[X86, AVX] replace vextractf128 intrinsics with generic shuffles

Now that we've replaced the vinsertf128 intrinsics,
do the same for their extract twins.

This is very much like D8086 (checked in at r231794):
We want to replace as much custom x86 shuffling via intrinsics
as possible because pushing the code down the generic shuffle
optimization path allows for better codegen and less complexity
in LLVM.

This is also the LLVM sibling to the cfe D8275 patch.

Differential Revision: http://reviews.llvm.org/D8276

llvm-svn: 232045

diff --git a/llvm/include/llvm/IR/IntrinsicsX86.td b/llvm/include/llvm/IR/IntrinsicsX86.td
index 998b249..b7ffeeb 100644 (file)
--- a/llvm/include/llvm/IR/IntrinsicsX86.td
+++ b/llvm/include/llvm/IR/IntrinsicsX86.td
@@ -1172,19 +1172,6 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
                   llvm_v8f32_ty, llvm_i8_ty], [IntrNoMem]>;
 }
 
-// Vector extract and insert
-let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
-  def int_x86_avx_vextractf128_pd_256 :
-        GCCBuiltin<"__builtin_ia32_vextractf128_pd256">,
-        Intrinsic<[llvm_v2f64_ty], [llvm_v4f64_ty, llvm_i8_ty], [IntrNoMem]>;
-  def int_x86_avx_vextractf128_ps_256 :
-        GCCBuiltin<"__builtin_ia32_vextractf128_ps256">,
-        Intrinsic<[llvm_v4f32_ty], [llvm_v8f32_ty, llvm_i8_ty], [IntrNoMem]>;
-  def int_x86_avx_vextractf128_si_256 :
-        GCCBuiltin<"__builtin_ia32_vextractf128_si256">,
-        Intrinsic<[llvm_v4i32_ty], [llvm_v8i32_ty, llvm_i8_ty], [IntrNoMem]>;
-}
-
 // Vector convert
 let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
   def int_x86_avx_cvtdq2_pd_256 : GCCBuiltin<"__builtin_ia32_cvtdq2pd256">,

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index b9e6e44..d4d449a 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -4978,9 +4978,6 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) {
     setValue(&I, Res);
     return nullptr;
   }
-  case Intrinsic::x86_avx_vextractf128_pd_256:
-  case Intrinsic::x86_avx_vextractf128_ps_256:
-  case Intrinsic::x86_avx_vextractf128_si_256:
   case Intrinsic::x86_avx2_vextracti128: {
     EVT DestVT = TLI.getValueType(I.getType());
     uint64_t Idx = (cast<ConstantInt>(I.getArgOperand(1))->getZExtValue() & 1) *

diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index f9493bc..4075705 100644 (file)
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -161,6 +161,9 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
         Name == "x86.avx.vinsertf128.pd.256" ||
         Name == "x86.avx.vinsertf128.ps.256" ||
         Name == "x86.avx.vinsertf128.si.256" ||
+        Name == "x86.avx.vextractf128.pd.256" ||
+        Name == "x86.avx.vextractf128.ps.256" ||
+        Name == "x86.avx.vextractf128.si.256" ||
         Name == "x86.avx.movnt.dq.256" ||
         Name == "x86.avx.movnt.pd.256" ||
         Name == "x86.avx.movnt.ps.256" ||
@@ -676,6 +679,26 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
         Idxs2.push_back(Builder.getInt32(Idx));
       }
       Rep = Builder.CreateShuffleVector(Op0, Rep, ConstantVector::get(Idxs2));
+    } else if (Name == "llvm.x86.avx.vextractf128.pd.256" ||
+               Name == "llvm.x86.avx.vextractf128.ps.256" ||
+               Name == "llvm.x86.avx.vextractf128.si.256") {
+      Value *Op0 = CI->getArgOperand(0);
+      unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
+      VectorType *VecTy = cast<VectorType>(CI->getType());
+      unsigned NumElts = VecTy->getNumElements();
+      
+      // Mask off the high bits of the immediate value; hardware ignores those.
+      Imm = Imm & 1;
+
+      // Get indexes for either the high half or low half of the input vector.
+      SmallVector<Constant*, 4> Idxs(NumElts);
+      for (unsigned i = 0; i != NumElts; ++i) {
+        unsigned Idx = Imm ? (i + NumElts) : i;
+        Idxs[i] = Builder.getInt32(Idx);
+      }
+
+      Value *UndefV = UndefValue::get(Op0->getType());
+      Rep = Builder.CreateShuffleVector(Op0, UndefV, ConstantVector::get(Idxs));
     } else {
       bool PD128 = false, PD256 = false, PS128 = false, PS256 = false;
       if (Name == "llvm.x86.avx.vpermil.pd.256")

diff --git a/llvm/test/CodeGen/X86/avx-intrinsics-x86-upgrade.ll b/llvm/test/CodeGen/X86/avx-intrinsics-x86-upgrade.ll
index a3908c6..e2f690b 100644 (file)
--- a/llvm/test/CodeGen/X86/avx-intrinsics-x86-upgrade.ll
+++ b/llvm/test/CodeGen/X86/avx-intrinsics-x86-upgrade.ll
@@ -36,6 +36,43 @@ define <8 x i32> @test_x86_avx_vinsertf128_si_256_2(<8 x i32> %a0, <4 x i32> %a1
 }
 declare <8 x i32> @llvm.x86.avx.vinsertf128.si.256(<8 x i32>, <4 x i32>, i8) nounwind readnone
 
+; We don't check any vextractf128 variant with immediate 0 because that's just a move. 
+
+define <2 x double> @test_x86_avx_vextractf128_pd_256_1(<4 x double> %a0) {
+; CHECK-LABEL:       test_x86_avx_vextractf128_pd_256_1: 
+; CHECK:             vextractf128 $1, %ymm0, %xmm0
+  %res = call <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double> %a0, i8 1)
+  ret <2 x double> %res
+}
+declare <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double>, i8) nounwind readnone
+
+define <4 x float> @test_x86_avx_vextractf128_ps_256_1(<8 x float> %a0) {
+; CHECK-LABEL:       test_x86_avx_vextractf128_ps_256_1: 
+; CHECK:             vextractf128 $1, %ymm0, %xmm0
+  %res = call <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float> %a0, i8 1)
+  ret <4 x float> %res
+}
+declare <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float>, i8) nounwind readnone
+
+define <4 x i32> @test_x86_avx_vextractf128_si_256_1(<8 x i32> %a0) {
+; CHECK-LABEL:    test_x86_avx_vextractf128_si_256_1: 
+; CHECK:          vextractf128 $1, %ymm0, %xmm0
+  %res = call <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32> %a0, i8 1)
+  ret <4 x i32> %res
+}
+declare <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32>, i8) nounwind readnone
+
+; Verify that high bits of the immediate are masked off. This should be the equivalent
+; of a vextractf128 $0 which should be optimized away, so just check that it's
+; not a vextractf128 of any kind.
+define <2 x double> @test_x86_avx_extractf128_pd_256_2(<4 x double> %a0) {
+; CHECK-LABEL:       test_x86_avx_extractf128_pd_256_2: 
+; CHECK-NOT:         vextractf128
+  %res = call <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double> %a0, i8 2)
+  ret <2 x double> %res
+}
+
+
 define <4 x double> @test_x86_avx_blend_pd_256(<4 x double> %a0, <4 x double> %a1) {
 ; CHECK-LABEL:       test_x86_avx_blend_pd_256: 
 ; CHECK:             vblendpd

diff --git a/llvm/test/CodeGen/X86/avx-intrinsics-x86.ll b/llvm/test/CodeGen/X86/avx-intrinsics-x86.ll
index 96d80ea..ecf8a19 100644 (file)
--- a/llvm/test/CodeGen/X86/avx-intrinsics-x86.ll
+++ b/llvm/test/CodeGen/X86/avx-intrinsics-x86.ll
@@ -2163,30 +2163,6 @@ define <8 x float> @test_x86_avx_vbroadcastf128_ps_256(i8* %a0) {
 declare <8 x float> @llvm.x86.avx.vbroadcastf128.ps.256(i8*) nounwind readonly
 
 
-define <2 x double> @test_x86_avx_vextractf128_pd_256(<4 x double> %a0) {
-  ; CHECK: vextractf128
-  %res = call <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double> %a0, i8 7) ; <<2 x double>> [#uses=1]
-  ret <2 x double> %res
-}
-declare <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double>, i8) nounwind readnone
-
-
-define <4 x float> @test_x86_avx_vextractf128_ps_256(<8 x float> %a0) {
-  ; CHECK: vextractf128
-  %res = call <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float> %a0, i8 7) ; <<4 x float>> [#uses=1]
-  ret <4 x float> %res
-}
-declare <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float>, i8) nounwind readnone
-
-
-define <4 x i32> @test_x86_avx_vextractf128_si_256(<8 x i32> %a0) {
-  ; CHECK: vextractf128
-  %res = call <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32> %a0, i8 7) ; <<4 x i32>> [#uses=1]
-  ret <4 x i32> %res
-}
-declare <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32>, i8) nounwind readnone
-
-
 define <4 x double> @test_x86_avx_vperm2f128_pd_256(<4 x double> %a0, <4 x double> %a1) {
   ; CHECK: vperm2f128
   %res = call <4 x double> @llvm.x86.avx.vperm2f128.pd.256(<4 x double> %a0, <4 x double> %a1, i8 7) ; <<4 x double>> [#uses=1]