[X86][SSE] Added non-temporal store tests for all 512-bit vector types

llvm-svn: 271393

test/CodeGen/X86/fast-isel-nontemporal.ll

@@ -2,7 +2,8 @@
 ; RUN: llc -verify-machineinstrs -mtriple=x86_64-unknown-unknown -mattr=+sse2 -fast-isel -O0 < %s | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE2
 ; RUN: llc -verify-machineinstrs -mtriple=x86_64-unknown-unknown -mattr=+sse4a -fast-isel -O0 < %s | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE4A
 ; RUN: llc -verify-machineinstrs -mtriple=x86_64-unknown-unknown -mattr=+avx -fast-isel -O0 < %s | FileCheck %s --check-prefix=ALL --check-prefix=AVX
-; RUN: llc -verify-machineinstrs -mtriple=x86_64-unknown-unknown -mcpu=knl -fast-isel -O0 < %s | FileCheck %s --check-prefix=ALL --check-prefix=KNL
+; RUN: llc -verify-machineinstrs -mtriple=x86_64-unknown-unknown -mattr=+avx512f -fast-isel -O0 < %s | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512F
+; RUN: llc -verify-machineinstrs -mtriple=x86_64-unknown-unknown -mattr=+avx512bw -fast-isel -O0 < %s | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW
 
 ;
 ; Scalar Stores
@@ -44,10 +45,10 @@ define void @test_ntfloat(float* nocapture %ptr, float %X) {
 ; AVX-NEXT:    vmovss %xmm0, (%rdi)
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_ntfloat:
+; AVX512-LABEL: test_ntfloat:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovss %xmm0, (%rdi)
+; AVX512-NEXT:    vmovss %xmm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store float %X, float* %ptr, align 4, !nontemporal !1
   ret void
@@ -69,10 +70,10 @@ define void @test_ntdouble(double* nocapture %ptr, double %X) {
 ; AVX-NEXT:    vmovsd %xmm0, (%rdi)
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_ntdouble:
+; AVX512-LABEL: test_ntdouble:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovsd %xmm0, (%rdi)
+; AVX512-NEXT:    vmovsd %xmm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store double %X, double* %ptr, align 8, !nontemporal !1
   ret void
@@ -93,10 +94,10 @@ define void @test_nt4xfloat(<4 x float>* nocapture %ptr, <4 x float> %X) {
 ; AVX-NEXT:    vmovntps %xmm0, (%rdi)
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt4xfloat:
+; AVX512-LABEL: test_nt4xfloat:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntps %xmm0, (%rdi)
+; AVX512-NEXT:    vmovntps %xmm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <4 x float> %X, <4 x float>* %ptr, align 16, !nontemporal !1
   ret void
@@ -113,10 +114,10 @@ define void @test_nt2xdouble(<2 x double>* nocapture %ptr, <2 x double> %X) {
 ; AVX-NEXT:    vmovntpd %xmm0, (%rdi)
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt2xdouble:
+; AVX512-LABEL: test_nt2xdouble:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntpd %xmm0, (%rdi)
+; AVX512-NEXT:    vmovntpd %xmm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <2 x double> %X, <2 x double>* %ptr, align 16, !nontemporal !1
   ret void
@@ -133,10 +134,10 @@ define void @test_nt16xi8(<16 x i8>* nocapture %ptr, <16 x i8> %X) {
 ; AVX-NEXT:    vmovntdq %xmm0, (%rdi)
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt16xi8:
+; AVX512-LABEL: test_nt16xi8:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntdq %xmm0, (%rdi)
+; AVX512-NEXT:    vmovntdq %xmm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <16 x i8> %X, <16 x i8>* %ptr, align 16, !nontemporal !1
   ret void
@@ -153,10 +154,10 @@ define void @test_nt8xi16(<8 x i16>* nocapture %ptr, <8 x i16> %X) {
 ; AVX-NEXT:    vmovntdq %xmm0, (%rdi)
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt8xi16:
+; AVX512-LABEL: test_nt8xi16:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntdq %xmm0, (%rdi)
+; AVX512-NEXT:    vmovntdq %xmm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <8 x i16> %X, <8 x i16>* %ptr, align 16, !nontemporal !1
   ret void
@@ -173,10 +174,10 @@ define void @test_nt4xi32(<4 x i32>* nocapture %ptr, <4 x i32> %X) {
 ; AVX-NEXT:    vmovntdq %xmm0, (%rdi)
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt4xi32:
+; AVX512-LABEL: test_nt4xi32:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntdq %xmm0, (%rdi)
+; AVX512-NEXT:    vmovntdq %xmm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <4 x i32> %X, <4 x i32>* %ptr, align 16, !nontemporal !1
   ret void
@@ -193,10 +194,10 @@ define void @test_nt2xi64(<2 x i64>* nocapture %ptr, <2 x i64> %X) {
 ; AVX-NEXT:    vmovntdq %xmm0, (%rdi)
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt2xi64:
+; AVX512-LABEL: test_nt2xi64:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntdq %xmm0, (%rdi)
+; AVX512-NEXT:    vmovntdq %xmm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <2 x i64> %X, <2 x i64>* %ptr, align 16, !nontemporal !1
   ret void
@@ -219,10 +220,10 @@ define void @test_nt8xfloat(<8 x float>* nocapture %ptr, <8 x float> %X) {
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt8xfloat:
+; AVX512-LABEL: test_nt8xfloat:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntps %ymm0, (%rdi)
+; AVX512-NEXT:    vmovntps %ymm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <8 x float> %X, <8 x float>* %ptr, align 32, !nontemporal !1
   ret void
@@ -241,10 +242,10 @@ define void @test_nt4xdouble(<4 x double>* nocapture %ptr, <4 x double> %X) {
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt4xdouble:
+; AVX512-LABEL: test_nt4xdouble:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntpd %ymm0, (%rdi)
+; AVX512-NEXT:    vmovntpd %ymm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <4 x double> %X, <4 x double>* %ptr, align 32, !nontemporal !1
   ret void
@@ -263,10 +264,10 @@ define void @test_nt32xi8(<32 x i8>* nocapture %ptr, <32 x i8> %X) {
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt32xi8:
+; AVX512-LABEL: test_nt32xi8:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntdq %ymm0, (%rdi)
+; AVX512-NEXT:    vmovntdq %ymm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <32 x i8> %X, <32 x i8>* %ptr, align 32, !nontemporal !1
   ret void
@@ -285,10 +286,10 @@ define void @test_nt16xi16(<16 x i16>* nocapture %ptr, <16 x i16> %X) {
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt16xi16:
+; AVX512-LABEL: test_nt16xi16:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntdq %ymm0, (%rdi)
+; AVX512-NEXT:    vmovntdq %ymm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <16 x i16> %X, <16 x i16>* %ptr, align 32, !nontemporal !1
   ret void
@@ -307,10 +308,10 @@ define void @test_nt8xi32(<8 x i32>* nocapture %ptr, <8 x i32> %X) {
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt8xi32:
+; AVX512-LABEL: test_nt8xi32:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntdq %ymm0, (%rdi)
+; AVX512-NEXT:    vmovntdq %ymm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <8 x i32> %X, <8 x i32>* %ptr, align 32, !nontemporal !1
   ret void
@@ -329,13 +330,179 @@ define void @test_nt4xi64(<4 x i64>* nocapture %ptr, <4 x i64> %X) {
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
-; KNL-LABEL: test_nt4xi64:
+; AVX512-LABEL: test_nt4xi64:
-; KNL:       # BB#0: # %entry
+; AVX512:       # BB#0: # %entry
-; KNL-NEXT:    vmovntdq %ymm0, (%rdi)
+; AVX512-NEXT:    vmovntdq %ymm0, (%rdi)
-; KNL-NEXT:    retq
+; AVX512-NEXT:    retq
 entry:
   store <4 x i64> %X, <4 x i64>* %ptr, align 32, !nontemporal !1
   ret void
 }
 
+;
+; 512-bit Vector Stores
+;
+
+define void @test_nt16xfloat(<16 x float>* nocapture %ptr, <16 x float> %X) {
+; SSE-LABEL: test_nt16xfloat:
+; SSE:       # BB#0: # %entry
+; SSE-NEXT:    movntps %xmm0, (%rdi)
+; SSE-NEXT:    movntps %xmm1, 16(%rdi)
+; SSE-NEXT:    movntps %xmm2, 32(%rdi)
+; SSE-NEXT:    movntps %xmm3, 48(%rdi)
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: test_nt16xfloat:
+; AVX:       # BB#0: # %entry
+; AVX-NEXT:    vmovntps %ymm0, (%rdi)
+; AVX-NEXT:    vmovntps %ymm1, 32(%rdi)
+; AVX-NEXT:    vzeroupper
+; AVX-NEXT:    retq
+;
+; AVX512-LABEL: test_nt16xfloat:
+; AVX512:       # BB#0: # %entry
+; AVX512-NEXT:    vmovntps %zmm0, (%rdi)
+; AVX512-NEXT:    retq
+entry:
+  store <16 x float> %X, <16 x float>* %ptr, align 64, !nontemporal !1
+  ret void
+}
+
+define void @test_nt8xdouble(<8 x double>* nocapture %ptr, <8 x double> %X) {
+; SSE-LABEL: test_nt8xdouble:
+; SSE:       # BB#0: # %entry
+; SSE-NEXT:    movntpd %xmm0, (%rdi)
+; SSE-NEXT:    movntpd %xmm1, 16(%rdi)
+; SSE-NEXT:    movntpd %xmm2, 32(%rdi)
+; SSE-NEXT:    movntpd %xmm3, 48(%rdi)
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: test_nt8xdouble:
+; AVX:       # BB#0: # %entry
+; AVX-NEXT:    vmovntpd %ymm0, (%rdi)
+; AVX-NEXT:    vmovntpd %ymm1, 32(%rdi)
+; AVX-NEXT:    vzeroupper
+; AVX-NEXT:    retq
+;
+; AVX512-LABEL: test_nt8xdouble:
+; AVX512:       # BB#0: # %entry
+; AVX512-NEXT:    vmovntpd %zmm0, (%rdi)
+; AVX512-NEXT:    retq
+entry:
+  store <8 x double> %X, <8 x double>* %ptr, align 64, !nontemporal !1
+  ret void
+}
+
+define void @test_nt64xi8(<64 x i8>* nocapture %ptr, <64 x i8> %X) {
+; SSE-LABEL: test_nt64xi8:
+; SSE:       # BB#0: # %entry
+; SSE-NEXT:    movntdq %xmm0, (%rdi)
+; SSE-NEXT:    movntdq %xmm1, 16(%rdi)
+; SSE-NEXT:    movntdq %xmm2, 32(%rdi)
+; SSE-NEXT:    movntdq %xmm3, 48(%rdi)
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: test_nt64xi8:
+; AVX:       # BB#0: # %entry
+; AVX-NEXT:    vmovntps %ymm0, (%rdi)
+; AVX-NEXT:    vmovntps %ymm1, 32(%rdi)
+; AVX-NEXT:    vzeroupper
+; AVX-NEXT:    retq
+;
+; AVX512F-LABEL: test_nt64xi8:
+; AVX512F:       # BB#0: # %entry
+; AVX512F-NEXT:    vmovntdq %ymm0, (%rdi)
+; AVX512F-NEXT:    vmovntdq %ymm1, 32(%rdi)
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: test_nt64xi8:
+; AVX512BW:       # BB#0: # %entry
+; AVX512BW-NEXT:    vmovntdq %zmm0, (%rdi)
+; AVX512BW-NEXT:    retq
+entry:
+  store <64 x i8> %X, <64 x i8>* %ptr, align 64, !nontemporal !1
+  ret void
+}
+
+define void @test_nt32xi16(<32 x i16>* nocapture %ptr, <32 x i16> %X) {
+; SSE-LABEL: test_nt32xi16:
+; SSE:       # BB#0: # %entry
+; SSE-NEXT:    movntdq %xmm0, (%rdi)
+; SSE-NEXT:    movntdq %xmm1, 16(%rdi)
+; SSE-NEXT:    movntdq %xmm2, 32(%rdi)
+; SSE-NEXT:    movntdq %xmm3, 48(%rdi)
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: test_nt32xi16:
+; AVX:       # BB#0: # %entry
+; AVX-NEXT:    vmovntps %ymm0, (%rdi)
+; AVX-NEXT:    vmovntps %ymm1, 32(%rdi)
+; AVX-NEXT:    vzeroupper
+; AVX-NEXT:    retq
+;
+; AVX512F-LABEL: test_nt32xi16:
+; AVX512F:       # BB#0: # %entry
+; AVX512F-NEXT:    vmovntdq %ymm0, (%rdi)
+; AVX512F-NEXT:    vmovntdq %ymm1, 32(%rdi)
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: test_nt32xi16:
+; AVX512BW:       # BB#0: # %entry
+; AVX512BW-NEXT:    vmovntdq %zmm0, (%rdi)
+; AVX512BW-NEXT:    retq
+entry:
+  store <32 x i16> %X, <32 x i16>* %ptr, align 64, !nontemporal !1
+  ret void
+}
+
+define void @test_nt16xi32(<16 x i32>* nocapture %ptr, <16 x i32> %X) {
+; SSE-LABEL: test_nt16xi32:
+; SSE:       # BB#0: # %entry
+; SSE-NEXT:    movntdq %xmm0, (%rdi)
+; SSE-NEXT:    movntdq %xmm1, 16(%rdi)
+; SSE-NEXT:    movntdq %xmm2, 32(%rdi)
+; SSE-NEXT:    movntdq %xmm3, 48(%rdi)
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: test_nt16xi32:
+; AVX:       # BB#0: # %entry
+; AVX-NEXT:    vmovntps %ymm0, (%rdi)
+; AVX-NEXT:    vmovntps %ymm1, 32(%rdi)
+; AVX-NEXT:    vzeroupper
+; AVX-NEXT:    retq
+;
+; AVX512-LABEL: test_nt16xi32:
+; AVX512:       # BB#0: # %entry
+; AVX512-NEXT:    vmovntdq %zmm0, (%rdi)
+; AVX512-NEXT:    retq
+entry:
+  store <16 x i32> %X, <16 x i32>* %ptr, align 64, !nontemporal !1
+  ret void
+}
+
+define void @test_nt8xi64(<8 x i64>* nocapture %ptr, <8 x i64> %X) {
+; SSE-LABEL: test_nt8xi64:
+; SSE:       # BB#0: # %entry
+; SSE-NEXT:    movntdq %xmm0, (%rdi)
+; SSE-NEXT:    movntdq %xmm1, 16(%rdi)
+; SSE-NEXT:    movntdq %xmm2, 32(%rdi)
+; SSE-NEXT:    movntdq %xmm3, 48(%rdi)
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: test_nt8xi64:
+; AVX:       # BB#0: # %entry
+; AVX-NEXT:    vmovntdq %ymm0, (%rdi)
+; AVX-NEXT:    vmovntdq %ymm1, 32(%rdi)
+; AVX-NEXT:    vzeroupper
+; AVX-NEXT:    retq
+;
+; AVX512-LABEL: test_nt8xi64:
+; AVX512:       # BB#0: # %entry
+; AVX512-NEXT:    vmovntdq %zmm0, (%rdi)
+; AVX512-NEXT:    retq
+entry:
+  store <8 x i64> %X, <8 x i64>* %ptr, align 64, !nontemporal !1
+  ret void
+}
+
 !1 = !{i32 1}