From 352ae7e65b6a74bcd768aa88b866a44c713284c8 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Wed, 26 Oct 2016 15:35:44 +0000
Subject: [PATCH] ADAM
---
src/cuda.c | 83 +++++++++++++++++++++++++++++++++++------
1 files changed, 70 insertions(+), 13 deletions(-)
diff --git a/src/cuda.c b/src/cuda.c
index 8849fb1..617e7b3 100644
--- a/src/cuda.c
+++ b/src/cuda.c
@@ -5,19 +5,46 @@
#include "cuda.h"
#include "utils.h"
#include "blas.h"
+#include "assert.h"
#include <stdlib.h>
+#include <time.h>
+void cuda_set_device(int n)
+{
+ gpu_index = n;
+ cudaError_t status = cudaSetDevice(n);
+ check_error(status);
+}
+
+int cuda_get_device()
+{
+ int n = 0;
+ cudaError_t status = cudaGetDevice(&n);
+ check_error(status);
+ return n;
+}
void check_error(cudaError_t status)
{
+ 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){
@@ -25,26 +52,41 @@
size_t x = k;
size_t y = 1;
if(x > 65535){
- x = ceil(sqrt(k));
- y = (n-1)/(x*BLOCK) + 1;
+ x = ceil(sqrt(k));
+ y = (n-1)/(x*BLOCK) + 1;
}
dim3 d = {x, y, 1};
//printf("%ld %ld %ld %ld\n", n, x, y, x*y*BLOCK);
return d;
}
+#ifdef CUDNN
+cudnnHandle_t cudnn_handle()
+{
+ static int init[16] = {0};
+ static cudnnHandle_t handle[16];
+ int i = cuda_get_device();
+ if(!init[i]) {
+ cudnnCreate(&handle[i]);
+ init[i] = 1;
+ }
+ return handle[i];
+}
+#endif
+
cublasHandle_t blas_handle()
{
- static int init = 0;
- static cublasHandle_t handle;
- if(!init) {
- cublasCreate(&handle);
- init = 1;
+ static int init[16] = {0};
+ static cublasHandle_t handle[16];
+ int i = cuda_get_device();
+ if(!init[i]) {
+ cublasCreate(&handle[i]);
+ init[i] = 1;
}
- return handle;
+ return handle[i];
}
-float *cuda_make_array(float *x, int n)
+float *cuda_make_array(float *x, size_t n)
{
float *x_gpu;
size_t size = sizeof(float)*n;
@@ -54,10 +96,25 @@
status = cudaMemcpy(x_gpu, x, size, cudaMemcpyHostToDevice);
check_error(status);
}
+ if(!x_gpu) error("Cuda malloc failed\n");
return x_gpu;
}
-float cuda_compare(float *x_gpu, float *x, int n, char *s)
+void cuda_random(float *x_gpu, size_t n)
+{
+ static curandGenerator_t gen[16];
+ static int init[16] = {0};
+ int i = cuda_get_device();
+ if(!init[i]){
+ curandCreateGenerator(&gen[i], CURAND_RNG_PSEUDO_DEFAULT);
+ curandSetPseudoRandomGeneratorSeed(gen[i], time(0));
+ init[i] = 1;
+ }
+ curandGenerateUniform(gen[i], x_gpu, n);
+ check_error(cudaPeekAtLastError());
+}
+
+float cuda_compare(float *x_gpu, float *x, size_t n, char *s)
{
float *tmp = calloc(n, sizeof(float));
cuda_pull_array(x_gpu, tmp, n);
@@ -70,7 +127,7 @@
return err;
}
-int *cuda_make_int_array(int n)
+int *cuda_make_int_array(size_t n)
{
int *x_gpu;
size_t size = sizeof(int)*n;
@@ -85,14 +142,14 @@
check_error(status);
}
-void cuda_push_array(float *x_gpu, float *x, int n)
+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);
check_error(status);
}
-void cuda_pull_array(float *x_gpu, float *x, int n)
+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);
--
Gitblit v1.10.0