Add AVX2 permute intrinsics. Also add parentheses on some macro arguments in other...

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@147242 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index c13fdb082dc127d00748808b359317e524c08afa..5aa2c1bd5b3987aaeca0422c3ec71df9734ba147 100644 (file)
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -562,5 +562,18 @@ BUILTIN(__builtin_ia32_vbroadcastsd_pd256, "V4dV2d", "")
 BUILTIN(__builtin_ia32_vbroadcastsi256, "V4LLiV2LLiC*", "")
 BUILTIN(__builtin_ia32_pblendd128, "V4iV4iV4iIi", "")
 BUILTIN(__builtin_ia32_pblendd256, "V8iV8iV8iIi", "")
+BUILTIN(__builtin_ia32_pbroadcastb256, "V32cV16c", "")
+BUILTIN(__builtin_ia32_pbroadcastw256, "V16sV8s", "")
+BUILTIN(__builtin_ia32_pbroadcastd256, "V8iV4i", "")
+BUILTIN(__builtin_ia32_pbroadcastq256, "V4LLiV2LLi", "")
+BUILTIN(__builtin_ia32_pbroadcastb128, "V16cV16c", "")
+BUILTIN(__builtin_ia32_pbroadcastw128, "V8sV8s", "")
+BUILTIN(__builtin_ia32_pbroadcastd128, "V4iV4i", "")
+BUILTIN(__builtin_ia32_pbroadcastq128, "V2LLiV2LLi", "")
+BUILTIN(__builtin_ia32_permvarsi256, "V8iV8iV8i", "")
+BUILTIN(__builtin_ia32_permdf256, "V4dV4dIc", "")
+BUILTIN(__builtin_ia32_permvarsf256, "V8fV8fV8f", "")
+BUILTIN(__builtin_ia32_permdi256, "V4LLiV4LLiIc", "")
+BUILTIN(__builtin_ia32_permti256, "V4LLiV4LLiV4LLiIc", "")
 
 #undef BUILTIN

diff --git a/test/CodeGen/avx2-builtins.c b/test/CodeGen/avx2-builtins.c
index dc9c5009c64252f75c1a9050e6bb040ba2d74b0a..caa27f639d0a3d373979c281855a38619888bcf8 100644 (file)
--- a/test/CodeGen/avx2-builtins.c
+++ b/test/CodeGen/avx2-builtins.c
@@ -597,7 +597,7 @@ __m256 test_mm256_broadcastss_ps(__m128 a) {
 }
 
 __m256d test_mm256_broadcastsd_pd(__m128d a) {
-  // CHECK: @llvm.x86.avx2.vbroadcast.sd.pd.256
+  // check: @llvm.x86.avx2.vbroadcast.sd.pd.256
   return _mm256_broadcastsd_pd(a);
 }
 
@@ -615,3 +615,68 @@ __m256i test_mm256_blend_epi32(__m256i a, __m256i b) {
   // CHECK: @llvm.x86.avx2.pblendd.256
   return _mm256_blend_epi32(a, b, 57);
 }
+
+__m256i test_mm256_broadcastb_epi8(__m128i a) {
+  // CHECK: @llvm.x86.avx2.pbroadcastb.256
+  return _mm256_broadcastb_epi8(a);
+}
+
+__m256i test_mm256_broadcastw_epi16(__m128i a) {
+  // CHECK: @llvm.x86.avx2.pbroadcastw.256
+  return _mm256_broadcastw_epi16(a);
+}
+
+__m256i test_mm256_broadcastd_epi32(__m128i a) {
+  // CHECK: @llvm.x86.avx2.pbroadcastd.256
+  return _mm256_broadcastd_epi32(a);
+}
+
+__m256i test_mm256_broadcastq_epi64(__m128i a) {
+  // CHECK: @llvm.x86.avx2.pbroadcastq.256
+  return _mm256_broadcastq_epi64(a);
+}
+
+__m128i test_mm_broadcastb_epi8(__m128i a) {
+  // CHECK: @llvm.x86.avx2.pbroadcastb.128
+  return _mm_broadcastb_epi8(a);
+}
+
+__m128i test_mm_broadcastw_epi16(__m128i a) {
+  // CHECK: @llvm.x86.avx2.pbroadcastw.128
+  return _mm_broadcastw_epi16(a);
+}
+
+__m128i test_mm_broadcastd_epi32(__m128i a) {
+  // CHECK: @llvm.x86.avx2.pbroadcastd.128
+  return _mm_broadcastd_epi32(a);
+}
+
+__m128i test_mm_broadcastq_epi64(__m128i a) {
+  // CHECK: @llvm.x86.avx2.pbroadcastq.128
+  return _mm_broadcastq_epi64(a);
+}
+
+__m256i test_mm256_permutevar8x32_epi32(__m256i a, __m256i b) {
+  // CHECK: @llvm.x86.avx2.permd
+  return _mm256_permutevar8x32_epi32(a, b);
+}
+
+__m256d test_mm256_permute4x64_pd(__m256d a) {
+  // CHECK: @llvm.x86.avx2.permpd
+  return _mm256_permute4x64_pd(a, 25);
+}
+
+__m256 test_mm256_permutevar8x32_ps(__m256 a, __m256 b) {
+  // CHECK: @llvm.x86.avx2.permps
+  return _mm256_permutevar8x32_ps(a, b);
+}
+
+__m256i test_mm256_permute4x64_epi64(__m256i a) {
+  // CHECK:  @llvm.x86.avx2.permq
+  return _mm256_permute4x64_epi64(a, 35);
+}
+
+__m256i test_mm256_permute2x128_si256(__m256i a, __m256i b) {
+  // CHECK: @llvm.x86.avx2.vperm2i128
+  return _mm256_permute2x128_si256(a, b, 94);
+}