[X86][AVX512BW] Added BROADCAST intrinsics fast-isel generic IR tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@274545 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll b/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll
index c3b9d30cda1b86479f6ba03a2002a6a550be65a2..50a9076163e8073058963ea3b2ced9fa34c76813 100644 (file)
--- a/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll
+++ b/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll
@@ -4,6 +4,126 @@
 
 ; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/avx512bw-builtins.c
 
+define <8 x i64> @test_mm512_broadcastb_epi8(<2 x i64> %a0) {
+; X32-LABEL: test_mm512_broadcastb_epi8:
+; X32:       # BB#0:
+; X32-NEXT:    vpbroadcastb %xmm0, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_broadcastb_epi8:
+; X64:       # BB#0:
+; X64-NEXT:    vpbroadcastb %xmm0, %zmm0
+; X64-NEXT:    retq
+  %arg0 = bitcast <2 x i64> %a0 to <16 x i8>
+  %res0 = shufflevector <16 x i8> %arg0, <16 x i8> undef, <64 x i32> zeroinitializer
+  %res1 = bitcast <64 x i8> %res0 to <8 x i64>
+  ret <8 x i64> %res1
+}
+
+define <8 x i64> @test_mm512_mask_broadcastb_epi8(<8 x i64> %a0, i64* %a1, <2 x i64> %a2) {
+; X32-LABEL: test_mm512_mask_broadcastb_epi8:
+; X32:       # BB#0:
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    kmovq (%eax), %k1
+; X32-NEXT:    vpbroadcastb %xmm1, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_broadcastb_epi8:
+; X64:       # BB#0:
+; X64-NEXT:    kmovq (%rdi), %k1
+; X64-NEXT:    vpbroadcastb %xmm1, %zmm0 {%k1}
+; X64-NEXT:    retq
+  %arg0 = bitcast <8 x i64> %a0 to <64 x i8>
+  %bc1 = bitcast i64* %a1 to <64 x i1>*
+  %arg1 = load <64 x i1>, <64 x i1>* %bc1
+  %arg2 = bitcast <2 x i64> %a2 to <16 x i8>
+  %res0 = shufflevector <16 x i8> %arg2, <16 x i8> undef, <64 x i32> zeroinitializer
+  %res1 = select <64 x i1> %arg1, <64 x i8> %res0, <64 x i8> %arg0
+  %res2 = bitcast <64 x i8> %res1 to <8 x i64>
+  ret <8 x i64> %res2
+}
+
+define <8 x i64> @test_mm512_maskz_broadcastb_epi8(i64* %a0, <2 x i64> %a1) {
+; X32-LABEL: test_mm512_maskz_broadcastb_epi8:
+; X32:       # BB#0:
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    kmovq (%eax), %k1
+; X32-NEXT:    vpbroadcastb %xmm0, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_broadcastb_epi8:
+; X64:       # BB#0:
+; X64-NEXT:    kmovq (%rdi), %k1
+; X64-NEXT:    vpbroadcastb %xmm0, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+  %bc0 = bitcast i64* %a0 to <64 x i1>*
+  %arg0 = load <64 x i1>, <64 x i1>* %bc0
+  %arg1 = bitcast <2 x i64> %a1 to <16 x i8>
+  %res0 = shufflevector <16 x i8> %arg1, <16 x i8> undef, <64 x i32> zeroinitializer
+  %res1 = select <64 x i1> %arg0, <64 x i8> %res0, <64 x i8> zeroinitializer
+  %res2 = bitcast <64 x i8> %res1 to <8 x i64>
+  ret <8 x i64> %res2
+}
+
+define <8 x i64> @test_mm512_broadcastw_epi16(<2 x i64> %a0) {
+; X32-LABEL: test_mm512_broadcastw_epi16:
+; X32:       # BB#0:
+; X32-NEXT:    vpbroadcastw %xmm0, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_broadcastw_epi16:
+; X64:       # BB#0:
+; X64-NEXT:    vpbroadcastw %xmm0, %zmm0
+; X64-NEXT:    retq
+  %arg0 = bitcast <2 x i64> %a0 to <8 x i16>
+  %res0 = shufflevector <8 x i16> %arg0, <8 x i16> undef, <32 x i32> zeroinitializer
+  %res1 = bitcast <32 x i16> %res0 to <8 x i64>
+  ret <8 x i64> %res1
+}
+
+define <8 x i64> @test_mm512_mask_broadcastw_epi16(<8 x i64> %a0, i32 %a1, <2 x i64> %a2) {
+; X32-LABEL: test_mm512_mask_broadcastw_epi16:
+; X32:       # BB#0:
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    kmovd %eax, %k1
+; X32-NEXT:    vpbroadcastw %xmm1, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_broadcastw_epi16:
+; X64:       # BB#0:
+; X64-NEXT:    kmovd %edi, %k1
+; X64-NEXT:    vpbroadcastw %xmm1, %zmm0 {%k1}
+; X64-NEXT:    retq
+  %arg0 = bitcast <8 x i64> %a0 to <32 x i16>
+  %arg1 = bitcast i32 %a1 to <32 x i1>
+  %arg2 = bitcast <2 x i64> %a2 to <8 x i16>
+  %res0 = shufflevector <8 x i16> %arg2, <8 x i16> undef, <32 x i32> zeroinitializer
+  %res1 = select <32 x i1> %arg1, <32 x i16> %res0, <32 x i16> %arg0
+  %res2 = bitcast <32 x i16> %res1 to <8 x i64>
+  ret <8 x i64> %res2
+}
+
+define <8 x i64> @test_mm512_maskz_broadcastw_epi16(i32 %a0, <2 x i64> %a1) {
+; X32-LABEL: test_mm512_maskz_broadcastw_epi16:
+; X32:       # BB#0:
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    kmovd %eax, %k1
+; X32-NEXT:    vpbroadcastw %xmm0, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_broadcastw_epi16:
+; X64:       # BB#0:
+; X64-NEXT:    kmovd %edi, %k1
+; X64-NEXT:    vpbroadcastw %xmm0, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+  %arg0 = bitcast i32 %a0 to <32 x i1>
+  %arg1 = bitcast <2 x i64> %a1 to <8 x i16>
+  %res0 = shufflevector <8 x i16> %arg1, <8 x i16> undef, <32 x i32> zeroinitializer
+  %res1 = select <32 x i1> %arg0, <32 x i16> %res0, <32 x i16> zeroinitializer
+  %res2 = bitcast <32 x i16> %res1 to <8 x i64>
+  ret <8 x i64> %res2
+}
+
 define <8 x i64> @test_mm512_bslli_epi128(<8 x i64> %a0) {
 ; X32-LABEL: test_mm512_bslli_epi128:
 ; X32:       # BB#0:

diff --git a/test/CodeGen/X86/avx512bwvl-intrinsics-fast-isel.ll b/test/CodeGen/X86/avx512bwvl-intrinsics-fast-isel.ll
new file mode 100644 (file)
index 0000000..7cd0da9
--- /dev/null
+++ b/test/CodeGen/X86/avx512bwvl-intrinsics-fast-isel.ll
@@ -0,0 +1,244 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -fast-isel -mtriple=i386-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=X32
+; RUN: llc < %s -fast-isel -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=X64
+
+; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/avx512vlbw-builtins.c
+
+define <2 x i64> @test_mm_broadcastb_epi8(<2 x i64> %a0) {
+; X32-LABEL: test_mm_broadcastb_epi8:
+; X32:       # BB#0:
+; X32-NEXT:    vpbroadcastb %xmm0, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_broadcastb_epi8:
+; X64:       # BB#0:
+; X64-NEXT:    vpbroadcastb %xmm0, %xmm0
+; X64-NEXT:    retq
+  %arg0 = bitcast <2 x i64> %a0 to <16 x i8>
+  %res0 = shufflevector <16 x i8> %arg0, <16 x i8> undef, <16 x i32> zeroinitializer
+  %res1 = bitcast <16 x i8> %res0 to <2 x i64>
+  ret <2 x i64> %res1
+}
+
+define <2 x i64> @test_mm_mask_broadcastb_epi8(<2 x i64> %a0, i16 %a1, <2 x i64> %a2) {
+; X32-LABEL: test_mm_mask_broadcastb_epi8:
+; X32:       # BB#0:
+; X32-NEXT:    movw {{[0-9]+}}(%esp), %ax
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vpbroadcastb %xmm1, %xmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_broadcastb_epi8:
+; X64:       # BB#0:
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vpbroadcastb %xmm1, %xmm0 {%k1}
+; X64-NEXT:    retq
+  %arg0 = bitcast <2 x i64> %a0 to <16 x i8>
+  %arg1 = bitcast i16 %a1 to <16 x i1>
+  %arg2 = bitcast <2 x i64> %a2 to <16 x i8>
+  %res0 = shufflevector <16 x i8> %arg2, <16 x i8> undef, <16 x i32> zeroinitializer
+  %res1 = select <16 x i1> %arg1, <16 x i8> %res0, <16 x i8> %arg0
+  %res2 = bitcast <16 x i8> %res1 to <2 x i64>
+  ret <2 x i64> %res2
+}
+
+define <2 x i64> @test_mm_maskz_broadcastb_epi8(i16 %a0, <2 x i64> %a1) {
+; X32-LABEL: test_mm_maskz_broadcastb_epi8:
+; X32:       # BB#0:
+; X32-NEXT:    movw {{[0-9]+}}(%esp), %ax
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vpbroadcastb %xmm0, %xmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_broadcastb_epi8:
+; X64:       # BB#0:
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vpbroadcastb %xmm0, %xmm0 {%k1} {z}
+; X64-NEXT:    retq
+  %arg0 = bitcast i16 %a0 to <16 x i1>
+  %arg1 = bitcast <2 x i64> %a1 to <16 x i8>
+  %res0 = shufflevector <16 x i8> %arg1, <16 x i8> undef, <16 x i32> zeroinitializer
+  %res1 = select <16 x i1> %arg0, <16 x i8> %res0, <16 x i8> zeroinitializer
+  %res2 = bitcast <16 x i8> %res1 to <2 x i64>
+  ret <2 x i64> %res2
+}
+
+define <4 x i64> @test_mm256_broadcastb_epi8(<2 x i64> %a0) {
+; X32-LABEL: test_mm256_broadcastb_epi8:
+; X32:       # BB#0:
+; X32-NEXT:    vpbroadcastb %xmm0, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_broadcastb_epi8:
+; X64:       # BB#0:
+; X64-NEXT:    vpbroadcastb %xmm0, %ymm0
+; X64-NEXT:    retq
+  %arg0 = bitcast <2 x i64> %a0 to <16 x i8>
+  %res0 = shufflevector <16 x i8> %arg0, <16 x i8> undef, <32 x i32> zeroinitializer
+  %res1 = bitcast <32 x i8> %res0 to <4 x i64>
+  ret <4 x i64> %res1
+}
+
+define <4 x i64> @test_mm256_mask_broadcastb_epi8(<4 x i64> %a0, i32 %a1, <2 x i64> %a2) {
+; X32-LABEL: test_mm256_mask_broadcastb_epi8:
+; X32:       # BB#0:
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    kmovd %eax, %k1
+; X32-NEXT:    vpbroadcastb %xmm1, %ymm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_broadcastb_epi8:
+; X64:       # BB#0:
+; X64-NEXT:    kmovd %edi, %k1
+; X64-NEXT:    vpbroadcastb %xmm1, %ymm0 {%k1}
+; X64-NEXT:    retq
+  %arg0 = bitcast <4 x i64> %a0 to <32 x i8>
+  %arg1 = bitcast i32 %a1 to <32 x i1>
+  %arg2 = bitcast <2 x i64> %a2 to <16 x i8>
+  %res0 = shufflevector <16 x i8> %arg2, <16 x i8> undef, <32 x i32> zeroinitializer
+  %res1 = select <32 x i1> %arg1, <32 x i8> %res0, <32 x i8> %arg0
+  %res2 = bitcast <32 x i8> %res1 to <4 x i64>
+  ret <4 x i64> %res2
+}
+
+define <4 x i64> @test_mm256_maskz_broadcastb_epi8(i32 %a0, <2 x i64> %a1) {
+; X32-LABEL: test_mm256_maskz_broadcastb_epi8:
+; X32:       # BB#0:
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    kmovd %eax, %k1
+; X32-NEXT:    vpbroadcastb %xmm0, %ymm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_broadcastb_epi8:
+; X64:       # BB#0:
+; X64-NEXT:    kmovd %edi, %k1
+; X64-NEXT:    vpbroadcastb %xmm0, %ymm0 {%k1} {z}
+; X64-NEXT:    retq
+  %arg0 = bitcast i32 %a0 to <32 x i1>
+  %arg1 = bitcast <2 x i64> %a1 to <16 x i8>
+  %res0 = shufflevector <16 x i8> %arg1, <16 x i8> undef, <32 x i32> zeroinitializer
+  %res1 = select <32 x i1> %arg0, <32 x i8> %res0, <32 x i8> zeroinitializer
+  %res2 = bitcast <32 x i8> %res1 to <4 x i64>
+  ret <4 x i64> %res2
+}
+
+define <2 x i64> @test_mm_broadcastw_epi16(<2 x i64> %a0) {
+; X32-LABEL: test_mm_broadcastw_epi16:
+; X32:       # BB#0:
+; X32-NEXT:    vpbroadcastw %xmm0, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_broadcastw_epi16:
+; X64:       # BB#0:
+; X64-NEXT:    vpbroadcastw %xmm0, %xmm0
+; X64-NEXT:    retq
+  %arg0 = bitcast <2 x i64> %a0 to <8 x i16>
+  %res0 = shufflevector <8 x i16> %arg0, <8 x i16> undef, <8 x i32> zeroinitializer
+  %res1 = bitcast <8 x i16> %res0 to <2 x i64>
+  ret <2 x i64> %res1
+}
+
+define <2 x i64> @test_mm_mask_broadcastw_epi16(<2 x i64> %a0, i8 %a1, <2 x i64> %a2) {
+; X32-LABEL: test_mm_mask_broadcastw_epi16:
+; X32:       # BB#0:
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vpbroadcastw %xmm1, %xmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_broadcastw_epi16:
+; X64:       # BB#0:
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vpbroadcastw %xmm1, %xmm0 {%k1}
+; X64-NEXT:    retq
+  %arg0 = bitcast <2 x i64> %a0 to <8 x i16>
+  %arg1 = bitcast i8 %a1 to <8 x i1>
+  %arg2 = bitcast <2 x i64> %a2 to <8 x i16>
+  %res0 = shufflevector <8 x i16> %arg2, <8 x i16> undef, <8 x i32> zeroinitializer
+  %res1 = select <8 x i1> %arg1, <8 x i16> %res0, <8 x i16> %arg0
+  %res2 = bitcast <8 x i16> %res1 to <2 x i64>
+  ret <2 x i64> %res2
+}
+
+define <2 x i64> @test_mm_maskz_broadcastw_epi16(i8 %a0, <2 x i64> %a1) {
+; X32-LABEL: test_mm_maskz_broadcastw_epi16:
+; X32:       # BB#0:
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vpbroadcastw %xmm0, %xmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_broadcastw_epi16:
+; X64:       # BB#0:
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vpbroadcastw %xmm0, %xmm0 {%k1} {z}
+; X64-NEXT:    retq
+  %arg0 = bitcast i8 %a0 to <8 x i1>
+  %arg1 = bitcast <2 x i64> %a1 to <8 x i16>
+  %res0 = shufflevector <8 x i16> %arg1, <8 x i16> undef, <8 x i32> zeroinitializer
+  %res1 = select <8 x i1> %arg0, <8 x i16> %res0, <8 x i16> zeroinitializer
+  %res2 = bitcast <8 x i16> %res1 to <2 x i64>
+  ret <2 x i64> %res2
+}
+
+define <4 x i64> @test_mm256_broadcastw_epi16(<2 x i64> %a0) {
+; X32-LABEL: test_mm256_broadcastw_epi16:
+; X32:       # BB#0:
+; X32-NEXT:    vpbroadcastw %xmm0, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_broadcastw_epi16:
+; X64:       # BB#0:
+; X64-NEXT:    vpbroadcastw %xmm0, %ymm0
+; X64-NEXT:    retq
+  %arg0 = bitcast <2 x i64> %a0 to <8 x i16>
+  %res0 = shufflevector <8 x i16> %arg0, <8 x i16> undef, <16 x i32> zeroinitializer
+  %res1 = bitcast <16 x i16> %res0 to <4 x i64>
+  ret <4 x i64> %res1
+}
+
+define <4 x i64> @test_mm256_mask_broadcastw_epi16(<4 x i64> %a0, i16 %a1, <2 x i64> %a2) {
+; X32-LABEL: test_mm256_mask_broadcastw_epi16:
+; X32:       # BB#0:
+; X32-NEXT:    movw {{[0-9]+}}(%esp), %ax
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vpbroadcastw %xmm1, %ymm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_broadcastw_epi16:
+; X64:       # BB#0:
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vpbroadcastw %xmm1, %ymm0 {%k1}
+; X64-NEXT:    retq
+  %arg0 = bitcast <4 x i64> %a0 to <16 x i16>
+  %arg1 = bitcast i16 %a1 to <16 x i1>
+  %arg2 = bitcast <2 x i64> %a2 to <8 x i16>
+  %res0 = shufflevector <8 x i16> %arg2, <8 x i16> undef, <16 x i32> zeroinitializer
+  %res1 = select <16 x i1> %arg1, <16 x i16> %res0, <16 x i16> %arg0
+  %res2 = bitcast <16 x i16> %res1 to <4 x i64>
+  ret <4 x i64> %res2
+}
+
+define <4 x i64> @test_mm256_maskz_broadcastw_epi16(i16 %a0, <2 x i64> %a1) {
+; X32-LABEL: test_mm256_maskz_broadcastw_epi16:
+; X32:       # BB#0:
+; X32-NEXT:    movw {{[0-9]+}}(%esp), %ax
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vpbroadcastw %xmm0, %ymm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_broadcastw_epi16:
+; X64:       # BB#0:
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vpbroadcastw %xmm0, %ymm0 {%k1} {z}
+; X64-NEXT:    retq
+  %arg0 = bitcast i16 %a0 to <16 x i1>
+  %arg1 = bitcast <2 x i64> %a1 to <8 x i16>
+  %res0 = shufflevector <8 x i16> %arg1, <8 x i16> undef, <16 x i32> zeroinitializer
+  %res1 = select <16 x i1> %arg0, <16 x i16> %res0, <16 x i16> zeroinitializer
+  %res2 = bitcast <16 x i16> %res1 to <4 x i64>
+  ret <4 x i64> %res2
+}
+
+!0 = !{i32 1}
+