summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll')
-rw-r--r--llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll60
1 files changed, 30 insertions, 30 deletions
diff --git a/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll
index f9126b4614e..83ef971724d 100644
--- a/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll
+++ b/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll
@@ -1509,7 +1509,7 @@ define <4 x i32> @test_mask_and_epi32_rrk_128(<4 x i32> %a, <4 x i32> %b, <4 x i
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
; CHECK-NEXT: vpandd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xdb,0xd1]
-; CHECK-NEXT: vmovaps %xmm2, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc2]
+; CHECK-NEXT: vmovdqa64 %xmm2, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc2]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res = call <4 x i32> @llvm.x86.avx512.mask.pand.d.128(<4 x i32> %a, <4 x i32> %b, <4 x i32> %passThru, i8 %mask)
ret <4 x i32> %res
@@ -1540,7 +1540,7 @@ define <4 x i32> @test_mask_and_epi32_rmk_128(<4 x i32> %a, <4 x i32>* %ptr_b, <
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandd (%rdi), %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xdb,0x0f]
-; CHECK-NEXT: vmovaps %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %xmm1, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%b = load <4 x i32>, <4 x i32>* %ptr_b
%res = call <4 x i32> @llvm.x86.avx512.mask.pand.d.128(<4 x i32> %a, <4 x i32> %b, <4 x i32> %passThru, i8 %mask)
@@ -1575,7 +1575,7 @@ define <4 x i32> @test_mask_and_epi32_rmbk_128(<4 x i32> %a, i32* %ptr_b, <4 x i
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandd (%rdi){1to4}, %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x19,0xdb,0x0f]
-; CHECK-NEXT: vmovaps %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %xmm1, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%q = load i32, i32* %ptr_b
%vecinit.i = insertelement <4 x i32> undef, i32 %q, i32 0
@@ -1613,7 +1613,7 @@ define <8 x i32> @test_mask_and_epi32_rrk_256(<8 x i32> %a, <8 x i32> %b, <8 x i
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
; CHECK-NEXT: vpandd %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xdb,0xd1]
-; CHECK-NEXT: vmovaps %ymm2, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc2]
+; CHECK-NEXT: vmovdqa64 %ymm2, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc2]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res = call <8 x i32> @llvm.x86.avx512.mask.pand.d.256(<8 x i32> %a, <8 x i32> %b, <8 x i32> %passThru, i8 %mask)
ret <8 x i32> %res
@@ -1644,7 +1644,7 @@ define <8 x i32> @test_mask_and_epi32_rmk_256(<8 x i32> %a, <8 x i32>* %ptr_b, <
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandd (%rdi), %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xdb,0x0f]
-; CHECK-NEXT: vmovaps %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%b = load <8 x i32>, <8 x i32>* %ptr_b
%res = call <8 x i32> @llvm.x86.avx512.mask.pand.d.256(<8 x i32> %a, <8 x i32> %b, <8 x i32> %passThru, i8 %mask)
@@ -1679,7 +1679,7 @@ define <8 x i32> @test_mask_and_epi32_rmbk_256(<8 x i32> %a, i32* %ptr_b, <8 x i
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandd (%rdi){1to8}, %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x39,0xdb,0x0f]
-; CHECK-NEXT: vmovaps %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%q = load i32, i32* %ptr_b
%vecinit.i = insertelement <8 x i32> undef, i32 %q, i32 0
@@ -1717,7 +1717,7 @@ define <4 x i32> @test_mask_or_epi32_rrk_128(<4 x i32> %a, <4 x i32> %b, <4 x i3
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
; CHECK-NEXT: vpord %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xeb,0xd1]
-; CHECK-NEXT: vmovaps %xmm2, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc2]
+; CHECK-NEXT: vmovdqa64 %xmm2, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc2]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res = call <4 x i32> @llvm.x86.avx512.mask.por.d.128(<4 x i32> %a, <4 x i32> %b, <4 x i32> %passThru, i8 %mask)
ret <4 x i32> %res
@@ -1748,7 +1748,7 @@ define <4 x i32> @test_mask_or_epi32_rmk_128(<4 x i32> %a, <4 x i32>* %ptr_b, <4
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpord (%rdi), %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xeb,0x0f]
-; CHECK-NEXT: vmovaps %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %xmm1, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%b = load <4 x i32>, <4 x i32>* %ptr_b
%res = call <4 x i32> @llvm.x86.avx512.mask.por.d.128(<4 x i32> %a, <4 x i32> %b, <4 x i32> %passThru, i8 %mask)
@@ -1783,7 +1783,7 @@ define <4 x i32> @test_mask_or_epi32_rmbk_128(<4 x i32> %a, i32* %ptr_b, <4 x i3
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpord (%rdi){1to4}, %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x19,0xeb,0x0f]
-; CHECK-NEXT: vmovaps %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %xmm1, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%q = load i32, i32* %ptr_b
%vecinit.i = insertelement <4 x i32> undef, i32 %q, i32 0
@@ -1821,7 +1821,7 @@ define <8 x i32> @test_mask_or_epi32_rrk_256(<8 x i32> %a, <8 x i32> %b, <8 x i3
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
; CHECK-NEXT: vpord %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xeb,0xd1]
-; CHECK-NEXT: vmovaps %ymm2, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc2]
+; CHECK-NEXT: vmovdqa64 %ymm2, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc2]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res = call <8 x i32> @llvm.x86.avx512.mask.por.d.256(<8 x i32> %a, <8 x i32> %b, <8 x i32> %passThru, i8 %mask)
ret <8 x i32> %res
@@ -1852,7 +1852,7 @@ define <8 x i32> @test_mask_or_epi32_rmk_256(<8 x i32> %a, <8 x i32>* %ptr_b, <8
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpord (%rdi), %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xeb,0x0f]
-; CHECK-NEXT: vmovaps %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%b = load <8 x i32>, <8 x i32>* %ptr_b
%res = call <8 x i32> @llvm.x86.avx512.mask.por.d.256(<8 x i32> %a, <8 x i32> %b, <8 x i32> %passThru, i8 %mask)
@@ -1887,7 +1887,7 @@ define <8 x i32> @test_mask_or_epi32_rmbk_256(<8 x i32> %a, i32* %ptr_b, <8 x i3
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpord (%rdi){1to8}, %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x39,0xeb,0x0f]
-; CHECK-NEXT: vmovaps %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%q = load i32, i32* %ptr_b
%vecinit.i = insertelement <8 x i32> undef, i32 %q, i32 0
@@ -1925,7 +1925,7 @@ define <4 x i32> @test_mask_xor_epi32_rrk_128(<4 x i32> %a, <4 x i32> %b, <4 x i
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
; CHECK-NEXT: vpxord %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xef,0xd1]
-; CHECK-NEXT: vmovaps %xmm2, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc2]
+; CHECK-NEXT: vmovdqa64 %xmm2, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc2]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res = call <4 x i32> @llvm.x86.avx512.mask.pxor.d.128(<4 x i32> %a, <4 x i32> %b, <4 x i32> %passThru, i8 %mask)
ret <4 x i32> %res
@@ -1956,7 +1956,7 @@ define <4 x i32> @test_mask_xor_epi32_rmk_128(<4 x i32> %a, <4 x i32>* %ptr_b, <
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpxord (%rdi), %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xef,0x0f]
-; CHECK-NEXT: vmovaps %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %xmm1, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%b = load <4 x i32>, <4 x i32>* %ptr_b
%res = call <4 x i32> @llvm.x86.avx512.mask.pxor.d.128(<4 x i32> %a, <4 x i32> %b, <4 x i32> %passThru, i8 %mask)
@@ -1991,7 +1991,7 @@ define <4 x i32> @test_mask_xor_epi32_rmbk_128(<4 x i32> %a, i32* %ptr_b, <4 x i
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpxord (%rdi){1to4}, %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x19,0xef,0x0f]
-; CHECK-NEXT: vmovaps %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %xmm1, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%q = load i32, i32* %ptr_b
%vecinit.i = insertelement <4 x i32> undef, i32 %q, i32 0
@@ -2029,7 +2029,7 @@ define <8 x i32> @test_mask_xor_epi32_rrk_256(<8 x i32> %a, <8 x i32> %b, <8 x i
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
; CHECK-NEXT: vpxord %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xef,0xd1]
-; CHECK-NEXT: vmovaps %ymm2, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc2]
+; CHECK-NEXT: vmovdqa64 %ymm2, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc2]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res = call <8 x i32> @llvm.x86.avx512.mask.pxor.d.256(<8 x i32> %a, <8 x i32> %b, <8 x i32> %passThru, i8 %mask)
ret <8 x i32> %res
@@ -2060,7 +2060,7 @@ define <8 x i32> @test_mask_xor_epi32_rmk_256(<8 x i32> %a, <8 x i32>* %ptr_b, <
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpxord (%rdi), %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xef,0x0f]
-; CHECK-NEXT: vmovaps %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%b = load <8 x i32>, <8 x i32>* %ptr_b
%res = call <8 x i32> @llvm.x86.avx512.mask.pxor.d.256(<8 x i32> %a, <8 x i32> %b, <8 x i32> %passThru, i8 %mask)
@@ -2095,7 +2095,7 @@ define <8 x i32> @test_mask_xor_epi32_rmbk_256(<8 x i32> %a, i32* %ptr_b, <8 x i
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpxord (%rdi){1to8}, %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x39,0xef,0x0f]
-; CHECK-NEXT: vmovaps %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%q = load i32, i32* %ptr_b
%vecinit.i = insertelement <8 x i32> undef, i32 %q, i32 0
@@ -2133,7 +2133,7 @@ define <4 x i32> @test_mask_andnot_epi32_rrk_128(<4 x i32> %a, <4 x i32> %b, <4
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
; CHECK-NEXT: vpandnd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xdf,0xd1]
-; CHECK-NEXT: vmovaps %xmm2, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc2]
+; CHECK-NEXT: vmovdqa64 %xmm2, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc2]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res = call <4 x i32> @llvm.x86.avx512.mask.pandn.d.128(<4 x i32> %a, <4 x i32> %b, <4 x i32> %passThru, i8 %mask)
ret <4 x i32> %res
@@ -2164,7 +2164,7 @@ define <4 x i32> @test_mask_andnot_epi32_rmk_128(<4 x i32> %a, <4 x i32>* %ptr_b
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandnd (%rdi), %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xdf,0x0f]
-; CHECK-NEXT: vmovaps %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %xmm1, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%b = load <4 x i32>, <4 x i32>* %ptr_b
%res = call <4 x i32> @llvm.x86.avx512.mask.pandn.d.128(<4 x i32> %a, <4 x i32> %b, <4 x i32> %passThru, i8 %mask)
@@ -2199,7 +2199,7 @@ define <4 x i32> @test_mask_andnot_epi32_rmbk_128(<4 x i32> %a, i32* %ptr_b, <4
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandnd (%rdi){1to4}, %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x19,0xdf,0x0f]
-; CHECK-NEXT: vmovaps %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %xmm1, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%q = load i32, i32* %ptr_b
%vecinit.i = insertelement <4 x i32> undef, i32 %q, i32 0
@@ -2237,7 +2237,7 @@ define <8 x i32> @test_mask_andnot_epi32_rrk_256(<8 x i32> %a, <8 x i32> %b, <8
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
; CHECK-NEXT: vpandnd %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xdf,0xd1]
-; CHECK-NEXT: vmovaps %ymm2, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc2]
+; CHECK-NEXT: vmovdqa64 %ymm2, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc2]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res = call <8 x i32> @llvm.x86.avx512.mask.pandn.d.256(<8 x i32> %a, <8 x i32> %b, <8 x i32> %passThru, i8 %mask)
ret <8 x i32> %res
@@ -2268,7 +2268,7 @@ define <8 x i32> @test_mask_andnot_epi32_rmk_256(<8 x i32> %a, <8 x i32>* %ptr_b
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandnd (%rdi), %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xdf,0x0f]
-; CHECK-NEXT: vmovaps %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%b = load <8 x i32>, <8 x i32>* %ptr_b
%res = call <8 x i32> @llvm.x86.avx512.mask.pandn.d.256(<8 x i32> %a, <8 x i32> %b, <8 x i32> %passThru, i8 %mask)
@@ -2303,7 +2303,7 @@ define <8 x i32> @test_mask_andnot_epi32_rmbk_256(<8 x i32> %a, i32* %ptr_b, <8
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandnd (%rdi){1to8}, %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x39,0xdf,0x0f]
-; CHECK-NEXT: vmovaps %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%q = load i32, i32* %ptr_b
%vecinit.i = insertelement <8 x i32> undef, i32 %q, i32 0
@@ -2341,7 +2341,7 @@ define <2 x i64> @test_mask_andnot_epi64_rrk_128(<2 x i64> %a, <2 x i64> %b, <2
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
; CHECK-NEXT: vpandnq %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf1,0xfd,0x09,0xdf,0xd1]
-; CHECK-NEXT: vmovaps %xmm2, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc2]
+; CHECK-NEXT: vmovdqa64 %xmm2, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc2]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res = call <2 x i64> @llvm.x86.avx512.mask.pandn.q.128(<2 x i64> %a, <2 x i64> %b, <2 x i64> %passThru, i8 %mask)
ret <2 x i64> %res
@@ -2372,7 +2372,7 @@ define <2 x i64> @test_mask_andnot_epi64_rmk_128(<2 x i64> %a, <2 x i64>* %ptr_b
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandnq (%rdi), %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x09,0xdf,0x0f]
-; CHECK-NEXT: vmovaps %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %xmm1, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%b = load <2 x i64>, <2 x i64>* %ptr_b
%res = call <2 x i64> @llvm.x86.avx512.mask.pandn.q.128(<2 x i64> %a, <2 x i64> %b, <2 x i64> %passThru, i8 %mask)
@@ -2407,7 +2407,7 @@ define <2 x i64> @test_mask_andnot_epi64_rmbk_128(<2 x i64> %a, i64* %ptr_b, <2
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandnq (%rdi){1to2}, %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x19,0xdf,0x0f]
-; CHECK-NEXT: vmovaps %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %xmm1, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%q = load i64, i64* %ptr_b
%vecinit.i = insertelement <2 x i64> undef, i64 %q, i32 0
@@ -2445,7 +2445,7 @@ define <4 x i64> @test_mask_andnot_epi64_rrk_256(<4 x i64> %a, <4 x i64> %b, <4
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
; CHECK-NEXT: vpandnq %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf1,0xfd,0x29,0xdf,0xd1]
-; CHECK-NEXT: vmovaps %ymm2, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc2]
+; CHECK-NEXT: vmovdqa64 %ymm2, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc2]
; CHECK-NEXT: retq ## encoding: [0xc3]
%res = call <4 x i64> @llvm.x86.avx512.mask.pandn.q.256(<4 x i64> %a, <4 x i64> %b, <4 x i64> %passThru, i8 %mask)
ret <4 x i64> %res
@@ -2476,7 +2476,7 @@ define <4 x i64> @test_mask_andnot_epi64_rmk_256(<4 x i64> %a, <4 x i64>* %ptr_b
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandnq (%rdi), %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x29,0xdf,0x0f]
-; CHECK-NEXT: vmovaps %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%b = load <4 x i64>, <4 x i64>* %ptr_b
%res = call <4 x i64> @llvm.x86.avx512.mask.pandn.q.256(<4 x i64> %a, <4 x i64> %b, <4 x i64> %passThru, i8 %mask)
@@ -2511,7 +2511,7 @@ define <4 x i64> @test_mask_andnot_epi64_rmbk_256(<4 x i64> %a, i64* %ptr_b, <4
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %esi, %k1 ## encoding: [0xc5,0xf8,0x92,0xce]
; CHECK-NEXT: vpandnq (%rdi){1to4}, %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x39,0xdf,0x0f]
-; CHECK-NEXT: vmovaps %ymm1, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x28,0xc1]
+; CHECK-NEXT: vmovdqa64 %ymm1, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc1]
; CHECK-NEXT: retq ## encoding: [0xc3]
%q = load i64, i64* %ptr_b
%vecinit.i = insertelement <4 x i64> undef, i64 %q, i32 0
OpenPOWER on IntegriCloud