From ae43c2bc32fbb838bfebeeaf2c2b058ccab5c83c Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@burninator.cs.washington.edu>
Date: Thu, 23 Jun 2016 05:31:14 +0000
Subject: [PATCH] hi
---
src/gemm.c | 56 ++++++++++++++++++++++++++++++++++++++++++++------------
1 files changed, 44 insertions(+), 12 deletions(-)
diff --git a/src/gemm.c b/src/gemm.c
index a923beb..3003be0 100644
--- a/src/gemm.c
+++ b/src/gemm.c
@@ -5,6 +5,28 @@
#include <stdio.h>
#include <math.h>
+void gemm_bin(int M, int N, int K, float ALPHA,
+ char *A, int lda,
+ float *B, int ldb,
+ float *C, int ldc)
+{
+ int i,j,k;
+ for(i = 0; i < M; ++i){
+ for(k = 0; k < K; ++k){
+ char A_PART = A[i*lda+k];
+ if(A_PART){
+ for(j = 0; j < N; ++j){
+ C[i*ldc+j] += B[k*ldb+j];
+ }
+ } else {
+ for(j = 0; j < N; ++j){
+ C[i*ldc+j] -= B[k*ldb+j];
+ }
+ }
+ }
+ }
+}
+
float *random_matrix(int rows, int cols)
{
int i;
@@ -151,7 +173,7 @@
{
cublasHandle_t handle = blas_handle();
cudaError_t status = cublasSgemm(handle, (TB ? CUBLAS_OP_T : CUBLAS_OP_N),
- (TA ? CUBLAS_OP_T : CUBLAS_OP_N), N, M, K, &ALPHA, B_gpu, ldb, A_gpu, lda, &BETA, C_gpu, ldc);
+ (TA ? CUBLAS_OP_T : CUBLAS_OP_N), N, M, K, &ALPHA, B_gpu, ldb, A_gpu, lda, &BETA, C_gpu, ldc);
check_error(status);
}
@@ -276,6 +298,7 @@
int test_gpu_blas()
{
+ /*
test_gpu_accuracy(0,0,10,576,75);
test_gpu_accuracy(0,0,17,10,10);
@@ -288,19 +311,28 @@
test_gpu_accuracy(0,1,1000,10,100);
test_gpu_accuracy(1,1,1000,10,100);
- test_gpu_accuracy(0,0,10,10,10);
+ test_gpu_accuracy(0,0,10,10,10);
- time_ongpu(0,0,64,2916,363);
- time_ongpu(0,0,64,2916,363);
- time_ongpu(0,0,64,2916,363);
- time_ongpu(0,0,192,729,1600);
- time_ongpu(0,0,384,196,1728);
- time_ongpu(0,0,256,196,3456);
- time_ongpu(0,0,256,196,2304);
- time_ongpu(0,0,128,4096,12544);
- time_ongpu(0,0,128,4096,4096);
+ time_ongpu(0,0,64,2916,363);
+ time_ongpu(0,0,64,2916,363);
+ time_ongpu(0,0,64,2916,363);
+ time_ongpu(0,0,192,729,1600);
+ time_ongpu(0,0,384,196,1728);
+ time_ongpu(0,0,256,196,3456);
+ time_ongpu(0,0,256,196,2304);
+ time_ongpu(0,0,128,4096,12544);
+ time_ongpu(0,0,128,4096,4096);
+ */
+ time_ongpu(0,0,64,75,12544);
+ time_ongpu(0,0,64,75,12544);
+ time_ongpu(0,0,64,75,12544);
+ time_ongpu(0,0,64,576,12544);
+ time_ongpu(0,0,256,2304,784);
+ time_ongpu(1,1,2304,256,784);
+ time_ongpu(0,0,512,4608,196);
+ time_ongpu(1,1,4608,512,196);
-return 0;
+ return 0;
}
#endif
--
Gitblit v1.10.0