[X86] Teach clang to lower __builtin_ia32_psrldqi256 and __builtin_ia32_pslldqi256 to vector shuffles the backend recognizes. This is a step towards removing the corresponding intrinsics from the backend.

llvm-svn: 229348

2 files changed, 56 insertions, 2 deletions

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index fe46e6fce9e..75c8716ecfa 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -6025,6 +6025,60 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
     // If palignr is shifting the pair of vectors more than 32 bytes, emit zero.
     return llvm::Constant::getNullValue(ConvertType(E->getType()));
   }
+  case X86::BI__builtin_ia32_pslldqi256: {
+    // Shift value is in bits so divide by 8.
+    unsigned shiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() >> 3;
+
+    // If pslldq is shifting the vector more than 15 bytes, emit zero.
+    if (shiftVal >= 16)
+      return llvm::Constant::getNullValue(ConvertType(E->getType()));
+
+    SmallVector<llvm::Constant*, 32> Indices;
+    // 256-bit pslldq operates on 128-bit lanes so we need to handle that
+    for (unsigned l = 0; l != 32; l += 16) {
+      for (unsigned i = 0; i != 16; ++i) {
+        unsigned Idx = 32 + i - shiftVal;
+        if (Idx < 32) Idx -= 16; // end of lane, switch operand.
+        Indices.push_back(llvm::ConstantInt::get(Int32Ty, Idx + l));
+      }
+    }
+
+    llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, 32);
+    Ops[0] = Builder.CreateBitCast(Ops[0], VecTy, "cast");
+    Value *Zero = llvm::Constant::getNullValue(VecTy);
+
+    Value *SV = llvm::ConstantVector::get(Indices);
+    SV = Builder.CreateShuffleVector(Zero, Ops[0], SV, "pslldq");
+    llvm::Type *ResultType = ConvertType(E->getType());
+    return Builder.CreateBitCast(SV, ResultType, "cast");
+  }
+  case X86::BI__builtin_ia32_psrldqi256: {
+    // Shift value is in bits so divide by 8.
+    unsigned shiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() >> 3;
+
+    // If psrldq is shifting the vector more than 15 bytes, emit zero.
+    if (shiftVal >= 16)
+      return llvm::Constant::getNullValue(ConvertType(E->getType()));
+
+    SmallVector<llvm::Constant*, 32> Indices;
+    // 256-bit psrldq operates on 128-bit lanes so we need to handle that
+    for (unsigned l = 0; l != 32; l += 16) {
+      for (unsigned i = 0; i != 16; ++i) {
+        unsigned Idx = i + shiftVal;
+        if (Idx >= 16) Idx += 16; // end of lane, switch operand.
+        Indices.push_back(llvm::ConstantInt::get(Int32Ty, Idx + l));
+      }
+    }
+
+    llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, 32);
+    Ops[0] = Builder.CreateBitCast(Ops[0], VecTy, "cast");
+    Value *Zero = llvm::Constant::getNullValue(VecTy);
+
+    Value *SV = llvm::ConstantVector::get(Indices);
+    SV = Builder.CreateShuffleVector(Ops[0], Zero, SV, "psrldq");
+    llvm::Type *ResultType = ConvertType(E->getType());
+    return Builder.CreateBitCast(SV, ResultType, "cast");
+  }
   case X86::BI__builtin_ia32_movntps:
   case X86::BI__builtin_ia32_movntps256:
   case X86::BI__builtin_ia32_movntpd:
diff --git a/clang/test/CodeGen/avx2-builtins.c b/clang/test/CodeGen/avx2-builtins.c
index 04825ffa2f6..a30f590ad16 100644
--- a/clang/test/CodeGen/avx2-builtins.c
+++ b/clang/test/CodeGen/avx2-builtins.c
@@ -462,7 +462,7 @@ __m256i test_mm256_sign_epi32(__m256i a, __m256i b) {
 }
 
 __m256i test_mm256_slli_si256(__m256i a) {
-  // CHECK: @llvm.x86.avx2.psll.dq
+  // CHECK: shufflevector <32 x i8> zeroinitializer, <32 x i8> %{{.*}}, <32 x i32> <i32 13, i32 14, i32 15, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 29, i32 30, i32 31, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60>
   return _mm256_slli_si256(a, 3);
 }
 
@@ -517,7 +517,7 @@ __m256i test_mm256_sra_epi32(__m256i a, __m128i b) {
 }
 
 __m256i test_mm256_srli_si256(__m256i a) {
-  // CHECK: @llvm.x86.avx2.psrl.dq
+  // CHECK: shufflevector <32 x i8> %{{.*}}, <32 x i8> zeroinitializer, <32 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 32, i32 33, i32 34, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 48, i32 49, i32 50>
   return _mm256_srli_si256(a, 3);
 }