Magic-Mirror/llvm-capstone
1
0
Fork 0
mirror of https://github.com/capstone-engine/llvm-capstone.git synced 2024-11-24 06:10:12 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[InstCombine][X86] Tweak generic expansion of PACKSS/PACKUS to shuffle then truncate. NFCI.

Browse Source 
This has no effect on constant folding but will be useful when we expand non-saturating PACKSS/PACKUS intrinsics.

llvm-svn: 359191
This commit is contained in:
Simon Pilgrim 2019-04-25 13:51:57 +00:00
parent 0fc09d0d25
commit 48a3b54572
1 changed files with 4 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
View File

@ -593,12 +593,7 @@ static Value *simplifyX86pack(IntrinsicInst &II,
Arg0 = Builder.CreateSelect(Builder.CreateICmpSGT(Arg0, MaxC), MaxC, Arg0); Arg0 = Builder.CreateSelect(Builder.CreateICmpSGT(Arg0, MaxC), MaxC, Arg0);
Arg1 = Builder.CreateSelect(Builder.CreateICmpSGT(Arg1, MaxC), MaxC, Arg1); Arg1 = Builder.CreateSelect(Builder.CreateICmpSGT(Arg1, MaxC), MaxC, Arg1);
// Truncate clamped args to dst size. // Shuffle clamped args together at the lane level.
auto *TruncTy = VectorType::get(ResTy->getScalarType(), NumSrcElts);
Arg0 = Builder.CreateTrunc(Arg0, TruncTy);
Arg1 = Builder.CreateTrunc(Arg1, TruncTy);
// Shuffle args together at the lane level.
SmallVector<unsigned, 32> PackMask; SmallVector<unsigned, 32> PackMask;
for (unsigned Lane = 0; Lane != NumLanes; ++Lane) { for (unsigned Lane = 0; Lane != NumLanes; ++Lane) {
for (unsigned Elt = 0; Elt != NumSrcEltsPerLane; ++Elt) for (unsigned Elt = 0; Elt != NumSrcEltsPerLane; ++Elt)
@ -606,8 +601,10 @@ static Value *simplifyX86pack(IntrinsicInst &II,
for (unsigned Elt = 0; Elt != NumSrcEltsPerLane; ++Elt) for (unsigned Elt = 0; Elt != NumSrcEltsPerLane; ++Elt)
PackMask.push_back(Elt + (Lane * NumSrcEltsPerLane) + NumSrcElts); PackMask.push_back(Elt + (Lane * NumSrcEltsPerLane) + NumSrcElts);
} }
auto *Shuffle = Builder.CreateShuffleVector(Arg0, Arg1, PackMask);
return Builder.CreateShuffleVector(Arg0, Arg1, PackMask); // Truncate to dst size.
return Builder.CreateTrunc(Shuffle, ResTy);
} }
// Replace X86-specific intrinsics with generic floor-ceil where applicable. // Replace X86-specific intrinsics with generic floor-ceil where applicable.