0, (%rdi) {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v8f64_v8i1: +; SSE2: ## %bb.0: +; SSE2-NEXT: movd %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB0_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movlpd %xmm0, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB0_2: ## %else +; SSE2-NEXT: shrl $16, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB0_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: movhpd %xmm0, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB0_4: ## %else2 +; SSE2-NEXT: pextrw $2, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB0_6 +; SSE2-NEXT: ## %bb.5: ## %cond.store4 +; SSE2-NEXT: movlpd %xmm1, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB0_6: ## %else5 +; SSE2-NEXT: pextrw $3, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB0_8 +; SSE2-NEXT: ## %bb.7: ## %cond.store7 +; SSE2-NEXT: movhpd %xmm1, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB0_8: ## %else8 +; SSE2-NEXT: pextrw $4, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB0_10 +; SSE2-NEXT: ## %bb.9: ## %cond.store10 +; SSE2-NEXT: movlpd %xmm2, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB0_10: ## %else11 +; SSE2-NEXT: pextrw $5, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB0_12 +; SSE2-NEXT: ## %bb.11: ## %cond.store13 +; SSE2-NEXT: movhpd %xmm2, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB0_12: ## %else14 +; SSE2-NEXT: pextrw $6, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB0_14 +; SSE2-NEXT: ## %bb.13: ## %cond.store16 +; SSE2-NEXT: movlpd %xmm3, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB0_14: ## %else17 +; SSE2-NEXT: pextrw $7, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB0_16 +; SSE2-NEXT: ## %bb.15: ## %cond.store19 +; SSE2-NEXT: movhpd %xmm3, (%rdi) +; SSE2-NEXT: LBB0_16: ## %else20 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v8f64_v8i1: +; SSE42: ## %bb.0: +; SSE42-NEXT: pextrb $0, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB0_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: movlpd %xmm0, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB0_2: ## %else +; SSE42-NEXT: pextrb $2, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB0_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: movhpd %xmm0, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB0_4: ## %else2 +; SSE42-NEXT: pextrb $4, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB0_6 +; SSE42-NEXT: ## %bb.5: ## %cond.store4 +; SSE42-NEXT: movlpd %xmm1, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB0_6: ## %else5 +; SSE42-NEXT: pextrb $6, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB0_8 +; SSE42-NEXT: ## %bb.7: ## %cond.store7 +; SSE42-NEXT: movhpd %xmm1, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB0_8: ## %else8 +; SSE42-NEXT: pextrb $8, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB0_10 +; SSE42-NEXT: ## %bb.9: ## %cond.store10 +; SSE42-NEXT: movlpd %xmm2, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB0_10: ## %else11 +; SSE42-NEXT: pextrb $10, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB0_12 +; SSE42-NEXT: ## %bb.11: ## %cond.store13 +; SSE42-NEXT: movhpd %xmm2, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB0_12: ## %else14 +; SSE42-NEXT: pextrb $12, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB0_14 +; SSE42-NEXT: ## %bb.13: ## %cond.store16 +; SSE42-NEXT: movlpd %xmm3, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB0_14: ## %else17 +; SSE42-NEXT: pextrb $14, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB0_16 +; SSE42-NEXT: ## %bb.15: ## %cond.store19 +; SSE42-NEXT: movhpd %xmm3, (%rdi) +; SSE42-NEXT: LBB0_16: ## %else20 +; SSE42-NEXT: retq +; +; AVX1-LABEL: compressstore_v8f64_v8i1: +; AVX1: ## %bb.0: +; AVX1-NEXT: vpextrb $0, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB0_2 +; AVX1-NEXT: ## %bb.1: ## %cond.store +; AVX1-NEXT: vmovlpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB0_2: ## %else +; AVX1-NEXT: vpextrb $2, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB0_4 +; AVX1-NEXT: ## %bb.3: ## %cond.store1 +; AVX1-NEXT: vmovhpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB0_4: ## %else2 +; AVX1-NEXT: vpextrb $4, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB0_6 +; AVX1-NEXT: ## %bb.5: ## %cond.store4 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 +; AVX1-NEXT: vmovlps %xmm3, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB0_6: ## %else5 +; AVX1-NEXT: vpextrb $6, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB0_8 +; AVX1-NEXT: ## %bb.7: ## %cond.store7 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; AVX1-NEXT: vmovhpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB0_8: ## %else8 +; AVX1-NEXT: vpextrb $8, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB0_10 +; AVX1-NEXT: ## %bb.9: ## %cond.store10 +; AVX1-NEXT: vmovlpd %xmm1, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB0_10: ## %else11 +; AVX1-NEXT: vpextrb $10, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB0_12 +; AVX1-NEXT: ## %bb.11: ## %cond.store13 +; AVX1-NEXT: vmovhpd %xmm1, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB0_12: ## %else14 +; AVX1-NEXT: vpextrb $12, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB0_14 +; AVX1-NEXT: ## %bb.13: ## %cond.store16 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX1-NEXT: vmovlpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB0_14: ## %else17 +; AVX1-NEXT: vpextrb $14, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB0_16 +; AVX1-NEXT: ## %bb.15: ## %cond.store19 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX1-NEXT: vmovhpd %xmm0, (%rdi) +; AVX1-NEXT: LBB0_16: ## %else20 +; AVX1-NEXT: vzeroupper +; AVX1-NEXT: retq +; +; AVX2-LABEL: compressstore_v8f64_v8i1: +; AVX2: ## %bb.0: +; AVX2-NEXT: vpextrb $0, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB0_2 +; AVX2-NEXT: ## %bb.1: ## %cond.store +; AVX2-NEXT: vmovlpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB0_2: ## %else +; AVX2-NEXT: vpextrb $2, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB0_4 +; AVX2-NEXT: ## %bb.3: ## %cond.store1 +; AVX2-NEXT: vmovhpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB0_4: ## %else2 +; AVX2-NEXT: vpextrb $4, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB0_6 +; AVX2-NEXT: ## %bb.5: ## %cond.store4 +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm3 +; AVX2-NEXT: vmovlpd %xmm3, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB0_6: ## %else5 +; AVX2-NEXT: vpextrb $6, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB0_8 +; AVX2-NEXT: ## %bb.7: ## %cond.store7 +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm0 +; AVX2-NEXT: vmovhpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB0_8: ## %else8 +; AVX2-NEXT: vpextrb $8, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB0_10 +; AVX2-NEXT: ## %bb.9: ## %cond.store10 +; AVX2-NEXT: vmovlpd %xmm1, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB0_10: ## %else11 +; AVX2-NEXT: vpextrb $10, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB0_12 +; AVX2-NEXT: ## %bb.11: ## %cond.store13 +; AVX2-NEXT: vmovhpd %xmm1, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB0_12: ## %else14 +; AVX2-NEXT: vpextrb $12, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB0_14 +; AVX2-NEXT: ## %bb.13: ## %cond.store16 +; AVX2-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX2-NEXT: vmovlpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB0_14: ## %else17 +; AVX2-NEXT: vpextrb $14, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB0_16 +; AVX2-NEXT: ## %bb.15: ## %cond.store19 +; AVX2-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX2-NEXT: vmovhpd %xmm0, (%rdi) +; AVX2-NEXT: LBB0_16: ## %else20 +; AVX2-NEXT: vzeroupper +; AVX2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v8f64_v8i1: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero +; AVX512F-NEXT: vpsllq $63, %zmm1, %zmm1 +; AVX512F-NEXT: vptestmq %zmm1, %zmm1, %k1 +; AVX512F-NEXT: vcompresspd %zmm0, (%rdi) {%k1} +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v8f64_v8i1: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vpsllw $15, %xmm1, %xmm1 +; AVX512VLBW-NEXT: vpmovw2m %xmm1, %k1 +; AVX512VLBW-NEXT: vcompresspd %zmm0, (%rdi) {%k1} +; AVX512VLBW-NEXT: vzeroupper +; AVX512VLBW-NEXT: retq call void @llvm.masked.compressstore.v8f64(<8 x double> %V, double* %base, <8 x i1> %mask) ret void } define void @compressstore_v16f64_v16i1(double* %base, <16 x double> %V, <16 x i1> %mask) { -; HSW-LABEL: compressstore_v16f64_v16i1: -; HSW: # %bb.0: -; HSW-NEXT: vpextrb $0, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vmovlpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_2: # %else -; HSW-NEXT: vpextrb $1, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vmovhpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_4: # %else2 -; HSW-NEXT: vpextrb $2, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_6 -; HSW-NEXT: # %bb.5: # %cond.store4 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm5 -; HSW-NEXT: vmovlpd %xmm5, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_6: # %else5 -; HSW-NEXT: vpextrb $3, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_8 -; HSW-NEXT: # %bb.7: # %cond.store7 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm0 -; HSW-NEXT: vmovhpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_8: # %else8 -; HSW-NEXT: vpextrb $4, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_10 -; HSW-NEXT: # %bb.9: # %cond.store10 -; HSW-NEXT: vmovlpd %xmm1, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_10: # %else11 -; HSW-NEXT: vpextrb $5, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_12 -; HSW-NEXT: # %bb.11: # %cond.store13 -; HSW-NEXT: vmovhpd %xmm1, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_12: # %else14 -; HSW-NEXT: vpextrb $6, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_14 -; HSW-NEXT: # %bb.13: # %cond.store16 -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm0 -; HSW-NEXT: vmovlpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_14: # %else17 -; HSW-NEXT: vpextrb $7, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_16 -; HSW-NEXT: # %bb.15: # %cond.store19 -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm0 -; HSW-NEXT: vmovhpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_16: # %else20 -; HSW-NEXT: vpextrb $8, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_18 -; HSW-NEXT: # %bb.17: # %cond.store22 -; HSW-NEXT: vmovlpd %xmm2, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_18: # %else23 -; HSW-NEXT: vpextrb $9, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_20 -; HSW-NEXT: # %bb.19: # %cond.store25 -; HSW-NEXT: vmovhpd %xmm2, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_20: # %else26 -; HSW-NEXT: vpextrb $10, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_22 -; HSW-NEXT: # %bb.21: # %cond.store28 -; HSW-NEXT: vextractf128 $1, %ymm2, %xmm0 -; HSW-NEXT: vmovlpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_22: # %else29 -; HSW-NEXT: vpextrb $11, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_24 -; HSW-NEXT: # %bb.23: # %cond.store31 -; HSW-NEXT: vextractf128 $1, %ymm2, %xmm0 -; HSW-NEXT: vmovhpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_24: # %else32 -; HSW-NEXT: vpextrb $12, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_26 -; HSW-NEXT: # %bb.25: # %cond.store34 -; HSW-NEXT: vmovlpd %xmm3, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_26: # %else35 -; HSW-NEXT: vpextrb $13, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_28 -; HSW-NEXT: # %bb.27: # %cond.store37 -; HSW-NEXT: vmovhpd %xmm3, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_28: # %else38 -; HSW-NEXT: vpextrb $14, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_30 -; HSW-NEXT: # %bb.29: # %cond.store40 -; HSW-NEXT: vextractf128 $1, %ymm3, %xmm0 -; HSW-NEXT: vmovlpd %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_30: # %else41 -; HSW-NEXT: vpextrb $15, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_32 -; HSW-NEXT: # %bb.31: # %cond.store43 -; HSW-NEXT: vextractf128 $1, %ymm3, %xmm0 -; HSW-NEXT: vmovhpd %xmm0, (%rdi) -; HSW-NEXT: .LBB1_32: # %else44 -; HSW-NEXT: vzeroupper -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v16f64_v16i1: -; SKX: # %bb.0: -; SKX-NEXT: vpsllw $7, %xmm2, %xmm2 -; SKX-NEXT: vpmovb2m %xmm2, %k1 -; SKX-NEXT: kshiftrw $8, %k1, %k2 -; SKX-NEXT: kmovb %k1, %eax -; SKX-NEXT: popcntl %eax, %eax -; SKX-NEXT: vcompresspd %zmm1, (%rdi,%rax,8) {%k2} -; SKX-NEXT: vcompresspd %zmm0, (%rdi) {%k1} -; SKX-NEXT: vzeroupper -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v16f64_v16i1: -; KNL: # %bb.0: -; KNL-NEXT: vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero -; KNL-NEXT: vpslld $31, %zmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k1 -; KNL-NEXT: kshiftrw $8, %k1, %k2 -; KNL-NEXT: vcompresspd %zmm0, (%rdi) {%k1} -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: movzbl %al, %eax -; KNL-NEXT: popcntl %eax, %eax -; KNL-NEXT: vcompresspd %zmm1, (%rdi,%rax,8) {%k2} -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v16f64_v16i1: +; SSE2: ## %bb.0: +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE2-NEXT: movdqa %xmm8, -{{[0-9]+}}(%rsp) +; SSE2-NEXT: testb $1, -{{[0-9]+}}(%rsp) +; SSE2-NEXT: je LBB1_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movlpd %xmm0, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_2: ## %else +; SSE2-NEXT: movd %xmm8, %eax +; SSE2-NEXT: movl %eax, %ecx +; SSE2-NEXT: shrl $8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB1_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: movhpd %xmm0, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_4: ## %else2 +; SSE2-NEXT: movl %eax, %ecx +; SSE2-NEXT: shrl $16, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB1_6 +; SSE2-NEXT: ## %bb.5: ## %cond.store4 +; SSE2-NEXT: movlpd %xmm1, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_6: ## %else5 +; SSE2-NEXT: shrl $24, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_8 +; SSE2-NEXT: ## %bb.7: ## %cond.store7 +; SSE2-NEXT: movhpd %xmm1, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_8: ## %else8 +; SSE2-NEXT: pextrw $2, %xmm8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_10 +; SSE2-NEXT: ## %bb.9: ## %cond.store10 +; SSE2-NEXT: movlpd %xmm2, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_10: ## %else11 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_12 +; SSE2-NEXT: ## %bb.11: ## %cond.store13 +; SSE2-NEXT: movhpd %xmm2, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_12: ## %else14 +; SSE2-NEXT: pextrw $3, %xmm8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_14 +; SSE2-NEXT: ## %bb.13: ## %cond.store16 +; SSE2-NEXT: movlpd %xmm3, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_14: ## %else17 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_16 +; SSE2-NEXT: ## %bb.15: ## %cond.store19 +; SSE2-NEXT: movhpd %xmm3, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_16: ## %else20 +; SSE2-NEXT: pextrw $4, %xmm8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_18 +; SSE2-NEXT: ## %bb.17: ## %cond.store22 +; SSE2-NEXT: movlpd %xmm4, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_18: ## %else23 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_20 +; SSE2-NEXT: ## %bb.19: ## %cond.store25 +; SSE2-NEXT: movhpd %xmm4, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_20: ## %else26 +; SSE2-NEXT: pextrw $5, %xmm8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_22 +; SSE2-NEXT: ## %bb.21: ## %cond.store28 +; SSE2-NEXT: movlpd %xmm5, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_22: ## %else29 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_24 +; SSE2-NEXT: ## %bb.23: ## %cond.store31 +; SSE2-NEXT: movhpd %xmm5, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_24: ## %else32 +; SSE2-NEXT: pextrw $6, %xmm8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_26 +; SSE2-NEXT: ## %bb.25: ## %cond.store34 +; SSE2-NEXT: movlpd %xmm6, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_26: ## %else35 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_28 +; SSE2-NEXT: ## %bb.27: ## %cond.store37 +; SSE2-NEXT: movhpd %xmm6, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_28: ## %else38 +; SSE2-NEXT: pextrw $7, %xmm8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_30 +; SSE2-NEXT: ## %bb.29: ## %cond.store40 +; SSE2-NEXT: movlpd %xmm7, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_30: ## %else41 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_32 +; SSE2-NEXT: ## %bb.31: ## %cond.store43 +; SSE2-NEXT: movhpd %xmm7, (%rdi) +; SSE2-NEXT: LBB1_32: ## %else44 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v16f64_v16i1: +; SSE42: ## %bb.0: +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE42-NEXT: pextrb $0, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: movlpd %xmm0, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_2: ## %else +; SSE42-NEXT: pextrb $1, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: movhpd %xmm0, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_4: ## %else2 +; SSE42-NEXT: pextrb $2, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_6 +; SSE42-NEXT: ## %bb.5: ## %cond.store4 +; SSE42-NEXT: movlpd %xmm1, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_6: ## %else5 +; SSE42-NEXT: pextrb $3, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_8 +; SSE42-NEXT: ## %bb.7: ## %cond.store7 +; SSE42-NEXT: movhpd %xmm1, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_8: ## %else8 +; SSE42-NEXT: pextrb $4, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_10 +; SSE42-NEXT: ## %bb.9: ## %cond.store10 +; SSE42-NEXT: movlpd %xmm2, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_10: ## %else11 +; SSE42-NEXT: pextrb $5, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_12 +; SSE42-NEXT: ## %bb.11: ## %cond.store13 +; SSE42-NEXT: movhpd %xmm2, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_12: ## %else14 +; SSE42-NEXT: pextrb $6, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_14 +; SSE42-NEXT: ## %bb.13: ## %cond.store16 +; SSE42-NEXT: movlpd %xmm3, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_14: ## %else17 +; SSE42-NEXT: pextrb $7, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_16 +; SSE42-NEXT: ## %bb.15: ## %cond.store19 +; SSE42-NEXT: movhpd %xmm3, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_16: ## %else20 +; SSE42-NEXT: pextrb $8, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_18 +; SSE42-NEXT: ## %bb.17: ## %cond.store22 +; SSE42-NEXT: movlpd %xmm4, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_18: ## %else23 +; SSE42-NEXT: pextrb $9, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_20 +; SSE42-NEXT: ## %bb.19: ## %cond.store25 +; SSE42-NEXT: movhpd %xmm4, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_20: ## %else26 +; SSE42-NEXT: pextrb $10, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_22 +; SSE42-NEXT: ## %bb.21: ## %cond.store28 +; SSE42-NEXT: movlpd %xmm5, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_22: ## %else29 +; SSE42-NEXT: pextrb $11, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_24 +; SSE42-NEXT: ## %bb.23: ## %cond.store31 +; SSE42-NEXT: movhpd %xmm5, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_24: ## %else32 +; SSE42-NEXT: pextrb $12, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_26 +; SSE42-NEXT: ## %bb.25: ## %cond.store34 +; SSE42-NEXT: movlpd %xmm6, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_26: ## %else35 +; SSE42-NEXT: pextrb $13, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_28 +; SSE42-NEXT: ## %bb.27: ## %cond.store37 +; SSE42-NEXT: movhpd %xmm6, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_28: ## %else38 +; SSE42-NEXT: pextrb $14, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_30 +; SSE42-NEXT: ## %bb.29: ## %cond.store40 +; SSE42-NEXT: movlpd %xmm7, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_30: ## %else41 +; SSE42-NEXT: pextrb $15, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_32 +; SSE42-NEXT: ## %bb.31: ## %cond.store43 +; SSE42-NEXT: movhpd %xmm7, (%rdi) +; SSE42-NEXT: LBB1_32: ## %else44 +; SSE42-NEXT: retq +; +; AVX1-LABEL: compressstore_v16f64_v16i1: +; AVX1: ## %bb.0: +; AVX1-NEXT: vpextrb $0, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_2 +; AVX1-NEXT: ## %bb.1: ## %cond.store +; AVX1-NEXT: vmovlpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_2: ## %else +; AVX1-NEXT: vpextrb $1, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_4 +; AVX1-NEXT: ## %bb.3: ## %cond.store1 +; AVX1-NEXT: vmovhpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_4: ## %else2 +; AVX1-NEXT: vpextrb $2, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_6 +; AVX1-NEXT: ## %bb.5: ## %cond.store4 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm5 +; AVX1-NEXT: vmovlps %xmm5, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_6: ## %else5 +; AVX1-NEXT: vpextrb $3, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_8 +; AVX1-NEXT: ## %bb.7: ## %cond.store7 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; AVX1-NEXT: vmovhpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_8: ## %else8 +; AVX1-NEXT: vpextrb $4, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_10 +; AVX1-NEXT: ## %bb.9: ## %cond.store10 +; AVX1-NEXT: vmovlpd %xmm1, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_10: ## %else11 +; AVX1-NEXT: vpextrb $5, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_12 +; AVX1-NEXT: ## %bb.11: ## %cond.store13 +; AVX1-NEXT: vmovhpd %xmm1, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_12: ## %else14 +; AVX1-NEXT: vpextrb $6, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_14 +; AVX1-NEXT: ## %bb.13: ## %cond.store16 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX1-NEXT: vmovlpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_14: ## %else17 +; AVX1-NEXT: vpextrb $7, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_16 +; AVX1-NEXT: ## %bb.15: ## %cond.store19 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX1-NEXT: vmovhpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_16: ## %else20 +; AVX1-NEXT: vpextrb $8, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_18 +; AVX1-NEXT: ## %bb.17: ## %cond.store22 +; AVX1-NEXT: vmovlpd %xmm2, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_18: ## %else23 +; AVX1-NEXT: vpextrb $9, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_20 +; AVX1-NEXT: ## %bb.19: ## %cond.store25 +; AVX1-NEXT: vmovhpd %xmm2, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_20: ## %else26 +; AVX1-NEXT: vpextrb $10, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_22 +; AVX1-NEXT: ## %bb.21: ## %cond.store28 +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm0 +; AVX1-NEXT: vmovlpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_22: ## %else29 +; AVX1-NEXT: vpextrb $11, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_24 +; AVX1-NEXT: ## %bb.23: ## %cond.store31 +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm0 +; AVX1-NEXT: vmovhpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_24: ## %else32 +; AVX1-NEXT: vpextrb $12, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_26 +; AVX1-NEXT: ## %bb.25: ## %cond.store34 +; AVX1-NEXT: vmovlpd %xmm3, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_26: ## %else35 +; AVX1-NEXT: vpextrb $13, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_28 +; AVX1-NEXT: ## %bb.27: ## %cond.store37 +; AVX1-NEXT: vmovhpd %xmm3, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_28: ## %else38 +; AVX1-NEXT: vpextrb $14, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_30 +; AVX1-NEXT: ## %bb.29: ## %cond.store40 +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm0 +; AVX1-NEXT: vmovlpd %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_30: ## %else41 +; AVX1-NEXT: vpextrb $15, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_32 +; AVX1-NEXT: ## %bb.31: ## %cond.store43 +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm0 +; AVX1-NEXT: vmovhpd %xmm0, (%rdi) +; AVX1-NEXT: LBB1_32: ## %else44 +; AVX1-NEXT: vzeroupper +; AVX1-NEXT: retq +; +; AVX2-LABEL: compressstore_v16f64_v16i1: +; AVX2: ## %bb.0: +; AVX2-NEXT: vpextrb $0, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_2 +; AVX2-NEXT: ## %bb.1: ## %cond.store +; AVX2-NEXT: vmovlpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_2: ## %else +; AVX2-NEXT: vpextrb $1, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_4 +; AVX2-NEXT: ## %bb.3: ## %cond.store1 +; AVX2-NEXT: vmovhpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_4: ## %else2 +; AVX2-NEXT: vpextrb $2, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_6 +; AVX2-NEXT: ## %bb.5: ## %cond.store4 +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm5 +; AVX2-NEXT: vmovlpd %xmm5, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_6: ## %else5 +; AVX2-NEXT: vpextrb $3, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_8 +; AVX2-NEXT: ## %bb.7: ## %cond.store7 +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm0 +; AVX2-NEXT: vmovhpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_8: ## %else8 +; AVX2-NEXT: vpextrb $4, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_10 +; AVX2-NEXT: ## %bb.9: ## %cond.store10 +; AVX2-NEXT: vmovlpd %xmm1, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_10: ## %else11 +; AVX2-NEXT: vpextrb $5, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_12 +; AVX2-NEXT: ## %bb.11: ## %cond.store13 +; AVX2-NEXT: vmovhpd %xmm1, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_12: ## %else14 +; AVX2-NEXT: vpextrb $6, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_14 +; AVX2-NEXT: ## %bb.13: ## %cond.store16 +; AVX2-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX2-NEXT: vmovlpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_14: ## %else17 +; AVX2-NEXT: vpextrb $7, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_16 +; AVX2-NEXT: ## %bb.15: ## %cond.store19 +; AVX2-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX2-NEXT: vmovhpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_16: ## %else20 +; AVX2-NEXT: vpextrb $8, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_18 +; AVX2-NEXT: ## %bb.17: ## %cond.store22 +; AVX2-NEXT: vmovlpd %xmm2, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_18: ## %else23 +; AVX2-NEXT: vpextrb $9, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_20 +; AVX2-NEXT: ## %bb.19: ## %cond.store25 +; AVX2-NEXT: vmovhpd %xmm2, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_20: ## %else26 +; AVX2-NEXT: vpextrb $10, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_22 +; AVX2-NEXT: ## %bb.21: ## %cond.store28 +; AVX2-NEXT: vextractf128 $1, %ymm2, %xmm0 +; AVX2-NEXT: vmovlpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_22: ## %else29 +; AVX2-NEXT: vpextrb $11, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_24 +; AVX2-NEXT: ## %bb.23: ## %cond.store31 +; AVX2-NEXT: vextractf128 $1, %ymm2, %xmm0 +; AVX2-NEXT: vmovhpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_24: ## %else32 +; AVX2-NEXT: vpextrb $12, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_26 +; AVX2-NEXT: ## %bb.25: ## %cond.store34 +; AVX2-NEXT: vmovlpd %xmm3, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_26: ## %else35 +; AVX2-NEXT: vpextrb $13, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_28 +; AVX2-NEXT: ## %bb.27: ## %cond.store37 +; AVX2-NEXT: vmovhpd %xmm3, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_28: ## %else38 +; AVX2-NEXT: vpextrb $14, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_30 +; AVX2-NEXT: ## %bb.29: ## %cond.store40 +; AVX2-NEXT: vextractf128 $1, %ymm3, %xmm0 +; AVX2-NEXT: vmovlpd %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_30: ## %else41 +; AVX2-NEXT: vpextrb $15, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_32 +; AVX2-NEXT: ## %bb.31: ## %cond.store43 +; AVX2-NEXT: vextractf128 $1, %ymm3, %xmm0 +; AVX2-NEXT: vmovhpd %xmm0, (%rdi) +; AVX2-NEXT: LBB1_32: ## %else44 +; AVX2-NEXT: vzeroupper +; AVX2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v16f64_v16i1: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: vpmovzxbd {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero +; AVX512F-NEXT: vpslld $31, %zmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k1 +; AVX512F-NEXT: kshiftrw $8, %k1, %k2 +; AVX512F-NEXT: vcompresspd %zmm0, (%rdi) {%k1} +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: movzbl %al, %eax +; AVX512F-NEXT: movl %eax, %ecx +; AVX512F-NEXT: shrl %ecx +; AVX512F-NEXT: andl $-43, %ecx +; AVX512F-NEXT: subl %ecx, %eax +; AVX512F-NEXT: movl %eax, %ecx +; AVX512F-NEXT: andl $858993459, %ecx ## imm = 0x33333333 +; AVX512F-NEXT: shrl $2, %eax +; AVX512F-NEXT: andl $858993459, %eax ## imm = 0x33333333 +; AVX512F-NEXT: addl %ecx, %eax +; AVX512F-NEXT: movl %eax, %ecx +; AVX512F-NEXT: shrl $4, %ecx +; AVX512F-NEXT: addl %eax, %ecx +; AVX512F-NEXT: andl $252645135, %ecx ## imm = 0xF0F0F0F +; AVX512F-NEXT: imull $16843009, %ecx, %eax ## imm = 0x1010101 +; AVX512F-NEXT: shrl $24, %eax +; AVX512F-NEXT: vcompresspd %zmm1, (%rdi,%rax,8) {%k2} +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v16f64_v16i1: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vpsllw $7, %xmm2, %xmm2 +; AVX512VLBW-NEXT: vpmovb2m %xmm2, %k1 +; AVX512VLBW-NEXT: kshiftrw $8, %k1, %k2 +; AVX512VLBW-NEXT: vcompresspd %zmm0, (%rdi) {%k1} +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: movzbl %al, %eax +; AVX512VLBW-NEXT: movl %eax, %ecx +; AVX512VLBW-NEXT: shrl %ecx +; AVX512VLBW-NEXT: andl $-43, %ecx +; AVX512VLBW-NEXT: subl %ecx, %eax +; AVX512VLBW-NEXT: movl %eax, %ecx +; AVX512VLBW-NEXT: andl $858993459, %ecx ## imm = 0x33333333 +; AVX512VLBW-NEXT: shrl $2, %eax +; AVX512VLBW-NEXT: andl $858993459, %eax ## imm = 0x33333333 +; AVX512VLBW-NEXT: addl %ecx, %eax +; AVX512VLBW-NEXT: movl %eax, %ecx +; AVX512VLBW-NEXT: shrl $4, %ecx +; AVX512VLBW-NEXT: addl %eax, %ecx +; AVX512VLBW-NEXT: andl $252645135, %ecx ## imm = 0xF0F0F0F +; AVX512VLBW-NEXT: imull $16843009, %ecx, %eax ## imm = 0x1010101 +; AVX512VLBW-NEXT: shrl $24, %eax +; AVX512VLBW-NEXT: vcompresspd %zmm1, (%rdi,%rax,8) {%k2} +; AVX512VLBW-NEXT: vzeroupper +; AVX512VLBW-NEXT: retq call void @llvm.masked.compressstore.v16f64(<16 x double> %V, double* %base, <16 x i1> %mask) ret void } @@ -249,587 +824,1839 @@ define void @compressstore_v16f64_v16i1(double* %base, <16 x double> %V, <16 x i ; define void @compressstore_v2f32_v2i32(float* %base, <2 x float> %V, <2 x i32> %trigger) { -; HSW-LABEL: compressstore_v2f32_v2i32: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; HSW-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $0, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB2_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vmovss %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB2_2: # %else -; HSW-NEXT: vpextrb $8, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB2_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vextractps $1, %xmm0, (%rdi) -; HSW-NEXT: .LBB2_4: # %else2 -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v2f32_v2i32: -; SKX: # %bb.0: -; SKX-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; SKX-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; SKX-NEXT: vptestnmq %xmm1, %xmm1, %k1 -; SKX-NEXT: vcompressps %xmm0, (%rdi) {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v2f32_v2i32: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; KNL-NEXT: vptestnmq %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftlw $14, %k0, %k0 -; KNL-NEXT: kshiftrw $14, %k0, %k1 -; KNL-NEXT: vcompressps %zmm0, (%rdi) {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v2f32_v2i32: +; SSE2: ## %bb.0: +; SSE2-NEXT: pand {{.*}}(%rip), %xmm1 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqd %xmm1, %xmm2 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,0,3,2] +; SSE2-NEXT: pand %xmm2, %xmm1 +; SSE2-NEXT: movd %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB2_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB2_2: ## %else +; SSE2-NEXT: pextrw $4, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB2_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,2,3] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: LBB2_4: ## %else2 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v2f32_v2i32: +; SSE42: ## %bb.0: +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; SSE42-NEXT: pcmpeqq %xmm2, %xmm1 +; SSE42-NEXT: pextrb $0, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB2_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: movss %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB2_2: ## %else +; SSE42-NEXT: pextrb $8, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB2_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: extractps $1, %xmm0, (%rdi) +; SSE42-NEXT: LBB2_4: ## %else2 +; SSE42-NEXT: retq +; +; AVX1-LABEL: compressstore_v2f32_v2i32: +; AVX1: ## %bb.0: +; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; AVX1-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm1 +; AVX1-NEXT: vpextrb $0, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB2_2 +; AVX1-NEXT: ## %bb.1: ## %cond.store +; AVX1-NEXT: vmovss %xmm0, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB2_2: ## %else +; AVX1-NEXT: vpextrb $8, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB2_4 +; AVX1-NEXT: ## %bb.3: ## %cond.store1 +; AVX1-NEXT: vextractps $1, %xmm0, (%rdi) +; AVX1-NEXT: LBB2_4: ## %else2 +; AVX1-NEXT: retq +; +; AVX2-LABEL: compressstore_v2f32_v2i32: +; AVX2: ## %bb.0: +; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX2-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm1 +; AVX2-NEXT: vpextrb $0, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB2_2 +; AVX2-NEXT: ## %bb.1: ## %cond.store +; AVX2-NEXT: vmovss %xmm0, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB2_2: ## %else +; AVX2-NEXT: vpextrb $8, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB2_4 +; AVX2-NEXT: ## %bb.3: ## %cond.store1 +; AVX2-NEXT: vextractps $1, %xmm0, (%rdi) +; AVX2-NEXT: LBB2_4: ## %else2 +; AVX2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v2f32_v2i32: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX512F-NEXT: vptestnmq %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftlw $14, %k0, %k0 +; AVX512F-NEXT: kshiftrw $14, %k0, %k1 +; AVX512F-NEXT: vcompressps %zmm0, (%rdi) {%k1} +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v2f32_v2i32: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512VLBW-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX512VLBW-NEXT: vptestnmq %xmm1, %xmm1, %k1 +; AVX512VLBW-NEXT: vcompressps %xmm0, (%rdi) {%k1} +; AVX512VLBW-NEXT: retq %mask = icmp eq <2 x i32> %trigger, zeroinitializer call void @llvm.masked.compressstore.v2f32(<2 x float> %V, float* %base, <2 x i1> %mask) ret void } define void @compressstore_v4f32_v4i1(float* %base, <4 x float> %V, <4 x i1> %mask) { -; HSW-LABEL: compressstore_v4f32_v4i1: -; HSW: # %bb.0: -; HSW-NEXT: vpextrb $0, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vmovss %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB3_2: # %else -; HSW-NEXT: vpextrb $4, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vextractps $1, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB3_4: # %else2 -; HSW-NEXT: vpextrb $8, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_6 -; HSW-NEXT: # %bb.5: # %cond.store4 -; HSW-NEXT: vextractps $2, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB3_6: # %else5 -; HSW-NEXT: vpextrb $12, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_8 -; HSW-NEXT: # %bb.7: # %cond.store7 -; HSW-NEXT: vextractps $3, %xmm0, (%rdi) -; HSW-NEXT: .LBB3_8: # %else8 -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v4f32_v4i1: -; SKX: # %bb.0: -; SKX-NEXT: vpslld $31, %xmm1, %xmm1 -; SKX-NEXT: vpmovd2m %xmm1, %k1 -; SKX-NEXT: vcompressps %xmm0, (%rdi) {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v4f32_v4i1: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; KNL-NEXT: vpslld $31, %xmm1, %xmm1 -; KNL-NEXT: vptestmd %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftlw $12, %k0, %k0 -; KNL-NEXT: kshiftrw $12, %k0, %k1 -; KNL-NEXT: vcompressps %zmm0, (%rdi) {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v4f32_v4i1: +; SSE2: ## %bb.0: +; SSE2-NEXT: movd %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB3_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB3_2: ## %else +; SSE2-NEXT: pextrw $2, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB3_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: movaps %xmm0, %xmm2 +; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1],xmm0[2,3] +; SSE2-NEXT: movss %xmm2, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB3_4: ## %else2 +; SSE2-NEXT: pextrw $4, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB3_6 +; SSE2-NEXT: ## %bb.5: ## %cond.store4 +; SSE2-NEXT: movaps %xmm0, %xmm2 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm0[1] +; SSE2-NEXT: movss %xmm2, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB3_6: ## %else5 +; SSE2-NEXT: pextrw $6, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB3_8 +; SSE2-NEXT: ## %bb.7: ## %cond.store7 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1,2,3] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: LBB3_8: ## %else8 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v4f32_v4i1: +; SSE42: ## %bb.0: +; SSE42-NEXT: pextrb $0, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB3_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: movss %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB3_2: ## %else +; SSE42-NEXT: pextrb $4, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB3_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: extractps $1, %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB3_4: ## %else2 +; SSE42-NEXT: pextrb $8, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB3_6 +; SSE42-NEXT: ## %bb.5: ## %cond.store4 +; SSE42-NEXT: extractps $2, %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB3_6: ## %else5 +; SSE42-NEXT: pextrb $12, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB3_8 +; SSE42-NEXT: ## %bb.7: ## %cond.store7 +; SSE42-NEXT: extractps $3, %xmm0, (%rdi) +; SSE42-NEXT: LBB3_8: ## %else8 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: compressstore_v4f32_v4i1: +; AVX1OR2: ## %bb.0: +; AVX1OR2-NEXT: vpextrb $0, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB3_2 +; AVX1OR2-NEXT: ## %bb.1: ## %cond.store +; AVX1OR2-NEXT: vmovss %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB3_2: ## %else +; AVX1OR2-NEXT: vpextrb $4, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB3_4 +; AVX1OR2-NEXT: ## %bb.3: ## %cond.store1 +; AVX1OR2-NEXT: vextractps $1, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB3_4: ## %else2 +; AVX1OR2-NEXT: vpextrb $8, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB3_6 +; AVX1OR2-NEXT: ## %bb.5: ## %cond.store4 +; AVX1OR2-NEXT: vextractps $2, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB3_6: ## %else5 +; AVX1OR2-NEXT: vpextrb $12, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB3_8 +; AVX1OR2-NEXT: ## %bb.7: ## %cond.store7 +; AVX1OR2-NEXT: vextractps $3, %xmm0, (%rdi) +; AVX1OR2-NEXT: LBB3_8: ## %else8 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v4f32_v4i1: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 +; AVX512F-NEXT: vpslld $31, %xmm1, %xmm1 +; AVX512F-NEXT: vptestmd %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftlw $12, %k0, %k0 +; AVX512F-NEXT: kshiftrw $12, %k0, %k1 +; AVX512F-NEXT: vcompressps %zmm0, (%rdi) {%k1} +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v4f32_v4i1: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vpslld $31, %xmm1, %xmm1 +; AVX512VLBW-NEXT: vptestmd %xmm1, %xmm1, %k1 +; AVX512VLBW-NEXT: vcompressps %xmm0, (%rdi) {%k1} +; AVX512VLBW-NEXT: retq call void @llvm.masked.compressstore.v4f32(<4 x float> %V, float* %base, <4 x i1> %mask) ret void } define void @compressstore_v8f32_v8i1(float* %base, <8 x float> %V, <8 x i1> %mask) { -; HSW-LABEL: compressstore_v8f32_v8i1: -; HSW: # %bb.0: -; HSW-NEXT: vpextrb $0, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB4_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vmovss %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB4_2: # %else -; HSW-NEXT: vpextrb $2, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB4_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vextractps $1, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB4_4: # %else2 -; HSW-NEXT: vpextrb $4, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB4_6 -; HSW-NEXT: # %bb.5: # %cond.store4 -; HSW-NEXT: vextractps $2, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB4_6: # %else5 -; HSW-NEXT: vpextrb $6, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB4_8 -; HSW-NEXT: # %bb.7: # %cond.store7 -; HSW-NEXT: vextractps $3, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB4_8: # %else8 -; HSW-NEXT: vpextrb $8, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB4_10 -; HSW-NEXT: # %bb.9: # %cond.store10 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm2 -; HSW-NEXT: vmovss %xmm2, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB4_10: # %else11 -; HSW-NEXT: vpextrb $10, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB4_12 -; HSW-NEXT: # %bb.11: # %cond.store13 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm2 -; HSW-NEXT: vextractps $1, %xmm2, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB4_12: # %else14 -; HSW-NEXT: vpextrb $12, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB4_14 -; HSW-NEXT: # %bb.13: # %cond.store16 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm2 -; HSW-NEXT: vextractps $2, %xmm2, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB4_14: # %else17 -; HSW-NEXT: vpextrb $14, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB4_16 -; HSW-NEXT: # %bb.15: # %cond.store19 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm0 -; HSW-NEXT: vextractps $3, %xmm0, (%rdi) -; HSW-NEXT: .LBB4_16: # %else20 -; HSW-NEXT: vzeroupper -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v8f32_v8i1: -; SKX: # %bb.0: -; SKX-NEXT: vpsllw $15, %xmm1, %xmm1 -; SKX-NEXT: vpmovw2m %xmm1, %k1 -; SKX-NEXT: vcompressps %ymm0, (%rdi) {%k1} -; SKX-NEXT: vzeroupper -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v8f32_v8i1: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; KNL-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero -; KNL-NEXT: vpsllq $63, %zmm1, %zmm1 -; KNL-NEXT: vptestmq %zmm1, %zmm1, %k1 -; KNL-NEXT: vcompressps %zmm0, (%rdi) {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v8f32_v8i1: +; SSE2: ## %bb.0: +; SSE2-NEXT: movd %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB4_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB4_2: ## %else +; SSE2-NEXT: shrl $16, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB4_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: movaps %xmm0, %xmm3 +; SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm0[2,3] +; SSE2-NEXT: movss %xmm3, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB4_4: ## %else2 +; SSE2-NEXT: pextrw $2, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB4_6 +; SSE2-NEXT: ## %bb.5: ## %cond.store4 +; SSE2-NEXT: movaps %xmm0, %xmm3 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm3 = xmm3[1],xmm0[1] +; SSE2-NEXT: movss %xmm3, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB4_6: ## %else5 +; SSE2-NEXT: pextrw $3, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB4_8 +; SSE2-NEXT: ## %bb.7: ## %cond.store7 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1,2,3] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB4_8: ## %else8 +; SSE2-NEXT: pextrw $4, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB4_10 +; SSE2-NEXT: ## %bb.9: ## %cond.store10 +; SSE2-NEXT: movss %xmm1, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB4_10: ## %else11 +; SSE2-NEXT: pextrw $5, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB4_12 +; SSE2-NEXT: ## %bb.11: ## %cond.store13 +; SSE2-NEXT: movaps %xmm1, %xmm0 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1],xmm1[2,3] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB4_12: ## %else14 +; SSE2-NEXT: pextrw $6, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB4_14 +; SSE2-NEXT: ## %bb.13: ## %cond.store16 +; SSE2-NEXT: movaps %xmm1, %xmm0 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB4_14: ## %else17 +; SSE2-NEXT: pextrw $7, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB4_16 +; SSE2-NEXT: ## %bb.15: ## %cond.store19 +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1,2,3] +; SSE2-NEXT: movss %xmm1, (%rdi) +; SSE2-NEXT: LBB4_16: ## %else20 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v8f32_v8i1: +; SSE42: ## %bb.0: +; SSE42-NEXT: pextrb $0, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB4_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: movss %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB4_2: ## %else +; SSE42-NEXT: pextrb $2, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB4_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: extractps $1, %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB4_4: ## %else2 +; SSE42-NEXT: pextrb $4, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB4_6 +; SSE42-NEXT: ## %bb.5: ## %cond.store4 +; SSE42-NEXT: extractps $2, %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB4_6: ## %else5 +; SSE42-NEXT: pextrb $6, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB4_8 +; SSE42-NEXT: ## %bb.7: ## %cond.store7 +; SSE42-NEXT: extractps $3, %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB4_8: ## %else8 +; SSE42-NEXT: pextrb $8, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB4_10 +; SSE42-NEXT: ## %bb.9: ## %cond.store10 +; SSE42-NEXT: movss %xmm1, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB4_10: ## %else11 +; SSE42-NEXT: pextrb $10, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB4_12 +; SSE42-NEXT: ## %bb.11: ## %cond.store13 +; SSE42-NEXT: extractps $1, %xmm1, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB4_12: ## %else14 +; SSE42-NEXT: pextrb $12, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB4_14 +; SSE42-NEXT: ## %bb.13: ## %cond.store16 +; SSE42-NEXT: extractps $2, %xmm1, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB4_14: ## %else17 +; SSE42-NEXT: pextrb $14, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB4_16 +; SSE42-NEXT: ## %bb.15: ## %cond.store19 +; SSE42-NEXT: extractps $3, %xmm1, (%rdi) +; SSE42-NEXT: LBB4_16: ## %else20 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: compressstore_v8f32_v8i1: +; AVX1OR2: ## %bb.0: +; AVX1OR2-NEXT: vpextrb $0, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB4_2 +; AVX1OR2-NEXT: ## %bb.1: ## %cond.store +; AVX1OR2-NEXT: vmovss %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB4_2: ## %else +; AVX1OR2-NEXT: vpextrb $2, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB4_4 +; AVX1OR2-NEXT: ## %bb.3: ## %cond.store1 +; AVX1OR2-NEXT: vextractps $1, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB4_4: ## %else2 +; AVX1OR2-NEXT: vpextrb $4, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB4_6 +; AVX1OR2-NEXT: ## %bb.5: ## %cond.store4 +; AVX1OR2-NEXT: vextractps $2, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB4_6: ## %else5 +; AVX1OR2-NEXT: vpextrb $6, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB4_8 +; AVX1OR2-NEXT: ## %bb.7: ## %cond.store7 +; AVX1OR2-NEXT: vextractps $3, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB4_8: ## %else8 +; AVX1OR2-NEXT: vpextrb $8, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB4_10 +; AVX1OR2-NEXT: ## %bb.9: ## %cond.store10 +; AVX1OR2-NEXT: vextractf128 $1, %ymm0, %xmm2 +; AVX1OR2-NEXT: vmovss %xmm2, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB4_10: ## %else11 +; AVX1OR2-NEXT: vpextrb $10, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB4_12 +; AVX1OR2-NEXT: ## %bb.11: ## %cond.store13 +; AVX1OR2-NEXT: vextractf128 $1, %ymm0, %xmm2 +; AVX1OR2-NEXT: vextractps $1, %xmm2, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB4_12: ## %else14 +; AVX1OR2-NEXT: vpextrb $12, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB4_14 +; AVX1OR2-NEXT: ## %bb.13: ## %cond.store16 +; AVX1OR2-NEXT: vextractf128 $1, %ymm0, %xmm2 +; AVX1OR2-NEXT: vextractps $2, %xmm2, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB4_14: ## %else17 +; AVX1OR2-NEXT: vpextrb $14, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB4_16 +; AVX1OR2-NEXT: ## %bb.15: ## %cond.store19 +; AVX1OR2-NEXT: vextractf128 $1, %ymm0, %xmm0 +; AVX1OR2-NEXT: vextractps $3, %xmm0, (%rdi) +; AVX1OR2-NEXT: LBB4_16: ## %else20 +; AVX1OR2-NEXT: vzeroupper +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v8f32_v8i1: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 +; AVX512F-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero +; AVX512F-NEXT: vpsllq $63, %zmm1, %zmm1 +; AVX512F-NEXT: vptestmq %zmm1, %zmm1, %k1 +; AVX512F-NEXT: vcompressps %zmm0, (%rdi) {%k1} +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v8f32_v8i1: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vpsllw $15, %xmm1, %xmm1 +; AVX512VLBW-NEXT: vpmovw2m %xmm1, %k1 +; AVX512VLBW-NEXT: vcompressps %ymm0, (%rdi) {%k1} +; AVX512VLBW-NEXT: vzeroupper +; AVX512VLBW-NEXT: retq call void @llvm.masked.compressstore.v8f32(<8 x float> %V, float* %base, <8 x i1> %mask) ret void } define void @compressstore_v16f32_const(float* %base, <16 x float> %V) { -; HSW-LABEL: compressstore_v16f32_const: -; HSW: # %bb.0: # %cond.store -; HSW-NEXT: vmovups %ymm0, (%rdi) -; HSW-NEXT: vmovaps {{.*#+}} xmm0 = [0,1,2,4] -; HSW-NEXT: vpermps %ymm1, %ymm0, %ymm0 -; HSW-NEXT: vmovups %xmm0, 32(%rdi) -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm0 -; HSW-NEXT: vextractps $1, %xmm0, 48(%rdi) -; HSW-NEXT: vextractps $2, %xmm0, 52(%rdi) -; HSW-NEXT: vextractps $3, %xmm0, 56(%rdi) -; HSW-NEXT: vzeroupper -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v16f32_const: -; SKX: # %bb.0: -; SKX-NEXT: movw $-2049, %ax # imm = 0xF7FF -; SKX-NEXT: kmovd %eax, %k1 -; SKX-NEXT: vcompressps %zmm0, (%rdi) {%k1} -; SKX-NEXT: vzeroupper -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v16f32_const: -; KNL: # %bb.0: -; KNL-NEXT: movw $-2049, %ax # imm = 0xF7FF -; KNL-NEXT: kmovw %eax, %k1 -; KNL-NEXT: vcompressps %zmm0, (%rdi) {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v16f32_const: +; SSE2: ## %bb.0: ## %cond.store +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: movaps %xmm0, %xmm4 +; SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[1,1],xmm0[2,3] +; SSE2-NEXT: movss %xmm4, 4(%rdi) +; SSE2-NEXT: movaps %xmm0, %xmm4 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm4 = xmm4[1],xmm0[1] +; SSE2-NEXT: movss %xmm4, 8(%rdi) +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1,2,3] +; SSE2-NEXT: movss %xmm0, 12(%rdi) +; SSE2-NEXT: movss %xmm1, 16(%rdi) +; SSE2-NEXT: movaps %xmm1, %xmm0 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1],xmm1[2,3] +; SSE2-NEXT: movss %xmm0, 20(%rdi) +; SSE2-NEXT: movaps %xmm1, %xmm0 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1] +; SSE2-NEXT: movss %xmm0, 24(%rdi) +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1,2,3] +; SSE2-NEXT: movss %xmm1, 28(%rdi) +; SSE2-NEXT: movss %xmm2, 32(%rdi) +; SSE2-NEXT: movaps %xmm2, %xmm0 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1],xmm2[2,3] +; SSE2-NEXT: movss %xmm0, 36(%rdi) +; SSE2-NEXT: movhlps {{.*#+}} xmm2 = xmm2[1,1] +; SSE2-NEXT: movss %xmm2, 40(%rdi) +; SSE2-NEXT: movss %xmm3, 44(%rdi) +; SSE2-NEXT: movaps %xmm3, %xmm0 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1],xmm3[2,3] +; SSE2-NEXT: movss %xmm0, 48(%rdi) +; SSE2-NEXT: movaps %xmm3, %xmm0 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm3[1] +; SSE2-NEXT: movss %xmm0, 52(%rdi) +; SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[3,1,2,3] +; SSE2-NEXT: movss %xmm3, 56(%rdi) +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v16f32_const: +; SSE42: ## %bb.0: ## %cond.store +; SSE42-NEXT: movups %xmm0, (%rdi) +; SSE42-NEXT: movups %xmm1, 16(%rdi) +; SSE42-NEXT: insertps {{.*#+}} xmm2 = xmm2[0,1,2],xmm3[0] +; SSE42-NEXT: movups %xmm2, 32(%rdi) +; SSE42-NEXT: extractps $1, %xmm3, 48(%rdi) +; SSE42-NEXT: extractps $2, %xmm3, 52(%rdi) +; SSE42-NEXT: extractps $3, %xmm3, 56(%rdi) +; SSE42-NEXT: retq +; +; AVX1-LABEL: compressstore_v16f32_const: +; AVX1: ## %bb.0: ## %cond.store +; AVX1-NEXT: vmovups %ymm0, (%rdi) +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX1-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm0[0] +; AVX1-NEXT: vmovups %xmm1, 32(%rdi) +; AVX1-NEXT: vextractps $1, %xmm0, 48(%rdi) +; AVX1-NEXT: vextractps $2, %xmm0, 52(%rdi) +; AVX1-NEXT: vextractps $3, %xmm0, 56(%rdi) +; AVX1-NEXT: vzeroupper +; AVX1-NEXT: retq +; +; AVX2-LABEL: compressstore_v16f32_const: +; AVX2: ## %bb.0: ## %cond.store +; AVX2-NEXT: vmovups %ymm0, (%rdi) +; AVX2-NEXT: vmovaps {{.*#+}} xmm0 = [0,1,2,4] +; AVX2-NEXT: vpermps %ymm1, %ymm0, %ymm0 +; AVX2-NEXT: vmovups %xmm0, 32(%rdi) +; AVX2-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX2-NEXT: vextractps $1, %xmm0, 48(%rdi) +; AVX2-NEXT: vextractps $2, %xmm0, 52(%rdi) +; AVX2-NEXT: vextractps $3, %xmm0, 56(%rdi) +; AVX2-NEXT: vzeroupper +; AVX2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v16f32_const: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: movw $-2049, %ax ## imm = 0xF7FF +; AVX512F-NEXT: kmovw %eax, %k1 +; AVX512F-NEXT: vcompressps %zmm0, (%rdi) {%k1} +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v16f32_const: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: movw $-2049, %ax ## imm = 0xF7FF +; AVX512VLBW-NEXT: kmovd %eax, %k1 +; AVX512VLBW-NEXT: vcompressps %zmm0, (%rdi) {%k1} +; AVX512VLBW-NEXT: vzeroupper +; AVX512VLBW-NEXT: retq call void @llvm.masked.compressstore.v16f32(<16 x float> %V, float* %base, <16 x i1> ) ret void } define void @compressstore_v32f32_v32i32(float* %base, <32 x float> %V, <32 x i32> %trigger) { -; HSW-LABEL: compressstore_v32f32_v32i32: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm8, %xmm8, %xmm8 -; HSW-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm9 -; HSW-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 -; HSW-NEXT: vpextrb $0, %xmm9, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vmovd %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_2: # %else -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 -; HSW-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 -; HSW-NEXT: vpextrb $1, %xmm8, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vpextrd $1, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_4: # %else2 -; HSW-NEXT: vpxor %xmm8, %xmm8, %xmm8 -; HSW-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm9 -; HSW-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 -; HSW-NEXT: vpextrb $2, %xmm9, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_6 -; HSW-NEXT: # %bb.5: # %cond.store4 -; HSW-NEXT: vpextrd $2, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_6: # %else5 -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 -; HSW-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 -; HSW-NEXT: vpextrb $3, %xmm8, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_8 -; HSW-NEXT: # %bb.7: # %cond.store7 -; HSW-NEXT: vpextrd $3, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_8: # %else8 -; HSW-NEXT: vpxor %xmm8, %xmm8, %xmm8 -; HSW-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 -; HSW-NEXT: vextracti128 $1, %ymm8, %xmm8 -; HSW-NEXT: vpackssdw %xmm8, %xmm0, %xmm9 -; HSW-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 -; HSW-NEXT: vpextrb $4, %xmm9, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_10 -; HSW-NEXT: # %bb.9: # %cond.store10 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm9 -; HSW-NEXT: vmovd %xmm9, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_10: # %else11 -; HSW-NEXT: vpackssdw %xmm8, %xmm0, %xmm8 -; HSW-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 -; HSW-NEXT: vpextrb $5, %xmm8, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_12 -; HSW-NEXT: # %bb.11: # %cond.store13 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm8 -; HSW-NEXT: vpextrd $1, %xmm8, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_12: # %else14 -; HSW-NEXT: vpxor %xmm8, %xmm8, %xmm8 -; HSW-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm4 -; HSW-NEXT: vextracti128 $1, %ymm4, %xmm8 -; HSW-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpextrb $6, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_14 -; HSW-NEXT: # %bb.13: # %cond.store16 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm4 -; HSW-NEXT: vpextrd $2, %xmm4, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_14: # %else17 -; HSW-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpextrb $7, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_16 -; HSW-NEXT: # %bb.15: # %cond.store19 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm0 -; HSW-NEXT: vpextrd $3, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_16: # %else20 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm5, %ymm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $8, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_18 -; HSW-NEXT: # %bb.17: # %cond.store22 -; HSW-NEXT: vmovd %xmm1, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_18: # %else23 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $9, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_20 -; HSW-NEXT: # %bb.19: # %cond.store25 -; HSW-NEXT: vpextrd $1, %xmm1, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_20: # %else26 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm5, %ymm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $10, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_22 -; HSW-NEXT: # %bb.21: # %cond.store28 -; HSW-NEXT: vpextrd $2, %xmm1, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_22: # %else29 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $11, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_24 -; HSW-NEXT: # %bb.23: # %cond.store31 -; HSW-NEXT: vpextrd $3, %xmm1, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_24: # %else32 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm5, %ymm0 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $12, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_26 -; HSW-NEXT: # %bb.25: # %cond.store34 -; HSW-NEXT: vextracti128 $1, %ymm1, %xmm4 -; HSW-NEXT: vmovd %xmm4, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_26: # %else35 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $13, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_28 -; HSW-NEXT: # %bb.27: # %cond.store37 -; HSW-NEXT: vextracti128 $1, %ymm1, %xmm0 -; HSW-NEXT: vpextrd $1, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_28: # %else38 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm5, %ymm0 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $14, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_30 -; HSW-NEXT: # %bb.29: # %cond.store40 -; HSW-NEXT: vextracti128 $1, %ymm1, %xmm4 -; HSW-NEXT: vpextrd $2, %xmm4, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_30: # %else41 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $15, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_32 -; HSW-NEXT: # %bb.31: # %cond.store43 -; HSW-NEXT: vextracti128 $1, %ymm1, %xmm0 -; HSW-NEXT: vpextrd $3, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_32: # %else44 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm6, %ymm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 -; HSW-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $0, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_34 -; HSW-NEXT: # %bb.33: # %cond.store46 -; HSW-NEXT: vmovd %xmm2, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_34: # %else47 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $1, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_36 -; HSW-NEXT: # %bb.35: # %cond.store49 -; HSW-NEXT: vpextrd $1, %xmm2, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_36: # %else50 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm6, %ymm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 -; HSW-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $2, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_38 -; HSW-NEXT: # %bb.37: # %cond.store52 -; HSW-NEXT: vpextrd $2, %xmm2, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_38: # %else53 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $3, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_40 -; HSW-NEXT: # %bb.39: # %cond.store55 -; HSW-NEXT: vpextrd $3, %xmm2, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_40: # %else56 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm6, %ymm0 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 -; HSW-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $4, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_42 -; HSW-NEXT: # %bb.41: # %cond.store58 -; HSW-NEXT: vextracti128 $1, %ymm2, %xmm1 -; HSW-NEXT: vmovd %xmm1, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_42: # %else59 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $5, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_44 -; HSW-NEXT: # %bb.43: # %cond.store61 -; HSW-NEXT: vextracti128 $1, %ymm2, %xmm0 -; HSW-NEXT: vpextrd $1, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_44: # %else62 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm6, %ymm0 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 -; HSW-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $6, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_46 -; HSW-NEXT: # %bb.45: # %cond.store64 -; HSW-NEXT: vextracti128 $1, %ymm2, %xmm1 -; HSW-NEXT: vpextrd $2, %xmm1, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_46: # %else65 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $7, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_48 -; HSW-NEXT: # %bb.47: # %cond.store67 -; HSW-NEXT: vextracti128 $1, %ymm2, %xmm0 -; HSW-NEXT: vpextrd $3, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_48: # %else68 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm7, %ymm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 -; HSW-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 -; HSW-NEXT: vpextrb $8, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_50 -; HSW-NEXT: # %bb.49: # %cond.store70 -; HSW-NEXT: vmovd %xmm3, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_50: # %else71 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $9, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_52 -; HSW-NEXT: # %bb.51: # %cond.store73 -; HSW-NEXT: vpextrd $1, %xmm3, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_52: # %else74 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm7, %ymm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 -; HSW-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 -; HSW-NEXT: vpextrb $10, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_54 -; HSW-NEXT: # %bb.53: # %cond.store76 -; HSW-NEXT: vpextrd $2, %xmm3, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_54: # %else77 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $11, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_56 -; HSW-NEXT: # %bb.55: # %cond.store79 -; HSW-NEXT: vpextrd $3, %xmm3, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_56: # %else80 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm7, %ymm0 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 -; HSW-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 -; HSW-NEXT: vpextrb $12, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_58 -; HSW-NEXT: # %bb.57: # %cond.store82 -; HSW-NEXT: vextracti128 $1, %ymm3, %xmm1 -; HSW-NEXT: vmovd %xmm1, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_58: # %else83 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $13, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_60 -; HSW-NEXT: # %bb.59: # %cond.store85 -; HSW-NEXT: vextracti128 $1, %ymm3, %xmm0 -; HSW-NEXT: vpextrd $1, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_60: # %else86 -; HSW-NEXT: vpxor %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpcmpeqd %ymm0, %ymm7, %ymm0 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm0 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 -; HSW-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 -; HSW-NEXT: vpextrb $14, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_62 -; HSW-NEXT: # %bb.61: # %cond.store88 -; HSW-NEXT: vextracti128 $1, %ymm3, %xmm1 -; HSW-NEXT: vpextrd $2, %xmm1, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB6_62: # %else89 -; HSW-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 -; HSW-NEXT: vpextrb $15, %xmm0, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB6_64 -; HSW-NEXT: # %bb.63: # %cond.store91 -; HSW-NEXT: vextracti128 $1, %ymm3, %xmm0 -; HSW-NEXT: vpextrd $3, %xmm0, (%rdi) -; HSW-NEXT: .LBB6_64: # %else92 -; HSW-NEXT: vzeroupper -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v32f32_v32i32: -; SKX: # %bb.0: -; SKX-NEXT: vptestnmd %zmm3, %zmm3, %k1 -; SKX-NEXT: vptestnmd %zmm2, %zmm2, %k2 -; SKX-NEXT: kmovw %k2, %eax -; SKX-NEXT: popcntl %eax, %eax -; SKX-NEXT: vcompressps %zmm1, (%rdi,%rax,4) {%k1} -; SKX-NEXT: vcompressps %zmm0, (%rdi) {%k2} -; SKX-NEXT: vzeroupper -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v32f32_v32i32: -; KNL: # %bb.0: -; KNL-NEXT: vptestnmd %zmm3, %zmm3, %k1 -; KNL-NEXT: vptestnmd %zmm2, %zmm2, %k2 -; KNL-NEXT: kmovw %k2, %eax -; KNL-NEXT: popcntl %eax, %eax -; KNL-NEXT: vcompressps %zmm1, (%rdi,%rax,4) {%k1} -; KNL-NEXT: vcompressps %zmm0, (%rdi) {%k2} -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v32f32_v32i32: +; SSE2: ## %bb.0: +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE2-NEXT: pxor %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: movd %xmm8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_2: ## %else +; SSE2-NEXT: pextrw $2, %xmm8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: movaps %xmm0, %xmm8 +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[1,1],xmm0[2,3] +; SSE2-NEXT: movss %xmm8, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_4: ## %else2 +; SSE2-NEXT: xorps %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: pextrw $4, %xmm9, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_6 +; SSE2-NEXT: ## %bb.5: ## %cond.store4 +; SSE2-NEXT: movaps %xmm0, %xmm8 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm8 = xmm8[1],xmm0[1] +; SSE2-NEXT: movss %xmm8, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_6: ## %else5 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE2-NEXT: pextrw $6, %xmm9, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_8 +; SSE2-NEXT: ## %bb.7: ## %cond.store7 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,1,2,3] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_8: ## %else8 +; SSE2-NEXT: xorps %xmm0, %xmm0 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm0 +; SSE2-NEXT: movd %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_10 +; SSE2-NEXT: ## %bb.9: ## %cond.store10 +; SSE2-NEXT: movss %xmm1, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_10: ## %else11 +; SSE2-NEXT: pextrw $2, %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_12 +; SSE2-NEXT: ## %bb.11: ## %cond.store13 +; SSE2-NEXT: movaps %xmm1, %xmm0 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1],xmm1[2,3] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_12: ## %else14 +; SSE2-NEXT: xorps %xmm0, %xmm0 +; SSE2-NEXT: pcmpeqd %xmm0, %xmm8 +; SSE2-NEXT: pextrw $4, %xmm8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_14 +; SSE2-NEXT: ## %bb.13: ## %cond.store16 +; SSE2-NEXT: movaps %xmm1, %xmm0 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_14: ## %else17 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm0 +; SSE2-NEXT: pextrw $6, %xmm8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_16 +; SSE2-NEXT: ## %bb.15: ## %cond.store19 +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,1,2,3] +; SSE2-NEXT: movss %xmm1, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_16: ## %else20 +; SSE2-NEXT: xorps %xmm1, %xmm1 +; SSE2-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE2-NEXT: movd %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_18 +; SSE2-NEXT: ## %bb.17: ## %cond.store22 +; SSE2-NEXT: movss %xmm2, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_18: ## %else23 +; SSE2-NEXT: pextrw $2, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_20 +; SSE2-NEXT: ## %bb.19: ## %cond.store25 +; SSE2-NEXT: movaps %xmm2, %xmm1 +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1],xmm2[2,3] +; SSE2-NEXT: movss %xmm1, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_20: ## %else26 +; SSE2-NEXT: xorps %xmm1, %xmm1 +; SSE2-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE2-NEXT: pextrw $4, %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_22 +; SSE2-NEXT: ## %bb.21: ## %cond.store28 +; SSE2-NEXT: movaps %xmm2, %xmm1 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm2[1] +; SSE2-NEXT: movss %xmm1, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_22: ## %else29 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm1 +; SSE2-NEXT: pextrw $6, %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_24 +; SSE2-NEXT: ## %bb.23: ## %cond.store31 +; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[3,1,2,3] +; SSE2-NEXT: movss %xmm2, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_24: ## %else32 +; SSE2-NEXT: pxor %xmm0, %xmm0 +; SSE2-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE2-NEXT: movd %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_26 +; SSE2-NEXT: ## %bb.25: ## %cond.store34 +; SSE2-NEXT: movss %xmm3, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_26: ## %else35 +; SSE2-NEXT: pextrw $2, %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_28 +; SSE2-NEXT: ## %bb.27: ## %cond.store37 +; SSE2-NEXT: movaps %xmm3, %xmm0 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1],xmm3[2,3] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_28: ## %else38 +; SSE2-NEXT: xorps %xmm0, %xmm0 +; SSE2-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE2-NEXT: pextrw $4, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_30 +; SSE2-NEXT: ## %bb.29: ## %cond.store40 +; SSE2-NEXT: movaps %xmm3, %xmm0 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm3[1] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_30: ## %else41 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm0 +; SSE2-NEXT: pextrw $6, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_32 +; SSE2-NEXT: ## %bb.31: ## %cond.store43 +; SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[3,1,2,3] +; SSE2-NEXT: movss %xmm3, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_32: ## %else44 +; SSE2-NEXT: pxor %xmm1, %xmm1 +; SSE2-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE2-NEXT: movd %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_34 +; SSE2-NEXT: ## %bb.33: ## %cond.store46 +; SSE2-NEXT: movss %xmm4, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_34: ## %else47 +; SSE2-NEXT: pextrw $2, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_36 +; SSE2-NEXT: ## %bb.35: ## %cond.store49 +; SSE2-NEXT: movaps %xmm4, %xmm1 +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1],xmm4[2,3] +; SSE2-NEXT: movss %xmm1, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_36: ## %else50 +; SSE2-NEXT: xorps %xmm1, %xmm1 +; SSE2-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE2-NEXT: pextrw $4, %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_38 +; SSE2-NEXT: ## %bb.37: ## %cond.store52 +; SSE2-NEXT: movaps %xmm4, %xmm1 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm4[1] +; SSE2-NEXT: movss %xmm1, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_38: ## %else53 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm1 +; SSE2-NEXT: pextrw $6, %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_40 +; SSE2-NEXT: ## %bb.39: ## %cond.store55 +; SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[3,1,2,3] +; SSE2-NEXT: movss %xmm4, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_40: ## %else56 +; SSE2-NEXT: pxor %xmm0, %xmm0 +; SSE2-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE2-NEXT: movd %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_42 +; SSE2-NEXT: ## %bb.41: ## %cond.store58 +; SSE2-NEXT: movss %xmm5, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_42: ## %else59 +; SSE2-NEXT: pextrw $2, %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_44 +; SSE2-NEXT: ## %bb.43: ## %cond.store61 +; SSE2-NEXT: movaps %xmm5, %xmm0 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1],xmm5[2,3] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_44: ## %else62 +; SSE2-NEXT: xorps %xmm0, %xmm0 +; SSE2-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE2-NEXT: pextrw $4, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_46 +; SSE2-NEXT: ## %bb.45: ## %cond.store64 +; SSE2-NEXT: movaps %xmm5, %xmm0 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm5[1] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_46: ## %else65 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm0 +; SSE2-NEXT: pextrw $6, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_48 +; SSE2-NEXT: ## %bb.47: ## %cond.store67 +; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[3,1,2,3] +; SSE2-NEXT: movss %xmm5, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_48: ## %else68 +; SSE2-NEXT: pxor %xmm1, %xmm1 +; SSE2-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE2-NEXT: movd %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_50 +; SSE2-NEXT: ## %bb.49: ## %cond.store70 +; SSE2-NEXT: movss %xmm6, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_50: ## %else71 +; SSE2-NEXT: pextrw $2, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_52 +; SSE2-NEXT: ## %bb.51: ## %cond.store73 +; SSE2-NEXT: movaps %xmm6, %xmm1 +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1],xmm6[2,3] +; SSE2-NEXT: movss %xmm1, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_52: ## %else74 +; SSE2-NEXT: xorps %xmm1, %xmm1 +; SSE2-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE2-NEXT: pextrw $4, %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_54 +; SSE2-NEXT: ## %bb.53: ## %cond.store76 +; SSE2-NEXT: movaps %xmm6, %xmm1 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm6[1] +; SSE2-NEXT: movss %xmm1, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_54: ## %else77 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm1 +; SSE2-NEXT: pextrw $6, %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_56 +; SSE2-NEXT: ## %bb.55: ## %cond.store79 +; SSE2-NEXT: shufps {{.*#+}} xmm6 = xmm6[3,1,2,3] +; SSE2-NEXT: movss %xmm6, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_56: ## %else80 +; SSE2-NEXT: pxor %xmm0, %xmm0 +; SSE2-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE2-NEXT: movd %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_58 +; SSE2-NEXT: ## %bb.57: ## %cond.store82 +; SSE2-NEXT: movss %xmm7, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_58: ## %else83 +; SSE2-NEXT: pextrw $2, %xmm0, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_60 +; SSE2-NEXT: ## %bb.59: ## %cond.store85 +; SSE2-NEXT: movaps %xmm7, %xmm0 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1],xmm7[2,3] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_60: ## %else86 +; SSE2-NEXT: xorps %xmm0, %xmm0 +; SSE2-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE2-NEXT: pextrw $4, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_62 +; SSE2-NEXT: ## %bb.61: ## %cond.store88 +; SSE2-NEXT: movaps %xmm7, %xmm0 +; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm7[1] +; SSE2-NEXT: movss %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB6_62: ## %else89 +; SSE2-NEXT: pextrw $6, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB6_64 +; SSE2-NEXT: ## %bb.63: ## %cond.store91 +; SSE2-NEXT: shufps {{.*#+}} xmm7 = xmm7[3,1,2,3] +; SSE2-NEXT: movss %xmm7, (%rdi) +; SSE2-NEXT: LBB6_64: ## %else92 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v32f32_v32i32: +; SSE42: ## %bb.0: +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $0, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: movd %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_2: ## %else +; SSE42-NEXT: pextrb $4, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: pextrd $1, %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_4: ## %else2 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $8, %xmm9, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_6 +; SSE42-NEXT: ## %bb.5: ## %cond.store4 +; SSE42-NEXT: pextrd $2, %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_6: ## %else5 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE42-NEXT: pextrb $12, %xmm9, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_8 +; SSE42-NEXT: ## %bb.7: ## %cond.store7 +; SSE42-NEXT: pextrd $3, %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_8: ## %else8 +; SSE42-NEXT: pxor %xmm0, %xmm0 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm0 +; SSE42-NEXT: pextrb $0, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_10 +; SSE42-NEXT: ## %bb.9: ## %cond.store10 +; SSE42-NEXT: movd %xmm1, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_10: ## %else11 +; SSE42-NEXT: pextrb $4, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_12 +; SSE42-NEXT: ## %bb.11: ## %cond.store13 +; SSE42-NEXT: pextrd $1, %xmm1, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_12: ## %else14 +; SSE42-NEXT: pxor %xmm0, %xmm0 +; SSE42-NEXT: pcmpeqd %xmm0, %xmm8 +; SSE42-NEXT: pextrb $8, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_14 +; SSE42-NEXT: ## %bb.13: ## %cond.store16 +; SSE42-NEXT: pextrd $2, %xmm1, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_14: ## %else17 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm0 +; SSE42-NEXT: pextrb $12, %xmm8, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_16 +; SSE42-NEXT: ## %bb.15: ## %cond.store19 +; SSE42-NEXT: pextrd $3, %xmm1, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_16: ## %else20 +; SSE42-NEXT: pxor %xmm1, %xmm1 +; SSE42-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE42-NEXT: pextrb $0, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_18 +; SSE42-NEXT: ## %bb.17: ## %cond.store22 +; SSE42-NEXT: movss %xmm2, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_18: ## %else23 +; SSE42-NEXT: pextrb $4, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_20 +; SSE42-NEXT: ## %bb.19: ## %cond.store25 +; SSE42-NEXT: extractps $1, %xmm2, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_20: ## %else26 +; SSE42-NEXT: pxor %xmm1, %xmm1 +; SSE42-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE42-NEXT: pextrb $8, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_22 +; SSE42-NEXT: ## %bb.21: ## %cond.store28 +; SSE42-NEXT: extractps $2, %xmm2, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_22: ## %else29 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm1 +; SSE42-NEXT: pextrb $12, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_24 +; SSE42-NEXT: ## %bb.23: ## %cond.store31 +; SSE42-NEXT: extractps $3, %xmm2, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_24: ## %else32 +; SSE42-NEXT: pxor %xmm0, %xmm0 +; SSE42-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE42-NEXT: pextrb $0, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_26 +; SSE42-NEXT: ## %bb.25: ## %cond.store34 +; SSE42-NEXT: movss %xmm3, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_26: ## %else35 +; SSE42-NEXT: pextrb $4, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_28 +; SSE42-NEXT: ## %bb.27: ## %cond.store37 +; SSE42-NEXT: extractps $1, %xmm3, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_28: ## %else38 +; SSE42-NEXT: pxor %xmm0, %xmm0 +; SSE42-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE42-NEXT: pextrb $8, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_30 +; SSE42-NEXT: ## %bb.29: ## %cond.store40 +; SSE42-NEXT: extractps $2, %xmm3, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_30: ## %else41 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm0 +; SSE42-NEXT: pextrb $12, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_32 +; SSE42-NEXT: ## %bb.31: ## %cond.store43 +; SSE42-NEXT: extractps $3, %xmm3, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_32: ## %else44 +; SSE42-NEXT: pxor %xmm1, %xmm1 +; SSE42-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE42-NEXT: pextrb $0, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_34 +; SSE42-NEXT: ## %bb.33: ## %cond.store46 +; SSE42-NEXT: movss %xmm4, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_34: ## %else47 +; SSE42-NEXT: pextrb $4, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_36 +; SSE42-NEXT: ## %bb.35: ## %cond.store49 +; SSE42-NEXT: extractps $1, %xmm4, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_36: ## %else50 +; SSE42-NEXT: pxor %xmm1, %xmm1 +; SSE42-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE42-NEXT: pextrb $8, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_38 +; SSE42-NEXT: ## %bb.37: ## %cond.store52 +; SSE42-NEXT: extractps $2, %xmm4, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_38: ## %else53 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm1 +; SSE42-NEXT: pextrb $12, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_40 +; SSE42-NEXT: ## %bb.39: ## %cond.store55 +; SSE42-NEXT: extractps $3, %xmm4, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_40: ## %else56 +; SSE42-NEXT: pxor %xmm0, %xmm0 +; SSE42-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE42-NEXT: pextrb $0, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_42 +; SSE42-NEXT: ## %bb.41: ## %cond.store58 +; SSE42-NEXT: movss %xmm5, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_42: ## %else59 +; SSE42-NEXT: pextrb $4, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_44 +; SSE42-NEXT: ## %bb.43: ## %cond.store61 +; SSE42-NEXT: extractps $1, %xmm5, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_44: ## %else62 +; SSE42-NEXT: pxor %xmm0, %xmm0 +; SSE42-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE42-NEXT: pextrb $8, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_46 +; SSE42-NEXT: ## %bb.45: ## %cond.store64 +; SSE42-NEXT: extractps $2, %xmm5, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_46: ## %else65 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm0 +; SSE42-NEXT: pextrb $12, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_48 +; SSE42-NEXT: ## %bb.47: ## %cond.store67 +; SSE42-NEXT: extractps $3, %xmm5, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_48: ## %else68 +; SSE42-NEXT: pxor %xmm1, %xmm1 +; SSE42-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE42-NEXT: pextrb $0, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_50 +; SSE42-NEXT: ## %bb.49: ## %cond.store70 +; SSE42-NEXT: movss %xmm6, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_50: ## %else71 +; SSE42-NEXT: pextrb $4, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_52 +; SSE42-NEXT: ## %bb.51: ## %cond.store73 +; SSE42-NEXT: extractps $1, %xmm6, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_52: ## %else74 +; SSE42-NEXT: pxor %xmm1, %xmm1 +; SSE42-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE42-NEXT: pextrb $8, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_54 +; SSE42-NEXT: ## %bb.53: ## %cond.store76 +; SSE42-NEXT: extractps $2, %xmm6, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_54: ## %else77 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm1 +; SSE42-NEXT: pextrb $12, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_56 +; SSE42-NEXT: ## %bb.55: ## %cond.store79 +; SSE42-NEXT: extractps $3, %xmm6, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_56: ## %else80 +; SSE42-NEXT: pxor %xmm0, %xmm0 +; SSE42-NEXT: pcmpeqd %xmm1, %xmm0 +; SSE42-NEXT: pextrb $0, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_58 +; SSE42-NEXT: ## %bb.57: ## %cond.store82 +; SSE42-NEXT: movss %xmm7, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_58: ## %else83 +; SSE42-NEXT: pextrb $4, %xmm0, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_60 +; SSE42-NEXT: ## %bb.59: ## %cond.store85 +; SSE42-NEXT: extractps $1, %xmm7, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_60: ## %else86 +; SSE42-NEXT: pxor %xmm0, %xmm0 +; SSE42-NEXT: pcmpeqd %xmm0, %xmm1 +; SSE42-NEXT: pextrb $8, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_62 +; SSE42-NEXT: ## %bb.61: ## %cond.store88 +; SSE42-NEXT: extractps $2, %xmm7, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB6_62: ## %else89 +; SSE42-NEXT: pextrb $12, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB6_64 +; SSE42-NEXT: ## %bb.63: ## %cond.store91 +; SSE42-NEXT: extractps $3, %xmm7, (%rdi) +; SSE42-NEXT: LBB6_64: ## %else92 +; SSE42-NEXT: retq +; +; AVX1-LABEL: compressstore_v32f32_v32i32: +; AVX1: ## %bb.0: +; AVX1-NEXT: vpxor %xmm8, %xmm8, %xmm8 +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm4, %xmm9 +; AVX1-NEXT: vpackssdw %xmm0, %xmm9, %xmm9 +; AVX1-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 +; AVX1-NEXT: vpextrb $0, %xmm9, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_2 +; AVX1-NEXT: ## %bb.1: ## %cond.store +; AVX1-NEXT: vmovd %xmm0, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_2: ## %else +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm4, %xmm8 +; AVX1-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 +; AVX1-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 +; AVX1-NEXT: vpextrb $1, %xmm8, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_4 +; AVX1-NEXT: ## %bb.3: ## %cond.store1 +; AVX1-NEXT: vpextrd $1, %xmm0, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_4: ## %else2 +; AVX1-NEXT: vpxor %xmm8, %xmm8, %xmm8 +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm4, %xmm9 +; AVX1-NEXT: vpackssdw %xmm0, %xmm9, %xmm9 +; AVX1-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 +; AVX1-NEXT: vpextrb $2, %xmm9, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_6 +; AVX1-NEXT: ## %bb.5: ## %cond.store4 +; AVX1-NEXT: vpextrd $2, %xmm0, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_6: ## %else5 +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm4, %xmm8 +; AVX1-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 +; AVX1-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 +; AVX1-NEXT: vpextrb $3, %xmm8, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_8 +; AVX1-NEXT: ## %bb.7: ## %cond.store7 +; AVX1-NEXT: vpextrd $3, %xmm0, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_8: ## %else8 +; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm8 +; AVX1-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm8, %xmm9 +; AVX1-NEXT: vpackssdw %xmm9, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $4, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_10 +; AVX1-NEXT: ## %bb.9: ## %cond.store10 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; AVX1-NEXT: vmovd %xmm4, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_10: ## %else11 +; AVX1-NEXT: vpackssdw %xmm9, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $5, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_12 +; AVX1-NEXT: ## %bb.11: ## %cond.store13 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; AVX1-NEXT: vpextrd $1, %xmm4, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_12: ## %else14 +; AVX1-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm8, %xmm8 +; AVX1-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $6, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_14 +; AVX1-NEXT: ## %bb.13: ## %cond.store16 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; AVX1-NEXT: vpextrd $2, %xmm4, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_14: ## %else17 +; AVX1-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $7, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_16 +; AVX1-NEXT: ## %bb.15: ## %cond.store19 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; AVX1-NEXT: vpextrd $3, %xmm0, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_16: ## %else20 +; AVX1-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm5, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $8, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_18 +; AVX1-NEXT: ## %bb.17: ## %cond.store22 +; AVX1-NEXT: vmovss %xmm1, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_18: ## %else23 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm5, %xmm0 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpextrb $9, %xmm0, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_20 +; AVX1-NEXT: ## %bb.19: ## %cond.store25 +; AVX1-NEXT: vextractps $1, %xmm1, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_20: ## %else26 +; AVX1-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm5, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $10, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_22 +; AVX1-NEXT: ## %bb.21: ## %cond.store28 +; AVX1-NEXT: vextractps $2, %xmm1, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_22: ## %else29 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm5, %xmm0 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpextrb $11, %xmm0, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_24 +; AVX1-NEXT: ## %bb.23: ## %cond.store31 +; AVX1-NEXT: vextractps $3, %xmm1, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_24: ## %else32 +; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm0 +; AVX1-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 +; AVX1-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX1-NEXT: vpextrb $12, %xmm5, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_26 +; AVX1-NEXT: ## %bb.25: ## %cond.store34 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5 +; AVX1-NEXT: vmovd %xmm5, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_26: ## %else35 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $13, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_28 +; AVX1-NEXT: ## %bb.27: ## %cond.store37 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm4 +; AVX1-NEXT: vpextrd $1, %xmm4, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_28: ## %else38 +; AVX1-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm0, %xmm0 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $14, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_30 +; AVX1-NEXT: ## %bb.29: ## %cond.store40 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm4 +; AVX1-NEXT: vpextrd $2, %xmm4, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_30: ## %else41 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpextrb $15, %xmm0, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_32 +; AVX1-NEXT: ## %bb.31: ## %cond.store43 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX1-NEXT: vpextrd $3, %xmm0, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_32: ## %else44 +; AVX1-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm6, %xmm1 +; AVX1-NEXT: vpackssdw %xmm0, %xmm1, %xmm1 +; AVX1-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 +; AVX1-NEXT: vpextrb $0, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_34 +; AVX1-NEXT: ## %bb.33: ## %cond.store46 +; AVX1-NEXT: vmovss %xmm2, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_34: ## %else47 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm6, %xmm0 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpextrb $1, %xmm0, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_36 +; AVX1-NEXT: ## %bb.35: ## %cond.store49 +; AVX1-NEXT: vextractps $1, %xmm2, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_36: ## %else50 +; AVX1-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm6, %xmm1 +; AVX1-NEXT: vpackssdw %xmm0, %xmm1, %xmm1 +; AVX1-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 +; AVX1-NEXT: vpextrb $2, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_38 +; AVX1-NEXT: ## %bb.37: ## %cond.store52 +; AVX1-NEXT: vextractps $2, %xmm2, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_38: ## %else53 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm6, %xmm0 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpextrb $3, %xmm0, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_40 +; AVX1-NEXT: ## %bb.39: ## %cond.store55 +; AVX1-NEXT: vextractps $3, %xmm2, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_40: ## %else56 +; AVX1-NEXT: vextractf128 $1, %ymm6, %xmm0 +; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm1 +; AVX1-NEXT: vpackssdw %xmm1, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $4, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_42 +; AVX1-NEXT: ## %bb.41: ## %cond.store58 +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm4 +; AVX1-NEXT: vmovd %xmm4, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_42: ## %else59 +; AVX1-NEXT: vpackssdw %xmm1, %xmm0, %xmm1 +; AVX1-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 +; AVX1-NEXT: vpextrb $5, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_44 +; AVX1-NEXT: ## %bb.43: ## %cond.store61 +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm1 +; AVX1-NEXT: vpextrd $1, %xmm1, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_44: ## %else62 +; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 +; AVX1-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 +; AVX1-NEXT: vpextrb $6, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_46 +; AVX1-NEXT: ## %bb.45: ## %cond.store64 +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm1 +; AVX1-NEXT: vpextrd $2, %xmm1, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_46: ## %else65 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpextrb $7, %xmm0, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_48 +; AVX1-NEXT: ## %bb.47: ## %cond.store67 +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm0 +; AVX1-NEXT: vpextrd $3, %xmm0, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_48: ## %else68 +; AVX1-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm7, %xmm1 +; AVX1-NEXT: vpackssdw %xmm0, %xmm1, %xmm1 +; AVX1-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 +; AVX1-NEXT: vpextrb $8, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_50 +; AVX1-NEXT: ## %bb.49: ## %cond.store70 +; AVX1-NEXT: vmovss %xmm3, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_50: ## %else71 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm7, %xmm0 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpextrb $9, %xmm0, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_52 +; AVX1-NEXT: ## %bb.51: ## %cond.store73 +; AVX1-NEXT: vextractps $1, %xmm3, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_52: ## %else74 +; AVX1-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm7, %xmm1 +; AVX1-NEXT: vpackssdw %xmm0, %xmm1, %xmm1 +; AVX1-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 +; AVX1-NEXT: vpextrb $10, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_54 +; AVX1-NEXT: ## %bb.53: ## %cond.store76 +; AVX1-NEXT: vextractps $2, %xmm3, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_54: ## %else77 +; AVX1-NEXT: vpcmpeqd %xmm0, %xmm7, %xmm0 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpextrb $11, %xmm0, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_56 +; AVX1-NEXT: ## %bb.55: ## %cond.store79 +; AVX1-NEXT: vextractps $3, %xmm3, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_56: ## %else80 +; AVX1-NEXT: vextractf128 $1, %ymm7, %xmm0 +; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm1 +; AVX1-NEXT: vpackssdw %xmm1, %xmm0, %xmm2 +; AVX1-NEXT: vpacksswb %xmm2, %xmm0, %xmm2 +; AVX1-NEXT: vpextrb $12, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_58 +; AVX1-NEXT: ## %bb.57: ## %cond.store82 +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm2 +; AVX1-NEXT: vmovd %xmm2, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_58: ## %else83 +; AVX1-NEXT: vpackssdw %xmm1, %xmm0, %xmm1 +; AVX1-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 +; AVX1-NEXT: vpextrb $13, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_60 +; AVX1-NEXT: ## %bb.59: ## %cond.store85 +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm1 +; AVX1-NEXT: vpextrd $1, %xmm1, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_60: ## %else86 +; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 +; AVX1-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 +; AVX1-NEXT: vpextrb $14, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_62 +; AVX1-NEXT: ## %bb.61: ## %cond.store88 +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm1 +; AVX1-NEXT: vpextrd $2, %xmm1, (%rdi) +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB6_62: ## %else89 +; AVX1-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX1-NEXT: vpextrb $15, %xmm0, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB6_64 +; AVX1-NEXT: ## %bb.63: ## %cond.store91 +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm0 +; AVX1-NEXT: vpextrd $3, %xmm0, (%rdi) +; AVX1-NEXT: LBB6_64: ## %else92 +; AVX1-NEXT: vzeroupper +; AVX1-NEXT: retq +; +; AVX2-LABEL: compressstore_v32f32_v32i32: +; AVX2: ## %bb.0: +; AVX2-NEXT: vpxor %xmm8, %xmm8, %xmm8 +; AVX2-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm9 +; AVX2-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 +; AVX2-NEXT: vpextrb $0, %xmm9, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_2 +; AVX2-NEXT: ## %bb.1: ## %cond.store +; AVX2-NEXT: vmovd %xmm0, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_2: ## %else +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 +; AVX2-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 +; AVX2-NEXT: vpextrb $1, %xmm8, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_4 +; AVX2-NEXT: ## %bb.3: ## %cond.store1 +; AVX2-NEXT: vpextrd $1, %xmm0, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_4: ## %else2 +; AVX2-NEXT: vpxor %xmm8, %xmm8, %xmm8 +; AVX2-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm9 +; AVX2-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 +; AVX2-NEXT: vpextrb $2, %xmm9, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_6 +; AVX2-NEXT: ## %bb.5: ## %cond.store4 +; AVX2-NEXT: vpextrd $2, %xmm0, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_6: ## %else5 +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 +; AVX2-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 +; AVX2-NEXT: vpextrb $3, %xmm8, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_8 +; AVX2-NEXT: ## %bb.7: ## %cond.store7 +; AVX2-NEXT: vpextrd $3, %xmm0, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_8: ## %else8 +; AVX2-NEXT: vpxor %xmm8, %xmm8, %xmm8 +; AVX2-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 +; AVX2-NEXT: vextracti128 $1, %ymm8, %xmm8 +; AVX2-NEXT: vpackssdw %xmm8, %xmm0, %xmm9 +; AVX2-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 +; AVX2-NEXT: vpextrb $4, %xmm9, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_10 +; AVX2-NEXT: ## %bb.9: ## %cond.store10 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm9 +; AVX2-NEXT: vmovd %xmm9, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_10: ## %else11 +; AVX2-NEXT: vpackssdw %xmm8, %xmm0, %xmm8 +; AVX2-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 +; AVX2-NEXT: vpextrb $5, %xmm8, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_12 +; AVX2-NEXT: ## %bb.11: ## %cond.store13 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm8 +; AVX2-NEXT: vpextrd $1, %xmm8, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_12: ## %else14 +; AVX2-NEXT: vpxor %xmm8, %xmm8, %xmm8 +; AVX2-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm4 +; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm8 +; AVX2-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpextrb $6, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_14 +; AVX2-NEXT: ## %bb.13: ## %cond.store16 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm4 +; AVX2-NEXT: vpextrd $2, %xmm4, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_14: ## %else17 +; AVX2-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpextrb $7, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_16 +; AVX2-NEXT: ## %bb.15: ## %cond.store19 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0 +; AVX2-NEXT: vpextrd $3, %xmm0, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_16: ## %else20 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm5, %ymm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $8, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_18 +; AVX2-NEXT: ## %bb.17: ## %cond.store22 +; AVX2-NEXT: vmovd %xmm1, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_18: ## %else23 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $9, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_20 +; AVX2-NEXT: ## %bb.19: ## %cond.store25 +; AVX2-NEXT: vpextrd $1, %xmm1, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_20: ## %else26 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm5, %ymm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $10, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_22 +; AVX2-NEXT: ## %bb.21: ## %cond.store28 +; AVX2-NEXT: vpextrd $2, %xmm1, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_22: ## %else29 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $11, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_24 +; AVX2-NEXT: ## %bb.23: ## %cond.store31 +; AVX2-NEXT: vpextrd $3, %xmm1, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_24: ## %else32 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm5, %ymm0 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $12, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_26 +; AVX2-NEXT: ## %bb.25: ## %cond.store34 +; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm4 +; AVX2-NEXT: vmovd %xmm4, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_26: ## %else35 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $13, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_28 +; AVX2-NEXT: ## %bb.27: ## %cond.store37 +; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm0 +; AVX2-NEXT: vpextrd $1, %xmm0, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_28: ## %else38 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm5, %ymm0 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $14, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_30 +; AVX2-NEXT: ## %bb.29: ## %cond.store40 +; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm4 +; AVX2-NEXT: vpextrd $2, %xmm4, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_30: ## %else41 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $15, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_32 +; AVX2-NEXT: ## %bb.31: ## %cond.store43 +; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm0 +; AVX2-NEXT: vpextrd $3, %xmm0, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_32: ## %else44 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm6, %ymm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 +; AVX2-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 +; AVX2-NEXT: vpextrb $0, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_34 +; AVX2-NEXT: ## %bb.33: ## %cond.store46 +; AVX2-NEXT: vmovd %xmm2, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_34: ## %else47 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $1, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_36 +; AVX2-NEXT: ## %bb.35: ## %cond.store49 +; AVX2-NEXT: vpextrd $1, %xmm2, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_36: ## %else50 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm6, %ymm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 +; AVX2-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 +; AVX2-NEXT: vpextrb $2, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_38 +; AVX2-NEXT: ## %bb.37: ## %cond.store52 +; AVX2-NEXT: vpextrd $2, %xmm2, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_38: ## %else53 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $3, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_40 +; AVX2-NEXT: ## %bb.39: ## %cond.store55 +; AVX2-NEXT: vpextrd $3, %xmm2, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_40: ## %else56 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm6, %ymm0 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 +; AVX2-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 +; AVX2-NEXT: vpextrb $4, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_42 +; AVX2-NEXT: ## %bb.41: ## %cond.store58 +; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm1 +; AVX2-NEXT: vmovd %xmm1, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_42: ## %else59 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $5, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_44 +; AVX2-NEXT: ## %bb.43: ## %cond.store61 +; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm0 +; AVX2-NEXT: vpextrd $1, %xmm0, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_44: ## %else62 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm6, %ymm0 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 +; AVX2-NEXT: vpacksswb %xmm0, %xmm1, %xmm1 +; AVX2-NEXT: vpextrb $6, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_46 +; AVX2-NEXT: ## %bb.45: ## %cond.store64 +; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm1 +; AVX2-NEXT: vpextrd $2, %xmm1, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_46: ## %else65 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $7, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_48 +; AVX2-NEXT: ## %bb.47: ## %cond.store67 +; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm0 +; AVX2-NEXT: vpextrd $3, %xmm0, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_48: ## %else68 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm7, %ymm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 +; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 +; AVX2-NEXT: vpextrb $8, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_50 +; AVX2-NEXT: ## %bb.49: ## %cond.store70 +; AVX2-NEXT: vmovd %xmm3, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_50: ## %else71 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $9, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_52 +; AVX2-NEXT: ## %bb.51: ## %cond.store73 +; AVX2-NEXT: vpextrd $1, %xmm3, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_52: ## %else74 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm7, %ymm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 +; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 +; AVX2-NEXT: vpextrb $10, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_54 +; AVX2-NEXT: ## %bb.53: ## %cond.store76 +; AVX2-NEXT: vpextrd $2, %xmm3, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_54: ## %else77 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $11, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_56 +; AVX2-NEXT: ## %bb.55: ## %cond.store79 +; AVX2-NEXT: vpextrd $3, %xmm3, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_56: ## %else80 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm7, %ymm0 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 +; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 +; AVX2-NEXT: vpextrb $12, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_58 +; AVX2-NEXT: ## %bb.57: ## %cond.store82 +; AVX2-NEXT: vextracti128 $1, %ymm3, %xmm1 +; AVX2-NEXT: vmovd %xmm1, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_58: ## %else83 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $13, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_60 +; AVX2-NEXT: ## %bb.59: ## %cond.store85 +; AVX2-NEXT: vextracti128 $1, %ymm3, %xmm0 +; AVX2-NEXT: vpextrd $1, %xmm0, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_60: ## %else86 +; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpcmpeqd %ymm0, %ymm7, %ymm0 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm1 +; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm1 +; AVX2-NEXT: vpextrb $14, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_62 +; AVX2-NEXT: ## %bb.61: ## %cond.store88 +; AVX2-NEXT: vextracti128 $1, %ymm3, %xmm1 +; AVX2-NEXT: vpextrd $2, %xmm1, (%rdi) +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB6_62: ## %else89 +; AVX2-NEXT: vpackssdw %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: vpextrb $15, %xmm0, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB6_64 +; AVX2-NEXT: ## %bb.63: ## %cond.store91 +; AVX2-NEXT: vextracti128 $1, %ymm3, %xmm0 +; AVX2-NEXT: vpextrd $3, %xmm0, (%rdi) +; AVX2-NEXT: LBB6_64: ## %else92 +; AVX2-NEXT: vzeroupper +; AVX2-NEXT: retq +; +; AVX512-LABEL: compressstore_v32f32_v32i32: +; AVX512: ## %bb.0: +; AVX512-NEXT: vptestnmd %zmm3, %zmm3, %k1 +; AVX512-NEXT: vptestnmd %zmm2, %zmm2, %k2 +; AVX512-NEXT: kmovw %k2, %eax +; AVX512-NEXT: movl %eax, %ecx +; AVX512-NEXT: shrl %ecx +; AVX512-NEXT: andl $21845, %ecx ## imm = 0x5555 +; AVX512-NEXT: subl %ecx, %eax +; AVX512-NEXT: movl %eax, %ecx +; AVX512-NEXT: andl $858993459, %ecx ## imm = 0x33333333 +; AVX512-NEXT: shrl $2, %eax +; AVX512-NEXT: andl $858993459, %eax ## imm = 0x33333333 +; AVX512-NEXT: addl %ecx, %eax +; AVX512-NEXT: movl %eax, %ecx +; AVX512-NEXT: shrl $4, %ecx +; AVX512-NEXT: addl %eax, %ecx +; AVX512-NEXT: andl $252645135, %ecx ## imm = 0xF0F0F0F +; AVX512-NEXT: imull $16843009, %ecx, %eax ## imm = 0x1010101 +; AVX512-NEXT: shrl $24, %eax +; AVX512-NEXT: vcompressps %zmm1, (%rdi,%rax,4) {%k1} +; AVX512-NEXT: vcompressps %zmm0, (%rdi) {%k2} +; AVX512-NEXT: vzeroupper +; AVX512-NEXT: retq %mask = icmp eq <32 x i32> %trigger, zeroinitializer call void @llvm.masked.compressstore.v32f32(<32 x float> %V, float* %base, <32 x i1> %mask) ret void @@ -840,179 +2667,501 @@ define void @compressstore_v32f32_v32i32(float* %base, <32 x float> %V, <32 x i3 ; define void @compressstore_v2i64_v2i1(i64* %base, <2 x i64> %V, <2 x i1> %mask) { -; HSW-LABEL: compressstore_v2i64_v2i1: -; HSW: # %bb.0: -; HSW-NEXT: vpextrb $0, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB7_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vmovq %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB7_2: # %else -; HSW-NEXT: vpextrb $8, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB7_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vpextrq $1, %xmm0, (%rdi) -; HSW-NEXT: .LBB7_4: # %else2 -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v2i64_v2i1: -; SKX: # %bb.0: -; SKX-NEXT: vpsllq $63, %xmm1, %xmm1 -; SKX-NEXT: vpmovq2m %xmm1, %k1 -; SKX-NEXT: vpcompressq %xmm0, (%rdi) {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v2i64_v2i1: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; KNL-NEXT: vpsllq $63, %xmm1, %xmm1 -; KNL-NEXT: vptestmq %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftlw $14, %k0, %k0 -; KNL-NEXT: kshiftrw $14, %k0, %k1 -; KNL-NEXT: vpcompressq %zmm0, (%rdi) {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v2i64_v2i1: +; SSE2: ## %bb.0: +; SSE2-NEXT: movd %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB7_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movq %xmm0, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB7_2: ## %else +; SSE2-NEXT: pextrw $4, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB7_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1] +; SSE2-NEXT: movq %xmm0, (%rdi) +; SSE2-NEXT: LBB7_4: ## %else2 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v2i64_v2i1: +; SSE42: ## %bb.0: +; SSE42-NEXT: pextrb $0, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB7_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: movq %xmm0, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB7_2: ## %else +; SSE42-NEXT: pextrb $8, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB7_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: pextrq $1, %xmm0, (%rdi) +; SSE42-NEXT: LBB7_4: ## %else2 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: compressstore_v2i64_v2i1: +; AVX1OR2: ## %bb.0: +; AVX1OR2-NEXT: vpextrb $0, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB7_2 +; AVX1OR2-NEXT: ## %bb.1: ## %cond.store +; AVX1OR2-NEXT: vmovq %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $8, %rdi +; AVX1OR2-NEXT: LBB7_2: ## %else +; AVX1OR2-NEXT: vpextrb $8, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB7_4 +; AVX1OR2-NEXT: ## %bb.3: ## %cond.store1 +; AVX1OR2-NEXT: vpextrq $1, %xmm0, (%rdi) +; AVX1OR2-NEXT: LBB7_4: ## %else2 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v2i64_v2i1: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 +; AVX512F-NEXT: vpsllq $63, %xmm1, %xmm1 +; AVX512F-NEXT: vptestmq %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftlw $14, %k0, %k0 +; AVX512F-NEXT: kshiftrw $14, %k0, %k1 +; AVX512F-NEXT: vpcompressq %zmm0, (%rdi) {%k1} +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v2i64_v2i1: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vpsllq $63, %xmm1, %xmm1 +; AVX512VLBW-NEXT: vptestmq %xmm1, %xmm1, %k1 +; AVX512VLBW-NEXT: vpcompressq %xmm0, (%rdi) {%k1} +; AVX512VLBW-NEXT: retq call void @llvm.masked.compressstore.v2i64(<2 x i64> %V, i64* %base, <2 x i1> %mask) ret void } define void @compressstore_v4i64_v4i1(i64* %base, <4 x i64> %V, <4 x i1> %mask) { -; HSW-LABEL: compressstore_v4i64_v4i1: -; HSW: # %bb.0: -; HSW-NEXT: vpextrb $0, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vmovq %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB8_2: # %else -; HSW-NEXT: vpextrb $4, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vpextrq $1, %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB8_4: # %else2 -; HSW-NEXT: vpextrb $8, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_6 -; HSW-NEXT: # %bb.5: # %cond.store4 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm2 -; HSW-NEXT: vmovq %xmm2, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB8_6: # %else5 -; HSW-NEXT: vpextrb $12, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_8 -; HSW-NEXT: # %bb.7: # %cond.store7 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm0 -; HSW-NEXT: vpextrq $1, %xmm0, (%rdi) -; HSW-NEXT: .LBB8_8: # %else8 -; HSW-NEXT: vzeroupper -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v4i64_v4i1: -; SKX: # %bb.0: -; SKX-NEXT: vpslld $31, %xmm1, %xmm1 -; SKX-NEXT: vpmovd2m %xmm1, %k1 -; SKX-NEXT: vpcompressq %ymm0, (%rdi) {%k1} -; SKX-NEXT: vzeroupper -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v4i64_v4i1: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; KNL-NEXT: vpslld $31, %xmm1, %xmm1 -; KNL-NEXT: vptestmd %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftlw $12, %k0, %k0 -; KNL-NEXT: kshiftrw $12, %k0, %k1 -; KNL-NEXT: vpcompressq %zmm0, (%rdi) {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v4i64_v4i1: +; SSE2: ## %bb.0: +; SSE2-NEXT: movd %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB8_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movq %xmm0, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB8_2: ## %else +; SSE2-NEXT: pextrw $2, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB8_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1] +; SSE2-NEXT: movq %xmm0, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB8_4: ## %else2 +; SSE2-NEXT: pextrw $4, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB8_6 +; SSE2-NEXT: ## %bb.5: ## %cond.store4 +; SSE2-NEXT: movq %xmm1, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB8_6: ## %else5 +; SSE2-NEXT: pextrw $6, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB8_8 +; SSE2-NEXT: ## %bb.7: ## %cond.store7 +; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1] +; SSE2-NEXT: movq %xmm0, (%rdi) +; SSE2-NEXT: LBB8_8: ## %else8 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v4i64_v4i1: +; SSE42: ## %bb.0: +; SSE42-NEXT: pextrb $0, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB8_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: movq %xmm0, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB8_2: ## %else +; SSE42-NEXT: pextrb $4, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB8_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: pextrq $1, %xmm0, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB8_4: ## %else2 +; SSE42-NEXT: pextrb $8, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB8_6 +; SSE42-NEXT: ## %bb.5: ## %cond.store4 +; SSE42-NEXT: movq %xmm1, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB8_6: ## %else5 +; SSE42-NEXT: pextrb $12, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB8_8 +; SSE42-NEXT: ## %bb.7: ## %cond.store7 +; SSE42-NEXT: pextrq $1, %xmm1, (%rdi) +; SSE42-NEXT: LBB8_8: ## %else8 +; SSE42-NEXT: retq +; +; AVX1-LABEL: compressstore_v4i64_v4i1: +; AVX1: ## %bb.0: +; AVX1-NEXT: vpextrb $0, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_2 +; AVX1-NEXT: ## %bb.1: ## %cond.store +; AVX1-NEXT: vmovq %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB8_2: ## %else +; AVX1-NEXT: vpextrb $4, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_4 +; AVX1-NEXT: ## %bb.3: ## %cond.store1 +; AVX1-NEXT: vpextrq $1, %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB8_4: ## %else2 +; AVX1-NEXT: vpextrb $8, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_6 +; AVX1-NEXT: ## %bb.5: ## %cond.store4 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; AVX1-NEXT: vmovlps %xmm2, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB8_6: ## %else5 +; AVX1-NEXT: vpextrb $12, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_8 +; AVX1-NEXT: ## %bb.7: ## %cond.store7 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; AVX1-NEXT: vpextrq $1, %xmm0, (%rdi) +; AVX1-NEXT: LBB8_8: ## %else8 +; AVX1-NEXT: vzeroupper +; AVX1-NEXT: retq +; +; AVX2-LABEL: compressstore_v4i64_v4i1: +; AVX2: ## %bb.0: +; AVX2-NEXT: vpextrb $0, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_2 +; AVX2-NEXT: ## %bb.1: ## %cond.store +; AVX2-NEXT: vmovq %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB8_2: ## %else +; AVX2-NEXT: vpextrb $4, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_4 +; AVX2-NEXT: ## %bb.3: ## %cond.store1 +; AVX2-NEXT: vpextrq $1, %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB8_4: ## %else2 +; AVX2-NEXT: vpextrb $8, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_6 +; AVX2-NEXT: ## %bb.5: ## %cond.store4 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm2 +; AVX2-NEXT: vmovq %xmm2, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB8_6: ## %else5 +; AVX2-NEXT: vpextrb $12, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_8 +; AVX2-NEXT: ## %bb.7: ## %cond.store7 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0 +; AVX2-NEXT: vpextrq $1, %xmm0, (%rdi) +; AVX2-NEXT: LBB8_8: ## %else8 +; AVX2-NEXT: vzeroupper +; AVX2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v4i64_v4i1: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 +; AVX512F-NEXT: vpslld $31, %xmm1, %xmm1 +; AVX512F-NEXT: vptestmd %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftlw $12, %k0, %k0 +; AVX512F-NEXT: kshiftrw $12, %k0, %k1 +; AVX512F-NEXT: vpcompressq %zmm0, (%rdi) {%k1} +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v4i64_v4i1: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vpslld $31, %xmm1, %xmm1 +; AVX512VLBW-NEXT: vptestmd %xmm1, %xmm1, %k1 +; AVX512VLBW-NEXT: vpcompressq %ymm0, (%rdi) {%k1} +; AVX512VLBW-NEXT: vzeroupper +; AVX512VLBW-NEXT: retq call void @llvm.masked.compressstore.v4i64(<4 x i64> %V, i64* %base, <4 x i1> %mask) ret void } define void @compressstore_v8i64_v8i1(i64* %base, <8 x i64> %V, <8 x i1> %mask) { -; HSW-LABEL: compressstore_v8i64_v8i1: -; HSW: # %bb.0: -; HSW-NEXT: vpextrb $0, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB9_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vmovq %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB9_2: # %else -; HSW-NEXT: vpextrb $2, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB9_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vpextrq $1, %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB9_4: # %else2 -; HSW-NEXT: vpextrb $4, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB9_6 -; HSW-NEXT: # %bb.5: # %cond.store4 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm3 -; HSW-NEXT: vmovq %xmm3, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB9_6: # %else5 -; HSW-NEXT: vpextrb $6, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB9_8 -; HSW-NEXT: # %bb.7: # %cond.store7 -; HSW-NEXT: vextracti128 $1, %ymm0, %xmm0 -; HSW-NEXT: vpextrq $1, %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB9_8: # %else8 -; HSW-NEXT: vpextrb $8, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB9_10 -; HSW-NEXT: # %bb.9: # %cond.store10 -; HSW-NEXT: vmovq %xmm1, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB9_10: # %else11 -; HSW-NEXT: vpextrb $10, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB9_12 -; HSW-NEXT: # %bb.11: # %cond.store13 -; HSW-NEXT: vpextrq $1, %xmm1, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB9_12: # %else14 -; HSW-NEXT: vpextrb $12, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB9_14 -; HSW-NEXT: # %bb.13: # %cond.store16 -; HSW-NEXT: vextracti128 $1, %ymm1, %xmm0 -; HSW-NEXT: vmovq %xmm0, (%rdi) -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB9_14: # %else17 -; HSW-NEXT: vpextrb $14, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB9_16 -; HSW-NEXT: # %bb.15: # %cond.store19 -; HSW-NEXT: vextracti128 $1, %ymm1, %xmm0 -; HSW-NEXT: vpextrq $1, %xmm0, (%rdi) -; HSW-NEXT: .LBB9_16: # %else20 -; HSW-NEXT: vzeroupper -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v8i64_v8i1: -; SKX: # %bb.0: -; SKX-NEXT: vpsllw $15, %xmm1, %xmm1 -; SKX-NEXT: vpmovw2m %xmm1, %k1 -; SKX-NEXT: vpcompressq %zmm0, (%rdi) {%k1} -; SKX-NEXT: vzeroupper -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v8i64_v8i1: -; KNL: # %bb.0: -; KNL-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero -; KNL-NEXT: vpsllq $63, %zmm1, %zmm1 -; KNL-NEXT: vptestmq %zmm1, %zmm1, %k1 -; KNL-NEXT: vpcompressq %zmm0, (%rdi) {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v8i64_v8i1: +; SSE2: ## %bb.0: +; SSE2-NEXT: movd %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB9_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movq %xmm0, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB9_2: ## %else +; SSE2-NEXT: shrl $16, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB9_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1] +; SSE2-NEXT: movq %xmm0, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB9_4: ## %else2 +; SSE2-NEXT: pextrw $2, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB9_6 +; SSE2-NEXT: ## %bb.5: ## %cond.store4 +; SSE2-NEXT: movq %xmm1, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB9_6: ## %else5 +; SSE2-NEXT: pextrw $3, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB9_8 +; SSE2-NEXT: ## %bb.7: ## %cond.store7 +; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1] +; SSE2-NEXT: movq %xmm0, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB9_8: ## %else8 +; SSE2-NEXT: pextrw $4, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB9_10 +; SSE2-NEXT: ## %bb.9: ## %cond.store10 +; SSE2-NEXT: movq %xmm2, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB9_10: ## %else11 +; SSE2-NEXT: pextrw $5, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB9_12 +; SSE2-NEXT: ## %bb.11: ## %cond.store13 +; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1] +; SSE2-NEXT: movq %xmm0, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB9_12: ## %else14 +; SSE2-NEXT: pextrw $6, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB9_14 +; SSE2-NEXT: ## %bb.13: ## %cond.store16 +; SSE2-NEXT: movq %xmm3, (%rdi) +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB9_14: ## %else17 +; SSE2-NEXT: pextrw $7, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB9_16 +; SSE2-NEXT: ## %bb.15: ## %cond.store19 +; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[2,3,0,1] +; SSE2-NEXT: movq %xmm0, (%rdi) +; SSE2-NEXT: LBB9_16: ## %else20 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v8i64_v8i1: +; SSE42: ## %bb.0: +; SSE42-NEXT: pextrb $0, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB9_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: movq %xmm0, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB9_2: ## %else +; SSE42-NEXT: pextrb $2, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB9_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: pextrq $1, %xmm0, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB9_4: ## %else2 +; SSE42-NEXT: pextrb $4, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB9_6 +; SSE42-NEXT: ## %bb.5: ## %cond.store4 +; SSE42-NEXT: movq %xmm1, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB9_6: ## %else5 +; SSE42-NEXT: pextrb $6, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB9_8 +; SSE42-NEXT: ## %bb.7: ## %cond.store7 +; SSE42-NEXT: pextrq $1, %xmm1, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB9_8: ## %else8 +; SSE42-NEXT: pextrb $8, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB9_10 +; SSE42-NEXT: ## %bb.9: ## %cond.store10 +; SSE42-NEXT: movq %xmm2, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB9_10: ## %else11 +; SSE42-NEXT: pextrb $10, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB9_12 +; SSE42-NEXT: ## %bb.11: ## %cond.store13 +; SSE42-NEXT: pextrq $1, %xmm2, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB9_12: ## %else14 +; SSE42-NEXT: pextrb $12, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB9_14 +; SSE42-NEXT: ## %bb.13: ## %cond.store16 +; SSE42-NEXT: movq %xmm3, (%rdi) +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB9_14: ## %else17 +; SSE42-NEXT: pextrb $14, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB9_16 +; SSE42-NEXT: ## %bb.15: ## %cond.store19 +; SSE42-NEXT: pextrq $1, %xmm3, (%rdi) +; SSE42-NEXT: LBB9_16: ## %else20 +; SSE42-NEXT: retq +; +; AVX1-LABEL: compressstore_v8i64_v8i1: +; AVX1: ## %bb.0: +; AVX1-NEXT: vpextrb $0, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB9_2 +; AVX1-NEXT: ## %bb.1: ## %cond.store +; AVX1-NEXT: vmovq %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB9_2: ## %else +; AVX1-NEXT: vpextrb $2, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB9_4 +; AVX1-NEXT: ## %bb.3: ## %cond.store1 +; AVX1-NEXT: vpextrq $1, %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB9_4: ## %else2 +; AVX1-NEXT: vpextrb $4, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB9_6 +; AVX1-NEXT: ## %bb.5: ## %cond.store4 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 +; AVX1-NEXT: vmovlps %xmm3, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB9_6: ## %else5 +; AVX1-NEXT: vpextrb $6, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB9_8 +; AVX1-NEXT: ## %bb.7: ## %cond.store7 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; AVX1-NEXT: vpextrq $1, %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB9_8: ## %else8 +; AVX1-NEXT: vpextrb $8, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB9_10 +; AVX1-NEXT: ## %bb.9: ## %cond.store10 +; AVX1-NEXT: vmovq %xmm1, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB9_10: ## %else11 +; AVX1-NEXT: vpextrb $10, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB9_12 +; AVX1-NEXT: ## %bb.11: ## %cond.store13 +; AVX1-NEXT: vpextrq $1, %xmm1, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB9_12: ## %else14 +; AVX1-NEXT: vpextrb $12, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB9_14 +; AVX1-NEXT: ## %bb.13: ## %cond.store16 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX1-NEXT: vmovq %xmm0, (%rdi) +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB9_14: ## %else17 +; AVX1-NEXT: vpextrb $14, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB9_16 +; AVX1-NEXT: ## %bb.15: ## %cond.store19 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm0 +; AVX1-NEXT: vpextrq $1, %xmm0, (%rdi) +; AVX1-NEXT: LBB9_16: ## %else20 +; AVX1-NEXT: vzeroupper +; AVX1-NEXT: retq +; +; AVX2-LABEL: compressstore_v8i64_v8i1: +; AVX2: ## %bb.0: +; AVX2-NEXT: vpextrb $0, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB9_2 +; AVX2-NEXT: ## %bb.1: ## %cond.store +; AVX2-NEXT: vmovq %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB9_2: ## %else +; AVX2-NEXT: vpextrb $2, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB9_4 +; AVX2-NEXT: ## %bb.3: ## %cond.store1 +; AVX2-NEXT: vpextrq $1, %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB9_4: ## %else2 +; AVX2-NEXT: vpextrb $4, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB9_6 +; AVX2-NEXT: ## %bb.5: ## %cond.store4 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm3 +; AVX2-NEXT: vmovq %xmm3, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB9_6: ## %else5 +; AVX2-NEXT: vpextrb $6, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB9_8 +; AVX2-NEXT: ## %bb.7: ## %cond.store7 +; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0 +; AVX2-NEXT: vpextrq $1, %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB9_8: ## %else8 +; AVX2-NEXT: vpextrb $8, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB9_10 +; AVX2-NEXT: ## %bb.9: ## %cond.store10 +; AVX2-NEXT: vmovq %xmm1, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB9_10: ## %else11 +; AVX2-NEXT: vpextrb $10, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB9_12 +; AVX2-NEXT: ## %bb.11: ## %cond.store13 +; AVX2-NEXT: vpextrq $1, %xmm1, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB9_12: ## %else14 +; AVX2-NEXT: vpextrb $12, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB9_14 +; AVX2-NEXT: ## %bb.13: ## %cond.store16 +; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm0 +; AVX2-NEXT: vmovq %xmm0, (%rdi) +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB9_14: ## %else17 +; AVX2-NEXT: vpextrb $14, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB9_16 +; AVX2-NEXT: ## %bb.15: ## %cond.store19 +; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm0 +; AVX2-NEXT: vpextrq $1, %xmm0, (%rdi) +; AVX2-NEXT: LBB9_16: ## %else20 +; AVX2-NEXT: vzeroupper +; AVX2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v8i64_v8i1: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero +; AVX512F-NEXT: vpsllq $63, %zmm1, %zmm1 +; AVX512F-NEXT: vptestmq %zmm1, %zmm1, %k1 +; AVX512F-NEXT: vpcompressq %zmm0, (%rdi) {%k1} +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v8i64_v8i1: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vpsllw $15, %xmm1, %xmm1 +; AVX512VLBW-NEXT: vpmovw2m %xmm1, %k1 +; AVX512VLBW-NEXT: vpcompressq %zmm0, (%rdi) {%k1} +; AVX512VLBW-NEXT: vzeroupper +; AVX512VLBW-NEXT: retq call void @llvm.masked.compressstore.v8i64(<8 x i64> %V, i64* %base, <8 x i1> %mask) ret void } @@ -1022,56 +3171,130 @@ define void @compressstore_v8i64_v8i1(i64* %base, <8 x i64> %V, <8 x i1> %mask) ; define void @compressstore_v4i32_v4i32(i32* %base, <4 x i32> %V, <4 x i32> %trigger) { -; HSW-LABEL: compressstore_v4i32_v4i32: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $0, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB10_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vmovss %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB10_2: # %else -; HSW-NEXT: vpextrb $4, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB10_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vextractps $1, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB10_4: # %else2 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $8, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB10_6 -; HSW-NEXT: # %bb.5: # %cond.store4 -; HSW-NEXT: vextractps $2, %xmm0, (%rdi) -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB10_6: # %else5 -; HSW-NEXT: vpextrb $12, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB10_8 -; HSW-NEXT: # %bb.7: # %cond.store7 -; HSW-NEXT: vextractps $3, %xmm0, (%rdi) -; HSW-NEXT: .LBB10_8: # %else8 -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v4i32_v4i32: -; SKX: # %bb.0: -; SKX-NEXT: vptestnmd %xmm1, %xmm1, %k1 -; SKX-NEXT: vpcompressd %xmm0, (%rdi) {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v4i32_v4i32: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; KNL-NEXT: vptestnmd %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftlw $12, %k0, %k0 -; KNL-NEXT: kshiftrw $12, %k0, %k1 -; KNL-NEXT: vpcompressd %zmm0, (%rdi) {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v4i32_v4i32: +; SSE2: ## %bb.0: +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqd %xmm1, %xmm2 +; SSE2-NEXT: movd %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB10_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movd %xmm0, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB10_2: ## %else +; SSE2-NEXT: pextrw $2, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB10_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,2,3] +; SSE2-NEXT: movd %xmm2, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB10_4: ## %else2 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqd %xmm2, %xmm1 +; SSE2-NEXT: pextrw $4, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB10_6 +; SSE2-NEXT: ## %bb.5: ## %cond.store4 +; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1] +; SSE2-NEXT: movd %xmm2, (%rdi) +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB10_6: ## %else5 +; SSE2-NEXT: pextrw $6, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB10_8 +; SSE2-NEXT: ## %bb.7: ## %cond.store7 +; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,1,2,3] +; SSE2-NEXT: movd %xmm0, (%rdi) +; SSE2-NEXT: LBB10_8: ## %else8 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v4i32_v4i32: +; SSE42: ## %bb.0: +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqd %xmm1, %xmm2 +; SSE42-NEXT: pextrb $0, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB10_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: movss %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB10_2: ## %else +; SSE42-NEXT: pextrb $4, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB10_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: extractps $1, %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB10_4: ## %else2 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqd %xmm2, %xmm1 +; SSE42-NEXT: pextrb $8, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB10_6 +; SSE42-NEXT: ## %bb.5: ## %cond.store4 +; SSE42-NEXT: extractps $2, %xmm0, (%rdi) +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB10_6: ## %else5 +; SSE42-NEXT: pextrb $12, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB10_8 +; SSE42-NEXT: ## %bb.7: ## %cond.store7 +; SSE42-NEXT: extractps $3, %xmm0, (%rdi) +; SSE42-NEXT: LBB10_8: ## %else8 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: compressstore_v4i32_v4i32: +; AVX1OR2: ## %bb.0: +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $0, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB10_2 +; AVX1OR2-NEXT: ## %bb.1: ## %cond.store +; AVX1OR2-NEXT: vmovss %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB10_2: ## %else +; AVX1OR2-NEXT: vpextrb $4, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB10_4 +; AVX1OR2-NEXT: ## %bb.3: ## %cond.store1 +; AVX1OR2-NEXT: vextractps $1, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB10_4: ## %else2 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm1 +; AVX1OR2-NEXT: vpextrb $8, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB10_6 +; AVX1OR2-NEXT: ## %bb.5: ## %cond.store4 +; AVX1OR2-NEXT: vextractps $2, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB10_6: ## %else5 +; AVX1OR2-NEXT: vpextrb $12, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB10_8 +; AVX1OR2-NEXT: ## %bb.7: ## %cond.store7 +; AVX1OR2-NEXT: vextractps $3, %xmm0, (%rdi) +; AVX1OR2-NEXT: LBB10_8: ## %else8 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v4i32_v4i32: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 +; AVX512F-NEXT: vptestnmd %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftlw $12, %k0, %k0 +; AVX512F-NEXT: kshiftrw $12, %k0, %k1 +; AVX512F-NEXT: vpcompressd %zmm0, (%rdi) {%k1} +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v4i32_v4i32: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vptestnmd %xmm1, %xmm1, %k1 +; AVX512VLBW-NEXT: vpcompressd %xmm0, (%rdi) {%k1} +; AVX512VLBW-NEXT: retq %mask = icmp eq <4 x i32> %trigger, zeroinitializer call void @llvm.masked.compressstore.v4i32(<4 x i32> %V, i32* %base, <4 x i1> %mask) ret void @@ -1082,221 +3305,364 @@ define void @compressstore_v4i32_v4i32(i32* %base, <4 x i32> %V, <4 x i32> %trig ; define void @compressstore_v8i16_v8i16(i16* %base, <8 x i16> %V, <8 x i16> %trigger) { -; HSW-LABEL: compressstore_v8i16_v8i16: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $0, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vpextrw $0, %xmm0, (%rdi) -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_2: # %else -; HSW-NEXT: vpextrb $2, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vpextrw $1, %xmm0, (%rdi) -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_4: # %else2 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $4, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_6 -; HSW-NEXT: # %bb.5: # %cond.store4 -; HSW-NEXT: vpextrw $2, %xmm0, (%rdi) -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_6: # %else5 -; HSW-NEXT: vpextrb $6, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_8 -; HSW-NEXT: # %bb.7: # %cond.store7 -; HSW-NEXT: vpextrw $3, %xmm0, (%rdi) -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_8: # %else8 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $8, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_10 -; HSW-NEXT: # %bb.9: # %cond.store10 -; HSW-NEXT: vpextrw $4, %xmm0, (%rdi) -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_10: # %else11 -; HSW-NEXT: vpextrb $10, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_12 -; HSW-NEXT: # %bb.11: # %cond.store13 -; HSW-NEXT: vpextrw $5, %xmm0, (%rdi) -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_12: # %else14 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $12, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_14 -; HSW-NEXT: # %bb.13: # %cond.store16 -; HSW-NEXT: vpextrw $6, %xmm0, (%rdi) -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_14: # %else17 -; HSW-NEXT: vpextrb $14, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_16 -; HSW-NEXT: # %bb.15: # %cond.store19 -; HSW-NEXT: vpextrw $7, %xmm0, (%rdi) -; HSW-NEXT: .LBB11_16: # %else20 -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v8i16_v8i16: -; SKX: # %bb.0: -; SKX-NEXT: vptestnmw %xmm1, %xmm1, %k0 -; SKX-NEXT: kmovd %k0, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_2 -; SKX-NEXT: # %bb.1: # %cond.store -; SKX-NEXT: vpextrw $0, %xmm0, (%rdi) -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_2: # %else -; SKX-NEXT: kshiftrb $1, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_4 -; SKX-NEXT: # %bb.3: # %cond.store1 -; SKX-NEXT: vpextrw $1, %xmm0, (%rdi) -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_4: # %else2 -; SKX-NEXT: kshiftrb $2, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_6 -; SKX-NEXT: # %bb.5: # %cond.store4 -; SKX-NEXT: vpextrw $2, %xmm0, (%rdi) -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_6: # %else5 -; SKX-NEXT: kshiftrb $3, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_8 -; SKX-NEXT: # %bb.7: # %cond.store7 -; SKX-NEXT: vpextrw $3, %xmm0, (%rdi) -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_8: # %else8 -; SKX-NEXT: kshiftrb $4, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_10 -; SKX-NEXT: # %bb.9: # %cond.store10 -; SKX-NEXT: vpextrw $4, %xmm0, (%rdi) -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_10: # %else11 -; SKX-NEXT: kshiftrb $5, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_12 -; SKX-NEXT: # %bb.11: # %cond.store13 -; SKX-NEXT: vpextrw $5, %xmm0, (%rdi) -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_12: # %else14 -; SKX-NEXT: kshiftrb $6, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_14 -; SKX-NEXT: # %bb.13: # %cond.store16 -; SKX-NEXT: vpextrw $6, %xmm0, (%rdi) -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_14: # %else17 -; SKX-NEXT: kshiftrb $7, %k0, %k0 -; SKX-NEXT: kmovd %k0, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_16 -; SKX-NEXT: # %bb.15: # %cond.store19 -; SKX-NEXT: vpextrw $7, %xmm0, (%rdi) -; SKX-NEXT: .LBB11_16: # %else20 -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v8i16_v8i16: -; KNL: # %bb.0: -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxwq %xmm2, %zmm2 -; KNL-NEXT: vptestmq %zmm2, %zmm2, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_2 -; KNL-NEXT: # %bb.1: # %cond.store -; KNL-NEXT: vpextrw $0, %xmm0, (%rdi) -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_2: # %else -; KNL-NEXT: kshiftrw $1, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_4 -; KNL-NEXT: # %bb.3: # %cond.store1 -; KNL-NEXT: vpextrw $1, %xmm0, (%rdi) -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_4: # %else2 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxwq %xmm2, %zmm2 -; KNL-NEXT: vptestmq %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $2, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_6 -; KNL-NEXT: # %bb.5: # %cond.store4 -; KNL-NEXT: vpextrw $2, %xmm0, (%rdi) -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_6: # %else5 -; KNL-NEXT: kshiftrw $3, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_8 -; KNL-NEXT: # %bb.7: # %cond.store7 -; KNL-NEXT: vpextrw $3, %xmm0, (%rdi) -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_8: # %else8 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxwq %xmm2, %zmm2 -; KNL-NEXT: vptestmq %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $4, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_10 -; KNL-NEXT: # %bb.9: # %cond.store10 -; KNL-NEXT: vpextrw $4, %xmm0, (%rdi) -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_10: # %else11 -; KNL-NEXT: kshiftrw $5, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_12 -; KNL-NEXT: # %bb.11: # %cond.store13 -; KNL-NEXT: vpextrw $5, %xmm0, (%rdi) -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_12: # %else14 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm1 -; KNL-NEXT: vpmovsxwq %xmm1, %zmm1 -; KNL-NEXT: vptestmq %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftrw $6, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_14 -; KNL-NEXT: # %bb.13: # %cond.store16 -; KNL-NEXT: vpextrw $6, %xmm0, (%rdi) -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_14: # %else17 -; KNL-NEXT: kshiftrw $7, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_16 -; KNL-NEXT: # %bb.15: # %cond.store19 -; KNL-NEXT: vpextrw $7, %xmm0, (%rdi) -; KNL-NEXT: .LBB11_16: # %else20 -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v8i16_v8i16: +; SSE2: ## %bb.0: +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE2-NEXT: movd %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movd %xmm0, %ecx +; SSE2-NEXT: movw %cx, (%rdi) +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_2: ## %else +; SSE2-NEXT: shrl $16, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: pextrw $1, %xmm0, %eax +; SSE2-NEXT: movw %ax, (%rdi) +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_4: ## %else2 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE2-NEXT: pextrw $2, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_6 +; SSE2-NEXT: ## %bb.5: ## %cond.store4 +; SSE2-NEXT: pextrw $2, %xmm0, %eax +; SSE2-NEXT: movw %ax, (%rdi) +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_6: ## %else5 +; SSE2-NEXT: pextrw $3, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_8 +; SSE2-NEXT: ## %bb.7: ## %cond.store7 +; SSE2-NEXT: pextrw $3, %xmm0, %eax +; SSE2-NEXT: movw %ax, (%rdi) +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_8: ## %else8 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE2-NEXT: pextrw $4, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_10 +; SSE2-NEXT: ## %bb.9: ## %cond.store10 +; SSE2-NEXT: pextrw $4, %xmm0, %eax +; SSE2-NEXT: movw %ax, (%rdi) +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_10: ## %else11 +; SSE2-NEXT: pextrw $5, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_12 +; SSE2-NEXT: ## %bb.11: ## %cond.store13 +; SSE2-NEXT: pextrw $5, %xmm0, %eax +; SSE2-NEXT: movw %ax, (%rdi) +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_12: ## %else14 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqw %xmm2, %xmm1 +; SSE2-NEXT: pextrw $6, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_14 +; SSE2-NEXT: ## %bb.13: ## %cond.store16 +; SSE2-NEXT: pextrw $6, %xmm0, %eax +; SSE2-NEXT: movw %ax, (%rdi) +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_14: ## %else17 +; SSE2-NEXT: pextrw $7, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_16 +; SSE2-NEXT: ## %bb.15: ## %cond.store19 +; SSE2-NEXT: pextrw $7, %xmm0, %eax +; SSE2-NEXT: movw %ax, (%rdi) +; SSE2-NEXT: LBB11_16: ## %else20 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v8i16_v8i16: +; SSE42: ## %bb.0: +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE42-NEXT: pextrb $0, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: pextrw $0, %xmm0, (%rdi) +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_2: ## %else +; SSE42-NEXT: pextrb $2, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: pextrw $1, %xmm0, (%rdi) +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_4: ## %else2 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE42-NEXT: pextrb $4, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_6 +; SSE42-NEXT: ## %bb.5: ## %cond.store4 +; SSE42-NEXT: pextrw $2, %xmm0, (%rdi) +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_6: ## %else5 +; SSE42-NEXT: pextrb $6, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_8 +; SSE42-NEXT: ## %bb.7: ## %cond.store7 +; SSE42-NEXT: pextrw $3, %xmm0, (%rdi) +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_8: ## %else8 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE42-NEXT: pextrb $8, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_10 +; SSE42-NEXT: ## %bb.9: ## %cond.store10 +; SSE42-NEXT: pextrw $4, %xmm0, (%rdi) +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_10: ## %else11 +; SSE42-NEXT: pextrb $10, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_12 +; SSE42-NEXT: ## %bb.11: ## %cond.store13 +; SSE42-NEXT: pextrw $5, %xmm0, (%rdi) +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_12: ## %else14 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqw %xmm2, %xmm1 +; SSE42-NEXT: pextrb $12, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_14 +; SSE42-NEXT: ## %bb.13: ## %cond.store16 +; SSE42-NEXT: pextrw $6, %xmm0, (%rdi) +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_14: ## %else17 +; SSE42-NEXT: pextrb $14, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_16 +; SSE42-NEXT: ## %bb.15: ## %cond.store19 +; SSE42-NEXT: pextrw $7, %xmm0, (%rdi) +; SSE42-NEXT: LBB11_16: ## %else20 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: compressstore_v8i16_v8i16: +; AVX1OR2: ## %bb.0: +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $0, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_2 +; AVX1OR2-NEXT: ## %bb.1: ## %cond.store +; AVX1OR2-NEXT: vpextrw $0, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_2: ## %else +; AVX1OR2-NEXT: vpextrb $2, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_4 +; AVX1OR2-NEXT: ## %bb.3: ## %cond.store1 +; AVX1OR2-NEXT: vpextrw $1, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_4: ## %else2 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $4, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_6 +; AVX1OR2-NEXT: ## %bb.5: ## %cond.store4 +; AVX1OR2-NEXT: vpextrw $2, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_6: ## %else5 +; AVX1OR2-NEXT: vpextrb $6, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_8 +; AVX1OR2-NEXT: ## %bb.7: ## %cond.store7 +; AVX1OR2-NEXT: vpextrw $3, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_8: ## %else8 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $8, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_10 +; AVX1OR2-NEXT: ## %bb.9: ## %cond.store10 +; AVX1OR2-NEXT: vpextrw $4, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_10: ## %else11 +; AVX1OR2-NEXT: vpextrb $10, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_12 +; AVX1OR2-NEXT: ## %bb.11: ## %cond.store13 +; AVX1OR2-NEXT: vpextrw $5, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_12: ## %else14 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm1 +; AVX1OR2-NEXT: vpextrb $12, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_14 +; AVX1OR2-NEXT: ## %bb.13: ## %cond.store16 +; AVX1OR2-NEXT: vpextrw $6, %xmm0, (%rdi) +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_14: ## %else17 +; AVX1OR2-NEXT: vpextrb $14, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_16 +; AVX1OR2-NEXT: ## %bb.15: ## %cond.store19 +; AVX1OR2-NEXT: vpextrw $7, %xmm0, (%rdi) +; AVX1OR2-NEXT: LBB11_16: ## %else20 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v8i16_v8i16: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxwq %xmm2, %zmm2 +; AVX512F-NEXT: vptestmq %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_2 +; AVX512F-NEXT: ## %bb.1: ## %cond.store +; AVX512F-NEXT: vpextrw $0, %xmm0, (%rdi) +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_2: ## %else +; AVX512F-NEXT: kshiftrw $1, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_4 +; AVX512F-NEXT: ## %bb.3: ## %cond.store1 +; AVX512F-NEXT: vpextrw $1, %xmm0, (%rdi) +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_4: ## %else2 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxwq %xmm2, %zmm2 +; AVX512F-NEXT: vptestmq %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $2, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_6 +; AVX512F-NEXT: ## %bb.5: ## %cond.store4 +; AVX512F-NEXT: vpextrw $2, %xmm0, (%rdi) +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_6: ## %else5 +; AVX512F-NEXT: kshiftrw $3, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_8 +; AVX512F-NEXT: ## %bb.7: ## %cond.store7 +; AVX512F-NEXT: vpextrw $3, %xmm0, (%rdi) +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_8: ## %else8 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxwq %xmm2, %zmm2 +; AVX512F-NEXT: vptestmq %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $4, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_10 +; AVX512F-NEXT: ## %bb.9: ## %cond.store10 +; AVX512F-NEXT: vpextrw $4, %xmm0, (%rdi) +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_10: ## %else11 +; AVX512F-NEXT: kshiftrw $5, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_12 +; AVX512F-NEXT: ## %bb.11: ## %cond.store13 +; AVX512F-NEXT: vpextrw $5, %xmm0, (%rdi) +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_12: ## %else14 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm1 +; AVX512F-NEXT: vpmovsxwq %xmm1, %zmm1 +; AVX512F-NEXT: vptestmq %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftrw $6, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_14 +; AVX512F-NEXT: ## %bb.13: ## %cond.store16 +; AVX512F-NEXT: vpextrw $6, %xmm0, (%rdi) +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_14: ## %else17 +; AVX512F-NEXT: kshiftrw $7, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_16 +; AVX512F-NEXT: ## %bb.15: ## %cond.store19 +; AVX512F-NEXT: vpextrw $7, %xmm0, (%rdi) +; AVX512F-NEXT: LBB11_16: ## %else20 +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v8i16_v8i16: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vptestnmw %xmm1, %xmm1, %k0 +; AVX512VLBW-NEXT: kmovd %k0, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_2 +; AVX512VLBW-NEXT: ## %bb.1: ## %cond.store +; AVX512VLBW-NEXT: vpextrw $0, %xmm0, (%rdi) +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_2: ## %else +; AVX512VLBW-NEXT: kshiftrw $1, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_4 +; AVX512VLBW-NEXT: ## %bb.3: ## %cond.store1 +; AVX512VLBW-NEXT: vpextrw $1, %xmm0, (%rdi) +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_4: ## %else2 +; AVX512VLBW-NEXT: kshiftrw $2, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_6 +; AVX512VLBW-NEXT: ## %bb.5: ## %cond.store4 +; AVX512VLBW-NEXT: vpextrw $2, %xmm0, (%rdi) +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_6: ## %else5 +; AVX512VLBW-NEXT: kshiftrw $3, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_8 +; AVX512VLBW-NEXT: ## %bb.7: ## %cond.store7 +; AVX512VLBW-NEXT: vpextrw $3, %xmm0, (%rdi) +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_8: ## %else8 +; AVX512VLBW-NEXT: kshiftrw $4, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_10 +; AVX512VLBW-NEXT: ## %bb.9: ## %cond.store10 +; AVX512VLBW-NEXT: vpextrw $4, %xmm0, (%rdi) +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_10: ## %else11 +; AVX512VLBW-NEXT: kshiftrw $5, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_12 +; AVX512VLBW-NEXT: ## %bb.11: ## %cond.store13 +; AVX512VLBW-NEXT: vpextrw $5, %xmm0, (%rdi) +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_12: ## %else14 +; AVX512VLBW-NEXT: kshiftrw $6, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_14 +; AVX512VLBW-NEXT: ## %bb.13: ## %cond.store16 +; AVX512VLBW-NEXT: vpextrw $6, %xmm0, (%rdi) +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_14: ## %else17 +; AVX512VLBW-NEXT: kshiftrw $7, %k0, %k0 +; AVX512VLBW-NEXT: kmovd %k0, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_16 +; AVX512VLBW-NEXT: ## %bb.15: ## %cond.store19 +; AVX512VLBW-NEXT: vpextrw $7, %xmm0, (%rdi) +; AVX512VLBW-NEXT: LBB11_16: ## %else20 +; AVX512VLBW-NEXT: retq %mask = icmp eq <8 x i16> %trigger, zeroinitializer call void @llvm.masked.compressstore.v8i16(<8 x i16> %V, i16* %base, <8 x i1> %mask) ret void @@ -1307,429 +3673,712 @@ define void @compressstore_v8i16_v8i16(i16* %base, <8 x i16> %V, <8 x i16> %trig ; define void @compressstore_v16i8_v16i8(i8* %base, <16 x i8> %V, <16 x i8> %trigger) { -; HSW-LABEL: compressstore_v16i8_v16i8: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $0, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_2 -; HSW-NEXT: # %bb.1: # %cond.store -; HSW-NEXT: vpextrb $0, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_2: # %else -; HSW-NEXT: vpextrb $1, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_4 -; HSW-NEXT: # %bb.3: # %cond.store1 -; HSW-NEXT: vpextrb $1, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_4: # %else2 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $2, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_6 -; HSW-NEXT: # %bb.5: # %cond.store4 -; HSW-NEXT: vpextrb $2, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_6: # %else5 -; HSW-NEXT: vpextrb $3, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_8 -; HSW-NEXT: # %bb.7: # %cond.store7 -; HSW-NEXT: vpextrb $3, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_8: # %else8 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $4, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_10 -; HSW-NEXT: # %bb.9: # %cond.store10 -; HSW-NEXT: vpextrb $4, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_10: # %else11 -; HSW-NEXT: vpextrb $5, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_12 -; HSW-NEXT: # %bb.11: # %cond.store13 -; HSW-NEXT: vpextrb $5, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_12: # %else14 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $6, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_14 -; HSW-NEXT: # %bb.13: # %cond.store16 -; HSW-NEXT: vpextrb $6, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_14: # %else17 -; HSW-NEXT: vpextrb $7, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_16 -; HSW-NEXT: # %bb.15: # %cond.store19 -; HSW-NEXT: vpextrb $7, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_16: # %else20 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $8, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_18 -; HSW-NEXT: # %bb.17: # %cond.store22 -; HSW-NEXT: vpextrb $8, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_18: # %else23 -; HSW-NEXT: vpextrb $9, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_20 -; HSW-NEXT: # %bb.19: # %cond.store25 -; HSW-NEXT: vpextrb $9, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_20: # %else26 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $10, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_22 -; HSW-NEXT: # %bb.21: # %cond.store28 -; HSW-NEXT: vpextrb $10, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_22: # %else29 -; HSW-NEXT: vpextrb $11, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_24 -; HSW-NEXT: # %bb.23: # %cond.store31 -; HSW-NEXT: vpextrb $11, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_24: # %else32 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $12, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_26 -; HSW-NEXT: # %bb.25: # %cond.store34 -; HSW-NEXT: vpextrb $12, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_26: # %else35 -; HSW-NEXT: vpextrb $13, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_28 -; HSW-NEXT: # %bb.27: # %cond.store37 -; HSW-NEXT: vpextrb $13, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_28: # %else38 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $14, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_30 -; HSW-NEXT: # %bb.29: # %cond.store40 -; HSW-NEXT: vpextrb $14, %xmm0, (%rdi) -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_30: # %else41 -; HSW-NEXT: vpextrb $15, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_32 -; HSW-NEXT: # %bb.31: # %cond.store43 -; HSW-NEXT: vpextrb $15, %xmm0, (%rdi) -; HSW-NEXT: .LBB12_32: # %else44 -; HSW-NEXT: retq -; -; SKX-LABEL: compressstore_v16i8_v16i8: -; SKX: # %bb.0: -; SKX-NEXT: vptestnmb %xmm1, %xmm1, %k0 -; SKX-NEXT: kmovd %k0, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_2 -; SKX-NEXT: # %bb.1: # %cond.store -; SKX-NEXT: vpextrb $0, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_2: # %else -; SKX-NEXT: kshiftrw $1, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_4 -; SKX-NEXT: # %bb.3: # %cond.store1 -; SKX-NEXT: vpextrb $1, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_4: # %else2 -; SKX-NEXT: kshiftrw $2, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_6 -; SKX-NEXT: # %bb.5: # %cond.store4 -; SKX-NEXT: vpextrb $2, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_6: # %else5 -; SKX-NEXT: kshiftrw $3, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_8 -; SKX-NEXT: # %bb.7: # %cond.store7 -; SKX-NEXT: vpextrb $3, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_8: # %else8 -; SKX-NEXT: kshiftrw $4, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_10 -; SKX-NEXT: # %bb.9: # %cond.store10 -; SKX-NEXT: vpextrb $4, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_10: # %else11 -; SKX-NEXT: kshiftrw $5, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_12 -; SKX-NEXT: # %bb.11: # %cond.store13 -; SKX-NEXT: vpextrb $5, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_12: # %else14 -; SKX-NEXT: kshiftrw $6, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_14 -; SKX-NEXT: # %bb.13: # %cond.store16 -; SKX-NEXT: vpextrb $6, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_14: # %else17 -; SKX-NEXT: kshiftrw $7, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_16 -; SKX-NEXT: # %bb.15: # %cond.store19 -; SKX-NEXT: vpextrb $7, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_16: # %else20 -; SKX-NEXT: kshiftrw $8, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_18 -; SKX-NEXT: # %bb.17: # %cond.store22 -; SKX-NEXT: vpextrb $8, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_18: # %else23 -; SKX-NEXT: kshiftrw $9, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_20 -; SKX-NEXT: # %bb.19: # %cond.store25 -; SKX-NEXT: vpextrb $9, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_20: # %else26 -; SKX-NEXT: kshiftrw $10, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_22 -; SKX-NEXT: # %bb.21: # %cond.store28 -; SKX-NEXT: vpextrb $10, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_22: # %else29 -; SKX-NEXT: kshiftrw $11, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_24 -; SKX-NEXT: # %bb.23: # %cond.store31 -; SKX-NEXT: vpextrb $11, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_24: # %else32 -; SKX-NEXT: kshiftrw $12, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_26 -; SKX-NEXT: # %bb.25: # %cond.store34 -; SKX-NEXT: vpextrb $12, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_26: # %else35 -; SKX-NEXT: kshiftrw $13, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_28 -; SKX-NEXT: # %bb.27: # %cond.store37 -; SKX-NEXT: vpextrb $13, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_28: # %else38 -; SKX-NEXT: kshiftrw $14, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_30 -; SKX-NEXT: # %bb.29: # %cond.store40 -; SKX-NEXT: vpextrb $14, %xmm0, (%rdi) -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_30: # %else41 -; SKX-NEXT: kshiftrw $15, %k0, %k0 -; SKX-NEXT: kmovd %k0, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_32 -; SKX-NEXT: # %bb.31: # %cond.store43 -; SKX-NEXT: vpextrb $15, %xmm0, (%rdi) -; SKX-NEXT: .LBB12_32: # %else44 -; SKX-NEXT: retq -; -; KNL-LABEL: compressstore_v16i8_v16i8: -; KNL: # %bb.0: -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_2 -; KNL-NEXT: # %bb.1: # %cond.store -; KNL-NEXT: vpextrb $0, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_2: # %else -; KNL-NEXT: kshiftrw $1, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_4 -; KNL-NEXT: # %bb.3: # %cond.store1 -; KNL-NEXT: vpextrb $1, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_4: # %else2 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $2, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_6 -; KNL-NEXT: # %bb.5: # %cond.store4 -; KNL-NEXT: vpextrb $2, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_6: # %else5 -; KNL-NEXT: kshiftrw $3, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_8 -; KNL-NEXT: # %bb.7: # %cond.store7 -; KNL-NEXT: vpextrb $3, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_8: # %else8 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $4, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_10 -; KNL-NEXT: # %bb.9: # %cond.store10 -; KNL-NEXT: vpextrb $4, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_10: # %else11 -; KNL-NEXT: kshiftrw $5, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_12 -; KNL-NEXT: # %bb.11: # %cond.store13 -; KNL-NEXT: vpextrb $5, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_12: # %else14 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $6, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_14 -; KNL-NEXT: # %bb.13: # %cond.store16 -; KNL-NEXT: vpextrb $6, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_14: # %else17 -; KNL-NEXT: kshiftrw $7, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_16 -; KNL-NEXT: # %bb.15: # %cond.store19 -; KNL-NEXT: vpextrb $7, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_16: # %else20 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $8, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_18 -; KNL-NEXT: # %bb.17: # %cond.store22 -; KNL-NEXT: vpextrb $8, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_18: # %else23 -; KNL-NEXT: kshiftrw $9, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_20 -; KNL-NEXT: # %bb.19: # %cond.store25 -; KNL-NEXT: vpextrb $9, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_20: # %else26 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $10, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_22 -; KNL-NEXT: # %bb.21: # %cond.store28 -; KNL-NEXT: vpextrb $10, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_22: # %else29 -; KNL-NEXT: kshiftrw $11, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_24 -; KNL-NEXT: # %bb.23: # %cond.store31 -; KNL-NEXT: vpextrb $11, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_24: # %else32 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $12, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_26 -; KNL-NEXT: # %bb.25: # %cond.store34 -; KNL-NEXT: vpextrb $12, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_26: # %else35 -; KNL-NEXT: kshiftrw $13, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_28 -; KNL-NEXT: # %bb.27: # %cond.store37 -; KNL-NEXT: vpextrb $13, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_28: # %else38 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1 -; KNL-NEXT: vpmovsxbd %xmm1, %zmm1 -; KNL-NEXT: vptestmd %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftrw $14, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_30 -; KNL-NEXT: # %bb.29: # %cond.store40 -; KNL-NEXT: vpextrb $14, %xmm0, (%rdi) -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_30: # %else41 -; KNL-NEXT: kshiftrw $15, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_32 -; KNL-NEXT: # %bb.31: # %cond.store43 -; KNL-NEXT: vpextrb $15, %xmm0, (%rdi) -; KNL-NEXT: .LBB12_32: # %else44 -; KNL-NEXT: retq +; SSE2-LABEL: compressstore_v16i8_v16i8: +; SSE2: ## %bb.0: +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: movd %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_2 +; SSE2-NEXT: ## %bb.1: ## %cond.store +; SSE2-NEXT: movd %xmm0, %ecx +; SSE2-NEXT: movb %cl, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_2: ## %else +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_4 +; SSE2-NEXT: ## %bb.3: ## %cond.store1 +; SSE2-NEXT: movd %xmm0, %eax +; SSE2-NEXT: movb %ah, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_4: ## %else2 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: movd %xmm2, %eax +; SSE2-NEXT: movl %eax, %ecx +; SSE2-NEXT: shrl $16, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB12_6 +; SSE2-NEXT: ## %bb.5: ## %cond.store4 +; SSE2-NEXT: movd %xmm0, %ecx +; SSE2-NEXT: shrl $16, %ecx +; SSE2-NEXT: movb %cl, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_6: ## %else5 +; SSE2-NEXT: shrl $24, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_8 +; SSE2-NEXT: ## %bb.7: ## %cond.store7 +; SSE2-NEXT: movd %xmm0, %eax +; SSE2-NEXT: shrl $24, %eax +; SSE2-NEXT: movb %al, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_8: ## %else8 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: pextrw $2, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_10 +; SSE2-NEXT: ## %bb.9: ## %cond.store10 +; SSE2-NEXT: pextrw $2, %xmm0, %ecx +; SSE2-NEXT: movb %cl, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_10: ## %else11 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_12 +; SSE2-NEXT: ## %bb.11: ## %cond.store13 +; SSE2-NEXT: pextrw $2, %xmm0, %eax +; SSE2-NEXT: movb %ah, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_12: ## %else14 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: pextrw $3, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_14 +; SSE2-NEXT: ## %bb.13: ## %cond.store16 +; SSE2-NEXT: pextrw $3, %xmm0, %ecx +; SSE2-NEXT: movb %cl, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_14: ## %else17 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_16 +; SSE2-NEXT: ## %bb.15: ## %cond.store19 +; SSE2-NEXT: pextrw $3, %xmm0, %eax +; SSE2-NEXT: movb %ah, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_16: ## %else20 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: pextrw $4, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_18 +; SSE2-NEXT: ## %bb.17: ## %cond.store22 +; SSE2-NEXT: pextrw $4, %xmm0, %ecx +; SSE2-NEXT: movb %cl, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_18: ## %else23 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_20 +; SSE2-NEXT: ## %bb.19: ## %cond.store25 +; SSE2-NEXT: pextrw $4, %xmm0, %eax +; SSE2-NEXT: movb %ah, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_20: ## %else26 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: pextrw $5, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_22 +; SSE2-NEXT: ## %bb.21: ## %cond.store28 +; SSE2-NEXT: pextrw $5, %xmm0, %ecx +; SSE2-NEXT: movb %cl, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_22: ## %else29 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_24 +; SSE2-NEXT: ## %bb.23: ## %cond.store31 +; SSE2-NEXT: pextrw $5, %xmm0, %eax +; SSE2-NEXT: movb %ah, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_24: ## %else32 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: pextrw $6, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_26 +; SSE2-NEXT: ## %bb.25: ## %cond.store34 +; SSE2-NEXT: pextrw $6, %xmm0, %ecx +; SSE2-NEXT: movb %cl, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_26: ## %else35 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_28 +; SSE2-NEXT: ## %bb.27: ## %cond.store37 +; SSE2-NEXT: pextrw $6, %xmm0, %eax +; SSE2-NEXT: movb %ah, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_28: ## %else38 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm2, %xmm1 +; SSE2-NEXT: pextrw $7, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_30 +; SSE2-NEXT: ## %bb.29: ## %cond.store40 +; SSE2-NEXT: pextrw $7, %xmm0, %ecx +; SSE2-NEXT: movb %cl, (%rdi) +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_30: ## %else41 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_32 +; SSE2-NEXT: ## %bb.31: ## %cond.store43 +; SSE2-NEXT: pextrw $7, %xmm0, %eax +; SSE2-NEXT: movb %ah, (%rdi) +; SSE2-NEXT: LBB12_32: ## %else44 +; SSE2-NEXT: retq +; +; SSE42-LABEL: compressstore_v16i8_v16i8: +; SSE42: ## %bb.0: +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $0, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_2 +; SSE42-NEXT: ## %bb.1: ## %cond.store +; SSE42-NEXT: pextrb $0, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_2: ## %else +; SSE42-NEXT: pextrb $1, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_4 +; SSE42-NEXT: ## %bb.3: ## %cond.store1 +; SSE42-NEXT: pextrb $1, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_4: ## %else2 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $2, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_6 +; SSE42-NEXT: ## %bb.5: ## %cond.store4 +; SSE42-NEXT: pextrb $2, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_6: ## %else5 +; SSE42-NEXT: pextrb $3, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_8 +; SSE42-NEXT: ## %bb.7: ## %cond.store7 +; SSE42-NEXT: pextrb $3, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_8: ## %else8 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $4, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_10 +; SSE42-NEXT: ## %bb.9: ## %cond.store10 +; SSE42-NEXT: pextrb $4, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_10: ## %else11 +; SSE42-NEXT: pextrb $5, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_12 +; SSE42-NEXT: ## %bb.11: ## %cond.store13 +; SSE42-NEXT: pextrb $5, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_12: ## %else14 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $6, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_14 +; SSE42-NEXT: ## %bb.13: ## %cond.store16 +; SSE42-NEXT: pextrb $6, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_14: ## %else17 +; SSE42-NEXT: pextrb $7, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_16 +; SSE42-NEXT: ## %bb.15: ## %cond.store19 +; SSE42-NEXT: pextrb $7, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_16: ## %else20 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $8, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_18 +; SSE42-NEXT: ## %bb.17: ## %cond.store22 +; SSE42-NEXT: pextrb $8, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_18: ## %else23 +; SSE42-NEXT: pextrb $9, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_20 +; SSE42-NEXT: ## %bb.19: ## %cond.store25 +; SSE42-NEXT: pextrb $9, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_20: ## %else26 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $10, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_22 +; SSE42-NEXT: ## %bb.21: ## %cond.store28 +; SSE42-NEXT: pextrb $10, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_22: ## %else29 +; SSE42-NEXT: pextrb $11, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_24 +; SSE42-NEXT: ## %bb.23: ## %cond.store31 +; SSE42-NEXT: pextrb $11, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_24: ## %else32 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $12, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_26 +; SSE42-NEXT: ## %bb.25: ## %cond.store34 +; SSE42-NEXT: pextrb $12, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_26: ## %else35 +; SSE42-NEXT: pextrb $13, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_28 +; SSE42-NEXT: ## %bb.27: ## %cond.store37 +; SSE42-NEXT: pextrb $13, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_28: ## %else38 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm2, %xmm1 +; SSE42-NEXT: pextrb $14, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_30 +; SSE42-NEXT: ## %bb.29: ## %cond.store40 +; SSE42-NEXT: pextrb $14, %xmm0, (%rdi) +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_30: ## %else41 +; SSE42-NEXT: pextrb $15, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_32 +; SSE42-NEXT: ## %bb.31: ## %cond.store43 +; SSE42-NEXT: pextrb $15, %xmm0, (%rdi) +; SSE42-NEXT: LBB12_32: ## %else44 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: compressstore_v16i8_v16i8: +; AVX1OR2: ## %bb.0: +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $0, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_2 +; AVX1OR2-NEXT: ## %bb.1: ## %cond.store +; AVX1OR2-NEXT: vpextrb $0, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_2: ## %else +; AVX1OR2-NEXT: vpextrb $1, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_4 +; AVX1OR2-NEXT: ## %bb.3: ## %cond.store1 +; AVX1OR2-NEXT: vpextrb $1, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_4: ## %else2 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $2, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_6 +; AVX1OR2-NEXT: ## %bb.5: ## %cond.store4 +; AVX1OR2-NEXT: vpextrb $2, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_6: ## %else5 +; AVX1OR2-NEXT: vpextrb $3, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_8 +; AVX1OR2-NEXT: ## %bb.7: ## %cond.store7 +; AVX1OR2-NEXT: vpextrb $3, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_8: ## %else8 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $4, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_10 +; AVX1OR2-NEXT: ## %bb.9: ## %cond.store10 +; AVX1OR2-NEXT: vpextrb $4, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_10: ## %else11 +; AVX1OR2-NEXT: vpextrb $5, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_12 +; AVX1OR2-NEXT: ## %bb.11: ## %cond.store13 +; AVX1OR2-NEXT: vpextrb $5, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_12: ## %else14 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $6, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_14 +; AVX1OR2-NEXT: ## %bb.13: ## %cond.store16 +; AVX1OR2-NEXT: vpextrb $6, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_14: ## %else17 +; AVX1OR2-NEXT: vpextrb $7, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_16 +; AVX1OR2-NEXT: ## %bb.15: ## %cond.store19 +; AVX1OR2-NEXT: vpextrb $7, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_16: ## %else20 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $8, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_18 +; AVX1OR2-NEXT: ## %bb.17: ## %cond.store22 +; AVX1OR2-NEXT: vpextrb $8, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_18: ## %else23 +; AVX1OR2-NEXT: vpextrb $9, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_20 +; AVX1OR2-NEXT: ## %bb.19: ## %cond.store25 +; AVX1OR2-NEXT: vpextrb $9, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_20: ## %else26 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $10, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_22 +; AVX1OR2-NEXT: ## %bb.21: ## %cond.store28 +; AVX1OR2-NEXT: vpextrb $10, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_22: ## %else29 +; AVX1OR2-NEXT: vpextrb $11, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_24 +; AVX1OR2-NEXT: ## %bb.23: ## %cond.store31 +; AVX1OR2-NEXT: vpextrb $11, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_24: ## %else32 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $12, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_26 +; AVX1OR2-NEXT: ## %bb.25: ## %cond.store34 +; AVX1OR2-NEXT: vpextrb $12, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_26: ## %else35 +; AVX1OR2-NEXT: vpextrb $13, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_28 +; AVX1OR2-NEXT: ## %bb.27: ## %cond.store37 +; AVX1OR2-NEXT: vpextrb $13, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_28: ## %else38 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1 +; AVX1OR2-NEXT: vpextrb $14, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_30 +; AVX1OR2-NEXT: ## %bb.29: ## %cond.store40 +; AVX1OR2-NEXT: vpextrb $14, %xmm0, (%rdi) +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_30: ## %else41 +; AVX1OR2-NEXT: vpextrb $15, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_32 +; AVX1OR2-NEXT: ## %bb.31: ## %cond.store43 +; AVX1OR2-NEXT: vpextrb $15, %xmm0, (%rdi) +; AVX1OR2-NEXT: LBB12_32: ## %else44 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: compressstore_v16i8_v16i8: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_2 +; AVX512F-NEXT: ## %bb.1: ## %cond.store +; AVX512F-NEXT: vpextrb $0, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_2: ## %else +; AVX512F-NEXT: kshiftrw $1, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_4 +; AVX512F-NEXT: ## %bb.3: ## %cond.store1 +; AVX512F-NEXT: vpextrb $1, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_4: ## %else2 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $2, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_6 +; AVX512F-NEXT: ## %bb.5: ## %cond.store4 +; AVX512F-NEXT: vpextrb $2, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_6: ## %else5 +; AVX512F-NEXT: kshiftrw $3, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_8 +; AVX512F-NEXT: ## %bb.7: ## %cond.store7 +; AVX512F-NEXT: vpextrb $3, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_8: ## %else8 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $4, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_10 +; AVX512F-NEXT: ## %bb.9: ## %cond.store10 +; AVX512F-NEXT: vpextrb $4, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_10: ## %else11 +; AVX512F-NEXT: kshiftrw $5, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_12 +; AVX512F-NEXT: ## %bb.11: ## %cond.store13 +; AVX512F-NEXT: vpextrb $5, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_12: ## %else14 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $6, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_14 +; AVX512F-NEXT: ## %bb.13: ## %cond.store16 +; AVX512F-NEXT: vpextrb $6, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_14: ## %else17 +; AVX512F-NEXT: kshiftrw $7, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_16 +; AVX512F-NEXT: ## %bb.15: ## %cond.store19 +; AVX512F-NEXT: vpextrb $7, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_16: ## %else20 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $8, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_18 +; AVX512F-NEXT: ## %bb.17: ## %cond.store22 +; AVX512F-NEXT: vpextrb $8, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_18: ## %else23 +; AVX512F-NEXT: kshiftrw $9, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_20 +; AVX512F-NEXT: ## %bb.19: ## %cond.store25 +; AVX512F-NEXT: vpextrb $9, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_20: ## %else26 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $10, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_22 +; AVX512F-NEXT: ## %bb.21: ## %cond.store28 +; AVX512F-NEXT: vpextrb $10, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_22: ## %else29 +; AVX512F-NEXT: kshiftrw $11, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_24 +; AVX512F-NEXT: ## %bb.23: ## %cond.store31 +; AVX512F-NEXT: vpextrb $11, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_24: ## %else32 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $12, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_26 +; AVX512F-NEXT: ## %bb.25: ## %cond.store34 +; AVX512F-NEXT: vpextrb $12, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_26: ## %else35 +; AVX512F-NEXT: kshiftrw $13, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_28 +; AVX512F-NEXT: ## %bb.27: ## %cond.store37 +; AVX512F-NEXT: vpextrb $13, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_28: ## %else38 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1 +; AVX512F-NEXT: vpmovsxbd %xmm1, %zmm1 +; AVX512F-NEXT: vptestmd %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftrw $14, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_30 +; AVX512F-NEXT: ## %bb.29: ## %cond.store40 +; AVX512F-NEXT: vpextrb $14, %xmm0, (%rdi) +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_30: ## %else41 +; AVX512F-NEXT: kshiftrw $15, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_32 +; AVX512F-NEXT: ## %bb.31: ## %cond.store43 +; AVX512F-NEXT: vpextrb $15, %xmm0, (%rdi) +; AVX512F-NEXT: LBB12_32: ## %else44 +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: compressstore_v16i8_v16i8: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vptestnmb %xmm1, %xmm1, %k0 +; AVX512VLBW-NEXT: kmovd %k0, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_2 +; AVX512VLBW-NEXT: ## %bb.1: ## %cond.store +; AVX512VLBW-NEXT: vpextrb $0, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_2: ## %else +; AVX512VLBW-NEXT: kshiftrw $1, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_4 +; AVX512VLBW-NEXT: ## %bb.3: ## %cond.store1 +; AVX512VLBW-NEXT: vpextrb $1, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_4: ## %else2 +; AVX512VLBW-NEXT: kshiftrw $2, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_6 +; AVX512VLBW-NEXT: ## %bb.5: ## %cond.store4 +; AVX512VLBW-NEXT: vpextrb $2, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_6: ## %else5 +; AVX512VLBW-NEXT: kshiftrw $3, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_8 +; AVX512VLBW-NEXT: ## %bb.7: ## %cond.store7 +; AVX512VLBW-NEXT: vpextrb $3, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_8: ## %else8 +; AVX512VLBW-NEXT: kshiftrw $4, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_10 +; AVX512VLBW-NEXT: ## %bb.9: ## %cond.store10 +; AVX512VLBW-NEXT: vpextrb $4, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_10: ## %else11 +; AVX512VLBW-NEXT: kshiftrw $5, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_12 +; AVX512VLBW-NEXT: ## %bb.11: ## %cond.store13 +; AVX512VLBW-NEXT: vpextrb $5, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_12: ## %else14 +; AVX512VLBW-NEXT: kshiftrw $6, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_14 +; AVX512VLBW-NEXT: ## %bb.13: ## %cond.store16 +; AVX512VLBW-NEXT: vpextrb $6, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_14: ## %else17 +; AVX512VLBW-NEXT: kshiftrw $7, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_16 +; AVX512VLBW-NEXT: ## %bb.15: ## %cond.store19 +; AVX512VLBW-NEXT: vpextrb $7, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_16: ## %else20 +; AVX512VLBW-NEXT: kshiftrw $8, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_18 +; AVX512VLBW-NEXT: ## %bb.17: ## %cond.store22 +; AVX512VLBW-NEXT: vpextrb $8, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_18: ## %else23 +; AVX512VLBW-NEXT: kshiftrw $9, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_20 +; AVX512VLBW-NEXT: ## %bb.19: ## %cond.store25 +; AVX512VLBW-NEXT: vpextrb $9, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_20: ## %else26 +; AVX512VLBW-NEXT: kshiftrw $10, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_22 +; AVX512VLBW-NEXT: ## %bb.21: ## %cond.store28 +; AVX512VLBW-NEXT: vpextrb $10, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_22: ## %else29 +; AVX512VLBW-NEXT: kshiftrw $11, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_24 +; AVX512VLBW-NEXT: ## %bb.23: ## %cond.store31 +; AVX512VLBW-NEXT: vpextrb $11, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_24: ## %else32 +; AVX512VLBW-NEXT: kshiftrw $12, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_26 +; AVX512VLBW-NEXT: ## %bb.25: ## %cond.store34 +; AVX512VLBW-NEXT: vpextrb $12, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_26: ## %else35 +; AVX512VLBW-NEXT: kshiftrw $13, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_28 +; AVX512VLBW-NEXT: ## %bb.27: ## %cond.store37 +; AVX512VLBW-NEXT: vpextrb $13, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_28: ## %else38 +; AVX512VLBW-NEXT: kshiftrw $14, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_30 +; AVX512VLBW-NEXT: ## %bb.29: ## %cond.store40 +; AVX512VLBW-NEXT: vpextrb $14, %xmm0, (%rdi) +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_30: ## %else41 +; AVX512VLBW-NEXT: kshiftrw $15, %k0, %k0 +; AVX512VLBW-NEXT: kmovd %k0, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_32 +; AVX512VLBW-NEXT: ## %bb.31: ## %cond.store43 +; AVX512VLBW-NEXT: vpextrb $15, %xmm0, (%rdi) +; AVX512VLBW-NEXT: LBB12_32: ## %else44 +; AVX512VLBW-NEXT: retq %mask = icmp eq <16 x i8> %trigger, zeroinitializer call void @llvm.masked.compressstore.v16i8(<16 x i8> %V, i8* %base, <16 x i1> %mask) ret void diff --git a/llvm/test/CodeGen/X86/masked_expandload.ll b/llvm/test/CodeGen/X86/masked_expandload.ll index 377f832..5b89d34 100644 --- a/llvm/test/CodeGen/X86/masked_expandload.ll +++ b/llvm/test/CodeGen/X86/masked_expandload.ll @@ -1,421 +1,1210 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=haswell| FileCheck %s --check-prefixes=ALL,HSW -; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=skylake-avx512 | FileCheck %s --check-prefixes=ALL,SKX -; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=knl | FileCheck %s --check-prefixes=ALL,KNL +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=sse2 | FileCheck %s --check-prefixes=SSE,SSE2 +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=sse4.2 | FileCheck %s --check-prefixes=SSE,SSE42 +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=avx | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX1 +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=avx2 | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX2 +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=avx512f | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F +; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=avx512f,avx512bw,avx512vl | FileCheck %s --check-prefixes=AVX,AVX512,AVX512VLBW ; ; vXf64 ; define <2 x double> @expandload_v2f64_v2i64(double* %base, <2 x double> %src0, <2 x i64> %trigger) { -; HSW-LABEL: expandload_v2f64_v2i64: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $0, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB0_2 -; HSW-NEXT: # %bb.1: # %cond.load -; HSW-NEXT: vmovlpd {{.*#+}} xmm0 = mem[0],xmm0[1] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB0_2: # %else -; HSW-NEXT: vpextrb $8, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB0_4 -; HSW-NEXT: # %bb.3: # %cond.load1 -; HSW-NEXT: vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0] -; HSW-NEXT: .LBB0_4: # %else2 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v2f64_v2i64: -; SKX: # %bb.0: -; SKX-NEXT: vptestnmq %xmm1, %xmm1, %k1 -; SKX-NEXT: vexpandpd (%rdi), %xmm0 {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v2f64_v2i64: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; KNL-NEXT: vptestnmq %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftlw $14, %k0, %k0 -; KNL-NEXT: kshiftrw $14, %k0, %k1 -; KNL-NEXT: vexpandpd (%rdi), %zmm0 {%k1} -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v2f64_v2i64: +; SSE2: ## %bb.0: +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqd %xmm1, %xmm2 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,0,3,2] +; SSE2-NEXT: pand %xmm2, %xmm1 +; SSE2-NEXT: movd %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB0_2 +; SSE2-NEXT: ## %bb.1: ## %cond.load +; SSE2-NEXT: movlpd {{.*#+}} xmm0 = mem[0],xmm0[1] +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB0_2: ## %else +; SSE2-NEXT: pextrw $4, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB0_4 +; SSE2-NEXT: ## %bb.3: ## %cond.load1 +; SSE2-NEXT: movhpd {{.*#+}} xmm0 = xmm0[0],mem[0] +; SSE2-NEXT: LBB0_4: ## %else2 +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v2f64_v2i64: +; SSE42: ## %bb.0: +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqq %xmm1, %xmm2 +; SSE42-NEXT: pextrb $0, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB0_2 +; SSE42-NEXT: ## %bb.1: ## %cond.load +; SSE42-NEXT: movlpd {{.*#+}} xmm0 = mem[0],xmm0[1] +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB0_2: ## %else +; SSE42-NEXT: pextrb $8, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB0_4 +; SSE42-NEXT: ## %bb.3: ## %cond.load1 +; SSE42-NEXT: movhpd {{.*#+}} xmm0 = xmm0[0],mem[0] +; SSE42-NEXT: LBB0_4: ## %else2 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: expandload_v2f64_v2i64: +; AVX1OR2: ## %bb.0: +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm1 +; AVX1OR2-NEXT: vpextrb $0, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB0_2 +; AVX1OR2-NEXT: ## %bb.1: ## %cond.load +; AVX1OR2-NEXT: vmovlpd {{.*#+}} xmm0 = mem[0],xmm0[1] +; AVX1OR2-NEXT: addq $8, %rdi +; AVX1OR2-NEXT: LBB0_2: ## %else +; AVX1OR2-NEXT: vpextrb $8, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB0_4 +; AVX1OR2-NEXT: ## %bb.3: ## %cond.load1 +; AVX1OR2-NEXT: vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0] +; AVX1OR2-NEXT: LBB0_4: ## %else2 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: expandload_v2f64_v2i64: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 +; AVX512F-NEXT: vptestnmq %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftlw $14, %k0, %k0 +; AVX512F-NEXT: kshiftrw $14, %k0, %k1 +; AVX512F-NEXT: vexpandpd (%rdi), %zmm0 {%k1} +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v2f64_v2i64: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vptestnmq %xmm1, %xmm1, %k1 +; AVX512VLBW-NEXT: vexpandpd (%rdi), %xmm0 {%k1} +; AVX512VLBW-NEXT: retq %mask = icmp eq <2 x i64> %trigger, zeroinitializer %res = call <2 x double> @llvm.masked.expandload.v2f64(double* %base, <2 x i1> %mask, <2 x double> %src0) ret <2 x double>%res } define <4 x double> @expandload_v4f64_v4i64(double* %base, <4 x double> %src0, <4 x i64> %trigger) { -; HSW-LABEL: expandload_v4f64_v4i64: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqq %ymm2, %ymm1, %ymm2 -; HSW-NEXT: vpextrb $0, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_2 -; HSW-NEXT: # %bb.1: # %cond.load -; HSW-NEXT: vmovsd {{.*#+}} xmm3 = mem[0],zero -; HSW-NEXT: vblendpd {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_2: # %else -; HSW-NEXT: vpextrb $8, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_4 -; HSW-NEXT: # %bb.3: # %cond.load1 -; HSW-NEXT: vmovhpd {{.*#+}} xmm2 = xmm0[0],mem[0] -; HSW-NEXT: vblendpd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_4: # %else2 -; HSW-NEXT: vxorpd %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqq %ymm2, %ymm1, %ymm1 -; HSW-NEXT: vextracti128 $1, %ymm1, %xmm1 -; HSW-NEXT: vpextrb $0, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_6 -; HSW-NEXT: # %bb.5: # %cond.load5 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm2 -; HSW-NEXT: vmovlpd {{.*#+}} xmm2 = mem[0],xmm2[1] -; HSW-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB1_6: # %else6 -; HSW-NEXT: vpextrb $8, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB1_8 -; HSW-NEXT: # %bb.7: # %cond.load9 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm1 -; HSW-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 -; HSW-NEXT: .LBB1_8: # %else10 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v4f64_v4i64: -; SKX: # %bb.0: -; SKX-NEXT: vptestnmq %ymm1, %ymm1, %k1 -; SKX-NEXT: vexpandpd (%rdi), %ymm0 {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v4f64_v4i64: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1 -; KNL-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; KNL-NEXT: vptestnmq %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftlw $12, %k0, %k0 -; KNL-NEXT: kshiftrw $12, %k0, %k1 -; KNL-NEXT: vexpandpd (%rdi), %zmm0 {%k1} -; KNL-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0 -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v4f64_v4i64: +; SSE2: ## %bb.0: +; SSE2-NEXT: pxor %xmm4, %xmm4 +; SSE2-NEXT: pcmpeqd %xmm2, %xmm4 +; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,0,3,2] +; SSE2-NEXT: pand %xmm4, %xmm2 +; SSE2-NEXT: movd %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_2 +; SSE2-NEXT: ## %bb.1: ## %cond.load +; SSE2-NEXT: movlpd {{.*#+}} xmm0 = mem[0],xmm0[1] +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_2: ## %else +; SSE2-NEXT: pextrw $4, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_4 +; SSE2-NEXT: ## %bb.3: ## %cond.load1 +; SSE2-NEXT: movhpd {{.*#+}} xmm0 = xmm0[0],mem[0] +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_4: ## %else2 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqd %xmm2, %xmm3 +; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2] +; SSE2-NEXT: pand %xmm3, %xmm2 +; SSE2-NEXT: pextrw $0, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_6 +; SSE2-NEXT: ## %bb.5: ## %cond.load5 +; SSE2-NEXT: movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB1_6: ## %else6 +; SSE2-NEXT: pextrw $4, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB1_8 +; SSE2-NEXT: ## %bb.7: ## %cond.load9 +; SSE2-NEXT: movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] +; SSE2-NEXT: LBB1_8: ## %else10 +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v4f64_v4i64: +; SSE42: ## %bb.0: +; SSE42-NEXT: pxor %xmm4, %xmm4 +; SSE42-NEXT: pcmpeqq %xmm2, %xmm4 +; SSE42-NEXT: pextrb $0, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_2 +; SSE42-NEXT: ## %bb.1: ## %cond.load +; SSE42-NEXT: movlpd {{.*#+}} xmm0 = mem[0],xmm0[1] +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_2: ## %else +; SSE42-NEXT: pextrb $8, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_4 +; SSE42-NEXT: ## %bb.3: ## %cond.load1 +; SSE42-NEXT: movhpd {{.*#+}} xmm0 = xmm0[0],mem[0] +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_4: ## %else2 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqq %xmm2, %xmm3 +; SSE42-NEXT: pextrb $0, %xmm3, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_6 +; SSE42-NEXT: ## %bb.5: ## %cond.load5 +; SSE42-NEXT: movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB1_6: ## %else6 +; SSE42-NEXT: pextrb $8, %xmm3, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB1_8 +; SSE42-NEXT: ## %bb.7: ## %cond.load9 +; SSE42-NEXT: movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] +; SSE42-NEXT: LBB1_8: ## %else10 +; SSE42-NEXT: retq +; +; AVX1-LABEL: expandload_v4f64_v4i64: +; AVX1: ## %bb.0: +; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm3 +; AVX1-NEXT: vpextrb $0, %xmm3, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_2 +; AVX1-NEXT: ## %bb.1: ## %cond.load +; AVX1-NEXT: vmovsd {{.*#+}} xmm3 = mem[0],zero +; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7] +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_2: ## %else +; AVX1-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm2 +; AVX1-NEXT: vpextrb $8, %xmm2, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_4 +; AVX1-NEXT: ## %bb.3: ## %cond.load1 +; AVX1-NEXT: vmovhpd {{.*#+}} xmm2 = xmm0[0],mem[0] +; AVX1-NEXT: vblendpd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3] +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_4: ## %else2 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm1 +; AVX1-NEXT: vxorpd %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm1 +; AVX1-NEXT: vpextrb $0, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_6 +; AVX1-NEXT: ## %bb.5: ## %cond.load5 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; AVX1-NEXT: vmovlpd {{.*#+}} xmm2 = mem[0],xmm2[1] +; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB1_6: ## %else6 +; AVX1-NEXT: vpextrb $8, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB1_8 +; AVX1-NEXT: ## %bb.7: ## %cond.load9 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1 +; AVX1-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] +; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; AVX1-NEXT: LBB1_8: ## %else10 +; AVX1-NEXT: retq +; +; AVX2-LABEL: expandload_v4f64_v4i64: +; AVX2: ## %bb.0: +; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX2-NEXT: vpcmpeqq %ymm2, %ymm1, %ymm2 +; AVX2-NEXT: vpextrb $0, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_2 +; AVX2-NEXT: ## %bb.1: ## %cond.load +; AVX2-NEXT: vmovsd {{.*#+}} xmm3 = mem[0],zero +; AVX2-NEXT: vblendpd {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3] +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_2: ## %else +; AVX2-NEXT: vpextrb $8, %xmm2, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_4 +; AVX2-NEXT: ## %bb.3: ## %cond.load1 +; AVX2-NEXT: vmovhpd {{.*#+}} xmm2 = xmm0[0],mem[0] +; AVX2-NEXT: vblendpd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3] +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_4: ## %else2 +; AVX2-NEXT: vxorpd %xmm2, %xmm2, %xmm2 +; AVX2-NEXT: vpcmpeqq %ymm2, %ymm1, %ymm1 +; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm1 +; AVX2-NEXT: vpextrb $0, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_6 +; AVX2-NEXT: ## %bb.5: ## %cond.load5 +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm2 +; AVX2-NEXT: vmovlpd {{.*#+}} xmm2 = mem[0],xmm2[1] +; AVX2-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB1_6: ## %else6 +; AVX2-NEXT: vpextrb $8, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB1_8 +; AVX2-NEXT: ## %bb.7: ## %cond.load9 +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm1 +; AVX2-NEXT: vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0] +; AVX2-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; AVX2-NEXT: LBB1_8: ## %else10 +; AVX2-NEXT: retq +; +; AVX512F-LABEL: expandload_v4f64_v4i64: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 +; AVX512F-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 +; AVX512F-NEXT: vptestnmq %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftlw $12, %k0, %k0 +; AVX512F-NEXT: kshiftrw $12, %k0, %k1 +; AVX512F-NEXT: vexpandpd (%rdi), %zmm0 {%k1} +; AVX512F-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v4f64_v4i64: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vptestnmq %ymm1, %ymm1, %k1 +; AVX512VLBW-NEXT: vexpandpd (%rdi), %ymm0 {%k1} +; AVX512VLBW-NEXT: retq %mask = icmp eq <4 x i64> %trigger, zeroinitializer %res = call <4 x double> @llvm.masked.expandload.v4f64(double* %base, <4 x i1> %mask, <4 x double> %src0) ret <4 x double>%res } define <8 x double> @expandload_v8f64_v8i1(double* %base, <8 x double> %src0, <8 x i1> %mask) { -; HSW-LABEL: expandload_v8f64_v8i1: -; HSW: # %bb.0: -; HSW-NEXT: vpextrb $0, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB2_2 -; HSW-NEXT: # %bb.1: # %cond.load -; HSW-NEXT: vmovsd {{.*#+}} xmm3 = mem[0],zero -; HSW-NEXT: vblendpd {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB2_2: # %else -; HSW-NEXT: vpextrb $2, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB2_4 -; HSW-NEXT: # %bb.3: # %cond.load1 -; HSW-NEXT: vmovhpd {{.*#+}} xmm3 = xmm0[0],mem[0] -; HSW-NEXT: vblendpd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB2_4: # %else2 -; HSW-NEXT: vpextrb $4, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB2_6 -; HSW-NEXT: # %bb.5: # %cond.load5 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm3 -; HSW-NEXT: vmovlpd {{.*#+}} xmm3 = mem[0],xmm3[1] -; HSW-NEXT: vinsertf128 $1, %xmm3, %ymm0, %ymm0 -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB2_6: # %else6 -; HSW-NEXT: vpextrb $6, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB2_8 -; HSW-NEXT: # %bb.7: # %cond.load9 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm3 -; HSW-NEXT: vmovhpd {{.*#+}} xmm3 = xmm3[0],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm3, %ymm0, %ymm0 -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB2_8: # %else10 -; HSW-NEXT: vpextrb $8, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB2_10 -; HSW-NEXT: # %bb.9: # %cond.load13 -; HSW-NEXT: vmovsd {{.*#+}} xmm3 = mem[0],zero -; HSW-NEXT: vblendpd {{.*#+}} ymm1 = ymm3[0],ymm1[1,2,3] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB2_10: # %else14 -; HSW-NEXT: vpextrb $10, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB2_12 -; HSW-NEXT: # %bb.11: # %cond.load17 -; HSW-NEXT: vmovhpd {{.*#+}} xmm3 = xmm1[0],mem[0] -; HSW-NEXT: vblendpd {{.*#+}} ymm1 = ymm3[0,1],ymm1[2,3] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB2_12: # %else18 -; HSW-NEXT: vpextrb $12, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB2_14 -; HSW-NEXT: # %bb.13: # %cond.load21 -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm3 -; HSW-NEXT: vmovlpd {{.*#+}} xmm3 = mem[0],xmm3[1] -; HSW-NEXT: vinsertf128 $1, %xmm3, %ymm1, %ymm1 -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB2_14: # %else22 -; HSW-NEXT: vpextrb $14, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB2_16 -; HSW-NEXT: # %bb.15: # %cond.load25 -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm2 -; HSW-NEXT: vmovhpd {{.*#+}} xmm2 = xmm2[0],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm2, %ymm1, %ymm1 -; HSW-NEXT: .LBB2_16: # %else26 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v8f64_v8i1: -; SKX: # %bb.0: -; SKX-NEXT: vpsllw $15, %xmm1, %xmm1 -; SKX-NEXT: vpmovw2m %xmm1, %k1 -; SKX-NEXT: vexpandpd (%rdi), %zmm0 {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v8f64_v8i1: -; KNL: # %bb.0: -; KNL-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero -; KNL-NEXT: vpsllq $63, %zmm1, %zmm1 -; KNL-NEXT: vptestmq %zmm1, %zmm1, %k1 -; KNL-NEXT: vexpandpd (%rdi), %zmm0 {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v8f64_v8i1: +; SSE2: ## %bb.0: +; SSE2-NEXT: movd %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB2_2 +; SSE2-NEXT: ## %bb.1: ## %cond.load +; SSE2-NEXT: movlpd {{.*#+}} xmm0 = mem[0],xmm0[1] +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB2_2: ## %else +; SSE2-NEXT: shrl $16, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB2_4 +; SSE2-NEXT: ## %bb.3: ## %cond.load1 +; SSE2-NEXT: movhpd {{.*#+}} xmm0 = xmm0[0],mem[0] +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB2_4: ## %else2 +; SSE2-NEXT: pextrw $2, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB2_6 +; SSE2-NEXT: ## %bb.5: ## %cond.load5 +; SSE2-NEXT: movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB2_6: ## %else6 +; SSE2-NEXT: pextrw $3, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB2_8 +; SSE2-NEXT: ## %bb.7: ## %cond.load9 +; SSE2-NEXT: movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB2_8: ## %else10 +; SSE2-NEXT: pextrw $4, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB2_10 +; SSE2-NEXT: ## %bb.9: ## %cond.load13 +; SSE2-NEXT: movlpd {{.*#+}} xmm2 = mem[0],xmm2[1] +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB2_10: ## %else14 +; SSE2-NEXT: pextrw $5, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB2_12 +; SSE2-NEXT: ## %bb.11: ## %cond.load17 +; SSE2-NEXT: movhpd {{.*#+}} xmm2 = xmm2[0],mem[0] +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB2_12: ## %else18 +; SSE2-NEXT: pextrw $6, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB2_14 +; SSE2-NEXT: ## %bb.13: ## %cond.load21 +; SSE2-NEXT: movlpd {{.*#+}} xmm3 = mem[0],xmm3[1] +; SSE2-NEXT: addq $8, %rdi +; SSE2-NEXT: LBB2_14: ## %else22 +; SSE2-NEXT: pextrw $7, %xmm4, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB2_16 +; SSE2-NEXT: ## %bb.15: ## %cond.load25 +; SSE2-NEXT: movhpd {{.*#+}} xmm3 = xmm3[0],mem[0] +; SSE2-NEXT: LBB2_16: ## %else26 +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v8f64_v8i1: +; SSE42: ## %bb.0: +; SSE42-NEXT: pextrb $0, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB2_2 +; SSE42-NEXT: ## %bb.1: ## %cond.load +; SSE42-NEXT: movlpd {{.*#+}} xmm0 = mem[0],xmm0[1] +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB2_2: ## %else +; SSE42-NEXT: pextrb $2, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB2_4 +; SSE42-NEXT: ## %bb.3: ## %cond.load1 +; SSE42-NEXT: movhpd {{.*#+}} xmm0 = xmm0[0],mem[0] +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB2_4: ## %else2 +; SSE42-NEXT: pextrb $4, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB2_6 +; SSE42-NEXT: ## %bb.5: ## %cond.load5 +; SSE42-NEXT: movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB2_6: ## %else6 +; SSE42-NEXT: pextrb $6, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB2_8 +; SSE42-NEXT: ## %bb.7: ## %cond.load9 +; SSE42-NEXT: movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB2_8: ## %else10 +; SSE42-NEXT: pextrb $8, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB2_10 +; SSE42-NEXT: ## %bb.9: ## %cond.load13 +; SSE42-NEXT: movlpd {{.*#+}} xmm2 = mem[0],xmm2[1] +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB2_10: ## %else14 +; SSE42-NEXT: pextrb $10, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB2_12 +; SSE42-NEXT: ## %bb.11: ## %cond.load17 +; SSE42-NEXT: movhpd {{.*#+}} xmm2 = xmm2[0],mem[0] +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB2_12: ## %else18 +; SSE42-NEXT: pextrb $12, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB2_14 +; SSE42-NEXT: ## %bb.13: ## %cond.load21 +; SSE42-NEXT: movlpd {{.*#+}} xmm3 = mem[0],xmm3[1] +; SSE42-NEXT: addq $8, %rdi +; SSE42-NEXT: LBB2_14: ## %else22 +; SSE42-NEXT: pextrb $14, %xmm4, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB2_16 +; SSE42-NEXT: ## %bb.15: ## %cond.load25 +; SSE42-NEXT: movhpd {{.*#+}} xmm3 = xmm3[0],mem[0] +; SSE42-NEXT: LBB2_16: ## %else26 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: expandload_v8f64_v8i1: +; AVX1OR2: ## %bb.0: +; AVX1OR2-NEXT: vpextrb $0, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB2_2 +; AVX1OR2-NEXT: ## %bb.1: ## %cond.load +; AVX1OR2-NEXT: vmovsd {{.*#+}} xmm3 = mem[0],zero +; AVX1OR2-NEXT: vblendpd {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3] +; AVX1OR2-NEXT: addq $8, %rdi +; AVX1OR2-NEXT: LBB2_2: ## %else +; AVX1OR2-NEXT: vpextrb $2, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB2_4 +; AVX1OR2-NEXT: ## %bb.3: ## %cond.load1 +; AVX1OR2-NEXT: vmovhpd {{.*#+}} xmm3 = xmm0[0],mem[0] +; AVX1OR2-NEXT: vblendpd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3] +; AVX1OR2-NEXT: addq $8, %rdi +; AVX1OR2-NEXT: LBB2_4: ## %else2 +; AVX1OR2-NEXT: vpextrb $4, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB2_6 +; AVX1OR2-NEXT: ## %bb.5: ## %cond.load5 +; AVX1OR2-NEXT: vextractf128 $1, %ymm0, %xmm3 +; AVX1OR2-NEXT: vmovlpd {{.*#+}} xmm3 = mem[0],xmm3[1] +; AVX1OR2-NEXT: vinsertf128 $1, %xmm3, %ymm0, %ymm0 +; AVX1OR2-NEXT: addq $8, %rdi +; AVX1OR2-NEXT: LBB2_6: ## %else6 +; AVX1OR2-NEXT: vpextrb $6, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB2_8 +; AVX1OR2-NEXT: ## %bb.7: ## %cond.load9 +; AVX1OR2-NEXT: vextractf128 $1, %ymm0, %xmm3 +; AVX1OR2-NEXT: vmovhpd {{.*#+}} xmm3 = xmm3[0],mem[0] +; AVX1OR2-NEXT: vinsertf128 $1, %xmm3, %ymm0, %ymm0 +; AVX1OR2-NEXT: addq $8, %rdi +; AVX1OR2-NEXT: LBB2_8: ## %else10 +; AVX1OR2-NEXT: vpextrb $8, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB2_10 +; AVX1OR2-NEXT: ## %bb.9: ## %cond.load13 +; AVX1OR2-NEXT: vmovsd {{.*#+}} xmm3 = mem[0],zero +; AVX1OR2-NEXT: vblendpd {{.*#+}} ymm1 = ymm3[0],ymm1[1,2,3] +; AVX1OR2-NEXT: addq $8, %rdi +; AVX1OR2-NEXT: LBB2_10: ## %else14 +; AVX1OR2-NEXT: vpextrb $10, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB2_12 +; AVX1OR2-NEXT: ## %bb.11: ## %cond.load17 +; AVX1OR2-NEXT: vmovhpd {{.*#+}} xmm3 = xmm1[0],mem[0] +; AVX1OR2-NEXT: vblendpd {{.*#+}} ymm1 = ymm3[0,1],ymm1[2,3] +; AVX1OR2-NEXT: addq $8, %rdi +; AVX1OR2-NEXT: LBB2_12: ## %else18 +; AVX1OR2-NEXT: vpextrb $12, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB2_14 +; AVX1OR2-NEXT: ## %bb.13: ## %cond.load21 +; AVX1OR2-NEXT: vextractf128 $1, %ymm1, %xmm3 +; AVX1OR2-NEXT: vmovlpd {{.*#+}} xmm3 = mem[0],xmm3[1] +; AVX1OR2-NEXT: vinsertf128 $1, %xmm3, %ymm1, %ymm1 +; AVX1OR2-NEXT: addq $8, %rdi +; AVX1OR2-NEXT: LBB2_14: ## %else22 +; AVX1OR2-NEXT: vpextrb $14, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB2_16 +; AVX1OR2-NEXT: ## %bb.15: ## %cond.load25 +; AVX1OR2-NEXT: vextractf128 $1, %ymm1, %xmm2 +; AVX1OR2-NEXT: vmovhpd {{.*#+}} xmm2 = xmm2[0],mem[0] +; AVX1OR2-NEXT: vinsertf128 $1, %xmm2, %ymm1, %ymm1 +; AVX1OR2-NEXT: LBB2_16: ## %else26 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: expandload_v8f64_v8i1: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero +; AVX512F-NEXT: vpsllq $63, %zmm1, %zmm1 +; AVX512F-NEXT: vptestmq %zmm1, %zmm1, %k1 +; AVX512F-NEXT: vexpandpd (%rdi), %zmm0 {%k1} +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v8f64_v8i1: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vpsllw $15, %xmm1, %xmm1 +; AVX512VLBW-NEXT: vpmovw2m %xmm1, %k1 +; AVX512VLBW-NEXT: vexpandpd (%rdi), %zmm0 {%k1} +; AVX512VLBW-NEXT: retq %res = call <8 x double> @llvm.masked.expandload.v8f64(double* %base, <8 x i1> %mask, <8 x double> %src0) ret <8 x double>%res } define <16 x double> @expandload_v16f64_v16i32(double* %base, <16 x double> %src0, <16 x i32> %trigger) { -; HSW-LABEL: expandload_v16f64_v16i32: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm6, %xmm6, %xmm6 -; HSW-NEXT: vpcmpeqd %ymm6, %ymm4, %ymm6 -; HSW-NEXT: vpackssdw %xmm0, %xmm6, %xmm7 -; HSW-NEXT: vpacksswb %xmm0, %xmm7, %xmm7 -; HSW-NEXT: vpextrb $0, %xmm7, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_2 -; HSW-NEXT: # %bb.1: # %cond.load -; HSW-NEXT: vmovq {{.*#+}} xmm7 = mem[0],zero -; HSW-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm0[2,3,4,5,6,7] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_2: # %else -; HSW-NEXT: vpackssdw %xmm0, %xmm6, %xmm6 -; HSW-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 -; HSW-NEXT: vpextrb $1, %xmm6, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_4 -; HSW-NEXT: # %bb.3: # %cond.load1 -; HSW-NEXT: vmovhpd {{.*#+}} xmm6 = xmm0[0],mem[0] -; HSW-NEXT: vblendpd {{.*#+}} ymm0 = ymm6[0,1],ymm0[2,3] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_4: # %else2 -; HSW-NEXT: vxorpd %xmm6, %xmm6, %xmm6 -; HSW-NEXT: vpcmpeqd %ymm6, %ymm4, %ymm6 -; HSW-NEXT: vpackssdw %xmm0, %xmm6, %xmm7 -; HSW-NEXT: vpacksswb %xmm0, %xmm7, %xmm7 -; HSW-NEXT: vpextrb $2, %xmm7, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_6 -; HSW-NEXT: # %bb.5: # %cond.load5 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm7 -; HSW-NEXT: vmovlpd {{.*#+}} xmm7 = mem[0],xmm7[1] -; HSW-NEXT: vinsertf128 $1, %xmm7, %ymm0, %ymm0 -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_6: # %else6 -; HSW-NEXT: vpackssdw %xmm0, %xmm6, %xmm6 -; HSW-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 -; HSW-NEXT: vpextrb $3, %xmm6, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_8 -; HSW-NEXT: # %bb.7: # %cond.load9 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm6 -; HSW-NEXT: vmovhpd {{.*#+}} xmm6 = xmm6[0],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm6, %ymm0, %ymm0 -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_8: # %else10 -; HSW-NEXT: vxorpd %xmm6, %xmm6, %xmm6 -; HSW-NEXT: vpcmpeqd %ymm6, %ymm4, %ymm6 -; HSW-NEXT: vextracti128 $1, %ymm6, %xmm6 -; HSW-NEXT: vpackssdw %xmm6, %xmm0, %xmm7 -; HSW-NEXT: vpacksswb %xmm0, %xmm7, %xmm7 -; HSW-NEXT: vpextrb $4, %xmm7, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_10 -; HSW-NEXT: # %bb.9: # %cond.load13 -; HSW-NEXT: vmovq {{.*#+}} xmm7 = mem[0],zero -; HSW-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm1[2,3,4,5,6,7] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_10: # %else14 -; HSW-NEXT: vpackssdw %xmm6, %xmm0, %xmm6 -; HSW-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 -; HSW-NEXT: vpextrb $5, %xmm6, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_12 -; HSW-NEXT: # %bb.11: # %cond.load17 -; HSW-NEXT: vmovhpd {{.*#+}} xmm6 = xmm1[0],mem[0] -; HSW-NEXT: vblendpd {{.*#+}} ymm1 = ymm6[0,1],ymm1[2,3] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_12: # %else18 -; HSW-NEXT: vxorpd %xmm6, %xmm6, %xmm6 -; HSW-NEXT: vpcmpeqd %ymm6, %ymm4, %ymm4 -; HSW-NEXT: vextracti128 $1, %ymm4, %xmm4 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm6 -; HSW-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 -; HSW-NEXT: vpextrb $6, %xmm6, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_14 -; HSW-NEXT: # %bb.13: # %cond.load21 -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm6 -; HSW-NEXT: vmovlpd {{.*#+}} xmm6 = mem[0],xmm6[1] -; HSW-NEXT: vinsertf128 $1, %xmm6, %ymm1, %ymm1 -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_14: # %else22 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpextrb $7, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_16 -; HSW-NEXT: # %bb.15: # %cond.load25 -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm4 -; HSW-NEXT: vmovhpd {{.*#+}} xmm4 = xmm4[0],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_16: # %else26 -; HSW-NEXT: vxorpd %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm6 -; HSW-NEXT: vpacksswb %xmm6, %xmm0, %xmm6 -; HSW-NEXT: vpextrb $8, %xmm6, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_18 -; HSW-NEXT: # %bb.17: # %cond.load29 -; HSW-NEXT: vmovq {{.*#+}} xmm6 = mem[0],zero -; HSW-NEXT: vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm2[2,3,4,5,6,7] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_18: # %else30 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $9, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_20 -; HSW-NEXT: # %bb.19: # %cond.load33 -; HSW-NEXT: vmovhpd {{.*#+}} xmm4 = xmm2[0],mem[0] -; HSW-NEXT: vblendpd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_20: # %else34 -; HSW-NEXT: vxorpd %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm6 -; HSW-NEXT: vpacksswb %xmm6, %xmm0, %xmm6 -; HSW-NEXT: vpextrb $10, %xmm6, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_22 -; HSW-NEXT: # %bb.21: # %cond.load37 -; HSW-NEXT: vextractf128 $1, %ymm2, %xmm6 -; HSW-NEXT: vmovlpd {{.*#+}} xmm6 = mem[0],xmm6[1] -; HSW-NEXT: vinsertf128 $1, %xmm6, %ymm2, %ymm2 -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_22: # %else38 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $11, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_24 -; HSW-NEXT: # %bb.23: # %cond.load41 -; HSW-NEXT: vextractf128 $1, %ymm2, %xmm4 -; HSW-NEXT: vmovhpd {{.*#+}} xmm4 = xmm4[0],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm2, %ymm2 -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_24: # %else42 -; HSW-NEXT: vxorpd %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 -; HSW-NEXT: vextracti128 $1, %ymm4, %xmm4 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm6 -; HSW-NEXT: vpacksswb %xmm6, %xmm0, %xmm6 -; HSW-NEXT: vpextrb $12, %xmm6, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_26 -; HSW-NEXT: # %bb.25: # %cond.load45 -; HSW-NEXT: vmovq {{.*#+}} xmm6 = mem[0],zero -; HSW-NEXT: vpblendd {{.*#+}} ymm3 = ymm6[0,1],ymm3[2,3,4,5,6,7] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_26: # %else46 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $13, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_28 -; HSW-NEXT: # %bb.27: # %cond.load49 -; HSW-NEXT: vmovhpd {{.*#+}} xmm4 = xmm3[0],mem[0] -; HSW-NEXT: vblendpd {{.*#+}} ymm3 = ymm4[0,1],ymm3[2,3] -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_28: # %else50 -; HSW-NEXT: vxorpd %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 -; HSW-NEXT: vextracti128 $1, %ymm4, %xmm4 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 -; HSW-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 -; HSW-NEXT: vpextrb $14, %xmm5, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_30 -; HSW-NEXT: # %bb.29: # %cond.load53 -; HSW-NEXT: vextractf128 $1, %ymm3, %xmm5 -; HSW-NEXT: vmovlpd {{.*#+}} xmm5 = mem[0],xmm5[1] -; HSW-NEXT: vinsertf128 $1, %xmm5, %ymm3, %ymm3 -; HSW-NEXT: addq $8, %rdi -; HSW-NEXT: .LBB3_30: # %else54 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $15, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB3_32 -; HSW-NEXT: # %bb.31: # %cond.load57 -; HSW-NEXT: vextractf128 $1, %ymm3, %xmm4 -; HSW-NEXT: vmovhpd {{.*#+}} xmm4 = xmm4[0],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm3, %ymm3 -; HSW-NEXT: .LBB3_32: # %else58 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v16f64_v16i32: -; SKX: # %bb.0: -; SKX-NEXT: vextracti64x4 $1, %zmm2, %ymm3 -; SKX-NEXT: vptestnmd %ymm3, %ymm3, %k1 -; SKX-NEXT: vptestnmd %ymm2, %ymm2, %k2 -; SKX-NEXT: kmovb %k2, %eax -; SKX-NEXT: popcntl %eax, %eax -; SKX-NEXT: vexpandpd (%rdi,%rax,8), %zmm1 {%k1} -; SKX-NEXT: vexpandpd (%rdi), %zmm0 {%k2} -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v16f64_v16i32: -; KNL: # %bb.0: -; KNL-NEXT: vextracti64x4 $1, %zmm2, %ymm3 -; KNL-NEXT: vptestnmd %zmm3, %zmm3, %k1 -; KNL-NEXT: vptestnmd %zmm2, %zmm2, %k2 -; KNL-NEXT: vexpandpd (%rdi), %zmm0 {%k2} -; KNL-NEXT: kmovw %k2, %eax -; KNL-NEXT: movzbl %al, %eax -; KNL-NEXT: popcntl %eax, %eax -; KNL-NEXT: vexpandpd (%rdi,%rax,8), %zmm1 {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v16f64_v16i32: +; SSE2: ## %bb.0: +; SSE2-NEXT: movq %rdi, %rax +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE2-NEXT: pxor %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: movd %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_2 +; SSE2-NEXT: ## %bb.1: ## %cond.load +; SSE2-NEXT: movlpd {{.*#+}} xmm0 = mem[0],xmm0[1] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_2: ## %else +; SSE2-NEXT: pextrw $2, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_4 +; SSE2-NEXT: ## %bb.3: ## %cond.load1 +; SSE2-NEXT: movhpd {{.*#+}} xmm0 = xmm0[0],mem[0] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_4: ## %else2 +; SSE2-NEXT: pxor %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: pextrw $4, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_6 +; SSE2-NEXT: ## %bb.5: ## %cond.load5 +; SSE2-NEXT: movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_6: ## %else6 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE2-NEXT: pextrw $6, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_8 +; SSE2-NEXT: ## %bb.7: ## %cond.load9 +; SSE2-NEXT: movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_8: ## %else10 +; SSE2-NEXT: pxor %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: movd %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_10 +; SSE2-NEXT: ## %bb.9: ## %cond.load13 +; SSE2-NEXT: movlpd {{.*#+}} xmm2 = mem[0],xmm2[1] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_10: ## %else14 +; SSE2-NEXT: pextrw $2, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_12 +; SSE2-NEXT: ## %bb.11: ## %cond.load17 +; SSE2-NEXT: movhpd {{.*#+}} xmm2 = xmm2[0],mem[0] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_12: ## %else18 +; SSE2-NEXT: pxor %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: pextrw $4, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_14 +; SSE2-NEXT: ## %bb.13: ## %cond.load21 +; SSE2-NEXT: movlpd {{.*#+}} xmm3 = mem[0],xmm3[1] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_14: ## %else22 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE2-NEXT: pextrw $6, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_16 +; SSE2-NEXT: ## %bb.15: ## %cond.load25 +; SSE2-NEXT: movhpd {{.*#+}} xmm3 = xmm3[0],mem[0] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_16: ## %else26 +; SSE2-NEXT: pxor %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: movd %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_18 +; SSE2-NEXT: ## %bb.17: ## %cond.load29 +; SSE2-NEXT: movlpd {{.*#+}} xmm4 = mem[0],xmm4[1] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_18: ## %else30 +; SSE2-NEXT: pextrw $2, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_20 +; SSE2-NEXT: ## %bb.19: ## %cond.load33 +; SSE2-NEXT: movhpd {{.*#+}} xmm4 = xmm4[0],mem[0] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_20: ## %else34 +; SSE2-NEXT: pxor %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: pextrw $4, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_22 +; SSE2-NEXT: ## %bb.21: ## %cond.load37 +; SSE2-NEXT: movlpd {{.*#+}} xmm5 = mem[0],xmm5[1] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_22: ## %else38 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE2-NEXT: pextrw $6, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_24 +; SSE2-NEXT: ## %bb.23: ## %cond.load41 +; SSE2-NEXT: movhpd {{.*#+}} xmm5 = xmm5[0],mem[0] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_24: ## %else42 +; SSE2-NEXT: pxor %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: movd %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_26 +; SSE2-NEXT: ## %bb.25: ## %cond.load45 +; SSE2-NEXT: movlpd {{.*#+}} xmm6 = mem[0],xmm6[1] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_26: ## %else46 +; SSE2-NEXT: pextrw $2, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_28 +; SSE2-NEXT: ## %bb.27: ## %cond.load49 +; SSE2-NEXT: movhpd {{.*#+}} xmm6 = xmm6[0],mem[0] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_28: ## %else50 +; SSE2-NEXT: pxor %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: pextrw $4, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_30 +; SSE2-NEXT: ## %bb.29: ## %cond.load53 +; SSE2-NEXT: movlpd {{.*#+}} xmm7 = mem[0],xmm7[1] +; SSE2-NEXT: addq $8, %rsi +; SSE2-NEXT: LBB3_30: ## %else54 +; SSE2-NEXT: pextrw $6, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB3_32 +; SSE2-NEXT: ## %bb.31: ## %cond.load57 +; SSE2-NEXT: movhpd {{.*#+}} xmm7 = xmm7[0],mem[0] +; SSE2-NEXT: LBB3_32: ## %else58 +; SSE2-NEXT: movapd %xmm0, (%rax) +; SSE2-NEXT: movapd %xmm1, 16(%rax) +; SSE2-NEXT: movapd %xmm2, 32(%rax) +; SSE2-NEXT: movapd %xmm3, 48(%rax) +; SSE2-NEXT: movapd %xmm4, 64(%rax) +; SSE2-NEXT: movapd %xmm5, 80(%rax) +; SSE2-NEXT: movapd %xmm6, 96(%rax) +; SSE2-NEXT: movapd %xmm7, 112(%rax) +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v16f64_v16i32: +; SSE42: ## %bb.0: +; SSE42-NEXT: movq %rdi, %rax +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $0, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_2 +; SSE42-NEXT: ## %bb.1: ## %cond.load +; SSE42-NEXT: movlpd {{.*#+}} xmm0 = mem[0],xmm0[1] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_2: ## %else +; SSE42-NEXT: pextrb $4, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_4 +; SSE42-NEXT: ## %bb.3: ## %cond.load1 +; SSE42-NEXT: movhpd {{.*#+}} xmm0 = xmm0[0],mem[0] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_4: ## %else2 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $8, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_6 +; SSE42-NEXT: ## %bb.5: ## %cond.load5 +; SSE42-NEXT: movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_6: ## %else6 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE42-NEXT: pextrb $12, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_8 +; SSE42-NEXT: ## %bb.7: ## %cond.load9 +; SSE42-NEXT: movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_8: ## %else10 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $0, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_10 +; SSE42-NEXT: ## %bb.9: ## %cond.load13 +; SSE42-NEXT: movlpd {{.*#+}} xmm2 = mem[0],xmm2[1] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_10: ## %else14 +; SSE42-NEXT: pextrb $4, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_12 +; SSE42-NEXT: ## %bb.11: ## %cond.load17 +; SSE42-NEXT: movhpd {{.*#+}} xmm2 = xmm2[0],mem[0] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_12: ## %else18 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $8, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_14 +; SSE42-NEXT: ## %bb.13: ## %cond.load21 +; SSE42-NEXT: movlpd {{.*#+}} xmm3 = mem[0],xmm3[1] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_14: ## %else22 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE42-NEXT: pextrb $12, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_16 +; SSE42-NEXT: ## %bb.15: ## %cond.load25 +; SSE42-NEXT: movhpd {{.*#+}} xmm3 = xmm3[0],mem[0] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_16: ## %else26 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $0, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_18 +; SSE42-NEXT: ## %bb.17: ## %cond.load29 +; SSE42-NEXT: movlpd {{.*#+}} xmm4 = mem[0],xmm4[1] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_18: ## %else30 +; SSE42-NEXT: pextrb $4, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_20 +; SSE42-NEXT: ## %bb.19: ## %cond.load33 +; SSE42-NEXT: movhpd {{.*#+}} xmm4 = xmm4[0],mem[0] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_20: ## %else34 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $8, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_22 +; SSE42-NEXT: ## %bb.21: ## %cond.load37 +; SSE42-NEXT: movlpd {{.*#+}} xmm5 = mem[0],xmm5[1] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_22: ## %else38 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE42-NEXT: pextrb $12, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_24 +; SSE42-NEXT: ## %bb.23: ## %cond.load41 +; SSE42-NEXT: movhpd {{.*#+}} xmm5 = xmm5[0],mem[0] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_24: ## %else42 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $0, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_26 +; SSE42-NEXT: ## %bb.25: ## %cond.load45 +; SSE42-NEXT: movlpd {{.*#+}} xmm6 = mem[0],xmm6[1] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_26: ## %else46 +; SSE42-NEXT: pextrb $4, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_28 +; SSE42-NEXT: ## %bb.27: ## %cond.load49 +; SSE42-NEXT: movhpd {{.*#+}} xmm6 = xmm6[0],mem[0] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_28: ## %else50 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $8, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_30 +; SSE42-NEXT: ## %bb.29: ## %cond.load53 +; SSE42-NEXT: movlpd {{.*#+}} xmm7 = mem[0],xmm7[1] +; SSE42-NEXT: addq $8, %rsi +; SSE42-NEXT: LBB3_30: ## %else54 +; SSE42-NEXT: pextrb $12, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB3_32 +; SSE42-NEXT: ## %bb.31: ## %cond.load57 +; SSE42-NEXT: movhpd {{.*#+}} xmm7 = xmm7[0],mem[0] +; SSE42-NEXT: LBB3_32: ## %else58 +; SSE42-NEXT: movapd %xmm0, (%rax) +; SSE42-NEXT: movapd %xmm1, 16(%rax) +; SSE42-NEXT: movapd %xmm2, 32(%rax) +; SSE42-NEXT: movapd %xmm3, 48(%rax) +; SSE42-NEXT: movapd %xmm4, 64(%rax) +; SSE42-NEXT: movapd %xmm5, 80(%rax) +; SSE42-NEXT: movapd %xmm6, 96(%rax) +; SSE42-NEXT: movapd %xmm7, 112(%rax) +; SSE42-NEXT: retq +; +; AVX1-LABEL: expandload_v16f64_v16i32: +; AVX1: ## %bb.0: +; AVX1-NEXT: vpxor %xmm6, %xmm6, %xmm6 +; AVX1-NEXT: vpcmpeqd %xmm6, %xmm4, %xmm7 +; AVX1-NEXT: vpackssdw %xmm0, %xmm7, %xmm7 +; AVX1-NEXT: vpacksswb %xmm0, %xmm7, %xmm7 +; AVX1-NEXT: vpextrb $0, %xmm7, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_2 +; AVX1-NEXT: ## %bb.1: ## %cond.load +; AVX1-NEXT: vmovsd {{.*#+}} xmm7 = mem[0],zero +; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm7[0,1],ymm0[2,3,4,5,6,7] +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_2: ## %else +; AVX1-NEXT: vpcmpeqd %xmm6, %xmm4, %xmm6 +; AVX1-NEXT: vpackssdw %xmm0, %xmm6, %xmm6 +; AVX1-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 +; AVX1-NEXT: vpextrb $1, %xmm6, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_4 +; AVX1-NEXT: ## %bb.3: ## %cond.load1 +; AVX1-NEXT: vmovhpd {{.*#+}} xmm6 = xmm0[0],mem[0] +; AVX1-NEXT: vblendpd {{.*#+}} ymm0 = ymm6[0,1],ymm0[2,3] +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_4: ## %else2 +; AVX1-NEXT: vxorpd %xmm6, %xmm6, %xmm6 +; AVX1-NEXT: vpcmpeqd %xmm6, %xmm4, %xmm7 +; AVX1-NEXT: vpackssdw %xmm0, %xmm7, %xmm7 +; AVX1-NEXT: vpacksswb %xmm0, %xmm7, %xmm7 +; AVX1-NEXT: vpextrb $2, %xmm7, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_6 +; AVX1-NEXT: ## %bb.5: ## %cond.load5 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm7 +; AVX1-NEXT: vmovlpd {{.*#+}} xmm7 = mem[0],xmm7[1] +; AVX1-NEXT: vinsertf128 $1, %xmm7, %ymm0, %ymm0 +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_6: ## %else6 +; AVX1-NEXT: vpcmpeqd %xmm6, %xmm4, %xmm6 +; AVX1-NEXT: vpackssdw %xmm0, %xmm6, %xmm6 +; AVX1-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 +; AVX1-NEXT: vpextrb $3, %xmm6, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_8 +; AVX1-NEXT: ## %bb.7: ## %cond.load9 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm6 +; AVX1-NEXT: vmovhpd {{.*#+}} xmm6 = xmm6[0],mem[0] +; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm0, %ymm0 +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_8: ## %else10 +; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm4 +; AVX1-NEXT: vxorpd %xmm6, %xmm6, %xmm6 +; AVX1-NEXT: vpcmpeqd %xmm6, %xmm4, %xmm6 +; AVX1-NEXT: vpackssdw %xmm6, %xmm0, %xmm7 +; AVX1-NEXT: vpacksswb %xmm0, %xmm7, %xmm7 +; AVX1-NEXT: vpextrb $4, %xmm7, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_10 +; AVX1-NEXT: ## %bb.9: ## %cond.load13 +; AVX1-NEXT: vmovsd {{.*#+}} xmm7 = mem[0],zero +; AVX1-NEXT: vblendps {{.*#+}} ymm1 = ymm7[0,1],ymm1[2,3,4,5,6,7] +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_10: ## %else14 +; AVX1-NEXT: vpackssdw %xmm6, %xmm0, %xmm6 +; AVX1-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 +; AVX1-NEXT: vpextrb $5, %xmm6, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_12 +; AVX1-NEXT: ## %bb.11: ## %cond.load17 +; AVX1-NEXT: vmovhpd {{.*#+}} xmm6 = xmm1[0],mem[0] +; AVX1-NEXT: vblendpd {{.*#+}} ymm1 = ymm6[0,1],ymm1[2,3] +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_12: ## %else18 +; AVX1-NEXT: vxorpd %xmm6, %xmm6, %xmm6 +; AVX1-NEXT: vpcmpeqd %xmm6, %xmm4, %xmm4 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm6 +; AVX1-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 +; AVX1-NEXT: vpextrb $6, %xmm6, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_14 +; AVX1-NEXT: ## %bb.13: ## %cond.load21 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm6 +; AVX1-NEXT: vmovlpd {{.*#+}} xmm6 = mem[0],xmm6[1] +; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm1, %ymm1 +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_14: ## %else22 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $7, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_16 +; AVX1-NEXT: ## %bb.15: ## %cond.load25 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm4 +; AVX1-NEXT: vmovhpd {{.*#+}} xmm4 = xmm4[0],mem[0] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_16: ## %else26 +; AVX1-NEXT: vxorpd %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm5, %xmm6 +; AVX1-NEXT: vpackssdw %xmm0, %xmm6, %xmm6 +; AVX1-NEXT: vpacksswb %xmm6, %xmm0, %xmm6 +; AVX1-NEXT: vpextrb $8, %xmm6, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_18 +; AVX1-NEXT: ## %bb.17: ## %cond.load29 +; AVX1-NEXT: vmovsd {{.*#+}} xmm6 = mem[0],zero +; AVX1-NEXT: vblendps {{.*#+}} ymm2 = ymm6[0,1],ymm2[2,3,4,5,6,7] +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_18: ## %else30 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm5, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $9, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_20 +; AVX1-NEXT: ## %bb.19: ## %cond.load33 +; AVX1-NEXT: vmovhpd {{.*#+}} xmm4 = xmm2[0],mem[0] +; AVX1-NEXT: vblendpd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3] +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_20: ## %else34 +; AVX1-NEXT: vxorpd %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm5, %xmm6 +; AVX1-NEXT: vpackssdw %xmm0, %xmm6, %xmm6 +; AVX1-NEXT: vpacksswb %xmm6, %xmm0, %xmm6 +; AVX1-NEXT: vpextrb $10, %xmm6, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_22 +; AVX1-NEXT: ## %bb.21: ## %cond.load37 +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm6 +; AVX1-NEXT: vmovlpd {{.*#+}} xmm6 = mem[0],xmm6[1] +; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm2, %ymm2 +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_22: ## %else38 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm5, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $11, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_24 +; AVX1-NEXT: ## %bb.23: ## %cond.load41 +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm4 +; AVX1-NEXT: vmovhpd {{.*#+}} xmm4 = xmm4[0],mem[0] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm2, %ymm2 +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_24: ## %else42 +; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm4 +; AVX1-NEXT: vpxor %xmm5, %xmm5, %xmm5 +; AVX1-NEXT: vpcmpeqd %xmm5, %xmm4, %xmm5 +; AVX1-NEXT: vpackssdw %xmm5, %xmm0, %xmm6 +; AVX1-NEXT: vpacksswb %xmm6, %xmm0, %xmm6 +; AVX1-NEXT: vpextrb $12, %xmm6, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_26 +; AVX1-NEXT: ## %bb.25: ## %cond.load45 +; AVX1-NEXT: vmovsd {{.*#+}} xmm6 = mem[0],zero +; AVX1-NEXT: vblendps {{.*#+}} ymm3 = ymm6[0,1],ymm3[2,3,4,5,6,7] +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_26: ## %else46 +; AVX1-NEXT: vpackssdw %xmm5, %xmm0, %xmm5 +; AVX1-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX1-NEXT: vpextrb $13, %xmm5, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_28 +; AVX1-NEXT: ## %bb.27: ## %cond.load49 +; AVX1-NEXT: vmovhpd {{.*#+}} xmm5 = xmm3[0],mem[0] +; AVX1-NEXT: vblendpd {{.*#+}} ymm3 = ymm5[0,1],ymm3[2,3] +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_28: ## %else50 +; AVX1-NEXT: vxorpd %xmm5, %xmm5, %xmm5 +; AVX1-NEXT: vpcmpeqd %xmm5, %xmm4, %xmm4 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 +; AVX1-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX1-NEXT: vpextrb $14, %xmm5, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_30 +; AVX1-NEXT: ## %bb.29: ## %cond.load53 +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm5 +; AVX1-NEXT: vmovlpd {{.*#+}} xmm5 = mem[0],xmm5[1] +; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm3, %ymm3 +; AVX1-NEXT: addq $8, %rdi +; AVX1-NEXT: LBB3_30: ## %else54 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $15, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB3_32 +; AVX1-NEXT: ## %bb.31: ## %cond.load57 +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm4 +; AVX1-NEXT: vmovhpd {{.*#+}} xmm4 = xmm4[0],mem[0] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm3, %ymm3 +; AVX1-NEXT: LBB3_32: ## %else58 +; AVX1-NEXT: retq +; +; AVX2-LABEL: expandload_v16f64_v16i32: +; AVX2: ## %bb.0: +; AVX2-NEXT: vpxor %xmm6, %xmm6, %xmm6 +; AVX2-NEXT: vpcmpeqd %ymm6, %ymm4, %ymm6 +; AVX2-NEXT: vpackssdw %xmm0, %xmm6, %xmm7 +; AVX2-NEXT: vpacksswb %xmm0, %xmm7, %xmm7 +; AVX2-NEXT: vpextrb $0, %xmm7, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_2 +; AVX2-NEXT: ## %bb.1: ## %cond.load +; AVX2-NEXT: vmovq {{.*#+}} xmm7 = mem[0],zero +; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm0[2,3,4,5,6,7] +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_2: ## %else +; AVX2-NEXT: vpackssdw %xmm0, %xmm6, %xmm6 +; AVX2-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 +; AVX2-NEXT: vpextrb $1, %xmm6, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_4 +; AVX2-NEXT: ## %bb.3: ## %cond.load1 +; AVX2-NEXT: vmovhpd {{.*#+}} xmm6 = xmm0[0],mem[0] +; AVX2-NEXT: vblendpd {{.*#+}} ymm0 = ymm6[0,1],ymm0[2,3] +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_4: ## %else2 +; AVX2-NEXT: vxorpd %xmm6, %xmm6, %xmm6 +; AVX2-NEXT: vpcmpeqd %ymm6, %ymm4, %ymm6 +; AVX2-NEXT: vpackssdw %xmm0, %xmm6, %xmm7 +; AVX2-NEXT: vpacksswb %xmm0, %xmm7, %xmm7 +; AVX2-NEXT: vpextrb $2, %xmm7, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_6 +; AVX2-NEXT: ## %bb.5: ## %cond.load5 +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm7 +; AVX2-NEXT: vmovlpd {{.*#+}} xmm7 = mem[0],xmm7[1] +; AVX2-NEXT: vinsertf128 $1, %xmm7, %ymm0, %ymm0 +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_6: ## %else6 +; AVX2-NEXT: vpackssdw %xmm0, %xmm6, %xmm6 +; AVX2-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 +; AVX2-NEXT: vpextrb $3, %xmm6, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_8 +; AVX2-NEXT: ## %bb.7: ## %cond.load9 +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm6 +; AVX2-NEXT: vmovhpd {{.*#+}} xmm6 = xmm6[0],mem[0] +; AVX2-NEXT: vinsertf128 $1, %xmm6, %ymm0, %ymm0 +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_8: ## %else10 +; AVX2-NEXT: vxorpd %xmm6, %xmm6, %xmm6 +; AVX2-NEXT: vpcmpeqd %ymm6, %ymm4, %ymm6 +; AVX2-NEXT: vextracti128 $1, %ymm6, %xmm6 +; AVX2-NEXT: vpackssdw %xmm6, %xmm0, %xmm7 +; AVX2-NEXT: vpacksswb %xmm0, %xmm7, %xmm7 +; AVX2-NEXT: vpextrb $4, %xmm7, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_10 +; AVX2-NEXT: ## %bb.9: ## %cond.load13 +; AVX2-NEXT: vmovq {{.*#+}} xmm7 = mem[0],zero +; AVX2-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm1[2,3,4,5,6,7] +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_10: ## %else14 +; AVX2-NEXT: vpackssdw %xmm6, %xmm0, %xmm6 +; AVX2-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 +; AVX2-NEXT: vpextrb $5, %xmm6, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_12 +; AVX2-NEXT: ## %bb.11: ## %cond.load17 +; AVX2-NEXT: vmovhpd {{.*#+}} xmm6 = xmm1[0],mem[0] +; AVX2-NEXT: vblendpd {{.*#+}} ymm1 = ymm6[0,1],ymm1[2,3] +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_12: ## %else18 +; AVX2-NEXT: vxorpd %xmm6, %xmm6, %xmm6 +; AVX2-NEXT: vpcmpeqd %ymm6, %ymm4, %ymm4 +; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm4 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm6 +; AVX2-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 +; AVX2-NEXT: vpextrb $6, %xmm6, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_14 +; AVX2-NEXT: ## %bb.13: ## %cond.load21 +; AVX2-NEXT: vextractf128 $1, %ymm1, %xmm6 +; AVX2-NEXT: vmovlpd {{.*#+}} xmm6 = mem[0],xmm6[1] +; AVX2-NEXT: vinsertf128 $1, %xmm6, %ymm1, %ymm1 +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_14: ## %else22 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpextrb $7, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_16 +; AVX2-NEXT: ## %bb.15: ## %cond.load25 +; AVX2-NEXT: vextractf128 $1, %ymm1, %xmm4 +; AVX2-NEXT: vmovhpd {{.*#+}} xmm4 = xmm4[0],mem[0] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_16: ## %else26 +; AVX2-NEXT: vxorpd %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm6 +; AVX2-NEXT: vpacksswb %xmm6, %xmm0, %xmm6 +; AVX2-NEXT: vpextrb $8, %xmm6, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_18 +; AVX2-NEXT: ## %bb.17: ## %cond.load29 +; AVX2-NEXT: vmovq {{.*#+}} xmm6 = mem[0],zero +; AVX2-NEXT: vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm2[2,3,4,5,6,7] +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_18: ## %else30 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $9, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_20 +; AVX2-NEXT: ## %bb.19: ## %cond.load33 +; AVX2-NEXT: vmovhpd {{.*#+}} xmm4 = xmm2[0],mem[0] +; AVX2-NEXT: vblendpd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3] +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_20: ## %else34 +; AVX2-NEXT: vxorpd %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm6 +; AVX2-NEXT: vpacksswb %xmm6, %xmm0, %xmm6 +; AVX2-NEXT: vpextrb $10, %xmm6, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_22 +; AVX2-NEXT: ## %bb.21: ## %cond.load37 +; AVX2-NEXT: vextractf128 $1, %ymm2, %xmm6 +; AVX2-NEXT: vmovlpd {{.*#+}} xmm6 = mem[0],xmm6[1] +; AVX2-NEXT: vinsertf128 $1, %xmm6, %ymm2, %ymm2 +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_22: ## %else38 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $11, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_24 +; AVX2-NEXT: ## %bb.23: ## %cond.load41 +; AVX2-NEXT: vextractf128 $1, %ymm2, %xmm4 +; AVX2-NEXT: vmovhpd {{.*#+}} xmm4 = xmm4[0],mem[0] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm2, %ymm2 +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_24: ## %else42 +; AVX2-NEXT: vxorpd %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 +; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm4 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm6 +; AVX2-NEXT: vpacksswb %xmm6, %xmm0, %xmm6 +; AVX2-NEXT: vpextrb $12, %xmm6, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_26 +; AVX2-NEXT: ## %bb.25: ## %cond.load45 +; AVX2-NEXT: vmovq {{.*#+}} xmm6 = mem[0],zero +; AVX2-NEXT: vpblendd {{.*#+}} ymm3 = ymm6[0,1],ymm3[2,3,4,5,6,7] +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_26: ## %else46 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $13, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_28 +; AVX2-NEXT: ## %bb.27: ## %cond.load49 +; AVX2-NEXT: vmovhpd {{.*#+}} xmm4 = xmm3[0],mem[0] +; AVX2-NEXT: vblendpd {{.*#+}} ymm3 = ymm4[0,1],ymm3[2,3] +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_28: ## %else50 +; AVX2-NEXT: vxorpd %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 +; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm4 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 +; AVX2-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX2-NEXT: vpextrb $14, %xmm5, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_30 +; AVX2-NEXT: ## %bb.29: ## %cond.load53 +; AVX2-NEXT: vextractf128 $1, %ymm3, %xmm5 +; AVX2-NEXT: vmovlpd {{.*#+}} xmm5 = mem[0],xmm5[1] +; AVX2-NEXT: vinsertf128 $1, %xmm5, %ymm3, %ymm3 +; AVX2-NEXT: addq $8, %rdi +; AVX2-NEXT: LBB3_30: ## %else54 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $15, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB3_32 +; AVX2-NEXT: ## %bb.31: ## %cond.load57 +; AVX2-NEXT: vextractf128 $1, %ymm3, %xmm4 +; AVX2-NEXT: vmovhpd {{.*#+}} xmm4 = xmm4[0],mem[0] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm3, %ymm3 +; AVX2-NEXT: LBB3_32: ## %else58 +; AVX2-NEXT: retq +; +; AVX512F-LABEL: expandload_v16f64_v16i32: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: vextracti64x4 $1, %zmm2, %ymm3 +; AVX512F-NEXT: vptestnmd %zmm3, %zmm3, %k1 +; AVX512F-NEXT: vptestnmd %zmm2, %zmm2, %k2 +; AVX512F-NEXT: vexpandpd (%rdi), %zmm0 {%k2} +; AVX512F-NEXT: kmovw %k2, %eax +; AVX512F-NEXT: movzbl %al, %eax +; AVX512F-NEXT: movl %eax, %ecx +; AVX512F-NEXT: shrl %ecx +; AVX512F-NEXT: andl $-43, %ecx +; AVX512F-NEXT: subl %ecx, %eax +; AVX512F-NEXT: movl %eax, %ecx +; AVX512F-NEXT: andl $858993459, %ecx ## imm = 0x33333333 +; AVX512F-NEXT: shrl $2, %eax +; AVX512F-NEXT: andl $858993459, %eax ## imm = 0x33333333 +; AVX512F-NEXT: addl %ecx, %eax +; AVX512F-NEXT: movl %eax, %ecx +; AVX512F-NEXT: shrl $4, %ecx +; AVX512F-NEXT: addl %eax, %ecx +; AVX512F-NEXT: andl $252645135, %ecx ## imm = 0xF0F0F0F +; AVX512F-NEXT: imull $16843009, %ecx, %eax ## imm = 0x1010101 +; AVX512F-NEXT: shrl $24, %eax +; AVX512F-NEXT: vexpandpd (%rdi,%rax,8), %zmm1 {%k1} +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v16f64_v16i32: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vextracti64x4 $1, %zmm2, %ymm3 +; AVX512VLBW-NEXT: vptestnmd %ymm3, %ymm3, %k1 +; AVX512VLBW-NEXT: vptestnmd %ymm2, %ymm2, %k2 +; AVX512VLBW-NEXT: vexpandpd (%rdi), %zmm0 {%k2} +; AVX512VLBW-NEXT: kmovd %k2, %eax +; AVX512VLBW-NEXT: movzbl %al, %eax +; AVX512VLBW-NEXT: movl %eax, %ecx +; AVX512VLBW-NEXT: shrl %ecx +; AVX512VLBW-NEXT: andl $-43, %ecx +; AVX512VLBW-NEXT: subl %ecx, %eax +; AVX512VLBW-NEXT: movl %eax, %ecx +; AVX512VLBW-NEXT: andl $858993459, %ecx ## imm = 0x33333333 +; AVX512VLBW-NEXT: shrl $2, %eax +; AVX512VLBW-NEXT: andl $858993459, %eax ## imm = 0x33333333 +; AVX512VLBW-NEXT: addl %ecx, %eax +; AVX512VLBW-NEXT: movl %eax, %ecx +; AVX512VLBW-NEXT: shrl $4, %ecx +; AVX512VLBW-NEXT: addl %eax, %ecx +; AVX512VLBW-NEXT: andl $252645135, %ecx ## imm = 0xF0F0F0F +; AVX512VLBW-NEXT: imull $16843009, %ecx, %eax ## imm = 0x1010101 +; AVX512VLBW-NEXT: shrl $24, %eax +; AVX512VLBW-NEXT: vexpandpd (%rdi,%rax,8), %zmm1 {%k1} +; AVX512VLBW-NEXT: retq %mask = icmp eq <16 x i32> %trigger, zeroinitializer %res = call <16 x double> @llvm.masked.expandload.v16f64(double* %base, <16 x i1> %mask, <16 x double> %src0) ret <16 x double> %res @@ -426,552 +1215,1709 @@ define <16 x double> @expandload_v16f64_v16i32(double* %base, <16 x double> %src ; define <2 x float> @expandload_v2f32_v2i1(float* %base, <2 x float> %src0, <2 x i32> %trigger) { -; HSW-LABEL: expandload_v2f32_v2i1: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; HSW-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $0, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB4_2 -; HSW-NEXT: # %bb.1: # %cond.load -; HSW-NEXT: vmovd {{.*#+}} xmm2 = mem[0],zero,zero,zero -; HSW-NEXT: vpblendd {{.*#+}} xmm0 = xmm2[0],xmm0[1,2,3] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB4_2: # %else -; HSW-NEXT: vpextrb $8, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB4_4 -; HSW-NEXT: # %bb.3: # %cond.load1 -; HSW-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3] -; HSW-NEXT: .LBB4_4: # %else2 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v2f32_v2i1: -; SKX: # %bb.0: -; SKX-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; SKX-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; SKX-NEXT: vptestnmq %xmm1, %xmm1, %k1 -; SKX-NEXT: vexpandps (%rdi), %xmm0 {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v2f32_v2i1: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; KNL-NEXT: vptestnmq %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftlw $14, %k0, %k0 -; KNL-NEXT: kshiftrw $14, %k0, %k1 -; KNL-NEXT: vexpandps (%rdi), %zmm0 {%k1} -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v2f32_v2i1: +; SSE2: ## %bb.0: +; SSE2-NEXT: pand {{.*}}(%rip), %xmm1 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqd %xmm1, %xmm2 +; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,0,3,2] +; SSE2-NEXT: pand %xmm2, %xmm1 +; SSE2-NEXT: movd %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB4_2 +; SSE2-NEXT: ## %bb.1: ## %cond.load +; SSE2-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm0 = xmm2[0],xmm0[1,2,3] +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB4_2: ## %else +; SSE2-NEXT: pextrw $4, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB4_4 +; SSE2-NEXT: ## %bb.3: ## %cond.load1 +; SSE2-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[0,0] +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3] +; SSE2-NEXT: movaps %xmm1, %xmm0 +; SSE2-NEXT: LBB4_4: ## %else2 +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v2f32_v2i1: +; SSE42: ## %bb.0: +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; SSE42-NEXT: pcmpeqq %xmm2, %xmm1 +; SSE42-NEXT: pextrb $0, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB4_2 +; SSE42-NEXT: ## %bb.1: ## %cond.load +; SSE42-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero +; SSE42-NEXT: pblendw {{.*#+}} xmm0 = xmm2[0,1],xmm0[2,3,4,5,6,7] +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB4_2: ## %else +; SSE42-NEXT: pextrb $8, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB4_4 +; SSE42-NEXT: ## %bb.3: ## %cond.load1 +; SSE42-NEXT: insertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3] +; SSE42-NEXT: LBB4_4: ## %else2 +; SSE42-NEXT: retq +; +; AVX1-LABEL: expandload_v2f32_v2i1: +; AVX1: ## %bb.0: +; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; AVX1-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm1 +; AVX1-NEXT: vpextrb $0, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB4_2 +; AVX1-NEXT: ## %bb.1: ## %cond.load +; AVX1-NEXT: vmovd {{.*#+}} xmm2 = mem[0],zero,zero,zero +; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1],xmm0[2,3,4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB4_2: ## %else +; AVX1-NEXT: vpextrb $8, %xmm1, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB4_4 +; AVX1-NEXT: ## %bb.3: ## %cond.load1 +; AVX1-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3] +; AVX1-NEXT: LBB4_4: ## %else2 +; AVX1-NEXT: retq +; +; AVX2-LABEL: expandload_v2f32_v2i1: +; AVX2: ## %bb.0: +; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX2-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm1 +; AVX2-NEXT: vpextrb $0, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB4_2 +; AVX2-NEXT: ## %bb.1: ## %cond.load +; AVX2-NEXT: vmovd {{.*#+}} xmm2 = mem[0],zero,zero,zero +; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm2[0],xmm0[1,2,3] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB4_2: ## %else +; AVX2-NEXT: vpextrb $8, %xmm1, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB4_4 +; AVX2-NEXT: ## %bb.3: ## %cond.load1 +; AVX2-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3] +; AVX2-NEXT: LBB4_4: ## %else2 +; AVX2-NEXT: retq +; +; AVX512F-LABEL: expandload_v2f32_v2i1: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX512F-NEXT: vptestnmq %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftlw $14, %k0, %k0 +; AVX512F-NEXT: kshiftrw $14, %k0, %k1 +; AVX512F-NEXT: vexpandps (%rdi), %zmm0 {%k1} +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v2f32_v2i1: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512VLBW-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] +; AVX512VLBW-NEXT: vptestnmq %xmm1, %xmm1, %k1 +; AVX512VLBW-NEXT: vexpandps (%rdi), %xmm0 {%k1} +; AVX512VLBW-NEXT: retq %mask = icmp eq <2 x i32> %trigger, zeroinitializer %res = call <2 x float> @llvm.masked.expandload.v2f32(float* %base, <2 x i1> %mask, <2 x float> %src0) ret <2 x float> %res } define <4 x float> @expandload_v4f32_const(float* %base, <4 x float> %src0) { -; HSW-LABEL: expandload_v4f32_const: -; HSW: # %bb.0: # %cond.load -; HSW-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; HSW-NEXT: vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3] -; HSW-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3] -; HSW-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v4f32_const: -; SKX: # %bb.0: -; SKX-NEXT: movb $7, %al -; SKX-NEXT: kmovd %eax, %k1 -; SKX-NEXT: vexpandps (%rdi), %xmm0 {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v4f32_const: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; KNL-NEXT: movw $7, %ax -; KNL-NEXT: kmovw %eax, %k1 -; KNL-NEXT: vexpandps (%rdi), %zmm0 {%k1} -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v4f32_const: +; SSE2: ## %bb.0: ## %cond.load +; SSE2-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm2[0,0] +; SSE2-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,0],xmm0[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[0,2] +; SSE2-NEXT: movaps %xmm1, %xmm0 +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v4f32_const: +; SSE42: ## %bb.0: ## %cond.load +; SSE42-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero +; SSE42-NEXT: blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3] +; SSE42-NEXT: insertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3] +; SSE42-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: expandload_v4f32_const: +; AVX1OR2: ## %bb.0: ## %cond.load +; AVX1OR2-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero +; AVX1OR2-NEXT: vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3] +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3] +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: expandload_v4f32_const: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 +; AVX512F-NEXT: movw $7, %ax +; AVX512F-NEXT: kmovw %eax, %k1 +; AVX512F-NEXT: vexpandps (%rdi), %zmm0 {%k1} +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v4f32_const: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: movb $7, %al +; AVX512VLBW-NEXT: kmovd %eax, %k1 +; AVX512VLBW-NEXT: vexpandps (%rdi), %xmm0 {%k1} +; AVX512VLBW-NEXT: retq %res = call <4 x float> @llvm.masked.expandload.v4f32(float* %base, <4 x i1> , <4 x float> %src0) ret <4 x float>%res } define <16 x float> @expandload_v16f32_const(float* %base, <16 x float> %src0) { -; HSW-LABEL: expandload_v16f32_const: -; HSW: # %bb.0: # %cond.load -; HSW-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero -; HSW-NEXT: vblendps {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7] -; HSW-NEXT: vinsertps {{.*#+}} xmm2 = xmm0[0],mem[0],xmm0[2,3] -; HSW-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3] -; HSW-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0] -; HSW-NEXT: vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7] -; HSW-NEXT: vmovsd {{.*#+}} xmm2 = mem[0],zero -; HSW-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3] -; HSW-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 -; HSW-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero -; HSW-NEXT: vblendps {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7] -; HSW-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm3 -; HSW-NEXT: vblendps {{.*#+}} xmm2 = xmm2[0],xmm3[1,2,3] -; HSW-NEXT: vinsertps {{.*#+}} xmm3 = xmm1[0],mem[0],xmm1[2,3] -; HSW-NEXT: vinsertps {{.*#+}} xmm3 = xmm3[0,1],mem[0],xmm3[3] -; HSW-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3] -; HSW-NEXT: vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7] -; HSW-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3] -; HSW-NEXT: vinsertf128 $1, %xmm2, %ymm1, %ymm1 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v16f32_const: -; SKX: # %bb.0: -; SKX-NEXT: movw $30719, %ax # imm = 0x77FF -; SKX-NEXT: kmovd %eax, %k1 -; SKX-NEXT: vexpandps (%rdi), %zmm0 {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v16f32_const: -; KNL: # %bb.0: -; KNL-NEXT: movw $30719, %ax # imm = 0x77FF -; KNL-NEXT: kmovw %eax, %k1 -; KNL-NEXT: vexpandps (%rdi), %zmm0 {%k1} -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v16f32_const: +; SSE2: ## %bb.0: ## %cond.load +; SSE2-NEXT: movups (%rdi), %xmm0 +; SSE2-NEXT: movups 16(%rdi), %xmm1 +; SSE2-NEXT: movss {{.*#+}} xmm5 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm4 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,0],xmm5[0,0] +; SSE2-NEXT: movss {{.*#+}} xmm5 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,0],xmm2[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm5[0,2] +; SSE2-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm5 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,0],xmm2[0,0] +; SSE2-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,0],xmm3[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[2,0],xmm2[0,2] +; SSE2-NEXT: movaps %xmm4, %xmm2 +; SSE2-NEXT: movaps %xmm5, %xmm3 +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v16f32_const: +; SSE42: ## %bb.0: ## %cond.load +; SSE42-NEXT: movups (%rdi), %xmm0 +; SSE42-NEXT: movups 16(%rdi), %xmm1 +; SSE42-NEXT: movss {{.*#+}} xmm4 = mem[0],zero,zero,zero +; SSE42-NEXT: blendps {{.*#+}} xmm2 = xmm4[0],xmm2[1,2,3] +; SSE42-NEXT: insertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3] +; SSE42-NEXT: insertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3] +; SSE42-NEXT: movss {{.*#+}} xmm4 = mem[0],zero,zero,zero +; SSE42-NEXT: blendps {{.*#+}} xmm3 = xmm4[0],xmm3[1,2,3] +; SSE42-NEXT: insertps {{.*#+}} xmm3 = xmm3[0],mem[0],xmm3[2,3] +; SSE42-NEXT: insertps {{.*#+}} xmm3 = xmm3[0,1],mem[0],xmm3[3] +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: expandload_v16f32_const: +; AVX1OR2: ## %bb.0: ## %cond.load +; AVX1OR2-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero +; AVX1OR2-NEXT: vblendps {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7] +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm2 = xmm0[0],mem[0],xmm0[2,3] +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3] +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0] +; AVX1OR2-NEXT: vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7] +; AVX1OR2-NEXT: vmovsd {{.*#+}} xmm2 = mem[0],zero +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3] +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0] +; AVX1OR2-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; AVX1OR2-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero +; AVX1OR2-NEXT: vblendps {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7] +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm2 = xmm1[0],mem[0],xmm1[2,3] +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3] +; AVX1OR2-NEXT: vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm1[4,5,6,7] +; AVX1OR2-NEXT: vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero +; AVX1OR2-NEXT: vextractf128 $1, %ymm1, %xmm1 +; AVX1OR2-NEXT: vblendps {{.*#+}} xmm1 = xmm3[0],xmm1[1,2,3] +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[2,3] +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],mem[0],xmm1[3] +; AVX1OR2-NEXT: vinsertf128 $1, %xmm1, %ymm2, %ymm1 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: expandload_v16f32_const: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: movw $30719, %ax ## imm = 0x77FF +; AVX512F-NEXT: kmovw %eax, %k1 +; AVX512F-NEXT: vexpandps (%rdi), %zmm0 {%k1} +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v16f32_const: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: movw $30719, %ax ## imm = 0x77FF +; AVX512VLBW-NEXT: kmovd %eax, %k1 +; AVX512VLBW-NEXT: vexpandps (%rdi), %zmm0 {%k1} +; AVX512VLBW-NEXT: retq %res = call <16 x float> @llvm.masked.expandload.v16f32(float* %base, <16 x i1> , <16 x float> %src0) ret <16 x float>%res } define <16 x float> @expandload_v16f32_const_undef(float* %base) { -; HSW-LABEL: expandload_v16f32_const_undef: -; HSW: # %bb.0: # %cond.load -; HSW-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; HSW-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] -; HSW-NEXT: vinsertf128 $1, 44(%rdi), %ymm0, %ymm1 -; HSW-NEXT: vmovups (%rdi), %ymm0 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v16f32_const_undef: -; SKX: # %bb.0: -; SKX-NEXT: movw $-2049, %ax # imm = 0xF7FF -; SKX-NEXT: kmovd %eax, %k1 -; SKX-NEXT: vexpandps (%rdi), %zmm0 {%k1} {z} -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v16f32_const_undef: -; KNL: # %bb.0: -; KNL-NEXT: movw $-2049, %ax # imm = 0xF7FF -; KNL-NEXT: kmovw %eax, %k1 -; KNL-NEXT: vexpandps (%rdi), %zmm0 {%k1} {z} -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v16f32_const_undef: +; SSE2: ## %bb.0: ## %cond.load +; SSE2-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; SSE2-NEXT: movsd {{.*#+}} xmm2 = mem[0],zero +; SSE2-NEXT: movlhps {{.*#+}} xmm2 = xmm2[0],xmm0[0] +; SSE2-NEXT: movups (%rdi), %xmm0 +; SSE2-NEXT: movups 16(%rdi), %xmm1 +; SSE2-NEXT: movups 44(%rdi), %xmm3 +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v16f32_const_undef: +; SSE42: ## %bb.0: ## %cond.load +; SSE42-NEXT: movsd {{.*#+}} xmm2 = mem[0],zero +; SSE42-NEXT: insertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3] +; SSE42-NEXT: movups (%rdi), %xmm0 +; SSE42-NEXT: movups 16(%rdi), %xmm1 +; SSE42-NEXT: movups 44(%rdi), %xmm3 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: expandload_v16f32_const_undef: +; AVX1OR2: ## %bb.0: ## %cond.load +; AVX1OR2-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero +; AVX1OR2-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] +; AVX1OR2-NEXT: vinsertf128 $1, 44(%rdi), %ymm0, %ymm1 +; AVX1OR2-NEXT: vmovups (%rdi), %ymm0 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: expandload_v16f32_const_undef: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: movw $-2049, %ax ## imm = 0xF7FF +; AVX512F-NEXT: kmovw %eax, %k1 +; AVX512F-NEXT: vexpandps (%rdi), %zmm0 {%k1} {z} +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v16f32_const_undef: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: movw $-2049, %ax ## imm = 0xF7FF +; AVX512VLBW-NEXT: kmovd %eax, %k1 +; AVX512VLBW-NEXT: vexpandps (%rdi), %zmm0 {%k1} {z} +; AVX512VLBW-NEXT: retq %res = call <16 x float> @llvm.masked.expandload.v16f32(float* %base, <16 x i1> , <16 x float> undef) ret <16 x float>%res } define <32 x float> @expandload_v32f32_v32i32(float* %base, <32 x float> %src0, <32 x i32> %trigger) { -; HSW-LABEL: expandload_v32f32_v32i32: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm8, %xmm8, %xmm8 -; HSW-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm9 -; HSW-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 -; HSW-NEXT: vpextrb $0, %xmm9, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_2 -; HSW-NEXT: # %bb.1: # %cond.load -; HSW-NEXT: vmovd {{.*#+}} xmm9 = mem[0],zero,zero,zero -; HSW-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0],ymm0[1,2,3,4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_2: # %else -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 -; HSW-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 -; HSW-NEXT: vpextrb $1, %xmm8, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_4 -; HSW-NEXT: # %bb.3: # %cond.load1 -; HSW-NEXT: vinsertps {{.*#+}} xmm8 = xmm0[0],mem[0],xmm0[2,3] -; HSW-NEXT: vblendps {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm0[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_4: # %else2 -; HSW-NEXT: vxorps %xmm8, %xmm8, %xmm8 -; HSW-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm9 -; HSW-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 -; HSW-NEXT: vpextrb $2, %xmm9, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_6 -; HSW-NEXT: # %bb.5: # %cond.load5 -; HSW-NEXT: vinsertps {{.*#+}} xmm9 = xmm0[0,1],mem[0],xmm0[3] -; HSW-NEXT: vblendps {{.*#+}} ymm0 = ymm9[0,1,2,3],ymm0[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_6: # %else6 -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 -; HSW-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 -; HSW-NEXT: vpextrb $3, %xmm8, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_8 -; HSW-NEXT: # %bb.7: # %cond.load9 -; HSW-NEXT: vinsertps {{.*#+}} xmm8 = xmm0[0,1,2],mem[0] -; HSW-NEXT: vblendps {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm0[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_8: # %else10 -; HSW-NEXT: vxorps %xmm8, %xmm8, %xmm8 -; HSW-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 -; HSW-NEXT: vextracti128 $1, %ymm8, %xmm8 -; HSW-NEXT: vpackssdw %xmm8, %xmm0, %xmm9 -; HSW-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 -; HSW-NEXT: vpextrb $4, %xmm9, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_10 -; HSW-NEXT: # %bb.9: # %cond.load13 -; HSW-NEXT: vmovss {{.*#+}} xmm9 = mem[0],zero,zero,zero -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm10 -; HSW-NEXT: vblendps {{.*#+}} xmm9 = xmm9[0],xmm10[1,2,3] -; HSW-NEXT: vinsertf128 $1, %xmm9, %ymm0, %ymm0 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_10: # %else14 -; HSW-NEXT: vpackssdw %xmm8, %xmm0, %xmm8 -; HSW-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 -; HSW-NEXT: vpextrb $5, %xmm8, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_12 -; HSW-NEXT: # %bb.11: # %cond.load17 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm8 -; HSW-NEXT: vinsertps {{.*#+}} xmm8 = xmm8[0],mem[0],xmm8[2,3] -; HSW-NEXT: vinsertf128 $1, %xmm8, %ymm0, %ymm0 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_12: # %else18 -; HSW-NEXT: vxorps %xmm8, %xmm8, %xmm8 -; HSW-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm4 -; HSW-NEXT: vextracti128 $1, %ymm4, %xmm8 -; HSW-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpextrb $6, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_14 -; HSW-NEXT: # %bb.13: # %cond.load21 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm4 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1],mem[0],xmm4[3] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm0, %ymm0 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_14: # %else22 -; HSW-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpextrb $7, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_16 -; HSW-NEXT: # %bb.15: # %cond.load25 -; HSW-NEXT: vextractf128 $1, %ymm0, %xmm4 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm0, %ymm0 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_16: # %else26 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm8 -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $8, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_18 -; HSW-NEXT: # %bb.17: # %cond.load29 -; HSW-NEXT: vmovd {{.*#+}} xmm4 = mem[0],zero,zero,zero -; HSW-NEXT: vpblendd {{.*#+}} ymm1 = ymm4[0],ymm1[1,2,3,4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_18: # %else30 -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $9, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_20 -; HSW-NEXT: # %bb.19: # %cond.load33 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm1[0],mem[0],xmm1[2,3] -; HSW-NEXT: vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_20: # %else34 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm8 -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $10, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_22 -; HSW-NEXT: # %bb.21: # %cond.load37 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm1[0,1],mem[0],xmm1[3] -; HSW-NEXT: vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_22: # %else38 -; HSW-NEXT: vpackssdw %xmm0, %xmm8, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $11, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_24 -; HSW-NEXT: # %bb.23: # %cond.load41 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm1[0,1,2],mem[0] -; HSW-NEXT: vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_24: # %else42 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 -; HSW-NEXT: vextracti128 $1, %ymm4, %xmm8 -; HSW-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $12, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_26 -; HSW-NEXT: # %bb.25: # %cond.load45 -; HSW-NEXT: vmovss {{.*#+}} xmm9 = mem[0],zero,zero,zero -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm4 -; HSW-NEXT: vblendps {{.*#+}} xmm4 = xmm9[0],xmm4[1,2,3] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_26: # %else46 -; HSW-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $13, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_28 -; HSW-NEXT: # %bb.27: # %cond.load49 -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm4 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0],mem[0],xmm4[2,3] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_28: # %else50 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 -; HSW-NEXT: vextracti128 $1, %ymm4, %xmm4 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 -; HSW-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 -; HSW-NEXT: vpextrb $14, %xmm5, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_30 -; HSW-NEXT: # %bb.29: # %cond.load53 -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm5 -; HSW-NEXT: vinsertps {{.*#+}} xmm5 = xmm5[0,1],mem[0],xmm5[3] -; HSW-NEXT: vinsertf128 $1, %xmm5, %ymm1, %ymm1 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_30: # %else54 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $15, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_32 -; HSW-NEXT: # %bb.31: # %cond.load57 -; HSW-NEXT: vextractf128 $1, %ymm1, %xmm4 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_32: # %else58 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm6, %ymm4 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm5 -; HSW-NEXT: vpacksswb %xmm0, %xmm5, %xmm5 -; HSW-NEXT: vpextrb $0, %xmm5, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_34 -; HSW-NEXT: # %bb.33: # %cond.load61 -; HSW-NEXT: vmovd {{.*#+}} xmm5 = mem[0],zero,zero,zero -; HSW-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_34: # %else62 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpextrb $1, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_36 -; HSW-NEXT: # %bb.35: # %cond.load65 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm2[0],mem[0],xmm2[2,3] -; HSW-NEXT: vblendps {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_36: # %else66 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm6, %ymm4 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm5 -; HSW-NEXT: vpacksswb %xmm0, %xmm5, %xmm5 -; HSW-NEXT: vpextrb $2, %xmm5, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_38 -; HSW-NEXT: # %bb.37: # %cond.load69 -; HSW-NEXT: vinsertps {{.*#+}} xmm5 = xmm2[0,1],mem[0],xmm2[3] -; HSW-NEXT: vblendps {{.*#+}} ymm2 = ymm5[0,1,2,3],ymm2[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_38: # %else70 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpextrb $3, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_40 -; HSW-NEXT: # %bb.39: # %cond.load73 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm2[0,1,2],mem[0] -; HSW-NEXT: vblendps {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_40: # %else74 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm6, %ymm4 -; HSW-NEXT: vextracti128 $1, %ymm4, %xmm4 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 -; HSW-NEXT: vpacksswb %xmm0, %xmm5, %xmm5 -; HSW-NEXT: vpextrb $4, %xmm5, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_42 -; HSW-NEXT: # %bb.41: # %cond.load77 -; HSW-NEXT: vmovss {{.*#+}} xmm8 = mem[0],zero,zero,zero -; HSW-NEXT: vextractf128 $1, %ymm2, %xmm5 -; HSW-NEXT: vblendps {{.*#+}} xmm5 = xmm8[0],xmm5[1,2,3] -; HSW-NEXT: vinsertf128 $1, %xmm5, %ymm2, %ymm2 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_42: # %else78 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpextrb $5, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_44 -; HSW-NEXT: # %bb.43: # %cond.load81 -; HSW-NEXT: vextractf128 $1, %ymm2, %xmm4 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0],mem[0],xmm4[2,3] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm2, %ymm2 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_44: # %else82 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm6, %ymm4 -; HSW-NEXT: vextracti128 $1, %ymm4, %xmm4 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 -; HSW-NEXT: vpacksswb %xmm0, %xmm5, %xmm5 -; HSW-NEXT: vpextrb $6, %xmm5, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_46 -; HSW-NEXT: # %bb.45: # %cond.load85 -; HSW-NEXT: vextractf128 $1, %ymm2, %xmm5 -; HSW-NEXT: vinsertps {{.*#+}} xmm5 = xmm5[0,1],mem[0],xmm5[3] -; HSW-NEXT: vinsertf128 $1, %xmm5, %ymm2, %ymm2 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_46: # %else86 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpextrb $7, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_48 -; HSW-NEXT: # %bb.47: # %cond.load89 -; HSW-NEXT: vextractf128 $1, %ymm2, %xmm4 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm2, %ymm2 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_48: # %else90 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm7, %ymm4 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm5 -; HSW-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 -; HSW-NEXT: vpextrb $8, %xmm5, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_50 -; HSW-NEXT: # %bb.49: # %cond.load93 -; HSW-NEXT: vmovd {{.*#+}} xmm5 = mem[0],zero,zero,zero -; HSW-NEXT: vpblendd {{.*#+}} ymm3 = ymm5[0],ymm3[1,2,3,4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_50: # %else94 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $9, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_52 -; HSW-NEXT: # %bb.51: # %cond.load97 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm3[0],mem[0],xmm3[2,3] -; HSW-NEXT: vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_52: # %else98 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm7, %ymm4 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm5 -; HSW-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 -; HSW-NEXT: vpextrb $10, %xmm5, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_54 -; HSW-NEXT: # %bb.53: # %cond.load101 -; HSW-NEXT: vinsertps {{.*#+}} xmm5 = xmm3[0,1],mem[0],xmm3[3] -; HSW-NEXT: vblendps {{.*#+}} ymm3 = ymm5[0,1,2,3],ymm3[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_54: # %else102 -; HSW-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $11, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_56 -; HSW-NEXT: # %bb.55: # %cond.load105 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm3[0,1,2],mem[0] -; HSW-NEXT: vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7] -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_56: # %else106 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm7, %ymm4 -; HSW-NEXT: vextracti128 $1, %ymm4, %xmm4 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 -; HSW-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 -; HSW-NEXT: vpextrb $12, %xmm5, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_58 -; HSW-NEXT: # %bb.57: # %cond.load109 -; HSW-NEXT: vmovss {{.*#+}} xmm5 = mem[0],zero,zero,zero -; HSW-NEXT: vextractf128 $1, %ymm3, %xmm6 -; HSW-NEXT: vblendps {{.*#+}} xmm5 = xmm5[0],xmm6[1,2,3] -; HSW-NEXT: vinsertf128 $1, %xmm5, %ymm3, %ymm3 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_58: # %else110 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $13, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_60 -; HSW-NEXT: # %bb.59: # %cond.load113 -; HSW-NEXT: vextractf128 $1, %ymm3, %xmm4 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0],mem[0],xmm4[2,3] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm3, %ymm3 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_60: # %else114 -; HSW-NEXT: vxorps %xmm4, %xmm4, %xmm4 -; HSW-NEXT: vpcmpeqd %ymm4, %ymm7, %ymm4 -; HSW-NEXT: vextracti128 $1, %ymm4, %xmm4 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 -; HSW-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 -; HSW-NEXT: vpextrb $14, %xmm5, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_62 -; HSW-NEXT: # %bb.61: # %cond.load117 -; HSW-NEXT: vextractf128 $1, %ymm3, %xmm5 -; HSW-NEXT: vinsertps {{.*#+}} xmm5 = xmm5[0,1],mem[0],xmm5[3] -; HSW-NEXT: vinsertf128 $1, %xmm5, %ymm3, %ymm3 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB8_62: # %else118 -; HSW-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 -; HSW-NEXT: vpextrb $15, %xmm4, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB8_64 -; HSW-NEXT: # %bb.63: # %cond.load121 -; HSW-NEXT: vextractf128 $1, %ymm3, %xmm4 -; HSW-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] -; HSW-NEXT: vinsertf128 $1, %xmm4, %ymm3, %ymm3 -; HSW-NEXT: .LBB8_64: # %else122 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v32f32_v32i32: -; SKX: # %bb.0: -; SKX-NEXT: vptestnmd %zmm3, %zmm3, %k1 -; SKX-NEXT: vptestnmd %zmm2, %zmm2, %k2 -; SKX-NEXT: kmovw %k2, %eax -; SKX-NEXT: popcntl %eax, %eax -; SKX-NEXT: vexpandps (%rdi,%rax,4), %zmm1 {%k1} -; SKX-NEXT: vexpandps (%rdi), %zmm0 {%k2} -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v32f32_v32i32: -; KNL: # %bb.0: -; KNL-NEXT: vptestnmd %zmm3, %zmm3, %k1 -; KNL-NEXT: vptestnmd %zmm2, %zmm2, %k2 -; KNL-NEXT: kmovw %k2, %eax -; KNL-NEXT: popcntl %eax, %eax -; KNL-NEXT: vexpandps (%rdi,%rax,4), %zmm1 {%k1} -; KNL-NEXT: vexpandps (%rdi), %zmm0 {%k2} -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v32f32_v32i32: +; SSE2: ## %bb.0: +; SSE2-NEXT: movq %rdi, %rax +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE2-NEXT: pxor %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: movd %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_2 +; SSE2-NEXT: ## %bb.1: ## %cond.load +; SSE2-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm0 = xmm10[0],xmm0[1,2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_2: ## %else +; SSE2-NEXT: pextrw $2, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_4 +; SSE2-NEXT: ## %bb.3: ## %cond.load1 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm0[0,0] +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[2,0],xmm0[2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: movaps %xmm8, %xmm0 +; SSE2-NEXT: LBB8_4: ## %else2 +; SSE2-NEXT: xorps %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: pextrw $4, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_6 +; SSE2-NEXT: ## %bb.5: ## %cond.load5 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm0[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm8[0,2] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_6: ## %else6 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE2-NEXT: pextrw $6, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_8 +; SSE2-NEXT: ## %bb.7: ## %cond.load9 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm0[2,0] +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm9[2,0] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_8: ## %else10 +; SSE2-NEXT: xorps %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: movd %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_10 +; SSE2-NEXT: ## %bb.9: ## %cond.load13 +; SSE2-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm1 = xmm10[0],xmm1[1,2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_10: ## %else14 +; SSE2-NEXT: pextrw $2, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_12 +; SSE2-NEXT: ## %bb.11: ## %cond.load17 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm1[0,0] +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[2,0],xmm1[2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: movaps %xmm9, %xmm1 +; SSE2-NEXT: LBB8_12: ## %else18 +; SSE2-NEXT: xorps %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: pextrw $4, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_14 +; SSE2-NEXT: ## %bb.13: ## %cond.load21 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm1[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm9[0,2] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_14: ## %else22 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE2-NEXT: pextrw $6, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_16 +; SSE2-NEXT: ## %bb.15: ## %cond.load25 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm1[2,0] +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm8[2,0] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_16: ## %else26 +; SSE2-NEXT: xorps %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: movd %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_18 +; SSE2-NEXT: ## %bb.17: ## %cond.load29 +; SSE2-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm2 = xmm10[0],xmm2[1,2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_18: ## %else30 +; SSE2-NEXT: pextrw $2, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_20 +; SSE2-NEXT: ## %bb.19: ## %cond.load33 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm2[0,0] +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[2,0],xmm2[2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: movaps %xmm8, %xmm2 +; SSE2-NEXT: LBB8_20: ## %else34 +; SSE2-NEXT: xorps %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: pextrw $4, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_22 +; SSE2-NEXT: ## %bb.21: ## %cond.load37 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm2[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm8[0,2] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_22: ## %else38 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE2-NEXT: pextrw $6, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_24 +; SSE2-NEXT: ## %bb.23: ## %cond.load41 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm2[2,0] +; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm9[2,0] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_24: ## %else42 +; SSE2-NEXT: xorps %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: movd %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_26 +; SSE2-NEXT: ## %bb.25: ## %cond.load45 +; SSE2-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm3 = xmm10[0],xmm3[1,2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_26: ## %else46 +; SSE2-NEXT: pextrw $2, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_28 +; SSE2-NEXT: ## %bb.27: ## %cond.load49 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm3[0,0] +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[2,0],xmm3[2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: movaps %xmm9, %xmm3 +; SSE2-NEXT: LBB8_28: ## %else50 +; SSE2-NEXT: xorps %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: pextrw $4, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_30 +; SSE2-NEXT: ## %bb.29: ## %cond.load53 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm3[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm9[0,2] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_30: ## %else54 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE2-NEXT: pextrw $6, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_32 +; SSE2-NEXT: ## %bb.31: ## %cond.load57 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm3[2,0] +; SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm8[2,0] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_32: ## %else58 +; SSE2-NEXT: xorps %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: movd %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_34 +; SSE2-NEXT: ## %bb.33: ## %cond.load61 +; SSE2-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm4 = xmm10[0],xmm4[1,2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_34: ## %else62 +; SSE2-NEXT: pextrw $2, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_36 +; SSE2-NEXT: ## %bb.35: ## %cond.load65 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm4[0,0] +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[2,0],xmm4[2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: movaps %xmm8, %xmm4 +; SSE2-NEXT: LBB8_36: ## %else66 +; SSE2-NEXT: xorps %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: pextrw $4, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_38 +; SSE2-NEXT: ## %bb.37: ## %cond.load69 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm4[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,1],xmm8[0,2] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_38: ## %else70 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE2-NEXT: pextrw $6, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_40 +; SSE2-NEXT: ## %bb.39: ## %cond.load73 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm4[2,0] +; SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,1],xmm9[2,0] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_40: ## %else74 +; SSE2-NEXT: xorps %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: movd %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_42 +; SSE2-NEXT: ## %bb.41: ## %cond.load77 +; SSE2-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm5 = xmm10[0],xmm5[1,2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_42: ## %else78 +; SSE2-NEXT: pextrw $2, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_44 +; SSE2-NEXT: ## %bb.43: ## %cond.load81 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm5[0,0] +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[2,0],xmm5[2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: movaps %xmm9, %xmm5 +; SSE2-NEXT: LBB8_44: ## %else82 +; SSE2-NEXT: xorps %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: pextrw $4, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_46 +; SSE2-NEXT: ## %bb.45: ## %cond.load85 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm5[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,1],xmm9[0,2] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_46: ## %else86 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE2-NEXT: pextrw $6, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_48 +; SSE2-NEXT: ## %bb.47: ## %cond.load89 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm5[2,0] +; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,1],xmm8[2,0] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_48: ## %else90 +; SSE2-NEXT: xorps %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: movd %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_50 +; SSE2-NEXT: ## %bb.49: ## %cond.load93 +; SSE2-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm6 = xmm10[0],xmm6[1,2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_50: ## %else94 +; SSE2-NEXT: pextrw $2, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_52 +; SSE2-NEXT: ## %bb.51: ## %cond.load97 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm6[0,0] +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[2,0],xmm6[2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: movaps %xmm8, %xmm6 +; SSE2-NEXT: LBB8_52: ## %else98 +; SSE2-NEXT: xorps %xmm8, %xmm8 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: pextrw $4, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_54 +; SSE2-NEXT: ## %bb.53: ## %cond.load101 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm6[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm6 = xmm6[0,1],xmm8[0,2] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_54: ## %else102 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE2-NEXT: pextrw $6, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_56 +; SSE2-NEXT: ## %bb.55: ## %cond.load105 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm6[2,0] +; SSE2-NEXT: shufps {{.*#+}} xmm6 = xmm6[0,1],xmm9[2,0] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_56: ## %else106 +; SSE2-NEXT: xorps %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE2-NEXT: movd %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_58 +; SSE2-NEXT: ## %bb.57: ## %cond.load109 +; SSE2-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm7 = xmm10[0],xmm7[1,2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_58: ## %else110 +; SSE2-NEXT: pextrw $2, %xmm9, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_60 +; SSE2-NEXT: ## %bb.59: ## %cond.load113 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm7[0,0] +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[2,0],xmm7[2,3] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: movaps %xmm9, %xmm7 +; SSE2-NEXT: LBB8_60: ## %else114 +; SSE2-NEXT: xorps %xmm9, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE2-NEXT: pextrw $4, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_62 +; SSE2-NEXT: ## %bb.61: ## %cond.load117 +; SSE2-NEXT: movss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,0],xmm7[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,1],xmm9[0,2] +; SSE2-NEXT: addq $4, %rsi +; SSE2-NEXT: LBB8_62: ## %else118 +; SSE2-NEXT: pextrw $6, %xmm8, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB8_64 +; SSE2-NEXT: ## %bb.63: ## %cond.load121 +; SSE2-NEXT: movss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,0],xmm7[2,0] +; SSE2-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,1],xmm8[2,0] +; SSE2-NEXT: LBB8_64: ## %else122 +; SSE2-NEXT: movaps %xmm0, (%rax) +; SSE2-NEXT: movaps %xmm1, 16(%rax) +; SSE2-NEXT: movaps %xmm2, 32(%rax) +; SSE2-NEXT: movaps %xmm3, 48(%rax) +; SSE2-NEXT: movaps %xmm4, 64(%rax) +; SSE2-NEXT: movaps %xmm5, 80(%rax) +; SSE2-NEXT: movaps %xmm6, 96(%rax) +; SSE2-NEXT: movaps %xmm7, 112(%rax) +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v32f32_v32i32: +; SSE42: ## %bb.0: +; SSE42-NEXT: movq %rdi, %rax +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $0, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_2 +; SSE42-NEXT: ## %bb.1: ## %cond.load +; SSE42-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE42-NEXT: blendps {{.*#+}} xmm0 = xmm10[0],xmm0[1,2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_2: ## %else +; SSE42-NEXT: pextrb $4, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_4 +; SSE42-NEXT: ## %bb.3: ## %cond.load1 +; SSE42-NEXT: insertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_4: ## %else2 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $8, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_6 +; SSE42-NEXT: ## %bb.5: ## %cond.load5 +; SSE42-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_6: ## %else6 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE42-NEXT: pextrb $12, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_8 +; SSE42-NEXT: ## %bb.7: ## %cond.load9 +; SSE42-NEXT: insertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_8: ## %else10 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $0, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_10 +; SSE42-NEXT: ## %bb.9: ## %cond.load13 +; SSE42-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE42-NEXT: blendps {{.*#+}} xmm1 = xmm10[0],xmm1[1,2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_10: ## %else14 +; SSE42-NEXT: pextrb $4, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_12 +; SSE42-NEXT: ## %bb.11: ## %cond.load17 +; SSE42-NEXT: insertps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_12: ## %else18 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $8, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_14 +; SSE42-NEXT: ## %bb.13: ## %cond.load21 +; SSE42-NEXT: insertps {{.*#+}} xmm1 = xmm1[0,1],mem[0],xmm1[3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_14: ## %else22 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE42-NEXT: pextrb $12, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_16 +; SSE42-NEXT: ## %bb.15: ## %cond.load25 +; SSE42-NEXT: insertps {{.*#+}} xmm1 = xmm1[0,1,2],mem[0] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_16: ## %else26 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $0, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_18 +; SSE42-NEXT: ## %bb.17: ## %cond.load29 +; SSE42-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE42-NEXT: blendps {{.*#+}} xmm2 = xmm10[0],xmm2[1,2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_18: ## %else30 +; SSE42-NEXT: pextrb $4, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_20 +; SSE42-NEXT: ## %bb.19: ## %cond.load33 +; SSE42-NEXT: insertps {{.*#+}} xmm2 = xmm2[0],mem[0],xmm2[2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_20: ## %else34 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $8, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_22 +; SSE42-NEXT: ## %bb.21: ## %cond.load37 +; SSE42-NEXT: insertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_22: ## %else38 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE42-NEXT: pextrb $12, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_24 +; SSE42-NEXT: ## %bb.23: ## %cond.load41 +; SSE42-NEXT: insertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_24: ## %else42 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $0, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_26 +; SSE42-NEXT: ## %bb.25: ## %cond.load45 +; SSE42-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE42-NEXT: blendps {{.*#+}} xmm3 = xmm10[0],xmm3[1,2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_26: ## %else46 +; SSE42-NEXT: pextrb $4, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_28 +; SSE42-NEXT: ## %bb.27: ## %cond.load49 +; SSE42-NEXT: insertps {{.*#+}} xmm3 = xmm3[0],mem[0],xmm3[2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_28: ## %else50 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $8, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_30 +; SSE42-NEXT: ## %bb.29: ## %cond.load53 +; SSE42-NEXT: insertps {{.*#+}} xmm3 = xmm3[0,1],mem[0],xmm3[3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_30: ## %else54 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE42-NEXT: pextrb $12, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_32 +; SSE42-NEXT: ## %bb.31: ## %cond.load57 +; SSE42-NEXT: insertps {{.*#+}} xmm3 = xmm3[0,1,2],mem[0] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_32: ## %else58 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $0, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_34 +; SSE42-NEXT: ## %bb.33: ## %cond.load61 +; SSE42-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE42-NEXT: blendps {{.*#+}} xmm4 = xmm10[0],xmm4[1,2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_34: ## %else62 +; SSE42-NEXT: pextrb $4, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_36 +; SSE42-NEXT: ## %bb.35: ## %cond.load65 +; SSE42-NEXT: insertps {{.*#+}} xmm4 = xmm4[0],mem[0],xmm4[2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_36: ## %else66 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $8, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_38 +; SSE42-NEXT: ## %bb.37: ## %cond.load69 +; SSE42-NEXT: insertps {{.*#+}} xmm4 = xmm4[0,1],mem[0],xmm4[3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_38: ## %else70 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE42-NEXT: pextrb $12, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_40 +; SSE42-NEXT: ## %bb.39: ## %cond.load73 +; SSE42-NEXT: insertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_40: ## %else74 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $0, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_42 +; SSE42-NEXT: ## %bb.41: ## %cond.load77 +; SSE42-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE42-NEXT: blendps {{.*#+}} xmm5 = xmm10[0],xmm5[1,2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_42: ## %else78 +; SSE42-NEXT: pextrb $4, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_44 +; SSE42-NEXT: ## %bb.43: ## %cond.load81 +; SSE42-NEXT: insertps {{.*#+}} xmm5 = xmm5[0],mem[0],xmm5[2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_44: ## %else82 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $8, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_46 +; SSE42-NEXT: ## %bb.45: ## %cond.load85 +; SSE42-NEXT: insertps {{.*#+}} xmm5 = xmm5[0,1],mem[0],xmm5[3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_46: ## %else86 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 +; SSE42-NEXT: pextrb $12, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_48 +; SSE42-NEXT: ## %bb.47: ## %cond.load89 +; SSE42-NEXT: insertps {{.*#+}} xmm5 = xmm5[0,1,2],mem[0] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_48: ## %else90 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $0, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_50 +; SSE42-NEXT: ## %bb.49: ## %cond.load93 +; SSE42-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE42-NEXT: blendps {{.*#+}} xmm6 = xmm10[0],xmm6[1,2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_50: ## %else94 +; SSE42-NEXT: pextrb $4, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_52 +; SSE42-NEXT: ## %bb.51: ## %cond.load97 +; SSE42-NEXT: insertps {{.*#+}} xmm6 = xmm6[0],mem[0],xmm6[2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_52: ## %else98 +; SSE42-NEXT: pxor %xmm8, %xmm8 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $8, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_54 +; SSE42-NEXT: ## %bb.53: ## %cond.load101 +; SSE42-NEXT: insertps {{.*#+}} xmm6 = xmm6[0,1],mem[0],xmm6[3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_54: ## %else102 +; SSE42-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8 +; SSE42-NEXT: pextrb $12, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_56 +; SSE42-NEXT: ## %bb.55: ## %cond.load105 +; SSE42-NEXT: insertps {{.*#+}} xmm6 = xmm6[0,1,2],mem[0] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_56: ## %else106 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm8, %xmm9 +; SSE42-NEXT: pextrb $0, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_58 +; SSE42-NEXT: ## %bb.57: ## %cond.load109 +; SSE42-NEXT: movss {{.*#+}} xmm10 = mem[0],zero,zero,zero +; SSE42-NEXT: blendps {{.*#+}} xmm7 = xmm10[0],xmm7[1,2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_58: ## %else110 +; SSE42-NEXT: pextrb $4, %xmm9, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_60 +; SSE42-NEXT: ## %bb.59: ## %cond.load113 +; SSE42-NEXT: insertps {{.*#+}} xmm7 = xmm7[0],mem[0],xmm7[2,3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_60: ## %else114 +; SSE42-NEXT: pxor %xmm9, %xmm9 +; SSE42-NEXT: pcmpeqd %xmm9, %xmm8 +; SSE42-NEXT: pextrb $8, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_62 +; SSE42-NEXT: ## %bb.61: ## %cond.load117 +; SSE42-NEXT: insertps {{.*#+}} xmm7 = xmm7[0,1],mem[0],xmm7[3] +; SSE42-NEXT: addq $4, %rsi +; SSE42-NEXT: LBB8_62: ## %else118 +; SSE42-NEXT: pextrb $12, %xmm8, %ecx +; SSE42-NEXT: testb $1, %cl +; SSE42-NEXT: je LBB8_64 +; SSE42-NEXT: ## %bb.63: ## %cond.load121 +; SSE42-NEXT: insertps {{.*#+}} xmm7 = xmm7[0,1,2],mem[0] +; SSE42-NEXT: LBB8_64: ## %else122 +; SSE42-NEXT: movaps %xmm0, (%rax) +; SSE42-NEXT: movaps %xmm1, 16(%rax) +; SSE42-NEXT: movaps %xmm2, 32(%rax) +; SSE42-NEXT: movaps %xmm3, 48(%rax) +; SSE42-NEXT: movaps %xmm4, 64(%rax) +; SSE42-NEXT: movaps %xmm5, 80(%rax) +; SSE42-NEXT: movaps %xmm6, 96(%rax) +; SSE42-NEXT: movaps %xmm7, 112(%rax) +; SSE42-NEXT: retq +; +; AVX1-LABEL: expandload_v32f32_v32i32: +; AVX1: ## %bb.0: +; AVX1-NEXT: vpxor %xmm8, %xmm8, %xmm8 +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm4, %xmm9 +; AVX1-NEXT: vpackssdw %xmm0, %xmm9, %xmm9 +; AVX1-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 +; AVX1-NEXT: vpextrb $0, %xmm9, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_2 +; AVX1-NEXT: ## %bb.1: ## %cond.load +; AVX1-NEXT: vmovss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm9[0],ymm0[1,2,3,4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_2: ## %else +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm4, %xmm8 +; AVX1-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 +; AVX1-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 +; AVX1-NEXT: vpextrb $1, %xmm8, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_4 +; AVX1-NEXT: ## %bb.3: ## %cond.load1 +; AVX1-NEXT: vinsertps {{.*#+}} xmm8 = xmm0[0],mem[0],xmm0[2,3] +; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm0[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_4: ## %else2 +; AVX1-NEXT: vxorps %xmm8, %xmm8, %xmm8 +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm4, %xmm9 +; AVX1-NEXT: vpackssdw %xmm0, %xmm9, %xmm9 +; AVX1-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 +; AVX1-NEXT: vpextrb $2, %xmm9, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_6 +; AVX1-NEXT: ## %bb.5: ## %cond.load5 +; AVX1-NEXT: vinsertps {{.*#+}} xmm9 = xmm0[0,1],mem[0],xmm0[3] +; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm9[0,1,2,3],ymm0[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_6: ## %else6 +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm4, %xmm8 +; AVX1-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 +; AVX1-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 +; AVX1-NEXT: vpextrb $3, %xmm8, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_8 +; AVX1-NEXT: ## %bb.7: ## %cond.load9 +; AVX1-NEXT: vinsertps {{.*#+}} xmm8 = xmm0[0,1,2],mem[0] +; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm0[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_8: ## %else10 +; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm8 +; AVX1-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm8, %xmm9 +; AVX1-NEXT: vpackssdw %xmm9, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $4, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_10 +; AVX1-NEXT: ## %bb.9: ## %cond.load13 +; AVX1-NEXT: vmovd {{.*#+}} xmm10 = mem[0],zero,zero,zero +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm10[0,1],xmm4[2,3,4,5,6,7] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm0, %ymm0 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_10: ## %else14 +; AVX1-NEXT: vpackssdw %xmm9, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $5, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_12 +; AVX1-NEXT: ## %bb.11: ## %cond.load17 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0],mem[0],xmm4[2,3] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm0, %ymm0 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_12: ## %else18 +; AVX1-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm8, %xmm8 +; AVX1-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $6, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_14 +; AVX1-NEXT: ## %bb.13: ## %cond.load21 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1],mem[0],xmm4[3] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm0, %ymm0 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_14: ## %else22 +; AVX1-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $7, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_16 +; AVX1-NEXT: ## %bb.15: ## %cond.load25 +; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm0, %ymm0 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_16: ## %else26 +; AVX1-NEXT: vpxor %xmm8, %xmm8, %xmm8 +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm5, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $8, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_18 +; AVX1-NEXT: ## %bb.17: ## %cond.load29 +; AVX1-NEXT: vmovss {{.*#+}} xmm4 = mem[0],zero,zero,zero +; AVX1-NEXT: vblendps {{.*#+}} ymm1 = ymm4[0],ymm1[1,2,3,4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_18: ## %else30 +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm5, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $9, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_20 +; AVX1-NEXT: ## %bb.19: ## %cond.load33 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm1[0],mem[0],xmm1[2,3] +; AVX1-NEXT: vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_20: ## %else34 +; AVX1-NEXT: vpxor %xmm8, %xmm8, %xmm8 +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm5, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $10, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_22 +; AVX1-NEXT: ## %bb.21: ## %cond.load37 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm1[0,1],mem[0],xmm1[3] +; AVX1-NEXT: vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_22: ## %else38 +; AVX1-NEXT: vpcmpeqd %xmm8, %xmm5, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $11, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_24 +; AVX1-NEXT: ## %bb.23: ## %cond.load41 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm1[0,1,2],mem[0] +; AVX1-NEXT: vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_24: ## %else42 +; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm8 +; AVX1-NEXT: vpxor %xmm5, %xmm5, %xmm5 +; AVX1-NEXT: vpcmpeqd %xmm5, %xmm8, %xmm5 +; AVX1-NEXT: vpackssdw %xmm5, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $12, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_26 +; AVX1-NEXT: ## %bb.25: ## %cond.load45 +; AVX1-NEXT: vmovd {{.*#+}} xmm9 = mem[0],zero,zero,zero +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm4 +; AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm9[0,1],xmm4[2,3,4,5,6,7] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_26: ## %else46 +; AVX1-NEXT: vpackssdw %xmm5, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $13, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_28 +; AVX1-NEXT: ## %bb.27: ## %cond.load49 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm4 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0],mem[0],xmm4[2,3] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_28: ## %else50 +; AVX1-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm8, %xmm4 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 +; AVX1-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX1-NEXT: vpextrb $14, %xmm5, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_30 +; AVX1-NEXT: ## %bb.29: ## %cond.load53 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5 +; AVX1-NEXT: vinsertps {{.*#+}} xmm5 = xmm5[0,1],mem[0],xmm5[3] +; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm1, %ymm1 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_30: ## %else54 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $15, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_32 +; AVX1-NEXT: ## %bb.31: ## %cond.load57 +; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm4 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_32: ## %else58 +; AVX1-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm6, %xmm5 +; AVX1-NEXT: vpackssdw %xmm0, %xmm5, %xmm5 +; AVX1-NEXT: vpacksswb %xmm0, %xmm5, %xmm5 +; AVX1-NEXT: vpextrb $0, %xmm5, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_34 +; AVX1-NEXT: ## %bb.33: ## %cond.load61 +; AVX1-NEXT: vmovss {{.*#+}} xmm5 = mem[0],zero,zero,zero +; AVX1-NEXT: vblendps {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_34: ## %else62 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm6, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $1, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_36 +; AVX1-NEXT: ## %bb.35: ## %cond.load65 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm2[0],mem[0],xmm2[2,3] +; AVX1-NEXT: vblendps {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_36: ## %else66 +; AVX1-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm6, %xmm5 +; AVX1-NEXT: vpackssdw %xmm0, %xmm5, %xmm5 +; AVX1-NEXT: vpacksswb %xmm0, %xmm5, %xmm5 +; AVX1-NEXT: vpextrb $2, %xmm5, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_38 +; AVX1-NEXT: ## %bb.37: ## %cond.load69 +; AVX1-NEXT: vinsertps {{.*#+}} xmm5 = xmm2[0,1],mem[0],xmm2[3] +; AVX1-NEXT: vblendps {{.*#+}} ymm2 = ymm5[0,1,2,3],ymm2[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_38: ## %else70 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm6, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $3, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_40 +; AVX1-NEXT: ## %bb.39: ## %cond.load73 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm2[0,1,2],mem[0] +; AVX1-NEXT: vblendps {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_40: ## %else74 +; AVX1-NEXT: vextractf128 $1, %ymm6, %xmm8 +; AVX1-NEXT: vxorps %xmm5, %xmm5, %xmm5 +; AVX1-NEXT: vpcmpeqd %xmm5, %xmm8, %xmm5 +; AVX1-NEXT: vpackssdw %xmm5, %xmm0, %xmm6 +; AVX1-NEXT: vpacksswb %xmm0, %xmm6, %xmm6 +; AVX1-NEXT: vpextrb $4, %xmm6, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_42 +; AVX1-NEXT: ## %bb.41: ## %cond.load77 +; AVX1-NEXT: vmovss {{.*#+}} xmm6 = mem[0],zero,zero,zero +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm4 +; AVX1-NEXT: vblendps {{.*#+}} xmm4 = xmm6[0],xmm4[1,2,3] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm2, %ymm2 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_42: ## %else78 +; AVX1-NEXT: vpackssdw %xmm5, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $5, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_44 +; AVX1-NEXT: ## %bb.43: ## %cond.load81 +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm4 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0],mem[0],xmm4[2,3] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm2, %ymm2 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_44: ## %else82 +; AVX1-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm8, %xmm4 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 +; AVX1-NEXT: vpacksswb %xmm0, %xmm5, %xmm5 +; AVX1-NEXT: vpextrb $6, %xmm5, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_46 +; AVX1-NEXT: ## %bb.45: ## %cond.load85 +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm5 +; AVX1-NEXT: vinsertps {{.*#+}} xmm5 = xmm5[0,1],mem[0],xmm5[3] +; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm2, %ymm2 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_46: ## %else86 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpextrb $7, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_48 +; AVX1-NEXT: ## %bb.47: ## %cond.load89 +; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm4 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm2, %ymm2 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_48: ## %else90 +; AVX1-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm7, %xmm5 +; AVX1-NEXT: vpackssdw %xmm0, %xmm5, %xmm5 +; AVX1-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX1-NEXT: vpextrb $8, %xmm5, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_50 +; AVX1-NEXT: ## %bb.49: ## %cond.load93 +; AVX1-NEXT: vmovss {{.*#+}} xmm5 = mem[0],zero,zero,zero +; AVX1-NEXT: vblendps {{.*#+}} ymm3 = ymm5[0],ymm3[1,2,3,4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_50: ## %else94 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm7, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $9, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_52 +; AVX1-NEXT: ## %bb.51: ## %cond.load97 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm3[0],mem[0],xmm3[2,3] +; AVX1-NEXT: vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_52: ## %else98 +; AVX1-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm7, %xmm5 +; AVX1-NEXT: vpackssdw %xmm0, %xmm5, %xmm5 +; AVX1-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX1-NEXT: vpextrb $10, %xmm5, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_54 +; AVX1-NEXT: ## %bb.53: ## %cond.load101 +; AVX1-NEXT: vinsertps {{.*#+}} xmm5 = xmm3[0,1],mem[0],xmm3[3] +; AVX1-NEXT: vblendps {{.*#+}} ymm3 = ymm5[0,1,2,3],ymm3[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_54: ## %else102 +; AVX1-NEXT: vpcmpeqd %xmm4, %xmm7, %xmm4 +; AVX1-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $11, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_56 +; AVX1-NEXT: ## %bb.55: ## %cond.load105 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm3[0,1,2],mem[0] +; AVX1-NEXT: vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7] +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_56: ## %else106 +; AVX1-NEXT: vextractf128 $1, %ymm7, %xmm4 +; AVX1-NEXT: vxorps %xmm5, %xmm5, %xmm5 +; AVX1-NEXT: vpcmpeqd %xmm5, %xmm4, %xmm5 +; AVX1-NEXT: vpackssdw %xmm5, %xmm0, %xmm6 +; AVX1-NEXT: vpacksswb %xmm6, %xmm0, %xmm6 +; AVX1-NEXT: vpextrb $12, %xmm6, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_58 +; AVX1-NEXT: ## %bb.57: ## %cond.load109 +; AVX1-NEXT: vmovd {{.*#+}} xmm6 = mem[0],zero,zero,zero +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm7 +; AVX1-NEXT: vpblendw {{.*#+}} xmm6 = xmm6[0,1],xmm7[2,3,4,5,6,7] +; AVX1-NEXT: vinsertf128 $1, %xmm6, %ymm3, %ymm3 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_58: ## %else110 +; AVX1-NEXT: vpackssdw %xmm5, %xmm0, %xmm5 +; AVX1-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX1-NEXT: vpextrb $13, %xmm5, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_60 +; AVX1-NEXT: ## %bb.59: ## %cond.load113 +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm5 +; AVX1-NEXT: vinsertps {{.*#+}} xmm5 = xmm5[0],mem[0],xmm5[2,3] +; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm3, %ymm3 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_60: ## %else114 +; AVX1-NEXT: vxorps %xmm5, %xmm5, %xmm5 +; AVX1-NEXT: vpcmpeqd %xmm5, %xmm4, %xmm4 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 +; AVX1-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX1-NEXT: vpextrb $14, %xmm5, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_62 +; AVX1-NEXT: ## %bb.61: ## %cond.load117 +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm5 +; AVX1-NEXT: vinsertps {{.*#+}} xmm5 = xmm5[0,1],mem[0],xmm5[3] +; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm3, %ymm3 +; AVX1-NEXT: addq $4, %rdi +; AVX1-NEXT: LBB8_62: ## %else118 +; AVX1-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX1-NEXT: vpextrb $15, %xmm4, %eax +; AVX1-NEXT: testb $1, %al +; AVX1-NEXT: je LBB8_64 +; AVX1-NEXT: ## %bb.63: ## %cond.load121 +; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm4 +; AVX1-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] +; AVX1-NEXT: vinsertf128 $1, %xmm4, %ymm3, %ymm3 +; AVX1-NEXT: LBB8_64: ## %else122 +; AVX1-NEXT: retq +; +; AVX2-LABEL: expandload_v32f32_v32i32: +; AVX2: ## %bb.0: +; AVX2-NEXT: vpxor %xmm8, %xmm8, %xmm8 +; AVX2-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm9 +; AVX2-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 +; AVX2-NEXT: vpextrb $0, %xmm9, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_2 +; AVX2-NEXT: ## %bb.1: ## %cond.load +; AVX2-NEXT: vmovd {{.*#+}} xmm9 = mem[0],zero,zero,zero +; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0],ymm0[1,2,3,4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_2: ## %else +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 +; AVX2-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 +; AVX2-NEXT: vpextrb $1, %xmm8, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_4 +; AVX2-NEXT: ## %bb.3: ## %cond.load1 +; AVX2-NEXT: vinsertps {{.*#+}} xmm8 = xmm0[0],mem[0],xmm0[2,3] +; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm0[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_4: ## %else2 +; AVX2-NEXT: vxorps %xmm8, %xmm8, %xmm8 +; AVX2-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm9 +; AVX2-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 +; AVX2-NEXT: vpextrb $2, %xmm9, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_6 +; AVX2-NEXT: ## %bb.5: ## %cond.load5 +; AVX2-NEXT: vinsertps {{.*#+}} xmm9 = xmm0[0,1],mem[0],xmm0[3] +; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm9[0,1,2,3],ymm0[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_6: ## %else6 +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm8 +; AVX2-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 +; AVX2-NEXT: vpextrb $3, %xmm8, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_8 +; AVX2-NEXT: ## %bb.7: ## %cond.load9 +; AVX2-NEXT: vinsertps {{.*#+}} xmm8 = xmm0[0,1,2],mem[0] +; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm0[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_8: ## %else10 +; AVX2-NEXT: vxorps %xmm8, %xmm8, %xmm8 +; AVX2-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm8 +; AVX2-NEXT: vextracti128 $1, %ymm8, %xmm8 +; AVX2-NEXT: vpackssdw %xmm8, %xmm0, %xmm9 +; AVX2-NEXT: vpacksswb %xmm0, %xmm9, %xmm9 +; AVX2-NEXT: vpextrb $4, %xmm9, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_10 +; AVX2-NEXT: ## %bb.9: ## %cond.load13 +; AVX2-NEXT: vmovss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm10 +; AVX2-NEXT: vblendps {{.*#+}} xmm9 = xmm9[0],xmm10[1,2,3] +; AVX2-NEXT: vinsertf128 $1, %xmm9, %ymm0, %ymm0 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_10: ## %else14 +; AVX2-NEXT: vpackssdw %xmm8, %xmm0, %xmm8 +; AVX2-NEXT: vpacksswb %xmm0, %xmm8, %xmm8 +; AVX2-NEXT: vpextrb $5, %xmm8, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_12 +; AVX2-NEXT: ## %bb.11: ## %cond.load17 +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm8 +; AVX2-NEXT: vinsertps {{.*#+}} xmm8 = xmm8[0],mem[0],xmm8[2,3] +; AVX2-NEXT: vinsertf128 $1, %xmm8, %ymm0, %ymm0 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_12: ## %else18 +; AVX2-NEXT: vxorps %xmm8, %xmm8, %xmm8 +; AVX2-NEXT: vpcmpeqd %ymm8, %ymm4, %ymm4 +; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm8 +; AVX2-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpextrb $6, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_14 +; AVX2-NEXT: ## %bb.13: ## %cond.load21 +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm4 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1],mem[0],xmm4[3] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm0, %ymm0 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_14: ## %else22 +; AVX2-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpextrb $7, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_16 +; AVX2-NEXT: ## %bb.15: ## %cond.load25 +; AVX2-NEXT: vextractf128 $1, %ymm0, %xmm4 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm0, %ymm0 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_16: ## %else26 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm8 +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $8, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_18 +; AVX2-NEXT: ## %bb.17: ## %cond.load29 +; AVX2-NEXT: vmovd {{.*#+}} xmm4 = mem[0],zero,zero,zero +; AVX2-NEXT: vpblendd {{.*#+}} ymm1 = ymm4[0],ymm1[1,2,3,4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_18: ## %else30 +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $9, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_20 +; AVX2-NEXT: ## %bb.19: ## %cond.load33 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm1[0],mem[0],xmm1[2,3] +; AVX2-NEXT: vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_20: ## %else34 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm8 +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $10, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_22 +; AVX2-NEXT: ## %bb.21: ## %cond.load37 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm1[0,1],mem[0],xmm1[3] +; AVX2-NEXT: vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_22: ## %else38 +; AVX2-NEXT: vpackssdw %xmm0, %xmm8, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $11, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_24 +; AVX2-NEXT: ## %bb.23: ## %cond.load41 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm1[0,1,2],mem[0] +; AVX2-NEXT: vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm1[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_24: ## %else42 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 +; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm8 +; AVX2-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $12, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_26 +; AVX2-NEXT: ## %bb.25: ## %cond.load45 +; AVX2-NEXT: vmovss {{.*#+}} xmm9 = mem[0],zero,zero,zero +; AVX2-NEXT: vextractf128 $1, %ymm1, %xmm4 +; AVX2-NEXT: vblendps {{.*#+}} xmm4 = xmm9[0],xmm4[1,2,3] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_26: ## %else46 +; AVX2-NEXT: vpackssdw %xmm8, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $13, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_28 +; AVX2-NEXT: ## %bb.27: ## %cond.load49 +; AVX2-NEXT: vextractf128 $1, %ymm1, %xmm4 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0],mem[0],xmm4[2,3] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_28: ## %else50 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm5, %ymm4 +; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm4 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 +; AVX2-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX2-NEXT: vpextrb $14, %xmm5, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_30 +; AVX2-NEXT: ## %bb.29: ## %cond.load53 +; AVX2-NEXT: vextractf128 $1, %ymm1, %xmm5 +; AVX2-NEXT: vinsertps {{.*#+}} xmm5 = xmm5[0,1],mem[0],xmm5[3] +; AVX2-NEXT: vinsertf128 $1, %xmm5, %ymm1, %ymm1 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_30: ## %else54 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $15, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_32 +; AVX2-NEXT: ## %bb.31: ## %cond.load57 +; AVX2-NEXT: vextractf128 $1, %ymm1, %xmm4 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm1, %ymm1 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_32: ## %else58 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm6, %ymm4 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm5 +; AVX2-NEXT: vpacksswb %xmm0, %xmm5, %xmm5 +; AVX2-NEXT: vpextrb $0, %xmm5, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_34 +; AVX2-NEXT: ## %bb.33: ## %cond.load61 +; AVX2-NEXT: vmovd {{.*#+}} xmm5 = mem[0],zero,zero,zero +; AVX2-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_34: ## %else62 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpextrb $1, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_36 +; AVX2-NEXT: ## %bb.35: ## %cond.load65 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm2[0],mem[0],xmm2[2,3] +; AVX2-NEXT: vblendps {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_36: ## %else66 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm6, %ymm4 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm5 +; AVX2-NEXT: vpacksswb %xmm0, %xmm5, %xmm5 +; AVX2-NEXT: vpextrb $2, %xmm5, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_38 +; AVX2-NEXT: ## %bb.37: ## %cond.load69 +; AVX2-NEXT: vinsertps {{.*#+}} xmm5 = xmm2[0,1],mem[0],xmm2[3] +; AVX2-NEXT: vblendps {{.*#+}} ymm2 = ymm5[0,1,2,3],ymm2[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_38: ## %else70 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpextrb $3, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_40 +; AVX2-NEXT: ## %bb.39: ## %cond.load73 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm2[0,1,2],mem[0] +; AVX2-NEXT: vblendps {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_40: ## %else74 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm6, %ymm4 +; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm4 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 +; AVX2-NEXT: vpacksswb %xmm0, %xmm5, %xmm5 +; AVX2-NEXT: vpextrb $4, %xmm5, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_42 +; AVX2-NEXT: ## %bb.41: ## %cond.load77 +; AVX2-NEXT: vmovss {{.*#+}} xmm8 = mem[0],zero,zero,zero +; AVX2-NEXT: vextractf128 $1, %ymm2, %xmm5 +; AVX2-NEXT: vblendps {{.*#+}} xmm5 = xmm8[0],xmm5[1,2,3] +; AVX2-NEXT: vinsertf128 $1, %xmm5, %ymm2, %ymm2 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_42: ## %else78 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpextrb $5, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_44 +; AVX2-NEXT: ## %bb.43: ## %cond.load81 +; AVX2-NEXT: vextractf128 $1, %ymm2, %xmm4 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0],mem[0],xmm4[2,3] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm2, %ymm2 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_44: ## %else82 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm6, %ymm4 +; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm4 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 +; AVX2-NEXT: vpacksswb %xmm0, %xmm5, %xmm5 +; AVX2-NEXT: vpextrb $6, %xmm5, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_46 +; AVX2-NEXT: ## %bb.45: ## %cond.load85 +; AVX2-NEXT: vextractf128 $1, %ymm2, %xmm5 +; AVX2-NEXT: vinsertps {{.*#+}} xmm5 = xmm5[0,1],mem[0],xmm5[3] +; AVX2-NEXT: vinsertf128 $1, %xmm5, %ymm2, %ymm2 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_46: ## %else86 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpextrb $7, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_48 +; AVX2-NEXT: ## %bb.47: ## %cond.load89 +; AVX2-NEXT: vextractf128 $1, %ymm2, %xmm4 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm2, %ymm2 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_48: ## %else90 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm7, %ymm4 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm5 +; AVX2-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX2-NEXT: vpextrb $8, %xmm5, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_50 +; AVX2-NEXT: ## %bb.49: ## %cond.load93 +; AVX2-NEXT: vmovd {{.*#+}} xmm5 = mem[0],zero,zero,zero +; AVX2-NEXT: vpblendd {{.*#+}} ymm3 = ymm5[0],ymm3[1,2,3,4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_50: ## %else94 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $9, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_52 +; AVX2-NEXT: ## %bb.51: ## %cond.load97 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm3[0],mem[0],xmm3[2,3] +; AVX2-NEXT: vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_52: ## %else98 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm7, %ymm4 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm5 +; AVX2-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX2-NEXT: vpextrb $10, %xmm5, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_54 +; AVX2-NEXT: ## %bb.53: ## %cond.load101 +; AVX2-NEXT: vinsertps {{.*#+}} xmm5 = xmm3[0,1],mem[0],xmm3[3] +; AVX2-NEXT: vblendps {{.*#+}} ymm3 = ymm5[0,1,2,3],ymm3[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_54: ## %else102 +; AVX2-NEXT: vpackssdw %xmm0, %xmm4, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $11, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_56 +; AVX2-NEXT: ## %bb.55: ## %cond.load105 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm3[0,1,2],mem[0] +; AVX2-NEXT: vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7] +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_56: ## %else106 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm7, %ymm4 +; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm4 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 +; AVX2-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX2-NEXT: vpextrb $12, %xmm5, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_58 +; AVX2-NEXT: ## %bb.57: ## %cond.load109 +; AVX2-NEXT: vmovss {{.*#+}} xmm5 = mem[0],zero,zero,zero +; AVX2-NEXT: vextractf128 $1, %ymm3, %xmm6 +; AVX2-NEXT: vblendps {{.*#+}} xmm5 = xmm5[0],xmm6[1,2,3] +; AVX2-NEXT: vinsertf128 $1, %xmm5, %ymm3, %ymm3 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_58: ## %else110 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $13, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_60 +; AVX2-NEXT: ## %bb.59: ## %cond.load113 +; AVX2-NEXT: vextractf128 $1, %ymm3, %xmm4 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0],mem[0],xmm4[2,3] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm3, %ymm3 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_60: ## %else114 +; AVX2-NEXT: vxorps %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpcmpeqd %ymm4, %ymm7, %ymm4 +; AVX2-NEXT: vextracti128 $1, %ymm4, %xmm4 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm5 +; AVX2-NEXT: vpacksswb %xmm5, %xmm0, %xmm5 +; AVX2-NEXT: vpextrb $14, %xmm5, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_62 +; AVX2-NEXT: ## %bb.61: ## %cond.load117 +; AVX2-NEXT: vextractf128 $1, %ymm3, %xmm5 +; AVX2-NEXT: vinsertps {{.*#+}} xmm5 = xmm5[0,1],mem[0],xmm5[3] +; AVX2-NEXT: vinsertf128 $1, %xmm5, %ymm3, %ymm3 +; AVX2-NEXT: addq $4, %rdi +; AVX2-NEXT: LBB8_62: ## %else118 +; AVX2-NEXT: vpackssdw %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpacksswb %xmm4, %xmm0, %xmm4 +; AVX2-NEXT: vpextrb $15, %xmm4, %eax +; AVX2-NEXT: testb $1, %al +; AVX2-NEXT: je LBB8_64 +; AVX2-NEXT: ## %bb.63: ## %cond.load121 +; AVX2-NEXT: vextractf128 $1, %ymm3, %xmm4 +; AVX2-NEXT: vinsertps {{.*#+}} xmm4 = xmm4[0,1,2],mem[0] +; AVX2-NEXT: vinsertf128 $1, %xmm4, %ymm3, %ymm3 +; AVX2-NEXT: LBB8_64: ## %else122 +; AVX2-NEXT: retq +; +; AVX512-LABEL: expandload_v32f32_v32i32: +; AVX512: ## %bb.0: +; AVX512-NEXT: vptestnmd %zmm3, %zmm3, %k2 +; AVX512-NEXT: vptestnmd %zmm2, %zmm2, %k1 +; AVX512-NEXT: kmovw %k1, %eax +; AVX512-NEXT: movl %eax, %ecx +; AVX512-NEXT: shrl %ecx +; AVX512-NEXT: andl $21845, %ecx ## imm = 0x5555 +; AVX512-NEXT: subl %ecx, %eax +; AVX512-NEXT: movl %eax, %ecx +; AVX512-NEXT: andl $858993459, %ecx ## imm = 0x33333333 +; AVX512-NEXT: shrl $2, %eax +; AVX512-NEXT: andl $858993459, %eax ## imm = 0x33333333 +; AVX512-NEXT: addl %ecx, %eax +; AVX512-NEXT: movl %eax, %ecx +; AVX512-NEXT: shrl $4, %ecx +; AVX512-NEXT: addl %eax, %ecx +; AVX512-NEXT: andl $252645135, %ecx ## imm = 0xF0F0F0F +; AVX512-NEXT: imull $16843009, %ecx, %eax ## imm = 0x1010101 +; AVX512-NEXT: shrl $24, %eax +; AVX512-NEXT: vexpandps (%rdi,%rax,4), %zmm1 {%k2} +; AVX512-NEXT: vexpandps (%rdi), %zmm0 {%k1} +; AVX512-NEXT: retq %mask = icmp eq <32 x i32> %trigger, zeroinitializer %res = call <32 x float> @llvm.masked.expandload.v32f32(float* %base, <32 x i1> %mask, <32 x float> %src0) ret <32 x float> %res @@ -982,26 +2928,38 @@ define <32 x float> @expandload_v32f32_v32i32(float* %base, <32 x float> %src0, ; define <2 x i64> @expandload_v2i64_const(i64* %base, <2 x i64> %src0) { -; HSW-LABEL: expandload_v2i64_const: -; HSW: # %bb.0: # %else -; HSW-NEXT: vpinsrq $1, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v2i64_const: -; SKX: # %bb.0: -; SKX-NEXT: movb $2, %al -; SKX-NEXT: kmovd %eax, %k1 -; SKX-NEXT: vpexpandq (%rdi), %xmm0 {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v2i64_const: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; KNL-NEXT: movb $2, %al -; KNL-NEXT: kmovw %eax, %k1 -; KNL-NEXT: vpexpandq (%rdi), %zmm0 {%k1} -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v2i64_const: +; SSE2: ## %bb.0: ## %else +; SSE2-NEXT: movsd {{.*#+}} xmm1 = mem[0],zero +; SSE2-NEXT: movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v2i64_const: +; SSE42: ## %bb.0: ## %else +; SSE42-NEXT: pinsrq $1, (%rdi), %xmm0 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: expandload_v2i64_const: +; AVX1OR2: ## %bb.0: ## %else +; AVX1OR2-NEXT: vpinsrq $1, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: expandload_v2i64_const: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 +; AVX512F-NEXT: movb $2, %al +; AVX512F-NEXT: kmovw %eax, %k1 +; AVX512F-NEXT: vpexpandq (%rdi), %zmm0 {%k1} +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v2i64_const: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: movb $2, %al +; AVX512VLBW-NEXT: kmovd %eax, %k1 +; AVX512VLBW-NEXT: vpexpandq (%rdi), %xmm0 {%k1} +; AVX512VLBW-NEXT: retq %res = call <2 x i64> @llvm.masked.expandload.v2i64(i64* %base, <2 x i1> , <2 x i64> %src0) ret <2 x i64>%res } @@ -1011,57 +2969,136 @@ define <2 x i64> @expandload_v2i64_const(i64* %base, <2 x i64> %src0) { ; define <4 x i32> @expandload_v4i32_v4i32(i32* %base, <4 x i32> %src0, <4 x i32> %trigger) { -; HSW-LABEL: expandload_v4i32_v4i32: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $0, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB10_2 -; HSW-NEXT: # %bb.1: # %cond.load -; HSW-NEXT: vpinsrd $0, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB10_2: # %else -; HSW-NEXT: vpextrb $4, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB10_4 -; HSW-NEXT: # %bb.3: # %cond.load1 -; HSW-NEXT: vpinsrd $1, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB10_4: # %else2 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $8, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB10_6 -; HSW-NEXT: # %bb.5: # %cond.load5 -; HSW-NEXT: vpinsrd $2, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: addq $4, %rdi -; HSW-NEXT: .LBB10_6: # %else6 -; HSW-NEXT: vpextrb $12, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB10_8 -; HSW-NEXT: # %bb.7: # %cond.load9 -; HSW-NEXT: vpinsrd $3, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: .LBB10_8: # %else10 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v4i32_v4i32: -; SKX: # %bb.0: -; SKX-NEXT: vptestnmd %xmm1, %xmm1, %k1 -; SKX-NEXT: vpexpandd (%rdi), %xmm0 {%k1} -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v4i32_v4i32: -; KNL: # %bb.0: -; KNL-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; KNL-NEXT: vptestnmd %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftlw $12, %k0, %k0 -; KNL-NEXT: kshiftrw $12, %k0, %k1 -; KNL-NEXT: vpexpandd (%rdi), %zmm0 {%k1} -; KNL-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v4i32_v4i32: +; SSE2: ## %bb.0: +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqd %xmm1, %xmm2 +; SSE2-NEXT: movd %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB10_2 +; SSE2-NEXT: ## %bb.1: ## %cond.load +; SSE2-NEXT: movss {{.*#+}} xmm3 = mem[0],zero,zero,zero +; SSE2-NEXT: movss {{.*#+}} xmm0 = xmm3[0],xmm0[1,2,3] +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB10_2: ## %else +; SSE2-NEXT: pextrw $2, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB10_4 +; SSE2-NEXT: ## %bb.3: ## %cond.load1 +; SSE2-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,0],xmm0[0,0] +; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm0[2,3] +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: movaps %xmm2, %xmm0 +; SSE2-NEXT: LBB10_4: ## %else2 +; SSE2-NEXT: xorps %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqd %xmm2, %xmm1 +; SSE2-NEXT: pextrw $4, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB10_6 +; SSE2-NEXT: ## %bb.5: ## %cond.load5 +; SSE2-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,0],xmm0[3,0] +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm2[0,2] +; SSE2-NEXT: addq $4, %rdi +; SSE2-NEXT: LBB10_6: ## %else6 +; SSE2-NEXT: pextrw $6, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB10_8 +; SSE2-NEXT: ## %bb.7: ## %cond.load9 +; SSE2-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0] +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0] +; SSE2-NEXT: LBB10_8: ## %else10 +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v4i32_v4i32: +; SSE42: ## %bb.0: +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqd %xmm1, %xmm2 +; SSE42-NEXT: pextrb $0, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB10_2 +; SSE42-NEXT: ## %bb.1: ## %cond.load +; SSE42-NEXT: pinsrd $0, (%rdi), %xmm0 +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB10_2: ## %else +; SSE42-NEXT: pextrb $4, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB10_4 +; SSE42-NEXT: ## %bb.3: ## %cond.load1 +; SSE42-NEXT: pinsrd $1, (%rdi), %xmm0 +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB10_4: ## %else2 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqd %xmm2, %xmm1 +; SSE42-NEXT: pextrb $8, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB10_6 +; SSE42-NEXT: ## %bb.5: ## %cond.load5 +; SSE42-NEXT: pinsrd $2, (%rdi), %xmm0 +; SSE42-NEXT: addq $4, %rdi +; SSE42-NEXT: LBB10_6: ## %else6 +; SSE42-NEXT: pextrb $12, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB10_8 +; SSE42-NEXT: ## %bb.7: ## %cond.load9 +; SSE42-NEXT: pinsrd $3, (%rdi), %xmm0 +; SSE42-NEXT: LBB10_8: ## %else10 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: expandload_v4i32_v4i32: +; AVX1OR2: ## %bb.0: +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $0, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB10_2 +; AVX1OR2-NEXT: ## %bb.1: ## %cond.load +; AVX1OR2-NEXT: vpinsrd $0, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB10_2: ## %else +; AVX1OR2-NEXT: vpextrb $4, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB10_4 +; AVX1OR2-NEXT: ## %bb.3: ## %cond.load1 +; AVX1OR2-NEXT: vpinsrd $1, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB10_4: ## %else2 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm1 +; AVX1OR2-NEXT: vpextrb $8, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB10_6 +; AVX1OR2-NEXT: ## %bb.5: ## %cond.load5 +; AVX1OR2-NEXT: vpinsrd $2, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: addq $4, %rdi +; AVX1OR2-NEXT: LBB10_6: ## %else6 +; AVX1OR2-NEXT: vpextrb $12, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB10_8 +; AVX1OR2-NEXT: ## %bb.7: ## %cond.load9 +; AVX1OR2-NEXT: vpinsrd $3, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: LBB10_8: ## %else10 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: expandload_v4i32_v4i32: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 +; AVX512F-NEXT: vptestnmd %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftlw $12, %k0, %k0 +; AVX512F-NEXT: kshiftrw $12, %k0, %k1 +; AVX512F-NEXT: vpexpandd (%rdi), %zmm0 {%k1} +; AVX512F-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v4i32_v4i32: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vptestnmd %xmm1, %xmm1, %k1 +; AVX512VLBW-NEXT: vpexpandd (%rdi), %xmm0 {%k1} +; AVX512VLBW-NEXT: retq %mask = icmp eq <4 x i32> %trigger, zeroinitializer %res = call <4 x i32> @llvm.masked.expandload.v4i32(i32* %base, <4 x i1> %mask, <4 x i32> %src0) ret <4 x i32>%res @@ -1072,221 +3109,356 @@ define <4 x i32> @expandload_v4i32_v4i32(i32* %base, <4 x i32> %src0, <4 x i32> ; define <8 x i16> @expandload_v8i16_v8i16(i16* %base, <8 x i16> %src0, <8 x i16> %trigger) { -; HSW-LABEL: expandload_v8i16_v8i16: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $0, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_2 -; HSW-NEXT: # %bb.1: # %cond.load -; HSW-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_2: # %else -; HSW-NEXT: vpextrb $2, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_4 -; HSW-NEXT: # %bb.3: # %cond.load1 -; HSW-NEXT: vpinsrw $1, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_4: # %else2 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $4, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_6 -; HSW-NEXT: # %bb.5: # %cond.load5 -; HSW-NEXT: vpinsrw $2, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_6: # %else6 -; HSW-NEXT: vpextrb $6, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_8 -; HSW-NEXT: # %bb.7: # %cond.load9 -; HSW-NEXT: vpinsrw $3, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_8: # %else10 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $8, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_10 -; HSW-NEXT: # %bb.9: # %cond.load13 -; HSW-NEXT: vpinsrw $4, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_10: # %else14 -; HSW-NEXT: vpextrb $10, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_12 -; HSW-NEXT: # %bb.11: # %cond.load17 -; HSW-NEXT: vpinsrw $5, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_12: # %else18 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $12, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_14 -; HSW-NEXT: # %bb.13: # %cond.load21 -; HSW-NEXT: vpinsrw $6, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: addq $2, %rdi -; HSW-NEXT: .LBB11_14: # %else22 -; HSW-NEXT: vpextrb $14, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB11_16 -; HSW-NEXT: # %bb.15: # %cond.load25 -; HSW-NEXT: vpinsrw $7, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: .LBB11_16: # %else26 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v8i16_v8i16: -; SKX: # %bb.0: -; SKX-NEXT: vptestnmw %xmm1, %xmm1, %k0 -; SKX-NEXT: kmovd %k0, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_2 -; SKX-NEXT: # %bb.1: # %cond.load -; SKX-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_2: # %else -; SKX-NEXT: kshiftrb $1, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_4 -; SKX-NEXT: # %bb.3: # %cond.load1 -; SKX-NEXT: vpinsrw $1, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_4: # %else2 -; SKX-NEXT: kshiftrb $2, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_6 -; SKX-NEXT: # %bb.5: # %cond.load5 -; SKX-NEXT: vpinsrw $2, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_6: # %else6 -; SKX-NEXT: kshiftrb $3, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_8 -; SKX-NEXT: # %bb.7: # %cond.load9 -; SKX-NEXT: vpinsrw $3, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_8: # %else10 -; SKX-NEXT: kshiftrb $4, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_10 -; SKX-NEXT: # %bb.9: # %cond.load13 -; SKX-NEXT: vpinsrw $4, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_10: # %else14 -; SKX-NEXT: kshiftrb $5, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_12 -; SKX-NEXT: # %bb.11: # %cond.load17 -; SKX-NEXT: vpinsrw $5, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_12: # %else18 -; SKX-NEXT: kshiftrb $6, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_14 -; SKX-NEXT: # %bb.13: # %cond.load21 -; SKX-NEXT: vpinsrw $6, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: addq $2, %rdi -; SKX-NEXT: .LBB11_14: # %else22 -; SKX-NEXT: kshiftrb $7, %k0, %k0 -; SKX-NEXT: kmovd %k0, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB11_16 -; SKX-NEXT: # %bb.15: # %cond.load25 -; SKX-NEXT: vpinsrw $7, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: .LBB11_16: # %else26 -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v8i16_v8i16: -; KNL: # %bb.0: -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxwq %xmm2, %zmm2 -; KNL-NEXT: vptestmq %zmm2, %zmm2, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_2 -; KNL-NEXT: # %bb.1: # %cond.load -; KNL-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_2: # %else -; KNL-NEXT: kshiftrw $1, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_4 -; KNL-NEXT: # %bb.3: # %cond.load1 -; KNL-NEXT: vpinsrw $1, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_4: # %else2 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxwq %xmm2, %zmm2 -; KNL-NEXT: vptestmq %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $2, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_6 -; KNL-NEXT: # %bb.5: # %cond.load5 -; KNL-NEXT: vpinsrw $2, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_6: # %else6 -; KNL-NEXT: kshiftrw $3, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_8 -; KNL-NEXT: # %bb.7: # %cond.load9 -; KNL-NEXT: vpinsrw $3, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_8: # %else10 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxwq %xmm2, %zmm2 -; KNL-NEXT: vptestmq %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $4, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_10 -; KNL-NEXT: # %bb.9: # %cond.load13 -; KNL-NEXT: vpinsrw $4, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_10: # %else14 -; KNL-NEXT: kshiftrw $5, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_12 -; KNL-NEXT: # %bb.11: # %cond.load17 -; KNL-NEXT: vpinsrw $5, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_12: # %else18 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm1 -; KNL-NEXT: vpmovsxwq %xmm1, %zmm1 -; KNL-NEXT: vptestmq %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftrw $6, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_14 -; KNL-NEXT: # %bb.13: # %cond.load21 -; KNL-NEXT: vpinsrw $6, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $2, %rdi -; KNL-NEXT: .LBB11_14: # %else22 -; KNL-NEXT: kshiftrw $7, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB11_16 -; KNL-NEXT: # %bb.15: # %cond.load25 -; KNL-NEXT: vpinsrw $7, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: .LBB11_16: # %else26 -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v8i16_v8i16: +; SSE2: ## %bb.0: +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE2-NEXT: movd %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_2 +; SSE2-NEXT: ## %bb.1: ## %cond.load +; SSE2-NEXT: pinsrw $0, (%rdi), %xmm0 +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_2: ## %else +; SSE2-NEXT: shrl $16, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_4 +; SSE2-NEXT: ## %bb.3: ## %cond.load1 +; SSE2-NEXT: pinsrw $1, (%rdi), %xmm0 +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_4: ## %else2 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE2-NEXT: pextrw $2, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_6 +; SSE2-NEXT: ## %bb.5: ## %cond.load5 +; SSE2-NEXT: pinsrw $2, (%rdi), %xmm0 +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_6: ## %else6 +; SSE2-NEXT: pextrw $3, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_8 +; SSE2-NEXT: ## %bb.7: ## %cond.load9 +; SSE2-NEXT: pinsrw $3, (%rdi), %xmm0 +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_8: ## %else10 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE2-NEXT: pextrw $4, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_10 +; SSE2-NEXT: ## %bb.9: ## %cond.load13 +; SSE2-NEXT: pinsrw $4, (%rdi), %xmm0 +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_10: ## %else14 +; SSE2-NEXT: pextrw $5, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_12 +; SSE2-NEXT: ## %bb.11: ## %cond.load17 +; SSE2-NEXT: pinsrw $5, (%rdi), %xmm0 +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_12: ## %else18 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqw %xmm2, %xmm1 +; SSE2-NEXT: pextrw $6, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_14 +; SSE2-NEXT: ## %bb.13: ## %cond.load21 +; SSE2-NEXT: pinsrw $6, (%rdi), %xmm0 +; SSE2-NEXT: addq $2, %rdi +; SSE2-NEXT: LBB11_14: ## %else22 +; SSE2-NEXT: pextrw $7, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB11_16 +; SSE2-NEXT: ## %bb.15: ## %cond.load25 +; SSE2-NEXT: pinsrw $7, (%rdi), %xmm0 +; SSE2-NEXT: LBB11_16: ## %else26 +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v8i16_v8i16: +; SSE42: ## %bb.0: +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE42-NEXT: pextrb $0, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_2 +; SSE42-NEXT: ## %bb.1: ## %cond.load +; SSE42-NEXT: pinsrw $0, (%rdi), %xmm0 +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_2: ## %else +; SSE42-NEXT: pextrb $2, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_4 +; SSE42-NEXT: ## %bb.3: ## %cond.load1 +; SSE42-NEXT: pinsrw $1, (%rdi), %xmm0 +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_4: ## %else2 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE42-NEXT: pextrb $4, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_6 +; SSE42-NEXT: ## %bb.5: ## %cond.load5 +; SSE42-NEXT: pinsrw $2, (%rdi), %xmm0 +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_6: ## %else6 +; SSE42-NEXT: pextrb $6, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_8 +; SSE42-NEXT: ## %bb.7: ## %cond.load9 +; SSE42-NEXT: pinsrw $3, (%rdi), %xmm0 +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_8: ## %else10 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqw %xmm1, %xmm2 +; SSE42-NEXT: pextrb $8, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_10 +; SSE42-NEXT: ## %bb.9: ## %cond.load13 +; SSE42-NEXT: pinsrw $4, (%rdi), %xmm0 +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_10: ## %else14 +; SSE42-NEXT: pextrb $10, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_12 +; SSE42-NEXT: ## %bb.11: ## %cond.load17 +; SSE42-NEXT: pinsrw $5, (%rdi), %xmm0 +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_12: ## %else18 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqw %xmm2, %xmm1 +; SSE42-NEXT: pextrb $12, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_14 +; SSE42-NEXT: ## %bb.13: ## %cond.load21 +; SSE42-NEXT: pinsrw $6, (%rdi), %xmm0 +; SSE42-NEXT: addq $2, %rdi +; SSE42-NEXT: LBB11_14: ## %else22 +; SSE42-NEXT: pextrb $14, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB11_16 +; SSE42-NEXT: ## %bb.15: ## %cond.load25 +; SSE42-NEXT: pinsrw $7, (%rdi), %xmm0 +; SSE42-NEXT: LBB11_16: ## %else26 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: expandload_v8i16_v8i16: +; AVX1OR2: ## %bb.0: +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $0, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_2 +; AVX1OR2-NEXT: ## %bb.1: ## %cond.load +; AVX1OR2-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_2: ## %else +; AVX1OR2-NEXT: vpextrb $2, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_4 +; AVX1OR2-NEXT: ## %bb.3: ## %cond.load1 +; AVX1OR2-NEXT: vpinsrw $1, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_4: ## %else2 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $4, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_6 +; AVX1OR2-NEXT: ## %bb.5: ## %cond.load5 +; AVX1OR2-NEXT: vpinsrw $2, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_6: ## %else6 +; AVX1OR2-NEXT: vpextrb $6, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_8 +; AVX1OR2-NEXT: ## %bb.7: ## %cond.load9 +; AVX1OR2-NEXT: vpinsrw $3, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_8: ## %else10 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $8, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_10 +; AVX1OR2-NEXT: ## %bb.9: ## %cond.load13 +; AVX1OR2-NEXT: vpinsrw $4, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_10: ## %else14 +; AVX1OR2-NEXT: vpextrb $10, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_12 +; AVX1OR2-NEXT: ## %bb.11: ## %cond.load17 +; AVX1OR2-NEXT: vpinsrw $5, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_12: ## %else18 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm1 +; AVX1OR2-NEXT: vpextrb $12, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_14 +; AVX1OR2-NEXT: ## %bb.13: ## %cond.load21 +; AVX1OR2-NEXT: vpinsrw $6, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: addq $2, %rdi +; AVX1OR2-NEXT: LBB11_14: ## %else22 +; AVX1OR2-NEXT: vpextrb $14, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB11_16 +; AVX1OR2-NEXT: ## %bb.15: ## %cond.load25 +; AVX1OR2-NEXT: vpinsrw $7, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: LBB11_16: ## %else26 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: expandload_v8i16_v8i16: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxwq %xmm2, %zmm2 +; AVX512F-NEXT: vptestmq %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_2 +; AVX512F-NEXT: ## %bb.1: ## %cond.load +; AVX512F-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_2: ## %else +; AVX512F-NEXT: kshiftrw $1, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_4 +; AVX512F-NEXT: ## %bb.3: ## %cond.load1 +; AVX512F-NEXT: vpinsrw $1, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_4: ## %else2 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxwq %xmm2, %zmm2 +; AVX512F-NEXT: vptestmq %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $2, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_6 +; AVX512F-NEXT: ## %bb.5: ## %cond.load5 +; AVX512F-NEXT: vpinsrw $2, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_6: ## %else6 +; AVX512F-NEXT: kshiftrw $3, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_8 +; AVX512F-NEXT: ## %bb.7: ## %cond.load9 +; AVX512F-NEXT: vpinsrw $3, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_8: ## %else10 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxwq %xmm2, %zmm2 +; AVX512F-NEXT: vptestmq %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $4, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_10 +; AVX512F-NEXT: ## %bb.9: ## %cond.load13 +; AVX512F-NEXT: vpinsrw $4, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_10: ## %else14 +; AVX512F-NEXT: kshiftrw $5, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_12 +; AVX512F-NEXT: ## %bb.11: ## %cond.load17 +; AVX512F-NEXT: vpinsrw $5, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_12: ## %else18 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqw %xmm2, %xmm1, %xmm1 +; AVX512F-NEXT: vpmovsxwq %xmm1, %zmm1 +; AVX512F-NEXT: vptestmq %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftrw $6, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_14 +; AVX512F-NEXT: ## %bb.13: ## %cond.load21 +; AVX512F-NEXT: vpinsrw $6, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: addq $2, %rdi +; AVX512F-NEXT: LBB11_14: ## %else22 +; AVX512F-NEXT: kshiftrw $7, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB11_16 +; AVX512F-NEXT: ## %bb.15: ## %cond.load25 +; AVX512F-NEXT: vpinsrw $7, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: LBB11_16: ## %else26 +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v8i16_v8i16: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vptestnmw %xmm1, %xmm1, %k0 +; AVX512VLBW-NEXT: kmovd %k0, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_2 +; AVX512VLBW-NEXT: ## %bb.1: ## %cond.load +; AVX512VLBW-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_2: ## %else +; AVX512VLBW-NEXT: kshiftrw $1, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_4 +; AVX512VLBW-NEXT: ## %bb.3: ## %cond.load1 +; AVX512VLBW-NEXT: vpinsrw $1, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_4: ## %else2 +; AVX512VLBW-NEXT: kshiftrw $2, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_6 +; AVX512VLBW-NEXT: ## %bb.5: ## %cond.load5 +; AVX512VLBW-NEXT: vpinsrw $2, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_6: ## %else6 +; AVX512VLBW-NEXT: kshiftrw $3, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_8 +; AVX512VLBW-NEXT: ## %bb.7: ## %cond.load9 +; AVX512VLBW-NEXT: vpinsrw $3, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_8: ## %else10 +; AVX512VLBW-NEXT: kshiftrw $4, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_10 +; AVX512VLBW-NEXT: ## %bb.9: ## %cond.load13 +; AVX512VLBW-NEXT: vpinsrw $4, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_10: ## %else14 +; AVX512VLBW-NEXT: kshiftrw $5, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_12 +; AVX512VLBW-NEXT: ## %bb.11: ## %cond.load17 +; AVX512VLBW-NEXT: vpinsrw $5, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_12: ## %else18 +; AVX512VLBW-NEXT: kshiftrw $6, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_14 +; AVX512VLBW-NEXT: ## %bb.13: ## %cond.load21 +; AVX512VLBW-NEXT: vpinsrw $6, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: addq $2, %rdi +; AVX512VLBW-NEXT: LBB11_14: ## %else22 +; AVX512VLBW-NEXT: kshiftrw $7, %k0, %k0 +; AVX512VLBW-NEXT: kmovd %k0, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB11_16 +; AVX512VLBW-NEXT: ## %bb.15: ## %cond.load25 +; AVX512VLBW-NEXT: vpinsrw $7, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: LBB11_16: ## %else26 +; AVX512VLBW-NEXT: retq %mask = icmp eq <8 x i16> %trigger, zeroinitializer %res = call <8 x i16> @llvm.masked.expandload.v8i16(i16* %base, <8 x i1> %mask, <8 x i16> %src0) ret <8 x i16>%res @@ -1297,429 +3469,787 @@ define <8 x i16> @expandload_v8i16_v8i16(i16* %base, <8 x i16> %src0, <8 x i16> ; define <16 x i8> @expandload_v16i8_v16i8(i8* %base, <16 x i8> %src0, <16 x i8> %trigger) { -; HSW-LABEL: expandload_v16i8_v16i8: -; HSW: # %bb.0: -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $0, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_2 -; HSW-NEXT: # %bb.1: # %cond.load -; HSW-NEXT: vpinsrb $0, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_2: # %else -; HSW-NEXT: vpextrb $1, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_4 -; HSW-NEXT: # %bb.3: # %cond.load1 -; HSW-NEXT: vpinsrb $1, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_4: # %else2 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $2, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_6 -; HSW-NEXT: # %bb.5: # %cond.load5 -; HSW-NEXT: vpinsrb $2, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_6: # %else6 -; HSW-NEXT: vpextrb $3, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_8 -; HSW-NEXT: # %bb.7: # %cond.load9 -; HSW-NEXT: vpinsrb $3, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_8: # %else10 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $4, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_10 -; HSW-NEXT: # %bb.9: # %cond.load13 -; HSW-NEXT: vpinsrb $4, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_10: # %else14 -; HSW-NEXT: vpextrb $5, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_12 -; HSW-NEXT: # %bb.11: # %cond.load17 -; HSW-NEXT: vpinsrb $5, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_12: # %else18 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $6, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_14 -; HSW-NEXT: # %bb.13: # %cond.load21 -; HSW-NEXT: vpinsrb $6, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_14: # %else22 -; HSW-NEXT: vpextrb $7, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_16 -; HSW-NEXT: # %bb.15: # %cond.load25 -; HSW-NEXT: vpinsrb $7, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_16: # %else26 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $8, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_18 -; HSW-NEXT: # %bb.17: # %cond.load29 -; HSW-NEXT: vpinsrb $8, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_18: # %else30 -; HSW-NEXT: vpextrb $9, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_20 -; HSW-NEXT: # %bb.19: # %cond.load33 -; HSW-NEXT: vpinsrb $9, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_20: # %else34 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $10, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_22 -; HSW-NEXT: # %bb.21: # %cond.load37 -; HSW-NEXT: vpinsrb $10, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_22: # %else38 -; HSW-NEXT: vpextrb $11, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_24 -; HSW-NEXT: # %bb.23: # %cond.load41 -; HSW-NEXT: vpinsrb $11, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_24: # %else42 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; HSW-NEXT: vpextrb $12, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_26 -; HSW-NEXT: # %bb.25: # %cond.load45 -; HSW-NEXT: vpinsrb $12, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_26: # %else46 -; HSW-NEXT: vpextrb $13, %xmm2, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_28 -; HSW-NEXT: # %bb.27: # %cond.load49 -; HSW-NEXT: vpinsrb $13, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_28: # %else50 -; HSW-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; HSW-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1 -; HSW-NEXT: vpextrb $14, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_30 -; HSW-NEXT: # %bb.29: # %cond.load53 -; HSW-NEXT: vpinsrb $14, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: incq %rdi -; HSW-NEXT: .LBB12_30: # %else54 -; HSW-NEXT: vpextrb $15, %xmm1, %eax -; HSW-NEXT: testb $1, %al -; HSW-NEXT: je .LBB12_32 -; HSW-NEXT: # %bb.31: # %cond.load57 -; HSW-NEXT: vpinsrb $15, (%rdi), %xmm0, %xmm0 -; HSW-NEXT: .LBB12_32: # %else58 -; HSW-NEXT: retq -; -; SKX-LABEL: expandload_v16i8_v16i8: -; SKX: # %bb.0: -; SKX-NEXT: vptestnmb %xmm1, %xmm1, %k0 -; SKX-NEXT: kmovd %k0, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_2 -; SKX-NEXT: # %bb.1: # %cond.load -; SKX-NEXT: vpinsrb $0, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_2: # %else -; SKX-NEXT: kshiftrw $1, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_4 -; SKX-NEXT: # %bb.3: # %cond.load1 -; SKX-NEXT: vpinsrb $1, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_4: # %else2 -; SKX-NEXT: kshiftrw $2, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_6 -; SKX-NEXT: # %bb.5: # %cond.load5 -; SKX-NEXT: vpinsrb $2, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_6: # %else6 -; SKX-NEXT: kshiftrw $3, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_8 -; SKX-NEXT: # %bb.7: # %cond.load9 -; SKX-NEXT: vpinsrb $3, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_8: # %else10 -; SKX-NEXT: kshiftrw $4, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_10 -; SKX-NEXT: # %bb.9: # %cond.load13 -; SKX-NEXT: vpinsrb $4, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_10: # %else14 -; SKX-NEXT: kshiftrw $5, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_12 -; SKX-NEXT: # %bb.11: # %cond.load17 -; SKX-NEXT: vpinsrb $5, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_12: # %else18 -; SKX-NEXT: kshiftrw $6, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_14 -; SKX-NEXT: # %bb.13: # %cond.load21 -; SKX-NEXT: vpinsrb $6, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_14: # %else22 -; SKX-NEXT: kshiftrw $7, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_16 -; SKX-NEXT: # %bb.15: # %cond.load25 -; SKX-NEXT: vpinsrb $7, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_16: # %else26 -; SKX-NEXT: kshiftrw $8, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_18 -; SKX-NEXT: # %bb.17: # %cond.load29 -; SKX-NEXT: vpinsrb $8, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_18: # %else30 -; SKX-NEXT: kshiftrw $9, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_20 -; SKX-NEXT: # %bb.19: # %cond.load33 -; SKX-NEXT: vpinsrb $9, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_20: # %else34 -; SKX-NEXT: kshiftrw $10, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_22 -; SKX-NEXT: # %bb.21: # %cond.load37 -; SKX-NEXT: vpinsrb $10, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_22: # %else38 -; SKX-NEXT: kshiftrw $11, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_24 -; SKX-NEXT: # %bb.23: # %cond.load41 -; SKX-NEXT: vpinsrb $11, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_24: # %else42 -; SKX-NEXT: kshiftrw $12, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_26 -; SKX-NEXT: # %bb.25: # %cond.load45 -; SKX-NEXT: vpinsrb $12, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_26: # %else46 -; SKX-NEXT: kshiftrw $13, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_28 -; SKX-NEXT: # %bb.27: # %cond.load49 -; SKX-NEXT: vpinsrb $13, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_28: # %else50 -; SKX-NEXT: kshiftrw $14, %k0, %k1 -; SKX-NEXT: kmovd %k1, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_30 -; SKX-NEXT: # %bb.29: # %cond.load53 -; SKX-NEXT: vpinsrb $14, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: incq %rdi -; SKX-NEXT: .LBB12_30: # %else54 -; SKX-NEXT: kshiftrw $15, %k0, %k0 -; SKX-NEXT: kmovd %k0, %eax -; SKX-NEXT: testb $1, %al -; SKX-NEXT: je .LBB12_32 -; SKX-NEXT: # %bb.31: # %cond.load57 -; SKX-NEXT: vpinsrb $15, (%rdi), %xmm0, %xmm0 -; SKX-NEXT: .LBB12_32: # %else58 -; SKX-NEXT: retq -; -; KNL-LABEL: expandload_v16i8_v16i8: -; KNL: # %bb.0: -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_2 -; KNL-NEXT: # %bb.1: # %cond.load -; KNL-NEXT: vpinsrb $0, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_2: # %else -; KNL-NEXT: kshiftrw $1, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_4 -; KNL-NEXT: # %bb.3: # %cond.load1 -; KNL-NEXT: vpinsrb $1, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_4: # %else2 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $2, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_6 -; KNL-NEXT: # %bb.5: # %cond.load5 -; KNL-NEXT: vpinsrb $2, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_6: # %else6 -; KNL-NEXT: kshiftrw $3, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_8 -; KNL-NEXT: # %bb.7: # %cond.load9 -; KNL-NEXT: vpinsrb $3, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_8: # %else10 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $4, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_10 -; KNL-NEXT: # %bb.9: # %cond.load13 -; KNL-NEXT: vpinsrb $4, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_10: # %else14 -; KNL-NEXT: kshiftrw $5, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_12 -; KNL-NEXT: # %bb.11: # %cond.load17 -; KNL-NEXT: vpinsrb $5, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_12: # %else18 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $6, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_14 -; KNL-NEXT: # %bb.13: # %cond.load21 -; KNL-NEXT: vpinsrb $6, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_14: # %else22 -; KNL-NEXT: kshiftrw $7, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_16 -; KNL-NEXT: # %bb.15: # %cond.load25 -; KNL-NEXT: vpinsrb $7, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_16: # %else26 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $8, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_18 -; KNL-NEXT: # %bb.17: # %cond.load29 -; KNL-NEXT: vpinsrb $8, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_18: # %else30 -; KNL-NEXT: kshiftrw $9, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_20 -; KNL-NEXT: # %bb.19: # %cond.load33 -; KNL-NEXT: vpinsrb $9, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_20: # %else34 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $10, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_22 -; KNL-NEXT: # %bb.21: # %cond.load37 -; KNL-NEXT: vpinsrb $10, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_22: # %else38 -; KNL-NEXT: kshiftrw $11, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_24 -; KNL-NEXT: # %bb.23: # %cond.load41 -; KNL-NEXT: vpinsrb $11, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_24: # %else42 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 -; KNL-NEXT: vpmovsxbd %xmm2, %zmm2 -; KNL-NEXT: vptestmd %zmm2, %zmm2, %k0 -; KNL-NEXT: kshiftrw $12, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_26 -; KNL-NEXT: # %bb.25: # %cond.load45 -; KNL-NEXT: vpinsrb $12, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_26: # %else46 -; KNL-NEXT: kshiftrw $13, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_28 -; KNL-NEXT: # %bb.27: # %cond.load49 -; KNL-NEXT: vpinsrb $13, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_28: # %else50 -; KNL-NEXT: vpxor %xmm2, %xmm2, %xmm2 -; KNL-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1 -; KNL-NEXT: vpmovsxbd %xmm1, %zmm1 -; KNL-NEXT: vptestmd %zmm1, %zmm1, %k0 -; KNL-NEXT: kshiftrw $14, %k0, %k1 -; KNL-NEXT: kmovw %k1, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_30 -; KNL-NEXT: # %bb.29: # %cond.load53 -; KNL-NEXT: vpinsrb $14, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: addq $1, %rdi -; KNL-NEXT: .LBB12_30: # %else54 -; KNL-NEXT: kshiftrw $15, %k0, %k0 -; KNL-NEXT: kmovw %k0, %eax -; KNL-NEXT: testb $1, %al -; KNL-NEXT: je .LBB12_32 -; KNL-NEXT: # %bb.31: # %cond.load57 -; KNL-NEXT: vpinsrb $15, (%rdi), %xmm0, %xmm0 -; KNL-NEXT: .LBB12_32: # %else58 -; KNL-NEXT: retq +; SSE2-LABEL: expandload_v16i8_v16i8: +; SSE2: ## %bb.0: +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: movd %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_2 +; SSE2-NEXT: ## %bb.1: ## %cond.load +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %ecx +; SSE2-NEXT: movd %ecx, %xmm3 +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_2: ## %else +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_4 +; SSE2-NEXT: ## %bb.3: ## %cond.load1 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,0,255,255,255,255,255,255,255,255,255,255,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %eax +; SSE2-NEXT: movd %eax, %xmm3 +; SSE2-NEXT: psllw $8, %xmm3 +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_4: ## %else2 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: movd %xmm2, %eax +; SSE2-NEXT: movl %eax, %ecx +; SSE2-NEXT: shrl $16, %ecx +; SSE2-NEXT: testb $1, %cl +; SSE2-NEXT: je LBB12_6 +; SSE2-NEXT: ## %bb.5: ## %cond.load5 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,0,255,255,255,255,255,255,255,255,255,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %ecx +; SSE2-NEXT: movd %ecx, %xmm3 +; SSE2-NEXT: pslld $16, %xmm3 +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_6: ## %else6 +; SSE2-NEXT: shrl $24, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_8 +; SSE2-NEXT: ## %bb.7: ## %cond.load9 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,0,255,255,255,255,255,255,255,255,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %eax +; SSE2-NEXT: movd %eax, %xmm3 +; SSE2-NEXT: pslld $24, %xmm3 +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_8: ## %else10 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: pextrw $2, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_10 +; SSE2-NEXT: ## %bb.9: ## %cond.load13 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,0,255,255,255,255,255,255,255,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %ecx +; SSE2-NEXT: movd %ecx, %xmm3 +; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,0,1,1] +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_10: ## %else14 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_12 +; SSE2-NEXT: ## %bb.11: ## %cond.load17 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,0,255,255,255,255,255,255,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %eax +; SSE2-NEXT: movd %eax, %xmm3 +; SSE2-NEXT: psllq $40, %xmm3 +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_12: ## %else18 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: pextrw $3, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_14 +; SSE2-NEXT: ## %bb.13: ## %cond.load21 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,0,255,255,255,255,255,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %ecx +; SSE2-NEXT: movd %ecx, %xmm3 +; SSE2-NEXT: psllq $48, %xmm3 +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_14: ## %else22 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_16 +; SSE2-NEXT: ## %bb.15: ## %cond.load25 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %eax +; SSE2-NEXT: movd %eax, %xmm3 +; SSE2-NEXT: psllq $56, %xmm3 +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_16: ## %else26 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: pextrw $4, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_18 +; SSE2-NEXT: ## %bb.17: ## %cond.load29 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %ecx +; SSE2-NEXT: movd %ecx, %xmm3 +; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,1,0,1] +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_18: ## %else30 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_20 +; SSE2-NEXT: ## %bb.19: ## %cond.load33 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255,255,0,255,255,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %eax +; SSE2-NEXT: movd %eax, %xmm3 +; SSE2-NEXT: pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1,2,3,4,5,6] +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_20: ## %else34 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: pextrw $5, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_22 +; SSE2-NEXT: ## %bb.21: ## %cond.load37 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255,255,255,0,255,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %ecx +; SSE2-NEXT: movd %ecx, %xmm3 +; SSE2-NEXT: pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1,2,3,4,5] +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_22: ## %else38 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_24 +; SSE2-NEXT: ## %bb.23: ## %cond.load41 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255,255,255,255,0,255,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %eax +; SSE2-NEXT: movd %eax, %xmm3 +; SSE2-NEXT: pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1,2,3,4] +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_24: ## %else42 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE2-NEXT: pextrw $6, %xmm2, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_26 +; SSE2-NEXT: ## %bb.25: ## %cond.load45 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %ecx +; SSE2-NEXT: movd %ecx, %xmm3 +; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,1,2,0] +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_26: ## %else46 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_28 +; SSE2-NEXT: ## %bb.27: ## %cond.load49 +; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255,255,255,255,255,255,0,255,255] +; SSE2-NEXT: pand %xmm2, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %eax +; SSE2-NEXT: movd %eax, %xmm3 +; SSE2-NEXT: pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1,2] +; SSE2-NEXT: pandn %xmm3, %xmm2 +; SSE2-NEXT: por %xmm2, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_28: ## %else50 +; SSE2-NEXT: pxor %xmm2, %xmm2 +; SSE2-NEXT: pcmpeqb %xmm2, %xmm1 +; SSE2-NEXT: pextrw $7, %xmm1, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_30 +; SSE2-NEXT: ## %bb.29: ## %cond.load53 +; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,255] +; SSE2-NEXT: pand %xmm1, %xmm0 +; SSE2-NEXT: movzbl (%rdi), %ecx +; SSE2-NEXT: movd %ecx, %xmm2 +; SSE2-NEXT: pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1] +; SSE2-NEXT: pandn %xmm2, %xmm1 +; SSE2-NEXT: por %xmm1, %xmm0 +; SSE2-NEXT: incq %rdi +; SSE2-NEXT: LBB12_30: ## %else54 +; SSE2-NEXT: shrl $8, %eax +; SSE2-NEXT: testb $1, %al +; SSE2-NEXT: je LBB12_32 +; SSE2-NEXT: ## %bb.31: ## %cond.load57 +; SSE2-NEXT: pand {{.*}}(%rip), %xmm0 +; SSE2-NEXT: movzbl (%rdi), %eax +; SSE2-NEXT: movd %eax, %xmm1 +; SSE2-NEXT: pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0] +; SSE2-NEXT: por %xmm1, %xmm0 +; SSE2-NEXT: LBB12_32: ## %else58 +; SSE2-NEXT: retq +; +; SSE42-LABEL: expandload_v16i8_v16i8: +; SSE42: ## %bb.0: +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $0, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_2 +; SSE42-NEXT: ## %bb.1: ## %cond.load +; SSE42-NEXT: pinsrb $0, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_2: ## %else +; SSE42-NEXT: pextrb $1, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_4 +; SSE42-NEXT: ## %bb.3: ## %cond.load1 +; SSE42-NEXT: pinsrb $1, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_4: ## %else2 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $2, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_6 +; SSE42-NEXT: ## %bb.5: ## %cond.load5 +; SSE42-NEXT: pinsrb $2, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_6: ## %else6 +; SSE42-NEXT: pextrb $3, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_8 +; SSE42-NEXT: ## %bb.7: ## %cond.load9 +; SSE42-NEXT: pinsrb $3, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_8: ## %else10 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $4, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_10 +; SSE42-NEXT: ## %bb.9: ## %cond.load13 +; SSE42-NEXT: pinsrb $4, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_10: ## %else14 +; SSE42-NEXT: pextrb $5, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_12 +; SSE42-NEXT: ## %bb.11: ## %cond.load17 +; SSE42-NEXT: pinsrb $5, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_12: ## %else18 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $6, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_14 +; SSE42-NEXT: ## %bb.13: ## %cond.load21 +; SSE42-NEXT: pinsrb $6, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_14: ## %else22 +; SSE42-NEXT: pextrb $7, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_16 +; SSE42-NEXT: ## %bb.15: ## %cond.load25 +; SSE42-NEXT: pinsrb $7, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_16: ## %else26 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $8, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_18 +; SSE42-NEXT: ## %bb.17: ## %cond.load29 +; SSE42-NEXT: pinsrb $8, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_18: ## %else30 +; SSE42-NEXT: pextrb $9, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_20 +; SSE42-NEXT: ## %bb.19: ## %cond.load33 +; SSE42-NEXT: pinsrb $9, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_20: ## %else34 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $10, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_22 +; SSE42-NEXT: ## %bb.21: ## %cond.load37 +; SSE42-NEXT: pinsrb $10, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_22: ## %else38 +; SSE42-NEXT: pextrb $11, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_24 +; SSE42-NEXT: ## %bb.23: ## %cond.load41 +; SSE42-NEXT: pinsrb $11, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_24: ## %else42 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm1, %xmm2 +; SSE42-NEXT: pextrb $12, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_26 +; SSE42-NEXT: ## %bb.25: ## %cond.load45 +; SSE42-NEXT: pinsrb $12, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_26: ## %else46 +; SSE42-NEXT: pextrb $13, %xmm2, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_28 +; SSE42-NEXT: ## %bb.27: ## %cond.load49 +; SSE42-NEXT: pinsrb $13, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_28: ## %else50 +; SSE42-NEXT: pxor %xmm2, %xmm2 +; SSE42-NEXT: pcmpeqb %xmm2, %xmm1 +; SSE42-NEXT: pextrb $14, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_30 +; SSE42-NEXT: ## %bb.29: ## %cond.load53 +; SSE42-NEXT: pinsrb $14, (%rdi), %xmm0 +; SSE42-NEXT: incq %rdi +; SSE42-NEXT: LBB12_30: ## %else54 +; SSE42-NEXT: pextrb $15, %xmm1, %eax +; SSE42-NEXT: testb $1, %al +; SSE42-NEXT: je LBB12_32 +; SSE42-NEXT: ## %bb.31: ## %cond.load57 +; SSE42-NEXT: pinsrb $15, (%rdi), %xmm0 +; SSE42-NEXT: LBB12_32: ## %else58 +; SSE42-NEXT: retq +; +; AVX1OR2-LABEL: expandload_v16i8_v16i8: +; AVX1OR2: ## %bb.0: +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $0, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_2 +; AVX1OR2-NEXT: ## %bb.1: ## %cond.load +; AVX1OR2-NEXT: vpinsrb $0, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_2: ## %else +; AVX1OR2-NEXT: vpextrb $1, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_4 +; AVX1OR2-NEXT: ## %bb.3: ## %cond.load1 +; AVX1OR2-NEXT: vpinsrb $1, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_4: ## %else2 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $2, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_6 +; AVX1OR2-NEXT: ## %bb.5: ## %cond.load5 +; AVX1OR2-NEXT: vpinsrb $2, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_6: ## %else6 +; AVX1OR2-NEXT: vpextrb $3, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_8 +; AVX1OR2-NEXT: ## %bb.7: ## %cond.load9 +; AVX1OR2-NEXT: vpinsrb $3, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_8: ## %else10 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $4, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_10 +; AVX1OR2-NEXT: ## %bb.9: ## %cond.load13 +; AVX1OR2-NEXT: vpinsrb $4, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_10: ## %else14 +; AVX1OR2-NEXT: vpextrb $5, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_12 +; AVX1OR2-NEXT: ## %bb.11: ## %cond.load17 +; AVX1OR2-NEXT: vpinsrb $5, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_12: ## %else18 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $6, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_14 +; AVX1OR2-NEXT: ## %bb.13: ## %cond.load21 +; AVX1OR2-NEXT: vpinsrb $6, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_14: ## %else22 +; AVX1OR2-NEXT: vpextrb $7, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_16 +; AVX1OR2-NEXT: ## %bb.15: ## %cond.load25 +; AVX1OR2-NEXT: vpinsrb $7, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_16: ## %else26 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $8, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_18 +; AVX1OR2-NEXT: ## %bb.17: ## %cond.load29 +; AVX1OR2-NEXT: vpinsrb $8, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_18: ## %else30 +; AVX1OR2-NEXT: vpextrb $9, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_20 +; AVX1OR2-NEXT: ## %bb.19: ## %cond.load33 +; AVX1OR2-NEXT: vpinsrb $9, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_20: ## %else34 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $10, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_22 +; AVX1OR2-NEXT: ## %bb.21: ## %cond.load37 +; AVX1OR2-NEXT: vpinsrb $10, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_22: ## %else38 +; AVX1OR2-NEXT: vpextrb $11, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_24 +; AVX1OR2-NEXT: ## %bb.23: ## %cond.load41 +; AVX1OR2-NEXT: vpinsrb $11, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_24: ## %else42 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX1OR2-NEXT: vpextrb $12, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_26 +; AVX1OR2-NEXT: ## %bb.25: ## %cond.load45 +; AVX1OR2-NEXT: vpinsrb $12, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_26: ## %else46 +; AVX1OR2-NEXT: vpextrb $13, %xmm2, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_28 +; AVX1OR2-NEXT: ## %bb.27: ## %cond.load49 +; AVX1OR2-NEXT: vpinsrb $13, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_28: ## %else50 +; AVX1OR2-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX1OR2-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1 +; AVX1OR2-NEXT: vpextrb $14, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_30 +; AVX1OR2-NEXT: ## %bb.29: ## %cond.load53 +; AVX1OR2-NEXT: vpinsrb $14, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: incq %rdi +; AVX1OR2-NEXT: LBB12_30: ## %else54 +; AVX1OR2-NEXT: vpextrb $15, %xmm1, %eax +; AVX1OR2-NEXT: testb $1, %al +; AVX1OR2-NEXT: je LBB12_32 +; AVX1OR2-NEXT: ## %bb.31: ## %cond.load57 +; AVX1OR2-NEXT: vpinsrb $15, (%rdi), %xmm0, %xmm0 +; AVX1OR2-NEXT: LBB12_32: ## %else58 +; AVX1OR2-NEXT: retq +; +; AVX512F-LABEL: expandload_v16i8_v16i8: +; AVX512F: ## %bb.0: +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_2 +; AVX512F-NEXT: ## %bb.1: ## %cond.load +; AVX512F-NEXT: vpinsrb $0, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_2: ## %else +; AVX512F-NEXT: kshiftrw $1, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_4 +; AVX512F-NEXT: ## %bb.3: ## %cond.load1 +; AVX512F-NEXT: vpinsrb $1, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_4: ## %else2 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $2, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_6 +; AVX512F-NEXT: ## %bb.5: ## %cond.load5 +; AVX512F-NEXT: vpinsrb $2, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_6: ## %else6 +; AVX512F-NEXT: kshiftrw $3, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_8 +; AVX512F-NEXT: ## %bb.7: ## %cond.load9 +; AVX512F-NEXT: vpinsrb $3, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_8: ## %else10 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $4, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_10 +; AVX512F-NEXT: ## %bb.9: ## %cond.load13 +; AVX512F-NEXT: vpinsrb $4, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_10: ## %else14 +; AVX512F-NEXT: kshiftrw $5, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_12 +; AVX512F-NEXT: ## %bb.11: ## %cond.load17 +; AVX512F-NEXT: vpinsrb $5, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_12: ## %else18 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $6, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_14 +; AVX512F-NEXT: ## %bb.13: ## %cond.load21 +; AVX512F-NEXT: vpinsrb $6, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_14: ## %else22 +; AVX512F-NEXT: kshiftrw $7, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_16 +; AVX512F-NEXT: ## %bb.15: ## %cond.load25 +; AVX512F-NEXT: vpinsrb $7, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_16: ## %else26 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $8, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_18 +; AVX512F-NEXT: ## %bb.17: ## %cond.load29 +; AVX512F-NEXT: vpinsrb $8, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_18: ## %else30 +; AVX512F-NEXT: kshiftrw $9, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_20 +; AVX512F-NEXT: ## %bb.19: ## %cond.load33 +; AVX512F-NEXT: vpinsrb $9, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_20: ## %else34 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $10, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_22 +; AVX512F-NEXT: ## %bb.21: ## %cond.load37 +; AVX512F-NEXT: vpinsrb $10, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_22: ## %else38 +; AVX512F-NEXT: kshiftrw $11, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_24 +; AVX512F-NEXT: ## %bb.23: ## %cond.load41 +; AVX512F-NEXT: vpinsrb $11, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_24: ## %else42 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm2 +; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm2 +; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k0 +; AVX512F-NEXT: kshiftrw $12, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_26 +; AVX512F-NEXT: ## %bb.25: ## %cond.load45 +; AVX512F-NEXT: vpinsrb $12, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_26: ## %else46 +; AVX512F-NEXT: kshiftrw $13, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_28 +; AVX512F-NEXT: ## %bb.27: ## %cond.load49 +; AVX512F-NEXT: vpinsrb $13, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_28: ## %else50 +; AVX512F-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm1 +; AVX512F-NEXT: vpmovsxbd %xmm1, %zmm1 +; AVX512F-NEXT: vptestmd %zmm1, %zmm1, %k0 +; AVX512F-NEXT: kshiftrw $14, %k0, %k1 +; AVX512F-NEXT: kmovw %k1, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_30 +; AVX512F-NEXT: ## %bb.29: ## %cond.load53 +; AVX512F-NEXT: vpinsrb $14, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: incq %rdi +; AVX512F-NEXT: LBB12_30: ## %else54 +; AVX512F-NEXT: kshiftrw $15, %k0, %k0 +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: testb $1, %al +; AVX512F-NEXT: je LBB12_32 +; AVX512F-NEXT: ## %bb.31: ## %cond.load57 +; AVX512F-NEXT: vpinsrb $15, (%rdi), %xmm0, %xmm0 +; AVX512F-NEXT: LBB12_32: ## %else58 +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: retq +; +; AVX512VLBW-LABEL: expandload_v16i8_v16i8: +; AVX512VLBW: ## %bb.0: +; AVX512VLBW-NEXT: vptestnmb %xmm1, %xmm1, %k0 +; AVX512VLBW-NEXT: kmovd %k0, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_2 +; AVX512VLBW-NEXT: ## %bb.1: ## %cond.load +; AVX512VLBW-NEXT: vpinsrb $0, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_2: ## %else +; AVX512VLBW-NEXT: kshiftrw $1, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_4 +; AVX512VLBW-NEXT: ## %bb.3: ## %cond.load1 +; AVX512VLBW-NEXT: vpinsrb $1, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_4: ## %else2 +; AVX512VLBW-NEXT: kshiftrw $2, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_6 +; AVX512VLBW-NEXT: ## %bb.5: ## %cond.load5 +; AVX512VLBW-NEXT: vpinsrb $2, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_6: ## %else6 +; AVX512VLBW-NEXT: kshiftrw $3, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_8 +; AVX512VLBW-NEXT: ## %bb.7: ## %cond.load9 +; AVX512VLBW-NEXT: vpinsrb $3, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_8: ## %else10 +; AVX512VLBW-NEXT: kshiftrw $4, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_10 +; AVX512VLBW-NEXT: ## %bb.9: ## %cond.load13 +; AVX512VLBW-NEXT: vpinsrb $4, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_10: ## %else14 +; AVX512VLBW-NEXT: kshiftrw $5, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_12 +; AVX512VLBW-NEXT: ## %bb.11: ## %cond.load17 +; AVX512VLBW-NEXT: vpinsrb $5, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_12: ## %else18 +; AVX512VLBW-NEXT: kshiftrw $6, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_14 +; AVX512VLBW-NEXT: ## %bb.13: ## %cond.load21 +; AVX512VLBW-NEXT: vpinsrb $6, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_14: ## %else22 +; AVX512VLBW-NEXT: kshiftrw $7, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_16 +; AVX512VLBW-NEXT: ## %bb.15: ## %cond.load25 +; AVX512VLBW-NEXT: vpinsrb $7, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_16: ## %else26 +; AVX512VLBW-NEXT: kshiftrw $8, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_18 +; AVX512VLBW-NEXT: ## %bb.17: ## %cond.load29 +; AVX512VLBW-NEXT: vpinsrb $8, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_18: ## %else30 +; AVX512VLBW-NEXT: kshiftrw $9, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_20 +; AVX512VLBW-NEXT: ## %bb.19: ## %cond.load33 +; AVX512VLBW-NEXT: vpinsrb $9, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_20: ## %else34 +; AVX512VLBW-NEXT: kshiftrw $10, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_22 +; AVX512VLBW-NEXT: ## %bb.21: ## %cond.load37 +; AVX512VLBW-NEXT: vpinsrb $10, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_22: ## %else38 +; AVX512VLBW-NEXT: kshiftrw $11, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_24 +; AVX512VLBW-NEXT: ## %bb.23: ## %cond.load41 +; AVX512VLBW-NEXT: vpinsrb $11, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_24: ## %else42 +; AVX512VLBW-NEXT: kshiftrw $12, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_26 +; AVX512VLBW-NEXT: ## %bb.25: ## %cond.load45 +; AVX512VLBW-NEXT: vpinsrb $12, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_26: ## %else46 +; AVX512VLBW-NEXT: kshiftrw $13, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_28 +; AVX512VLBW-NEXT: ## %bb.27: ## %cond.load49 +; AVX512VLBW-NEXT: vpinsrb $13, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_28: ## %else50 +; AVX512VLBW-NEXT: kshiftrw $14, %k0, %k1 +; AVX512VLBW-NEXT: kmovd %k1, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_30 +; AVX512VLBW-NEXT: ## %bb.29: ## %cond.load53 +; AVX512VLBW-NEXT: vpinsrb $14, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: incq %rdi +; AVX512VLBW-NEXT: LBB12_30: ## %else54 +; AVX512VLBW-NEXT: kshiftrw $15, %k0, %k0 +; AVX512VLBW-NEXT: kmovd %k0, %eax +; AVX512VLBW-NEXT: testb $1, %al +; AVX512VLBW-NEXT: je LBB12_32 +; AVX512VLBW-NEXT: ## %bb.31: ## %cond.load57 +; AVX512VLBW-NEXT: vpinsrb $15, (%rdi), %xmm0, %xmm0 +; AVX512VLBW-NEXT: LBB12_32: ## %else58 +; AVX512VLBW-NEXT: retq %mask = icmp eq <16 x i8> %trigger, zeroinitializer %res = call <16 x i8> @llvm.masked.expandload.v16i8(i8* %base, <16 x i1> %mask, <16 x i8> %src0) ret <16 x i8>%res -- 2.7.4