From cb1f33c6ae840e8dc0f43518daf76e6ed01034f0 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Mon, 08 Dec 2014 19:48:57 +0000
Subject: [PATCH] Fixed race condition in server
---
src/opencl.c | 60 +++++++++++++++++++++++++++++++-----------------------------
1 files changed, 31 insertions(+), 29 deletions(-)
diff --git a/src/opencl.c b/src/opencl.c
index fc7310c..994b8d6 100644
--- a/src/opencl.c
+++ b/src/opencl.c
@@ -4,25 +4,30 @@
#include <string.h>
#include <time.h>
#include <unistd.h>
-//#include <clBLAS.h>
+
+#ifdef CLBLAS
+#include <clBLAS.h>
+#endif
#include "opencl.h"
#include "utils.h"
+#include "activations.h"
cl_info cl = {0};
void check_error(cl_info info)
{
- clFinish(cl.queue);
+ // clFinish(cl.queue);
if (info.error != CL_SUCCESS) {
printf("\n Error number %d", info.error);
+ abort();
exit(1);
}
}
#define MAX_DEVICES 10
-cl_info cl_init()
+cl_info cl_init(int index)
{
cl_info info;
info.initialized = 0;
@@ -69,6 +74,8 @@
printf(" DEVICE_MAX_CLOCK_FREQUENCY = %u\n", (unsigned int)buf_uint);
clGetDeviceInfo(devices[i], CL_DEVICE_GLOBAL_MEM_SIZE, sizeof(buf_ulong), &buf_ulong, NULL);
printf(" DEVICE_GLOBAL_MEM_SIZE = %llu\n", (unsigned long long)buf_ulong);
+ clGetDeviceInfo(devices[i], CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(buf_ulong), &buf_ulong, NULL);
+ printf(" DEVICE_MAX_MEM_ALLOC_SIZE = %llu\n", (unsigned long long)buf_ulong);
clGetDeviceInfo(devices[i], CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(buf_ulong), &buf_ulong, NULL);
printf(" DEVICE_MAX_WORK_GROUP_SIZE = %llu\n", (unsigned long long)buf_ulong);
cl_uint items;
@@ -80,8 +87,7 @@
printf(" DEVICE_MAX_WORK_ITEM_SIZES = %u / %u / %u \n", (unsigned int)workitem_size[0], (unsigned int)workitem_size[1], (unsigned int)workitem_size[2]);
}
- int index = getpid()%num_devices;
- index = 1;
+ index = index%num_devices;
printf("%d rand, %d devices, %d index\n", getpid(), num_devices, index);
info.device = devices[index];
fprintf(stderr, "Found %d device(s)\n", num_devices);
@@ -95,24 +101,14 @@
check_error(info);
info.queue = clCreateCommandQueue(info.context, info.device, 0, &info.error);
check_error(info);
- for(i = 0; i < NUM_QUEUES; ++i){
- info.queues[i] = clCreateCommandQueue(info.context, info.device, 0, &info.error);
- check_error(info);
- }
- //info.error = clblasSetup();
+ #ifdef CLBLAS
+ info.error = clblasSetup();
+ #endif
check_error(info);
info.initialized = 1;
return info;
}
-void wait_for_queues()
-{
- int i;
- for(i = 0; i < NUM_QUEUES; ++i){
- clFinish(cl.queues[i]);
- }
-}
-
cl_program cl_fprog(char *filename, char *options, cl_info info)
{
size_t srcsize;
@@ -138,17 +134,16 @@
return prog;
}
-void cl_setup()
+void cl_setup(int index)
{
if(!cl.initialized){
printf("initializing\n");
- cl = cl_init();
+ cl = cl_init(index);
}
}
cl_kernel get_kernel(char *filename, char *kernelname, char *options)
{
- cl_setup();
cl_program prog = cl_fprog(filename, options, cl);
cl_kernel kernel=clCreateKernel(prog, kernelname, &cl.error);
check_error(cl);
@@ -157,22 +152,29 @@
void cl_read_array(cl_mem mem, float *x, int n)
{
- cl_setup();
- clEnqueueReadBuffer(cl.queue, mem, CL_TRUE, 0, sizeof(float)*n,x,0,0,0);
+ cl.error = clEnqueueReadBuffer(cl.queue, mem, CL_TRUE, 0, sizeof(float)*n,x,0,0,0);
check_error(cl);
}
+float cl_checksum(cl_mem mem, int n)
+{
+
+ float *x = calloc(n, sizeof(float));
+ cl_read_array(mem, x, n);
+ float sum = sum_array(x, n);
+ free(x);
+ return sum;
+}
+
void cl_write_array(cl_mem mem, float *x, int n)
{
- cl_setup();
- clEnqueueWriteBuffer(cl.queue, mem, CL_TRUE, 0,sizeof(float)*n,x,0,0,0);
+ cl.error = clEnqueueWriteBuffer(cl.queue, mem, CL_TRUE, 0,sizeof(float)*n,x,0,0,0);
check_error(cl);
}
void cl_copy_array(cl_mem src, cl_mem dst, int n)
{
- cl_setup();
- clEnqueueCopyBuffer(cl.queue, src, dst, 0, 0, sizeof(float)*n,0,0,0);
+ cl.error = clEnqueueCopyBuffer(cl.queue, src, dst, 0, 0, sizeof(float)*n,0,0,0);
check_error(cl);
}
@@ -186,19 +188,19 @@
return sub;
}
+
cl_mem cl_make_array(float *x, int n)
{
- cl_setup();
cl_mem mem = clCreateBuffer(cl.context,
CL_MEM_READ_WRITE|CL_MEM_COPY_HOST_PTR,
sizeof(float)*n, x, &cl.error);
check_error(cl);
+ activate_array_ongpu(mem, n, LINEAR);
return mem;
}
cl_mem cl_make_int_array(int *x, int n)
{
- cl_setup();
cl_mem mem = clCreateBuffer(cl.context,
CL_MEM_READ_WRITE|CL_MEM_COPY_HOST_PTR,
sizeof(int)*n, x, &cl.error);
--
Gitblit v1.10.0