From: Adam Nemet <anemet@apple.com>
Date: Thu, 14 Aug 2014 17:17:57 +0000 (+0000)
Subject: [AVX512] Add FMA intrinsics
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=42a4388e580f437b97ddfc211f35cc81b0cbc86a;p=clang

[AVX512] Add FMA intrinsics

Part of <rdar://problem/17688758>

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@215666 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index 117232cfc4..8c591713d9 100644
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -655,7 +655,7 @@ BUILTIN(__builtin_ia32_sha256rnds2, "V4iV4iV4iV4i", "")
 BUILTIN(__builtin_ia32_sha256msg1, "V4iV4iV4i", "")
 BUILTIN(__builtin_ia32_sha256msg2, "V4iV4iV4i", "")
 
-// FMA4
+// FMA
 BUILTIN(__builtin_ia32_vfmaddps, "V4fV4fV4fV4f", "")
 BUILTIN(__builtin_ia32_vfmaddpd, "V2dV2dV2dV2d", "")
 BUILTIN(__builtin_ia32_vfmaddss, "V4fV4fV4fV4f", "")
@@ -688,6 +688,12 @@ BUILTIN(__builtin_ia32_vfmaddsubps256, "V8fV8fV8fV8f", "")
 BUILTIN(__builtin_ia32_vfmaddsubpd256, "V4dV4dV4dV4d", "")
 BUILTIN(__builtin_ia32_vfmsubaddps256, "V8fV8fV8fV8f", "")
 BUILTIN(__builtin_ia32_vfmsubaddpd256, "V4dV4dV4dV4d", "")
+BUILTIN(__builtin_ia32_vfmaddpd512_mask,  "V8dV8dV8dV8dUci", "")
+BUILTIN(__builtin_ia32_vfmsubpd512_mask,  "V8dV8dV8dV8dUci", "")
+BUILTIN(__builtin_ia32_vfnmaddpd512_mask, "V8dV8dV8dV8dUci", "")
+BUILTIN(__builtin_ia32_vfmaddps512_mask,  "V16fV16fV16fV16fUsi", "")
+BUILTIN(__builtin_ia32_vfmsubps512_mask,  "V16fV16fV16fV16fUsi", "")
+BUILTIN(__builtin_ia32_vfnmaddps512_mask, "V16fV16fV16fV16fUsi", "")
 
 // XOP
 BUILTIN(__builtin_ia32_vpmacssww, "V8sV8sV8sV8s", "")
diff --git a/lib/Headers/avx512fintrin.h b/lib/Headers/avx512fintrin.h
index 1b558a0404..5736fbe555 100644
--- a/lib/Headers/avx512fintrin.h
+++ b/lib/Headers/avx512fintrin.h
@@ -507,6 +507,72 @@ _mm512_roundscale_pd(__m512d __A, const int __imm)
                _MM_FROUND_CUR_DIRECTION);
 }
 
+static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
+_mm512_fmadd_pd(__m512d __A, __m512d __B, __m512d __C)
+{
+  return (__m512d)
+    __builtin_ia32_vfmaddpd512_mask(__A,
+                                    __B,
+                                    __C,
+                                    (__mmask8) -1,
+                                    _MM_FROUND_CUR_DIRECTION);
+}
+
+static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
+_mm512_fmsub_pd(__m512d __A, __m512d __B, __m512d __C)
+{
+  return (__m512d)
+    __builtin_ia32_vfmsubpd512_mask(__A,
+                                    __B,
+                                    __C,
+                                    (__mmask8) -1,
+                                    _MM_FROUND_CUR_DIRECTION);
+}
+
+static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
+_mm512_fnmadd_pd(__m512d __A, __m512d __B, __m512d __C)
+{
+  return (__m512d)
+    __builtin_ia32_vfnmaddpd512_mask(__A,
+                                     __B,
+                                     __C,
+                                     (__mmask8) -1,
+                                     _MM_FROUND_CUR_DIRECTION);
+}
+
+static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
+_mm512_fmadd_ps(__m512 __A, __m512 __B, __m512 __C)
+{
+  return (__m512)
+    __builtin_ia32_vfmaddps512_mask(__A,
+                                    __B,
+                                    __C,
+                                    (__mmask16) -1,
+                                    _MM_FROUND_CUR_DIRECTION);
+}
+
+static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
+_mm512_fmsub_ps(__m512 __A, __m512 __B, __m512 __C)
+{
+  return (__m512)
+    __builtin_ia32_vfmsubps512_mask(__A,
+                                    __B,
+                                    __C,
+                                    (__mmask16) -1,
+                                    _MM_FROUND_CUR_DIRECTION);
+}
+
+static __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
+_mm512_fnmadd_ps(__m512 __A, __m512 __B, __m512 __C)
+{
+  return (__m512)
+    __builtin_ia32_vfnmaddps512_mask(__A,
+                                     __B,
+                                     __C,
+                                     (__mmask16) -1,
+                                     _MM_FROUND_CUR_DIRECTION);
+}
+
 /* Vector permutations */
 
 static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
diff --git a/test/CodeGen/avx512f-builtins.c b/test/CodeGen/avx512f-builtins.c
index b5b4ff2391..bde93fb271 100644
--- a/test/CodeGen/avx512f-builtins.c
+++ b/test/CodeGen/avx512f-builtins.c
@@ -151,3 +151,10 @@ __m512d test_mm512_broadcastsd_pd(__m128d a)
   // CHECK: insertelement <8 x double> {{.*}}, i32 7
   return _mm512_broadcastsd_pd(a);
 }
+
+__m512i test_mm512_fmadd_pd(__m512d a, __m512d b, __m512d c)
+{
+  // CHECK-LABEL: @test_mm512_fmadd_pd
+  // CHECK: @llvm.x86.fma.mask.vfmadd.pd.512
+  return _mm512_fmadd_pd(a, b, c);
+}