Merge branch 'amd-master' into rkebichi-trace-compare
This commit is contained in:
@@ -1,8 +1,7 @@
|
||||
ROOT_PATH = ../..
|
||||
LIB_PATH = $(ROOT_PATH)/build
|
||||
ROC_LIBS = -L$(LIB_PATH) -lroctracer64 -lroctx64
|
||||
HSA_KMT_INC_PATH ?= /opt/rocm/include
|
||||
export LD_LIBRARY_PATH=$(LIB_PATH)
|
||||
ROC_LIBS = -Wl,--rpath,${LIB_PATH} $(LIB_PATH)/libroctracer64.so $(LIB_PATH)/libroctx64.so
|
||||
ROCM_INC_PATH ?= /opt/rocm/include
|
||||
HIP_VDI ?= 0
|
||||
ITERATIONS ?= 100
|
||||
|
||||
@@ -12,27 +11,37 @@ ifeq (,$(HIP_PATH))
|
||||
endif
|
||||
|
||||
HIPCC=$(HIP_PATH)/bin/hipcc
|
||||
|
||||
TARGET=hcc
|
||||
|
||||
SOURCES = MatrixTranspose.cpp
|
||||
OBJECTS = $(SOURCES:.cpp=.o)
|
||||
|
||||
EXECUTABLE=./MatrixTranspose
|
||||
OBJECTS = MatrixTranspose.o
|
||||
FLAGS =-g -I$(ROOT_PATH) -I$(ROOT_PATH)/inc -I${ROCM_INC_PATH}/hsa -I${ROCM_INC_PATH} -DLOCAL_BUILD=1 -DHIP_VDI=${HIP_VDI} -DITERATIONS=$(ITERATIONS) -DAMD_INTERNAL_BUILD=1
|
||||
|
||||
ifeq ($(C_TEST), 1)
|
||||
COMP=gcc
|
||||
SOURCES = MatrixTranspose.c
|
||||
FLAGS += -DHIP_TEST=0 -D__HIP_PLATFORM_HCC__=1 -I${ROCM_INC_PATH}/hcc
|
||||
else
|
||||
COMP=$(HIPCC)
|
||||
SOURCES = MatrixTranspose.cpp
|
||||
FLAGS += -DHIP_TEST=1
|
||||
endif
|
||||
ifeq ($(MGPU_TEST), 1)
|
||||
FLAGS += -DMGPU_TEST=1
|
||||
endif
|
||||
|
||||
.PHONY: test
|
||||
|
||||
|
||||
all: clean $(EXECUTABLE)
|
||||
|
||||
CXXFLAGS =-g -I$(ROOT_PATH) -I$(ROOT_PATH)/inc -I${HSA_KMT_INC_PATH} -DLOCAL_BUILD=1 -DHIP_VDI=${HIP_VDI} -DITERATIONS=$(ITERATIONS)
|
||||
CXX=$(HIPCC)
|
||||
$(OBJECTS): $(SOURCES)
|
||||
$(COMP) $(FLAGS) -c -o $@ $<
|
||||
|
||||
$(EXECUTABLE): $(OBJECTS)
|
||||
$(HIPCC) $(OBJECTS) -o $@ $(ROC_LIBS)
|
||||
|
||||
test: $(EXECUTABLE)
|
||||
$(EXECUTABLE)
|
||||
LD_PRELOAD="$(LIB_PATH)/libkfdwrapper64.so librocprofiler64.so" $(EXECUTABLE)
|
||||
|
||||
clean:
|
||||
rm -f $(EXECUTABLE)
|
||||
|
||||
Szimbolikus hivatkozás
@@ -0,0 +1 @@
|
||||
MatrixTranspose.cpp
|
||||
@@ -238,20 +238,22 @@ void api_callback(
|
||||
fprintf(stdout, "<rocTX \"%s\">\n", data->args.message);
|
||||
return;
|
||||
}
|
||||
|
||||
if (domain == ACTIVITY_DOMAIN_KFD_API) {
|
||||
const kfd_api_data_t* data = (const kfd_api_data_t*)(callback_data);
|
||||
fprintf(stdout, "<%s id(%u)\tcorrelation_id(%lu) %s pid(%u) tid(%u)> \n",
|
||||
fprintf(stdout, "<%s id(%u)\tcorrelation_id(%lu) %s> \n",
|
||||
roctracer_op_string(ACTIVITY_DOMAIN_KFD_API, cid, 0),
|
||||
cid,
|
||||
data->correlation_id,
|
||||
(data->phase == ACTIVITY_API_PHASE_ENTER) ? "on-enter" : "on-exit",
|
||||
GetPid(),
|
||||
GetTid()
|
||||
);
|
||||
return;
|
||||
}
|
||||
const hip_api_data_t* data = (const hip_api_data_t*)(callback_data);
|
||||
fprintf(stdout, "<%s id(%u)\tcorrelation_id(%lu) %s> ",
|
||||
roctracer_op_string(ACTIVITY_DOMAIN_HIP_API, cid, 0),
|
||||
cid,
|
||||
data->correlation_id,
|
||||
(data->phase == ACTIVITY_API_PHASE_ENTER) ? "on-enter" : "on-exit");
|
||||
if (data->phase == ACTIVITY_API_PHASE_ENTER) {
|
||||
switch (cid) {
|
||||
case HIP_API_ID_hipMemcpy:
|
||||
@@ -326,7 +328,7 @@ void activity_callback(const char* begin, const char* end, void* arg) {
|
||||
record->end_ns
|
||||
);
|
||||
if ((record->domain == ACTIVITY_DOMAIN_HIP_API) || (record->domain == ACTIVITY_DOMAIN_KFD_API)) {
|
||||
fprintf(stdout, " process_id(%u) thread_id(%u)\n",
|
||||
fprintf(stdout, " process_id(%u) thread_id(%u)",
|
||||
record->process_id,
|
||||
record->thread_id
|
||||
);
|
||||
@@ -335,7 +337,13 @@ void activity_callback(const char* begin, const char* end, void* arg) {
|
||||
record->device_id,
|
||||
record->queue_id
|
||||
);
|
||||
if (record->op == HIP_OP_ID_COPY) fprintf(stdout, " bytes(0x%zx)\n", record->bytes);
|
||||
if (record->op == HIP_OP_ID_COPY) fprintf(stdout, " bytes(0x%zx)", record->bytes);
|
||||
} else if (record->domain == ACTIVITY_DOMAIN_HSA_OPS) {
|
||||
fprintf(stdout, " se(%u) cycle(%lu) pc(%lx)",
|
||||
record->pc_sample.se,
|
||||
record->pc_sample.cycle,
|
||||
record->pc_sample.pc
|
||||
);
|
||||
} else if (record->domain == ACTIVITY_DOMAIN_EXT_API) {
|
||||
fprintf(stdout, " external_id(%lu)\n",
|
||||
record->external_id
|
||||
@@ -365,9 +373,10 @@ void init_tracing() {
|
||||
// Enable HIP activity tracing
|
||||
ROCTRACER_CALL(roctracer_enable_domain_activity(ACTIVITY_DOMAIN_HIP_API));
|
||||
ROCTRACER_CALL(roctracer_enable_domain_activity(ACTIVITY_DOMAIN_HCC_OPS));
|
||||
// Enable PC sampling
|
||||
ROCTRACER_CALL(roctracer_enable_op_activity(ACTIVITY_DOMAIN_HSA_OPS, HSA_OP_ID_PCSAMPLE));
|
||||
// Enable KFD API tracing
|
||||
ROCTRACER_CALL(roctracer_enable_domain_callback(ACTIVITY_DOMAIN_KFD_API, api_callback, NULL));
|
||||
ROCTRACER_CALL(roctracer_enable_domain_activity(ACTIVITY_DOMAIN_KFD_API));
|
||||
// Enable rocTX
|
||||
ROCTRACER_CALL(roctracer_enable_domain_callback(ACTIVITY_DOMAIN_ROCTX, api_callback, NULL));
|
||||
}
|
||||
@@ -385,7 +394,9 @@ void stop_tracing() {
|
||||
ROCTRACER_CALL(roctracer_disable_domain_callback(ACTIVITY_DOMAIN_HIP_API));
|
||||
ROCTRACER_CALL(roctracer_disable_domain_activity(ACTIVITY_DOMAIN_HIP_API));
|
||||
ROCTRACER_CALL(roctracer_disable_domain_activity(ACTIVITY_DOMAIN_HCC_OPS));
|
||||
ROCTRACER_CALL(roctracer_disable_domain_activity(ACTIVITY_DOMAIN_KFD_API));
|
||||
ROCTRACER_CALL(roctracer_disable_domain_activity(ACTIVITY_DOMAIN_HSA_OPS));
|
||||
ROCTRACER_CALL(roctracer_disable_domain_callback(ACTIVITY_DOMAIN_KFD_API));
|
||||
ROCTRACER_CALL(roctracer_disable_domain_callback(ACTIVITY_DOMAIN_ROCTX));
|
||||
ROCTRACER_CALL(roctracer_flush_activity());
|
||||
printf("# STOP #############################\n");
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user