From 8b9c4f2855067bdf56f52ca68e52f8f0add1088d Mon Sep 17 00:00:00 2001 From: Simon Pilgrim Date: Mon, 7 Aug 2017 15:50:43 +0000 Subject: [PATCH] [X86][AVX] Cleanup subvector broadcast tests - remove old prefixes. llvm-svn: 310265 --- llvm/test/CodeGen/X86/subvector-broadcast.ll | 366 +++++++++++++-------------- 1 file changed, 183 insertions(+), 183 deletions(-) diff --git a/llvm/test/CodeGen/X86/subvector-broadcast.ll b/llvm/test/CodeGen/X86/subvector-broadcast.ll index 09c76e6..d826509 100644 --- a/llvm/test/CodeGen/X86/subvector-broadcast.ll +++ b/llvm/test/CodeGen/X86/subvector-broadcast.ll @@ -1,14 +1,14 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+avx | FileCheck %s --check-prefix=X32 --check-prefix=X32-AVX --check-prefix=X32-AVX1 -; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+avx2 | FileCheck %s --check-prefix=X32 --check-prefix=X32-AVX --check-prefix=X32-AVX2 -; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+avx512vl | FileCheck %s --check-prefix=X32 --check-prefix=X32-AVX512 --check-prefix=X32-AVX512F -; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+avx512bw,+avx512vl | FileCheck %s --check-prefix=X32 --check-prefix=X32-AVX512 --check-prefix=X32-AVX512BW -; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+avx512dq,+avx512vl | FileCheck %s --check-prefix=X32 --check-prefix=X32-AVX512 --check-prefix=X32-AVX512DQ -; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx | FileCheck %s --check-prefix=X64 --check-prefix=X64-AVX --check-prefix=X64-AVX1 -; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx2 | FileCheck %s --check-prefix=X64 --check-prefix=X64-AVX --check-prefix=X64-AVX2 -; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=X64-AVX512 --check-prefix=X64-AVX512F -; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512bw,+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=X64-AVX512 --check-prefix=X64-AVX512BW -; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512dq,+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=X64-AVX512 --check-prefix=X64-AVX512DQ +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X32 --check-prefix=X32-AVX --check-prefix=X32-AVX1 +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X32 --check-prefix=X32-AVX --check-prefix=X32-AVX2 +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=X32 --check-prefix=X32-AVX512 --check-prefix=X32-AVX512F +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512bw,+avx512vl | FileCheck %s --check-prefix=X32 --check-prefix=X32-AVX512 --check-prefix=X32-AVX512BW +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512dq,+avx512vl | FileCheck %s --check-prefix=X32 --check-prefix=X32-AVX512 --check-prefix=X32-AVX512DQ +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X64 --check-prefix=X64-AVX --check-prefix=X64-AVX1 +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X64 --check-prefix=X64-AVX --check-prefix=X64-AVX2 +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=X64-AVX512 --check-prefix=X64-AVX512F +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=X64-AVX512 --check-prefix=X64-AVX512BW +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq,+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=X64-AVX512 --check-prefix=X64-AVX512DQ ; ; Subvector Load + Broadcast @@ -16,46 +16,46 @@ define <4 x double> @test_broadcast_2f64_4f64(<2 x double> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_2f64_4f64: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_2f64_4f64: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_2f64_4f64: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_2f64_4f64: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: vbroadcastf64x2 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_2f64_4f64: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_2f64_4f64: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_2f64_4f64: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_2f64_4f64: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vbroadcastf64x2 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512DQ-NEXT: retq %1 = load <2 x double>, <2 x double> *%p @@ -65,48 +65,48 @@ define <4 x double> @test_broadcast_2f64_4f64(<2 x double> *%p) nounwind { define <8 x double> @test_broadcast_2f64_8f64(<2 x double> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_2f64_8f64: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX-NEXT: vmovdqa %ymm0, %ymm1 ; X32-AVX-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_2f64_8f64: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: vbroadcastf32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_2f64_8f64: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: vbroadcastf32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_2f64_8f64: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: vbroadcastf64x2 {{.*#+}} zmm0 = mem[0,1,0,1,0,1,0,1] ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_2f64_8f64: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX-NEXT: vmovdqa %ymm0, %ymm1 ; X64-AVX-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_2f64_8f64: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vbroadcastf32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_2f64_8f64: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vbroadcastf32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_2f64_8f64: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vbroadcastf64x2 {{.*#+}} zmm0 = mem[0,1,0,1,0,1,0,1] ; X64-AVX512DQ-NEXT: retq %1 = load <2 x double>, <2 x double> *%p @@ -116,26 +116,26 @@ define <8 x double> @test_broadcast_2f64_8f64(<2 x double> *%p) nounwind { define <8 x double> @test_broadcast_4f64_8f64(<4 x double> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_4f64_8f64: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vmovaps (%eax), %ymm0 ; X32-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX-NEXT: retl ; ; X32-AVX512-LABEL: test_broadcast_4f64_8f64: -; X32-AVX512: ## BB#0: +; X32-AVX512: # BB#0: ; X32-AVX512-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512-NEXT: vbroadcastf64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X32-AVX512-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_4f64_8f64: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vmovaps (%rdi), %ymm0 ; X64-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX-NEXT: retq ; ; X64-AVX512-LABEL: test_broadcast_4f64_8f64: -; X64-AVX512: ## BB#0: +; X64-AVX512: # BB#0: ; X64-AVX512-NEXT: vbroadcastf64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X64-AVX512-NEXT: retq %1 = load <4 x double>, <4 x double> *%p @@ -145,46 +145,46 @@ define <8 x double> @test_broadcast_4f64_8f64(<4 x double> *%p) nounwind { define <4 x i64> @test_broadcast_2i64_4i64(<2 x i64> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_2i64_4i64: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_2i64_4i64: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_2i64_4i64: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_2i64_4i64: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_2i64_4i64: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_2i64_4i64: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_2i64_4i64: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_2i64_4i64: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512DQ-NEXT: retq %1 = load <2 x i64>, <2 x i64> *%p @@ -194,61 +194,61 @@ define <4 x i64> @test_broadcast_2i64_4i64(<2 x i64> *%p) nounwind { define <8 x i64> @test_broadcast_2i64_8i64(<2 x i64> *%p) nounwind { ; X32-AVX1-LABEL: test_broadcast_2i64_8i64: -; X32-AVX1: ## BB#0: +; X32-AVX1: # BB#0: ; X32-AVX1-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX1-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX1-NEXT: vmovdqa %ymm0, %ymm1 ; X32-AVX1-NEXT: retl ; ; X32-AVX2-LABEL: test_broadcast_2i64_8i64: -; X32-AVX2: ## BB#0: +; X32-AVX2: # BB#0: ; X32-AVX2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX2-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX2-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX2-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_2i64_8i64: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_2i64_8i64: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_2i64_8i64: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 = mem[0,1,0,1,0,1,0,1] ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX1-LABEL: test_broadcast_2i64_8i64: -; X64-AVX1: ## BB#0: +; X64-AVX1: # BB#0: ; X64-AVX1-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX1-NEXT: vmovdqa %ymm0, %ymm1 ; X64-AVX1-NEXT: retq ; ; X64-AVX2-LABEL: test_broadcast_2i64_8i64: -; X64-AVX2: ## BB#0: +; X64-AVX2: # BB#0: ; X64-AVX2-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX2-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX2-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_2i64_8i64: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_2i64_8i64: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_2i64_8i64: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 = mem[0,1,0,1,0,1,0,1] ; X64-AVX512DQ-NEXT: retq %1 = load <2 x i64>, <2 x i64> *%p @@ -258,26 +258,26 @@ define <8 x i64> @test_broadcast_2i64_8i64(<2 x i64> *%p) nounwind { define <8 x i64> @test_broadcast_4i64_8i64(<4 x i64> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_4i64_8i64: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vmovaps (%eax), %ymm0 ; X32-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX-NEXT: retl ; ; X32-AVX512-LABEL: test_broadcast_4i64_8i64: -; X32-AVX512: ## BB#0: +; X32-AVX512: # BB#0: ; X32-AVX512-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X32-AVX512-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_4i64_8i64: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vmovaps (%rdi), %ymm0 ; X64-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX-NEXT: retq ; ; X64-AVX512-LABEL: test_broadcast_4i64_8i64: -; X64-AVX512: ## BB#0: +; X64-AVX512: # BB#0: ; X64-AVX512-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X64-AVX512-NEXT: retq %1 = load <4 x i64>, <4 x i64> *%p @@ -287,13 +287,13 @@ define <8 x i64> @test_broadcast_4i64_8i64(<4 x i64> *%p) nounwind { define <8 x float> @test_broadcast_4f32_8f32(<4 x float> *%p) nounwind { ; X32-LABEL: test_broadcast_4f32_8f32: -; X32: ## BB#0: +; X32: # BB#0: ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-NEXT: retl ; ; X64-LABEL: test_broadcast_4f32_8f32: -; X64: ## BB#0: +; X64: # BB#0: ; X64-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-NEXT: retq %1 = load <4 x float>, <4 x float> *%p @@ -303,26 +303,26 @@ define <8 x float> @test_broadcast_4f32_8f32(<4 x float> *%p) nounwind { define <16 x float> @test_broadcast_4f32_16f32(<4 x float> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_4f32_16f32: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX-NEXT: vmovdqa %ymm0, %ymm1 ; X32-AVX-NEXT: retl ; ; X32-AVX512-LABEL: test_broadcast_4f32_16f32: -; X32-AVX512: ## BB#0: +; X32-AVX512: # BB#0: ; X32-AVX512-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512-NEXT: vbroadcastf32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X32-AVX512-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_4f32_16f32: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX-NEXT: vmovdqa %ymm0, %ymm1 ; X64-AVX-NEXT: retq ; ; X64-AVX512-LABEL: test_broadcast_4f32_16f32: -; X64-AVX512: ## BB#0: +; X64-AVX512: # BB#0: ; X64-AVX512-NEXT: vbroadcastf32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X64-AVX512-NEXT: retq %1 = load <4 x float>, <4 x float> *%p @@ -332,48 +332,48 @@ define <16 x float> @test_broadcast_4f32_16f32(<4 x float> *%p) nounwind { define <16 x float> @test_broadcast_8f32_16f32(<8 x float> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_8f32_16f32: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vmovaps (%eax), %ymm0 ; X32-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_8f32_16f32: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: vbroadcastf64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_8f32_16f32: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: vbroadcastf64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_8f32_16f32: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: vbroadcastf32x8 {{.*#+}} zmm0 = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7] ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_8f32_16f32: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vmovaps (%rdi), %ymm0 ; X64-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_8f32_16f32: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vbroadcastf64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_8f32_16f32: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vbroadcastf64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_8f32_16f32: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vbroadcastf32x8 {{.*#+}} zmm0 = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7] ; X64-AVX512DQ-NEXT: retq %1 = load <8 x float>, <8 x float> *%p @@ -383,24 +383,24 @@ define <16 x float> @test_broadcast_8f32_16f32(<8 x float> *%p) nounwind { define <8 x i32> @test_broadcast_4i32_8i32(<4 x i32> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_4i32_8i32: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX-NEXT: retl ; ; X32-AVX512-LABEL: test_broadcast_4i32_8i32: -; X32-AVX512: ## BB#0: +; X32-AVX512: # BB#0: ; X32-AVX512-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_4i32_8i32: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX-NEXT: retq ; ; X64-AVX512-LABEL: test_broadcast_4i32_8i32: -; X64-AVX512: ## BB#0: +; X64-AVX512: # BB#0: ; X64-AVX512-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512-NEXT: retq %1 = load <4 x i32>, <4 x i32> *%p @@ -410,39 +410,39 @@ define <8 x i32> @test_broadcast_4i32_8i32(<4 x i32> *%p) nounwind { define <16 x i32> @test_broadcast_4i32_16i32(<4 x i32> *%p) nounwind { ; X32-AVX1-LABEL: test_broadcast_4i32_16i32: -; X32-AVX1: ## BB#0: +; X32-AVX1: # BB#0: ; X32-AVX1-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX1-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX1-NEXT: vmovdqa %ymm0, %ymm1 ; X32-AVX1-NEXT: retl ; ; X32-AVX2-LABEL: test_broadcast_4i32_16i32: -; X32-AVX2: ## BB#0: +; X32-AVX2: # BB#0: ; X32-AVX2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX2-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX2-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX2-NEXT: retl ; ; X32-AVX512-LABEL: test_broadcast_4i32_16i32: -; X32-AVX512: ## BB#0: +; X32-AVX512: # BB#0: ; X32-AVX512-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X32-AVX512-NEXT: retl ; ; X64-AVX1-LABEL: test_broadcast_4i32_16i32: -; X64-AVX1: ## BB#0: +; X64-AVX1: # BB#0: ; X64-AVX1-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX1-NEXT: vmovdqa %ymm0, %ymm1 ; X64-AVX1-NEXT: retq ; ; X64-AVX2-LABEL: test_broadcast_4i32_16i32: -; X64-AVX2: ## BB#0: +; X64-AVX2: # BB#0: ; X64-AVX2-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX2-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX2-NEXT: retq ; ; X64-AVX512-LABEL: test_broadcast_4i32_16i32: -; X64-AVX512: ## BB#0: +; X64-AVX512: # BB#0: ; X64-AVX512-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X64-AVX512-NEXT: retq %1 = load <4 x i32>, <4 x i32> *%p @@ -452,48 +452,48 @@ define <16 x i32> @test_broadcast_4i32_16i32(<4 x i32> *%p) nounwind { define <16 x i32> @test_broadcast_8i32_16i32(<8 x i32> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_8i32_16i32: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vmovaps (%eax), %ymm0 ; X32-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_8i32_16i32: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_8i32_16i32: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_8i32_16i32: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7] ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_8i32_16i32: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vmovaps (%rdi), %ymm0 ; X64-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_8i32_16i32: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_8i32_16i32: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_8i32_16i32: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7] ; X64-AVX512DQ-NEXT: retq %1 = load <8 x i32>, <8 x i32> *%p @@ -503,24 +503,24 @@ define <16 x i32> @test_broadcast_8i32_16i32(<8 x i32> *%p) nounwind { define <16 x i16> @test_broadcast_8i16_16i16(<8 x i16> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_8i16_16i16: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX-NEXT: retl ; ; X32-AVX512-LABEL: test_broadcast_8i16_16i16: -; X32-AVX512: ## BB#0: +; X32-AVX512: # BB#0: ; X32-AVX512-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_8i16_16i16: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX-NEXT: retq ; ; X64-AVX512-LABEL: test_broadcast_8i16_16i16: -; X64-AVX512: ## BB#0: +; X64-AVX512: # BB#0: ; X64-AVX512-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512-NEXT: retq %1 = load <8 x i16>, <8 x i16> *%p @@ -530,64 +530,64 @@ define <16 x i16> @test_broadcast_8i16_16i16(<8 x i16> *%p) nounwind { define <32 x i16> @test_broadcast_8i16_32i16(<8 x i16> *%p) nounwind { ; X32-AVX1-LABEL: test_broadcast_8i16_32i16: -; X32-AVX1: ## BB#0: +; X32-AVX1: # BB#0: ; X32-AVX1-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX1-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX1-NEXT: vmovdqa %ymm0, %ymm1 ; X32-AVX1-NEXT: retl ; ; X32-AVX2-LABEL: test_broadcast_8i16_32i16: -; X32-AVX2: ## BB#0: +; X32-AVX2: # BB#0: ; X32-AVX2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX2-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX2-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX2-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_8i16_32i16: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512F-NEXT: vmovdqa %ymm0, %ymm1 ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_8i16_32i16: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_8i16_32i16: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512DQ-NEXT: vmovdqa %ymm0, %ymm1 ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX1-LABEL: test_broadcast_8i16_32i16: -; X64-AVX1: ## BB#0: +; X64-AVX1: # BB#0: ; X64-AVX1-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX1-NEXT: vmovdqa %ymm0, %ymm1 ; X64-AVX1-NEXT: retq ; ; X64-AVX2-LABEL: test_broadcast_8i16_32i16: -; X64-AVX2: ## BB#0: +; X64-AVX2: # BB#0: ; X64-AVX2-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX2-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX2-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_8i16_32i16: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512F-NEXT: vmovdqa %ymm0, %ymm1 ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_8i16_32i16: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_8i16_32i16: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512DQ-NEXT: vmovdqa %ymm0, %ymm1 ; X64-AVX512DQ-NEXT: retq @@ -598,51 +598,51 @@ define <32 x i16> @test_broadcast_8i16_32i16(<8 x i16> *%p) nounwind { define <32 x i16> @test_broadcast_16i16_32i16(<16 x i16> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_16i16_32i16: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vmovaps (%eax), %ymm0 ; X32-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_16i16_32i16: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: vmovaps (%eax), %ymm0 ; X32-AVX512F-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_16i16_32i16: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_16i16_32i16: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: vmovaps (%eax), %ymm0 ; X32-AVX512DQ-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_16i16_32i16: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vmovaps (%rdi), %ymm0 ; X64-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_16i16_32i16: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vmovaps (%rdi), %ymm0 ; X64-AVX512F-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_16i16_32i16: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_16i16_32i16: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vmovaps (%rdi), %ymm0 ; X64-AVX512DQ-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX512DQ-NEXT: retq @@ -653,24 +653,24 @@ define <32 x i16> @test_broadcast_16i16_32i16(<16 x i16> *%p) nounwind { define <32 x i8> @test_broadcast_16i8_32i8(<16 x i8> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_16i8_32i8: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX-NEXT: retl ; ; X32-AVX512-LABEL: test_broadcast_16i8_32i8: -; X32-AVX512: ## BB#0: +; X32-AVX512: # BB#0: ; X32-AVX512-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_16i8_32i8: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX-NEXT: retq ; ; X64-AVX512-LABEL: test_broadcast_16i8_32i8: -; X64-AVX512: ## BB#0: +; X64-AVX512: # BB#0: ; X64-AVX512-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512-NEXT: retq %1 = load <16 x i8>, <16 x i8> *%p @@ -680,64 +680,64 @@ define <32 x i8> @test_broadcast_16i8_32i8(<16 x i8> *%p) nounwind { define <64 x i8> @test_broadcast_16i8_64i8(<16 x i8> *%p) nounwind { ; X32-AVX1-LABEL: test_broadcast_16i8_64i8: -; X32-AVX1: ## BB#0: +; X32-AVX1: # BB#0: ; X32-AVX1-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX1-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX1-NEXT: vmovdqa %ymm0, %ymm1 ; X32-AVX1-NEXT: retl ; ; X32-AVX2-LABEL: test_broadcast_16i8_64i8: -; X32-AVX2: ## BB#0: +; X32-AVX2: # BB#0: ; X32-AVX2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX2-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX2-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX2-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_16i8_64i8: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512F-NEXT: vmovdqa %ymm0, %ymm1 ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_16i8_64i8: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_16i8_64i8: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X32-AVX512DQ-NEXT: vmovdqa %ymm0, %ymm1 ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX1-LABEL: test_broadcast_16i8_64i8: -; X64-AVX1: ## BB#0: +; X64-AVX1: # BB#0: ; X64-AVX1-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX1-NEXT: vmovdqa %ymm0, %ymm1 ; X64-AVX1-NEXT: retq ; ; X64-AVX2-LABEL: test_broadcast_16i8_64i8: -; X64-AVX2: ## BB#0: +; X64-AVX2: # BB#0: ; X64-AVX2-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX2-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX2-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_16i8_64i8: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512F-NEXT: vmovdqa %ymm0, %ymm1 ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_16i8_64i8: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3] ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_16i8_64i8: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1] ; X64-AVX512DQ-NEXT: vmovdqa %ymm0, %ymm1 ; X64-AVX512DQ-NEXT: retq @@ -748,51 +748,51 @@ define <64 x i8> @test_broadcast_16i8_64i8(<16 x i8> *%p) nounwind { define <64 x i8> @test_broadcast_32i8_64i8(<32 x i8> *%p) nounwind { ; X32-AVX-LABEL: test_broadcast_32i8_64i8: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: vmovaps (%eax), %ymm0 ; X32-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_32i8_64i8: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: vmovaps (%eax), %ymm0 ; X32-AVX512F-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_32i8_64i8: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_32i8_64i8: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: vmovaps (%eax), %ymm0 ; X32-AVX512DQ-NEXT: vmovaps %ymm0, %ymm1 ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_32i8_64i8: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vmovaps (%rdi), %ymm0 ; X64-AVX-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_32i8_64i8: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vmovaps (%rdi), %ymm0 ; X64-AVX512F-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_32i8_64i8: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3] ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_32i8_64i8: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vmovaps (%rdi), %ymm0 ; X64-AVX512DQ-NEXT: vmovaps %ymm0, %ymm1 ; X64-AVX512DQ-NEXT: retq @@ -807,7 +807,7 @@ define <64 x i8> @test_broadcast_32i8_64i8(<32 x i8> *%p) nounwind { define <4 x double> @test_broadcast_2f64_4f64_reuse(<2 x double>* %p0, <2 x double>* %p1) { ; X32-LABEL: test_broadcast_2f64_4f64_reuse: -; X32: ## BB#0: +; X32: # BB#0: ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-NEXT: vmovaps (%ecx), %xmm0 @@ -816,7 +816,7 @@ define <4 x double> @test_broadcast_2f64_4f64_reuse(<2 x double>* %p0, <2 x doub ; X32-NEXT: retl ; ; X64-LABEL: test_broadcast_2f64_4f64_reuse: -; X64: ## BB#0: +; X64: # BB#0: ; X64-NEXT: vmovaps (%rdi), %xmm0 ; X64-NEXT: vmovaps %xmm0, (%rsi) ; X64-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 @@ -829,7 +829,7 @@ define <4 x double> @test_broadcast_2f64_4f64_reuse(<2 x double>* %p0, <2 x doub define <4 x i64> @test_broadcast_2i64_4i64_reuse(<2 x i64>* %p0, <2 x i64>* %p1) { ; X32-LABEL: test_broadcast_2i64_4i64_reuse: -; X32: ## BB#0: +; X32: # BB#0: ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-NEXT: vmovaps (%ecx), %xmm0 @@ -838,7 +838,7 @@ define <4 x i64> @test_broadcast_2i64_4i64_reuse(<2 x i64>* %p0, <2 x i64>* %p1) ; X32-NEXT: retl ; ; X64-LABEL: test_broadcast_2i64_4i64_reuse: -; X64: ## BB#0: +; X64: # BB#0: ; X64-NEXT: vmovaps (%rdi), %xmm0 ; X64-NEXT: vmovaps %xmm0, (%rsi) ; X64-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 @@ -851,7 +851,7 @@ define <4 x i64> @test_broadcast_2i64_4i64_reuse(<2 x i64>* %p0, <2 x i64>* %p1) define <8 x float> @test_broadcast_4f32_8f32_reuse(<4 x float>* %p0, <4 x float>* %p1) { ; X32-LABEL: test_broadcast_4f32_8f32_reuse: -; X32: ## BB#0: +; X32: # BB#0: ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-NEXT: vmovaps (%ecx), %xmm0 @@ -860,7 +860,7 @@ define <8 x float> @test_broadcast_4f32_8f32_reuse(<4 x float>* %p0, <4 x float> ; X32-NEXT: retl ; ; X64-LABEL: test_broadcast_4f32_8f32_reuse: -; X64: ## BB#0: +; X64: # BB#0: ; X64-NEXT: vmovaps (%rdi), %xmm0 ; X64-NEXT: vmovaps %xmm0, (%rsi) ; X64-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 @@ -873,7 +873,7 @@ define <8 x float> @test_broadcast_4f32_8f32_reuse(<4 x float>* %p0, <4 x float> define <8 x i32> @test_broadcast_4i32_8i32_reuse(<4 x i32>* %p0, <4 x i32>* %p1) { ; X32-LABEL: test_broadcast_4i32_8i32_reuse: -; X32: ## BB#0: +; X32: # BB#0: ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-NEXT: vmovaps (%ecx), %xmm0 @@ -882,7 +882,7 @@ define <8 x i32> @test_broadcast_4i32_8i32_reuse(<4 x i32>* %p0, <4 x i32>* %p1) ; X32-NEXT: retl ; ; X64-LABEL: test_broadcast_4i32_8i32_reuse: -; X64: ## BB#0: +; X64: # BB#0: ; X64-NEXT: vmovaps (%rdi), %xmm0 ; X64-NEXT: vmovaps %xmm0, (%rsi) ; X64-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 @@ -895,7 +895,7 @@ define <8 x i32> @test_broadcast_4i32_8i32_reuse(<4 x i32>* %p0, <4 x i32>* %p1) define <16 x i16> @test_broadcast_8i16_16i16_reuse(<8 x i16> *%p0, <8 x i16> *%p1) nounwind { ; X32-LABEL: test_broadcast_8i16_16i16_reuse: -; X32: ## BB#0: +; X32: # BB#0: ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-NEXT: vmovaps (%ecx), %xmm0 @@ -904,7 +904,7 @@ define <16 x i16> @test_broadcast_8i16_16i16_reuse(<8 x i16> *%p0, <8 x i16> *%p ; X32-NEXT: retl ; ; X64-LABEL: test_broadcast_8i16_16i16_reuse: -; X64: ## BB#0: +; X64: # BB#0: ; X64-NEXT: vmovaps (%rdi), %xmm0 ; X64-NEXT: vmovaps %xmm0, (%rsi) ; X64-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 @@ -917,7 +917,7 @@ define <16 x i16> @test_broadcast_8i16_16i16_reuse(<8 x i16> *%p0, <8 x i16> *%p define <32 x i8> @test_broadcast_16i8_32i8_reuse(<16 x i8> *%p0, <16 x i8> *%p1) nounwind { ; X32-LABEL: test_broadcast_16i8_32i8_reuse: -; X32: ## BB#0: +; X32: # BB#0: ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-NEXT: vmovaps (%ecx), %xmm0 @@ -926,7 +926,7 @@ define <32 x i8> @test_broadcast_16i8_32i8_reuse(<16 x i8> *%p0, <16 x i8> *%p1) ; X32-NEXT: retl ; ; X64-LABEL: test_broadcast_16i8_32i8_reuse: -; X64: ## BB#0: +; X64: # BB#0: ; X64-NEXT: vmovaps (%rdi), %xmm0 ; X64-NEXT: vmovaps %xmm0, (%rsi) ; X64-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 @@ -943,7 +943,7 @@ define <32 x i8> @test_broadcast_16i8_32i8_reuse(<16 x i8> *%p0, <16 x i8> *%p1) define <8 x i32> @test_broadcast_4i32_8i32_chain(<4 x i32>* %p0, <4 x float>* %p1) { ; X32-AVX-LABEL: test_broadcast_4i32_8i32_chain: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-AVX-NEXT: vmovaps (%ecx), %xmm0 @@ -953,7 +953,7 @@ define <8 x i32> @test_broadcast_4i32_8i32_chain(<4 x i32>* %p0, <4 x float>* %p ; X32-AVX-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_4i32_8i32_chain: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-AVX512F-NEXT: vmovaps (%ecx), %xmm0 @@ -963,7 +963,7 @@ define <8 x i32> @test_broadcast_4i32_8i32_chain(<4 x i32>* %p0, <4 x float>* %p ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_4i32_8i32_chain: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-AVX512BW-NEXT: vmovaps (%ecx), %xmm0 @@ -973,7 +973,7 @@ define <8 x i32> @test_broadcast_4i32_8i32_chain(<4 x i32>* %p0, <4 x float>* %p ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_4i32_8i32_chain: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-AVX512DQ-NEXT: vmovaps (%ecx), %xmm0 @@ -983,7 +983,7 @@ define <8 x i32> @test_broadcast_4i32_8i32_chain(<4 x i32>* %p0, <4 x float>* %p ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_4i32_8i32_chain: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vmovaps (%rdi), %xmm0 ; X64-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1 ; X64-AVX-NEXT: vmovaps %xmm1, (%rsi) @@ -991,7 +991,7 @@ define <8 x i32> @test_broadcast_4i32_8i32_chain(<4 x i32>* %p0, <4 x float>* %p ; X64-AVX-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_4i32_8i32_chain: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vmovaps (%rdi), %xmm0 ; X64-AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; X64-AVX512F-NEXT: vmovdqa %xmm1, (%rsi) @@ -999,7 +999,7 @@ define <8 x i32> @test_broadcast_4i32_8i32_chain(<4 x i32>* %p0, <4 x float>* %p ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_4i32_8i32_chain: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vmovaps (%rdi), %xmm0 ; X64-AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; X64-AVX512BW-NEXT: vmovdqa %xmm1, (%rsi) @@ -1007,7 +1007,7 @@ define <8 x i32> @test_broadcast_4i32_8i32_chain(<4 x i32>* %p0, <4 x float>* %p ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_4i32_8i32_chain: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vmovaps (%rdi), %xmm0 ; X64-AVX512DQ-NEXT: vxorps %xmm1, %xmm1, %xmm1 ; X64-AVX512DQ-NEXT: vmovaps %xmm1, (%rsi) @@ -1021,7 +1021,7 @@ define <8 x i32> @test_broadcast_4i32_8i32_chain(<4 x i32>* %p0, <4 x float>* %p define <16 x i32> @test_broadcast_4i32_16i32_chain(<4 x i32>* %p0, <4 x float>* %p1) { ; X32-AVX-LABEL: test_broadcast_4i32_16i32_chain: -; X32-AVX: ## BB#0: +; X32-AVX: # BB#0: ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-AVX-NEXT: vmovaps (%ecx), %xmm0 @@ -1032,7 +1032,7 @@ define <16 x i32> @test_broadcast_4i32_16i32_chain(<4 x i32>* %p0, <4 x float>* ; X32-AVX-NEXT: retl ; ; X32-AVX512F-LABEL: test_broadcast_4i32_16i32_chain: -; X32-AVX512F: ## BB#0: +; X32-AVX512F: # BB#0: ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-AVX512F-NEXT: vmovdqa (%ecx), %xmm0 @@ -1042,7 +1042,7 @@ define <16 x i32> @test_broadcast_4i32_16i32_chain(<4 x i32>* %p0, <4 x float>* ; X32-AVX512F-NEXT: retl ; ; X32-AVX512BW-LABEL: test_broadcast_4i32_16i32_chain: -; X32-AVX512BW: ## BB#0: +; X32-AVX512BW: # BB#0: ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-AVX512BW-NEXT: vmovdqa (%ecx), %xmm0 @@ -1052,7 +1052,7 @@ define <16 x i32> @test_broadcast_4i32_16i32_chain(<4 x i32>* %p0, <4 x float>* ; X32-AVX512BW-NEXT: retl ; ; X32-AVX512DQ-LABEL: test_broadcast_4i32_16i32_chain: -; X32-AVX512DQ: ## BB#0: +; X32-AVX512DQ: # BB#0: ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-AVX512DQ-NEXT: vmovdqa (%ecx), %xmm0 @@ -1062,7 +1062,7 @@ define <16 x i32> @test_broadcast_4i32_16i32_chain(<4 x i32>* %p0, <4 x float>* ; X32-AVX512DQ-NEXT: retl ; ; X64-AVX-LABEL: test_broadcast_4i32_16i32_chain: -; X64-AVX: ## BB#0: +; X64-AVX: # BB#0: ; X64-AVX-NEXT: vmovaps (%rdi), %xmm0 ; X64-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1 ; X64-AVX-NEXT: vmovaps %xmm1, (%rsi) @@ -1071,7 +1071,7 @@ define <16 x i32> @test_broadcast_4i32_16i32_chain(<4 x i32>* %p0, <4 x float>* ; X64-AVX-NEXT: retq ; ; X64-AVX512F-LABEL: test_broadcast_4i32_16i32_chain: -; X64-AVX512F: ## BB#0: +; X64-AVX512F: # BB#0: ; X64-AVX512F-NEXT: vmovdqa (%rdi), %xmm0 ; X64-AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; X64-AVX512F-NEXT: vmovdqa %xmm1, (%rsi) @@ -1079,7 +1079,7 @@ define <16 x i32> @test_broadcast_4i32_16i32_chain(<4 x i32>* %p0, <4 x float>* ; X64-AVX512F-NEXT: retq ; ; X64-AVX512BW-LABEL: test_broadcast_4i32_16i32_chain: -; X64-AVX512BW: ## BB#0: +; X64-AVX512BW: # BB#0: ; X64-AVX512BW-NEXT: vmovdqa (%rdi), %xmm0 ; X64-AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1 ; X64-AVX512BW-NEXT: vmovdqa %xmm1, (%rsi) @@ -1087,7 +1087,7 @@ define <16 x i32> @test_broadcast_4i32_16i32_chain(<4 x i32>* %p0, <4 x float>* ; X64-AVX512BW-NEXT: retq ; ; X64-AVX512DQ-LABEL: test_broadcast_4i32_16i32_chain: -; X64-AVX512DQ: ## BB#0: +; X64-AVX512DQ: # BB#0: ; X64-AVX512DQ-NEXT: vmovdqa (%rdi), %xmm0 ; X64-AVX512DQ-NEXT: vxorps %xmm1, %xmm1, %xmm1 ; X64-AVX512DQ-NEXT: vmovaps %xmm1, (%rsi) @@ -1109,7 +1109,7 @@ define <16 x i32> @test_broadcast_4i32_16i32_chain(<4 x i32>* %p0, <4 x float>* define void @fallback_broadcast_v4i64_to_v8i64(<4 x i64> %a, <8 x i64> %b) { ; X32-AVX1-LABEL: fallback_broadcast_v4i64_to_v8i64: -; X32-AVX1: ## BB#0: ## %entry +; X32-AVX1: # BB#0: # %entry ; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 ; X32-AVX1-NEXT: vmovdqa {{.*#+}} ymm4 = [1,0,2,0,3,0,4,0] ; X32-AVX1-NEXT: vextractf128 $1, %ymm4, %xmm5 @@ -1126,39 +1126,39 @@ define void @fallback_broadcast_v4i64_to_v8i64(<4 x i64> %a, <8 x i64> %b) { ; X32-AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm1, %ymm1 ; X32-AVX1-NEXT: vandps %ymm4, %ymm1, %ymm1 ; X32-AVX1-NEXT: vandps %ymm4, %ymm2, %ymm2 -; X32-AVX1-NEXT: vmovups %ymm0, _ga4 -; X32-AVX1-NEXT: vmovups %ymm2, _gb4+32 -; X32-AVX1-NEXT: vmovups %ymm1, _gb4 +; X32-AVX1-NEXT: vmovups %ymm0, ga4 +; X32-AVX1-NEXT: vmovups %ymm2, gb4+32 +; X32-AVX1-NEXT: vmovups %ymm1, gb4 ; X32-AVX1-NEXT: vzeroupper ; X32-AVX1-NEXT: retl ; ; X32-AVX2-LABEL: fallback_broadcast_v4i64_to_v8i64: -; X32-AVX2: ## BB#0: ## %entry +; X32-AVX2: # BB#0: # %entry ; X32-AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [1,0,2,0,3,0,4,0] ; X32-AVX2-NEXT: vpaddq %ymm3, %ymm0, %ymm0 ; X32-AVX2-NEXT: vpaddq %ymm3, %ymm2, %ymm2 ; X32-AVX2-NEXT: vpaddq %ymm3, %ymm1, %ymm1 ; X32-AVX2-NEXT: vpand %ymm3, %ymm1, %ymm1 ; X32-AVX2-NEXT: vpand %ymm3, %ymm2, %ymm2 -; X32-AVX2-NEXT: vmovdqu %ymm0, _ga4 -; X32-AVX2-NEXT: vmovdqu %ymm2, _gb4+32 -; X32-AVX2-NEXT: vmovdqu %ymm1, _gb4 +; X32-AVX2-NEXT: vmovdqu %ymm0, ga4 +; X32-AVX2-NEXT: vmovdqu %ymm2, gb4+32 +; X32-AVX2-NEXT: vmovdqu %ymm1, gb4 ; X32-AVX2-NEXT: vzeroupper ; X32-AVX2-NEXT: retl ; ; X32-AVX512-LABEL: fallback_broadcast_v4i64_to_v8i64: -; X32-AVX512: ## BB#0: ## %entry -; X32-AVX512-NEXT: vpaddq LCPI26_0, %ymm0, %ymm0 +; X32-AVX512: # BB#0: # %entry +; X32-AVX512-NEXT: vpaddq {{\.LCPI.*}}, %ymm0, %ymm0 ; X32-AVX512-NEXT: vmovdqa64 {{.*#+}} zmm2 = [1,0,2,0,3,0,4,0,1,0,2,0,3,0,4,0] ; X32-AVX512-NEXT: vpaddq %zmm2, %zmm1, %zmm1 ; X32-AVX512-NEXT: vpandq %zmm2, %zmm1, %zmm1 -; X32-AVX512-NEXT: vmovdqu %ymm0, _ga4 -; X32-AVX512-NEXT: vmovdqu64 %zmm1, _gb4 +; X32-AVX512-NEXT: vmovdqu %ymm0, ga4 +; X32-AVX512-NEXT: vmovdqu64 %zmm1, gb4 ; X32-AVX512-NEXT: vzeroupper ; X32-AVX512-NEXT: retl ; ; X64-AVX1-LABEL: fallback_broadcast_v4i64_to_v8i64: -; X64-AVX1: ## BB#0: ## %entry +; X64-AVX1: # BB#0: # %entry ; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 ; X64-AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [3,4] ; X64-AVX1-NEXT: vpaddq %xmm4, %xmm3, %xmm3 @@ -1177,13 +1177,13 @@ define void @fallback_broadcast_v4i64_to_v8i64(<4 x i64> %a, <8 x i64> %b) { ; X64-AVX1-NEXT: vandps %ymm3, %ymm1, %ymm1 ; X64-AVX1-NEXT: vandps %ymm3, %ymm2, %ymm2 ; X64-AVX1-NEXT: vmovups %ymm0, {{.*}}(%rip) -; X64-AVX1-NEXT: vmovups %ymm2, _gb4+{{.*}}(%rip) +; X64-AVX1-NEXT: vmovups %ymm2, gb4+{{.*}}(%rip) ; X64-AVX1-NEXT: vmovups %ymm1, {{.*}}(%rip) ; X64-AVX1-NEXT: vzeroupper ; X64-AVX1-NEXT: retq ; ; X64-AVX2-LABEL: fallback_broadcast_v4i64_to_v8i64: -; X64-AVX2: ## BB#0: ## %entry +; X64-AVX2: # BB#0: # %entry ; X64-AVX2-NEXT: vmovdqa {{.*#+}} ymm3 = [1,2,3,4] ; X64-AVX2-NEXT: vpaddq %ymm3, %ymm0, %ymm0 ; X64-AVX2-NEXT: vpaddq %ymm3, %ymm2, %ymm2 @@ -1191,13 +1191,13 @@ define void @fallback_broadcast_v4i64_to_v8i64(<4 x i64> %a, <8 x i64> %b) { ; X64-AVX2-NEXT: vpand %ymm3, %ymm1, %ymm1 ; X64-AVX2-NEXT: vpand %ymm3, %ymm2, %ymm2 ; X64-AVX2-NEXT: vmovdqu %ymm0, {{.*}}(%rip) -; X64-AVX2-NEXT: vmovdqu %ymm2, _gb4+{{.*}}(%rip) +; X64-AVX2-NEXT: vmovdqu %ymm2, gb4+{{.*}}(%rip) ; X64-AVX2-NEXT: vmovdqu %ymm1, {{.*}}(%rip) ; X64-AVX2-NEXT: vzeroupper ; X64-AVX2-NEXT: retq ; ; X64-AVX512-LABEL: fallback_broadcast_v4i64_to_v8i64: -; X64-AVX512: ## BB#0: ## %entry +; X64-AVX512: # BB#0: # %entry ; X64-AVX512-NEXT: vmovdqa {{.*#+}} ymm2 = [1,2,3,4] ; X64-AVX512-NEXT: vpaddq %ymm2, %ymm0, %ymm0 ; X64-AVX512-NEXT: vinserti64x4 $1, %ymm2, %zmm2, %zmm2 @@ -1222,33 +1222,33 @@ entry: define void @fallback_broadcast_v4f64_to_v8f64(<4 x double> %a, <8 x double> %b) { ; X32-AVX-LABEL: fallback_broadcast_v4f64_to_v8f64: -; X32-AVX: ## BB#0: ## %entry +; X32-AVX: # BB#0: # %entry ; X32-AVX-NEXT: vmovapd {{.*#+}} ymm3 = [1.000000e+00,2.000000e+00,3.000000e+00,4.000000e+00] ; X32-AVX-NEXT: vaddpd %ymm3, %ymm0, %ymm0 ; X32-AVX-NEXT: vaddpd %ymm3, %ymm2, %ymm2 ; X32-AVX-NEXT: vaddpd %ymm3, %ymm1, %ymm1 ; X32-AVX-NEXT: vdivpd %ymm3, %ymm1, %ymm1 ; X32-AVX-NEXT: vdivpd %ymm3, %ymm2, %ymm2 -; X32-AVX-NEXT: vmovupd %ymm0, _ga2 -; X32-AVX-NEXT: vmovupd %ymm2, _gb2+32 -; X32-AVX-NEXT: vmovupd %ymm1, _gb2 +; X32-AVX-NEXT: vmovupd %ymm0, ga2 +; X32-AVX-NEXT: vmovupd %ymm2, gb2+32 +; X32-AVX-NEXT: vmovupd %ymm1, gb2 ; X32-AVX-NEXT: vzeroupper ; X32-AVX-NEXT: retl ; ; X32-AVX512-LABEL: fallback_broadcast_v4f64_to_v8f64: -; X32-AVX512: ## BB#0: ## %entry +; X32-AVX512: # BB#0: # %entry ; X32-AVX512-NEXT: vmovapd {{.*#+}} ymm2 = [1.000000e+00,2.000000e+00,3.000000e+00,4.000000e+00] ; X32-AVX512-NEXT: vaddpd %ymm2, %ymm0, %ymm0 ; X32-AVX512-NEXT: vinsertf64x4 $1, %ymm2, %zmm2, %zmm2 ; X32-AVX512-NEXT: vaddpd %zmm2, %zmm1, %zmm1 ; X32-AVX512-NEXT: vdivpd %zmm2, %zmm1, %zmm1 -; X32-AVX512-NEXT: vmovupd %ymm0, _ga2 -; X32-AVX512-NEXT: vmovupd %zmm1, _gb2 +; X32-AVX512-NEXT: vmovupd %ymm0, ga2 +; X32-AVX512-NEXT: vmovupd %zmm1, gb2 ; X32-AVX512-NEXT: vzeroupper ; X32-AVX512-NEXT: retl ; ; X64-AVX-LABEL: fallback_broadcast_v4f64_to_v8f64: -; X64-AVX: ## BB#0: ## %entry +; X64-AVX: # BB#0: # %entry ; X64-AVX-NEXT: vmovapd {{.*#+}} ymm3 = [1.000000e+00,2.000000e+00,3.000000e+00,4.000000e+00] ; X64-AVX-NEXT: vaddpd %ymm3, %ymm0, %ymm0 ; X64-AVX-NEXT: vaddpd %ymm3, %ymm2, %ymm2 @@ -1256,13 +1256,13 @@ define void @fallback_broadcast_v4f64_to_v8f64(<4 x double> %a, <8 x double> %b) ; X64-AVX-NEXT: vdivpd %ymm3, %ymm1, %ymm1 ; X64-AVX-NEXT: vdivpd %ymm3, %ymm2, %ymm2 ; X64-AVX-NEXT: vmovupd %ymm0, {{.*}}(%rip) -; X64-AVX-NEXT: vmovupd %ymm2, _gb2+{{.*}}(%rip) +; X64-AVX-NEXT: vmovupd %ymm2, gb2+{{.*}}(%rip) ; X64-AVX-NEXT: vmovupd %ymm1, {{.*}}(%rip) ; X64-AVX-NEXT: vzeroupper ; X64-AVX-NEXT: retq ; ; X64-AVX512-LABEL: fallback_broadcast_v4f64_to_v8f64: -; X64-AVX512: ## BB#0: ## %entry +; X64-AVX512: # BB#0: # %entry ; X64-AVX512-NEXT: vmovapd {{.*#+}} ymm2 = [1.000000e+00,2.000000e+00,3.000000e+00,4.000000e+00] ; X64-AVX512-NEXT: vaddpd %ymm2, %ymm0, %ymm0 ; X64-AVX512-NEXT: vinsertf64x4 $1, %ymm2, %zmm2, %zmm2 -- 2.7.4