//pteracuda_nifs:gemv(Ctx, _m, _n, _alpha, A, X, _betha, Y),
ERL_NIF_TERM pteracuda_nifs_gemv(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
PCudaContextRef *ctxRef;
PCudaBufferRef *ref_A, *ref_X, *ref_Y;
unsigned long transpose;
unsigned long m, n;
double alpha, beta;
if (argc != 9 ||
!enif_get_resource(env, argv[0], pteracuda_context_resource, (void **) &ctxRef) ||
!enif_get_ulong(env, argv[1], &transpose)||
!enif_get_ulong(env, argv[2], &m)||
!enif_get_ulong(env, argv[3], &n)||
!enif_get_double(env, argv[4], &alpha)||
!enif_get_resource(env, argv[5], pteracuda_buffer_resource, (void **) &ref_A) ||
!enif_get_resource(env, argv[6], pteracuda_buffer_resource, (void **) &ref_X)||
!enif_get_double(env, argv[7], &beta)||
!enif_get_resource(env, argv[8], pteracuda_buffer_resource, (void **) &ref_Y)) {
return enif_make_badarg(env);
}
if(((PCudaMatrixFloatBuffer*)ref_A->buffer)->rows() != m || ((PCudaMatrixFloatBuffer*)ref_A->buffer)->cols() != n){
return enif_make_tuple2(env, ATOM_ERROR, enif_make_atom(env, "Matrix A dimensions do not match m,n parameters"));
}
cuCtxSetCurrent(ctxRef->ctx);
pcuda_gemv(transpose, m, n, alpha, ((PCudaMatrixFloatBuffer *)ref_A->buffer)->get_data(), ((PCudaFloatBuffer *)ref_X->buffer)->get_data(), beta, ((PCudaFloatBuffer *)ref_Y->buffer)->get_data());
return ATOM_OK;
}
ERL_NIF_TERM pteracuda_nifs_transpose(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
PCudaContextRef *ctxRef;
PCudaBufferRef *ref_A, *ref_B;
unsigned long m, n;
if (argc != 3 ||
!enif_get_resource(env, argv[0], pteracuda_context_resource, (void **) &ctxRef) ||
!enif_get_resource(env, argv[1], pteracuda_buffer_resource, (void **) &ref_A)||
!enif_get_resource(env, argv[2], pteracuda_buffer_resource, (void **) &ref_B)) {
return enif_make_badarg(env);
}
if(((PCudaMatrixFloatBuffer *)ref_A->buffer)->rows() != ((PCudaMatrixFloatBuffer *)ref_B->buffer)->cols() ||
((PCudaMatrixFloatBuffer *)ref_A->buffer)->cols() != ((PCudaMatrixFloatBuffer *)ref_B->buffer)->rows() ){
return enif_make_tuple2(env, ATOM_ERROR, enif_make_atom(env, "Size A does not match the transpose size B."));
}
m = ((PCudaMatrixFloatBuffer *)ref_A->buffer)->rows();
n = ((PCudaMatrixFloatBuffer *)ref_A->buffer)->cols();
cuCtxSetCurrent(ctxRef->ctx);
//as the internal representation of a matrix buffer is "column major", the actual Rows x Columns is N x M
pcuda_transpose(n, m, ((PCudaMatrixFloatBuffer *)ref_A->buffer)->get_data(), ((PCudaMatrixFloatBuffer *)ref_B->buffer)->get_data());
return ATOM_OK;
}
ERL_NIF_TERM pteracuda_ml_gd_learn(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
PCudaContextRef *ctxRef;
PCudaBufferRef *ref_Theta, *ref_X, *ref_Y;
unsigned long num_features;
unsigned long num_samples;
unsigned long iterations;
double learning_rate;
if (argc != 8 || !enif_get_resource(env, argv[0], pteracuda_context_resource, (void **) &ctxRef) ||
!enif_get_resource(env, argv[1], pteracuda_buffer_resource, (void **) &ref_Theta) ||
!enif_get_resource(env, argv[2], pteracuda_buffer_resource, (void **) &ref_X) ||
!enif_get_resource(env, argv[3], pteracuda_buffer_resource, (void **) &ref_Y) ||
!enif_get_ulong(env, argv[4], &num_features) ||
!enif_get_ulong(env, argv[5], &num_samples) ||
!enif_get_double(env, argv[6], &learning_rate) ||
!enif_get_ulong(env, argv[7], &iterations)
) {
return enif_make_badarg(env);
}
cuCtxSetCurrent(ctxRef->ctx);
pcuda_gd_learn(((PCudaFloatBuffer*)ref_Theta->buffer)->get_data(), ((PCudaFloatBuffer*)ref_X->buffer)->get_data(), ((PCudaFloatBuffer*)ref_Y->buffer)->get_data(), num_features, num_samples, (float)learning_rate, iterations);
return ATOM_OK;
}
ERL_NIF_TERM pteracuda_nifs_saxpy(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
PCudaContextRef *ctxRef;
PCudaBufferRef *ref_X, *ref_Y;
double a;
if (argc != 4 ||
!enif_get_resource(env, argv[0], pteracuda_context_resource, (void **) &ctxRef) ||
!enif_get_double(env, argv[1], &a)||
!enif_get_resource(env, argv[2], pteracuda_buffer_resource, (void **) &ref_X)||
!enif_get_resource(env, argv[3], pteracuda_buffer_resource, (void **) &ref_Y)) {
return enif_make_badarg(env);
}
if(((PCudaFloatBuffer *)ref_X->buffer)->get_data()->size() != ((PCudaFloatBuffer *)ref_Y->buffer)->get_data()->size()){
return enif_make_tuple2(env, ATOM_ERROR, enif_make_atom(env, "Size X does not match size Y."));
}
cuCtxSetCurrent(ctxRef->ctx);
pcuda_saxpy(a, ((PCudaFloatBuffer *)ref_X->buffer)->get_data(), ((PCudaFloatBuffer *)ref_Y->buffer)->get_data());
return ATOM_OK;
}
void cuda_enter(cuda_context *ctx) {
ASSERT_CTX(ctx);
cuCtxGetCurrent(&ctx->old);
if (ctx->old != ctx->ctx)
ctx->err = cuCtxSetCurrent(ctx->ctx);
/* If no context was there in the first place, then we take over
to avoid the set dance on the thread */
if (ctx->old == NULL) ctx->old = ctx->ctx;
}
ERL_NIF_TERM pteracuda_nifs_buffer_intersection(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
PCudaContextRef *ctxRef;
PCudaBufferRef *first, *second;
if (argc !=3 || !enif_get_resource(env, argv[0], pteracuda_context_resource, (void **) &ctxRef) ||
!enif_get_resource(env, argv[1], pteracuda_buffer_resource, (void **) &first) ||
!enif_get_resource(env, argv[2], pteracuda_buffer_resource, (void **) &second)) {
return enif_make_badarg(env);
}
cuCtxSetCurrent(ctxRef->ctx);
return enif_make_tuple2(env, ATOM_OK, first->buffer->intersect(env, second->buffer));
}
SEXP R_cuCtxSetCurrent(SEXP r_ctx)
{
SEXP r_ans = R_NilValue;
CUcontext ctx = (CUcontext) getRReference(r_ctx);
CUresult ans;
ans = cuCtxSetCurrent(ctx);
r_ans = Renum_convert_CUresult(ans) ;
return(r_ans);
}
///////////////////Matrix operations
// C(m,n) = A(m,k) * B(k,n)
//gemm(_Ctx, _transpose_op_A, _transpose_op_B, _m, _n, _k, _alpha, _A, _B, _beta, _C )
ERL_NIF_TERM pteracuda_nifs_gemm(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
PCudaContextRef *ctxRef;
PCudaBufferRef *ref_A, *ref_B, *ref_C;
unsigned long transpose_a, transpose_b;
unsigned long m, n, k;
double alpha, beta;
if (argc != 11 || !enif_get_resource(env, argv[0], pteracuda_context_resource, (void **) &ctxRef) ||
!enif_get_ulong(env, argv[1], &transpose_a)||
!enif_get_ulong(env, argv[2], &transpose_b)||
!enif_get_ulong(env, argv[3], &m)||
!enif_get_ulong(env, argv[4], &n)||
!enif_get_ulong(env, argv[5], &k)||
!enif_get_double(env, argv[6], &alpha)||
!enif_get_resource(env, argv[7], pteracuda_buffer_resource, (void **) &ref_A) ||
!enif_get_resource(env, argv[8], pteracuda_buffer_resource, (void **) &ref_B)||
!enif_get_double(env, argv[9], &beta)||
!enif_get_resource(env, argv[10], pteracuda_buffer_resource, (void **) &ref_C)
) {
return enif_make_badarg(env);
}
if(transpose_a == CUBLAS_OP_N){
if(((PCudaMatrixFloatBuffer*)ref_A->buffer)->rows() != m || ((PCudaMatrixFloatBuffer*)ref_A->buffer)->cols() != k){
return enif_make_tuple2(env, ATOM_ERROR, enif_make_atom(env, "Matrix A dimensions do not match m,k parameters"));
}
}else{
if(((PCudaMatrixFloatBuffer*)ref_A->buffer)->rows() != k || ((PCudaMatrixFloatBuffer*)ref_A->buffer)->cols() != n){
return enif_make_tuple2(env, ATOM_ERROR, enif_make_atom(env, "Matrix A dimensions do not match m,k parameters"));
}
}
if(transpose_b == CUBLAS_OP_N){
if(((PCudaMatrixFloatBuffer*)ref_B->buffer)->rows() != k || ((PCudaMatrixFloatBuffer*)ref_B->buffer)->cols() != n){
return enif_make_tuple2(env, ATOM_ERROR, enif_make_atom(env, "Matrix B dimensions do not match k,n parameters"));
}
}else{
if(((PCudaMatrixFloatBuffer*)ref_B->buffer)->rows() != n || ((PCudaMatrixFloatBuffer*)ref_B->buffer)->cols() != k){
return enif_make_tuple2(env, ATOM_ERROR, enif_make_atom(env, "Matrix B dimensions do not match k,n parameters"));
}
}
if(((PCudaMatrixFloatBuffer*)ref_C->buffer)->rows() != m || ((PCudaMatrixFloatBuffer*)ref_C->buffer)->cols() != n){
return enif_make_tuple2(env, ATOM_ERROR, enif_make_atom(env, "Matrix C dimensions do not match m,n parameters"));
}
cuCtxSetCurrent(ctxRef->ctx);
//pcuda_mmul(((PCudaMatrixFloatBuffer*)ref_A->buffer)->get_data(), ((PCudaMatrixFloatBuffer*)ref_B->buffer)->get_data(), ((PCudaMatrixFloatBuffer*)ref_C->buffer)->get_data(), m, k, n);
pcuda_gemm(transpose_a, transpose_b, m, n, k, alpha, ((PCudaMatrixFloatBuffer*)ref_A->buffer)->get_data(), ((PCudaMatrixFloatBuffer*)ref_B->buffer)->get_data(), beta, ((PCudaMatrixFloatBuffer*)ref_C->buffer)->get_data());
return ATOM_OK;
}
ERL_NIF_TERM pteracuda_nifs_buffer_minmax(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
PCudaContextRef *ctxRef;
PCudaBufferRef *bufRef;
if (argc !=2 || !enif_get_resource(env, argv[0], pteracuda_context_resource, (void **) &ctxRef) ||
!enif_get_resource(env, argv[1], pteracuda_buffer_resource, (void **) &bufRef)) {
return enif_make_badarg(env);
}
if (bufRef->buffer->size() == 0) {
return enif_make_tuple2(env, ATOM_OK, enif_make_tuple2(env, enif_make_int(env, 0),
enif_make_int(env, 0)));
}
cuCtxSetCurrent(ctxRef->ctx);
return enif_make_tuple2(env, ATOM_OK, bufRef->buffer->minmax(env));
}
ERL_NIF_TERM pteracuda_nifs_sort_buffer(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
PCudaContextRef *ctxRef;
PCudaBufferRef *ref;
if (argc != 2 || !enif_get_resource(env, argv[0], pteracuda_context_resource, (void **) &ctxRef) ||
!enif_get_resource(env, argv[1], pteracuda_buffer_resource, (void **) &ref)) {
return enif_make_badarg(env);
}
cuCtxSetCurrent(ctxRef->ctx);
if (ref->buffer->sort()) {
return ATOM_OK;
}
else {
return ATOM_ERROR;
}
}
void
pocl_cuda_free (cl_device_id device, cl_mem mem_obj)
{
cuCtxSetCurrent (((pocl_cuda_device_data_t *)device->data)->context);
if (mem_obj->flags & CL_MEM_ALLOC_HOST_PTR)
{
cuMemFreeHost (mem_obj->mem_host_ptr);
mem_obj->mem_host_ptr = NULL;
}
else
{
void *ptr = mem_obj->device_ptrs[device->dev_id].mem_ptr;
cuMemFree ((CUdeviceptr)ptr);
}
}
ERL_NIF_TERM pteracuda_nifs_buffer_contains(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
PCudaContextRef *ctxRef;
PCudaBufferRef *ref;
if (argc !=3 || !enif_get_resource(env, argv[0], pteracuda_context_resource, (void **) &ctxRef) ||
!enif_get_resource(env, argv[1], pteracuda_buffer_resource, (void **) &ref)) {
return enif_make_badarg(env);
}
if (ref->buffer->size() > 0) {
cuCtxSetCurrent(ctxRef->ctx);
if (ref->buffer->contains(env, argv[2])) {
return ATOM_TRUE;
}
else {
return ATOM_FALSE;
}
}
else {
return ATOM_FALSE;
}
}
void Initialize()
{
cuCtxSetCurrent(m_Context);
// Encode a dummy frame since it seems that some initialization is done on the first encoding
m_PictureParameters.encodePicFlags = NV_ENC_PIC_FLAG_FORCEIDR;
#ifdef ASYNCHRONOUS
// Sanity check
_ASSERT(WaitForSingleObject(m_PictureParameters.completionEvent, 0) == WAIT_TIMEOUT);
#endif
CHECK_NVENC_STATUS(m_FunctionList.nvEncEncodePicture(m_pEncoder, &m_PictureParameters));
#ifdef ASYNCHRONOUS
DWORD nWaitResult = WaitForSingleObject(m_PictureParameters.completionEvent, INFINITE);
// Sanity check
_ASSERT(nWaitResult == WAIT_OBJECT_0);
#endif
NV_ENC_LOCK_BITSTREAM LockBitstream = { NV_ENC_LOCK_BITSTREAM_VER, 0 };
LockBitstream.sliceOffsets = NULL;
LockBitstream.outputBitstream = m_PictureParameters.outputBitstream;
CHECK_NVENC_STATUS(m_FunctionList.nvEncLockBitstream(m_pEncoder, &LockBitstream));
CHECK_NVENC_STATUS(m_FunctionList.nvEncUnlockBitstream(m_pEncoder, LockBitstream.outputBitstream));
}
ERL_NIF_TERM pteracuda_nifs_log(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
PCudaContextRef *ctxRef;
PCudaBufferRef *ref_A, *ref_B;
if (argc != 3 || !enif_get_resource(env, argv[0], pteracuda_context_resource, (void **) &ctxRef) ||
!enif_get_resource(env, argv[1], pteracuda_buffer_resource, (void **) &ref_A) ||
!enif_get_resource(env, argv[2], pteracuda_buffer_resource, (void **) &ref_B)
) {
return enif_make_badarg(env);
}
if(((PCudaFloatBuffer*)ref_A->buffer)->size() != ((PCudaFloatBuffer*)ref_B->buffer)->size()){
return enif_make_tuple2(env, ATOM_ERROR, enif_make_atom(env, "Buffer A size does not match buffer B size"));
}
cuCtxSetCurrent(ctxRef->ctx);
pcuda_log(((PCudaFloatBuffer*)ref_A->buffer)->get_data(), ((PCudaFloatBuffer*)ref_B->buffer)->get_data());
return ATOM_OK;
}
cl_int
pocl_cuda_alloc_mem_obj (cl_device_id device, cl_mem mem_obj, void *host_ptr)
{
cuCtxSetCurrent (((pocl_cuda_device_data_t *)device->data)->context);
CUresult result;
void *b = NULL;
/* if memory for this global memory is not yet allocated -> do it */
if (mem_obj->device_ptrs[device->global_mem_id].mem_ptr == NULL)
{
cl_mem_flags flags = mem_obj->flags;
if (flags & CL_MEM_USE_HOST_PTR)
{
#if defined __arm__
// cuMemHostRegister is not supported on ARN
// Allocate device memory and perform explicit copies
// before and after running a kernel
result = cuMemAlloc ((CUdeviceptr *)&b, mem_obj->size);
CUDA_CHECK (result, "cuMemAlloc");
#else
result = cuMemHostRegister (host_ptr, mem_obj->size,
CU_MEMHOSTREGISTER_DEVICEMAP);
if (result != CUDA_SUCCESS
&& result != CUDA_ERROR_HOST_MEMORY_ALREADY_REGISTERED)
CUDA_CHECK (result, "cuMemHostRegister");
result = cuMemHostGetDevicePointer ((CUdeviceptr *)&b, host_ptr, 0);
CUDA_CHECK (result, "cuMemHostGetDevicePointer");
#endif
}
else if (flags & CL_MEM_ALLOC_HOST_PTR)
{
result = cuMemHostAlloc (&mem_obj->mem_host_ptr, mem_obj->size,
CU_MEMHOSTREGISTER_DEVICEMAP);
CUDA_CHECK (result, "cuMemHostAlloc");
result = cuMemHostGetDevicePointer ((CUdeviceptr *)&b,
mem_obj->mem_host_ptr, 0);
CUDA_CHECK (result, "cuMemHostGetDevicePointer");
}
else
{
result = cuMemAlloc ((CUdeviceptr *)&b, mem_obj->size);
if (result != CUDA_SUCCESS)
{
const char *err;
cuGetErrorName (result, &err);
POCL_MSG_PRINT2 (__FUNCTION__, __LINE__,
"-> Failed to allocate memory: %s\n", err);
return CL_MEM_OBJECT_ALLOCATION_FAILURE;
}
}
if (flags & CL_MEM_COPY_HOST_PTR)
{
result = cuMemcpyHtoD ((CUdeviceptr)b, host_ptr, mem_obj->size);
CUDA_CHECK (result, "cuMemcpyHtoD");
}
mem_obj->device_ptrs[device->global_mem_id].mem_ptr = b;
mem_obj->device_ptrs[device->global_mem_id].global_mem_id
= device->global_mem_id;
}
/* copy already allocated global mem info to devices own slot */
mem_obj->device_ptrs[device->dev_id]
= mem_obj->device_ptrs[device->global_mem_id];
return CL_SUCCESS;
}
void cuda_pop_context()
{
cuda_assert(cuCtxSetCurrent(NULL));
}
return true;
}
#define cuda_error(stmt) cuda_error_(stmt, #stmt)
void cuda_error_message(const string& message)
{
if(error_msg == "")
error_msg = message;
fprintf(stderr, "%s\n", message.c_str());
cuda_error_documentation();
}
void cuda_push_context()
{
cuda_assert(cuCtxSetCurrent(cuContext))
}
void cuda_pop_context()
{
cuda_assert(cuCtxSetCurrent(NULL));
}
CUDADevice(DeviceInfo& info, Stats &stats, bool background_) : Device(stats)
{
first_error = true;
background = background_;
cuDevId = info.num;
cuDevice = 0;
cuContext = 0;
static void dispose(CUcontext context) {
cuda_check( cuCtxSetCurrent(context) );
cuda_check( cuCtxSynchronize() );
cuda_check( cuCtxDestroy(context) );
}
static CUT_THREADPROC dt_thread_func(void *p)
{
dt_partition *pt = (dt_partition *)p;
struct timeval tv;
CUresult res;
int thread_num_x=0, thread_num_y=0;
int block_num_x=0, block_num_y=0;
res = cuCtxSetCurrent(ctx[pt->pid]);
if(res != CUDA_SUCCESS) {
printf("cuCtxSetCurrent(ctx[%d]) failed: res = %s\n", pt->pid, cuda_response_to_string(res));
exit(1);
}
/* allocate GPU memory */
//printf("part_error_array_num = %d\n",part_error_array_num);
if(pt->pid == 0){
gettimeofday(&tv_memcpy_start, NULL);
}
res = cuMemcpyHtoD(part_C_dev[pt->pid], dst_C, SUM_SIZE_C);
if(res != CUDA_SUCCESS) {
printf("cuMemcpyHtoD(part_C_dev) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
res = cuMemcpyHtoD(part_error_array_dev[pt->pid], part_error_array, part_error_array_num*sizeof(int));
if(res != CUDA_SUCCESS) {
printf("cuMemcpyHtoD(part_error_array_dev) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
res = cuMemcpyHtoD(pm_size_array_dev[pt->pid], &pt->size_array[0][0], pt->NoP*2*pt->L_MAX*sizeof(int));
if(res != CUDA_SUCCESS) {
printf("cuMemcpyHtoD(pm_size_array_dev) falied: res = %s\n", cuda_response_to_string(res));
exit(1);
}
res = cuMemcpyHtoD(def_array_dev[pt->pid], pt->def, sum_size_def_array);
if(res != CUDA_SUCCESS) {
printf("cuMemcpyHtoD(def_array_dev) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
res = cuMemcpyHtoD(numpart_dev[pt->pid], pt->numpart, pt->NoC*sizeof(int));
if(res != CUDA_SUCCESS) {
printf("cuMemcpyHtoD(cuMemcpyHtoD(numpart_dev) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
res = cuMemcpyHtoD(PIDX_array_dev[pt->pid], pt->dst_PIDX, pt->tmp_array_size);
if(res != CUDA_SUCCESS) {
printf("cuMemcpyHtoD(PIDX_array) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
res = cuMemcpyHtoD(DID_4_array_dev[pt->pid], pt->dst_DID_4, pt->tmp_array_size);
if(res != CUDA_SUCCESS) {
printf("cuMemcpyHtoD(DID_4__array) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
if(pt->pid == 0){
gettimeofday(&tv_memcpy_end, NULL);
tvsub(&tv_memcpy_end, &tv_memcpy_start, &tv);
time_memcpy += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
}
int sharedMemBytes = 0;
/* get max thread num per block */
int max_threads_num = 0;
res = cuDeviceGetAttribute(&max_threads_num, CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK, dev[pt->pid]);
if(res != CUDA_SUCCESS){
printf("\ncuDeviceGetAttribute() failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
/* prepare for launch inverse_Q */
void* kernel_args_inverse[] = {
&part_C_dev[pt->pid],
&pm_size_array_dev[pt->pid],
&part_error_array_dev[pt->pid],
&part_error_array_num,
(void*)&(pt->NoP),
&PIDX_array_dev[pt->pid],
&numpart_dev[pt->pid],
(void*)&(pt->NoC),
(void*)&(pt->max_numpart),
(void*)&(pt->interval),
(void*)&(pt->L_MAX),
(void*)&(pt->pid),
(void*)&(device_num)
};
/* define CUDA block shape */
int upper_limit_th_num_x = max_threads_num/(pt->max_numpart*pt->NoC);
int upper_limit_th_num_y = max_threads_num/upper_limit_th_num_x;
if(upper_limit_th_num_x < 1) upper_limit_th_num_x++;
if(upper_limit_th_num_y < 1) upper_limit_th_num_y++;
thread_num_x = (pt->max_dim0*pt->max_dim1 < upper_limit_th_num_x) ? (pt->max_dim0*pt->max_dim1) : upper_limit_th_num_x;
thread_num_y = (pt->max_numpart < upper_limit_th_num_y) ? pt->max_numpart : upper_limit_th_num_y;
block_num_x = (pt->max_dim0*pt->max_dim1) / thread_num_x;
block_num_y = (pt->max_numpart) / thread_num_y;
if((pt->max_dim0*pt->max_dim1) % thread_num_x != 0) block_num_x++;
if(pt->max_numpart % thread_num_y != 0) block_num_y++;
int blockDimY = thread_num_y / device_num;
if(thread_num_y%device_num != 0){
blockDimY++;
}
/* launch iverse_Q */
if(pt->pid == 0){
gettimeofday(&tv_kernel_start, NULL);
}
res = cuLaunchKernel(
func_inverse_Q[pt->pid], // call function
block_num_x, // gridDimX
block_num_y, // gridDimY
pt->L_MAX-pt->interval, // gridDimZ
thread_num_x, // blockDimX
blockDimY, // blockDimY
pt->NoC, // blockDimZ
sharedMemBytes, // sharedMemBytes
NULL, // hStream
kernel_args_inverse, // kernelParams
NULL // extra
);
if(res != CUDA_SUCCESS) {
printf("block_num_x %d, block_num_y %d, thread_num_x %d, thread_num_y %d\n", block_num_x, block_num_y, thread_num_x, thread_num_y);
printf("cuLaunchKernel(inverse_Q) failed : res = %s\n", cuda_response_to_string(res));
exit(1);
}
res = cuCtxSynchronize();
if(res != CUDA_SUCCESS) {
printf("cuCtxSynchronize(inverse_Q) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
if(pt->pid == 0){
gettimeofday(&tv_kernel_end, NULL);
tvsub(&tv_kernel_end, &tv_kernel_start, &tv);
time_kernel += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
}
/* prepare for launch dt1d_x */
void* kernel_args_x[] = {
&part_C_dev[pt->pid], // FLOAT *src_start
&tmpM_dev[pt->pid], // FLOTA *dst
&tmpIy_dev[pt->pid], // int *ptr
&DID_4_array_dev[pt->pid], // int *DID_4_array,
&def_array_dev[pt->pid], // FLOAT *def_array,
&pm_size_array_dev[pt->pid], // int *size_array
(void*)&(pt->NoP), // int NoP
&PIDX_array_dev[pt->pid], // int *PIDX_array
&part_error_array_dev[pt->pid], // int *error_array
(void*)&(part_error_array_num), // int error_array_num
&numpart_dev[pt->pid], // int *numpart
(void*)&(pt->NoC), // int NoC
(void*)&(pt->max_numpart), // int max_numpart
(void*)&(pt->interval), // int interval
(void*)&(pt->L_MAX), // int L_MAX
(void*)&(pt->pid), // int pid
(void*)&(device_num) // int device_num
};
max_threads_num = 64/pt->NoC;
if(max_threads_num < 1) max_threads_num++;
thread_num_x = (pt->max_dim1 < max_threads_num) ? pt->max_dim1 : max_threads_num;
thread_num_y = (pt->max_numpart < max_threads_num) ? pt->max_numpart : max_threads_num;
block_num_x = pt->max_dim1 / thread_num_x;
block_num_y = pt->max_numpart / thread_num_y;
if(pt->max_dim1 % thread_num_x != 0) block_num_x++;
if(pt->max_numpart % thread_num_y != 0) block_num_y++;
blockDimY = thread_num_y / device_num;
if(thread_num_y%device_num != 0){
blockDimY++;
}
/* launch dt1d_x */
if(pt->pid == 0){
gettimeofday(&tv_kernel_start, NULL);
}
res = cuLaunchKernel(
func_dt1d_x[pt->pid], // call function
block_num_x, // gridDimX
block_num_y, // gridDimY
pt->L_MAX-pt->interval, // gridDimZ
thread_num_x, // blockDimX
blockDimY, // blockDimY
pt->NoC, // blockDimZ
sharedMemBytes, // sharedMemBytes
NULL, // hStream
kernel_args_x, // kernelParams
NULL // extra
);
if(res != CUDA_SUCCESS) {
printf("block_num_x %d, block_num_y %d, thread_num_x %d, thread_num_y %d\n", block_num_x, block_num_y, thread_num_x, thread_num_y);
printf("cuLaunchKernel(dt1d_x) failed : res = %s\n", cuda_response_to_string(res));
exit(1);
}
res = cuCtxSynchronize();
if(res != CUDA_SUCCESS) {
printf("cuCtxSynchronize(dt1d_x) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
if(pt->pid == 0){
gettimeofday(&tv_kernel_end, NULL);
tvsub(&tv_kernel_end, &tv_kernel_start, &tv);
time_kernel += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
}
/* prepare for launch dt1d_y */
void* kernel_args_y[] = {
&tmpM_dev[pt->pid], // FLOAT *src_start
&M_dev[pt->pid], // FLOAT *dst_start
&tmpIx_dev[pt->pid], // int *ptr_start
&DID_4_array_dev[pt->pid], // int *DID_4_array,
&def_array_dev[pt->pid], // FLOAT *def_array,
(void*)&(pt->NoP), // int NoP
&pm_size_array_dev[pt->pid], // int *size_array
&numpart_dev[pt->pid], // int *numpart,
&PIDX_array_dev[pt->pid], // int *PIDX_array,
(void*)&(pt->NoC), // int NoC
(void*)&(pt->max_numpart), // int max_numpart
(void*)&(pt->interval), // int interval
(void*)&(pt->L_MAX), // int L_MAX
&part_error_array_dev[pt->pid], // int *error_array
(void*)&(part_error_array_num), // int error_array_num
(void*)&(pt->pid), // int pid
(void*)&(device_num) // int device_num
};
thread_num_x = (pt->max_dim0 < max_threads_num) ? pt->max_dim0 : max_threads_num;
thread_num_y = (pt->max_numpart < max_threads_num) ? pt->max_numpart : max_threads_num;
block_num_x = pt->max_dim0 / thread_num_x;
block_num_y = pt->max_numpart / thread_num_y;
if(pt->max_dim0 % thread_num_x != 0) block_num_x++;
if(pt->max_numpart % thread_num_y != 0) block_num_y++;
blockDimY = thread_num_y / device_num;
if(thread_num_y%device_num != 0){
blockDimY++;
}
/* prepare for launch dt1d_y */
if(pt->pid == 0){
gettimeofday(&tv_kernel_start, NULL);
}
res = cuLaunchKernel(
func_dt1d_y[pt->pid], // call functions
block_num_x, // gridDimX
block_num_y, // gridDimY
pt->L_MAX-pt->interval, // gridDimZ
thread_num_x, // blockDimX
blockDimY, // blockDimY
pt->NoC, // blockDimZ
sharedMemBytes, // sharedMemBytes
NULL, // hStream
kernel_args_y, // kernelParams
NULL // extra
);
if(res != CUDA_SUCCESS) {
printf("cuLaunchKernel(dt1d_y failed : res = %s\n", cuda_response_to_string(res));
exit(1);
}
res = cuCtxSynchronize();
if(res != CUDA_SUCCESS) {
printf("cuCtxSynchronize(dt1d_y) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
if(pt->pid == 0){
gettimeofday(&tv_kernel_end, NULL);
tvsub(&tv_kernel_end, &tv_kernel_start, &tv);
time_kernel += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
}
/* downloads datas from GPU */
/* downloads M from GPU */
int sum_part_size = 0;
int sum_pointer_size = 0;
int sum_move_size = 0;
int part_size = 0;
int pointer_size = 0;
int part_y = 0;
int move_size = 0;
int start_kk = 0;
int end_kk = 0;
int part_end_kk = 0;
unsigned long long int pointer_dst_M = (unsigned long long int)pt->dst_M;
unsigned long long int pointer_M_dev = (unsigned long long int)M_dev[pt->pid];
for(int L=0; L<(pt->L_MAX-pt->interval); L++) {
/**************************************************************************/
/* loop condition */
if( (pt->FSIZE[(L+pt->interval)*2]+2*pt->pady < pt->max_Y) || (pt->FSIZE[(L+pt->interval)*2+1]+2*pt->padx < pt->max_X) )
{
continue;
}
/* loop conditon */
/**************************************************************************/
for(int jj=0; jj<pt->NoC; jj++) {
part_y = pt->numpart[jj] / device_num;
if(pt->numpart[jj]%device_num != 0){
part_y++;
}
start_kk = part_y * pt->pid;
end_kk = part_y * (pt->pid + 1);
if(end_kk > pt->numpart[jj]){
end_kk = pt->numpart[jj];
}
if(pt->pid > 0){
part_end_kk = part_y * pt->pid;
}
for(int kk=0; kk<pt->numpart[jj]; kk++) {
int PIDX = pt->PIDX_array[L][jj][kk];
int dims0 = pt->size_array[L][PIDX*2];
int dims1 = pt->size_array[L][PIDX*2+1];
if(start_kk <= kk && kk < end_kk){
part_size += dims0 * dims1;
}
//if(pt->pid > 0 && part_start_kk <= kk && kk < part_end_kk){
if(pt->pid > 0 && 0 <= kk && kk < part_end_kk){
pointer_size += dims0 * dims1;
}
move_size += dims0 * dims1;
}
sum_part_size += part_size;
sum_pointer_size += pointer_size;
sum_move_size += move_size;
// error pt->pid == 2 && L == 24 && jj == 1
if(pt->pid*part_y < pt->numpart[jj]){
if(pt->pid == 0){
gettimeofday(&tv_memcpy_start, NULL);
}
res = cuMemcpyDtoH((void *)(pointer_dst_M+(unsigned long long int)(pointer_size*sizeof(FLOAT))), (CUdeviceptr)(pointer_M_dev+(unsigned long long int)(pointer_size*sizeof(FLOAT))), part_size*sizeof(FLOAT));
if(res != CUDA_SUCCESS) {
printf("error pid = %d\n",pt->pid);
printf("cuMemcpyDtoH(dst_M) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
if(pt->pid == 0){
gettimeofday(&tv_memcpy_end, NULL);
tvsub(&tv_memcpy_end, &tv_memcpy_start, &tv);
time_memcpy += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
}
}
pointer_dst_M += (unsigned long long int)(move_size * sizeof(FLOAT));
pointer_M_dev += (unsigned long long int)(move_size * sizeof(FLOAT));
part_size = 0;
pointer_size = 0;
move_size = 0;
}
}
/* downloads tmpIx from GPU */
sum_part_size = 0;
sum_pointer_size = 0;
part_size = 0;
pointer_size = 0;
part_y = 0;
move_size = 0;
start_kk = 0;
end_kk = 0;
part_end_kk = 0;
unsigned long long int pointer_dst_tmpIx = (unsigned long long int)pt->dst_tmpIx;
unsigned long long int pointer_tmpIx_dev = (unsigned long long int)tmpIx_dev[pt->pid];
for(int L=0; L<(pt->L_MAX-pt->interval); L++) {
/**************************************************************************/
/* loop condition */
if( (pt->FSIZE[(L+pt->interval)*2]+2*pt->pady < pt->max_Y) || (pt->FSIZE[(L+pt->interval)*2+1]+2*pt->padx < pt->max_X) )
{
continue;
}
/* loop conditon */
/**************************************************************************/
for(int jj=0; jj<pt->NoC; jj++) {
part_y = pt->numpart[jj] / device_num;
if(pt->numpart[jj]%device_num != 0){
part_y++;
}
start_kk = part_y * pt->pid;
end_kk = part_y * (pt->pid + 1);
if(end_kk > pt->numpart[jj]){
end_kk = pt->numpart[jj];
}
if(pt->pid > 0){
part_end_kk = part_y * pt->pid;
}
for(int kk=0; kk<pt->numpart[jj]; kk++) {
int PIDX = pt->PIDX_array[L][jj][kk];
int dims0 = pt->size_array[L][PIDX*2];
int dims1 = pt->size_array[L][PIDX*2+1];
if(start_kk <= kk && kk < end_kk){
part_size += dims0 * dims1;
}
if(pt->pid > 0){
if(0 <= kk && kk < part_end_kk){
pointer_size += dims0 * dims1;
}
}
move_size += dims0 * dims1;
}
sum_part_size += part_size;
sum_pointer_size += pointer_size;
if(pt->pid*part_y < pt->numpart[jj]){
if(pt->pid == 0){
gettimeofday(&tv_memcpy_start, NULL);
}
res = cuMemcpyDtoH((void *)(pointer_dst_tmpIx+(unsigned long long int)(pointer_size*sizeof(int))), (CUdeviceptr)(pointer_tmpIx_dev+(unsigned long long int)(pointer_size*sizeof(int))), part_size*sizeof(int));
if(res != CUDA_SUCCESS) {
printf("error pid = %d\n",pt->pid);
printf("cuMemcpyDtoH(tmpIx) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
if(pt->pid == 0){
gettimeofday(&tv_memcpy_end, NULL);
tvsub(&tv_memcpy_end, &tv_memcpy_start, &tv);
time_memcpy += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
}
}
pointer_dst_tmpIx += (unsigned long long int)(move_size * sizeof(int));
pointer_tmpIx_dev += (unsigned long long int)(move_size * sizeof(int));
part_size = 0;
pointer_size = 0;
move_size = 0;
}
}
/* downloads tmpIy from GPU */
sum_part_size = 0;
sum_pointer_size = 0;
part_size = 0;
pointer_size = 0;
part_y = 0;
move_size = 0;
start_kk = 0;
end_kk = 0;
part_end_kk = 0;
unsigned long long int pointer_dst_tmpIy = (unsigned long long int)pt->dst_tmpIy;
unsigned long long int pointer_tmpIy_dev = (unsigned long long int)tmpIy_dev[pt->pid];
for(int L=0; L<(pt->L_MAX-pt->interval); L++) {
/**************************************************************************/
/* loop condition */
if( (pt->FSIZE[(L+pt->interval)*2]+2*pt->pady < pt->max_Y) || (pt->FSIZE[(L+pt->interval)*2+1]+2*pt->padx < pt->max_X) )
{
continue;
}
/* loop conditon */
/**************************************************************************/
for(int jj=0; jj<pt->NoC; jj++) {
part_y = pt->numpart[jj] / device_num;
if(pt->numpart[jj]%device_num != 0){
part_y++;
}
start_kk = part_y * pt->pid;
end_kk = part_y * (pt->pid + 1);
if(end_kk > pt->numpart[jj]){
end_kk = pt->numpart[jj];
}
if(pt->pid > 0){
part_end_kk = part_y * pt->pid;
}
for(int kk=0; kk<pt->numpart[jj]; kk++) {
int PIDX = pt->PIDX_array[L][jj][kk];
int dims0 = pt->size_array[L][PIDX*2];
int dims1 = pt->size_array[L][PIDX*2+1];
if(start_kk <= kk && kk < end_kk){
part_size += dims0 * dims1;
}
if(pt->pid > 0){
if(0 <= kk && kk < part_end_kk){
pointer_size += dims0 * dims1;
}
}
move_size += dims0 * dims1;
}
sum_part_size += part_size;
sum_pointer_size += pointer_size;
if(pt->pid*part_y < pt->numpart[jj]){
if(pt->pid == 0){
gettimeofday(&tv_memcpy_start, NULL);
}
res = cuMemcpyDtoH((void *)(pointer_dst_tmpIy+(unsigned long long int)(pointer_size*sizeof(int))), (CUdeviceptr)(pointer_tmpIy_dev+(unsigned long long int)(pointer_size*sizeof(int))), part_size*sizeof(int));
if(res != CUDA_SUCCESS) {
printf("error pid = %d\n",pt->pid);
printf("cuMemcpyDtoH(tmpIy) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
if(pt->pid == 0){
gettimeofday(&tv_memcpy_end, NULL);
tvsub(&tv_memcpy_end, &tv_memcpy_start, &tv);
time_memcpy += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
}
}
pointer_dst_tmpIy += (unsigned long long int)(move_size * sizeof(int));
pointer_tmpIy_dev += (unsigned long long int)(move_size * sizeof(int));
part_size = 0;
pointer_size = 0;
move_size = 0;
}
}
/* end of thread */
CUT_THREADEND;
}
void operator()(Handle handle) const {
if (ctx) cuda_check( cuCtxSetCurrent(ctx) );
deleter_impl<Handle>::dispose(handle);
}
/// Binds the context to the calling CPU thread.
void set_current() const {
cuda_check( cuCtxSetCurrent( c.get() ) );
}
void cuda_exit(cuda_context *ctx) {
if (ctx->old != ctx->ctx)
cuCtxSetCurrent(ctx->old);
}
void init_cuda(void)
{
CUresult res;
std::string cubin_path(STR(CUBIN_PATH));
// initnialize GPU
res = cuInit(0);
CUDA_CHECK(res, "cuInit()");
// count the number of usable GPU
res = cuDeviceGetCount(&device_num);
CUDA_CHECK(res, "cuDeviceGetCount()");
// unsupported multi GPU
device_num = 1;
// get device
dev = (CUdevice*) malloc(device_num * sizeof(CUdevice));
for (int i = 0; i < device_num; i++)
{
res = cuDeviceGet(&dev[i], i);
CUDA_CHECK(res, "cuDeviceGet()");
}
ctx = (CUcontext*) malloc(device_num * sizeof(CUcontext));
module = (CUmodule*) malloc(device_num * sizeof(CUmodule));
ConvolutionKernel_func = (CUfunction*) malloc(
device_num * sizeof(CUfunction));
DistanceTransformTwoDimensionalProblem_func = (CUfunction*) malloc(
device_num * sizeof(CUfunction));
BilinearKernelTex32F_func = (CUfunction*) malloc(
device_num * sizeof(CUfunction));
calculateHistogram_func = (CUfunction*) malloc(
device_num * sizeof(CUfunction));
getFeatureMaps_func = (CUfunction*) malloc(device_num * sizeof(CUfunction));
calculateNorm_func = (CUfunction*) malloc(device_num * sizeof(CUfunction));
normalizeAndTruncate_func = (CUfunction*) malloc(
device_num * sizeof(CUfunction));
PCAFeatureMapsAddNullableBorder_func = (CUfunction*) malloc(
device_num * sizeof(CUfunction));
for (int i = 0; i < device_num; i++)
{
res = cuCtxCreate(&ctx[i], 0, dev[i]);
CUDA_CHECK(res, "cuCtxCreate()");
}
for (int i = 0; i < device_num; i++)
{
res = cuCtxSetCurrent(ctx[i]);
CUDA_CHECK(res, "cuCtxSetCurrent()");
// load .cubin file
res = cuModuleLoad(&module[i], cubin_path.c_str());
CUDA_CHECK(res, "cuModuleLoad()");
res = cuModuleGetFunction(&ConvolutionKernel_func[i], module[i],
"ConvolutionKernel");
CUDA_CHECK(res, "cuModuleGetFunction(ConvolutionKernel)");
res = cuModuleGetFunction(
&DistanceTransformTwoDimensionalProblem_func[i], module[i],
"DistanceTransformTwoDimensionalProblemKernel");
CUDA_CHECK(res, "cuModuleGetFunction(DistanceTransformTwoDimensionalProblemKernel)");
res = cuModuleGetFunction(&BilinearKernelTex32F_func[i], module[i],
"BilinearKernelTex32F");
CUDA_CHECK(res, "cuModuleGetFunction(BilinearKernelTex32F)");
res = cuModuleGetFunction(&calculateHistogram_func[i], module[i],
"calculateHistogram");
CUDA_CHECK(res, "cuModuleGetFunction(calculateHistogram)");
res = cuModuleGetFunction(&getFeatureMaps_func[i], module[i],
"getFeatureMaps");
CUDA_CHECK(res, "cuModuleGetFunction(getFeatureMaps)");
res = cuModuleGetFunction(&calculateNorm_func[i], module[i],
"calculateNorm");
CUDA_CHECK(res, "cuModuleGetFunction(calculateNorm)");
res = cuModuleGetFunction(&normalizeAndTruncate_func[i], module[i],
"normalizeAndTruncate");
CUDA_CHECK(res, "cuModuleGetFunction(normalizeAndTruncate)");
res = cuModuleGetFunction(&PCAFeatureMapsAddNullableBorder_func[i],
module[i], "PCAFeatureMapsAddNullableBorder");
CUDA_CHECK(res, "cuModuleGetFunction(PCAFeatureMapsAddNullableBorder)");
}
NR_MAXTHREADS_X = (int*) malloc(device_num * sizeof(int));
NR_MAXTHREADS_Y = (int*) malloc(device_num * sizeof(int));
for (int i = 0; i < device_num; i++)
{
// get max thread num per block
max_threads_num = 0;
res = cuDeviceGetAttribute(&max_threads_num,
CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK, dev[i]);
CUDA_CHECK(res, "cuDeviceGetAttribute()");
NR_MAXTHREADS_X[i] = (int) sqrt((double) max_threads_num);
NR_MAXTHREADS_Y[i] = (int) sqrt((double) max_threads_num);
}
}
//detect boundary box
FLOAT *dpm_ttic_gpu_get_boxes(FLOAT **features,FLOAT *scales,int *feature_size, GPUModel *MO,
int *detected_count, FLOAT *acc_score, FLOAT thresh)
{
//constant parameters
const int max_scale = MO->MI->max_scale;
const int interval = MO->MI->interval;
const int sbin = MO->MI->sbin;
const int padx = MO->MI->padx;
const int pady = MO->MI->pady;
const int NoR = MO->RF->NoR;
const int NoP = MO->PF->NoP;
const int NoC = MO->MI->numcomponent;
const int *numpart = MO->MI->numpart;
const int LofFeat=(max_scale+interval)*NoC;
const int L_MAX = max_scale+interval;
/* for measurement */
struct timeval tv;
struct timeval tv_make_c_start, tv_make_c_end;
struct timeval tv_nucom_start, tv_nucom_end;
struct timeval tv_box_start, tv_box_end;
float time_box=0;
struct timeval tv_root_score_start, tv_root_score_end;
float time_root_score = 0;
struct timeval tv_part_score_start, tv_part_score_end;
float time_part_score = 0;
struct timeval tv_dt_start, tv_dt_end;
float time_dt = 0;
struct timeval tv_calc_a_score_start, tv_calc_a_score_end;
float time_calc_a_score = 0;
gettimeofday(&tv_make_c_start, nullptr);
int **RF_size = MO->RF->root_size;
int *rootsym = MO->RF->rootsym;
int *part_sym = MO->PF->part_sym;
int **part_size = MO->PF->part_size;
FLOAT **rootfilter = MO->RF->rootfilter;
FLOAT **partfilter=MO->PF->partfilter;
int **psize = MO->MI->psize;
int **rm_size_array = (int **)malloc(sizeof(int *)*L_MAX);
int **pm_size_array = (int **)malloc(sizeof(int *)*L_MAX);
pm_size_array = (int **)malloc(sizeof(int *)*L_MAX);
FLOAT **Tboxes=(FLOAT**)calloc(LofFeat,sizeof(FLOAT*)); //box coordinate information(Temp)
int *b_nums =(int*)calloc(LofFeat,sizeof(int)); //length of Tboxes
int count = 0;
int detected_boxes=0;
CUresult res;
/* matched score (root and part) */
FLOAT ***rootmatch,***partmatch = nullptr;
int *new_PADsize; // need new_PADsize[L_MAX*3]
size_t SUM_SIZE_feat = 0;
FLOAT **featp2 = (FLOAT **)malloc(L_MAX*sizeof(FLOAT *));
if(featp2 == nullptr) { // error semantics
printf("allocate featp2 failed\n");
exit(1);
}
/* allocate required memory for new_PADsize */
new_PADsize = (int *)malloc(L_MAX*3*sizeof(int));
if(new_PADsize == nullptr) { // error semantics
printf("allocate new_PADsize failed\n");
exit(1);
}
/* do padarray once and reuse it at calculating root and part time */
/* calculate sum of size of padded feature */
for(int tmpL=0; tmpL<L_MAX; tmpL++) {
int PADsize[3] = { feature_size[tmpL*2], feature_size[tmpL*2+1], 31 };
int NEW_Y = PADsize[0] + pady*2;
int NEW_X = PADsize[1] + padx*2;
SUM_SIZE_feat += (NEW_X*NEW_Y*PADsize[2])*sizeof(FLOAT);
}
/* allocate region for padded feat in a lump */
FLOAT *dst_feat;
res = cuMemHostAlloc((void **)&dst_feat, SUM_SIZE_feat, CU_MEMHOSTALLOC_DEVICEMAP);
if(res != CUDA_SUCCESS) {
printf("cuMemHostAlloc(dst_feat) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
memset(dst_feat, 0, SUM_SIZE_feat); // zero clear
/* distribute allocated region */
uintptr_t pointer_feat = (uintptr_t)dst_feat;
for(int tmpL=0; tmpL<L_MAX; tmpL++) {
featp2[tmpL] = (FLOAT *)pointer_feat;
int PADsize[3] = { feature_size[tmpL*2], feature_size[tmpL*2+1], 31 };
int NEW_Y = PADsize[0] + pady*2;
int NEW_X = PADsize[1] + padx*2;
pointer_feat += (uintptr_t)(NEW_X*NEW_Y*PADsize[2]*sizeof(FLOAT));
}
/* copy feat to feat2 */
for(int tmpL=0; tmpL<L_MAX; tmpL++) {
int PADsize[3] = { feature_size[tmpL*2], feature_size[tmpL*2+1], 31 };
int NEW_Y = PADsize[0] + pady*2;
int NEW_X = PADsize[1] + padx*2;
int L = NEW_Y*padx;
int SPL = PADsize[0] + pady;
int M_S = sizeof(FLOAT)*PADsize[0];
FLOAT *P = featp2[tmpL];
FLOAT *S = features[tmpL];
for(int i=0; i<PADsize[2]; i++)
{
P += L;
for(int j=0; j<PADsize[1]; j++)
{
P += pady;
memcpy(P, S, M_S);
S += PADsize[0];
P += SPL;
}
P += L;
}
new_PADsize[tmpL*3] = NEW_Y;
new_PADsize[tmpL*3 + 1] = NEW_X;
new_PADsize[tmpL*3 + 2] = PADsize[2];
}
/* do padarray once and reuse it at calculating root and part time */
/* allocation in a lump */
int *dst_rm_size = (int *)malloc(sizeof(int)*NoC*2*L_MAX);
if(dst_rm_size == nullptr) {
printf("allocate dst_rm_size failed\n");
exit(1);
}
/* distribution to rm_size_array[L_MAX] */
uintptr_t ptr = (uintptr_t)dst_rm_size;
for(int i=0; i<L_MAX; i++) {
rm_size_array[i] = (int *)ptr;
ptr += (uintptr_t)(NoC*2*sizeof(int));
}
/* allocation in a lump */
int *dst_pm_size = (int *)malloc(sizeof(int)*NoP*2*L_MAX);
if(dst_pm_size == nullptr) {
printf("allocate dst_pm_size failed\n");
exit(1);
}
/* distribution to pm_size_array[L_MAX] */
ptr = (uintptr_t)dst_pm_size;
for(int i=0; i<L_MAX; i++) {
pm_size_array[i] = (int *)ptr;
ptr += (uintptr_t)(NoP*2*sizeof(int));
}
///////level
for (int level=interval; level<L_MAX; level++) // feature's loop(A's loop) 1level 1picture
{
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X))
{
Tboxes[count]=nullptr;
count++;
continue;
}
} //for (level) // feature's loop(A's loop) 1level 1picture
///////root calculation/////////
/* calculate model score (only root) */
gettimeofday(&tv_root_score_start, nullptr);
rootmatch = fconvsMT_GPU(
featp2,
SUM_SIZE_feat,
rootfilter,
rootsym,
1,
NoR,
new_PADsize,
RF_size, rm_size_array,
L_MAX,
interval,
feature_size,
padx,
pady,
MO->MI->max_X,
MO->MI->max_Y,
ROOT
);
gettimeofday(&tv_root_score_end, nullptr);
tvsub(&tv_root_score_end, &tv_root_score_start, &tv);
time_root_score += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
///////part calculation/////////
if(NoP>0)
{
/* calculate model score (only part) */
gettimeofday(&tv_part_score_start, nullptr);
partmatch = fconvsMT_GPU(
featp2,
SUM_SIZE_feat,
partfilter,
part_sym,
1,
NoP,
new_PADsize,
part_size,
pm_size_array,
L_MAX,
interval,
feature_size,
padx,
pady,
MO->MI->max_X,
MO->MI->max_Y,
PART
);
gettimeofday(&tv_part_score_end, nullptr);
tvsub(&tv_part_score_end, &tv_part_score_start, &tv);
time_part_score += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
}
res = cuCtxSetCurrent(ctx[0]);
if(res != CUDA_SUCCESS) {
printf("cuCtxSetCurrent(ctx[0]) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
gettimeofday(&tv_make_c_end, nullptr);
gettimeofday(&tv_nucom_start, nullptr);
count = 0;
detected_boxes = 0;
int **RL_array = (int **)malloc((L_MAX-interval)*sizeof(int*));
int *dst_RL = (int *) malloc(NoC*(L_MAX-interval)*sizeof(int));
int **RI_array = (int **)malloc((L_MAX-interval)*sizeof(int*));
int *dst_RI = (int *)malloc(NoC*(L_MAX-interval)*sizeof(int));
int **OI_array = (int **)malloc((L_MAX-interval)*sizeof(int*));
int *dst_OI = (int *)malloc((NoC)*(L_MAX-interval)*sizeof(int));
int **RL_S_array = (int **)malloc((L_MAX-interval)*sizeof(int*));
int *dst_RL_S = (int *)malloc(NoC*(L_MAX-interval)*sizeof(int));
FLOAT **OFF_array = (FLOAT **)malloc((L_MAX-interval)*sizeof(FLOAT*));
FLOAT *dst_OFF = (FLOAT *)malloc(NoC*(L_MAX-interval)*sizeof(FLOAT));
FLOAT ***SCORE_array = (FLOAT ***)malloc((L_MAX-interval)*sizeof(FLOAT **));
FLOAT **sub_dst_SCORE = (FLOAT **)malloc(NoC*(L_MAX-interval)*sizeof(FLOAT*));
uintptr_t pointer_RL = (uintptr_t)dst_RL;
uintptr_t pointer_RI = (uintptr_t)dst_RI;
uintptr_t pointer_OI = (uintptr_t)dst_OI;
uintptr_t pointer_RL_S = (uintptr_t)dst_RL_S;
uintptr_t pointer_OFF = (uintptr_t)dst_OFF;
uintptr_t pointer_SCORE = (uintptr_t)sub_dst_SCORE;
for (int level=interval; level<L_MAX; level++) {
int L=level-interval;
RL_array[L] = (int *)pointer_RL;
pointer_RL += (uintptr_t)NoC*sizeof(int);
RI_array[L] = (int *)pointer_RI;
pointer_RI += (uintptr_t)NoC*sizeof(int);
OI_array[L] = (int *)pointer_OI;
pointer_OI += (uintptr_t)NoC*sizeof(int);
RL_S_array[L] = (int *)pointer_RL_S;
pointer_RL_S += (uintptr_t)NoC*sizeof(int);
OFF_array[L] = (FLOAT *)pointer_OFF;
pointer_OFF += (uintptr_t)NoC*sizeof(FLOAT);
SCORE_array[L] = (FLOAT **)pointer_SCORE;
pointer_SCORE += (uintptr_t)NoC*sizeof(FLOAT*);
}
int sum_RL_S = 0;
int sum_SNJ = 0;
/* prepare for parallel execution */
for(int level=interval; level<L_MAX; level++) {
int L = level - interval;
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X))
{
continue;
}
for(int j=0; j<NoC; j++) {
/* root score + offset */
RL_array[L][j] = rm_size_array[level][j*2]*rm_size_array[level][j*2+1]; //length of root-matching
RI_array[L][j] = MO->MI->ridx[j]; //root-index
OI_array[L][j] = MO->MI->oidx[j]; //offset-index
RL_S_array[L][j] =sizeof(FLOAT)*RL_array[L][j];
OFF_array[L][j] = MO->MI->offw[RI_array[L][j]]; //offset information
/* search max values */
max_RL_S = (max_RL_S < RL_S_array[L][j]) ? RL_S_array[L][j] : max_RL_S;
max_numpart = (max_numpart < numpart[j]) ? numpart[j] : max_numpart;
}
}
sum_RL_S = max_RL_S*NoC*(L_MAX-interval);
/* root matching size */
sum_SNJ = sizeof(int*)*max_numpart*NoC*(L_MAX-interval);
/* consolidated allocation for SCORE_array and distribute region */
FLOAT *dst_SCORE = (FLOAT *)malloc(sum_RL_S);
pointer_SCORE = (uintptr_t)dst_SCORE;
for(int level=interval; level<L_MAX; level++) {
int L = level - interval;
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X))
{
continue;
}
for(int j=0; j<NoC; j++) {
SCORE_array[L][j] = (FLOAT *)pointer_SCORE;
pointer_SCORE += (uintptr_t)max_RL_S;
}
}
/* add offset */
for(int level=interval; level<L_MAX; level++) {
int L = level - interval;
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X))
{
continue;
}
for(int j=0; j<NoC; j++) {
memcpy(SCORE_array[L][j], rootmatch[level][j], RL_S_array[L][j]);
FLOAT *SC_S = SCORE_array[L][j];
FLOAT *SC_E = SCORE_array[L][j]+RL_array[L][j];
while(SC_S<SC_E) *(SC_S++)+=OFF_array[L][j];
}
}
/* anchor matrix */ // consolidated allocation
int ***ax_array = (int ***)malloc((L_MAX-interval)*sizeof(int **));
int **sub_dst_ax = (int **)malloc(NoC*(L_MAX-interval)*sizeof(int *));
int *dst_ax = (int *)malloc(sum_SNJ);
int ***ay_array = (int ***)malloc((L_MAX-interval)*sizeof(int **));
int **sub_dst_ay = (int **)malloc(NoC*(L_MAX-interval)*sizeof(int *));
int *dst_ay = (int *)malloc(sum_SNJ);
/* boudary index */ // consolidated allocation
int ****Ix_array =(int ****)malloc((L_MAX-interval)*sizeof(int ***));
int ***sub_dst_Ix = (int ***)malloc(NoC*(L_MAX-interval)*sizeof(int **));
int **dst_Ix = (int **)malloc(sum_SNJ);
int ****Iy_array = (int ****)malloc((L_MAX-interval)*sizeof(int ***));
int ***sub_dst_Iy = (int ***)malloc(NoC*(L_MAX-interval)*sizeof(int **));
int **dst_Iy = (int **)malloc(sum_SNJ);
/* distribute region */
uintptr_t pointer_ax = (uintptr_t)sub_dst_ax;
uintptr_t pointer_ay = (uintptr_t)sub_dst_ay;
uintptr_t pointer_Ix = (uintptr_t)sub_dst_Ix;
uintptr_t pointer_Iy = (uintptr_t)sub_dst_Iy;
for(int level=interval; level<L_MAX; level++) {
int L = level - interval;
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X))
{
continue;
}
ax_array[L] = (int **)pointer_ax;
pointer_ax += (uintptr_t)(NoC*sizeof(int*));
ay_array[L] = (int **)pointer_ay;
pointer_ay += (uintptr_t)(NoC*sizeof(int*));
Ix_array[L] = (int ***)pointer_Ix;
pointer_Ix += (uintptr_t)(NoC*sizeof(int**));
Iy_array[L] = (int ***)pointer_Iy;
pointer_Iy += (uintptr_t)(NoC*sizeof(int**));
}
pointer_ax = (uintptr_t)dst_ax;
pointer_ay = (uintptr_t)dst_ay;
pointer_Ix = (uintptr_t)dst_Ix;
pointer_Iy = (uintptr_t)dst_Iy;
for(int level=interval; level<L_MAX; level++) {
int L = level - interval;
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X))
{
continue;
}
for(int j=0; j<NoC; j++) {
uintptr_t pointer_offset = sizeof(int*)*max_numpart;
ax_array[L][j] = (int *)pointer_ax;
pointer_ax += pointer_offset;
ay_array[L][j] = (int *)pointer_ay;
pointer_ay += pointer_offset;
Ix_array[L][j] = (int **)pointer_Ix;
pointer_Ix += pointer_offset;
Iy_array[L][j] = (int **)pointer_Iy;
pointer_Iy += pointer_offset;
}
}
/* add parts */
if(NoP>0)
{
/* arrays to store temporary loop variables */
int tmp_array_size = 0;
for(int level=interval; level<L_MAX; level++) {
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X))
{
continue;
}
for(int j=0; j<NoC; j++) {
tmp_array_size += max_numpart*sizeof(int);
}
}
int ***DIDX_array = (int ***)malloc((L_MAX-interval)*sizeof(int**));
int **sub_dst_DIDX = (int **)malloc(NoC*(L_MAX-interval)*sizeof(int*));
int *dst_DIDX = (int *)malloc(tmp_array_size);
int ***DID_4_array = (int ***)malloc((L_MAX-interval)*sizeof(int **));
int **sub_dst_DID_4 = (int **)malloc(NoC*(L_MAX-interval)*sizeof(int*));
int *dst_DID_4;
res = cuMemHostAlloc((void **)&dst_DID_4, tmp_array_size, CU_MEMHOSTALLOC_DEVICEMAP);
if(res != CUDA_SUCCESS) {
printf("cuMemHostAlloc(dst_DID_4) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
int ***PIDX_array = (int ***)malloc((L_MAX-interval)*sizeof(int **));
int **sub_dst_PIDX = (int **)malloc(NoC*(L_MAX-interval)*sizeof(int*));
int *dst_PIDX;
res = cuMemHostAlloc((void **)&dst_PIDX, tmp_array_size, CU_MEMHOSTALLOC_DEVICEMAP);
if(res != CUDA_SUCCESS) {
printf("cuMemHostAlloc(dst_PIDX) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
/* distribute consolidated region */
uintptr_t pointer_DIDX = (uintptr_t)sub_dst_DIDX;
uintptr_t pointer_DID_4 = (uintptr_t)sub_dst_DID_4;
uintptr_t pointer_PIDX = (uintptr_t)sub_dst_PIDX;
for(int level=interval; level<L_MAX; level++) {
int L = level - interval;
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X))
{
continue;
}
DIDX_array[L] = (int **)pointer_DIDX;
pointer_DIDX += (uintptr_t)(NoC*sizeof(int*));
DID_4_array[L] = (int **)pointer_DID_4;
pointer_DID_4 += (uintptr_t)(NoC*sizeof(int*));
PIDX_array[L] = (int **)pointer_PIDX;
pointer_PIDX += (uintptr_t)(NoC*sizeof(int*));
}
pointer_DIDX = (uintptr_t)dst_DIDX;
pointer_DID_4 = (uintptr_t)dst_DID_4;
pointer_PIDX = (uintptr_t)dst_PIDX;
for(int level=interval; level<L_MAX; level++) {
int L = level - interval;
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X)) {
continue;
}
for(int j=0; j<NoC; j++) {
uintptr_t pointer_offset = (uintptr_t)(max_numpart*sizeof(int));
DIDX_array[L][j] = (int *)pointer_DIDX;
pointer_DIDX += pointer_offset;
DID_4_array[L][j] = (int *)pointer_DID_4;
pointer_DID_4 += pointer_offset;
PIDX_array[L][j] = (int *)pointer_PIDX;
pointer_PIDX += pointer_offset;
}
}
/* prepare for parallel execution */
int sum_size_index_matrix = 0;
for(int level=interval; level<L_MAX; level++) {
int L = level - interval;
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X)) {
continue;
}
for(int j=0; j<NoC; j++) {
for (int k=0;k<numpart[j];k++) {
/* assign values to each element */
DIDX_array[L][j][k] = MO->MI->didx[j][k];
DID_4_array[L][j][k] = DIDX_array[L][j][k]*4;
PIDX_array[L][j][k] = MO->MI->pidx[j][k];
/* anchor */
ax_array[L][j][k] = MO->MI->anchor[DIDX_array[L][j][k]*2]+1;
ay_array[L][j][k] = MO->MI->anchor[DIDX_array[L][j][k]*2+1]+1;
int PSSIZE[2] ={pm_size_array[L][PIDX_array[L][j][k]*2], pm_size_array[L][PIDX_array[L][j][k]*2+1]}; // size of C
/* index matrix */
sum_size_index_matrix += sizeof(int)*PSSIZE[0]*PSSIZE[1];
}
}
}
int *dst_Ix_kk = (int *)malloc(sum_size_index_matrix);
int *dst_Iy_kk = (int *)malloc(sum_size_index_matrix);
uintptr_t pointer_Ix_kk = (uintptr_t)dst_Ix_kk;
uintptr_t pointer_Iy_kk = (uintptr_t)dst_Iy_kk;
for(int level=interval; level<L_MAX; level++) {
int L = level - interval;
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X))
{
continue;
}
for(int j=0; j<NoC; j++) {
for (int k=0;k<numpart[j];k++) {
int PSSIZE[2] ={pm_size_array[L][PIDX_array[L][j][k]*2], pm_size_array[L][PIDX_array[L][j][k]*2+1]}; // size of C
Ix_array[L][j][k] = (int *)pointer_Ix_kk;
Iy_array[L][j][k] = (int *)pointer_Iy_kk;
pointer_Ix_kk += (uintptr_t)(sizeof(int)*PSSIZE[0]*PSSIZE[1]);
pointer_Iy_kk += (uintptr_t)(sizeof(int)*PSSIZE[0]*PSSIZE[1]);
}
}
}
gettimeofday(&tv_dt_start, nullptr);
FLOAT ****M_array = dt_GPU(
Ix_array, // int ****Ix_array
Iy_array, // int ****Iy_array
PIDX_array, // int ***PIDX_array
pm_size_array, // int **size_array
NoP, // int NoP
numpart, // int *numpart
NoC, // int NoC
interval, // int interval
L_MAX, // int L_MAX
feature_size, // int *feature_size,
padx, // int padx,
pady, // int pady,
MO->MI->max_X, // int max_X
MO->MI->max_Y, // int max_Y
MO->MI->def, // FLOAT *def
tmp_array_size, // int tmp_array_size
dst_PIDX, // int *dst_PIDX
dst_DID_4 // int *DID_4
);
gettimeofday(&tv_dt_end, nullptr);
tvsub(&tv_dt_end, &tv_dt_start, &tv);
time_dt += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
/* add part score */
for(int level=interval; level<L_MAX; level++){
int L = level - interval;
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X))
{
continue;
}
for(int j=0; j<NoC; j++) {
for(int k=0; k<numpart[j]; k++) {
int PSSIZE[2] ={pm_size_array[L][PIDX_array[L][j][k]*2],
pm_size_array[L][PIDX_array[L][j][k]*2+1]}; // Size of C
int R_S[2]={rm_size_array[level][j*2], rm_size_array[level][j*2+1]};
dpm_ttic_add_part_calculation(SCORE_array[L][j], M_array[L][j][k], R_S,
PSSIZE, ax_array[L][j][k], ay_array[L][j][k]);
}
}
}
s_free(M_array[0][0][0]);
s_free(M_array[0][0]);
s_free(M_array[0]);
s_free(M_array);
/* free temporary arrays */
free(dst_DIDX);
free(sub_dst_DIDX);
free(DIDX_array);
res = cuMemFreeHost(dst_DID_4);
if(res != CUDA_SUCCESS) {
printf("cuMemFreeHost(dst_DID_4) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
free(sub_dst_DID_4);
free(DID_4_array);
res = cuMemFreeHost(dst_PIDX);
if(res != CUDA_SUCCESS) {
printf("cuMemFreeHost(dst_PIDX) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
free(sub_dst_PIDX);
free(PIDX_array);
res = cuCtxSetCurrent(ctx[0]);
if(res != CUDA_SUCCESS) {
printf("cuCtxSetCurrent(ctx[0]) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
} // start from if(NoP>0)
/* combine root and part score and detect boundary box for each-component */
FLOAT *scale_array = (FLOAT *)malloc((L_MAX-interval)*sizeof(FLOAT));
for(int level=interval; level<L_MAX; level++) {
int L = level - interval;
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X))
{
Tboxes[count]=nullptr;
count++;
continue;
}
scale_array[L] = (FLOAT)sbin/scales[level];
}
for (int level=interval; level<L_MAX; level++) // feature's loop(A's loop) 1level 1picture
{
/* parameters (related for level) */
int L=level-interval;
/* matched score size matrix */
FLOAT scale=(FLOAT)sbin/scales[level];
/* loop conditon */
if(feature_size[level*2]+2*pady<MO->MI->max_Y ||(feature_size[level*2+1]+2*padx<MO->MI->max_X)) {
Tboxes[count]=nullptr;
count++;
continue;
}
/* calculate accumulated score */
gettimeofday(&tv_calc_a_score_start, nullptr);
calc_a_score_GPU(
acc_score, // FLOAT *ac_score
SCORE_array[L], // FLOAT **score
rm_size_array[level], // int *ssize_start
MO->MI, // Model_info *MI
scale, // FLOAT scale
RL_S_array[L], // int *size_score_array
NoC // int NoC
);
gettimeofday(&tv_calc_a_score_end, nullptr);
tvsub(&tv_calc_a_score_end, &tv_calc_a_score_start, &tv);
time_calc_a_score += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
for(int j = 0; j <NoC; j++) {
int R_S[2]={rm_size_array[level][j*2], rm_size_array[level][j*2+1]};
/* get all good matches */
int GMN;
int *GMPC = get_gmpc(SCORE_array[L][j],thresh,R_S,&GMN);
int RSIZE[2]={MO->MI->rsize[j*2], MO->MI->rsize[j*2+1]};
int GL = (numpart[j]+1)*4+3; //31
/* detected box coordinate(current level) */
FLOAT *t_boxes = (FLOAT*)calloc(GMN*GL,sizeof(FLOAT));
gettimeofday(&tv_box_start, nullptr);
// NO NEED TO USE GPU
for(int k = 0;k < GMN;k++) {
FLOAT *P_temp = t_boxes+GL*k;
int y = GMPC[2*k];
int x = GMPC[2*k+1];
/* calculate root box coordinate */
FLOAT *RB =rootbox(x,y,scale,padx,pady,RSIZE);
memcpy(P_temp, RB,sizeof(FLOAT)*4);
s_free(RB);
P_temp+=4;
for(int pp=0;pp<numpart[j];pp++) {
int PBSIZE[2]={psize[j][pp*2], psize[j][pp*2+1]};
int Isize[2]={pm_size_array[L][MO->MI->pidx[j][pp]*2], pm_size_array[L][MO->MI->pidx[j][pp]*2+1]};
/* calculate part box coordinate */
FLOAT *PB = partbox(x,y,ax_array[L][j][pp],ay_array[L][j][pp],scale,padx,pady,PBSIZE,Ix_array[L][j][pp],Iy_array[L][j][pp],Isize);
memcpy(P_temp, PB,sizeof(FLOAT)*4);
P_temp+=4;
s_free(PB);
}
/* component number and score */
*(P_temp++)=(FLOAT)j; //component number
*(P_temp++)=SCORE_array[L][j][x*R_S[0]+y]; //score of good match
*P_temp = scale;
}
// NO NEED TO USE GPU
gettimeofday(&tv_box_end, nullptr);
tvsub(&tv_box_end, &tv_box_start, &tv);
time_box += tv.tv_sec * 1000.0 + (float)tv.tv_usec / 1000.0;
/* save box information */
if (GMN > 0)
Tboxes[count] = t_boxes;
else
Tboxes[count] = nullptr;
b_nums[count]=GMN;
count++;
detected_boxes+=GMN; //number of detected box
/* release */
s_free(GMPC);
}
////numcom
}
////level
/* free temporary arrays */
free(dst_RL);
free(RL_array);
free(dst_RI);
free(RI_array);
free(dst_OI);
free(OI_array);
free(dst_RL_S);
free(RL_S_array);
free(dst_OFF);
free(OFF_array);
free(dst_SCORE);
free(sub_dst_SCORE);
free(SCORE_array);
free(dst_ax);
free(sub_dst_ax);
free(ax_array);
free(dst_ay);
free(sub_dst_ay);
free(ay_array);
free(Ix_array[0][0][0]);
free(dst_Ix);
free(sub_dst_Ix);
free(Ix_array);
free(Iy_array[0][0][0]);
free(dst_Iy);
free(sub_dst_Iy);
free(Iy_array);
free(scale_array);
gettimeofday(&tv_nucom_end, nullptr);
#ifdef PRINT_INFO
printf("root SCORE : %f\n", time_root_score);
printf("part SCORE : %f\n", time_part_score);
printf("dt : %f\n", time_dt);
printf("calc_a_score : %f\n", time_calc_a_score);
#endif
res = cuCtxSetCurrent(ctx[0]);
if(res != CUDA_SUCCESS) {
printf("cuCtxSetCurrent(ctx[0]) failed: res = %s\n",cuda_response_to_string(res));
exit(1);
}
/* free memory regions */
res = cuMemFreeHost((void *)featp2[0]);
if(res != CUDA_SUCCESS) {
printf("cuMemFreeHost(featp2[0]) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
s_free(featp2);
res = cuMemFreeHost((void *)rootmatch[interval][0]);
if(res != CUDA_SUCCESS) {
printf("cuMemFreeHost(rootmatch[0][0]) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
s_free(rootmatch[0]);
s_free(rootmatch);
if (partmatch != nullptr) {
res = cuMemFreeHost((void *)partmatch[0][0]);
if(res != CUDA_SUCCESS) {
printf("cuMemFreeHost(partmatch[0][0]) failed: res = %s\n", cuda_response_to_string(res));
exit(1);
}
s_free(partmatch[0]);
s_free(partmatch);
s_free(new_PADsize);
}
/* release */
s_free(rm_size_array[0]);
s_free(rm_size_array);
s_free(pm_size_array[0]);
s_free(pm_size_array);
/* Output boundary-box coorinate information */
int GL=(numpart[0]+1)*4+3;
FLOAT *boxes=(FLOAT*)calloc(detected_boxes*GL,sizeof(FLOAT)); //box coordinate information(Temp)
FLOAT *T1 = boxes;
for(int i = 0; i < LofFeat; i++) {
int num_t = b_nums[i]*GL;
if(num_t > 0) {
FLOAT *T2 = Tboxes[i];
//memcpy_s(T1,sizeof(FLOAT)*num_t,T2,sizeof(FLOAT)*num_t);
memcpy(T1, T2,sizeof(FLOAT)*num_t);
T1 += num_t;
}
}
FLOAT abs_threshold = abs(thresh);
/* accumulated score calculation */
FLOAT max_score = 0.0;
/* add offset to accumulated score */
for(int i = 0; i < MO->MI->IM_HEIGHT*MO->MI->IM_WIDTH; i++) {
if (acc_score[i] < thresh) {
acc_score[i] = 0.0;
} else {
acc_score[i] += abs_threshold;
if (acc_score[i] > max_score)
max_score = acc_score[i];
}
}
/* normalization */
if (max_score > 0.0) {
FLOAT ac_ratio = 1.0 / max_score;
for (int i = 0; i < MO->MI->IM_HEIGHT*MO->MI->IM_WIDTH; i++) {
acc_score[i] *= ac_ratio;
}
}
/* release */
free_boxes(Tboxes,LofFeat);
s_free(b_nums);
/* output result */
*detected_count = detected_boxes;
return boxes;
}
value spoc_getCudaDevice(value i)
{
CAMLparam1(i);
CAMLlocal4(general_info, cuda_info, specific_info, gc_info);
CAMLlocal3(device, maxT, maxG);
int nb_devices;
CUdevprop dev_infos;
CUdevice dev;
CUcontext ctx;
CUstream queue[2];
spoc_cu_context *spoc_ctx;
//CUcontext gl_ctx;
char infoStr[1024];
int infoInt;
size_t infoUInt;
int major, minor;
enum cudaError_enum cuda_error;
cuDeviceGetCount (&nb_devices);
if ((Int_val(i)) > nb_devices)
raise_constant(*caml_named_value("no_cuda_device")) ;
CUDA_CHECK_CALL(cuDeviceGet(&dev, Int_val(i)));
CUDA_CHECK_CALL(cuDeviceGetProperties(&dev_infos, dev));
general_info = caml_alloc (9, 0);
CUDA_CHECK_CALL(cuDeviceGetName(infoStr, sizeof(infoStr), dev));
Store_field(general_info,0, copy_string(infoStr));//
CUDA_CHECK_CALL(cuDeviceTotalMem(&infoUInt, dev));
Store_field(general_info,1, Val_int(infoUInt));//
Store_field(general_info,2, Val_int(dev_infos.sharedMemPerBlock));//
Store_field(general_info,3, Val_int(dev_infos.clockRate));//
Store_field(general_info,4, Val_int(dev_infos.totalConstantMemory));//
CUDA_CHECK_CALL(cuDeviceGetAttribute(&infoInt, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT, dev));
Store_field(general_info,5, Val_int(infoInt));//
CUDA_CHECK_CALL(cuDeviceGetAttribute(&infoInt, CU_DEVICE_ATTRIBUTE_ECC_ENABLED, dev));
Store_field(general_info,6, Val_bool(infoInt));//
Store_field(general_info,7, i);
CUDA_CHECK_CALL(cuCtxCreate (&ctx,
CU_CTX_SCHED_BLOCKING_SYNC | CU_CTX_MAP_HOST,
dev));
spoc_ctx = malloc(sizeof(spoc_cl_context));
spoc_ctx->ctx = ctx;
CUDA_CHECK_CALL(cuStreamCreate(&queue[0], 0));
CUDA_CHECK_CALL(cuStreamCreate(&queue[1], 0));
spoc_ctx->queue[0] = queue[0];
spoc_ctx->queue[1] = queue[1];
Store_field(general_info,8, (value)spoc_ctx);
CUDA_CHECK_CALL(cuCtxSetCurrent(ctx));
cuda_info = caml_alloc(1, 0); //0 -> Cuda
specific_info = caml_alloc(18, 0);
cuDeviceComputeCapability(&major, &minor, dev);
Store_field(specific_info,0, Val_int(major));//
Store_field(specific_info,1, Val_int(minor));//
Store_field(specific_info,2, Val_int(dev_infos.regsPerBlock));//
Store_field(specific_info,3, Val_int(dev_infos.SIMDWidth));//
Store_field(specific_info,4, Val_int(dev_infos.memPitch));//
Store_field(specific_info,5, Val_int(dev_infos.maxThreadsPerBlock));//
maxT = caml_alloc(3, 0);
Store_field(maxT,0, Val_int(dev_infos.maxThreadsDim[0]));//
Store_field(maxT,1, Val_int(dev_infos.maxThreadsDim[1]));//
Store_field(maxT,2, Val_int(dev_infos.maxThreadsDim[2]));//
Store_field(specific_info,6, maxT);
maxG = caml_alloc(3, 0);
Store_field(maxG,0, Val_int(dev_infos.maxGridSize[0]));//
Store_field(maxG,1, Val_int(dev_infos.maxGridSize[1]));//
Store_field(maxG,2, Val_int(dev_infos.maxGridSize[2]));//
Store_field(specific_info,7, maxG);
Store_field(specific_info,8, Val_int(dev_infos.textureAlign));//
cuDeviceGetAttribute(&infoInt, CU_DEVICE_ATTRIBUTE_GPU_OVERLAP, dev);
Store_field(specific_info,9, Val_bool(infoInt));//
cuDeviceGetAttribute(&infoInt, CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT, dev);
Store_field(specific_info,10, Val_bool(infoInt));//
cuDeviceGetAttribute(&infoInt, CU_DEVICE_ATTRIBUTE_INTEGRATED, dev);
Store_field(specific_info,11, Val_bool(infoInt));//
cuDeviceGetAttribute(&infoInt, CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY, dev);
Store_field(specific_info,12, Val_bool(infoInt));//
cuDeviceGetAttribute(&infoInt, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE, dev);
Store_field(specific_info,13, Val_int(infoInt));//
cuDeviceGetAttribute(&infoInt, CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS, dev);
Store_field(specific_info,14, Val_bool(infoInt));//
cuDeviceGetAttribute(&infoInt, CU_DEVICE_ATTRIBUTE_PCI_BUS_ID, dev);
Store_field(specific_info,15, Val_int(infoInt));
cuDeviceGetAttribute(&infoInt, CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID, dev);
Store_field(specific_info,16, Val_int(infoInt));
cuDriverGetVersion(&infoInt);
Store_field(specific_info, 17, Val_int(infoInt));
Store_field(cuda_info, 0, specific_info);
device = caml_alloc(4, 0);
Store_field(device, 0, general_info);
Store_field(device, 1, cuda_info);
{spoc_cuda_gc_info* gcInfo = (spoc_cuda_gc_info*)malloc(sizeof(spoc_cuda_gc_info));
CUDA_CHECK_CALL(cuMemGetInfo(&infoUInt, NULL));
infoUInt -= (32*1024*1024);
Store_field(device, 2, (value)gcInfo);
{cuda_event_list* events = NULL;
Store_field(device, 3, (value)events);
CAMLreturn(device);}}
}
/*
* Create a VampirTrace CUPTI context. If the CUDA context is not given, the
* current context will be requested and used.
*
* @param cuCtx CUDA context
* @param cuDev CUDA device
* @param ctxID ID of the CUDA context
* @param devID ID of the CUDA device
*
* @return pointer to created VampirTrace CUPTI context
*/
vt_cupti_ctx_t* vt_cupti_createCtx(CUcontext cuCtx, CUdevice cuDev,
uint32_t cuCtxID, uint32_t cuDevID)
{
vt_cupti_ctx_t* vtCtx = NULL;
/* create new context */
vtCtx = (vt_cupti_ctx_t *)malloc(sizeof(vt_cupti_ctx_t));
if(vtCtx == NULL)
vt_error_msg("[CUPTI] Could not allocate memory for VT CUPTI context!");
vtCtx->ctxID = cuCtxID;
#if (defined(VT_CUPTI_ACTIVITY) || defined(VT_CUPTI_CALLBACKS))
vtCtx->gpuMemAllocated = 0;
vtCtx->gpuMemList = NULL;
vtCtx->strmList = NULL;
#endif
vtCtx->next = NULL;
VT_CHECK_THREAD;
vtCtx->ptid = VT_MY_THREAD;
/* try to get CUDA device (ID), if they are not given */
if(cuDevID == VT_CUPTI_NO_DEVICE_ID){
if(cuDev == VT_CUPTI_NO_CUDA_DEVICE){
CUcontext cuCurrCtx;
if(cuCtx != NULL){
cuCtxGetCurrent(&cuCurrCtx);
/* if given context does not match the current one, get the device for
the given one */
if(cuCtx != cuCurrCtx)
VT_CUDRV_CALL(cuCtxSetCurrent(cuCtx), NULL);
}
if(CUDA_SUCCESS == cuCtxGetDevice(&cuDev))
cuDevID = (uint32_t)cuDev;
/* reset the active context */
if(cuCtx != NULL && cuCtx != cuCurrCtx)
VT_CUDRV_CALL(cuCtxSetCurrent(cuCurrCtx), NULL);
}else{
/* no device ID, but CUDA device is given */
cuDevID = (uint32_t)cuDev;
}
}
vtCtx->devID = cuDevID;
vtCtx->cuDev = cuDev;
/* get the current CUDA context, if it is not given */
if(cuCtx == NULL)
VT_CUDRV_CALL(cuCtxGetCurrent(&cuCtx), NULL);
/* set the CUDA context */
vtCtx->cuCtx = cuCtx;
#if defined(VT_CUPTI_ACTIVITY)
vtCtx->activity = NULL;
#endif
#if defined(VT_CUPTI_CALLBACKS)
vtCtx->callbacks = NULL;
#endif
#if defined(VT_CUPTI_EVENTS)
vtCtx->events = NULL;
#endif
vt_cntl_msg(2, "[CUPTI] Created context for CUcontext %d, CUdevice %d",
cuCtx, cuDev);
return vtCtx;
}
static void vq_handle_output(VirtIODevice *vdev, VirtQueue *vq)
{
VirtQueueElement elem;
while(virtqueue_pop(vq, &elem)) {
struct param *p = elem.out_sg[0].iov_base;
//for all library routines: get required arguments from buffer, execute, and push results back in virtqueue
switch (p->syscall_type) {
case CUINIT: {
p->result = cuInit(p->flags);
break;
}
case CUDRIVERGETVERSION: {
p->result = cuDriverGetVersion(&p->val1);
break;
}
case CUDEVICEGETCOUNT: {
p->result = cuDeviceGetCount(&p->val1);
break;
}
case CUDEVICEGET: {
p->result = cuDeviceGet(&p->device, p->val1);
break;
}
case CUDEVICECOMPUTECAPABILITY: {
p->result = cuDeviceComputeCapability(&p->val1, &p->val2, p->device);
break;
}
case CUDEVICEGETNAME: {
p->result = cuDeviceGetName(elem.in_sg[0].iov_base, p->val1, p->device);
break;
}
case CUDEVICEGETATTRIBUTE: {
p->result = cuDeviceGetAttribute(&p->val1, p->attrib, p->device);
break;
}
case CUCTXCREATE: {
p->result = cuCtxCreate(&p->ctx, p->flags, p->device);
break;
}
case CUCTXDESTROY: {
p->result = cuCtxDestroy(p->ctx);
break;
}
case CUCTXGETCURRENT: {
p->result = cuCtxGetCurrent(&p->ctx);
break;
}
case CUCTXGETDEVICE: {
p->result = cuCtxGetDevice(&p->device);
break;
}
case CUCTXPOPCURRENT: {
p->result = cuCtxPopCurrent(&p->ctx);
break;
}
case CUCTXSETCURRENT: {
p->result = cuCtxSetCurrent(p->ctx);
break;
}
case CUCTXSYNCHRONIZE: {
p->result = cuCtxSynchronize();
break;
}
case CUMODULELOAD: {
//hardcoded path - needs improvement
//all .cubin files should be stored in $QEMU_NFS_PATH - currently $QEMU_NFS_PATH is shared between host and guest with NFS
char *binname = malloc((strlen((char *)elem.out_sg[1].iov_base)+strlen(getenv("QEMU_NFS_PATH")+1))*sizeof(char));
if (!binname) {
p->result = 0;
virtqueue_push(vq, &elem, 0);
break;
}
strcpy(binname, getenv("QEMU_NFS_PATH"));
strcat(binname, (char *)elem.out_sg[1].iov_base);
//change current CUDA context
//each CUDA contets has its own virtual memory space - isolation is ensured by switching contexes
if (cuCtxSetCurrent(p->ctx) != 0) {
p->result = 999;
break;
}
p->result = cuModuleLoad(&p->module, binname);
free(binname);
break;
}
case CUMODULEGETGLOBAL: {
char *name = malloc(100*sizeof(char));
if (!name) {
p->result = 999;
break;
}
strcpy(name, (char *)elem.out_sg[1].iov_base);
p->result = cuModuleGetGlobal(&p->dptr,&p->size1,p->module,(const char *)name);
break;
}
case CUMODULEUNLOAD: {
p->result = cuModuleUnload(p->module);
break;
}
case CUMEMALLOC: {
if (cuCtxSetCurrent(p->ctx) != 0) {
p->result = 999;
break;
}
p->result = cuMemAlloc(&p->dptr, p->bytesize);
break;
}
case CUMEMALLOCPITCH: {
if (cuCtxSetCurrent(p->ctx) != 0) {
p->result = 999;
break;
}
p->result = cuMemAllocPitch(&p->dptr, &p->size3, p->size1, p->size2, p->bytesize);
break;
}
//large buffers are alocated in smaller chuncks in guest kernel space
//gets each chunck seperately and copies it to device memory
case CUMEMCPYHTOD: {
int i;
size_t offset;
unsigned long s, nr_pages = p->nr_pages;
if (cuCtxSetCurrent(p->ctx) != 0) {
p->result = 999;
break;
}
offset = 0;
for (i=0; i<nr_pages; i++) {
s = *(long *)elem.out_sg[1+2*i+1].iov_base;
p->result = cuMemcpyHtoD(p->dptr+offset, elem.out_sg[1+2*i].iov_base, s);
if (p->result != 0) break;
offset += s;
}
break;
}
case CUMEMCPYHTODASYNC: {
int i;
size_t offset;
unsigned long s, nr_pages = p->nr_pages;
if (cuCtxSetCurrent(p->ctx) != 0) {
p->result = 999;
break;
}
offset = 0;
for (i=0; i<nr_pages; i++) {
s = *(long *)elem.out_sg[1+2*i+1].iov_base;
p->result = cuMemcpyHtoDAsync(p->dptr+offset, elem.out_sg[1+2*i].iov_base, s, p->stream);
if (p->result != 0) break;
offset += s;
}
break;
}
case CUMEMCPYDTODASYNC: {
p->result = cuMemcpyDtoDAsync(p->dptr, p->dptr1, p->size1, p->stream);
break;
}
case CUMEMCPYDTOH: {
int i;
unsigned long s, nr_pages = p->nr_pages;
if (cuCtxSetCurrent(p->ctx) != 0) {
p->result = 999;
break;
}
size_t offset = 0;
for (i=0; i<nr_pages; i++) {
s = *(long *)elem.in_sg[0+2*i+1].iov_base;
p->result = cuMemcpyDtoH(elem.in_sg[0+2*i].iov_base, p->dptr+offset, s);
if (p->result != 0) break;
offset += s;
}
break;
}
case CUMEMCPYDTOHASYNC: {
int i;
unsigned long s, nr_pages = p->nr_pages;
if (cuCtxSetCurrent(p->ctx) != 0) {
p->result = 999;
break;
}
size_t offset = 0;
for (i=0; i<nr_pages; i++) {
s = *(long *)elem.in_sg[0+2*i+1].iov_base;
p->result = cuMemcpyDtoHAsync(elem.in_sg[0+2*i].iov_base, p->dptr+offset, s, p->stream);
if (p->result != 0) break;
offset += s;
}
break;
}
case CUMEMSETD32: {
p->result = cuMemsetD32(p->dptr, p->bytecount, p->bytesize);
break;
}
case CUMEMFREE: {
p->result = cuMemFree(p->dptr);
break;
}
case CUMODULEGETFUNCTION: {
char *name = (char *)elem.out_sg[1].iov_base;
name[p->length] = '\0';
if (cuCtxSetCurrent(p->ctx) != 0) {
p->result = 999;
break;
}
p->result = cuModuleGetFunction(&p->function, p->module, name);
break;
}
case CULAUNCHKERNEL: {
void **args = malloc(p->val1*sizeof(void *));
if (!args) {
p->result = 9999;
break;
}
int i;
for (i=0; i<p->val1; i++) {
args[i] = elem.out_sg[1+i].iov_base;
}
if (cuCtxSetCurrent(p->ctx) != 0) {
p->result = 999;
break;
}
p->result = cuLaunchKernel(p->function,
p->gridDimX, p->gridDimY, p->gridDimZ,
p->blockDimX, p->blockDimY, p->blockDimZ,
p->bytecount, 0, args, 0);
free(args);
break;
}
case CUEVENTCREATE: {
p->result = cuEventCreate(&p->event1, p->flags);
break;
}
case CUEVENTDESTROY: {
p->result = cuEventDestroy(p->event1);
break;
}
case CUEVENTRECORD: {
p->result = cuEventRecord(p->event1, p->stream);
break;
}
case CUEVENTSYNCHRONIZE: {
p->result = cuEventSynchronize(p->event1);
break;
}
case CUEVENTELAPSEDTIME: {
p->result = cuEventElapsedTime(&p->pMilliseconds, p->event1, p->event2);
break;
}
case CUSTREAMCREATE: {
p->result = cuStreamCreate(&p->stream, 0);
break;
}
case CUSTREAMSYNCHRONIZE: {
p->result = cuStreamSynchronize(p->stream);
break;
}
case CUSTREAMQUERY: {
p->result = cuStreamQuery(p->stream);
break;
}
case CUSTREAMDESTROY: {
p->result = cuStreamDestroy(p->stream);
break;
}
default:
printf("Unknown syscall_type\n");
}
virtqueue_push(vq, &elem, 0);
}
//notify frontend - trigger virtual interrupt
virtio_notify(vdev, vq);
return;
}
