[AVX-512] Remove vextract intrinsics and autoupgrade to native shufflevectors. This unfortunately generates some really terrible code without VLX support due to v2i1 and v4i1 not being legal.

Hopefully we can improve that in future patches.

llvm-svn: 290863

1 files changed, 16 insertions, 10 deletions

diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index 2d9d0f95efa..c6c6f0055d6 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -344,6 +344,7 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
          Name == "avx2.vinserti128" || // Added in 3.7
          Name.startswith("avx.vextractf128.") || // Added in 3.7
          Name == "avx2.vextracti128" || // Added in 3.7
+         Name.startswith("avx512.mask.vextract") || // Added in 4.0
          Name.startswith("sse4a.movnt.") || // Added in 3.9
          Name.startswith("avx.movnt.") || // Added in 3.2
          Name.startswith("avx512.storent.") || // Added in 3.9
@@ -1188,23 +1189,28 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
         Idxs[i] = Imm ? (i + NumElts / 2) : i;
       Rep = Builder.CreateShuffleVector(Op0, Rep, Idxs);
     } else if (IsX86 && (Name.startswith("avx.vextractf128.") ||
-                         Name == "avx2.vextracti128")) {
+                         Name == "avx2.vextracti128" ||
+                         Name.startswith("avx512.mask.vextract"))) {
       Value *Op0 = CI->getArgOperand(0);
       unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
-      VectorType *VecTy = cast<VectorType>(CI->getType());
-      unsigned NumElts = VecTy->getNumElements();
+      unsigned DstNumElts = CI->getType()->getVectorNumElements();
+      unsigned SrcNumElts = Op0->getType()->getVectorNumElements();
+      unsigned Scale = SrcNumElts / DstNumElts;
 
       // Mask off the high bits of the immediate value; hardware ignores those.
-      Imm = Imm & 1;
+      Imm = Imm % Scale;
 
-      // Get indexes for either the high half or low half of the input vector.
-      SmallVector<uint32_t, 4> Idxs(NumElts);
-      for (unsigned i = 0; i != NumElts; ++i) {
-        Idxs[i] = Imm ? (i + NumElts) : i;
+      // Get indexes for the subvector of the input vector.
+      SmallVector<uint32_t, 8> Idxs(DstNumElts);
+      for (unsigned i = 0; i != DstNumElts; ++i) {
+        Idxs[i] = i + (Imm * DstNumElts);
       }
+      Rep = Builder.CreateShuffleVector(Op0, Op0, Idxs);
 
-      Value *UndefV = UndefValue::get(Op0->getType());
-      Rep = Builder.CreateShuffleVector(Op0, UndefV, Idxs);
+      // If the intrinsic has a mask operand, handle that.
+      if (CI->getNumArgOperands() == 4)
+        Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
+                            CI->getArgOperand(2));
     } else if (!IsX86 && Name == "stackprotectorcheck") {
       Rep = nullptr;
     } else if (IsX86 && (Name.startswith("avx512.mask.perm.df.") ||