summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/X86/avx512vl-intrinsics.ll')
-rw-r--r--llvm/test/CodeGen/X86/avx512vl-intrinsics.ll32
1 files changed, 32 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
index bc51af03e4e..686a06f23fe 100644
--- a/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
@@ -3010,7 +3010,9 @@ declare <2 x double> @llvm.x86.avx512.mask.unpckh.pd.128(<2 x double>, <2 x doub
define <2 x double>@test_int_x86_avx512_mask_unpckh_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_unpckh_pd_128:
; CHECK: vunpckhpd %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT: ## xmm2 = xmm2[1],k1[1]
; CHECK-NEXT: vunpckhpd %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x15,0xc1]
+; CHECK-NEXT: ## xmm0 = xmm0[1],xmm1[1]
%res = call <2 x double> @llvm.x86.avx512.mask.unpckh.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
%res1 = call <2 x double> @llvm.x86.avx512.mask.unpckh.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1)
%res2 = fadd <2 x double> %res, %res1
@@ -3022,7 +3024,9 @@ declare <4 x double> @llvm.x86.avx512.mask.unpckh.pd.256(<4 x double>, <4 x doub
define <4 x double>@test_int_x86_avx512_mask_unpckh_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_unpckh_pd_256:
; CHECK: vunpckhpd %ymm1, %ymm0, %ymm2 {%k1}
+; CHECK-NEXT: ## ymm2 = ymm2[1],k1[1],ymm2[3],k1[3]
; CHECK-NEXT: vunpckhpd %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x15,0xc1]
+; CHECK-NEXT: ## ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
%res = call <4 x double> @llvm.x86.avx512.mask.unpckh.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
%res1 = call <4 x double> @llvm.x86.avx512.mask.unpckh.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 -1)
%res2 = fadd <4 x double> %res, %res1
@@ -3034,7 +3038,9 @@ declare <4 x float> @llvm.x86.avx512.mask.unpckh.ps.128(<4 x float>, <4 x float>
define <4 x float>@test_int_x86_avx512_mask_unpckh_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_unpckh_ps_128:
; CHECK: vunpckhps %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT: ## xmm2 = xmm2[2],k1[2],xmm2[3],k1[3]
; CHECK-NEXT: vunpckhps %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x15,0xc1]
+; CHECK-NEXT: ## xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
%res = call <4 x float> @llvm.x86.avx512.mask.unpckh.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
%res1 = call <4 x float> @llvm.x86.avx512.mask.unpckh.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1)
%res2 = fadd <4 x float> %res, %res1
@@ -3047,7 +3053,9 @@ define <8 x float>@test_int_x86_avx512_mask_unpckh_ps_256(<8 x float> %x0, <8 x
; CHECK-LABEL: test_int_x86_avx512_mask_unpckh_ps_256:
; CHECK: ## BB#0:
; CHECK: vunpckhps %ymm1, %ymm0, %ymm2 {%k1}
+; CHECK-NEXT: ## ymm2 = ymm2[2],k1[2],ymm2[3],k1[3],ymm2[6],k1[6],ymm2[7],k1[7]
; CHECK-NEXT: vunpckhps %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x15,0xc1]
+; CHECK-NEXT: ## ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
%res = call <8 x float> @llvm.x86.avx512.mask.unpckh.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
%res1 = call <8 x float> @llvm.x86.avx512.mask.unpckh.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 -1)
%res2 = fadd <8 x float> %res, %res1
@@ -3059,7 +3067,9 @@ declare <2 x double> @llvm.x86.avx512.mask.unpckl.pd.128(<2 x double>, <2 x doub
define <2 x double>@test_int_x86_avx512_mask_unpckl_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_unpckl_pd_128:
; CHECK: vunpcklpd %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT: ## xmm2 = xmm2[0],k1[0]
; CHECK-NEXT: vunpcklpd %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x14,0xc1]
+; CHECK-NEXT: ## xmm0 = xmm0[0],xmm1[0]
%res = call <2 x double> @llvm.x86.avx512.mask.unpckl.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
%res1 = call <2 x double> @llvm.x86.avx512.mask.unpckl.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1)
%res2 = fadd <2 x double> %res, %res1
@@ -3071,7 +3081,9 @@ declare <4 x double> @llvm.x86.avx512.mask.unpckl.pd.256(<4 x double>, <4 x doub
define <4 x double>@test_int_x86_avx512_mask_unpckl_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_unpckl_pd_256:
; CHECK: vunpcklpd %ymm1, %ymm0, %ymm2 {%k1}
+; CHECK-NEXT: ## ymm2 = ymm2[0],k1[0],ymm2[2],k1[2]
; CHECK-NEXT: vunpcklpd %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x14,0xc1]
+; CHECK-NEXT: ## ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
%res = call <4 x double> @llvm.x86.avx512.mask.unpckl.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
%res1 = call <4 x double> @llvm.x86.avx512.mask.unpckl.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 -1)
%res2 = fadd <4 x double> %res, %res1
@@ -3083,7 +3095,9 @@ declare <4 x float> @llvm.x86.avx512.mask.unpckl.ps.128(<4 x float>, <4 x float>
define <4 x float>@test_int_x86_avx512_mask_unpckl_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_unpckl_ps_128:
; CHECK: vunpcklps %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT: ## xmm2 = xmm2[0],k1[0],xmm2[1],k1[1]
; CHECK-NEXT: vunpcklps %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf1,0x7c,0x08,0x14,0xc1]
+; CHECK-NEXT: ## xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
%res = call <4 x float> @llvm.x86.avx512.mask.unpckl.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
%res1 = call <4 x float> @llvm.x86.avx512.mask.unpckl.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1)
%res2 = fadd <4 x float> %res, %res1
@@ -3095,7 +3109,9 @@ declare <8 x float> @llvm.x86.avx512.mask.unpckl.ps.256(<8 x float>, <8 x float>
define <8 x float>@test_int_x86_avx512_mask_unpckl_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_unpckl_ps_256:
; CHECK: vunpcklps %ymm1, %ymm0, %ymm2 {%k1}
+; CHECK-NEXT: ## ymm2 = ymm2[0],k1[0],ymm2[1],k1[1],ymm2[4],k1[4],ymm2[5],k1[5]
; CHECK-NEXT: vunpcklps %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf1,0x7c,0x28,0x14,0xc1]
+; CHECK-NEXT: ## ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
%res = call <8 x float> @llvm.x86.avx512.mask.unpckl.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
%res1 = call <8 x float> @llvm.x86.avx512.mask.unpckl.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 -1)
%res2 = fadd <8 x float> %res, %res1
@@ -3107,7 +3123,9 @@ declare <4 x i32> @llvm.x86.avx512.mask.punpckhd.q.128(<4 x i32>, <4 x i32>, <4
define <4 x i32>@test_int_x86_avx512_mask_punpckhd_q_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_punpckhd_q_128:
; CHECK: vpunpckhdq %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT: ## xmm2 = xmm2[2],k1[2],xmm2[3],k1[3]
; CHECK-NEXT: vpunpckhdq %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf1,0x7d,0x08,0x6a,0xc1]
+; CHECK-NEXT: ## xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
%res = call <4 x i32> @llvm.x86.avx512.mask.punpckhd.q.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3)
%res1 = call <4 x i32> @llvm.x86.avx512.mask.punpckhd.q.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 -1)
%res2 = add <4 x i32> %res, %res1
@@ -3119,7 +3137,9 @@ declare <4 x i32> @llvm.x86.avx512.mask.punpckld.q.128(<4 x i32>, <4 x i32>, <4
define <4 x i32>@test_int_x86_avx512_mask_punpckld_q_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_punpckld_q_128:
; CHECK: vpunpckldq %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT: ## xmm2 = xmm2[0],k1[0],xmm2[1],k1[1]
; CHECK-NEXT: vpunpckldq %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf1,0x7d,0x08,0x62,0xc1]
+; CHECK-NEXT: ## xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
%res = call <4 x i32> @llvm.x86.avx512.mask.punpckld.q.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3)
%res1 = call <4 x i32> @llvm.x86.avx512.mask.punpckld.q.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 -1)
%res2 = add <4 x i32> %res, %res1
@@ -3132,7 +3152,9 @@ define <8 x i32>@test_int_x86_avx512_mask_punpckhd_q_256(<8 x i32> %x0, <8 x i32
; CHECK-LABEL: test_int_x86_avx512_mask_punpckhd_q_256:
; CHECK: ## BB#0:
; CHECK: vpunpckhdq %ymm1, %ymm0, %ymm2 {%k1}
+; CHECK-NEXT: ## ymm2 = ymm2[2],k1[2],ymm2[3],k1[3],ymm2[6],k1[6],ymm2[7],k1[7]
; CHECK-NEXT: vpunpckhdq %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf1,0x7d,0x28,0x6a,0xc1]
+; CHECK-NEXT: ## ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
%res = call <8 x i32> @llvm.x86.avx512.mask.punpckhd.q.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3)
%res1 = call <8 x i32> @llvm.x86.avx512.mask.punpckhd.q.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 -1)
%res2 = add <8 x i32> %res, %res1
@@ -3144,7 +3166,9 @@ declare <8 x i32> @llvm.x86.avx512.mask.punpckld.q.256(<8 x i32>, <8 x i32>, <8
define <8 x i32>@test_int_x86_avx512_mask_punpckld_q_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_punpckld_q_256:
; CHECK: vpunpckldq %ymm1, %ymm0, %ymm2 {%k1}
+; CHECK-NEXT: ## ymm2 = ymm2[0],k1[0],ymm2[1],k1[1],ymm2[4],k1[4],ymm2[5],k1[5]
; CHECK-NEXT: vpunpckldq %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf1,0x7d,0x28,0x62,0xc1]
+; CHECK-NEXT: ## ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
%res = call <8 x i32> @llvm.x86.avx512.mask.punpckld.q.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3)
%res1 = call <8 x i32> @llvm.x86.avx512.mask.punpckld.q.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 -1)
%res2 = add <8 x i32> %res, %res1
@@ -3156,7 +3180,9 @@ declare <2 x i64> @llvm.x86.avx512.mask.punpckhqd.q.128(<2 x i64>, <2 x i64>, <2
define <2 x i64>@test_int_x86_avx512_mask_punpckhqd_q_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_punpckhqd_q_128:
; CHECK: vpunpckhqdq %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT: ## xmm2 = xmm2[1],k1[1]
; CHECK-NEXT: vpunpckhqdq %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6d,0xc1]
+; CHECK-NEXT: ## xmm0 = xmm0[1],xmm1[1]
%res = call <2 x i64> @llvm.x86.avx512.mask.punpckhqd.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3)
%res1 = call <2 x i64> @llvm.x86.avx512.mask.punpckhqd.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 -1)
%res2 = add <2 x i64> %res, %res1
@@ -3168,7 +3194,9 @@ declare <2 x i64> @llvm.x86.avx512.mask.punpcklqd.q.128(<2 x i64>, <2 x i64>, <2
define <2 x i64>@test_int_x86_avx512_mask_punpcklqd_q_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_punpcklqd_q_128:
; CHECK: vpunpcklqdq %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT: ## xmm2 = xmm2[0],k1[0]
; CHECK-NEXT: vpunpcklqdq %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6c,0xc1]
+; CHECK-NEXT: ## xmm0 = xmm0[0],xmm1[0]
%res = call <2 x i64> @llvm.x86.avx512.mask.punpcklqd.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3)
%res1 = call <2 x i64> @llvm.x86.avx512.mask.punpcklqd.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 -1)
%res2 = add <2 x i64> %res, %res1
@@ -3180,7 +3208,9 @@ declare <4 x i64> @llvm.x86.avx512.mask.punpcklqd.q.256(<4 x i64>, <4 x i64>, <4
define <4 x i64>@test_int_x86_avx512_mask_punpcklqd_q_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_punpcklqd_q_256:
; CHECK: vpunpcklqdq %ymm1, %ymm0, %ymm2 {%k1}
+; CHECK-NEXT: ## ymm2 = ymm2[0],k1[0],ymm2[2],k1[2]
; CHECK-NEXT: vpunpcklqdq %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6c,0xc1]
+; CHECK-NEXT: ## ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
%res = call <4 x i64> @llvm.x86.avx512.mask.punpcklqd.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %x3)
%res1 = call <4 x i64> @llvm.x86.avx512.mask.punpcklqd.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 -1)
%res2 = add <4 x i64> %res, %res1
@@ -3192,7 +3222,9 @@ declare <4 x i64> @llvm.x86.avx512.mask.punpckhqd.q.256(<4 x i64>, <4 x i64>, <4
define <4 x i64>@test_int_x86_avx512_mask_punpckhqd_q_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_punpckhqd_q_256:
; CHECK: vpunpckhqdq %ymm1, %ymm0, %ymm2 {%k1}
+; CHECK-NEXT: ## ymm2 = ymm2[1],k1[1],ymm2[3],k1[3]
; CHECK-NEXT: vpunpckhqdq %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6d,0xc1]
+; CHECK-NEXT: ## ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
%res = call <4 x i64> @llvm.x86.avx512.mask.punpckhqd.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %x3)
%res1 = call <4 x i64> @llvm.x86.avx512.mask.punpckhqd.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 -1)
%res2 = add <4 x i64> %res, %res1
OpenPOWER on IntegriCloud