From 354b0cbdcb48a66de1af3aca9fc2f5687e46ee42 Mon Sep 17 00:00:00 2001 From: Joseph Redmon Date: Sat, 19 Apr 2014 13:10:59 -0700 Subject: [PATCH] GPU flag in Makefile --- Makefile | 10 +++++- src/gpu_gemm.c | 83 ++++++++++++++++++++++++++++++++++++++++++++++ src/mini_blas.c | 88 +------------------------------------------------ src/tests.c | 6 ++-- 4 files changed, 96 insertions(+), 91 deletions(-) diff --git a/Makefile b/Makefile index 68ed3ee1..2c47fdfb 100644 --- a/Makefile +++ b/Makefile @@ -1,21 +1,29 @@ CC=gcc +GPU=0 COMMON=-Wall `pkg-config --cflags opencv` -I/usr/local/cuda/include/ UNAME = $(shell uname) OPTS=-O3 ifeq ($(UNAME), Darwin) COMMON+= -isystem /usr/local/Cellar/opencv/2.4.6.1/include/opencv -isystem /usr/local/Cellar/opencv/2.4.6.1/include +ifeq ($(GPU), 1) LDFLAGS= -framework OpenCL +endif else OPTS+= -march=native +ifeq ($(GPU), 1) LDFLAGS= -lOpenCL endif +endif CFLAGS= $(COMMON) $(OPTS) #CFLAGS= $(COMMON) -O0 -g LDFLAGS+=`pkg-config --libs opencv` -lm VPATH=./src/ EXEC=cnn -OBJ=network.o image.o tests.o connected_layer.o maxpool_layer.o activations.o list.o option_list.o parser.o utils.o data.o matrix.o softmax_layer.o mini_blas.o convolutional_layer.o opencl.o gpu_gemm.o cpu_gemm.o normalization_layer.o +OBJ=network.o image.o tests.o connected_layer.o maxpool_layer.o activations.o list.o option_list.o parser.o utils.o data.o matrix.o softmax_layer.o mini_blas.o convolutional_layer.o cpu_gemm.o normalization_layer.o +ifeq ($(GPU), 1) +OBJ+=gpu_gemm.o opencl.o +endif all: $(EXEC) diff --git a/src/gpu_gemm.c b/src/gpu_gemm.c index 26bb3fea..4a8aaca8 100644 --- a/src/gpu_gemm.c +++ b/src/gpu_gemm.c @@ -83,6 +83,89 @@ void gpu_gemm(int TA, int TB, int M, int N, int K, float ALPHA, } +void time_gpu_random_matrix(int TA, int TB, int m, int k, int n) +{ + float *a; + if(!TA) a = random_matrix(m,k); + else a = random_matrix(k,m); + int lda = (!TA)?k:m; + float *b; + if(!TB) b = random_matrix(k,n); + else b = random_matrix(n,k); + int ldb = (!TB)?n:k; + + float *c = random_matrix(m,n); + int i; + clock_t start = clock(), end; + for(i = 0; i<1000; ++i){ + gpu_gemm(TA,TB,m,n,k,1,a,lda,b,ldb,1,c,n); + } + end = clock(); + printf("Matrix Multiplication %dx%d * %dx%d, TA=%d, TB=%d: %lf ms\n",m,k,k,n, TA, TB, (float)(end-start)/CLOCKS_PER_SEC); + free(a); + free(b); + free(c); +} + +void test_gpu_accuracy(int TA, int TB, int m, int k, int n) +{ + srand(0); + float *a; + if(!TA) a = random_matrix(m,k); + else a = random_matrix(k,m); + int lda = (!TA)?k:m; + float *b; + if(!TB) b = random_matrix(k,n); + else b = random_matrix(n,k); + int ldb = (!TB)?n:k; + + float *c = random_matrix(m,n); + float *c_gpu = random_matrix(m,n); + memset(c, 0, m*n*sizeof(float)); + memset(c_gpu, 0, m*n*sizeof(float)); + int i; + //pm(m,k,b); + gpu_gemm(TA,TB,m,n,k,1,a,lda,b,ldb,1,c_gpu,n); + //pm(m, n, c_gpu); + cpu_gemm(TA,TB,m,n,k,1,a,lda,b,ldb,1,c,n); + //pm(m, n, c); + double sse = 0; + for(i = 0; i < m*n; ++i) { + //printf("%f %f\n", c[i], c_gpu[i]); + sse += pow(c[i]-c_gpu[i], 2); + } + printf("Matrix Multiplication %dx%d * %dx%d, TA=%d, TB=%d: %g MSE\n",m,k,k,n, TA, TB, sse/(m*n)); + free(a); + free(b); + free(c); +} + +void test_gpu_blas() +{ + test_gpu_accuracy(0,0,17,10,10); + test_gpu_accuracy(1,0,17,10,10); + test_gpu_accuracy(0,1,17,10,10); + test_gpu_accuracy(1,1,17,10,10); + + test_gpu_accuracy(0,0,1000,10,100); + test_gpu_accuracy(1,0,1000,10,100); + test_gpu_accuracy(0,1,1000,10,100); + test_gpu_accuracy(1,1,1000,10,100); + + time_gpu_random_matrix(0,0,1000,1000,100); + time_random_matrix(0,0,1000,1000,100); + + time_gpu_random_matrix(0,1,1000,1000,100); + time_random_matrix(0,1,1000,1000,100); + + time_gpu_random_matrix(1,0,1000,1000,100); + time_random_matrix(1,0,1000,1000,100); + + time_gpu_random_matrix(1,1,1000,1000,100); + time_random_matrix(1,1,1000,1000,100); + +} + /* cl_kernel get_gemm_kernel_slow() { diff --git a/src/mini_blas.c b/src/mini_blas.c index bac3e226..4c7d3d06 100644 --- a/src/mini_blas.c +++ b/src/mini_blas.c @@ -24,7 +24,7 @@ void gemm(int TA, int TB, int M, int N, int K, float ALPHA, float BETA, float *C, int ldc) { - gpu_gemm( TA, TB, M, N, K, ALPHA,A,lda, B, ldb,BETA,C,ldc); + cpu_gemm( TA, TB, M, N, K, ALPHA,A,lda, B, ldb,BETA,C,ldc); } void im2row(float *image, int h, int w, int c, int size, int stride, float *matrix) @@ -155,91 +155,5 @@ void test_blas() time_random_matrix(1,0,1000,100,100); time_random_matrix(0,1,1000,100,100); time_random_matrix(1,1,1000,100,100); - - } -void time_gpu_random_matrix(int TA, int TB, int m, int k, int n) -{ - float *a; - if(!TA) a = random_matrix(m,k); - else a = random_matrix(k,m); - int lda = (!TA)?k:m; - float *b; - if(!TB) b = random_matrix(k,n); - else b = random_matrix(n,k); - int ldb = (!TB)?n:k; - - float *c = random_matrix(m,n); - int i; - clock_t start = clock(), end; - for(i = 0; i<1000; ++i){ - gpu_gemm(TA,TB,m,n,k,1,a,lda,b,ldb,1,c,n); - } - end = clock(); - printf("Matrix Multiplication %dx%d * %dx%d, TA=%d, TB=%d: %lf ms\n",m,k,k,n, TA, TB, (float)(end-start)/CLOCKS_PER_SEC); - free(a); - free(b); - free(c); -} - -void test_gpu_accuracy(int TA, int TB, int m, int k, int n) -{ - srand(0); - float *a; - if(!TA) a = random_matrix(m,k); - else a = random_matrix(k,m); - int lda = (!TA)?k:m; - float *b; - if(!TB) b = random_matrix(k,n); - else b = random_matrix(n,k); - int ldb = (!TB)?n:k; - - float *c = random_matrix(m,n); - float *c_gpu = random_matrix(m,n); - memset(c, 0, m*n*sizeof(float)); - memset(c_gpu, 0, m*n*sizeof(float)); - int i; - //pm(m,k,b); - gpu_gemm(TA,TB,m,n,k,1,a,lda,b,ldb,1,c_gpu,n); - //pm(m, n, c_gpu); - cpu_gemm(TA,TB,m,n,k,1,a,lda,b,ldb,1,c,n); - //pm(m, n, c); - double sse = 0; - for(i = 0; i < m*n; ++i) { - //printf("%f %f\n", c[i], c_gpu[i]); - sse += pow(c[i]-c_gpu[i], 2); - } - printf("Matrix Multiplication %dx%d * %dx%d, TA=%d, TB=%d: %g MSE\n",m,k,k,n, TA, TB, sse/(m*n)); - free(a); - free(b); - free(c); -} - -void test_gpu_blas() -{ - test_gpu_accuracy(0,0,17,10,10); - test_gpu_accuracy(1,0,17,10,10); - test_gpu_accuracy(0,1,17,10,10); - test_gpu_accuracy(1,1,17,10,10); - - test_gpu_accuracy(0,0,1000,10,100); - test_gpu_accuracy(1,0,1000,10,100); - test_gpu_accuracy(0,1,1000,10,100); - test_gpu_accuracy(1,1,1000,10,100); - - time_gpu_random_matrix(0,0,1000,1000,100); - time_random_matrix(0,0,1000,1000,100); - - time_gpu_random_matrix(0,1,1000,1000,100); - time_random_matrix(0,1,1000,1000,100); - - time_gpu_random_matrix(1,0,1000,1000,100); - time_random_matrix(1,0,1000,1000,100); - - time_gpu_random_matrix(1,1,1000,1000,100); - time_random_matrix(1,1,1000,1000,100); - -} - - diff --git a/src/tests.c b/src/tests.c index e91fb63c..851d781c 100644 --- a/src/tests.c +++ b/src/tests.c @@ -789,9 +789,9 @@ int main(int argc, char *argv[]) //test_vince(); //test_full(); //train_VOC(); - features_VOC_image(argv[1], argv[2], argv[3], 0); - features_VOC_image(argv[1], argv[2], argv[3], 1); - //features_VOC_image_size(argv[1], atoi(argv[2]), atoi(argv[3])); + //features_VOC_image(argv[1], argv[2], argv[3], 0); + //features_VOC_image(argv[1], argv[2], argv[3], 1); + features_VOC_image_size(argv[1], atoi(argv[2]), atoi(argv[3])); //visualize_imagenet_features("data/assira/train.list"); //visualize_imagenet_topk("data/VOC2012.list"); //visualize_cat();