mirror of
https://github.com/NVIDIA/cuda-samples.git
synced 2024-11-25 01:19:19 +08:00
899 lines
32 KiB
Plaintext
899 lines
32 KiB
Plaintext
/* Copyright (c) 2023, NVIDIA CORPORATION. All rights reserved.
|
|
*
|
|
* Redistribution and use in source and binary forms, with or without
|
|
* modification, are permitted provided that the following conditions
|
|
* are met:
|
|
* * Redistributions of source code must retain the above copyright
|
|
* notice, this list of conditions and the following disclaimer.
|
|
* * Redistributions in binary form must reproduce the above copyright
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
* documentation and/or other materials provided with the distribution.
|
|
* * Neither the name of NVIDIA CORPORATION nor the names of its
|
|
* contributors may be used to endorse or promote products derived
|
|
* from this software without specific prior written permission.
|
|
*
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
|
|
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
|
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
|
|
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
|
|
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
|
|
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
|
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
|
|
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
*/
|
|
|
|
#include "cudla.h"
|
|
#include "cuda_runtime.h"
|
|
#include "cudlaExternalEtbl.hpp"
|
|
|
|
#include <cstdio>
|
|
#include <cstdlib>
|
|
#include <cstring>
|
|
#include <sys/stat.h>
|
|
#include <fstream>
|
|
#include <sstream>
|
|
|
|
#define MAX_FILENAME_LEN 200
|
|
#define RESERVED_SUFFIX_LEN 10
|
|
|
|
#define DPRINTF(...) printf(__VA_ARGS__)
|
|
|
|
static void printTensorDesc(cudlaModuleTensorDescriptor* tensorDesc) {
|
|
DPRINTF("\tTENSOR NAME : %s\n", tensorDesc->name);
|
|
DPRINTF("\tsize: %lu\n", tensorDesc->size);
|
|
|
|
DPRINTF("\tdims: [%lu, %lu, %lu, %lu]\n", tensorDesc->n, tensorDesc->c,
|
|
tensorDesc->h, tensorDesc->w);
|
|
|
|
DPRINTF("\tdata fmt: %d\n", tensorDesc->dataFormat);
|
|
DPRINTF("\tdata type: %d\n", tensorDesc->dataType);
|
|
DPRINTF("\tdata category: %d\n", tensorDesc->dataCategory);
|
|
DPRINTF("\tpixel fmt: %d\n", tensorDesc->pixelFormat);
|
|
DPRINTF("\tpixel mapping: %d\n", tensorDesc->pixelMapping);
|
|
DPRINTF("\tstride[0]: %d\n", tensorDesc->stride[0]);
|
|
DPRINTF("\tstride[1]: %d\n", tensorDesc->stride[1]);
|
|
DPRINTF("\tstride[2]: %d\n", tensorDesc->stride[2]);
|
|
DPRINTF("\tstride[3]: %d\n", tensorDesc->stride[3]);
|
|
}
|
|
|
|
typedef struct {
|
|
cudlaDevHandle devHandle;
|
|
cudlaModule moduleHandle;
|
|
unsigned char* loadableData;
|
|
cudaStream_t stream;
|
|
uint32_t numInputTensors;
|
|
uint32_t numOutputTensors;
|
|
uint32_t numOutputTaskStatistics;
|
|
unsigned char** inputBuffer;
|
|
unsigned char** outputBuffer;
|
|
unsigned char** statisticsOutputBuffer;
|
|
void** inputBufferGPU;
|
|
void** outputBufferGPU;
|
|
void** outputTaskStatisticsGPU;
|
|
void **csv;
|
|
cudlaModuleTensorDescriptor* inputTensorDesc;
|
|
cudlaModuleTensorDescriptor* outputTensorDesc;
|
|
cudlaModuleTensorDescriptor* outputTaskStatisticsDesc;
|
|
uint64_t** inputBufferRegisteredPtr;
|
|
uint64_t** outputBufferRegisteredPtr;
|
|
uint64_t** outputTaskStatisticsRegisteredPtr;
|
|
uint64_t** outputStatisticsBufferRegisteredPtr;
|
|
} ResourceList;
|
|
|
|
void cleanUp(ResourceList* resourceList);
|
|
|
|
void cleanUp(ResourceList* resourceList) {
|
|
uint32_t ii = 0;
|
|
if (resourceList->inputTensorDesc != NULL) {
|
|
free(resourceList->inputTensorDesc);
|
|
resourceList->inputTensorDesc = NULL;
|
|
}
|
|
|
|
if (resourceList->outputTensorDesc != NULL) {
|
|
free(resourceList->outputTensorDesc);
|
|
resourceList->outputTensorDesc = NULL;
|
|
}
|
|
|
|
if (resourceList->outputTaskStatisticsDesc != NULL) {
|
|
free(resourceList->outputTaskStatisticsDesc);
|
|
resourceList->outputTaskStatisticsDesc = NULL;
|
|
}
|
|
|
|
if (resourceList->loadableData != NULL) {
|
|
free(resourceList->loadableData);
|
|
resourceList->loadableData = NULL;
|
|
}
|
|
|
|
if (resourceList->moduleHandle != NULL) {
|
|
cudlaModuleUnload(resourceList->moduleHandle, 0);
|
|
resourceList->moduleHandle = NULL;
|
|
}
|
|
|
|
if (resourceList->devHandle != NULL) {
|
|
cudlaDestroyDevice(resourceList->devHandle);
|
|
resourceList->devHandle = NULL;
|
|
}
|
|
|
|
if (resourceList->inputBufferGPU != NULL) {
|
|
for (ii = 0; ii < resourceList->numInputTensors; ii++) {
|
|
if ((resourceList->inputBufferGPU)[ii] != NULL) {
|
|
cudaFree((resourceList->inputBufferGPU)[ii]);
|
|
(resourceList->inputBufferGPU)[ii] = NULL;
|
|
}
|
|
}
|
|
free(resourceList->inputBufferGPU);
|
|
resourceList->inputBufferGPU = NULL;
|
|
}
|
|
|
|
if (resourceList->outputBufferGPU != NULL) {
|
|
for (ii = 0; ii < resourceList->numOutputTensors; ii++) {
|
|
if ((resourceList->outputBufferGPU)[ii] != NULL) {
|
|
cudaFree((resourceList->outputBufferGPU)[ii]);
|
|
(resourceList->outputBufferGPU)[ii] = NULL;
|
|
}
|
|
}
|
|
free(resourceList->outputBufferGPU);
|
|
resourceList->outputBufferGPU = NULL;
|
|
}
|
|
|
|
if (resourceList->outputTaskStatisticsGPU != NULL) {
|
|
for (ii = 0; ii < resourceList->numOutputTaskStatistics; ii++) {
|
|
if ((resourceList->outputTaskStatisticsGPU)[ii] != NULL) {
|
|
cudaFree((resourceList->outputTaskStatisticsGPU)[ii]);
|
|
(resourceList->outputTaskStatisticsGPU)[ii] = NULL;
|
|
}
|
|
}
|
|
free(resourceList->outputTaskStatisticsGPU);
|
|
resourceList->outputTaskStatisticsGPU = NULL;
|
|
}
|
|
|
|
if (resourceList->csv != NULL) {
|
|
for (ii = 0; ii < resourceList->numOutputTaskStatistics; ii++) {
|
|
if ((resourceList->csv)[ii] != NULL)
|
|
{
|
|
free((resourceList->csv)[ii]);
|
|
(resourceList->csv)[ii] = NULL;
|
|
}
|
|
}
|
|
free(resourceList->csv);
|
|
resourceList->csv = NULL;
|
|
}
|
|
|
|
if (resourceList->inputBuffer != NULL) {
|
|
for (ii = 0; ii < resourceList->numInputTensors; ii++) {
|
|
if ((resourceList->inputBuffer)[ii] != NULL) {
|
|
free((resourceList->inputBuffer)[ii]);
|
|
(resourceList->inputBuffer)[ii] = NULL;
|
|
}
|
|
}
|
|
free(resourceList->inputBuffer);
|
|
resourceList->inputBuffer = NULL;
|
|
}
|
|
|
|
if (resourceList->outputBuffer != NULL) {
|
|
for (ii = 0; ii < resourceList->numOutputTensors; ii++) {
|
|
if ((resourceList->outputBuffer)[ii] != NULL)
|
|
{
|
|
free((resourceList->outputBuffer)[ii]);
|
|
(resourceList->outputBuffer)[ii] = NULL;
|
|
}
|
|
}
|
|
free(resourceList->outputBuffer);
|
|
resourceList->outputBuffer = NULL;
|
|
}
|
|
|
|
if (resourceList->statisticsOutputBuffer != NULL) {
|
|
for (ii = 0; ii < resourceList->numOutputTaskStatistics; ii++) {
|
|
if ((resourceList->statisticsOutputBuffer)[ii] != NULL) {
|
|
free((resourceList->statisticsOutputBuffer)[ii]);
|
|
(resourceList->statisticsOutputBuffer)[ii] = NULL;
|
|
}
|
|
}
|
|
free(resourceList->statisticsOutputBuffer);
|
|
resourceList->statisticsOutputBuffer = NULL;
|
|
}
|
|
|
|
if (resourceList->stream != NULL) {
|
|
cudaStreamDestroy(resourceList->stream);
|
|
resourceList->stream = NULL;
|
|
}
|
|
|
|
if (resourceList->inputBufferRegisteredPtr != NULL) {
|
|
free(resourceList->inputBufferRegisteredPtr);
|
|
resourceList->inputBufferRegisteredPtr = NULL;
|
|
}
|
|
|
|
if (resourceList->outputBufferRegisteredPtr != NULL) {
|
|
free(resourceList->outputBufferRegisteredPtr);
|
|
resourceList->outputBufferRegisteredPtr = NULL;
|
|
}
|
|
|
|
if (resourceList->outputTaskStatisticsRegisteredPtr != NULL) {
|
|
free(resourceList->outputTaskStatisticsRegisteredPtr);
|
|
resourceList->outputTaskStatisticsRegisteredPtr = NULL;
|
|
}
|
|
|
|
if (resourceList->outputStatisticsBufferRegisteredPtr != NULL) {
|
|
free(resourceList->outputStatisticsBufferRegisteredPtr);
|
|
resourceList->outputStatisticsBufferRegisteredPtr = NULL;
|
|
}
|
|
|
|
resourceList->numInputTensors = 0;
|
|
resourceList->numOutputTensors = 0;
|
|
resourceList->numOutputTaskStatistics = 0;
|
|
}
|
|
|
|
int main(int argc, char** argv) {
|
|
cudlaDevHandle devHandle;
|
|
cudlaModule moduleHandle;
|
|
cudlaStatus err;
|
|
uint32_t statSupport = 0;
|
|
uint32_t dlaFreqInMHz = 0;
|
|
FILE* fp = NULL;
|
|
struct stat st;
|
|
size_t file_size;
|
|
size_t actually_read = 0;
|
|
unsigned char *loadableData = NULL;
|
|
char filename[MAX_FILENAME_LEN];
|
|
const char* suffix = ".csv";
|
|
|
|
cudaStream_t stream;
|
|
cudaError_t result;
|
|
const char* errPtr = NULL;
|
|
|
|
ResourceList resourceList;
|
|
|
|
memset(&resourceList, 0x00, sizeof(ResourceList));
|
|
|
|
if ((argc != 4) && (argc != 5)) {
|
|
DPRINTF("Usage : ./test_cudla_layerwise_stats_L0_hybrid_test1 <loadable> <freqMHZ> <statSupport> <filename prefix>\n");
|
|
return 1;
|
|
}
|
|
|
|
if (argc == 5) {
|
|
if((strlen(argv[4])) > (MAX_FILENAME_LEN - RESERVED_SUFFIX_LEN))
|
|
{
|
|
DPRINTF("Filename prefix length is too big, greater than maximum permissible prefix length of %u \n",(MAX_FILENAME_LEN - RESERVED_SUFFIX_LEN));
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
// Read loadable into buffer.
|
|
fp = fopen(argv[1], "rb");
|
|
if (fp == NULL) {
|
|
DPRINTF("Cannot open file %s\n", argv[1]);
|
|
return 1;
|
|
}
|
|
|
|
if (stat(argv[1], &st) != 0) {
|
|
DPRINTF("Cannot stat file\n");
|
|
return 1;
|
|
}
|
|
|
|
file_size = st.st_size;
|
|
DPRINTF("The file size = %ld\n", file_size);
|
|
|
|
dlaFreqInMHz = atoi(argv[2]);
|
|
statSupport = atoi(argv[3]);
|
|
|
|
loadableData = (unsigned char *)malloc(file_size);
|
|
if (loadableData == NULL) {
|
|
DPRINTF("Cannot Allocate memory for loadable\n");
|
|
return 1;
|
|
}
|
|
|
|
actually_read = fread(loadableData, 1, file_size, fp);
|
|
if ( actually_read != file_size ) {
|
|
free(loadableData);
|
|
DPRINTF("Read wrong size\n");
|
|
return 1;
|
|
}
|
|
fclose(fp);
|
|
|
|
resourceList.loadableData = loadableData;
|
|
|
|
// Initialize CUDA.
|
|
result = cudaFree(0);
|
|
if (result != cudaSuccess) {
|
|
errPtr = cudaGetErrorName(result);
|
|
DPRINTF("Error in creating cudaFree = %s\n", errPtr);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
result = cudaSetDevice(0);
|
|
if (result != cudaSuccess) {
|
|
errPtr = cudaGetErrorName(result);
|
|
DPRINTF("Error in creating cudaSetDevice = %s\n", errPtr);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
err = cudlaCreateDevice(0, &devHandle, CUDLA_CUDA_DLA);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in cuDLA create device = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
DPRINTF("Device created successfully\n");
|
|
resourceList.devHandle = devHandle;
|
|
|
|
err = cudlaModuleLoadFromMemory(devHandle, loadableData, file_size, &moduleHandle, 0);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in cudlaModuleLoadFromMemory = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
} else {
|
|
DPRINTF("Successfully loaded module\n");
|
|
}
|
|
|
|
resourceList.moduleHandle = moduleHandle;
|
|
|
|
// Create CUDA stream.
|
|
result = cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking);
|
|
|
|
if (result != cudaSuccess) {
|
|
errPtr = cudaGetErrorName(result);
|
|
DPRINTF("Error in creating cuda stream = %s\n", errPtr);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
resourceList.stream = stream;
|
|
|
|
// Get tensor attributes.
|
|
uint32_t numInputTensors = 0;
|
|
uint32_t numOutputTensors = 0;
|
|
uint32_t numOutputTaskStatistics = 0;
|
|
|
|
cudlaModuleAttribute attribute;
|
|
|
|
err = cudlaModuleGetAttributes(moduleHandle, CUDLA_NUM_INPUT_TENSORS, &attribute);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in getting numInputTensors = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
numInputTensors = attribute.numInputTensors;
|
|
DPRINTF("numInputTensors = %d\n", numInputTensors);
|
|
|
|
err = cudlaModuleGetAttributes(moduleHandle, CUDLA_NUM_OUTPUT_TENSORS, &attribute);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in getting numOutputTensors = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
numOutputTensors = attribute.numOutputTensors;
|
|
DPRINTF("numOutputTensors = %d\n", numOutputTensors);
|
|
|
|
// using the same attributes to get num_output_task_statistics_tensors
|
|
attribute.numOutputTensors = 0;
|
|
|
|
err = cudlaModuleGetAttributes(moduleHandle, CUDLA_NUM_OUTPUT_TASK_STATISTICS, &attribute);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in getting numOutputTensors = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
numOutputTaskStatistics = attribute.numOutputTensors;
|
|
DPRINTF("numOutputTaskStatistics = %d\n", numOutputTaskStatistics);
|
|
|
|
if(numOutputTaskStatistics == 0) {
|
|
DPRINTF("Layerwise stats is not supported for this Loadable \n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
resourceList.numInputTensors = numInputTensors;
|
|
resourceList.numOutputTensors = numOutputTensors;
|
|
resourceList.numOutputTaskStatistics = numOutputTaskStatistics;
|
|
|
|
cudlaModuleTensorDescriptor* inputTensorDesc =
|
|
(cudlaModuleTensorDescriptor*)malloc(sizeof(cudlaModuleTensorDescriptor)*numInputTensors);
|
|
cudlaModuleTensorDescriptor* outputTensorDesc =
|
|
(cudlaModuleTensorDescriptor*)malloc(sizeof(cudlaModuleTensorDescriptor)*numOutputTensors);
|
|
|
|
if ((inputTensorDesc == NULL) || (outputTensorDesc == NULL)) {
|
|
if (inputTensorDesc != NULL) {
|
|
free(inputTensorDesc);
|
|
inputTensorDesc = NULL;
|
|
}
|
|
|
|
if (outputTensorDesc != NULL) {
|
|
free(outputTensorDesc);
|
|
outputTensorDesc = NULL;
|
|
}
|
|
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
resourceList.inputTensorDesc = inputTensorDesc;
|
|
resourceList.outputTensorDesc = outputTensorDesc;
|
|
|
|
cudlaModuleTensorDescriptor* outputTaskStatisticsDesc =
|
|
(cudlaModuleTensorDescriptor*)malloc(sizeof(cudlaModuleTensorDescriptor)*numOutputTaskStatistics);
|
|
if (outputTaskStatisticsDesc == NULL) {
|
|
free(outputTaskStatisticsDesc);
|
|
outputTaskStatisticsDesc = NULL;
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
resourceList.outputTaskStatisticsDesc = outputTaskStatisticsDesc;
|
|
|
|
attribute.inputTensorDesc = inputTensorDesc;
|
|
err = cudlaModuleGetAttributes(moduleHandle,
|
|
CUDLA_INPUT_TENSOR_DESCRIPTORS,
|
|
&attribute);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in getting input tensor descriptor = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
DPRINTF("Printing input tensor descriptor\n");
|
|
printTensorDesc(inputTensorDesc);
|
|
|
|
attribute.outputTensorDesc = outputTensorDesc;
|
|
err = cudlaModuleGetAttributes(moduleHandle,
|
|
CUDLA_OUTPUT_TENSOR_DESCRIPTORS,
|
|
&attribute);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in getting output tensor descriptor = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
DPRINTF("Printing output tensor descriptor\n");
|
|
printTensorDesc(outputTensorDesc);
|
|
|
|
attribute.outputTensorDesc = outputTaskStatisticsDesc;
|
|
err = cudlaModuleGetAttributes(moduleHandle,
|
|
CUDLA_OUTPUT_TASK_STATISTICS_DESCRIPTORS,
|
|
&attribute);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in getting task statistics descriptor = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
DPRINTF("Printing output task statistics descriptor size\n");
|
|
for (uint32_t ii = 0; ii < numOutputTaskStatistics; ii++) {
|
|
DPRINTF("The size of %u descriptor is %lu\n", ii,outputTaskStatisticsDesc[ii].size);
|
|
}
|
|
|
|
// Setup the input and output buffers which will be used as an input to CUDA.
|
|
unsigned char** inputBuffer = (unsigned char **)malloc(sizeof(unsigned char *)*numInputTensors);
|
|
if (inputBuffer == NULL) {
|
|
DPRINTF("Error in allocating memory for input buffer array\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
memset(inputBuffer, 0x00, sizeof(unsigned char *)*numInputTensors);
|
|
resourceList.inputBuffer = inputBuffer;
|
|
for (uint32_t ii = 0; ii < numInputTensors; ii++) {
|
|
inputBuffer[ii] = (unsigned char* )malloc(inputTensorDesc[ii].size);
|
|
if (inputBuffer[ii] == NULL) {
|
|
DPRINTF("Error in allocating input memory\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
memset(inputBuffer[ii], 0x01, inputTensorDesc[ii].size);
|
|
}
|
|
|
|
unsigned char** outputBuffer = (unsigned char **)malloc(sizeof(unsigned char *)*numOutputTensors);
|
|
if (outputBuffer == NULL) {
|
|
DPRINTF("Error in allocating memory for output buffer array\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
memset(outputBuffer, 0x00, sizeof(unsigned char *)*numOutputTensors);
|
|
resourceList.outputBuffer = outputBuffer;
|
|
|
|
for (uint32_t ii = 0; ii < numOutputTensors; ii++) {
|
|
outputBuffer[ii] = (unsigned char* )malloc(outputTensorDesc[ii].size);
|
|
if (outputBuffer[ii] == NULL) {
|
|
DPRINTF("Error in allocating output memory\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
memset(outputBuffer[ii], 0x00, outputTensorDesc[ii].size);
|
|
}
|
|
|
|
unsigned char** statisticsOutputBuffer = (unsigned char **)malloc(sizeof(unsigned char *)*numOutputTaskStatistics);
|
|
if (statisticsOutputBuffer == NULL) {
|
|
DPRINTF("Error in allocating memory for output buffer array\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
memset(statisticsOutputBuffer, 0x00, sizeof(unsigned char *)*numOutputTaskStatistics);
|
|
resourceList.statisticsOutputBuffer = statisticsOutputBuffer;
|
|
|
|
for (uint32_t ii = 0; ii < numOutputTaskStatistics; ii++) {
|
|
statisticsOutputBuffer[ii] = (unsigned char* )malloc(outputTaskStatisticsDesc[ii].size);
|
|
if (outputBuffer[ii] == NULL) {
|
|
DPRINTF("Error in allocating output memory\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
memset(statisticsOutputBuffer[ii], 0x00, outputTaskStatisticsDesc[ii].size);
|
|
}
|
|
|
|
// Allocate memory on GPU.
|
|
void** inputBufferGPU = (void **)malloc(sizeof(void *)*numInputTensors);
|
|
if (inputBufferGPU == NULL) {
|
|
DPRINTF("Error in allocating memory for input buffer GPU array\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
memset(inputBufferGPU, 0x00, sizeof(void *)*numInputTensors);
|
|
resourceList.inputBufferGPU = inputBufferGPU;
|
|
|
|
for (uint32_t ii = 0; ii < numInputTensors; ii++) {
|
|
result = cudaMalloc(&(inputBufferGPU[ii]), inputTensorDesc[ii].size);
|
|
if (result != cudaSuccess)
|
|
{
|
|
DPRINTF("Error in allocating input memory on GPU\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
void** outputBufferGPU = (void **)malloc(sizeof(void *)*numOutputTensors);
|
|
if (outputBufferGPU == NULL) {
|
|
DPRINTF("Error in allocating memory for output buffer GPU array\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
memset(outputBufferGPU, 0x00, sizeof(void *)*numOutputTensors);
|
|
resourceList.outputBufferGPU = outputBufferGPU;
|
|
|
|
for (uint32_t ii = 0; ii < numOutputTensors; ii++) {
|
|
result = cudaMalloc(&(outputBufferGPU[ii]), outputTensorDesc[ii].size);
|
|
if (result != cudaSuccess) {
|
|
DPRINTF("Error in allocating output memory on GPU\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
void** outputTaskStatisticsGPU = (void **)malloc(sizeof(void *)*numOutputTaskStatistics);
|
|
if (outputTaskStatisticsGPU == NULL) {
|
|
DPRINTF("Error in allocating memory for output task statistics GPU array\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
memset(outputTaskStatisticsGPU, 0x00, sizeof(void *)*numOutputTaskStatistics);
|
|
resourceList.outputTaskStatisticsGPU = outputTaskStatisticsGPU;
|
|
|
|
for (uint32_t ii = 0; ii < numOutputTaskStatistics; ii++) {
|
|
result = cudaMalloc(&(outputTaskStatisticsGPU[ii]), outputTaskStatisticsDesc[ii].size);
|
|
if (result != cudaSuccess) {
|
|
DPRINTF("Error in allocating task statistics memory on GPU\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
uint64_t** inputBufferRegisteredPtr = (uint64_t **)malloc(sizeof(uint64_t*)*numInputTensors);
|
|
uint64_t** outputBufferRegisteredPtr = (uint64_t **)malloc(sizeof(uint64_t*)*numOutputTensors);
|
|
uint64_t** outputTaskStatisticsRegisteredPtr = (uint64_t **)malloc(sizeof(uint64_t*)*numOutputTaskStatistics);
|
|
|
|
if ((inputBufferRegisteredPtr == NULL) || (outputBufferRegisteredPtr == NULL) || (outputTaskStatisticsRegisteredPtr == NULL)) {
|
|
if (inputBufferRegisteredPtr != NULL) {
|
|
free(inputBufferRegisteredPtr);
|
|
inputBufferRegisteredPtr = NULL;
|
|
}
|
|
|
|
if (outputBufferRegisteredPtr != NULL) {
|
|
free(outputBufferRegisteredPtr);
|
|
outputBufferRegisteredPtr = NULL;
|
|
}
|
|
|
|
if (outputTaskStatisticsRegisteredPtr != NULL) {
|
|
free(outputTaskStatisticsRegisteredPtr);
|
|
outputTaskStatisticsRegisteredPtr = NULL;
|
|
}
|
|
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
resourceList.inputBufferRegisteredPtr = inputBufferRegisteredPtr;
|
|
resourceList.outputBufferRegisteredPtr = outputBufferRegisteredPtr;
|
|
resourceList.outputTaskStatisticsRegisteredPtr = outputTaskStatisticsRegisteredPtr;
|
|
|
|
// Register the CUDA-allocated buffers.
|
|
for (uint32_t ii = 0; ii < numInputTensors; ii++) {
|
|
err = cudlaMemRegister(devHandle,
|
|
(uint64_t* )(inputBufferGPU[ii]),
|
|
inputTensorDesc[ii].size,
|
|
&(inputBufferRegisteredPtr[ii]),
|
|
0);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in registering input memory = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
for (uint32_t ii = 0; ii < numOutputTensors; ii++) {
|
|
err = cudlaMemRegister(devHandle,
|
|
(uint64_t* )(outputBufferGPU[ii]),
|
|
outputTensorDesc[ii].size,
|
|
&(outputBufferRegisteredPtr[ii]),
|
|
0);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in registering output memory = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
for (uint32_t ii = 0; ii < numOutputTaskStatistics; ii++) {
|
|
err = cudlaMemRegister(devHandle,
|
|
(uint64_t* )(outputTaskStatisticsGPU[ii]),
|
|
outputTaskStatisticsDesc[ii].size,
|
|
&(outputTaskStatisticsRegisteredPtr[ii]),
|
|
CUDLA_TASK_STATISTICS);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in registering statistics output memory = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
DPRINTF("ALL MEMORY REGISTERED SUCCESSFULLY\n");
|
|
|
|
// Copy data from CPU buffers to GPU buffers.
|
|
for (uint32_t ii = 0; ii < numInputTensors; ii++) {
|
|
result = cudaMemcpyAsync(inputBufferGPU[ii], inputBuffer[ii], inputTensorDesc[ii].size, cudaMemcpyHostToDevice, stream);
|
|
if (result != cudaSuccess) {
|
|
DPRINTF("Error in enqueueing memcpy for input\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
for (uint32_t ii = 0; ii < numOutputTensors; ii++) {
|
|
result = cudaMemsetAsync(outputBufferGPU[ii], 0, outputTensorDesc[ii].size, stream);
|
|
if (result != cudaSuccess) {
|
|
DPRINTF("Error in enqueueing memset for output\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
for (uint32_t ii = 0; ii < numOutputTaskStatistics; ii++) {
|
|
result = cudaMemsetAsync(outputTaskStatisticsGPU[ii], 0, outputTaskStatisticsDesc[ii].size, stream);
|
|
if (result != cudaSuccess) {
|
|
DPRINTF("Error in enqueueing memset for statistics output\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
uint64_t *outputStatisticsBufferRegisteredPtr[numOutputTensors + numOutputTaskStatistics] = {0};
|
|
uint32_t index = 0;
|
|
for (; index < numOutputTensors ; index++) {
|
|
outputStatisticsBufferRegisteredPtr[index] = ((outputBufferRegisteredPtr[index]));
|
|
}
|
|
|
|
for (uint32_t jj=0; jj < numOutputTaskStatistics ; jj++) {
|
|
outputStatisticsBufferRegisteredPtr[index++] = ((outputTaskStatisticsRegisteredPtr[jj]));
|
|
}
|
|
|
|
// Enqueue a cuDLA task.
|
|
cudlaTask task;
|
|
task.moduleHandle = moduleHandle;
|
|
task.outputTensor = (uint64_t * const*)&outputStatisticsBufferRegisteredPtr;
|
|
|
|
if(statSupport == 1) {
|
|
task.numOutputTensors = (numOutputTensors + numOutputTaskStatistics);
|
|
DPRINTF("Layerwise profiling is requested \n");
|
|
} else {
|
|
task.numOutputTensors = numOutputTensors;
|
|
DPRINTF("Layerwise profiling is not requested \n");
|
|
}
|
|
|
|
task.numInputTensors = numInputTensors;
|
|
task.inputTensor = inputBufferRegisteredPtr;
|
|
task.waitEvents = NULL;
|
|
task.signalEvents = NULL;
|
|
|
|
err = cudlaSubmitTask(devHandle, &task, 1, stream, 0);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("no of output tensor %u \n",(task.numOutputTensors));
|
|
DPRINTF("Error in submitting task\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
DPRINTF("SUBMIT IS DONE !!!\n");
|
|
|
|
result = cudaStreamSynchronize(stream);
|
|
if (result != cudaSuccess) {
|
|
DPRINTF("Error in synchronizing stream = %s\n", cudaGetErrorName(result));
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
// Wait for stream operations to finish and bring output buffer to CPU.
|
|
for (uint32_t ii = 0; ii < numOutputTensors; ii++) {
|
|
result = cudaMemcpyAsync(outputBuffer[ii], outputBufferGPU[ii],
|
|
outputTensorDesc[ii].size, cudaMemcpyDeviceToHost, stream);
|
|
if (result != cudaSuccess) {
|
|
DPRINTF("Error in bringing result back to CPU\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
result = cudaStreamSynchronize(stream);
|
|
if (result != cudaSuccess) {
|
|
DPRINTF("Error in synchronizing stream\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
if(statSupport == 1) {
|
|
// copy statistics data to cpu
|
|
for (uint32_t ii = 0; ii < numOutputTaskStatistics; ii++) {
|
|
result = cudaMemcpyAsync(statisticsOutputBuffer[ii], outputTaskStatisticsGPU[ii],
|
|
outputTaskStatisticsDesc[ii].size, cudaMemcpyDeviceToHost, stream);
|
|
if (result != cudaSuccess) {
|
|
DPRINTF("Error in bringing result back to CPU\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
result = cudaStreamSynchronize(stream);
|
|
if (result != cudaSuccess) {
|
|
DPRINTF("Error in synchronizing stream\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
// To get the last index of the filename prefix in which statistics will be dumped
|
|
uint32_t index = 0;
|
|
if (argc == 5) {
|
|
while(argv[4][index]!='\0') {
|
|
index++;
|
|
}
|
|
}
|
|
|
|
const cudlaExternalEtbl* etbl = NULL;
|
|
if (cudlaGetExternalExportTable(&etbl,0) != cudlaSuccess) {
|
|
DPRINTF("Error in getting export table\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
void** csv = (void **)malloc(sizeof(void *)*numOutputTaskStatistics);
|
|
if (csv == NULL) {
|
|
DPRINTF("Error in allocating memory for csv stream\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
memset(csv, 0x00, sizeof(void *)*numOutputTaskStatistics);
|
|
resourceList.csv = csv;
|
|
|
|
for (uint32_t ii = 0; ii < numOutputTaskStatistics; ii++) {
|
|
cudlaTranslateCsvAttribute csvAttribute;
|
|
uint64_t csvStreamLength = 0;
|
|
|
|
err = etbl->etiTranslateStats(devHandle,statisticsOutputBuffer[ii],dlaFreqInMHz,ii,CUDLA_GET_CSV_LENGTH,&csvAttribute);
|
|
csv[ii] = (void* )malloc(csvAttribute.csvStreamLength);
|
|
csvStreamLength = csvAttribute.csvStreamLength;
|
|
DPRINTF("size for statistics buffer %u is %lu \n",ii,csvStreamLength);
|
|
|
|
if (csv[ii] == NULL) {
|
|
DPRINTF("Error in allocating memory for csv stream\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
memset(csv[ii], 0x00, csvAttribute.csvStreamLength);
|
|
|
|
csvAttribute.csvStreamStats = csv[ii];
|
|
err = etbl->etiTranslateStats(devHandle,statisticsOutputBuffer[ii],dlaFreqInMHz,ii,CUDLA_GET_CSV_STATS,&csvAttribute);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in translating stats\n");
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
if (argc == 5) {
|
|
sprintf(filename,"%s%u%s", argv[4],(ii+1),suffix);
|
|
fp = fopen(filename, "w+");
|
|
if (fp == NULL) {
|
|
DPRINTF("Cannot open file %s\n", filename);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
uint32_t ret_val = fwrite(csv[ii],sizeof(char),csvStreamLength,fp);
|
|
if(ret_val != csvStreamLength) {
|
|
DPRINTF("number of elements written to file is %u \n", ret_val);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
fclose(fp);
|
|
} else {
|
|
DPRINTF("%s \n",(char *)csv[ii]);
|
|
}
|
|
}
|
|
}
|
|
|
|
// unregister the CUDA-allocated buffers.
|
|
for (uint32_t ii = 0; ii < numInputTensors; ii++) {
|
|
err = cudlaMemUnregister(devHandle,
|
|
(inputBufferRegisteredPtr[ii]));
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in registering input memory = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
for (uint32_t ii = 0; ii < numOutputTensors; ii++) {
|
|
err = cudlaMemUnregister(devHandle,
|
|
(outputBufferRegisteredPtr[ii]));
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in registering output memory = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
for (uint32_t ii = 0; ii < numOutputTaskStatistics; ii++) {
|
|
err = cudlaMemUnregister(devHandle,
|
|
(outputTaskStatisticsRegisteredPtr[ii]));
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in registering output memory = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
DPRINTF("ALL MEMORY UNREGISTERED SUCCESSFULLY\n");
|
|
|
|
result = cudaStreamDestroy(stream);
|
|
if (result != cudaSuccess) {
|
|
errPtr = cudaGetErrorName(result);
|
|
DPRINTF("Error in destroying cuda stream = %s\n", errPtr);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
}
|
|
|
|
resourceList.stream = NULL;
|
|
|
|
err = cudlaModuleUnload(moduleHandle, 0);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in cudlaModuleUnload = %d\n", err);
|
|
cleanUp(&resourceList);
|
|
return 1;
|
|
} else {
|
|
DPRINTF("Successfully unloaded module\n");
|
|
}
|
|
|
|
resourceList.moduleHandle = NULL;
|
|
|
|
err = cudlaDestroyDevice(devHandle);
|
|
if (err != cudlaSuccess) {
|
|
DPRINTF("Error in cuDLA destroy device = %d\n", err);
|
|
return 1;
|
|
}
|
|
DPRINTF("Device destroyed successfully\n");
|
|
|
|
resourceList.devHandle = NULL;
|
|
|
|
cleanUp(&resourceList);
|
|
|
|
DPRINTF("cuDLALayerwiseStatsHybrid DONE !!!\n");
|
|
|
|
return 0;
|
|
}
|