RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-04-03 00:01:39 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction.

Browse Source 
code example , previous implementation.
    movzbl  %dil, %eax
    kmovw  %eax, %k0
  new code
    kmovw  %edi, %k0

Differential Revision: http://reviews.llvm.org/D16287

llvm-svn: 258045
This commit is contained in:
Igor Breger 2016-01-18 12:02:45 +00:00
parent ec1b7475d8
commit 74d74d20c2
12 changed files with 1124 additions and 1358 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
lib/Target/X86/X86InstrAVX512.td
View File

@ -2172,7 +2172,7 @@ let Predicates = [HasAVX512, NoDQI] in {
// GR from/to 8-bit mask without native support
def : Pat<(v8i1 (bitconvert (i8 GR8:$src))),
(COPY_TO_REGCLASS
(KMOVWkr (MOVZX32rr8 GR8 :$src)), VK8)>;
(KMOVWkr (SUBREG_TO_REG (i32 0), GR8:$src, sub_8bit)), VK8)>;
def : Pat<(i8 (bitconvert (v8i1 VK8:$src))),
(EXTRACT_SUBREG
(KMOVWrk (COPY_TO_REGCLASS VK8:$src, VK16)),

2
test/CodeGen/X86/avx512-calling-conv.ll
View File

@ -277,7 +277,6 @@ define <8 x i1> @test7a(<8 x i32>%a, <8 x i32>%b) {
; KNL-NEXT: vpmovsxwq %xmm0, %zmm0
; KNL-NEXT: vpsllq $63, %zmm0, %zmm0
; KNL-NEXT: movb $85, %al
; KNL-NEXT: movzbl %al, %eax
; KNL-NEXT: kmovw %eax, %k1
; KNL-NEXT: vptestmq %zmm0, %zmm0, %k1 {%k1}
; KNL-NEXT: vpbroadcastq {{.*}}(%rip), %zmm0 {%k1} {z}
@ -313,7 +312,6 @@ define <8 x i1> @test7a(<8 x i32>%a, <8 x i32>%b) {
; KNL_X32-NEXT: vpmovsxwq %xmm0, %zmm0
; KNL_X32-NEXT: vpsllvq LCPI7_0, %zmm0, %zmm0
; KNL_X32-NEXT: movb $85, %al
; KNL_X32-NEXT: movzbl %al, %eax
; KNL_X32-NEXT: kmovw %eax, %k1
; KNL_X32-NEXT: vptestmq %zmm0, %zmm0, %k1 {%k1}
; KNL_X32-NEXT: vpbroadcastd LCPI7_1, %zmm0

3
test/CodeGen/X86/avx512-ext.ll
View File

@ -1312,8 +1312,7 @@ define <16 x i32> @zext_16i1_to_16xi32(i16 %b) {
define <8 x i64> @zext_8i1_to_8xi64(i8 %b) {
; KNL-LABEL: zext_8i1_to_8xi64:
; KNL: ## BB#0:
; KNL-NEXT: movzbl %dil, %eax
; KNL-NEXT: kmovw %eax, %k1
; KNL-NEXT: kmovw %edi, %k1
; KNL-NEXT: vpbroadcastq {{.*}}(%rip), %zmm0 {%k1} {z}
; KNL-NEXT: retq
;

307
test/CodeGen/X86/avx512-fma-intrinsics.ll
View File

@ -1,78 +1,104 @@
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512f --show-mc-encoding | FileCheck %s
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512f | FileCheck %s
declare <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
declare <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
define <16 x float> @test_x86_vfnmadd_ps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_x86_vfnmadd_ps_z
; CHECK: vfnmadd213ps %zmm
; CHECK-LABEL: test_x86_vfnmadd_ps_z:
; CHECK: ## BB#0:
; CHECK-NEXT: vfnmadd213ps %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfnmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 4) nounwind
ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.mask.vfnmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32) nounwind readnone
define <16 x float> @test_mask_vfnmadd_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
; CHECK-LABEL: test_mask_vfnmadd_ps
; CHECK: vfnmadd213ps %zmm
; CHECK-LABEL: test_mask_vfnmadd_ps:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfnmadd213ps %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfnmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 4) nounwind
ret <16 x float> %res
}
define <8 x double> @test_x86_vfnmadd_pd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_x86_vfnmadd_pd_z
; CHECK: vfnmadd213pd %zmm
; CHECK-LABEL: test_x86_vfnmadd_pd_z:
; CHECK: ## BB#0:
; CHECK-NEXT: vfnmadd213pd %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 4) nounwind
ret <8 x double> %res
}
declare <8 x double> @llvm.x86.avx512.mask.vfnmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32) nounwind readnone
define <8 x double> @test_mask_vfnmadd_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_vfnmadd_pd
; CHECK: vfnmadd213pd %zmm
; CHECK-LABEL: test_mask_vfnmadd_pd:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfnmadd213pd %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
ret <8 x double> %res
}
define <16 x float> @test_x86_vfnmsubps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_x86_vfnmsubps_z
; CHECK: vfnmsub213ps %zmm
; CHECK-LABEL: test_x86_vfnmsubps_z:
; CHECK: ## BB#0:
; CHECK-NEXT: vfnmsub213ps %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfnmsub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 4) nounwind
ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.mask.vfnmsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32) nounwind readnone
define <16 x float> @test_mask_vfnmsub_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
; CHECK-LABEL: test_mask_vfnmsub_ps
; CHECK: vfnmsub213ps %zmm
; CHECK-LABEL: test_mask_vfnmsub_ps:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfnmsub213ps %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfnmsub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 4) nounwind
ret <16 x float> %res
}
define <8 x double> @test_x86_vfnmsubpd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_x86_vfnmsubpd_z
; CHECK: vfnmsub213pd %zmm
; CHECK-LABEL: test_x86_vfnmsubpd_z:
; CHECK: ## BB#0:
; CHECK-NEXT: vfnmsub213pd %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 4) nounwind
ret <8 x double> %res
}
declare <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32) nounwind readnone
define <8 x double> @test_mask_vfnmsub_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_vfnmsub_pd
; CHECK: vfnmsub213pd %zmm
; CHECK-LABEL: test_mask_vfnmsub_pd:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfnmsub213pd %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
ret <8 x double> %res
}
define <16 x float> @test_x86_vfmaddsubps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_x86_vfmaddsubps_z
; CHECK: vfmaddsub213ps %zmm
; CHECK-LABEL: test_x86_vfmaddsubps_z:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmaddsub213ps %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 4) nounwind
ret <16 x float> %res
}
define <16 x float> @test_mask_fmaddsub_ps(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
; CHECK-LABEL: test_mask_fmaddsub_ps:
; CHECK: vfmaddsub213ps %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0xa6,0xc2]
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmaddsub213ps %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 4)
ret <16 x float> %res
}
@ -80,16 +106,21 @@ define <16 x float> @test_mask_fmaddsub_ps(<16 x float> %a, <16 x float> %b, <16
declare <16 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32) nounwind readnone
define <8 x double> @test_x86_vfmaddsubpd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_x86_vfmaddsubpd_z
; CHECK: vfmaddsub213pd %zmm
; CHECK-LABEL: test_x86_vfmaddsubpd_z:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmaddsub213pd %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 4) nounwind
ret <8 x double> %res
}
declare <8 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32) nounwind readnone
define <8 x double> @test_mask_vfmaddsub_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_vfmaddsub_pd
; CHECK: vfmaddsub213pd %zmm
; CHECK-LABEL: test_mask_vfmaddsub_pd:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmaddsub213pd %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
ret <8 x double> %res
}
@ -97,8 +128,7 @@ define <8 x double> @test_mask_vfmaddsub_pd(<8 x double> %a0, <8 x double> %a1,
define <8 x double>@test_int_x86_avx512_mask_vfmaddsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
; CHECK-LABEL: test_int_x86_avx512_mask_vfmaddsub_pd_512:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmaddsub213pd %zmm2, %zmm1, %zmm3 {%k1}
; CHECK-NEXT: vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
@ -115,8 +145,7 @@ declare <8 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.512(<8 x double>, <8 x
define <8 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_512:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmaddsub231pd %zmm1, %zmm0, %zmm3 {%k1}
; CHECK-NEXT: vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
@ -133,8 +162,7 @@ declare <8 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.512(<8 x double>, <8 x
define <8 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_512:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmaddsub213pd %zmm2, %zmm1, %zmm3 {%k1} {z}
; CHECK-NEXT: vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
@ -200,8 +228,7 @@ declare <8 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.512(<8 x double>, <8 x
define <8 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_512:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmsubadd231pd %zmm1, %zmm0, %zmm3 {%k1}
; CHECK-NEXT: vfmsubadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
@ -231,71 +258,96 @@ define <16 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_512(<16 x float> %x0,
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rne(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rne
; CHECK: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x19,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rne:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 0) nounwind
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rtn(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rtn
; CHECK: vfmadd213ps {rd-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x39,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rtn:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmadd213ps {rd-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 1) nounwind
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rtp(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rtp
; CHECK: vfmadd213ps {ru-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x59,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rtp:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmadd213ps {ru-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 2) nounwind
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rtz(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rtz
; CHECK: vfmadd213ps {rz-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x79,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rtz:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmadd213ps {rz-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 3) nounwind
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrb_current(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_current
; CHECK: vfmadd213ps %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_current:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmadd213ps %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 4) nounwind
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rne(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rne
; CHECK: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x18,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rne:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 0) nounwind
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rtn(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtn
; CHECK: vfmadd213ps {rd-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x38,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtn:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmadd213ps {rd-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 1) nounwind
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rtp(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtp
; CHECK: vfmadd213ps {ru-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x58,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtp:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmadd213ps {ru-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 2) nounwind
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rtz(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtz
; CHECK: vfmadd213ps {rz-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x78,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtz:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmadd213ps {rz-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 3) nounwind
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_current(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_current
; CHECK: vfmadd213ps %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x48,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_current:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmadd213ps %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 4) nounwind
ret <16 x float> %res
}
@ -305,8 +357,7 @@ declare <8 x double> @llvm.x86.avx512.mask3.vfmsub.pd.512(<8 x double>, <8 x dou
define <8 x double>@test_int_x86_avx512_mask3_vfmsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_512:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmsub231pd %zmm1, %zmm0, %zmm3 {%k1}
; CHECK-NEXT: vfmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
@ -336,71 +387,96 @@ define <16 x float>@test_int_x86_avx512_mask3_vfmsub_ps_512(<16 x float> %x0, <1
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rne
; CHECK: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x19,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rne:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 0) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rtn
; CHECK: vfmadd213pd {rd-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x39,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rtn:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmadd213pd {rd-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 1) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rtp
; CHECK: vfmadd213pd {ru-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x59,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rtp:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmadd213pd {ru-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 2) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rtz
; CHECK: vfmadd213pd {rz-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x79,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rtz:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmadd213pd {rz-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 3) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrb_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_current
; CHECK: vfmadd213pd %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_current:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmadd213pd %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rne
; CHECK: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x18,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rne:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 0) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtn
; CHECK: vfmadd213pd {rd-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x38,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtn:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmadd213pd {rd-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 1) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtp
; CHECK: vfmadd213pd {ru-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x58,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtp:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmadd213pd {ru-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 2) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtz
; CHECK: vfmadd213pd {rz-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x78,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtz:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmadd213pd {rz-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 3) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_current
; CHECK: vfmadd213pd %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x48,0xa8,0xc2]
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_current:
; CHECK: ## BB#0:
; CHECK-NEXT: vfmadd213pd %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 4) nounwind
ret <8 x double> %res
}
@ -408,8 +484,7 @@ define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_current(<8 x double> %a0,
define <8 x double>@test_int_x86_avx512_mask_vfmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
; CHECK-LABEL: test_int_x86_avx512_mask_vfmadd_pd_512:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmadd213pd %zmm2, %zmm1, %zmm3 {%k1}
; CHECK-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
@ -426,8 +501,7 @@ declare <8 x double> @llvm.x86.avx512.mask3.vfmadd.pd.512(<8 x double>, <8 x dou
define <8 x double>@test_int_x86_avx512_mask3_vfmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_512:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmadd231pd %zmm1, %zmm0, %zmm3 {%k1}
; CHECK-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
@ -444,8 +518,7 @@ declare <8 x double> @llvm.x86.avx512.maskz.vfmadd.pd.512(<8 x double>, <8 x dou
define <8 x double>@test_int_x86_avx512_maskz_vfmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_512:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmadd213pd %zmm2, %zmm1, %zmm3 {%k1} {z}
; CHECK-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
@ -508,71 +581,96 @@ define <16 x float>@test_int_x86_avx512_maskz_vfmadd_ps_512(<16 x float> %x0, <1
define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrb_rne
; CHECK: vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x19,0xae,0xc2]
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrb_rne:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 0) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtn
; CHECK: vfnmsub213pd {rd-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x39,0xae,0xc2]
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtn:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfnmsub213pd {rd-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 1) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtp
; CHECK: vfnmsub213pd {ru-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x59,0xae,0xc2]
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtp:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfnmsub213pd {ru-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 2) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtz
; CHECK: vfnmsub213pd {rz-sae}, %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x79,0xae,0xc2]
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtz:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfnmsub213pd {rz-sae}, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 3) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrb_current
; CHECK: vfnmsub213pd %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0xae,0xc2]
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrb_current:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfnmsub213pd %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rne
; CHECK: vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x18,0xae,0xc2]
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rne:
; CHECK: ## BB#0:
; CHECK-NEXT: vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 0) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rtn
; CHECK: vfnmsub213pd {rd-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x38,0xae,0xc2]
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rtn:
; CHECK: ## BB#0:
; CHECK-NEXT: vfnmsub213pd {rd-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 1) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rtp
; CHECK: vfnmsub213pd {ru-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x58,0xae,0xc2]
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rtp:
; CHECK: ## BB#0:
; CHECK-NEXT: vfnmsub213pd {ru-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 2) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rtz
; CHECK: vfnmsub213pd {rz-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x78,0xae,0xc2]
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rtz:
; CHECK: ## BB#0:
; CHECK-NEXT: vfnmsub213pd {rz-sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 3) nounwind
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_current
; CHECK: vfnmsub213pd %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x48,0xae,0xc2]
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_current:
; CHECK: ## BB#0:
; CHECK-NEXT: vfnmsub213pd %zmm2, %zmm1, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 4) nounwind
ret <8 x double> %res
}
@ -580,8 +678,7 @@ define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_current(<8 x double> %a0
define <8 x double>@test_int_x86_avx512_mask_vfnmsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
; CHECK-LABEL: test_int_x86_avx512_mask_vfnmsub_pd_512:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfnmsub213pd %zmm2, %zmm1, %zmm3 {%k1}
; CHECK-NEXT: vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
@ -598,8 +695,7 @@ declare <8 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.512(<8 x double>, <8 x do
define <8 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
; CHECK-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_512:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfnmsub231pd %zmm1, %zmm0, %zmm3 {%k1}
; CHECK-NEXT: vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
@ -646,8 +742,7 @@ define <16 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_512(<16 x float> %x0, <
define <8 x double>@test_int_x86_avx512_mask_vfnmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
; CHECK-LABEL: test_int_x86_avx512_mask_vfnmadd_pd_512:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfnmadd213pd %zmm2, %zmm1, %zmm3 {%k1}
; CHECK-NEXT: vfnmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0

728
test/CodeGen/X86/avx512-intrinsics.ll
View File

File diff suppressed because it is too large Load Diff

7
test/CodeGen/X86/avx512-mask-op.ll
View File

@ -18,8 +18,7 @@ define i16 @mask16(i16 %x) {
define i8 @mask8(i8 %x) {
; KNL-LABEL: mask8:
; KNL: ## BB#0:
; KNL-NEXT: movzbl %dil, %eax
; KNL-NEXT: kmovw %eax, %k0
; KNL-NEXT: kmovw %edi, %k0
; KNL-NEXT: knotw %k0, %k0
; KNL-NEXT: kmovw %k0, %eax
; KNL-NEXT: retq
@ -232,7 +231,6 @@ define void @test7(<8 x i1> %mask) {
; KNL-NEXT: vpsllq $63, %zmm0, %zmm0
; KNL-NEXT: vptestmq %zmm0, %zmm0, %k0
; KNL-NEXT: movb $85, %al
; KNL-NEXT: movzbl %al, %eax
; KNL-NEXT: kmovw %eax, %k1
; KNL-NEXT: korw %k1, %k0, %k0
; KNL-NEXT: kmovw %k0, %eax
@ -1337,8 +1335,7 @@ define <64 x i8> @test17(i64 %x, i32 %y, i32 %z) {
define <8 x i1> @test18(i8 %a, i16 %y) {
; KNL-LABEL: test18:
; KNL: ## BB#0:
; KNL-NEXT: movzbl %dil, %eax
; KNL-NEXT: kmovw %eax, %k0
; KNL-NEXT: kmovw %edi, %k0
; KNL-NEXT: kmovw %esi, %k1
; KNL-NEXT: kshiftlw $7, %k1, %k2
; KNL-NEXT: kshiftrw $15, %k2, %k2

106
test/CodeGen/X86/avx512-select.ll
View File

@ -1,62 +1,81 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s
; CHECK-LABEL: select00
; CHECK: vmovaps
; CHECK-NEXT: LBB
define <16 x i32> @select00(i32 %a, <16 x i32> %b) nounwind {
; CHECK-LABEL: select00:
; CHECK: ## BB#0:
; CHECK-NEXT: vpxord %zmm1, %zmm1, %zmm1
; CHECK-NEXT: cmpl $255, %edi
; CHECK-NEXT: je LBB0_2
; CHECK-NEXT: ## BB#1:
; CHECK-NEXT: vmovaps %zmm0, %zmm1
; CHECK-NEXT: LBB0_2:
; CHECK-NEXT: vpxord %zmm1, %zmm0, %zmm0
; CHECK-NEXT: retq
%cmpres = icmp eq i32 %a, 255
%selres = select i1 %cmpres, <16 x i32> zeroinitializer, <16 x i32> %b
%res = xor <16 x i32> %b, %selres
ret <16 x i32> %res
}
; CHECK-LABEL: select01
; CHECK: vmovaps
; CHECK-NEXT: LBB
define <8 x i64> @select01(i32 %a, <8 x i64> %b) nounwind {
; CHECK-LABEL: select01:
; CHECK: ## BB#0:
; CHECK-NEXT: vpxord %zmm1, %zmm1, %zmm1
; CHECK-NEXT: cmpl $255, %edi
; CHECK-NEXT: je LBB1_2
; CHECK-NEXT: ## BB#1:
; CHECK-NEXT: vmovaps %zmm0, %zmm1
; CHECK-NEXT: LBB1_2:
; CHECK-NEXT: vpxorq %zmm1, %zmm0, %zmm0
; CHECK-NEXT: retq
%cmpres = icmp eq i32 %a, 255
%selres = select i1 %cmpres, <8 x i64> zeroinitializer, <8 x i64> %b
%res = xor <8 x i64> %b, %selres
ret <8 x i64> %res
}
; CHECK-LABEL: @select02
; CHECK: cmpless %xmm0, %xmm3, %k1
; CHECK-NEXT: vmovss %xmm2, {{.*}}%xmm1 {%k1}
; CHECK: ret
define float @select02(float %a, float %b, float %c, float %eps) {
; CHECK-LABEL: select02:
; CHECK: ## BB#0:
; CHECK-NEXT: vcmpless %xmm0, %xmm3, %k1
; CHECK-NEXT: vmovss %xmm2, %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vmovaps %zmm1, %zmm0
; CHECK-NEXT: retq
%cmp = fcmp oge float %a, %eps
%cond = select i1 %cmp, float %c, float %b
ret float %cond
}
; CHECK-LABEL: @select03
; CHECK: cmplesd %xmm0, %xmm3, %k1
; CHECK-NEXT: vmovsd %xmm2, {{.*}}%xmm1 {%k1}
; CHECK: ret
define double @select03(double %a, double %b, double %c, double %eps) {
; CHECK-LABEL: select03:
; CHECK: ## BB#0:
; CHECK-NEXT: vcmplesd %xmm0, %xmm3, %k1
; CHECK-NEXT: vmovsd %xmm2, %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vmovaps %zmm1, %zmm0
; CHECK-NEXT: retq
%cmp = fcmp oge double %a, %eps
%cond = select i1 %cmp, double %c, double %b
ret double %cond
}
; CHECK-LABEL: @select04
; CHECK: vmovaps %zmm3, %zmm1
; CHECK-NEXT: ret
; PR20677
define <16 x double> @select04(<16 x double> %a, <16 x double> %b) {
; CHECK-LABEL: select04:
; CHECK: ## BB#0:
; CHECK-NEXT: vmovaps %zmm3, %zmm1
; CHECK-NEXT: retq
%sel = select <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false>, <16 x double> %a, <16 x double> %b
ret <16 x double> %sel
}
; CHECK-LABEL: select05
; CHECK: movzbl %sil, %eax
; CHECK: kmovw %eax, %k0
; CHECK: movzbl %dil, %eax
; CHECK: kmovw %eax, %k1
; CHECK-NEXT: korw %k1, %k0, %k0
; CHECK-NEXT: kmovw %k0, %eax
define i8 @select05(i8 %a.0, i8 %m) {
; CHECK-LABEL: select05:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k0
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: korw %k1, %k0, %k0
; CHECK-NEXT: kmovw %k0, %eax
; CHECK-NEXT: retq
%mask = bitcast i8 %m to <8 x i1>
%a = bitcast i8 %a.0 to <8 x i1>
%r = select <8 x i1> %mask, <8 x i1> <i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1>, <8 x i1> %a
@ -64,14 +83,14 @@ define i8 @select05(i8 %a.0, i8 %m) {
ret i8 %res;
}
; CHECK-LABEL: select06
; CHECK: movzbl %sil, %eax
; CHECK: kmovw %eax, %k0
; CHECK: movzbl %dil, %eax
; CHECK: kmovw %eax, %k1
; CHECK-NEXT: kandw %k1, %k0, %k0
; CHECK-NEXT: kmovw %k0, %eax
define i8 @select06(i8 %a.0, i8 %m) {
; CHECK-LABEL: select06:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k0
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: kandw %k1, %k0, %k0
; CHECK-NEXT: kmovw %k0, %eax
; CHECK-NEXT: retq
%mask = bitcast i8 %m to <8 x i1>
%a = bitcast i8 %a.0 to <8 x i1>
%r = select <8 x i1> %mask, <8 x i1> %a, <8 x i1> zeroinitializer
@ -79,19 +98,18 @@ define i8 @select06(i8 %a.0, i8 %m) {
ret i8 %res;
}
; CHECK-LABEL: select07
; CHECK-DAG: movzbl %dl, %eax
; CHECK-DAG: kmovw %eax, %k0
; CHECK-DAG: movzbl %dil, %eax
; CHECK-DAG: kmovw %eax, %k1
; CHECK-DAG: movzbl %sil, %eax
; CHECK-DAG: kmovw %eax, %k2
; CHECK: kandw %k0, %k1, %k1
; CHECK-NEXT: knotw %k0, %k0
; CHECK-NEXT: kandw %k0, %k2, %k0
; CHECK-NEXT: korw %k0, %k1, %k0
; CHECK-NEXT: kmovw %k0, %eax
define i8 @select07(i8 %a.0, i8 %b.0, i8 %m) {
; CHECK-LABEL: select07:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edx, %k0
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: kmovw %esi, %k2
; CHECK-NEXT: kandw %k0, %k1, %k1
; CHECK-NEXT: knotw %k0, %k0
; CHECK-NEXT: kandw %k0, %k2, %k0
; CHECK-NEXT: korw %k0, %k1, %k0
; CHECK-NEXT: kmovw %k0, %eax
; CHECK-NEXT: retq
%mask = bitcast i8 %m to <8 x i1>
%a = bitcast i8 %a.0 to <8 x i1>
%b = bitcast i8 %b.0 to <8 x i1>

204
test/CodeGen/X86/avx512bwvl-intrinsics.ll
View File

@ -651,8 +651,7 @@ define <2 x double> @test_mask_fmadd128_pd(<2 x double> %a, <2 x double> %b, <2
define <2 x double>@test_int_x86_avx512_mask_vfmadd_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfmadd_pd_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm3 {%k1}
; CHECK-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm0
@ -669,8 +668,7 @@ declare <2 x double> @llvm.x86.avx512.mask3.vfmadd.pd.128(<2 x double>, <2 x dou
define <2 x double>@test_int_x86_avx512_mask3_vfmadd_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmadd231pd %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm0
@ -687,8 +685,7 @@ declare <2 x double> @llvm.x86.avx512.maskz.vfmadd.pd.128(<2 x double>, <2 x dou
define <2 x double>@test_int_x86_avx512_maskz_vfmadd_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm3 {%k1} {z}
; CHECK-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm0
@ -703,8 +700,7 @@ define <2 x double>@test_int_x86_avx512_maskz_vfmadd_pd_128(<2 x double> %x0, <2
define <4 x double>@test_int_x86_avx512_mask_vfmadd_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfmadd_pd_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm3 {%k1}
; CHECK-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm0
@ -721,8 +717,7 @@ declare <4 x double> @llvm.x86.avx512.mask3.vfmadd.pd.256(<4 x double>, <4 x dou
define <4 x double>@test_int_x86_avx512_mask3_vfmadd_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmadd231pd %ymm1, %ymm0, %ymm3 {%k1}
; CHECK-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm0
@ -739,8 +734,7 @@ declare <4 x double> @llvm.x86.avx512.maskz.vfmadd.pd.256(<4 x double>, <4 x dou
define <4 x double>@test_int_x86_avx512_maskz_vfmadd_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm3 {%k1} {z}
; CHECK-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm0
@ -755,8 +749,7 @@ define <4 x double>@test_int_x86_avx512_maskz_vfmadd_pd_256(<4 x double> %x0, <4
define <4 x float>@test_int_x86_avx512_mask_vfmadd_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfmadd_ps_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm3 {%k1}
; CHECK-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0
@ -773,8 +766,7 @@ declare <4 x float> @llvm.x86.avx512.mask3.vfmadd.ps.128(<4 x float>, <4 x float
define <4 x float>@test_int_x86_avx512_mask3_vfmadd_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmadd231ps %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0
@ -791,8 +783,7 @@ declare <4 x float> @llvm.x86.avx512.maskz.vfmadd.ps.128(<4 x float>, <4 x float
define <4 x float>@test_int_x86_avx512_maskz_vfmadd_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmadd_ps_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm3 {%k1} {z}
; CHECK-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0
@ -807,8 +798,7 @@ define <4 x float>@test_int_x86_avx512_maskz_vfmadd_ps_128(<4 x float> %x0, <4 x
define <8 x float>@test_int_x86_avx512_mask_vfmadd_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfmadd_ps_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm3 {%k1}
; CHECK-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm0
@ -825,8 +815,7 @@ declare <8 x float> @llvm.x86.avx512.mask3.vfmadd.ps.256(<8 x float>, <8 x float
define <8 x float>@test_int_x86_avx512_mask3_vfmadd_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmadd231ps %ymm1, %ymm0, %ymm3 {%k1}
; CHECK-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm0
@ -843,8 +832,7 @@ declare <8 x float> @llvm.x86.avx512.maskz.vfmadd.ps.256(<8 x float>, <8 x float
define <8 x float>@test_int_x86_avx512_maskz_vfmadd_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmadd_ps_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm3 {%k1} {z}
; CHECK-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm0
@ -862,8 +850,7 @@ declare <2 x double> @llvm.x86.avx512.mask3.vfmsub.pd.128(<2 x double>, <2 x dou
define <2 x double>@test_int_x86_avx512_mask3_vfmsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmsub231pd %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT: vfmsub213pd %xmm2, %xmm1, %xmm0
@ -881,8 +868,7 @@ declare <4 x double> @llvm.x86.avx512.mask3.vfmsub.pd.256(<4 x double>, <4 x dou
define <4 x double>@test_int_x86_avx512_mask3_vfmsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmsub231pd %ymm1, %ymm0, %ymm3 {%k1}
; CHECK-NEXT: vfmsub213pd %ymm2, %ymm1, %ymm0
@ -899,8 +885,7 @@ declare <4 x float> @llvm.x86.avx512.mask3.vfmsub.ps.128(<4 x float>, <4 x float
define <4 x float>@test_int_x86_avx512_mask3_vfmsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_ps_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmsub231ps %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT: vfmsub213ps %xmm2, %xmm1, %xmm0
@ -917,8 +902,7 @@ declare <8 x float> @llvm.x86.avx512.mask3.vfmsub.ps.256(<8 x float>, <8 x float
define <8 x float>@test_int_x86_avx512_mask3_vfmsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_ps_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmsub231ps %ymm1, %ymm0, %ymm3 {%k1}
; CHECK-NEXT: vfmsub213ps %ymm2, %ymm1, %ymm0
@ -1006,8 +990,7 @@ define <2 x double> @test_mask_vfnmsub128_pd(<2 x double> %a0, <2 x double> %a1,
define <2 x double>@test_int_x86_avx512_mask_vfnmsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfnmsub_pd_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfnmsub213pd %xmm2, %xmm1, %xmm3 {%k1}
; CHECK-NEXT: vfnmsub213pd %xmm2, %xmm1, %xmm0
@ -1024,8 +1007,7 @@ declare <2 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.128(<2 x double>, <2 x do
define <2 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfnmsub231pd %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT: vfnmsub213pd %xmm2, %xmm1, %xmm0
@ -1040,8 +1022,7 @@ define <2 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_128(<2 x double> %x0, <
define <4 x double>@test_int_x86_avx512_mask_vfnmsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfnmsub_pd_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfnmsub213pd %ymm2, %ymm1, %ymm3 {%k1}
; CHECK-NEXT: vfnmsub213pd %ymm2, %ymm1, %ymm0
@ -1058,8 +1039,7 @@ declare <4 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.256(<4 x double>, <4 x do
define <4 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfnmsub231pd %ymm1, %ymm0, %ymm3 {%k1}
; CHECK-NEXT: vfnmsub213pd %ymm2, %ymm1, %ymm0
@ -1074,8 +1054,7 @@ define <4 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_256(<4 x double> %x0, <
define <4 x float>@test_int_x86_avx512_mask_vfnmsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfnmsub_ps_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfnmsub213ps %xmm2, %xmm1, %xmm3 {%k1}
; CHECK-NEXT: vfnmsub213ps %xmm2, %xmm1, %xmm0
@ -1092,8 +1071,7 @@ declare <4 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.128(<4 x float>, <4 x floa
define <4 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfnmsub_ps_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfnmsub231ps %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT: vfnmsub213ps %xmm2, %xmm1, %xmm0
@ -1108,8 +1086,7 @@ define <4 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_128(<4 x float> %x0, <4
define <8 x float>@test_int_x86_avx512_mask_vfnmsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfnmsub_ps_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfnmsub213ps %ymm2, %ymm1, %ymm3 {%k1}
; CHECK-NEXT: vfnmsub213ps %ymm2, %ymm1, %ymm0
@ -1126,8 +1103,7 @@ declare <8 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.256(<8 x float>, <8 x floa
define <8 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfnmsub_ps_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfnmsub231ps %ymm1, %ymm0, %ymm3 {%k1}
; CHECK-NEXT: vfnmsub213ps %ymm2, %ymm1, %ymm0
@ -1142,8 +1118,7 @@ define <8 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_256(<8 x float> %x0, <8
define <2 x double>@test_int_x86_avx512_mask_vfnmadd_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfnmadd_pd_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfnmadd213pd %xmm2, %xmm1, %xmm3 {%k1}
; CHECK-NEXT: vfnmadd213pd %xmm2, %xmm1, %xmm0
@ -1158,8 +1133,7 @@ define <2 x double>@test_int_x86_avx512_mask_vfnmadd_pd_128(<2 x double> %x0, <2
define <4 x double>@test_int_x86_avx512_mask_vfnmadd_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfnmadd_pd_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfnmadd213pd %ymm2, %ymm1, %ymm3 {%k1}
; CHECK-NEXT: vfnmadd213pd %ymm2, %ymm1, %ymm0
@ -1174,8 +1148,7 @@ define <4 x double>@test_int_x86_avx512_mask_vfnmadd_pd_256(<4 x double> %x0, <4
define <4 x float>@test_int_x86_avx512_mask_vfnmadd_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfnmadd_ps_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfnmadd213ps %xmm2, %xmm1, %xmm3 {%k1}
; CHECK-NEXT: vfnmadd213ps %xmm2, %xmm1, %xmm0
@ -1190,8 +1163,7 @@ define <4 x float>@test_int_x86_avx512_mask_vfnmadd_ps_128(<4 x float> %x0, <4 x
define <8 x float>@test_int_x86_avx512_mask_vfnmadd_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfnmadd_ps_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfnmadd213ps %ymm2, %ymm1, %ymm3 {%k1}
; CHECK-NEXT: vfnmadd213ps %ymm2, %ymm1, %ymm0
@ -1242,8 +1214,7 @@ define <2 x double> @test_mask_vfmaddsub128_pd(<2 x double> %a0, <2 x double> %a
define <2 x double>@test_int_x86_avx512_mask_vfmaddsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfmaddsub_pd_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmaddsub213pd %xmm2, %xmm1, %xmm3 {%k1}
; CHECK-NEXT: vfmaddsub213pd %xmm2, %xmm1, %xmm0
@ -1260,8 +1231,7 @@ declare <2 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.128(<2 x double>, <2 x
define <2 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmaddsub231pd %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT: vfmaddsub213pd %xmm2, %xmm1, %xmm0
@ -1278,8 +1248,7 @@ declare <2 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.128(<2 x double>, <2 x
define <2 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmaddsub213pd %xmm2, %xmm1, %xmm3 {%k1} {z}
; CHECK-NEXT: vfmaddsub213pd %xmm2, %xmm1, %xmm0
@ -1294,8 +1263,7 @@ define <2 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_128(<2 x double> %x0,
define <4 x double>@test_int_x86_avx512_mask_vfmaddsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfmaddsub_pd_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmaddsub213pd %ymm2, %ymm1, %ymm3 {%k1}
; CHECK-NEXT: vfmaddsub213pd %ymm2, %ymm1, %ymm0
@ -1312,8 +1280,7 @@ declare <4 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.256(<4 x double>, <4 x
define <4 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmaddsub231pd %ymm1, %ymm0, %ymm3 {%k1}
; CHECK-NEXT: vfmaddsub213pd %ymm2, %ymm1, %ymm0
@ -1330,8 +1297,7 @@ declare <4 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.256(<4 x double>, <4 x
define <4 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmaddsub213pd %ymm2, %ymm1, %ymm3 {%k1} {z}
; CHECK-NEXT: vfmaddsub213pd %ymm2, %ymm1, %ymm0
@ -1346,8 +1312,7 @@ define <4 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_256(<4 x double> %x0,
define <4 x float>@test_int_x86_avx512_mask_vfmaddsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfmaddsub_ps_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmaddsub213ps %xmm2, %xmm1, %xmm3 {%k1}
; CHECK-NEXT: vfmaddsub213ps %xmm2, %xmm1, %xmm0
@ -1364,8 +1329,7 @@ declare <4 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.128(<4 x float>, <4 x fl
define <4 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmaddsub231ps %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT: vfmaddsub213ps %xmm2, %xmm1, %xmm0
@ -1382,8 +1346,7 @@ declare <4 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.128(<4 x float>, <4 x fl
define <4 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ps_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmaddsub213ps %xmm2, %xmm1, %xmm3 {%k1} {z}
; CHECK-NEXT: vfmaddsub213ps %xmm2, %xmm1, %xmm0
@ -1398,8 +1361,7 @@ define <4 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_128(<4 x float> %x0, <
define <8 x float>@test_int_x86_avx512_mask_vfmaddsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfmaddsub_ps_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmaddsub213ps %ymm2, %ymm1, %ymm3 {%k1}
; CHECK-NEXT: vfmaddsub213ps %ymm2, %ymm1, %ymm0
@ -1416,8 +1378,7 @@ declare <8 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.256(<8 x float>, <8 x fl
define <8 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmaddsub231ps %ymm1, %ymm0, %ymm3 {%k1}
; CHECK-NEXT: vfmaddsub213ps %ymm2, %ymm1, %ymm0
@ -1434,8 +1395,7 @@ declare <8 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.256(<8 x float>, <8 x fl
define <8 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ps_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm0, %zmm3
; CHECK-NEXT: vfmaddsub213ps %ymm2, %ymm1, %ymm3 {%k1} {z}
; CHECK-NEXT: vfmaddsub213ps %ymm2, %ymm1, %ymm0
@ -1452,8 +1412,7 @@ declare <2 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.128(<2 x double>, <2 x
define <2 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmsubadd231pd %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT: vfmsubadd213pd %xmm2, %xmm1, %xmm0
@ -1470,8 +1429,7 @@ declare <4 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.256(<4 x double>, <4 x
define <4 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmsubadd231pd %ymm1, %ymm0, %ymm3 {%k1}
; CHECK-NEXT: vfmsubadd213pd %ymm2, %ymm1, %ymm0
@ -1488,8 +1446,7 @@ declare <4 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.128(<4 x float>, <4 x fl
define <4 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmsubadd231ps %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT: vfmsubadd213ps %xmm2, %xmm1, %xmm0
@ -1506,8 +1463,7 @@ declare <8 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.256(<8 x float>, <8 x fl
define <8 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vmovaps %zmm2, %zmm3
; CHECK-NEXT: vfmsubadd231ps %ymm1, %ymm0, %ymm3 {%k1}
; CHECK-NEXT: vfmsubadd213ps %ymm2, %ymm1, %ymm0
@ -4037,8 +3993,7 @@ declare <4 x i32> @llvm.x86.avx512.mask.pmaddw.d.128(<8 x i16>, <8 x i16>, <4 x
define <4 x i32>@test_int_x86_avx512_mask_pmaddw_d_128(<8 x i16> %x0, <8 x i16> %x1, <4 x i32> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmaddw_d_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpmaddwd %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT: vpmaddwd %xmm1, %xmm0, %xmm0
; CHECK-NEXT: vpaddd %xmm0, %xmm2, %xmm0
@ -4054,8 +4009,7 @@ declare <8 x i32> @llvm.x86.avx512.mask.pmaddw.d.256(<16 x i16>, <16 x i16>, <8
define <8 x i32>@test_int_x86_avx512_mask_pmaddw_d_256(<16 x i16> %x0, <16 x i16> %x1, <8 x i32> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmaddw_d_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpmaddwd %ymm1, %ymm0, %ymm2 {%k1}
; CHECK-NEXT: vpmaddwd %ymm1, %ymm0, %ymm0
; CHECK-NEXT: vpaddd %ymm0, %ymm2, %ymm0
@ -4071,8 +4025,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.pmaddubs.w.128(<16 x i8>, <16 x i8>, <8
define <8 x i16>@test_int_x86_avx512_mask_pmaddubs_w_128(<16 x i8> %x0, <16 x i8> %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmaddubs_w_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT: vpmaddubsw %xmm1, %xmm0, %xmm0
; CHECK-NEXT: vpaddw %xmm0, %xmm2, %xmm0
@ -4256,8 +4209,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.dbpsadbw.128(<16 x i8>, <16 x i8>, i32,
define <8 x i16>@test_int_x86_avx512_mask_dbpsadbw_128(<16 x i8> %x0, <16 x i8> %x1, <8 x i16> %x3, i8 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_dbpsadbw_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vdbpsadbw $2, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT: vdbpsadbw $2, %xmm1, %xmm0, %xmm3 {%k1} {z}
; CHECK-NEXT: vdbpsadbw $2, %xmm1, %xmm0, %xmm0
@ -4357,8 +4309,7 @@ declare <8 x i16> @llvm.x86.avx512.pbroadcastw.128(<8 x i16>, <8 x i16>, i8)
define <8 x i16>@test_int_x86_avx512_pbroadcastw_128(<8 x i16> %x0, <8 x i16> %x1, i8 %mask) {
; CHECK-LABEL: test_int_x86_avx512_pbroadcastw_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpbroadcastw %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpbroadcastw %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vpbroadcastw %xmm0, %xmm0
@ -4490,8 +4441,7 @@ declare <8 x i16> @llvm.x86.avx512.cvtmask2w.128(i8)
define <8 x i16>@test_int_x86_avx512_cvtmask2w_128(i8 %x0) {
; CHECK-LABEL: test_int_x86_avx512_cvtmask2w_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k0
; CHECK-NEXT: kmovw %edi, %k0
; CHECK-NEXT: vpmovm2w %k0, %xmm0
; CHECK-NEXT: retq
%res = call <8 x i16> @llvm.x86.avx512.cvtmask2w.128(i8 %x0)
@ -4515,8 +4465,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.psrl.w.128(<8 x i16>, <8 x i16>, <8 x i1
define <8 x i16>@test_int_x86_avx512_mask_psrl_w_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_psrl_w_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpsrlw %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT: vpsrlw %xmm1, %xmm0, %xmm3 {%k1} {z}
; CHECK-NEXT: vpsrlw %xmm1, %xmm0, %xmm0
@ -4556,8 +4505,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.psrl.wi.128(<8 x i16>, i8, <8 x i16>, i8
define <8 x i16>@test_int_x86_avx512_mask_psrl_wi_128(<8 x i16> %x0, i8 %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_psrl_wi_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %sil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %esi, %k1
; CHECK-NEXT: vpsrlw $3, %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpsrlw $3, %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vpsrlw $3, %xmm0, %xmm0
@ -4617,8 +4565,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.psrlv8.hi(<8 x i16>, <8 x i16>, <8 x i16
define <8 x i16>@test_int_x86_avx512_mask_psrlv8_hi(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_psrlv8_hi:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpsrlvw %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT: vpsrlvw %xmm1, %xmm0, %xmm3 {%k1} {z}
; CHECK-NEXT: vpsrlvw %xmm1, %xmm0, %xmm0
@ -4638,8 +4585,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.psra.w.128(<8 x i16>, <8 x i16>, <8 x i1
define <8 x i16>@test_int_x86_avx512_mask_psra_w_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_psra_w_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpsraw %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT: vpsraw %xmm1, %xmm0, %xmm3 {%k1} {z}
; CHECK-NEXT: vpsraw %xmm1, %xmm0, %xmm0
@ -4659,8 +4605,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.psra.wi.128(<8 x i16>, i8, <8 x i16>, i8
define <8 x i16>@test_int_x86_avx512_mask_psra_wi_128(<8 x i16> %x0, i8 %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_psra_wi_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %sil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %esi, %k1
; CHECK-NEXT: vpsraw $3, %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpsraw $3, %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vpsraw $3, %xmm0, %xmm0
@ -4720,8 +4665,7 @@ declare <4 x i32> @llvm.x86.avx512.mask.pshuf.d.128(<4 x i32>, i16, <4 x i32>, i
define <4 x i32>@test_int_x86_avx512_mask_pshuf_d_128(<4 x i32> %x0, i16 %x1, <4 x i32> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_pshuf_d_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %sil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %esi, %k1
; CHECK-NEXT: vpshufd $3, %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpshufd $3, %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vpshufd $3, %xmm0, %xmm0
@ -4742,8 +4686,7 @@ declare <8 x i32> @llvm.x86.avx512.mask.pshuf.d.256(<8 x i32>, i16, <8 x i32>, i
define <8 x i32>@test_int_x86_avx512_mask_pshuf_d_256(<8 x i32> %x0, i16 %x1, <8 x i32> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_pshuf_d_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %sil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %esi, %k1
; CHECK-NEXT: vpshufd $3, %ymm0, %ymm1 {%k1}
; CHECK-NEXT: vpshufd $3, %ymm0, %ymm2 {%k1} {z}
; CHECK-NEXT: vpshufd $3, %ymm0, %ymm0
@ -4764,8 +4707,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.pshufh.w.128(<8 x i16>, i8, <8 x i16>, i
define <8 x i16>@test_int_x86_avx512_mask_pshufh_w_128(<8 x i16> %x0, i8 %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_pshufh_w_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %sil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %esi, %k1
; CHECK-NEXT: vpshufhw $3, %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpshufhw $3, %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vpshufhw $3, %xmm0, %xmm0
@ -4807,8 +4749,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.pshufl.w.128(<8 x i16>, i8, <8 x i16>, i
define <8 x i16>@test_int_x86_avx512_mask_pshufl_w_128(<8 x i16> %x0, i8 %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_pshufl_w_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %sil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %esi, %k1
; CHECK-NEXT: vpshuflw $3, %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpshuflw $3, %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vpshuflw $3, %xmm0, %xmm0
@ -4870,8 +4811,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.psrav8.hi(<8 x i16>, <8 x i16>, <8 x i16
define <8 x i16>@test_int_x86_avx512_mask_psrav8_hi(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_psrav8_hi:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpsravw %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT: vpsravw %xmm1, %xmm0, %xmm3 {%k1} {z}
; CHECK-NEXT: vpsravw %xmm1, %xmm0, %xmm0
@ -4892,8 +4832,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.psll.w.128(<8 x i16>, <8 x i16>, <8 x i1
define <8 x i16>@test_int_x86_avx512_mask_psll_w_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_psll_w_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpsllw %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT: vpsllw %xmm1, %xmm0, %xmm3 {%k1} {z}
; CHECK-NEXT: vpsllw %xmm1, %xmm0, %xmm0
@ -4933,8 +4872,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.psll.wi.128(<8 x i16>, i8, <8 x i16>, i8
define <8 x i16>@test_int_x86_avx512_mask_psll_wi_128(<8 x i16> %x0, i8 %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_psll_wi_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %sil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %esi, %k1
; CHECK-NEXT: vpsllw $3, %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpsllw $3, %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vpsllw $3, %xmm0, %xmm0
@ -4994,8 +4932,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.psllv8.hi(<8 x i16>, <8 x i16>, <8 x i16
define <8 x i16>@test_int_x86_avx512_mask_psllv8_hi(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_psllv8_hi:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpsllvw %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT: vpsllvw %xmm1, %xmm0, %xmm3 {%k1} {z}
; CHECK-NEXT: vpsllvw %xmm1, %xmm0, %xmm0
@ -5015,8 +4952,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.pmovzxb.w.128(<16 x i8>, <8 x i16>, i8)
define <8 x i16>@test_int_x86_avx512_mask_pmovzxb_w_128(<16 x i8> %x0, <8 x i16> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovzxb_w_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpmovzxbw %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpmovzxbw %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vpmovzxbw %xmm0, %xmm0
@ -5057,8 +4993,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.pmovsxb.w.128(<16 x i8>, <8 x i16>, i8)
define <8 x i16>@test_int_x86_avx512_mask_pmovsxb_w_128(<16 x i8> %x0, <8 x i16> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovsxb_w_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpmovsxbw %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpmovsxbw %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vpmovsxbw %xmm0, %xmm0
@ -5098,8 +5033,7 @@ declare <2 x i64> @llvm.x86.avx512.mask.pmovsxd.q.128(<4 x i32>, <2 x i64>, i8)
define <2 x i64>@test_int_x86_avx512_mask_pmovsxd_q_128(<4 x i32> %x0, <2 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovsxd_q_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpmovsxdq %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpmovsxdq %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vpmovsxdq %xmm0, %xmm0
@ -5119,8 +5053,7 @@ declare <4 x i64> @llvm.x86.avx512.mask.pmovsxd.q.256(<4 x i32>, <4 x i64>, i8)
define <4 x i64>@test_int_x86_avx512_mask_pmovsxd_q_256(<4 x i32> %x0, <4 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovsxd_q_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpmovsxdq %xmm0, %ymm1 {%k1}
; CHECK-NEXT: vpmovsxdq %xmm0, %ymm2 {%k1} {z}
; CHECK-NEXT: vpmovsxdq %xmm0, %ymm0
@ -5140,8 +5073,7 @@ declare <8 x i16> @llvm.x86.avx512.mask.permvar.hi.128(<8 x i16>, <8 x i16>, <8
define <8 x i16>@test_int_x86_avx512_mask_permvar_hi_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_permvar_hi_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpermw %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT: vpermw %xmm1, %xmm0, %xmm3 {%k1} {z}
; CHECK-NEXT: vpermw %xmm1, %xmm0, %xmm0

61
test/CodeGen/X86/avx512cdvl-intrinsics.ll
View File

@ -1,3 +1,4 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512cd -mattr=+avx512vl| FileCheck %s
declare <2 x i64> @llvm.ctlz.v2i64(<2 x i64>, i1) nounwind readonly
@ -7,8 +8,7 @@ declare <4 x i32> @llvm.x86.avx512.mask.lzcnt.d.128(<4 x i32>, <4 x i32>, i8)
define <4 x i32>@test_int_x86_avx512_mask_vplzcnt_d_128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vplzcnt_d_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vplzcntd %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vplzcntd %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vplzcntd %xmm0, %xmm0
@ -28,8 +28,7 @@ declare <8 x i32> @llvm.x86.avx512.mask.lzcnt.d.256(<8 x i32>, <8 x i32>, i8)
define <8 x i32>@test_int_x86_avx512_mask_vplzcnt_d_256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vplzcnt_d_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vplzcntd %ymm0, %ymm1 {%k1}
; CHECK-NEXT: vplzcntd %ymm0, %ymm0
; CHECK-NEXT: vpaddd %ymm0, %ymm1, %ymm0
@ -45,8 +44,7 @@ declare <2 x i64> @llvm.x86.avx512.mask.lzcnt.q.128(<2 x i64>, <2 x i64>, i8)
define <2 x i64>@test_int_x86_avx512_mask_vplzcnt_q_128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vplzcnt_q_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vplzcntq %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vplzcntq %xmm0, %xmm0
; CHECK-NEXT: vpaddq %xmm0, %xmm1, %xmm0
@ -62,8 +60,7 @@ declare <4 x i64> @llvm.x86.avx512.mask.lzcnt.q.256(<4 x i64>, <4 x i64>, i8)
define <4 x i64>@test_int_x86_avx512_mask_vplzcnt_q_256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vplzcnt_q_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vplzcntq %ymm0, %ymm1 {%k1}
; CHECK-NEXT: vplzcntq %ymm0, %ymm0
; CHECK-NEXT: vpaddq %ymm0, %ymm1, %ymm0
@ -79,8 +76,7 @@ declare <4 x i32> @llvm.x86.avx512.mask.conflict.d.128(<4 x i32>, <4 x i32>, i8)
define <4 x i32>@test_int_x86_avx512_mask_vpconflict_d_128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpconflict_d_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpconflictd %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpconflictd %xmm0, %xmm2 {%k1} {z}
; CHECK-NEXT: vpconflictd %xmm0, %xmm0
@ -100,8 +96,7 @@ declare <8 x i32> @llvm.x86.avx512.mask.conflict.d.256(<8 x i32>, <8 x i32>, i8)
define <8 x i32>@test_int_x86_avx512_mask_vpconflict_d_256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpconflict_d_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpconflictd %ymm0, %ymm1 {%k1}
; CHECK-NEXT: vpconflictd %ymm0, %ymm0
; CHECK-NEXT: vpaddd %ymm0, %ymm1, %ymm0
@ -117,8 +112,7 @@ declare <2 x i64> @llvm.x86.avx512.mask.conflict.q.128(<2 x i64>, <2 x i64>, i8)
define <2 x i64>@test_int_x86_avx512_mask_vpconflict_q_128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpconflict_q_128:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpconflictq %xmm0, %xmm1 {%k1}
; CHECK-NEXT: vpconflictq %xmm0, %xmm0
; CHECK-NEXT: vpaddq %xmm0, %xmm1, %xmm0
@ -134,8 +128,7 @@ declare <4 x i64> @llvm.x86.avx512.mask.conflict.q.256(<4 x i64>, <4 x i64>, i8)
define <4 x i64>@test_int_x86_avx512_mask_vpconflict_q_256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpconflict_q_256:
; CHECK: ## BB#0:
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: kmovw %eax, %k1
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vpconflictq %ymm0, %ymm1 {%k1}
; CHECK-NEXT: vpconflictq %ymm0, %ymm0
; CHECK-NEXT: vpaddq %ymm0, %ymm1, %ymm0
@ -147,33 +140,45 @@ define <4 x i64>@test_int_x86_avx512_mask_vpconflict_q_256(<4 x i64> %x0, <4 x i
}
define <8 x i32> @test_x86_vbroadcastmw_256(i16 %a0) {
; CHECK: test_x86_vbroadcastmw_256
; CHECK: vpbroadcastmw2d %k0, %ymm0
%res = call <8 x i32> @llvm.x86.avx512.broadcastmw.256(i16 %a0) ;
; CHECK-LABEL: test_x86_vbroadcastmw_256:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k0
; CHECK-NEXT: vpbroadcastmw2d %k0, %ymm0
; CHECK-NEXT: retq
%res = call <8 x i32> @llvm.x86.avx512.broadcastmw.256(i16 %a0) ;
ret <8 x i32> %res
}
declare <8 x i32> @llvm.x86.avx512.broadcastmw.256(i16)
define <4 x i32> @test_x86_vbroadcastmw_128(i16 %a0) {
; CHECK: test_x86_vbroadcastmw_128
; CHECK: vpbroadcastmw2d %k0, %xmm0
%res = call <4 x i32> @llvm.x86.avx512.broadcastmw.128(i16 %a0) ;
; CHECK-LABEL: test_x86_vbroadcastmw_128:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k0
; CHECK-NEXT: vpbroadcastmw2d %k0, %xmm0
; CHECK-NEXT: retq
%res = call <4 x i32> @llvm.x86.avx512.broadcastmw.128(i16 %a0) ;
ret <4 x i32> %res
}
declare <4 x i32> @llvm.x86.avx512.broadcastmw.128(i16)
define <4 x i64> @test_x86_broadcastmb_256(i8 %a0) {
; CHECK: test_x86_broadcastmb_256
; CHECK: vpbroadcastmb2q %k0, %ymm0
%res = call <4 x i64> @llvm.x86.avx512.broadcastmb.256(i8 %a0) ;
; CHECK-LABEL: test_x86_broadcastmb_256:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k0
; CHECK-NEXT: vpbroadcastmb2q %k0, %ymm0
; CHECK-NEXT: retq
%res = call <4 x i64> @llvm.x86.avx512.broadcastmb.256(i8 %a0) ;
ret <4 x i64> %res
}
declare <4 x i64> @llvm.x86.avx512.broadcastmb.256(i8)
define <2 x i64> @test_x86_broadcastmb_128(i8 %a0) {
; CHECK: test_x86_broadcastmb_128
; CHECK: vpbroadcastmb2q %k0, %xmm0
%res = call <2 x i64> @llvm.x86.avx512.broadcastmb.128(i8 %a0) ;
; CHECK-LABEL: test_x86_broadcastmb_128:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k0
; CHECK-NEXT: vpbroadcastmb2q %k0, %xmm0
; CHECK-NEXT: retq
%res = call <2 x i64> @llvm.x86.avx512.broadcastmb.128(i8 %a0) ;
ret <2 x i64> %res
}
declare <2 x i64> @llvm.x86.avx512.broadcastmb.128(i8)

1035
test/CodeGen/X86/avx512vl-intrinsics.ll
View File

File diff suppressed because it is too large Load Diff

8
test/CodeGen/X86/masked_gather_scatter.ll
View File

@ -279,8 +279,7 @@ define <8 x i32> @test7(i32* %base, <8 x i32> %ind, i8 %mask) {
;
; KNL_64-LABEL: test7:
; KNL_64: # BB#0:
; KNL_64-NEXT: movzbl %sil, %eax
; KNL_64-NEXT: kmovw %eax, %k1
; KNL_64-NEXT: kmovw %esi, %k1
; KNL_64-NEXT: vpmovsxdq %ymm0, %zmm0
; KNL_64-NEXT: kmovw %k1, %k2
; KNL_64-NEXT: vpgatherqd (%rdi,%zmm0,4), %ymm1 {%k2}
@ -1128,7 +1127,6 @@ define <2 x i32> @test24(i32* %base, <2 x i32> %ind) {
; KNL_64-LABEL: test24:
; KNL_64: # BB#0:
; KNL_64-NEXT: movb $3, %al
; KNL_64-NEXT: movzbl %al, %eax
; KNL_64-NEXT: kmovw %eax, %k1
; KNL_64-NEXT: vpgatherqq (%rdi,%zmm0,8), %zmm1 {%k1}
; KNL_64-NEXT: vmovaps %zmm1, %zmm0
@ -1215,7 +1213,6 @@ define <2 x i64> @test26(i64* %base, <2 x i32> %ind, <2 x i64> %src0) {
; KNL_64-LABEL: test26:
; KNL_64: # BB#0:
; KNL_64-NEXT: movb $3, %al
; KNL_64-NEXT: movzbl %al, %eax
; KNL_64-NEXT: kmovw %eax, %k1
; KNL_64-NEXT: vpgatherqq (%rdi,%zmm0,8), %zmm1 {%k1}
; KNL_64-NEXT: vmovaps %zmm1, %zmm0
@ -1260,7 +1257,6 @@ define <2 x float> @test27(float* %base, <2 x i32> %ind) {
; KNL_64-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
; KNL_64-NEXT: vpmovsxdq %ymm0, %zmm1
; KNL_64-NEXT: movb $3, %al
; KNL_64-NEXT: movzbl %al, %eax
; KNL_64-NEXT: kmovw %eax, %k1
; KNL_64-NEXT: vgatherqps (%rdi,%zmm1,4), %ymm0 {%k1}
; KNL_64-NEXT: retq
@ -1271,7 +1267,6 @@ define <2 x float> @test27(float* %base, <2 x i32> %ind) {
; KNL_32-NEXT: movl {{[0-9]+}}(%esp), %eax
; KNL_32-NEXT: vpmovsxdq %ymm0, %zmm1
; KNL_32-NEXT: movb $3, %cl
; KNL_32-NEXT: movzbl %cl, %ecx
; KNL_32-NEXT: kmovw %ecx, %k1
; KNL_32-NEXT: vgatherqps (%eax,%zmm1,4), %ymm0 {%k1}
; KNL_32-NEXT: retl
@ -1297,7 +1292,6 @@ define void @test28(<2 x i32>%a1, <2 x i32*> %ptr) {
; KNL_64: # BB#0:
; KNL_64-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
; KNL_64-NEXT: movb $3, %al
; KNL_64-NEXT: movzbl %al, %eax
; KNL_64-NEXT: kmovw %eax, %k1
; KNL_64-NEXT: vpscatterqd %ymm0, (,%zmm1) {%k1}
; KNL_64-NEXT: retq

19
test/CodeGen/X86/vector-shuffle-v1.ll
View File

@ -162,8 +162,7 @@ define <32 x i1> @shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0
define <8 x i1> @shuf8i1_u_2_u_u_2_u_2_u(i8 %a) {
; AVX512F-LABEL: shuf8i1_u_2_u_u_2_u_2_u:
; AVX512F: # BB#0:
; AVX512F-NEXT: movzbl %dil, %eax
; AVX512F-NEXT: kmovw %eax, %k1
; AVX512F-NEXT: kmovw %edi, %k1
; AVX512F-NEXT: vpbroadcastq {{.*}}(%rip), %zmm0
; AVX512F-NEXT: vmovdqu64 %zmm0, %zmm1 {%k1} {z}
; AVX512F-NEXT: vpbroadcastq {{.*}}(%rip), %zmm2
@ -192,8 +191,7 @@ define <8 x i1> @shuf8i1_u_2_u_u_2_u_2_u(i8 %a) {
define i8 @shuf8i1_10_2_9_u_3_u_2_u(i8 %a) {
; AVX512F-LABEL: shuf8i1_10_2_9_u_3_u_2_u:
; AVX512F: # BB#0:
; AVX512F-NEXT: movzbl %dil, %eax
; AVX512F-NEXT: kmovw %eax, %k1
; AVX512F-NEXT: kmovw %edi, %k1
; AVX512F-NEXT: vpbroadcastq {{.*}}(%rip), %zmm0 {%k1} {z}
; AVX512F-NEXT: vpxord %zmm1, %zmm1, %zmm1
; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm2 = <8,2,10,u,3,u,2,u>
@ -223,8 +221,7 @@ define i8 @shuf8i1_10_2_9_u_3_u_2_u(i8 %a) {
define i8 @shuf8i1_0_1_4_5_u_u_u_u(i8 %a) {
; AVX512F-LABEL: shuf8i1_0_1_4_5_u_u_u_u:
; AVX512F: # BB#0:
; AVX512F-NEXT: movzbl %dil, %eax
; AVX512F-NEXT: kmovw %eax, %k1
; AVX512F-NEXT: kmovw %edi, %k1
; AVX512F-NEXT: vpbroadcastq {{.*}}(%rip), %zmm0 {%k1} {z}
; AVX512F-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,4,5,0,1,0,1]
; AVX512F-NEXT: vpsllq $63, %zmm0, %zmm0
@ -250,8 +247,7 @@ define i8 @shuf8i1_0_1_4_5_u_u_u_u(i8 %a) {
define i8 @shuf8i1_9_6_1_0_3_7_7_0(i8 %a) {
; AVX512F-LABEL: shuf8i1_9_6_1_0_3_7_7_0:
; AVX512F: # BB#0:
; AVX512F-NEXT: movzbl %dil, %eax
; AVX512F-NEXT: kmovw %eax, %k1
; AVX512F-NEXT: kmovw %edi, %k1
; AVX512F-NEXT: vpbroadcastq {{.*}}(%rip), %zmm0 {%k1} {z}
; AVX512F-NEXT: vpxord %zmm1, %zmm1, %zmm1
; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm2 = [8,6,1,0,3,7,7,0]
@ -281,8 +277,7 @@ define i8 @shuf8i1_9_6_1_0_3_7_7_0(i8 %a) {
define i8 @shuf8i1_9_6_1_10_3_7_7_0(i8 %a) {
; AVX512F-LABEL: shuf8i1_9_6_1_10_3_7_7_0:
; AVX512F: # BB#0:
; AVX512F-NEXT: movzbl %dil, %eax
; AVX512F-NEXT: kmovw %eax, %k1
; AVX512F-NEXT: kmovw %edi, %k1
; AVX512F-NEXT: vpbroadcastq {{.*}}(%rip), %zmm0 {%k1} {z}
; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm1 = [9,1,2,10,4,5,6,7]
; AVX512F-NEXT: vpxord %zmm2, %zmm2, %zmm2
@ -312,10 +307,8 @@ define i8 @shuf8i1_9_6_1_10_3_7_7_0(i8 %a) {
define i8 @shuf8i1__9_6_1_10_3_7_7_1(i8 %a) {
; AVX512F-LABEL: shuf8i1__9_6_1_10_3_7_7_1:
; AVX512F: # BB#0:
; AVX512F-NEXT: movzbl %dil, %eax
; AVX512F-NEXT: kmovw %eax, %k1
; AVX512F-NEXT: kmovw %edi, %k1
; AVX512F-NEXT: movb $51, %al
; AVX512F-NEXT: movzbl %al, %eax
; AVX512F-NEXT: kmovw %eax, %k2
; AVX512F-NEXT: vpbroadcastq {{.*}}(%rip), %zmm0
; AVX512F-NEXT: vmovdqu64 %zmm0, %zmm1 {%k2} {z}