[X86] Add builtins for vpermilps/pd instructions to enable target feature checking.

llvm-svn: 334256

4 files changed, 39 insertions, 51 deletions

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 00020584955..4331005cb35 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -9256,6 +9256,33 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
                                        makeArrayRef(Indices, NumElts),
                                        "blend");
   }
+  case X86::BI__builtin_ia32_vpermilpd:
+  case X86::BI__builtin_ia32_vpermilps:
+  case X86::BI__builtin_ia32_vpermilpd256:
+  case X86::BI__builtin_ia32_vpermilps256:
+  case X86::BI__builtin_ia32_vpermilpd512:
+  case X86::BI__builtin_ia32_vpermilps512: {
+    uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
+    llvm::Type *Ty = Ops[0]->getType();
+    unsigned NumElts = Ty->getVectorNumElements();
+    unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128;
+    unsigned NumLaneElts = NumElts / NumLanes;
+
+    // Splat the 8-bits of immediate 4 times to help the loop wrap around.
+    Imm = (Imm & 0xff) * 0x01010101;
+
+    uint32_t Indices[16];
+    for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
+      for (unsigned i = 0; i != NumLaneElts; ++i) {
+        Indices[i + l] = (Imm % NumLaneElts) + l;
+        Imm /= NumLaneElts;
+      }
+    }
+
+    return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty),
+                                       makeArrayRef(Indices, NumElts),
+                                       "permil");
+  }
   case X86::BI__builtin_ia32_palignr128:
   case X86::BI__builtin_ia32_palignr256:
   case X86::BI__builtin_ia32_palignr512: {
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 074bf61c7ba..ccc445a6acd 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -6302,16 +6302,7 @@ _mm_cvttss_u64 (__m128 __A)
 #endif
 
 #define _mm512_permute_pd(X, C) \
-  (__m512d)__builtin_shufflevector((__v8df)(__m512d)(X), \
-                                   (__v8df)_mm512_undefined_pd(), \
-                                   0 + (((C) >> 0) & 0x1), \
-                                   0 + (((C) >> 1) & 0x1), \
-                                   2 + (((C) >> 2) & 0x1), \
-                                   2 + (((C) >> 3) & 0x1), \
-                                   4 + (((C) >> 4) & 0x1), \
-                                   4 + (((C) >> 5) & 0x1), \
-                                   6 + (((C) >> 6) & 0x1), \
-                                   6 + (((C) >> 7) & 0x1))
+  (__m512d)__builtin_ia32_vpermilpd512((__v8df)(__m512d)(X), (int)(C))
 
 #define _mm512_mask_permute_pd(W, U, X, C) \
   (__m512d)__builtin_ia32_selectpd_512((__mmask8)(U), \
@@ -6324,24 +6315,7 @@ _mm_cvttss_u64 (__m128 __A)
                                        (__v8df)_mm512_setzero_pd())
 
 #define _mm512_permute_ps(X, C) \
-  (__m512)__builtin_shufflevector((__v16sf)(__m512)(X), \
-                                  (__v16sf)_mm512_undefined_ps(), \
-                                   0  + (((C) >> 0) & 0x3), \
-                                   0  + (((C) >> 2) & 0x3), \
-                                   0  + (((C) >> 4) & 0x3), \
-                                   0  + (((C) >> 6) & 0x3), \
-                                   4  + (((C) >> 0) & 0x3), \
-                                   4  + (((C) >> 2) & 0x3), \
-                                   4  + (((C) >> 4) & 0x3), \
-                                   4  + (((C) >> 6) & 0x3), \
-                                   8  + (((C) >> 0) & 0x3), \
-                                   8  + (((C) >> 2) & 0x3), \
-                                   8  + (((C) >> 4) & 0x3), \
-                                   8  + (((C) >> 6) & 0x3), \
-                                   12 + (((C) >> 0) & 0x3), \
-                                   12 + (((C) >> 2) & 0x3), \
-                                   12 + (((C) >> 4) & 0x3), \
-                                   12 + (((C) >> 6) & 0x3))
+  (__m512)__builtin_ia32_vpermilps512((__v16sf)(__m512)(X), (int)(C))
 
 #define _mm512_mask_permute_ps(W, U, X, C) \
   (__m512)__builtin_ia32_selectps_512((__mmask16)(U), \
diff --git a/clang/lib/Headers/avxintrin.h b/clang/lib/Headers/avxintrin.h
index 7e3c51ffb69..7c85893ba13 100644
--- a/clang/lib/Headers/avxintrin.h
+++ b/clang/lib/Headers/avxintrin.h
@@ -998,9 +998,7 @@ _mm256_permutevar_ps(__m256 __a, __m256i __c)
 ///         returned vector.
 /// \returns A 128-bit vector of [2 x double] containing the copied values.
 #define _mm_permute_pd(A, C) \
-  (__m128d)__builtin_shufflevector((__v2df)(__m128d)(A), \
-                                   (__v2df)_mm_undefined_pd(), \
-                                   ((C) >> 0) & 0x1, ((C) >> 1) & 0x1)
+  (__m128d)__builtin_ia32_vpermilpd((__v2df)(__m128d)(A), (int)(C))
 
 /// Copies the values in a 256-bit vector of [4 x double] as specified by
 ///    the immediate integer operand.
@@ -1040,12 +1038,7 @@ _mm256_permutevar_ps(__m256 __a, __m256i __c)
 ///         returned vector.
 /// \returns A 256-bit vector of [4 x double] containing the copied values.
 #define _mm256_permute_pd(A, C) \
-  (__m256d)__builtin_shufflevector((__v4df)(__m256d)(A), \
-                                   (__v4df)_mm256_undefined_pd(), \
-                                   0 + (((C) >> 0) & 0x1), \
-                                   0 + (((C) >> 1) & 0x1), \
-                                   2 + (((C) >> 2) & 0x1), \
-                                   2 + (((C) >> 3) & 0x1))
+  (__m256d)__builtin_ia32_vpermilpd256((__v4df)(__m256d)(A), (int)(C))
 
 /// Copies the values in a 128-bit vector of [4 x float] as specified by
 ///    the immediate integer operand.
@@ -1101,10 +1094,7 @@ _mm256_permutevar_ps(__m256 __a, __m256i __c)
 ///          returned vector.
 /// \returns A 128-bit vector of [4 x float] containing the copied values.
 #define _mm_permute_ps(A, C) \
-  (__m128)__builtin_shufflevector((__v4sf)(__m128)(A), \
-                                  (__v4sf)_mm_undefined_ps(), \
-                                  ((C) >> 0) & 0x3, ((C) >> 2) & 0x3, \
-                                  ((C) >> 4) & 0x3, ((C) >> 6) & 0x3)
+  (__m128)__builtin_ia32_vpermilps((__v4sf)(__m128)(A), (int)(C))
 
 /// Copies the values in a 256-bit vector of [8 x float] as specified by
 ///    the immediate integer operand.
@@ -1196,16 +1186,7 @@ _mm256_permutevar_ps(__m256 __a, __m256i __c)
 ///          returned vector.
 /// \returns A 256-bit vector of [8 x float] containing the copied values.
 #define _mm256_permute_ps(A, C) \
-  (__m256)__builtin_shufflevector((__v8sf)(__m256)(A), \
-                                  (__v8sf)_mm256_undefined_ps(), \
-                                  0 + (((C) >> 0) & 0x3), \
-                                  0 + (((C) >> 2) & 0x3), \
-                                  0 + (((C) >> 4) & 0x3), \
-                                  0 + (((C) >> 6) & 0x3), \
-                                  4 + (((C) >> 0) & 0x3), \
-                                  4 + (((C) >> 2) & 0x3), \
-                                  4 + (((C) >> 4) & 0x3), \
-                                  4 + (((C) >> 6) & 0x3))
+  (__m256)__builtin_ia32_vpermilps256((__v8sf)(__m256)(A), (int)(C))
 
 /// Permutes 128-bit data values stored in two 256-bit vectors of
 ///    [4 x double], as specified by the immediate integer operand.
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index db004607cc0..d16921c6c70 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -2612,6 +2612,7 @@ bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case X86::BI__builtin_ia32_vec_set_v2di:
     i = 2; l = 0; u = 1;
     break;
+  case X86::BI__builtin_ia32_vpermilpd:
   case X86::BI__builtin_ia32_vec_ext_v4hi:
   case X86::BI__builtin_ia32_vec_ext_v4si:
   case X86::BI__builtin_ia32_vec_ext_v4sf:
@@ -2676,6 +2677,7 @@ bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case X86::BI__builtin_ia32_vec_set_v8si:
     i = 2; l = 0; u = 7;
     break;
+  case X86::BI__builtin_ia32_vpermilpd256:
   case X86::BI__builtin_ia32_roundps:
   case X86::BI__builtin_ia32_roundpd:
   case X86::BI__builtin_ia32_roundps256:
@@ -2721,6 +2723,10 @@ bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case X86::BI__builtin_ia32_vec_set_v32qi:
     i = 2; l = 0; u = 31;
     break;
+  case X86::BI__builtin_ia32_vpermilps:
+  case X86::BI__builtin_ia32_vpermilps256:
+  case X86::BI__builtin_ia32_vpermilpd512:
+  case X86::BI__builtin_ia32_vpermilps512:
   case X86::BI__builtin_ia32_vcvtps2ph:
   case X86::BI__builtin_ia32_vcvtps2ph_mask:
   case X86::BI__builtin_ia32_vcvtps2ph256: