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

5 files changed, 138 insertions, 17 deletions

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index 3dc7271469..e72d86bb7f 100644
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -210,12 +210,6 @@ BUILTIN(__builtin_ia32_psubusw128, "V8sV8sV8s", "")
 BUILTIN(__builtin_ia32_pmulhw128, "V8sV8sV8s", "")
 BUILTIN(__builtin_ia32_pavgb128, "V16cV16cV16c", "")
 BUILTIN(__builtin_ia32_pavgw128, "V8sV8sV8s", "")
-BUILTIN(__builtin_ia32_pcmpeqb128, "V16cV16cV16c", "")
-BUILTIN(__builtin_ia32_pcmpeqw128, "V8sV8sV8s", "")
-BUILTIN(__builtin_ia32_pcmpeqd128, "V4iV4iV4i", "")
-BUILTIN(__builtin_ia32_pcmpgtb128, "V16cV16cV16c", "")
-BUILTIN(__builtin_ia32_pcmpgtw128, "V8sV8sV8s", "")
-BUILTIN(__builtin_ia32_pcmpgtd128, "V4iV4iV4i", "")
 BUILTIN(__builtin_ia32_pmaxub128, "V16cV16cV16c", "")
 BUILTIN(__builtin_ia32_pmaxsw128, "V8sV8sV8s", "")
 BUILTIN(__builtin_ia32_pminub128, "V16cV16cV16c", "")
@@ -353,7 +347,6 @@ BUILTIN(__builtin_ia32_movntdqa, "V2LLiV2LLi*", "")
 BUILTIN(__builtin_ia32_ptestz128, "iV2LLiV2LLi", "")
 BUILTIN(__builtin_ia32_ptestc128, "iV2LLiV2LLi", "")
 BUILTIN(__builtin_ia32_ptestnzc128, "iV2LLiV2LLi", "")
-BUILTIN(__builtin_ia32_pcmpeqq, "V2LLiV2LLiV2LLi", "")
 BUILTIN(__builtin_ia32_mpsadbw128, "V16cV16cV16ci", "")
 
 // SSE 4.2
@@ -374,8 +367,6 @@ BUILTIN(__builtin_ia32_pcmpestri128, "iV16ciV16ciIc","")
 //BUILTIN(__builtin_ia32_pcmpestris128, "iV16ciV16ciIc","")
 //BUILTIN(__builtin_ia32_pcmpestriz128, "iV16ciV16ciIc","")
 
-BUILTIN(__builtin_ia32_pcmpgtq, "V2LLiV2LLiV2LLi", "")
-
 BUILTIN(__builtin_ia32_crc32qi, "UiUiUc", "")
 BUILTIN(__builtin_ia32_crc32hi, "UiUiUs", "")
 BUILTIN(__builtin_ia32_crc32si, "UiUiUi", "")
@@ -499,5 +490,9 @@ BUILTIN(__builtin_ia32_paddusw256, "V16sV16sV16s", "")
 BUILTIN(__builtin_ia32_psubusb256, "V32cV32cV32c", "")
 BUILTIN(__builtin_ia32_psubusw256, "V16sV16sV16s", "")
 BUILTIN(__builtin_ia32_palignr256, "V32cV32cV32cIc", "")
+BUILTIN(__builtin_ia32_pavgb256, "V32cV32cV32c", "")
+BUILTIN(__builtin_ia32_pavgw256, "V16sV16sV16s", "")
+BUILTIN(__builtin_ia32_pblendvb256, "V32cV32cV32cV32c", "")
+BUILTIN(__builtin_ia32_pblendw256, "V16sV16sV16sIi", "")
 
 #undef BUILTIN
diff --git a/lib/Headers/avx2intrin.h b/lib/Headers/avx2intrin.h
index df0450d7d1..813c602527 100644
--- a/lib/Headers/avx2intrin.h
+++ b/lib/Headers/avx2intrin.h
@@ -136,6 +136,78 @@ _mm256_andnot_si256(__m256i a, __m256i b)
 }
 
 static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_avg_epu8(__m256i a, __m256i b)
+{
+  return (__m256i)__builtin_ia32_pavgb256((__v32qi)a, (__v32qi)b);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_avg_epu16(__m256i a, __m256i b)
+{
+  return (__m256i)__builtin_ia32_pavgw256((__v16hi)a, (__v16hi)b);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_blendv_epi8(__m256i __V1, __m256i __V2, __m256i __M)
+{
+  return (__m256i)__builtin_ia32_pblendvb256((__v32qi)__V1, (__v32qi)__V2,
+                                              (__v32qi)__M);
+}
+
+#define _mm256_blend_epi16(V1, V2, M) __extension__ ({ \
+  __m256i __V1 = (V1); \
+  __m256i __V2 = (V2); \
+  (__m256i)__builtin_ia32_pblendw256((__v16hi)__V1, (__v16hi)__V2, M); })
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpeq_epi8(__m256i a, __m256i b)
+{
+  return (__m256i)((__v32qi)a == (__v32qi)b);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpeq_epi16(__m256i a, __m256i b)
+{
+  return (__m256i)((__v16hi)a == (__v16hi)b);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpeq_epi32(__m256i a, __m256i b)
+{
+  return (__m256i)((__v8si)a == (__v8si)b);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpeq_epi64(__m256i a, __m256i b)
+{
+  return (__m256i)((__v4di)a == (__v4di)b);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epi8(__m256i a, __m256i b)
+{
+  return (__m256i)((__v32qi)a > (__v32qi)b);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epi16(__m256i a, __m256i b)
+{
+  return (__m256i)((__v16hi)a > (__v16hi)b);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epi32(__m256i a, __m256i b)
+{
+  return (__m256i)((__v8si)a > (__v8si)b);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epi64(__m256i a, __m256i b)
+{
+  return (__m256i)((__v4di)a > (__v4di)b);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_or_si256(__m256i a, __m256i b)
 {
   return a | b;
diff --git a/lib/Headers/smmintrin.h b/lib/Headers/smmintrin.h
index a2bbd5da87..2e376ddb46 100644
--- a/lib/Headers/smmintrin.h
+++ b/lib/Headers/smmintrin.h
@@ -245,7 +245,7 @@ _mm_testnzc_si128(__m128i __M, __m128i __V)
 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_cmpeq_epi64(__m128i __V1, __m128i __V2)
 {
-  return (__m128i) __builtin_ia32_pcmpeqq((__v2di)__V1, (__v2di)__V2);
+  return (__m128i)((__v2di)__V1 == (__v2di)__V2);
 }
 
 /* SSE4 Packed Integer Sign-Extension.  */
@@ -398,7 +398,7 @@ _mm_packus_epi32(__m128i __V1, __m128i __V2)
 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_cmpgt_epi64(__m128i __V1, __m128i __V2)
 {
-  return __builtin_ia32_pcmpgtq((__v2di)__V1, (__v2di)__V2);
+  return (__m128i)((__v2di)__V1 > (__v2di)__V2);
 }
 
 /* SSE4.2 Accumulate CRC32.  */
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);