; 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
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 {{.*#+}} ymm0 = <0,1,2,4,u,u,u,u>
+; 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
}
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 .LBB1_2
+; HSW-NEXT: # %bb.1: # %cond.store
+; HSW-NEXT: vmovss %xmm0, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB1_2: # %else
+; HSW-NEXT: vpextrb $2, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB1_4
+; HSW-NEXT: # %bb.3: # %cond.store1
+; HSW-NEXT: vextractps $1, %xmm0, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB1_4: # %else2
+; HSW-NEXT: vpextrb $4, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB1_6
+; HSW-NEXT: # %bb.5: # %cond.store4
+; HSW-NEXT: vextractps $2, %xmm0, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB1_6: # %else5
+; HSW-NEXT: vpextrb $6, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB1_8
+; HSW-NEXT: # %bb.7: # %cond.store7
+; HSW-NEXT: vextractps $3, %xmm0, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB1_8: # %else8
+; HSW-NEXT: vpextrb $8, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB1_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: .LBB1_10: # %else11
+; HSW-NEXT: vpextrb $10, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB1_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: .LBB1_12: # %else14
+; HSW-NEXT: vpextrb $12, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB1_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: .LBB1_14: # %else17
+; HSW-NEXT: vpextrb $14, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB1_16
+; HSW-NEXT: # %bb.15: # %cond.store19
+; HSW-NEXT: vextractf128 $1, %ymm0, %xmm0
+; HSW-NEXT: vextractps $3, %xmm0, (%rdi)
+; HSW-NEXT: .LBB1_16: # %else20
+; HSW-NEXT: vzeroupper
+; HSW-NEXT: retq
+;
; SKX-LABEL: compressstore_v8f32_v8i1:
; SKX: # %bb.0:
; SKX-NEXT: vpsllw $15, %xmm1, %xmm1
}
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 .LBB2_2
+; HSW-NEXT: # %bb.1: # %cond.store
+; HSW-NEXT: vmovlpd %xmm0, (%rdi)
+; 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.store1
+; HSW-NEXT: vmovhpd %xmm0, (%rdi)
+; 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.store4
+; HSW-NEXT: vextractf128 $1, %ymm0, %xmm3
+; HSW-NEXT: vmovlpd %xmm3, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB2_6: # %else5
+; HSW-NEXT: vpextrb $6, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB2_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: .LBB2_8: # %else8
+; HSW-NEXT: vpextrb $8, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB2_10
+; HSW-NEXT: # %bb.9: # %cond.store10
+; HSW-NEXT: vmovlpd %xmm1, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB2_10: # %else11
+; HSW-NEXT: vpextrb $10, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB2_12
+; HSW-NEXT: # %bb.11: # %cond.store13
+; HSW-NEXT: vmovhpd %xmm1, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB2_12: # %else14
+; HSW-NEXT: vpextrb $12, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB2_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: .LBB2_14: # %else17
+; HSW-NEXT: vpextrb $14, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB2_16
+; HSW-NEXT: # %bb.15: # %cond.store19
+; HSW-NEXT: vextractf128 $1, %ymm1, %xmm0
+; HSW-NEXT: vmovhpd %xmm0, (%rdi)
+; HSW-NEXT: .LBB2_16: # %else20
+; HSW-NEXT: vzeroupper
+; HSW-NEXT: retq
+;
; SKX-LABEL: compressstore_v8f64_v8i1:
; SKX: # %bb.0:
; SKX-NEXT: vpsllw $15, %xmm1, %xmm1
}
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 .LBB3_2
+; HSW-NEXT: # %bb.1: # %cond.store
+; HSW-NEXT: vmovq %xmm0, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB3_2: # %else
+; HSW-NEXT: vpextrb $2, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB3_4
+; HSW-NEXT: # %bb.3: # %cond.store1
+; HSW-NEXT: vpextrq $1, %xmm0, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB3_4: # %else2
+; HSW-NEXT: vpextrb $4, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB3_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: .LBB3_6: # %else5
+; HSW-NEXT: vpextrb $6, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB3_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: .LBB3_8: # %else8
+; HSW-NEXT: vpextrb $8, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB3_10
+; HSW-NEXT: # %bb.9: # %cond.store10
+; HSW-NEXT: vmovq %xmm1, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB3_10: # %else11
+; HSW-NEXT: vpextrb $10, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB3_12
+; HSW-NEXT: # %bb.11: # %cond.store13
+; HSW-NEXT: vpextrq $1, %xmm1, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB3_12: # %else14
+; HSW-NEXT: vpextrb $12, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB3_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: .LBB3_14: # %else17
+; HSW-NEXT: vpextrb $14, %xmm2, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB3_16
+; HSW-NEXT: # %bb.15: # %cond.store19
+; HSW-NEXT: vextracti128 $1, %ymm1, %xmm0
+; HSW-NEXT: vpextrq $1, %xmm0, (%rdi)
+; HSW-NEXT: .LBB3_16: # %else20
+; HSW-NEXT: vzeroupper
+; HSW-NEXT: retq
+;
; SKX-LABEL: compressstore_v8i64_v8i1:
; SKX: # %bb.0:
; SKX-NEXT: vpsllw $15, %xmm1, %xmm1
}
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 .LBB4_2
+; HSW-NEXT: # %bb.1: # %cond.store
+; HSW-NEXT: vmovq %xmm0, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB4_2: # %else
+; HSW-NEXT: vpextrb $4, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB4_4
+; HSW-NEXT: # %bb.3: # %cond.store1
+; HSW-NEXT: vpextrq $1, %xmm0, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB4_4: # %else2
+; HSW-NEXT: vpextrb $8, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB4_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: .LBB4_6: # %else5
+; HSW-NEXT: vpextrb $12, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB4_8
+; HSW-NEXT: # %bb.7: # %cond.store7
+; HSW-NEXT: vextracti128 $1, %ymm0, %xmm0
+; HSW-NEXT: vpextrq $1, %xmm0, (%rdi)
+; HSW-NEXT: .LBB4_8: # %else8
+; HSW-NEXT: vzeroupper
+; HSW-NEXT: retq
+;
; SKX-LABEL: compressstore_v4i64_v4i1:
; SKX: # %bb.0:
; SKX-NEXT: vpslld $31, %xmm1, %xmm1
}
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 .LBB5_2
+; HSW-NEXT: # %bb.1: # %cond.store
+; HSW-NEXT: vmovq %xmm0, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB5_2: # %else
+; HSW-NEXT: vpextrb $8, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB5_4
+; HSW-NEXT: # %bb.3: # %cond.store1
+; HSW-NEXT: vpextrq $1, %xmm0, (%rdi)
+; HSW-NEXT: .LBB5_4: # %else2
+; HSW-NEXT: retq
+;
; SKX-LABEL: compressstore_v2i64_v2i1:
; SKX: # %bb.0:
; SKX-NEXT: vpsllq $63, %xmm1, %xmm1
}
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 .LBB6_2
+; HSW-NEXT: # %bb.1: # %cond.store
+; HSW-NEXT: vmovss %xmm0, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB6_2: # %else
+; HSW-NEXT: vpextrb $4, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB6_4
+; HSW-NEXT: # %bb.3: # %cond.store1
+; HSW-NEXT: vextractps $1, %xmm0, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB6_4: # %else2
+; HSW-NEXT: vpextrb $8, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB6_6
+; HSW-NEXT: # %bb.5: # %cond.store4
+; HSW-NEXT: vextractps $2, %xmm0, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB6_6: # %else5
+; HSW-NEXT: vpextrb $12, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB6_8
+; HSW-NEXT: # %bb.7: # %cond.store7
+; HSW-NEXT: vextractps $3, %xmm0, (%rdi)
+; HSW-NEXT: .LBB6_8: # %else8
+; HSW-NEXT: retq
+;
; SKX-LABEL: compressstore_v4f32_v4i1:
; SKX: # %bb.0:
; SKX-NEXT: vpslld $31, %xmm1, %xmm1
}
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 .LBB7_2
+; HSW-NEXT: # %bb.1: # %cond.store
+; HSW-NEXT: vmovss %xmm0, (%rdi)
+; HSW-NEXT: addq $4, %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: vextractps $1, %xmm0, (%rdi)
+; HSW-NEXT: .LBB7_4: # %else2
+; HSW-NEXT: retq
+;
; SKX-LABEL: compressstore_v2f32_v2i32:
; SKX: # %bb.0:
; SKX-NEXT: vpxor %xmm2, %xmm2, %xmm2
}
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 .LBB8_2
+; HSW-NEXT: # %bb.1: # %cond.store
+; HSW-NEXT: vmovd %xmm0, (%rdi)
+; 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.store1
+; HSW-NEXT: vpextrd $1, %xmm0, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_6
+; HSW-NEXT: # %bb.5: # %cond.store4
+; HSW-NEXT: vpextrd $2, %xmm0, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_8
+; HSW-NEXT: # %bb.7: # %cond.store7
+; HSW-NEXT: vpextrd $3, %xmm0, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_18
+; HSW-NEXT: # %bb.17: # %cond.store22
+; HSW-NEXT: vmovd %xmm1, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_20
+; HSW-NEXT: # %bb.19: # %cond.store25
+; HSW-NEXT: vpextrd $1, %xmm1, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_22
+; HSW-NEXT: # %bb.21: # %cond.store28
+; HSW-NEXT: vpextrd $2, %xmm1, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_24
+; HSW-NEXT: # %bb.23: # %cond.store31
+; HSW-NEXT: vpextrd $3, %xmm1, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_34
+; HSW-NEXT: # %bb.33: # %cond.store46
+; HSW-NEXT: vmovd %xmm2, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_36
+; HSW-NEXT: # %bb.35: # %cond.store49
+; HSW-NEXT: vpextrd $1, %xmm2, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_38
+; HSW-NEXT: # %bb.37: # %cond.store52
+; HSW-NEXT: vpextrd $2, %xmm2, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_40
+; HSW-NEXT: # %bb.39: # %cond.store55
+; HSW-NEXT: vpextrd $3, %xmm2, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_50
+; HSW-NEXT: # %bb.49: # %cond.store70
+; HSW-NEXT: vmovd %xmm3, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_52
+; HSW-NEXT: # %bb.51: # %cond.store73
+; HSW-NEXT: vpextrd $1, %xmm3, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_54
+; HSW-NEXT: # %bb.53: # %cond.store76
+; HSW-NEXT: vpextrd $2, %xmm3, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_56
+; HSW-NEXT: # %bb.55: # %cond.store79
+; HSW-NEXT: vpextrd $3, %xmm3, (%rdi)
+; HSW-NEXT: addq $4, %rdi
+; HSW-NEXT: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_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: .LBB8_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 .LBB8_64
+; HSW-NEXT: # %bb.63: # %cond.store91
+; HSW-NEXT: vextracti128 $1, %ymm3, %xmm0
+; HSW-NEXT: vpextrd $3, %xmm0, (%rdi)
+; HSW-NEXT: .LBB8_64: # %else92
+; HSW-NEXT: vzeroupper
+; HSW-NEXT: retq
+;
; SKX-LABEL: compressstore_v32f32_v32i32:
; SKX: # %bb.0:
; SKX-NEXT: vptestnmd %zmm3, %zmm3, %k1
}
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 .LBB9_2
+; HSW-NEXT: # %bb.1: # %cond.store
+; HSW-NEXT: vmovlpd %xmm0, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB9_2: # %else
+; HSW-NEXT: vpextrb $1, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_4
+; HSW-NEXT: # %bb.3: # %cond.store1
+; HSW-NEXT: vmovhpd %xmm0, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB9_4: # %else2
+; HSW-NEXT: vpextrb $2, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_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: .LBB9_6: # %else5
+; HSW-NEXT: vpextrb $3, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_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: .LBB9_8: # %else8
+; HSW-NEXT: vpextrb $4, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_10
+; HSW-NEXT: # %bb.9: # %cond.store10
+; HSW-NEXT: vmovlpd %xmm1, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB9_10: # %else11
+; HSW-NEXT: vpextrb $5, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_12
+; HSW-NEXT: # %bb.11: # %cond.store13
+; HSW-NEXT: vmovhpd %xmm1, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB9_12: # %else14
+; HSW-NEXT: vpextrb $6, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_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: .LBB9_14: # %else17
+; HSW-NEXT: vpextrb $7, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_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: .LBB9_16: # %else20
+; HSW-NEXT: vpextrb $8, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_18
+; HSW-NEXT: # %bb.17: # %cond.store22
+; HSW-NEXT: vmovlpd %xmm2, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB9_18: # %else23
+; HSW-NEXT: vpextrb $9, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_20
+; HSW-NEXT: # %bb.19: # %cond.store25
+; HSW-NEXT: vmovhpd %xmm2, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB9_20: # %else26
+; HSW-NEXT: vpextrb $10, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_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: .LBB9_22: # %else29
+; HSW-NEXT: vpextrb $11, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_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: .LBB9_24: # %else32
+; HSW-NEXT: vpextrb $12, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_26
+; HSW-NEXT: # %bb.25: # %cond.store34
+; HSW-NEXT: vmovlpd %xmm3, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB9_26: # %else35
+; HSW-NEXT: vpextrb $13, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_28
+; HSW-NEXT: # %bb.27: # %cond.store37
+; HSW-NEXT: vmovhpd %xmm3, (%rdi)
+; HSW-NEXT: addq $8, %rdi
+; HSW-NEXT: .LBB9_28: # %else38
+; HSW-NEXT: vpextrb $14, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_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: .LBB9_30: # %else41
+; HSW-NEXT: vpextrb $15, %xmm4, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB9_32
+; HSW-NEXT: # %bb.31: # %cond.store43
+; HSW-NEXT: vextractf128 $1, %ymm3, %xmm0
+; HSW-NEXT: vmovhpd %xmm0, (%rdi)
+; HSW-NEXT: .LBB9_32: # %else44
+; HSW-NEXT: vzeroupper
+; HSW-NEXT: retq
+;
; SKX-LABEL: compressstore_v16f64_v16i1:
; SKX: # %bb.0:
; SKX-NEXT: vpsllw $7, %xmm2, %xmm2
; 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
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
}
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
}
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
}
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
}
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
}
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 .LBB5_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: .LBB5_2: # %else
+; HSW-NEXT: vpextrb $8, %xmm1, %eax
+; HSW-NEXT: testb $1, %al
+; HSW-NEXT: je .LBB5_4
+; HSW-NEXT: # %bb.3: # %cond.load1
+; HSW-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3]
+; HSW-NEXT: .LBB5_4: # %else2
+; HSW-NEXT: retq
+;
; SKX-LABEL: expandload_v2f32_v2i1:
; SKX: # %bb.0:
; SKX-NEXT: vpxor %xmm2, %xmm2, %xmm2
}
define <32 x float> @expandload_v32f32_v32i32(float* %base, <32 x float> %src0, <32 x i32> %trigger) {
-; ALL-LABEL: expandload_v32f32_v32i32:
-; ALL: # %bb.0:
-; ALL-NEXT: vptestnmd %zmm3, %zmm3, %k1
-; ALL-NEXT: vptestnmd %zmm2, %zmm2, %k2
-; ALL-NEXT: kmovw %k2, %eax
-; ALL-NEXT: popcntl %eax, %eax
-; ALL-NEXT: vexpandps (%rdi,%rax,4), %zmm1 {%k1}
-; ALL-NEXT: vexpandps (%rdi), %zmm0 {%k2}
-; ALL-NEXT: retq
+; 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 .LBB6_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: .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.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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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 .LBB6_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: .LBB6_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
%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
}
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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_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 .LBB7_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: .LBB7_32: # %else58
+; HSW-NEXT: retq
+;
; SKX-LABEL: expandload_v16f64_v16i32:
; SKX: # %bb.0:
; SKX-NEXT: vextracti64x4 $1, %zmm2, %ymm3