[AVX512] Add set1 intrinsics

(Dropped the byte and word variants from the patch.  Turns out these are not
part of AVX512F but only AVX512BW/VL.)

Part of <rdar://problem/17688758>

llvm-svn: 214314

diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index cebb0d5..a336f6d 100644 (file)
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -91,6 +91,32 @@ _mm512_setzero_pd(void)
   return (__m512d){ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 };
 }
 
+static __inline __m512 __attribute__((__always_inline__, __nodebug__))
+_mm512_set1_ps(float __w)
+{
+  return (__m512){ __w, __w, __w, __w, __w, __w, __w, __w,
+                   __w, __w, __w, __w, __w, __w, __w, __w  };
+}
+
+static __inline __m512d __attribute__((__always_inline__, __nodebug__))
+_mm512_set1_pd(double __w)
+{
+  return (__m512d){ __w, __w, __w, __w, __w, __w, __w, __w };
+}
+
+static __inline __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_set1_epi32(int __s)
+{
+  return (__m512i)(__v16si){ __s, __s, __s, __s, __s, __s, __s, __s,
+                             __s, __s, __s, __s, __s, __s, __s, __s };
+}
+
+static __inline __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_set1_epi64(long long __d)
+{
+  return (__m512i)(__v8di){ __d, __d, __d, __d, __d, __d, __d, __d };
+}
+
 /* Arithmetic */
 
 static __inline __m512d __attribute__((__always_inline__, __nodebug__))

diff --git a/clang/test/CodeGen/avx512f-builtins.c b/clang/test/CodeGen/avx512f-builtins.c
index ea84583..ddeb3b6 100644 (file)
--- a/clang/test/CodeGen/avx512f-builtins.c
+++ b/clang/test/CodeGen/avx512f-builtins.c
@@ -88,3 +88,17 @@ void test_mm512_store_pd(void *p, __m512d a)
   // CHECK: store <8 x double>
   _mm512_store_pd(p, a);
 }
+
+__m512d test_mm512_set1_pd(double d)
+{
+  // CHECK-LABEL: @test_mm512_set1_pd
+  // CHECK: insertelement <8 x double> {{.*}}, i32 0
+  // CHECK: insertelement <8 x double> {{.*}}, i32 1
+  // CHECK: insertelement <8 x double> {{.*}}, i32 2
+  // CHECK: insertelement <8 x double> {{.*}}, i32 3
+  // CHECK: insertelement <8 x double> {{.*}}, i32 4
+  // CHECK: insertelement <8 x double> {{.*}}, i32 5
+  // CHECK: insertelement <8 x double> {{.*}}, i32 6
+  // CHECK: insertelement <8 x double> {{.*}}, i32 7
+  return _mm512_set1_pd(d);
+}