[MLIR][Timing] Fix timing report of nested pass pipeline and negative Rest field when using -mlir-timing (Try to close #169443)

[Xubaidu]

This PR tries to close [MLIR][Timing] Nested pipeline is displayed at top level and show negative Rest field when using -mlir-timing

Here is the timing report of the inliner pass w/o and w/ this change:

# w/o this change
build/bin/mlir-opt -inline mlir/test/Transforms/inlining-repeated-use.mlir -mlir-timing -o 1.txt
===-------------------------------------------------------------------------===
                         ... Execution time report ...
===-------------------------------------------------------------------------===
  Total Execution Time: 0.0009 seconds

  ----Wall Time----  ----Name----
    0.0003 ( 40.1%)  Parser
    0.0003 ( 37.5%)  Inliner
    0.0000 (  1.3%)    (A) CallGraph
    0.0001 (  7.9%)  'func.func' Pipeline
    0.0001 (  7.7%)    Canonicalizer
    0.0002 ( 19.9%)  Output
   -0.0000 ( -5.5%)  Rest
    0.0009 (100.0%)  Total

# w/ this change
build/bin/mlir-opt -inline mlir/test/Transforms/inlining-repeated-use.mlir -mlir-timing -o 1.txt
===-------------------------------------------------------------------------===
                         ... Execution time report ...
===-------------------------------------------------------------------------===
  Total Execution Time: 0.0007 seconds

  ----Wall Time----  ----Name----
    0.0003 ( 46.5%)  Parser
    0.0003 ( 37.1%)  Inliner
    0.0000 (  1.1%)    (A) CallGraph
    0.0001 (  7.7%)    'func.func' Pipeline
    0.0001 (  7.4%)      Canonicalizer
    0.0001 ( 12.2%)  Output
    0.0000 (  4.2%)  Rest
    0.0007 (100.0%)  Total

Also take mlir/test/Pass/dynamic-pipeline-nested.mlir as an example:

# w/o this change
build/install/bin/mlir-opt mlir/test/Pass/dynamic-pipeline-nested.mlir -pass-pipeline='builtin.module(builtin.module(test-dynamic-pipeline{op-name=inner_mod1 dynamic-pipeline=cse}),inline)'  --mlir-disable-threading -mlir-timing > /dev/null
Dynamic execute 'cse' on builtin.module
===-------------------------------------------------------------------------===
                         ... Execution time report ...
===-------------------------------------------------------------------------===
  Total Execution Time: 0.0005 seconds

  ----Wall Time----  ----Name----
    0.0003 ( 58.3%)  Parser
    0.0000 (  4.7%)  'builtin.module' Pipeline
    0.0000 (  4.6%)    TestDynamicPipelinePass
    0.0000 (  1.6%)  'builtin.module' Pipeline
    0.0000 (  1.5%)    CSE
    0.0000 (  0.1%)      (A) DominanceInfo
    0.0001 ( 22.7%)  Inliner
    0.0000 (  1.3%)    (A) CallGraph
    0.0000 (  3.5%)  'func.func' Pipeline
    0.0000 (  3.3%)    Canonicalizer
    0.0000 (  6.6%)  Output
    0.0000 (  1.9%)  Rest
    0.0005 (100.0%)  Total

# w/ this change
Dynamic execute 'cse' on builtin.module
===-------------------------------------------------------------------------===
                         ... Execution time report ...
===-------------------------------------------------------------------------===
  Total Execution Time: 0.0023 seconds

  ----Wall Time----  ----Name----
    0.0017 ( 71.5%)  Parser
    0.0001 (  2.6%)  'builtin.module' Pipeline
    0.0001 (  2.6%)    (anonymous namespace)::TestDynamicPipelinePass
    0.0000 (  1.5%)      'builtin.module' Pipeline
    0.0000 (  1.5%)        CSE
    0.0000 (  0.0%)          (A) DominanceInfo
    0.0003 ( 14.9%)  Inliner
    0.0000 (  1.0%)    (A) CallGraph
    0.0001 (  3.2%)    'func.func' Pipeline
    0.0001 (  3.2%)      Canonicalizer
    0.0001 (  4.8%)  Output
    0.0001 (  5.6%)  Rest
    0.0023 (100.0%)  Total

[github-actions]

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

[llvmbot]

@llvm/pr-subscribers-mlir

@llvm/pr-subscribers-mlir-core

Author: ChenyangXu (Xubaidu)

Changes

This PR tries to close [MLIR][Timing] Nested pipeline is displayed at top level and show negative Rest field when using -mlir-timing

---

Full diff: https://github.com/llvm/llvm-project/pull/169615.diff

1 Files Affected:

• (modified) mlir/lib/Pass/PassTiming.cpp (+5-2)

diff --git a/mlir/lib/Pass/PassTiming.cpp b/mlir/lib/Pass/PassTiming.cpp
index dab56f09a72eb..8b78a99069cc9 100644
--- a/mlir/lib/Pass/PassTiming.cpp
+++ b/mlir/lib/Pass/PassTiming.cpp
@@ -90,6 +90,9 @@ struct PassTiming : public PassInstrumentation {
     auto &activeTimers = activeThreadTimers[tid];
     auto &parentScope = activeTimers.empty() ? rootScope : activeTimers.back();
 
+    // Record the parent information for this pass so nested pipelines can find it,
+    // regardless of whether this is an adaptor pass or a regular pass.
+    parentTimerIndices[{tid, pass}] = activeTimers.size();
     if (auto *adaptor = dyn_cast<OpToOpPassAdaptor>(pass)) {
       parentTimerIndices[{tid, pass}] = activeTimers.size();
       auto scope =
@@ -107,8 +110,8 @@ struct PassTiming : public PassInstrumentation {
 
   void runAfterPass(Pass *pass, Operation *) override {
     auto tid = llvm::get_threadid();
-    if (isa<OpToOpPassAdaptor>(pass))
-      parentTimerIndices.erase({tid, pass});
+    // Erase the parent index mapping for this pass (adaptor or not).
+    parentTimerIndices.erase({tid, pass});
     auto &activeTimers = activeThreadTimers[tid];
     assert(!activeTimers.empty() && "expected active timer");
     activeTimers.pop_back();

[joker-eph]

Isn't this line redundant now?

[Xubaidu]

Removed the redundant code now. Please take a look again.

[joker-eph]

That seems fine, how could this be tested though?

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[Xubaidu]

That seems fine, how could this be tested though?

@joker-eph I added two tests in the test file but I am not sure if they are sufficient. Would you mind adding more reviewers of dynamic pass pipeline and MLIR core to review it?

[joker-eph]

This test is passing at HEAD apparently, so may not be testing what you think it is?