mirror of
https://github.com/wolfpld/tracy.git
synced 2026-07-31 18:39:24 +00:00
12dc23f67e
Tests whether CUPTI recycles graphId values after cudaGraphExecDestroy, which would be the only scenario where the graphLaunchCache in TracyCUDA could serve stale entries for a non-matching exec handle. Result (H100, CUDA 12, CUPTI): graphId is a monotonically increasing counter that is never recycled. 22 create/instantiate/launch/destroy cycles produced unique IDs ranging from 2 to 65 (incrementing by 3 per cycle — one unit per node created during graph construction). This confirms that the stale-cache concern raised in code review is not a real risk in practice: two distinct exec handles always have distinct graphIds. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
50 lines
1.5 KiB
Makefile
50 lines
1.5 KiB
Makefile
TRACY_PUBLIC := ../../public
|
|
NVCC := nvcc
|
|
CXX := g++
|
|
CUPTI_INC := /usr/local/cuda/include
|
|
CUPTI_LIB := /usr/local/cuda/lib64
|
|
|
|
TRACY_SRCS := $(TRACY_PUBLIC)/TracyClient.cpp
|
|
INCLUDES := -I$(TRACY_PUBLIC) -I$(CUPTI_INC)
|
|
LIBS := -L$(CUPTI_LIB) -lcuda -lcupti -lpthread -ldl
|
|
|
|
CXXFLAGS_REL := -O2 -DTRACY_ENABLE
|
|
CXXFLAGS_DBG := -g -O0 -DTRACY_ENABLE
|
|
NVCCFLAGS_REL := -arch=native -O2 -DTRACY_ENABLE
|
|
NVCCFLAGS_DBG := -arch=native -g -O0 -DTRACY_ENABLE
|
|
|
|
.PHONY: all debug investigate investigate2 clean
|
|
|
|
all: repro
|
|
|
|
debug: repro_debug
|
|
|
|
investigate: test_corr_reuse
|
|
|
|
investigate2: test_graphid_recycle
|
|
|
|
# Release build
|
|
repro: repro.cu tracy_client.o
|
|
$(NVCC) $(NVCCFLAGS_REL) $(INCLUDES) -o $@ $< tracy_client.o $(LIBS)
|
|
|
|
tracy_client.o: $(TRACY_SRCS)
|
|
$(CXX) $(CXXFLAGS_REL) $(INCLUDES) -c -o $@ $<
|
|
|
|
# Debug build (asserts enabled, no NDEBUG)
|
|
repro_debug: repro.cu tracy_client_debug.o
|
|
$(NVCC) $(NVCCFLAGS_DBG) $(INCLUDES) -o $@ $< tracy_client_debug.o $(LIBS)
|
|
|
|
tracy_client_debug.o: $(TRACY_SRCS)
|
|
$(CXX) $(CXXFLAGS_DBG) $(INCLUDES) -c -o $@ $<
|
|
|
|
# Investigation: correlationId uniqueness per graph launch (no Tracy dependency)
|
|
test_corr_reuse: test_corr_reuse.cu
|
|
$(NVCC) $(NVCCFLAGS_REL) $(INCLUDES) -o $@ $< $(LIBS)
|
|
|
|
# Investigation: does CUPTI recycle graphId values after cudaGraphExecDestroy?
|
|
test_graphid_recycle: test_graphid_recycle.cu
|
|
$(NVCC) $(NVCCFLAGS_REL) $(INCLUDES) -o $@ $< $(LIBS)
|
|
|
|
clean:
|
|
rm -f repro repro_debug test_corr_reuse test_graphid_recycle tracy_client.o tracy_client_debug.o
|