From 8467662e1a795229f7c903a91f0eab27a303feb0 Mon Sep 17 00:00:00 2001
From: Adam Nemet <anemet@apple.com>
Date: Tue, 5 Aug 2014 17:28:23 +0000
Subject: [PATCH] [AVX512] Add intrinsic for valignd/q

Note that similar to palingr, we could further optimize these to emit
shufflevector when the shift count is <=64.  This however does not
change the overall design that unlike palignr we would still need the LLVM
intrinsic corresponding to this intruction to handle the >64 cases.  (palignr
uses the psrldq intrinsic in this case.)

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@214891 91177308-0d34-0410-b5e6-96231b3b80d8
---
 include/clang/Basic/BuiltinsX86.def |  2 ++
 lib/Headers/avx512fintrin.h         | 20 ++++++++++++++++++++
 test/CodeGen/avx512f-builtins.c     |  7 +++++++
 3 files changed, 29 insertions(+)

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index d546989738..117232cfc4 100644
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -838,6 +838,8 @@ BUILTIN(__builtin_ia32_vpermt2vard512_mask, "V16iV16iV16iV16iUs", "")
 BUILTIN(__builtin_ia32_vpermt2varq512_mask, "V8LLiV8LLiV8LLiV8LLiUc", "")
 BUILTIN(__builtin_ia32_vpermt2varps512_mask, "V16fV16iV16fV16fUs", "")
 BUILTIN(__builtin_ia32_vpermt2varpd512_mask, "V8dV8LLiV8dV8dUc", "")
+BUILTIN(__builtin_ia32_alignq512_mask, "V8LLiV8LLiV8LLiUcV8LLiUc", "")
+BUILTIN(__builtin_ia32_alignd512_mask, "V16iV16iV16iUcV16iUc", "")
 BUILTIN(__builtin_ia32_gathersiv8df, "V8dV8dv*V8iUcCi", "")
 BUILTIN(__builtin_ia32_gathersiv16sf, "V16fV16fv*UsCi", "")
 BUILTIN(__builtin_ia32_gatherdiv8df, "V8dV8dv*V8LLiUcCi", "")
diff --git a/lib/Headers/avx512fintrin.h b/lib/Headers/avx512fintrin.h
index c205662c16..ad92fe7b64 100644
--- a/lib/Headers/avx512fintrin.h
+++ b/lib/Headers/avx512fintrin.h
@@ -529,6 +529,26 @@ _mm512_permutex2var_ps(__m512 __A, __m512i __I, __m512 __B)
                                                        (__mmask16) -1);
 }
 
+static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_valign_epi64(__m512i __A, __m512i __B, const int __I)
+{
+  return (__m512i) __builtin_ia32_alignq512_mask((__v8di)__A,
+                                                 (__v8di)__B,
+                                                 __I,
+                                                 (__v8di)_mm512_setzero_si512(),
+                                                 (__mmask8) -1);
+}
+
+static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_valign_epi32(__m512i __A, __m512i __B, const int __I)
+{
+  return (__m512i)__builtin_ia32_alignd512_mask((__v16si)__A,
+                                                (__v16si)__B,
+                                                __I,
+                                                (__v16si)_mm512_setzero_si512(),
+                                                (__mmask16) -1);
+}
+
 /* Vector Blend */
 
 static __inline __m512d __attribute__ ((__always_inline__, __nodebug__))
diff --git a/test/CodeGen/avx512f-builtins.c b/test/CodeGen/avx512f-builtins.c
index 4c4a064ff5..35d79a8bff 100644
--- a/test/CodeGen/avx512f-builtins.c
+++ b/test/CodeGen/avx512f-builtins.c
@@ -130,3 +130,10 @@ __mmask16 test_mm512_knot(__mmask16 a)
   // CHECK: @llvm.x86.avx512.knot.w
   return _mm512_knot(a);
 }
+
+__m512i test_mm512_valign_epi64(__m512i a, __m512i b)
+{
+  // CHECK-LABEL: @test_mm512_valign_epi64
+  // CHECK: @llvm.x86.avx512.mask.valign.q.512
+  return _mm512_valign_epi64(a, b, 2);
+}
-- 
2.40.0