[X86] Add builtins for pshufd, pshuflw, and pshufhw to enable target feature and immediate range checking.

llvm-svn: 334265

6 files changed, 69 insertions, 112 deletions

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 025b34e809c..741f36b095d 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -9325,6 +9325,57 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
                                        makeArrayRef(Indices, NumElts),
                                        "blend");
   }
+  case X86::BI__builtin_ia32_pshuflw:
+  case X86::BI__builtin_ia32_pshuflw256:
+  case X86::BI__builtin_ia32_pshuflw512: {
+    uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
+    llvm::Type *Ty = Ops[0]->getType();
+    unsigned NumElts = Ty->getVectorNumElements();
+
+    // Splat the 8-bits of immediate 4 times to help the loop wrap around.
+    Imm = (Imm & 0xff) * 0x01010101;
+
+    uint32_t Indices[32];
+    for (unsigned l = 0; l != NumElts; l += 8) {
+      for (unsigned i = 0; i != 4; ++i) {
+        Indices[l + i] = l + (Imm & 3);
+        Imm >>= 2;
+      }
+      for (unsigned i = 4; i != 8; ++i)
+        Indices[l + i] = l + i;
+    }
+
+    return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty),
+                                       makeArrayRef(Indices, NumElts),
+                                       "pshuflw");
+  }
+  case X86::BI__builtin_ia32_pshufhw:
+  case X86::BI__builtin_ia32_pshufhw256:
+  case X86::BI__builtin_ia32_pshufhw512: {
+    uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
+    llvm::Type *Ty = Ops[0]->getType();
+    unsigned NumElts = Ty->getVectorNumElements();
+
+    // Splat the 8-bits of immediate 4 times to help the loop wrap around.
+    Imm = (Imm & 0xff) * 0x01010101;
+
+    uint32_t Indices[32];
+    for (unsigned l = 0; l != NumElts; l += 8) {
+      for (unsigned i = 0; i != 4; ++i)
+        Indices[l + i] = l + i;
+      for (unsigned i = 4; i != 8; ++i) {
+        Indices[l + i] = l + 4 + (Imm & 3);
+        Imm >>= 2;
+      }
+    }
+
+    return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty),
+                                       makeArrayRef(Indices, NumElts),
+                                       "pshufhw");
+  }
+  case X86::BI__builtin_ia32_pshufd:
+  case X86::BI__builtin_ia32_pshufd256:
+  case X86::BI__builtin_ia32_pshufd512:
   case X86::BI__builtin_ia32_vpermilpd:
   case X86::BI__builtin_ia32_vpermilps:
   case X86::BI__builtin_ia32_vpermilpd256:
diff --git a/clang/lib/Headers/avx2intrin.h b/clang/lib/Headers/avx2intrin.h
index 3867af08ccc..e7fdd2b1e59 100644
--- a/clang/lib/Headers/avx2intrin.h
+++ b/clang/lib/Headers/avx2intrin.h
@@ -488,44 +488,13 @@ _mm256_shuffle_epi8(__m256i __a, __m256i __b)
 }
 
 #define _mm256_shuffle_epi32(a, imm) \
-  (__m256i)__builtin_shufflevector((__v8si)(__m256i)(a), \
-                                   (__v8si)_mm256_undefined_si256(), \
-                                   0 + (((imm) >> 0) & 0x3), \
-                                   0 + (((imm) >> 2) & 0x3), \
-                                   0 + (((imm) >> 4) & 0x3), \
-                                   0 + (((imm) >> 6) & 0x3), \
-                                   4 + (((imm) >> 0) & 0x3), \
-                                   4 + (((imm) >> 2) & 0x3), \
-                                   4 + (((imm) >> 4) & 0x3), \
-                                   4 + (((imm) >> 6) & 0x3))
+  (__m256i)__builtin_ia32_pshufd256((__v8si)(__m256i)(a), (int)(imm))
 
 #define _mm256_shufflehi_epi16(a, imm) \
-  (__m256i)__builtin_shufflevector((__v16hi)(__m256i)(a), \
-                                   (__v16hi)_mm256_undefined_si256(), \
-                                   0, 1, 2, 3, \
-                                   4  + (((imm) >> 0) & 0x3), \
-                                   4  + (((imm) >> 2) & 0x3), \
-                                   4  + (((imm) >> 4) & 0x3), \
-                                   4  + (((imm) >> 6) & 0x3), \
-                                   8, 9, 10, 11, \
-                                   12 + (((imm) >> 0) & 0x3), \
-                                   12 + (((imm) >> 2) & 0x3), \
-                                   12 + (((imm) >> 4) & 0x3), \
-                                   12 + (((imm) >> 6) & 0x3))
+  (__m256i)__builtin_ia32_pshufhw256((__v16hi)(__m256i)(a), (int)(imm))
 
 #define _mm256_shufflelo_epi16(a, imm) \
-  (__m256i)__builtin_shufflevector((__v16hi)(__m256i)(a), \
-                                   (__v16hi)_mm256_undefined_si256(), \
-                                   0 + (((imm) >> 0) & 0x3), \
-                                   0 + (((imm) >> 2) & 0x3), \
-                                   0 + (((imm) >> 4) & 0x3), \
-                                   0 + (((imm) >> 6) & 0x3), \
-                                   4, 5, 6, 7, \
-                                   8 + (((imm) >> 0) & 0x3), \
-                                   8 + (((imm) >> 2) & 0x3), \
-                                   8 + (((imm) >> 4) & 0x3), \
-                                   8 + (((imm) >> 6) & 0x3), \
-                                   12, 13, 14, 15)
+  (__m256i)__builtin_ia32_pshuflw256((__v16hi)(__m256i)(a), (int)(imm))
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm256_sign_epi8(__m256i __a, __m256i __b)
diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index aabaabd648d..0e21a57f311 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -1287,28 +1287,7 @@ _mm512_maskz_cvtepu8_epi16(__mmask32 __U, __m256i __A)
 
 
 #define _mm512_shufflehi_epi16(A, imm) \
-  (__m512i)__builtin_shufflevector((__v32hi)(__m512i)(A), \
-                                   (__v32hi)_mm512_undefined_epi32(), \
-                                   0, 1, 2, 3, \
-                                   4  + (((imm) >> 0) & 0x3), \
-                                   4  + (((imm) >> 2) & 0x3), \
-                                   4  + (((imm) >> 4) & 0x3), \
-                                   4  + (((imm) >> 6) & 0x3), \
-                                   8, 9, 10, 11, \
-                                   12 + (((imm) >> 0) & 0x3), \
-                                   12 + (((imm) >> 2) & 0x3), \
-                                   12 + (((imm) >> 4) & 0x3), \
-                                   12 + (((imm) >> 6) & 0x3), \
-                                   16, 17, 18, 19, \
-                                   20 + (((imm) >> 0) & 0x3), \
-                                   20 + (((imm) >> 2) & 0x3), \
-                                   20 + (((imm) >> 4) & 0x3), \
-                                   20 + (((imm) >> 6) & 0x3), \
-                                   24, 25, 26, 27, \
-                                   28 + (((imm) >> 0) & 0x3), \
-                                   28 + (((imm) >> 2) & 0x3), \
-                                   28 + (((imm) >> 4) & 0x3), \
-                                   28 + (((imm) >> 6) & 0x3))
+  (__m512i)__builtin_ia32_pshufhw512((__v32hi)(__m512i)(A), (int)(imm))
 
 #define _mm512_mask_shufflehi_epi16(W, U, A, imm) \
   (__m512i)__builtin_ia32_selectw_512((__mmask32)(U), \
@@ -1323,28 +1302,7 @@ _mm512_maskz_cvtepu8_epi16(__mmask32 __U, __m256i __A)
                                       (__v32hi)_mm512_setzero_si512())
 
 #define _mm512_shufflelo_epi16(A, imm) \
-  (__m512i)__builtin_shufflevector((__v32hi)(__m512i)(A), \
-                                   (__v32hi)_mm512_undefined_epi32(), \
-                                   0 + (((imm) >> 0) & 0x3), \
-                                   0 + (((imm) >> 2) & 0x3), \
-                                   0 + (((imm) >> 4) & 0x3), \
-                                   0 + (((imm) >> 6) & 0x3), \
-                                   4, 5, 6, 7, \
-                                   8 + (((imm) >> 0) & 0x3), \
-                                   8 + (((imm) >> 2) & 0x3), \
-                                   8 + (((imm) >> 4) & 0x3), \
-                                   8 + (((imm) >> 6) & 0x3), \
-                                   12, 13, 14, 15, \
-                                   16 + (((imm) >> 0) & 0x3), \
-                                   16 + (((imm) >> 2) & 0x3), \
-                                   16 + (((imm) >> 4) & 0x3), \
-                                   16 + (((imm) >> 6) & 0x3), \
-                                   20, 21, 22, 23, \
-                                   24 + (((imm) >> 0) & 0x3), \
-                                   24 + (((imm) >> 2) & 0x3), \
-                                   24 + (((imm) >> 4) & 0x3), \
-                                   24 + (((imm) >> 6) & 0x3), \
-                                   28, 29, 30, 31)
+  (__m512i)__builtin_ia32_pshuflw512((__v32hi)(__m512i)(A), (int)(imm))
 
 
 #define _mm512_mask_shufflelo_epi16(W, U, A, imm) \
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 4ae235e6330..fe730ac27f2 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -8849,24 +8849,7 @@ _mm_maskz_load_sd (__mmask8 __U, const double* __A)
 }
 
 #define _mm512_shuffle_epi32(A, I) \
-  (__m512i)__builtin_shufflevector((__v16si)(__m512i)(A), \
-                                   (__v16si)_mm512_undefined_epi32(), \
-                                   0  + (((I) >> 0) & 0x3), \
-                                   0  + (((I) >> 2) & 0x3), \
-                                   0  + (((I) >> 4) & 0x3), \
-                                   0  + (((I) >> 6) & 0x3), \
-                                   4  + (((I) >> 0) & 0x3), \
-                                   4  + (((I) >> 2) & 0x3), \
-                                   4  + (((I) >> 4) & 0x3), \
-                                   4  + (((I) >> 6) & 0x3), \
-                                   8  + (((I) >> 0) & 0x3), \
-                                   8  + (((I) >> 2) & 0x3), \
-                                   8  + (((I) >> 4) & 0x3), \
-                                   8  + (((I) >> 6) & 0x3), \
-                                   12 + (((I) >> 0) & 0x3), \
-                                   12 + (((I) >> 2) & 0x3), \
-                                   12 + (((I) >> 4) & 0x3), \
-                                   12 + (((I) >> 6) & 0x3))
+  (__m512i)__builtin_ia32_pshufd512((__v16si)(__m512i)(A), (int)(I))
 
 #define _mm512_mask_shuffle_epi32(W, U, A, I) \
   (__m512i)__builtin_ia32_selectd_512((__mmask16)(U), \
diff --git a/clang/lib/Headers/emmintrin.h b/clang/lib/Headers/emmintrin.h
index ad836b1e637..8942790d5c5 100644
--- a/clang/lib/Headers/emmintrin.h
+++ b/clang/lib/Headers/emmintrin.h
@@ -4343,10 +4343,7 @@ _mm_movemask_epi8(__m128i __a)
 ///    11: assign values from bits [127:96] of \a a.
 /// \returns A 128-bit integer vector containing the shuffled values.
 #define _mm_shuffle_epi32(a, imm) \
-  (__m128i)__builtin_shufflevector((__v4si)(__m128i)(a), \
-                                   (__v4si)_mm_undefined_si128(), \
-                                   ((imm) >> 0) & 0x3, ((imm) >> 2) & 0x3, \
-                                   ((imm) >> 4) & 0x3, ((imm) >> 6) & 0x3)
+  (__m128i)__builtin_ia32_pshufd((__v4si)(__m128i)(a), (int)(imm))
 
 /// Constructs a 128-bit integer vector by shuffling four lower 16-bit
 ///    elements of a 128-bit integer vector of [8 x i16], using the immediate
@@ -4376,11 +4373,7 @@ _mm_movemask_epi8(__m128i __a)
 ///    11: assign values from bits [63:48] of \a a. \n
 /// \returns A 128-bit integer vector containing the shuffled values.
 #define _mm_shufflelo_epi16(a, imm) \
-  (__m128i)__builtin_shufflevector((__v8hi)(__m128i)(a), \
-                                   (__v8hi)_mm_undefined_si128(), \
-                                   ((imm) >> 0) & 0x3, ((imm) >> 2) & 0x3, \
-                                   ((imm) >> 4) & 0x3, ((imm) >> 6) & 0x3, \
-                                   4, 5, 6, 7)
+  (__m128i)__builtin_ia32_pshuflw((__v8hi)(__m128i)(a), (int)(imm))
 
 /// Constructs a 128-bit integer vector by shuffling four upper 16-bit
 ///    elements of a 128-bit integer vector of [8 x i16], using the immediate
@@ -4410,13 +4403,7 @@ _mm_movemask_epi8(__m128i __a)
 ///    11: assign values from bits [127:112] of \a a. \n
 /// \returns A 128-bit integer vector containing the shuffled values.
 #define _mm_shufflehi_epi16(a, imm) \
-  (__m128i)__builtin_shufflevector((__v8hi)(__m128i)(a), \
-                                   (__v8hi)_mm_undefined_si128(), \
-                                   0, 1, 2, 3, \
-                                   4 + (((imm) >> 0) & 0x3), \
-                                   4 + (((imm) >> 2) & 0x3), \
-                                   4 + (((imm) >> 4) & 0x3), \
-                                   4 + (((imm) >> 6) & 0x3))
+  (__m128i)__builtin_ia32_pshufhw((__v8hi)(__m128i)(a), (int)(imm))
 
 /// Unpacks the high-order (index 8-15) values from two 128-bit vectors
 ///    of [16 x i8] and interleaves them into a 128-bit vector of [16 x i8].
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index 4f5405935e2..020d82edf72 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -2759,6 +2759,15 @@ bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case X86::BI__builtin_ia32_vpermilps256:
   case X86::BI__builtin_ia32_vpermilpd512:
   case X86::BI__builtin_ia32_vpermilps512:
+  case X86::BI__builtin_ia32_pshufd:
+  case X86::BI__builtin_ia32_pshufd256:
+  case X86::BI__builtin_ia32_pshufd512:
+  case X86::BI__builtin_ia32_pshufhw:
+  case X86::BI__builtin_ia32_pshufhw256:
+  case X86::BI__builtin_ia32_pshufhw512:
+  case X86::BI__builtin_ia32_pshuflw:
+  case X86::BI__builtin_ia32_pshuflw256:
+  case X86::BI__builtin_ia32_pshuflw512:
   case X86::BI__builtin_ia32_vcvtps2ph:
   case X86::BI__builtin_ia32_vcvtps2ph_mask:
   case X86::BI__builtin_ia32_vcvtps2ph256: