mirror/darknet
1
0
Fork 0
mirror of https://github.com/pjreddie/darknet.git synced 2023-08-10 21:13:14 +03:00
Code Issues Projects Releases Wiki Activity

I am so done with opencl, switching to cuda

Browse Source
This commit is contained in:
Joseph Redmon 2015-01-20 13:26:46 -08:00
parent 6e1d5b45de
commit 4ac78c8926
3 changed files with 19 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/cnn.c
View File

@ -210,7 +210,7 @@ void train_imagenet(char *cfgfile)
//network net = parse_network_cfg("/home/pjreddie/imagenet_backup/alexnet_1270.cfg"); //network net = parse_network_cfg("/home/pjreddie/imagenet_backup/alexnet_1270.cfg");
srand(time(0)); srand(time(0));
network net = parse_network_cfg(cfgfile); network net = parse_network_cfg(cfgfile);
set_learning_network(&net, net.learning_rate*10., net.momentum, net.decay); set_learning_network(&net, net.learning_rate*100., net.momentum, net.decay);
printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay); printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
int imgs = 1024; int imgs = 1024;
int i = 6600; int i = 6600;

6
src/gemm.c
View File

@ -164,8 +164,7 @@ cl_kernel get_gemm_nn_kernel()
#define TILE 64 #define TILE 64
#define TILE_K 16 #define TILE_K 16
#define WPT 8 #define THREADS 64
#define THREADS (TILE*TILE)/(WPT*WPT)
cl_kernel get_gemm_nn_fast_kernel() cl_kernel get_gemm_nn_fast_kernel()
{ {
@ -175,7 +174,6 @@ cl_kernel get_gemm_nn_fast_kernel()
gemm_kernel = get_kernel("src/gemm_fast.cl", "gemm_nn_fast", "-D TILE=" STR(TILE) gemm_kernel = get_kernel("src/gemm_fast.cl", "gemm_nn_fast", "-D TILE=" STR(TILE)
" -cl-nv-verbose " " -cl-nv-verbose "
" -D TILE_K=" STR(TILE_K) " -D TILE_K=" STR(TILE_K)
" -D WPT=" STR(WPT)
" -D THREADS=" STR(THREADS)); " -D THREADS=" STR(THREADS));
init = 1; init = 1;
} }
@ -464,7 +462,6 @@ void test_gpu_blas()
test_gpu_accuracy(0,0,128,128,128); test_gpu_accuracy(0,0,128,128,128);
/*
time_ongpu(0,0,64,2916,363); time_ongpu(0,0,64,2916,363);
time_ongpu_fast(0,0,64,2916,363); time_ongpu_fast(0,0,64,2916,363);
time_ongpu(0,0,64,2916,363); time_ongpu(0,0,64,2916,363);
@ -483,7 +480,6 @@ void test_gpu_blas()
time_ongpu_fast(0,0,128,4096,12544); time_ongpu_fast(0,0,128,4096,12544);
time_ongpu(0,0,128,4096,4096); time_ongpu(0,0,128,4096,4096);
time_ongpu_fast(0,0,128,4096,4096); time_ongpu_fast(0,0,128,4096,4096);
*/
// time_ongpu(1,0,2304,196,256); // time_ongpu(1,0,2304,196,256);
// time_ongpu_fast(1,0,2304,196,256); // time_ongpu_fast(1,0,2304,196,256);
// time_ongpu(0,1,256,2304,196); // time_ongpu(0,1,256,2304,196);

37
src/gemm_fast.cl
View File

@ -16,16 +16,15 @@ __kernel void gemm_nn_fast(int TA, int TB, int M, int N, int K, float ALPHA,
int ctile = get_group_id(0); int ctile = get_group_id(0);
int rtile = get_group_id(1); int rtile = get_group_id(1);
float Breg; float Areg[TILE];
float Areg[WPT]; float acc[TILE][TILE/THREADS];
float acc[WPT][WPT];
A += rtile*TILE*lda; A += rtile*TILE*lda;
B += ctile*TILE; B += ctile*TILE;
C += rtile*TILE*ldc + ctile*TILE; C += rtile*TILE*ldc + ctile*TILE;
for(i = 0; i < WPT; ++i){ for(i = 0; i < TILE; ++i){
for(j = 0; j < WPT; ++j){ for(j = 0; j < TILE/THREADS; ++j){
acc[i][j] = 0; acc[i][j] = 0;
} }
} }
@ -51,28 +50,26 @@ __kernel void gemm_nn_fast(int TA, int TB, int M, int N, int K, float ALPHA,
barrier(CLK_LOCAL_MEM_FENCE); barrier(CLK_LOCAL_MEM_FENCE);
for(k = 0; k < TILE_K; ++k){ for(k = 0; k < TILE_K; ++k){
for(y = 0; y < WPT; ++y){ #pragma unroll
int row = (offset + (y*WPT)*THREADS)/TILE; for(y = 0; y < TILE; ++y){
//Areg[y] = Asub[y*WPT][k]; Areg[y] = Asub[y][k];
} }
for(y = 0; y < WPT; ++y){ for(x = 0; x < TILE; x += THREADS){
for(x = 0; x < WPT; ++x){ float Breg = Bsub[k][x+offset];
int index = offset + (y*WPT + x)*THREADS; #pragma unroll
int row = index / TILE; for(y = 0; y < TILE; ++y){
int col = index % TILE; acc[y][x/THREADS] += Breg * Areg[y];
acc[y][x] += Asub[row][k]*Bsub[k][col];
} }
} }
} }
barrier(CLK_LOCAL_MEM_FENCE); barrier(CLK_LOCAL_MEM_FENCE);
} }
for(y = 0; y < WPT; ++y){ for(i = 0; i < TILE; ++i){
for(x = 0; x < WPT; ++x){ for(j = 0; j < TILE/THREADS; ++j){
int index = offset + (y*WPT + x)*THREADS; int col = j*THREADS + offset;
int row = index / TILE; int row = i;
int col = index % TILE; C[row*ldc+col] = ALPHA*acc[i][j] + BETA*C[row*ldc+col];
C[row*ldc+col] = ALPHA*acc[y][x] + BETA*C[row*ldc+col];
} }
} }
} }