5 files changed, 195 insertions, 1 deletions
diff --git a/clang/test/CodeGen/attr-target-x86.c b/clang/test/CodeGen/attr-target-x86.c
index 5bd82f8bb7c..147f405e1df 100644
--- a/clang/test/CodeGen/attr-target-x86.c
+++ b/clang/test/CodeGen/attr-target-x86.c
@@ -38,7 +38,7 @@ int __attribute__((target("arch=lakemont,mmx"))) lake(int a) { return 4; }
 // CHECK: lake{{.*}} #7
 // CHECK: #0 = {{.*}}"target-cpu"="i686" "target-features"="+x87"
 // CHECK: #1 = {{.*}}"target-cpu"="ivybridge" "target-features"="+aes,+avx,+cx16,+f16c,+fsgsbase,+fxsr,+mmx,+pclmul,+popcnt,+rdrnd,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave,+xsaveopt"
-// CHECK: #2 = {{.*}}"target-cpu"="i686" "target-features"="+x87,-aes,-avx,-avx2,-avx512bw,-avx512cd,-avx512dq,-avx512er,-avx512f,-avx512ifma,-avx512pf,-avx512vbmi,-avx512vl,-avx512vpopcntdq,-f16c,-fma,-fma4,-pclmul,-sha,-sse2,-sse3,-sse4.1,-sse4.2,-sse4a,-ssse3,-vaes,-xop,-xsave,-xsaveopt"
+// CHECK: #2 = {{.*}}"target-cpu"="i686" "target-features"="+x87,-aes,-avx,-avx2,-avx512bw,-avx512cd,-avx512dq,-avx512er,-avx512f,-avx512ifma,-avx512pf,-avx512vbmi,-avx512vl,-avx512vpopcntdq,-f16c,-fma,-fma4,-gfni,-pclmul,-sha,-sse2,-sse3,-sse4.1,-sse4.2,-sse4a,-ssse3,-vaes,-xop,-xsave,-xsaveopt"
 // CHECK: #3 = {{.*}}"target-cpu"="i686" "target-features"="+mmx,+popcnt,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87"
 // CHECK: #4 = {{.*}}"target-cpu"="i686" "target-features"="+x87,-avx,-avx2,-avx512bw,-avx512cd,-avx512dq,-avx512er,-avx512f,-avx512ifma,-avx512pf,-avx512vbmi,-avx512vl,-avx512vpopcntdq,-f16c,-fma,-fma4,-sse4.1,-sse4.2,-vaes,-xop,-xsave,-xsaveopt"
 // CHECK: #5 = {{.*}}"target-cpu"="ivybridge" "target-features"="+avx,+cx16,+f16c,+fsgsbase,+fxsr,+mmx,+pclmul,+popcnt,+rdrnd,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave,+xsaveopt,-aes,-vaes"
diff --git a/clang/test/CodeGen/gfni-builtins.c b/clang/test/CodeGen/gfni-builtins.c
new file mode 100644
index 00000000000..95cfd4fcd5c
--- /dev/null
+++ b/clang/test/CodeGen/gfni-builtins.c
@@ -0,0 +1,182 @@
+// RUN: %clang_cc1 -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +gfni -emit-llvm -o - | FileCheck %s --check-prefix SSE
+// RUN: %clang_cc1 -ffreestanding %s -triple=x86_64-apple-darwin -DAVX -target-feature +gfni -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes SSE,AVX
+// RUN: %clang_cc1 -ffreestanding %s -triple=x86_64-apple-darwin -DAVX512 -target-feature +gfni -target-feature +avx512bw -target-feature +avx512vl -emit-llvm -o - | FileCheck %s --check-prefixes SSE,AVX,AVX512
+
+#include <immintrin.h>
+
+__m128i test_mm_gf2p8affineinv_epi64_epi8(__m128i A, __m128i B) {
+  // SSE-LABEL: @test_mm_gf2p8affineinv_epi64_epi8
+  // SSE: @llvm.x86.vgf2p8affineinvqb.128
+  return _mm_gf2p8affineinv_epi64_epi8(A, B, 1);
+}
+
+__m128i test_mm_gf2p8affine_epi64_epi8(__m128i A, __m128i B) {
+  // SSE-LABEL: @test_mm_gf2p8affine_epi64_epi8
+  // SSE: @llvm.x86.vgf2p8affineqb.128
+  return _mm_gf2p8affine_epi64_epi8(A, B, 1);
+}
+
+__m128i test_mm_gf2p8mul_epi8(__m128i A, __m128i B) {
+  // SSE-LABEL: @test_mm_gf2p8mul_epi8
+  // SSE: @llvm.x86.vgf2p8mulb.128
+  return _mm_gf2p8mul_epi8(A, B);
+}
+
+#if defined(AVX) || defined(AVX512)
+__m256i test_mm256_gf2p8affineinv_epi64_epi8(__m256i A, __m256i B) {
+  // AVX-LABEL: @test_mm256_gf2p8affineinv_epi64_epi8
+  // AVX: @llvm.x86.vgf2p8affineinvqb.256
+  return _mm256_gf2p8affineinv_epi64_epi8(A, B, 1);
+}
+
+__m256i test_mm256_gf2p8affine_epi64_epi8(__m256i A, __m256i B) {
+  // AVX-LABEL: @test_mm256_gf2p8affine_epi64_epi8
+  // AVX: @llvm.x86.vgf2p8affineqb.256
+  return _mm256_gf2p8affine_epi64_epi8(A, B, 1);
+}
+
+__m256i test_mm256_gf2p8mul_epi8(__m256i A, __m256i B) {
+  // AVX-LABEL: @test_mm256_gf2p8mul_epi8
+  // AVX: @llvm.x86.vgf2p8mulb.256
+  return _mm256_gf2p8mul_epi8(A, B);
+}
+#endif // AVX
+
+#ifdef AVX512
+__m512i test_mm512_gf2p8affineinv_epi64_epi8(__m512i A, __m512i B) {
+  // AVX512-LABEL: @test_mm512_gf2p8affineinv_epi64_epi8
+  // AVX512: @llvm.x86.vgf2p8affineinvqb.512
+  return _mm512_gf2p8affineinv_epi64_epi8(A, B, 1);
+}
+
+__m512i test_mm512_mask_gf2p8affineinv_epi64_epi8(__m512i S, __mmask64 U, __m512i A, __m512i B) {
+  // AVX512-LABEL: @test_mm512_mask_gf2p8affineinv_epi64_epi8
+  // AVX512: @llvm.x86.vgf2p8affineinvqb.512
+  // AVX512: select <64 x i1> %{{[0-9]+}}, <64 x i8> %{{[0-9]+}}, <64 x i8> {{.*}}
+  return _mm512_mask_gf2p8affineinv_epi64_epi8(S, U, A, B, 1);
+}
+
+__m512i test_mm512_maskz_gf2p8affineinv_epi64_epi8(__mmask64 U, __m512i A, __m512i B) {
+  // AVX512-LABEL: @test_mm512_maskz_gf2p8affineinv_epi64_epi8
+  // AVX512: @llvm.x86.vgf2p8affineinvqb.512
+  // AVX512: select <64 x i1> %{{[0-9]+}}, <64 x i8> %{{[0-9]+}}, <64 x i8> {{.*}}
+  return _mm512_maskz_gf2p8affineinv_epi64_epi8(U, A, B, 1);
+}
+
+__m256i test_mm256_mask_gf2p8affineinv_epi64_epi8(__m256i S, __mmask32 U, __m256i A, __m256i B) {
+  // AVX256-LABEL: @test_mm256_mask_gf2p8affineinv_epi64_epi8
+  // AVX256: @llvm.x86.vgf2p8affineinvqb.256
+  // AVX256: select <32 x i1> %{{[0-9]+}}, <32 x i8> %{{[0-9]+}}, <32 x i8> {{.*}}
+  return _mm256_mask_gf2p8affineinv_epi64_epi8(S, U, A, B, 1);
+}
+
+__m256i test_mm256_maskz_gf2p8affineinv_epi64_epi8(__mmask32 U, __m256i A, __m256i B) {
+  // AVX256-LABEL: @test_mm256_maskz_gf2p8affineinv_epi64_epi8
+  // AVX256: @llvm.x86.vgf2p8affineinvqb.256
+  // AVX256: select <32 x i1> %{{[0-9]+}}, <32 x i8> %{{[0-9]+}}, <32 x i8> {{.*}}
+  return _mm256_maskz_gf2p8affineinv_epi64_epi8(U, A, B, 1);
+}
+
+__m128i test_mm_mask_gf2p8affineinv_epi64_epi8(__m128i S, __mmask16 U, __m128i A, __m128i B) {
+  // AVX512-LABEL: @test_mm_mask_gf2p8affineinv_epi64_epi8
+  // AVX512: @llvm.x86.vgf2p8affineinvqb.128
+  // AVX512: select <16 x i1> %{{[0-9]+}}, <16 x i8> %{{[0-9]+}}, <16 x i8> {{.*}}
+  return _mm_mask_gf2p8affineinv_epi64_epi8(S, U, A, B, 1);
+}
+
+__m128i test_mm_maskz_gf2p8affineinv_epi64_epi8(__mmask16 U, __m128i A, __m128i B) {
+  // AVX512-LABEL: @test_mm_maskz_gf2p8affineinv_epi64_epi8
+  // AVX512: @llvm.x86.vgf2p8affineinvqb.128
+  // AVX512: select <16 x i1> %{{[0-9]+}}, <16 x i8> %{{[0-9]+}}, <16 x i8> {{.*}}
+  return _mm_maskz_gf2p8affineinv_epi64_epi8(U, A, B, 1);
+}
+
+__m512i test_mm512_gf2p8affine_epi64_epi8(__m512i A, __m512i B) {
+  // AVX512-LABEL: @test_mm512_gf2p8affine_epi64_epi8
+  // AVX512: @llvm.x86.vgf2p8affineqb.512
+  return _mm512_gf2p8affine_epi64_epi8(A, B, 1);
+}
+
+__m512i test_mm512_mask_gf2p8affine_epi64_epi8(__m512i S, __mmask64 U, __m512i A, __m512i B) {
+  // AVX512-LABEL: @test_mm512_mask_gf2p8affine_epi64_epi8
+  // AVX512: @llvm.x86.vgf2p8affineqb.512
+  // AVX512: select <64 x i1> %{{[0-9]+}}, <64 x i8> %{{[0-9]+}}, <64 x i8> {{.*}}
+  return _mm512_mask_gf2p8affine_epi64_epi8(S, U, A, B, 1);
+}
+
+__m512i test_mm512_maskz_gf2p8affine_epi64_epi8(__mmask64 U, __m512i A, __m512i B) {
+  // AVX512-LABEL: @test_mm512_maskz_gf2p8affine_epi64_epi8
+  // AVX512: @llvm.x86.vgf2p8affineqb.512
+  // AVX512: select <64 x i1> %{{[0-9]+}}, <64 x i8> %{{[0-9]+}}, <64 x i8> {{.*}}
+  return _mm512_maskz_gf2p8affine_epi64_epi8(U, A, B, 1);
+}
+
+__m256i test_mm256_mask_gf2p8affine_epi64_epi8(__m256i S, __mmask32 U, __m256i A, __m256i B) {
+  // AVX256-LABEL: @test_mm256_mask_gf2p8affine_epi64_epi8
+  // AVX256: @llvm.x86.vgf2p8affineqb.256
+  // AVX256: select <32 x i1> %{{[0-9]+}}, <32 x i8> %{{[0-9]+}}, <32 x i8> {{.*}}
+  return _mm256_mask_gf2p8affine_epi64_epi8(S, U, A, B, 1);
+}
+
+__m256i test_mm256_maskz_gf2p8affine_epi64_epi8(__mmask32 U, __m256i A, __m256i B) {
+  // AVX256-LABEL: @test_mm256_maskz_gf2p8affine_epi64_epi8
+  // AVX256: @llvm.x86.vgf2p8affineqb.256
+  // AVX256: select <32 x i1> %{{[0-9]+}}, <32 x i8> %{{[0-9]+}}, <32 x i8> {{.*}}
+  return _mm256_maskz_gf2p8affine_epi64_epi8(U, A, B, 1);
+}
+
+__m128i test_mm_mask_gf2p8affine_epi64_epi8(__m128i S, __mmask16 U, __m128i A, __m128i B) {
+  // AVX512-LABEL: @test_mm_mask_gf2p8affine_epi64_epi8
+  // AVX512: @llvm.x86.vgf2p8affineqb.128
+  // AVX512: select <16 x i1> %{{[0-9]+}}, <16 x i8> %{{[0-9]+}}, <16 x i8> {{.*}}
+  return _mm_mask_gf2p8affine_epi64_epi8(S, U, A, B, 1);
+}
+
+__m128i test_mm_maskz_gf2p8affine_epi64_epi8(__mmask16 U, __m128i A, __m128i B) {
+  // AVX512-LABEL: @test_mm_maskz_gf2p8affine_epi64_epi8
+  // AVX512: @llvm.x86.vgf2p8affineqb.128
+  // AVX512: select <16 x i1> %{{[0-9]+}}, <16 x i8> %{{[0-9]+}}, <16 x i8> {{.*}}
+  return _mm_maskz_gf2p8affine_epi64_epi8(U, A, B, 1);
+}
+
+__m512i test_mm512_gf2p8mul_epi8(__m512i A, __m512i B) {
+  // AVX512-LABEL: @test_mm512_gf2p8mul_epi8
+  // AVX512: @llvm.x86.vgf2p8mulb.512
+  return _mm512_gf2p8mul_epi8(A, B);
+}
+
+__m512i test_mm512_mask_gf2p8mul_epi8(__m512i S, __mmask64 U, __m512i A, __m512i B) {
+  // AVX512-LABEL: @test_mm512_mask_gf2p8mul_epi8
+  // AVX512: @llvm.x86.vgf2p8mulb.512
+  // AVX512: select <64 x i1> %{{[0-9]+}}, <64 x i8> %{{[0-9]+}}, <64 x i8> {{.*}}
+  return _mm512_mask_gf2p8mul_epi8(S, U, A, B);
+}
+
+__m512i test_mm512_maskz_gf2p8mul_epi8(__mmask64 U, __m512i A, __m512i B) {
+  // AVX512-LABEL: @test_mm512_maskz_gf2p8mul_epi8
+  // AVX512: @llvm.x86.vgf2p8mulb.512
+  // AVX512: select <64 x i1> %{{[0-9]+}}, <64 x i8> %{{[0-9]+}}, <64 x i8> {{.*}}
+  return _mm512_maskz_gf2p8mul_epi8(U, A, B);
+}
+
+__m256i test_mm256_mask_gf2p8mul_epi8(__m256i S, __mmask32 U, __m256i A, __m256i B) {
+  // AVX256-LABEL: @test_mm256_mask_gf2p8mul_epi8
+  // AVX256: @llvm.x86.vgf2p8mulb.256
+  // AVX256: select <32 x i1> %{{[0-9]+}}, <32 x i8> %{{[0-9]+}}, <32 x i8> {{.*}}
+  return _mm256_mask_gf2p8mul_epi8(S, U, A, B);
+}
+
+__m256i test_mm256_maskz_gf2p8mul_epi8(__mmask32 U, __m256i A, __m256i B) {
+  // AVX256-LABEL: @test_mm256_maskz_gf2p8mul_epi8
+  // AVX256: @llvm.x86.vgf2p8mulb.256
+  // AVX256: select <32 x i1> %{{[0-9]+}}, <32 x i8> %{{[0-9]+}}, <32 x i8> {{.*}}
+  return _mm256_maskz_gf2p8mul_epi8(U, A, B);
+}
+
+__m128i test_mm_mask_gf2p8mul_epi8(__m128i S, __mmask16 U, __m128i A, __m128i B) {
+  // AVX512-LABEL: @test_mm_mask_gf2p8mul_epi8
+  // AVX512: @llvm.x86.vgf2p8mulb.128
+  // AVX512: select <16 x i1> %{{[0-9]+}}, <16 x i8> %{{[0-9]+}}, <16 x i8> {{.*}}
+  return _mm_mask_gf2p8mul_epi8(S, U, A, B);
+}
+#endif // AVX512
diff --git a/clang/test/Driver/x86-target-features.c b/clang/test/Driver/x86-target-features.c
index 065a21e1f16..b8b202c28ab 100644
--- a/clang/test/Driver/x86-target-features.c
+++ b/clang/test/Driver/x86-target-features.c
@@ -100,3 +100,8 @@
 // VAES: "-target-feature" "+vaes"
 // NO-VAES: "-target-feature" "-vaes"
 
+// RUN: %clang -target i386-unknown-linux-gnu -march=i386 -mgfni %s -### -o %t.o 2>&1 | FileCheck -check-prefix=GFNI %s
+// RUN: %clang -target i386-unknown-linux-gnu -march=i386 -mno-gfni %s -### -o %t.o 2>&1 | FileCheck -check-prefix=NO-GFNI %s
+// GFNI: "-target-feature" "+gfni"
+// NO-GFNI: "-target-feature" "-gfni
+
diff --git a/clang/test/Preprocessor/predefined-arch-macros.c b/clang/test/Preprocessor/predefined-arch-macros.c
index 81b9ad66ff0..5c440901897 100644
--- a/clang/test/Preprocessor/predefined-arch-macros.c
+++ b/clang/test/Preprocessor/predefined-arch-macros.c
@@ -1063,6 +1063,7 @@
 // CHECK_ICL_M32: #define __CLFLUSHOPT__ 1
 // CHECK_ICL_M32: #define __F16C__ 1
 // CHECK_ICL_M32: #define __FMA__ 1
+// CHECK_ICL_M32: #define __GFNI__ 1
 // CHECK_ICL_M32: #define __LZCNT__ 1
 // CHECK_ICL_M32: #define __MMX__ 1
 // CHECK_ICL_M32: #define __MPX__ 1
@@ -1109,6 +1110,7 @@
 // CHECK_ICL_M64: #define __CLFLUSHOPT__ 1
 // CHECK_ICL_M64: #define __F16C__ 1
 // CHECK_ICL_M64: #define __FMA__ 1
+// CHECK_ICL_M64: #define __GFNI__ 1
 // CHECK_ICL_M64: #define __LZCNT__ 1
 // CHECK_ICL_M64: #define __MMX__ 1
 // CHECK_ICL_M64: #define __MPX__ 1
diff --git a/clang/test/Preprocessor/x86_target_features.c b/clang/test/Preprocessor/x86_target_features.c
index 18e2b2be9a2..1db49743d88 100644
--- a/clang/test/Preprocessor/x86_target_features.c
+++ b/clang/test/Preprocessor/x86_target_features.c
@@ -379,3 +379,8 @@
 // VAESNOAES-NOT: #define __AES__ 1
 // VAESNOAES-NOT: #define __VAES__ 1
 
+// RUN: %clang -target i386-unknown-unknown -march=atom -mgfni -x c -E -dM -o - %s | FileCheck -match-full-lines --check-prefix=GFNI %s
+
+// GFNI: #define __GFNI__ 1
+// GFNI: #define __SSE2__ 1
+