From 1c0fd9bb4726f28b5ccf4491b8d108b00c884ec3 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Thu, 30 Oct 2014 06:26:41 +0000
Subject: [PATCH] im2col slightly faster

---
 src/opencl.c |   34 +++++++++++++++++++++++++++++++---
 1 files changed, 31 insertions(+), 3 deletions(-)

diff --git a/src/opencl.c b/src/opencl.c
index bcc0f09..fc7310c 100644
--- a/src/opencl.c
+++ b/src/opencl.c
@@ -1,11 +1,13 @@
 #ifdef GPU
-#include "opencl.h"
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <time.h>
 #include <unistd.h>
+//#include <clBLAS.h>
 
+#include "opencl.h"
+#include "utils.h"
 
 cl_info cl = {0};
 
@@ -79,9 +81,9 @@
 
     }
     int index = getpid()%num_devices;
+    index = 1;
     printf("%d rand, %d devices, %d index\n", getpid(), num_devices, index);
-    //info.device = devices[index];
-    info.device = devices[0];
+    info.device = devices[index];
     fprintf(stderr, "Found %d device(s)\n", num_devices);
     check_error(info);
 
@@ -93,16 +95,31 @@
     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();
+    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;
 	char src[64*1024];
 	memset(src, 0, 64*1024);
 	FILE *fil=fopen(filename,"r");
+    if(fil == 0) file_error(filename);
 	srcsize=fread(src, sizeof src, 1, fil);
 	fclose(fil);
 	const char *srcptr[]={src};
@@ -124,6 +141,7 @@
 void cl_setup()
 {
 	if(!cl.initialized){
+        printf("initializing\n");
 		cl = cl_init();
 	}
 }
@@ -178,4 +196,14 @@
     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);
+    check_error(cl);
+    return mem;
+}
+
 #endif

--
Gitblit v1.10.0