Pass pipeline logging instrumentation#4182
Conversation
Signed-off-by: Adam Geller <adgeller@nvidia.com>
Signed-off-by: Adam Geller <adgeller@nvidia.com>
Signed-off-by: Adam Geller <adgeller@nvidia.com>
Signed-off-by: Adam Geller <adgeller@nvidia.com>
Signed-off-by: Adam Geller <adgeller@nvidia.com>
Signed-off-by: Adam Geller <adgeller@nvidia.com>
|
CUDA Quantum Docs Bot: A preview of the documentation can be found here. |
…y on source file Signed-off-by: Adam Geller <adgeller@nvidia.com>
Signed-off-by: Adam Geller <adgeller@nvidia.com>
|
CUDA Quantum Docs Bot: A preview of the documentation can be found here. |
Signed-off-by: Adam Geller <adgeller@nvidia.com>
| ~PipelineRecorder() override { flushToLog(); } | ||
|
|
||
| void runBeforePass(mlir::Pass *pass, mlir::Operation *op) override { | ||
| records.push_back( |
There was a problem hiding this comment.
A question. Is this records vector threadsafe? Seems like we will need to ensure that disableMultithreading is set before calling pm.run()?
| @@ -973,6 +975,7 @@ jitCode(ImplicitLocOpBuilder &builder, ExecutionEngine *jit, | |||
| pm.addNestedPass<func::FuncOp>(createCSEPass()); | |||
| pm.addPass(cudaq::opt::createConvertToQIR()); | |||
| pm.addPass(createCanonicalizerPass()); | |||
| cudaq_internal::maybeLogPassPipeline(pm, "cudaq::builder JIT"); | |||
There was a problem hiding this comment.
Seems like we should have different labels for these so that it will be helpful to distinguish?
| cudaq.reset_target() | ||
| yield | ||
| cudaq.reset_target() |
There was a problem hiding this comment.
Why are we calling reset_target() before and after the test? Should it be just after the test?
| @pytest.mark.parametrize("target,target_kwargs,supports_run", TARGET_CONFIGS) | ||
| def test_pipeline_logging_decorator(tmp_path, monkeypatch, target, | ||
| target_kwargs, supports_run): | ||
| log_path = tmp_path / "pipeline.jsonl" |
There was a problem hiding this comment.
Should we move out this hard coded value in a constant outside?
|
|
||
|
|
||
| def test_pipeline_logging_builder_default(tmp_path, monkeypatch): | ||
| log_path = tmp_path / "pipeline.jsonl" |
There was a problem hiding this comment.
same here. Wondering if the same log_file can be reused for these tests?
| tm.setEnabled(cudaq::isTimingTagEnabled(cudaq::TIMING_JIT_PASSES)); | ||
| auto timingScope = tm.getRootScope(); // starts the timer | ||
| pm.enableTiming(timingScope); // do this right before pm.run | ||
| cudaq_internal::maybeLogPassPipeline(pm, name + ":client"); |
| mlir::PassManager pm(ctx); | ||
| // For now, the server side expects full-QIR. | ||
| opt::addAOTPipelineConvertToQIR(pm); | ||
| cudaq_internal::maybeLogPassPipeline(pm, name + ":aot-qir"); |
| std::string jsonEscape(llvm::StringRef s) { | ||
| std::string out; | ||
| out.reserve(s.size()); | ||
| for (char c : s) { | ||
| switch (c) { | ||
| case '"': | ||
| out += "\\\""; | ||
| break; | ||
| case '\\': | ||
| out += "\\\\"; | ||
| break; | ||
| case '\n': | ||
| out += "\\n"; | ||
| break; | ||
| case '\r': | ||
| out += "\\r"; | ||
| break; | ||
| case '\t': | ||
| out += "\\t"; | ||
| break; | ||
| default: | ||
| out += c; | ||
| break; | ||
| } | ||
| } | ||
| return out; | ||
| } |
There was a problem hiding this comment.
I think we can optimize(instead of using +=) this function, we can collect all characters in some string and then flush it once at the end.
| std::string record = "{\"type\":\"executed\",\"label\":\"" + | ||
| jsonEscape(label) + "\",\"passes\":["; | ||
| for (size_t i = 0; i < records.size(); ++i) { | ||
| if (i > 0) | ||
| record += ","; | ||
| record += "{\"pass\":\"" + jsonEscape(records[i].passArg) + | ||
| "\",\"op\":\"" + jsonEscape(records[i].opName) + "\""; | ||
| if (records[i].failed) | ||
| record += ",\"failed\":true"; | ||
| record += "}"; | ||
| } | ||
| record += "]}"; | ||
| appendToLogFile(logPath, record); |
There was a problem hiding this comment.
Should we use llvm::raw_string_ostream here? This will avoid the creation of intermediate std::string (with +). We can also get rid of \".
| std::string record = "{\"type\":\"configured\",\"label\":\"" + | ||
| jsonEscape(effectiveLabel) + "\",\"pipeline\":\"" + | ||
| jsonEscape(pipeline) + "\"}"; |
|
CUDA Quantum Docs Bot: A preview of the documentation can be found here. |
Signed-off-by: Adam Geller <adgeller@nvidia.com>
3 participants
Adds instrumentation to log pass pipelines when enabled. Use
CUDAQ_PIPELINE_LOG=path/to/log/fileto enable logging. Currently, this will only log JIT passes.