lowering broadcastm

Change-Id: I0661abea3e3742860e0a03ff9e4fcdc367eff7db
llvm-svn: 317456

5 files changed, 82 insertions, 33 deletions

diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
index d57ea8f1194..cbf0778d451 100644
--- a/clang/include/clang/Basic/BuiltinsX86.def
+++ b/clang/include/clang/Basic/BuiltinsX86.def
@@ -1580,12 +1580,6 @@ TARGET_BUILTIN(__builtin_ia32_cvtmask2q128, "V2LLiUc","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_cvtmask2q256, "V4LLiUc","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_cvtq2mask128, "UcV2LLi","","avx512dq,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_cvtq2mask256, "UcV4LLi","","avx512dq,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_broadcastmb512, "V8LLiUc","","avx512cd")
-TARGET_BUILTIN(__builtin_ia32_broadcastmw512, "V16iUs","","avx512cd")
-TARGET_BUILTIN(__builtin_ia32_broadcastmb128, "V2LLiUc","","avx512cd,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_broadcastmb256, "V4LLiUc","","avx512cd,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_broadcastmw128, "V4iUs","","avx512cd,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_broadcastmw256, "V8iUs","","avx512cd,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovsdb512_mask, "V16cV16iV16cUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovsdb512mem_mask, "vV16c*V16iUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovswb512mem_mask, "vV32c*V32sUi","","avx512bw")
diff --git a/clang/lib/Headers/avx512cdintrin.h b/clang/lib/Headers/avx512cdintrin.h
index 23c423584a7..ec7e0cd443b 100644
--- a/clang/lib/Headers/avx512cdintrin.h
+++ b/clang/lib/Headers/avx512cdintrin.h
@@ -130,13 +130,14 @@ _mm512_maskz_lzcnt_epi64 (__mmask8 __U, __m512i __A)
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_broadcastmb_epi64 (__mmask8 __A)
 {
-  return (__m512i) __builtin_ia32_broadcastmb512 (__A);
+  return (__m512i) _mm512_set1_epi64((long long) __A);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_broadcastmw_epi32 (__mmask16 __A)
 {
-  return (__m512i) __builtin_ia32_broadcastmw512 (__A);
+  return (__m512i) _mm512_set1_epi32((int) __A);
+
 }
 
 #undef __DEFAULT_FN_ATTRS
diff --git a/clang/lib/Headers/avx512vlcdintrin.h b/clang/lib/Headers/avx512vlcdintrin.h
index 7b02e2e1f92..8f1cd25f0b5 100644
--- a/clang/lib/Headers/avx512vlcdintrin.h
+++ b/clang/lib/Headers/avx512vlcdintrin.h
@@ -33,26 +33,26 @@
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm_broadcastmb_epi64 (__mmask8 __A)
-{
-  return (__m128i) __builtin_ia32_broadcastmb128 (__A);
+{ 
+  return (__m128i) _mm_set1_epi64x((long long) __A);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm256_broadcastmb_epi64 (__mmask8 __A)
 {
-  return (__m256i) __builtin_ia32_broadcastmb256 (__A);
+  return (__m256i) _mm256_set1_epi64x((long long)__A);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm_broadcastmw_epi32 (__mmask16 __A)
 {
-  return (__m128i) __builtin_ia32_broadcastmw128 (__A);
+  return (__m128i) _mm_set1_epi32((int)__A);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm256_broadcastmw_epi32 (__mmask16 __A)
 {
-  return (__m256i) __builtin_ia32_broadcastmw256 (__A);
+  return (__m256i) _mm256_set1_epi32((int)__A);
 }
 
 
diff --git a/clang/test/CodeGen/avx512cdintrin.c b/clang/test/CodeGen/avx512cdintrin.c
index a28601895be..e01d277be9f 100644
--- a/clang/test/CodeGen/avx512cdintrin.c
+++ b/clang/test/CodeGen/avx512cdintrin.c
@@ -68,14 +68,40 @@ __m512i test_mm512_maskz_lzcnt_epi64(__mmask8 __U, __m512i __A) {
   return _mm512_maskz_lzcnt_epi64(__U,__A); 
 }
 
-__m512i test_mm512_broadcastmb_epi64(__mmask8 __A) {
+__m512i test_mm512_broadcastmb_epi64(__m512i a, __m512i b) {
   // CHECK-LABEL: @test_mm512_broadcastmb_epi64
-  // CHECK: @llvm.x86.avx512.broadcastmb.512
-  return _mm512_broadcastmb_epi64(__A); 
+  // CHECK: icmp eq <8 x i64> %{{.*}}, %{{.*}}
+  // CHECK: zext i8 %{{.*}} to i64
+  // CHECK: insertelement <8 x i64> undef, i64 %{{.*}}, i32 0
+  // CHECK: insertelement <8 x i64> %{{.*}}, i64 %{{.*}}, i32 1
+  // CHECK: insertelement <8 x i64> %{{.*}}, i64 %{{.*}}, i32 2
+  // CHECK: insertelement <8 x i64> %{{.*}}, i64 %{{.*}}, i32 3
+  // CHECK: insertelement <8 x i64> %{{.*}}, i64 %{{.*}}, i32 4
+  // CHECK: insertelement <8 x i64> %{{.*}}, i64 %{{.*}}, i32 5
+  // CHECK: insertelement <8 x i64> %{{.*}}, i64 %{{.*}}, i32 6
+  // CHECK: insertelement <8 x i64> %{{.*}}, i64 %{{.*}}, i32 7
+  return _mm512_broadcastmb_epi64(_mm512_cmpeq_epu64_mask ( a, b)); 
 }
 
-__m512i test_mm512_broadcastmw_epi32(__mmask16 __A) {
+__m512i test_mm512_broadcastmw_epi32(__m512i a, __m512i b) {
   // CHECK-LABEL: @test_mm512_broadcastmw_epi32
-  // CHECK: @llvm.x86.avx512.broadcastmw.512
-  return _mm512_broadcastmw_epi32(__A); 
+  // CHECK: icmp eq <16 x i32> %{{.*}}, %{{.*}}
+  // CHECK: zext i16 %{{.*}} to i32
+  // CHECK: insertelement <16 x i32> undef, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  // CHECK: insertelement <16 x i32> %{{.*}}, i32 %{{.*}}
+  return _mm512_broadcastmw_epi32(_mm512_cmpeq_epi32_mask ( a, b)); 
 }
diff --git a/clang/test/CodeGen/avx512vlcd-builtins.c b/clang/test/CodeGen/avx512vlcd-builtins.c
index 643f24f1d22..376a342f76e 100644
--- a/clang/test/CodeGen/avx512vlcd-builtins.c
+++ b/clang/test/CodeGen/avx512vlcd-builtins.c
@@ -3,28 +3,56 @@
 
 #include <immintrin.h>
 
-__m128i test_mm_broadcastmb_epi64(__mmask8 __A) {
+__m128i test_mm_broadcastmb_epi64(__m128i a,__m128i b) {
   // CHECK-LABEL: @test_mm_broadcastmb_epi64
-  // CHECK: @llvm.x86.avx512.broadcastmb.128
-  return _mm_broadcastmb_epi64(__A); 
+  // CHECK: icmp eq <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: shufflevector <4 x i1> %{{.*}}, <4 x i1> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  // CHECK: bitcast <8 x i1> %{{.*}} to i8
+  // CHECK: zext i8 %{{.*}} to i64
+  // CHECK: insertelement <2 x i64> undef, i64 %{{.*}}, i32 0
+  // CHECK: insertelement <2 x i64> %{{.*}}, i64 %{{.*}}, i32 1
+  return _mm_broadcastmb_epi64(_mm_cmpeq_epi32_mask (a, b)); 
 }
 
-__m256i test_mm256_broadcastmb_epi64(__mmask8 __A) {
+__m256i test_mm256_broadcastmb_epi64(__m256i a, __m256i b) {
   // CHECK-LABEL: @test_mm256_broadcastmb_epi64
-  // CHECK: @llvm.x86.avx512.broadcastmb.256
-  return _mm256_broadcastmb_epi64(__A); 
-}
-
-__m128i test_mm_broadcastmw_epi32(__mmask16 __A) {
+  // CHECK: icmp eq <4 x i64> %{{.*}}, %{{.*}}
+  // CHECK: shufflevector <4 x i1> %{{.*}}, <4 x i1> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  // CHECK: bitcast <8 x i1> %{{.*}} to i8
+  // CHECK: zext i8 %{{.*}} to i64
+  // CHECK: insertelement <4 x i64> undef, i64 %{{.*}}, i32 0
+  // CHECK: insertelement <4 x i64> %{{.*}}, i64 %{{.*}}, i32 1
+  // CHECK: insertelement <4 x i64> %{{.*}}, i64 %{{.*}}, i32 2
+  // CHECK: insertelement <4 x i64> %{{.*}}, i64 %{{.*}}, i32 3
+  return _mm256_broadcastmb_epi64(_mm256_cmpeq_epi64_mask ( a, b)); 
+}
+
+__m128i test_mm_broadcastmw_epi32(__m512i a, __m512i b) {
   // CHECK-LABEL: @test_mm_broadcastmw_epi32
-  // CHECK: @llvm.x86.avx512.broadcastmw.128
-  return _mm_broadcastmw_epi32(__A); 
+  // CHECK: icmp eq <16 x i32> %{{.*}}, %{{.*}}
+  // CHECK: bitcast <16 x i1> %{{.*}} to i16
+  // CHECK: zext i16 %{{.*}} to i32
+  // CHECK: insertelement <4 x i32> undef, i32 %{{.*}}, i32 0
+  // CHECK: insertelement <4 x i32> %{{.*}}, i32 %{{.*}}, i32 1
+  // CHECK: insertelement <4 x i32> %{{.*}}, i32 %{{.*}}, i32 2
+  // CHECK: insertelement <4 x i32> %{{.*}}, i32 %{{.*}}, i32 3
+  return _mm_broadcastmw_epi32(_mm512_cmpeq_epi32_mask ( a, b));
 }
 
-__m256i test_mm256_broadcastmw_epi32(__mmask16 __A) {
+__m256i test_mm256_broadcastmw_epi32(__m512i a, __m512i b) {
   // CHECK-LABEL: @test_mm256_broadcastmw_epi32
-  // CHECK: @llvm.x86.avx512.broadcastmw.256
-  return _mm256_broadcastmw_epi32(__A); 
+  // CHECK: icmp eq <16 x i32> %{{.*}}, %{{.*}}
+  // CHECK: bitcast <16 x i1> %{{.*}} to i16
+  // CHECK: zext i16 %{{.*}} to i32
+  // CHECK: insertelement <8 x i32> undef, i32 %{{.*}}, i32 0
+  // CHECK: insertelement <8 x i32> %{{.*}}, i32 %{{.*}}, i32 1
+  // CHECK: insertelement <8 x i32> %{{.*}}, i32 %{{.*}}, i32 2
+  // CHECK: insertelement <8 x i32> %{{.*}}, i32 %{{.*}}, i32 3
+  // CHECK: insertelement <8 x i32> %{{.*}}, i32 %{{.*}}, i32 4
+  // CHECK: insertelement <8 x i32> %{{.*}}, i32 %{{.*}}, i32 5
+  // CHECK: insertelement <8 x i32> %{{.*}}, i32 %{{.*}}, i32 6
+  // CHECK: insertelement <8 x i32> %{{.*}}, i32 %{{.*}}, i32 7
+  return _mm256_broadcastmw_epi32(_mm512_cmpeq_epi32_mask ( a, b)); 
 }
 
 __m128i test_mm_conflict_epi64(__m128i __A) {