From b8e6e80c6d411d05a9e09f1e3676eb9a7f3ea0e8 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Fri, 03 Aug 2018 11:35:03 +0000
Subject: [PATCH] Added spatial Yolo v3 yolov3-spp.cfg
---
src/cuda.c | 49 +++++++++++++++++++++++++++++++++++++++++--------
1 files changed, 41 insertions(+), 8 deletions(-)
diff --git a/src/cuda.c b/src/cuda.c
index 617e7b3..2284dad 100644
--- a/src/cuda.c
+++ b/src/cuda.c
@@ -26,25 +26,26 @@
void check_error(cudaError_t status)
{
+ //cudaDeviceSynchronize();
cudaError_t status2 = cudaGetLastError();
if (status != cudaSuccess)
- {
+ {
const char *s = cudaGetErrorString(status);
char buffer[256];
printf("CUDA Error: %s\n", s);
assert(0);
snprintf(buffer, 256, "CUDA Error: %s", s);
error(buffer);
- }
+ }
if (status2 != cudaSuccess)
- {
+ {
const char *s = cudaGetErrorString(status);
char buffer[256];
printf("CUDA Error Prev: %s\n", s);
assert(0);
snprintf(buffer, 256, "CUDA Error Prev: %s", s);
error(buffer);
- }
+ }
}
dim3 cuda_gridsize(size_t n){
@@ -60,6 +61,28 @@
return d;
}
+static cudaStream_t streamsArray[16]; // cudaStreamSynchronize( get_cuda_stream() );
+static int streamInit[16] = { 0 };
+
+cudaStream_t get_cuda_stream() {
+ int i = cuda_get_device();
+ if (!streamInit[i]) {
+ cudaError_t status = cudaStreamCreate(&streamsArray[i]);
+ //cudaError_t status = cudaStreamCreateWithFlags(&streamsArray[i], cudaStreamNonBlocking);
+ if (status != cudaSuccess) {
+ printf(" cudaStreamCreate error: %d \n", status);
+ const char *s = cudaGetErrorString(status);
+ char buffer[256];
+ printf("CUDA Error: %s\n", s);
+ status = cudaStreamCreateWithFlags(&streamsArray[i], cudaStreamDefault);
+ check_error(status);
+ }
+ streamInit[i] = 1;
+ }
+ return streamsArray[i];
+}
+
+
#ifdef CUDNN
cudnnHandle_t cudnn_handle()
{
@@ -69,6 +92,7 @@
if(!init[i]) {
cudnnCreate(&handle[i]);
init[i] = 1;
+ cudnnStatus_t status = cudnnSetStream(handle[i], get_cuda_stream());
}
return handle[i];
}
@@ -81,6 +105,7 @@
int i = cuda_get_device();
if(!init[i]) {
cublasCreate(&handle[i]);
+ cublasStatus_t status = cublasSetStream(handle[i], get_cuda_stream());
init[i] = 1;
}
return handle[i];
@@ -93,7 +118,8 @@
cudaError_t status = cudaMalloc((void **)&x_gpu, size);
check_error(status);
if(x){
- status = cudaMemcpy(x_gpu, x, size, cudaMemcpyHostToDevice);
+ //status = cudaMemcpy(x_gpu, x, size, cudaMemcpyHostToDevice);
+ status = cudaMemcpyAsync(x_gpu, x, size, cudaMemcpyHostToDevice, get_cuda_stream());
check_error(status);
}
if(!x_gpu) error("Cuda malloc failed\n");
@@ -138,6 +164,7 @@
void cuda_free(float *x_gpu)
{
+ //cudaStreamSynchronize(get_cuda_stream());
cudaError_t status = cudaFree(x_gpu);
check_error(status);
}
@@ -145,15 +172,21 @@
void cuda_push_array(float *x_gpu, float *x, size_t n)
{
size_t size = sizeof(float)*n;
- cudaError_t status = cudaMemcpy(x_gpu, x, size, cudaMemcpyHostToDevice);
+ //cudaError_t status = cudaMemcpy(x_gpu, x, size, cudaMemcpyHostToDevice);
+ cudaError_t status = cudaMemcpyAsync(x_gpu, x, size, cudaMemcpyHostToDevice, get_cuda_stream());
check_error(status);
}
void cuda_pull_array(float *x_gpu, float *x, size_t n)
{
size_t size = sizeof(float)*n;
- cudaError_t status = cudaMemcpy(x, x_gpu, size, cudaMemcpyDeviceToHost);
+ //cudaError_t status = cudaMemcpy(x, x_gpu, size, cudaMemcpyDeviceToHost);
+ cudaError_t status = cudaMemcpyAsync(x, x_gpu, size, cudaMemcpyDeviceToHost, get_cuda_stream());
check_error(status);
+ cudaStreamSynchronize(get_cuda_stream());
}
-#endif
+#else // GPU
+#include "cuda.h"
+void cuda_set_device(int n) {}
+#endif // GPU
--
Gitblit v1.10.0