gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[OPTIMIZER] Better pipeline tests (#660)

Browse Source
This commit is contained in:
Da Yan
2022-09-15 14:26:40 +08:00
committed by GitHub
parent 297d27e1c8
commit 2e08450c80
2 changed files with 56 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
lib/Dialect/TritonGPU/Transforms/Pipeline.cpp
View File

@@ -347,16 +347,13 @@ void LoopPipeliner::emitPrologue() {
// async.wait & extract_slice
Operation *asyncWait = builder.create<triton::gpu::AsyncWaitOp>(
loads[0].getLoc(), loads.size() * (numStages - 2));
loopIterIdx = builder.create<arith::ConstantIntOp>(iv.getLoc(), 0, 32);
for (Value loadOp : loads) {
Value extractSlice = builder.create<triton::gpu::ExtractSliceOp>(
loadOp.getLoc(), loadsMapping[loadOp].getType(),
loadStageBuffer[loadOp][numStages - 1],
builder.create<arith::ConstantIntOp>(loadOp.getLoc(), 0, 32),
/*axis*/ 0);
loadStageBuffer[loadOp][numStages - 1], loopIterIdx, /*axis*/ 0);
loadsExtract[loadOp] = extractSlice;
}
loopIterIdx = builder.create<arith::ConstantIntOp>(iv.getLoc(), 0, 32);
}
scf::ForOp LoopPipeliner::createNewForOp() {