[TOPI]fix scatterND large shape problem #12200
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Do you know why cc @tkonolige |
|
Looks like this is an issue in the TVM thread pool. I ran with TVM compiled with address sanitizer and I get the following error. @tqchen I think you have the most experience here. @yidawang wrote the code but I'm not sure if they are still active in the community. |
|
Would be good to know why. To see if it is an issue with particular thread pool, we can try to switch to openmp https://github.com/apache/tvm/blob/main/cmake/config.cmake#L178 |
|
It also errors with the openmp thread pool: |
|
That might hint something is wrong higher up? |
|
Following the hint of sanitizer that something related to stack overflow happens, there is a possible llvm issue found. The full dumped ll: scatter_nd_stackoverflow.ll.txt ; Function Attrs: noinline
define internal fastcc i32 @tvmgen_default_fused_scatter_nd_compute_(i8* noalias align 128 %0, i8* noalias nocapture readonly align 128 %1, i8* noalias align 128 %2, i8* noalias align 128 %3) unnamed_addr #2 {
entry:
call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 128 dereferenceable(1080000) %0, i8* noundef nonnull align 128 dereferenceable(1080000) %1, i64 1080000, i1 false)
br label %for_body_j
for_body_j: ; preds = %for_body_j, %entry
%j1 = phi i32 [ 0, %entry ], [ %13, %for_body_j ]
%4 = alloca %closure_loop_parallel_k, align 8
%5 = getelementptr inbounds %closure_loop_parallel_k, %closure_loop_parallel_k* %4, i64 0, i32 0
store i8* %0, i8** %5, align 8
%6 = getelementptr inbounds %closure_loop_parallel_k, %closure_loop_parallel_k* %4, i64 0, i32 1
store i8* %2, i8** %6, align 8
%7 = getelementptr inbounds %closure_loop_parallel_k, %closure_loop_parallel_k* %4, i64 0, i32 2
store i32 %j1, i32* %7, align 8
%8 = getelementptr inbounds %closure_loop_parallel_k, %closure_loop_parallel_k* %4, i64 0, i32 3
store i8* %3, i8** %8, align 8
%9 = load i32 (i32 (i32, %0*, i8*)*, i8*, i32)*, i32 (i32 (i32, %0*, i8*)*, i8*, i32)** @__TVMBackendParallelLaunch, align 8, !tbaa !5
%10 = bitcast %closure_loop_parallel_k* %4 to i8*
%11 = call i32 %9(i32 (i32, %0*, i8*)* nonnull @__tvm_parallel_lambda, i8* nonnull %10, i32 0)
%12 = icmp ne i32 %11, 0
%13 = add nuw nsw i32 %j1, 1
%exitcond.not = icmp eq i32 %13, 270000
%or.cond = select i1 %12, i1 true, i1 %exitcond.not
br i1 %or.cond, label %common.ret, label %for_body_j, !prof !142
common.ret: ; preds = %for_body_j
ret i32 %11
}The alloca A quick experimental modification to https://github.com/apache/tvm/blob/main/src/target/llvm/codegen_cpu.cc#L645 make segfault disappear in my environment. auto cur_pt = builder_->GetInsertBlock();
builder_->SetInsertPoint(&(*(function_->getEntryBlock().getFirstInsertionPt())));
llvm::Value* cvalue = builder_->CreateAlloca(ctype, ConstInt32(1)); // alloca at function begin
builder_->SetInsertPoint(cur_pt );So I think we could either
|
|
@wrongtest-intellif This sounds like a bug in StorageRewrite and/or PlanAndUpdateBufferAllocationLocation where the allocation should be lifted outside the loop but isn't. |
Emmmm I think it is not. https://github.com/apache/tvm/blob/main/src/target/llvm/codegen_cpu.cc#L645 is an dedicated usage of llvm alloca for thread pool, not from tvm level allocation. |
|
cc @tkonolige @kparzysz-quic Hi~ could you kindly review the fix again? |
added 2 commits
August 15, 2022 14:23
tkonolige
reviewed
Aug 16, 2022
src/target/llvm/codegen_cpu.cc
Outdated
| @@ -642,7 +642,16 @@ CodeGenLLVM::TypedPointer CodeGenCPU::PackClosureData(const Array<Var>& vfields, | |||
| } | |||
| llvm::StructType* ctype = struct_name.size() ? llvm::StructType::create(fields, struct_name) | |||
| : llvm::StructType::create(fields); | |||
| llvm::Value* cvalue = builder_->CreateAlloca(ctype, ConstInt32(1)); | |||
| // create ctype alloca at function entry | |||
There was a problem hiding this comment.
Can you add an explanation of why we allocate at the function entry.
|
@kparzysz-quic Looks like CI has a error. Can you help check it ? thanks. |
|
@tvm-bot rerun |
1 similar comment
|
@tvm-bot rerun |
|
@driazati Could you take a look at what happened here with the bot? We couldn't get it to rerun the CI. |
|
@masahi Could you review the fix again? thanks. |
masahi
approved these changes
Aug 19, 2022
xinetzone
pushed a commit
to daobook/tvm
that referenced
this pull request
Nov 25, 2022
* fix scatterND large shape problem * fix thread pool alloca * add scatternd unit test * update with comment * Empty Co-authored-by: wrongtest <wrongtest0@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
test case:
when i run the case, I will get a Segmentation fault.
Then I remove the
kindparam, it can run success.I don't know whether the
kindparam use 'unroll' is better.