Add AVX2 intrinsics for pavg, pblend, and pcmp instructions. Also remove unneeded builtins for SSE pcmp. Change SSE pcmpeqq and pcmpgtq to not use builtins and just use vector == and >.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@146969 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 60 insertions, 6 deletions

diff --git a/test/CodeGen/avx2-builtins.c b/test/CodeGen/avx2-builtins.c
index 3aa374e448..aa13b3c268 100644
--- a/test/CodeGen/avx2-builtins.c
+++ b/test/CodeGen/avx2-builtins.c
@@ -155,3 +155,63 @@ __m256 test_mm256_xor_si256(__m256 a, __m256 b) {
   // CHECK: xor <4 x i64>
   return _mm256_xor_si256(a, b);
 }
+
+__m256 test_mm256_avg_epu8(__m256 a, __m256 b) {
+  // CHECK: @llvm.x86.avx2.pavg.b
+  return _mm256_avg_epu8(a, b);
+}
+
+__m256 test_mm256_avg_epu16(__m256 a, __m256 b) {
+  // CHECK: @llvm.x86.avx2.pavg.w
+  return _mm256_avg_epu16(a, b);
+}
+
+__m256 test_mm256_blendv_epi8(__m256 a, __m256 b, __m256 m) {
+  // CHECK: @llvm.x86.avx2.pblendvb
+  return _mm256_blendv_epi8(a, b, m);
+}
+
+__m256 test_mm256_blend_epi16(__m256 a, __m256 b) {
+  // CHECK: @llvm.x86.avx2.pblendw(<16 x i16> %{{.*}}, <16 x i16> %{{.*}}, i32 2)
+  return _mm256_blend_epi16(a, b, 2);
+}
+
+__m256 test_mm256_cmpeq_epi8(__m256 a, __m256 b) {
+  // CHECK: icmp eq <32 x i8>
+  return _mm256_cmpeq_epi8(a, b);
+}
+
+__m256 test_mm256_cmpeq_epi16(__m256 a, __m256 b) {
+  // CHECK: icmp eq <16 x i16>
+  return _mm256_cmpeq_epi16(a, b);
+}
+
+__m256 test_mm256_cmpeq_epi32(__m256 a, __m256 b) {
+  // CHECK: icmp eq <8 x i32>
+  return _mm256_cmpeq_epi32(a, b);
+}
+
+__m256 test_mm256_cmpeq_epi64(__m256 a, __m256 b) {
+  // CHECK: icmp eq <4 x i64>
+  return _mm256_cmpeq_epi64(a, b);
+}
+
+__m256 test_mm256_cmpgt_epi8(__m256 a, __m256 b) {
+  // CHECK: icmp sgt <32 x i8>
+  return _mm256_cmpgt_epi8(a, b);
+}
+
+__m256 test_mm256_cmpgt_epi16(__m256 a, __m256 b) {
+  // CHECK: icmp sgt <16 x i16>
+  return _mm256_cmpgt_epi16(a, b);
+}
+
+__m256 test_mm256_cmpgt_epi32(__m256 a, __m256 b) {
+  // CHECK: icmp sgt <8 x i32>
+  return _mm256_cmpgt_epi32(a, b);
+}
+
+__m256 test_mm256_cmpgt_epi64(__m256 a, __m256 b) {
+  // CHECK: icmp sgt <4 x i64>
+  return _mm256_cmpgt_epi64(a, b);
+}
diff --git a/test/CodeGen/builtins-x86.c b/test/CodeGen/builtins-x86.c
index 7f028e5932..772ab105db 100644
--- a/test/CodeGen/builtins-x86.c
+++ b/test/CodeGen/builtins-x86.c
@@ -199,12 +199,6 @@ void f0() {
   tmp_V8s = __builtin_ia32_pmulhw128(tmp_V8s, tmp_V8s);
   tmp_V16c = __builtin_ia32_pavgb128(tmp_V16c, tmp_V16c);
   tmp_V8s = __builtin_ia32_pavgw128(tmp_V8s, tmp_V8s);
-  tmp_V16c = __builtin_ia32_pcmpeqb128(tmp_V16c, tmp_V16c);
-  tmp_V8s = __builtin_ia32_pcmpeqw128(tmp_V8s, tmp_V8s);
-  tmp_V4i = __builtin_ia32_pcmpeqd128(tmp_V4i, tmp_V4i);
-  tmp_V16c = __builtin_ia32_pcmpgtb128(tmp_V16c, tmp_V16c);
-  tmp_V8s = __builtin_ia32_pcmpgtw128(tmp_V8s, tmp_V8s);
-  tmp_V4i = __builtin_ia32_pcmpgtd128(tmp_V4i, tmp_V4i);
   tmp_V16c = __builtin_ia32_pmaxub128(tmp_V16c, tmp_V16c);
   tmp_V8s = __builtin_ia32_pmaxsw128(tmp_V8s, tmp_V8s);
   tmp_V16c = __builtin_ia32_pminub128(tmp_V16c, tmp_V16c);