Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[WIP] High-Level (Coase grain) parallelization with ONNXParallelOp and ONNXForkOp based on OpenMP #2756

Draft
wants to merge 37 commits into
base: main
Choose a base branch
from
Draft
Changes from 1 commit
Commits
Show all changes
37 commits
Select commit Hold shift + click to select a range
c9f68fd
Added ONNXParallelOp and ONNXForkOp
imaihal Mar 15, 2024
8f87a9c
Add lowering pass for ONNXParallelOp and ONNXForkOp
imaihal Mar 15, 2024
9234f25
Remove SCFToCF Pass
imaihal Mar 15, 2024
98aef8a
New files for ONNXParallelOp and ONNXForkOp
imaihal Mar 15, 2024
796aa80
Revert: enable deallocation op as ilegal op
imaihal Mar 18, 2024
84f9f0d
Clean up by creating a function for moving alloc op
imaihal Mar 21, 2024
f24dac6
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Mar 21, 2024
e6591f0
Update KrnlToAffine pass to support scf::ifOp in body of KrnlIterateOp.
imaihal Mar 26, 2024
0db91dc
Update comments.
imaihal Mar 26, 2024
22cbc08
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Mar 26, 2024
f24c9a6
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Mar 27, 2024
c4f0a3b
Dynamic dim support v1.
imaihal Mar 29, 2024
f7fc4ca
Fix static dim case.
imaihal Apr 1, 2024
38161c1
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Apr 1, 2024
c8f8075
Add KrnlRegionOp in region of scf::ifOp.
imaihal Apr 2, 2024
0fbfa9d
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Apr 2, 2024
8d10b0b
Update dynamic dim support.
imaihal Apr 3, 2024
ed5fec1
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Apr 3, 2024
9077e7d
Update condition to apply unstickStickRemoval optimization
imaihal Apr 4, 2024
91b6f46
Update comments
imaihal Apr 4, 2024
0e06aec
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Apr 4, 2024
c73d8ad
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Apr 5, 2024
d934273
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Apr 12, 2024
9a45f0f
Added lit test for shape inference.
imaihal Apr 12, 2024
dad1f99
Fix lit test.
imaihal Apr 13, 2024
6d111a7
Update dynamic dim support
imaihal Apr 17, 2024
44396c4
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Apr 17, 2024
37e12b2
Fix insertion of KrnlRegionOp.
imaihal Apr 22, 2024
c5e7663
Add shapeHelpr for ONNXParallelOp and ONNXForkOp.
imaihal Apr 25, 2024
2f31a16
Add dimAnalysis for ONNXParallelOp and ONNXForkOp.
imaihal Apr 25, 2024
b70fbcc
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Apr 25, 2024
b6ee509
Merge branch 'main' into parallelop_forkop_with_omp
imaihal May 1, 2024
06cc16c
Merge branch 'main' into parallelop_forkop_with_omp
imaihal May 27, 2024
2bcabe8
Fix inserting regionOp in iterateOp.
imaihal May 29, 2024
c837809
Add scripts for rewriting a model for operator-level parallelization.
imaihal Jun 6, 2024
53805d9
Merge branch 'main' into parallelop_forkop_with_omp
imaihal Jun 6, 2024
17ac56d
black format
imaihal Jun 6, 2024
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Update comments.
Signed-off-by: Haruki Imai <imaihal@jp.ibm.com>
imaihal committed Mar 26, 2024
commit 0db91dc2b05d8ac1288ad1cd694d094fb79b1818
9 changes: 4 additions & 5 deletions src/Conversion/ONNXToKrnl/Additional/Parallel.cpp
Original file line number Diff line number Diff line change
@@ -94,9 +94,7 @@ struct ONNXParallelOpLowering : public OpConversionPattern<ONNXParallelOp> {
if (!isa<ONNXYieldOp>(yieldOp))
return failure();

// ForkOps in ParallelOp region
// Move alloc op for return values and replace uses with them.
// Still not replace ForkOps yet.
// Move alloc ops included ForkOps
SmallVector<ONNXForkOp, 4> forkOps;
for (Operation &bOp : bodyBlock.getOperations()) {
if (auto forkOp = dyn_cast<ONNXForkOp>(bOp)) {
@@ -115,7 +113,8 @@ struct ONNXParallelOpLowering : public OpConversionPattern<ONNXParallelOp> {
if (failed(moveAllocOpBeforeAndReplaceAllUses(rewriter, op, yieldOp)))
return failure();

// 2. Create KrnlIterateOp and replace Parallel Op with it.
// Create KrnlIterateOp and replace ParallelOp with it.
rewriter.setInsertionPoint(op);
std::vector<Value> loop;
defineLoops(rewriter, loc, loop, 1);
krnl::KrnlIterateOperandPack pack(rewriter, loop);
@@ -130,7 +129,7 @@ struct ONNXParallelOpLowering : public OpConversionPattern<ONNXParallelOp> {
rewriter.eraseOp(yieldOp);
rewriter.inlineBlockBefore(&bodyBlock, iterationBlock.getTerminator());

// 3. Create SCFIfOp and replace ForkOp with it.
// Create SCFIfOp and replace ForkOp with it.
int64_t id = 0;
for (auto forkOp : forkOps) {
rewriter.setInsertionPoint(forkOp);