Temporarily back out binary building till it's working more reliably.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227
diff --git a/ocl.c b/ocl.c
index 64497e5..1c496cc 100644
--- a/ocl.c
+++ b/ocl.c
@@ -10,8 +10,6 @@
#include <time.h>
#include <sys/time.h>
#include <pthread.h>
-#include <sys/stat.h>
-#include <unistd.h>
#include "findnonce.h"
#include "ocl.h"
@@ -309,116 +307,41 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
if (clState->max_work_size > 512)
clState->max_work_size = 512;
- /* For some reason 2 vectors is still better even if the card says
- * otherwise */
- if (clState->preferred_vwidth > 1)
- clState->preferred_vwidth = 2;
- if (opt_vectors)
- clState->preferred_vwidth = opt_vectors;
- if (opt_worksize && opt_worksize <= clState->max_work_size)
- clState->work_size = opt_worksize;
- else
- clState->work_size = clState->max_work_size / clState->preferred_vwidth;
+ /////////////////////////////////////////////////////////////////
+ // Load CL file, build CL program object, create CL kernel object
+ /////////////////////////////////////////////////////////////////
- /* Create binary filename based on parameters passed to opencl
- * compiler to ensure we only load a binary that matches what would
- * have otherwise created. The filename is:
- * kernelname +/i bitalign + v + vectors + w + work_size + sizeof(long) + .bin
- */
- char binaryfilename[255];
- char numbuf[10];
+ /* Load a different kernel depending on whether it supports
+ * cl_amd_media_ops or not */
char filename[10];
+
if (clState->hasBitAlign)
strcpy(filename, "phatk.cl");
else
strcpy(filename, "poclbm.cl");
- FILE *binaryfile;
- size_t *binary_sizes;
- char **binaries;
- size_t nDevices = 1;
+
int pl;
char *source, *rawsource = file_contents(filename, &pl);
size_t sourceSize[] = {(size_t)pl};
-
source = malloc(pl);
+retry:
if (!source) {
applog(LOG_ERR, "Unable to malloc source");
return NULL;
}
-
- binary_sizes = (size_t *)malloc(sizeof(size_t)*nDevices);
- if (unlikely(!binary_sizes)) {
- applog(LOG_ERR, "Unable to malloc binary_sizes");
- return NULL;
- }
- binaries = (char **)malloc(sizeof(char *)*nDevices);
- if (unlikely(!binaries)) {
- applog(LOG_ERR, "Unable to malloc binaries");
- return NULL;
- }
-
- if (clState->hasBitAlign) {
- strcpy(binaryfilename, "phatk");
- strcat(binaryfilename, "bitalign");
- } else
- strcpy(binaryfilename, "poclbm");
- strcat(binaryfilename, "v");
- sprintf(numbuf, "%d", clState->preferred_vwidth);
- strcat(binaryfilename, numbuf);
- strcat(binaryfilename, "w");
- sprintf(numbuf, "%d", (int)clState->work_size);
- strcat(binaryfilename, numbuf);
- strcat(binaryfilename, "long");
- sprintf(numbuf, "%d", (int)sizeof(long));
- strcat(binaryfilename, numbuf);
- strcat(binaryfilename, ".bin");
-
- binaryfile = fopen(binaryfilename, "r");
- if (!binaryfile) {
- if (opt_debug)
- applog(LOG_DEBUG, "No binary found, generating from source");
- } else {
- struct stat binary_stat;
-
- if (unlikely(stat(binaryfilename, &binary_stat))) {
- if (opt_debug)
- applog(LOG_DEBUG, "Unable to stat binary, generating from source");
- fclose(binaryfile);
- goto build;
- }
- binary_sizes[gpu] = binary_stat.st_size;
- binaries[gpu] = (char *)malloc(binary_sizes[gpu]);
- if (unlikely(!binaries[gpu])) {
- applog(LOG_ERR, "Unable to malloc binaries");
- fclose(binaryfile);
- return NULL;
- }
-
- if (fread(binaries[gpu], 1, binary_sizes[gpu], binaryfile) != binary_sizes[gpu]) {
- applog(LOG_ERR, "Unable to fread binaries[gpu]");
- fclose(binaryfile);
- return NULL;
- }
- fclose(binaryfile);
-
- clState->program = clCreateProgramWithBinary(clState->context, 1, &devices[gpu], &binary_sizes[gpu], (const unsigned char **)&binaries[gpu], &status, NULL);
- if (status != CL_SUCCESS)
- {
- applog(LOG_ERR, "Error: Loading Binary into cl_program (clCreateProgramWithBinary)");
- return NULL;
- }
- if (opt_debug)
- applog(LOG_DEBUG, "Loaded binary image %s", binaryfilename);
- goto built;
- }
-
- /////////////////////////////////////////////////////////////////
- // Load CL file, build CL program object, create CL kernel object
- /////////////////////////////////////////////////////////////////
-
-build:
memcpy(source, rawsource, pl);
+ /* For some reason 2 vectors is still better even if the card says
+ * otherwise */
+ if (clState->preferred_vwidth > 1)
+ clState->preferred_vwidth = 2;
+ if (opt_vectors)
+ clState->preferred_vwidth = opt_vectors;
+ if (opt_worksize && opt_worksize <= clState->max_work_size)
+ clState->work_size = opt_worksize;
+ else
+ clState->work_size = clState->max_work_size / clState->preferred_vwidth;
+
/* Patch the source file with the preferred_vwidth */
if (clState->preferred_vwidth > 1) {
char *find = strstr(source, "VECTORSX");
@@ -488,24 +411,22 @@ build:
/* Patch the kernel if the hardware supports BFI_INT */
if (patchbfi) {
- /* figure out the size of the binary for each device. */
- status = clGetProgramInfo( clState->program, CL_PROGRAM_BINARY_SIZES, sizeof(size_t)*nDevices, binary_sizes, NULL );
- if (unlikely(status != CL_SUCCESS))
- {
- applog(LOG_ERR, "Error: Getting program info. (clGetPlatformInfo)");
- return NULL;
- }
+ size_t nDevices;
+ size_t * binary_sizes;
+ char ** binaries;
+ int err;
+
+ /* figure out number of devices and the sizes of the binary for each device. */
+ err = clGetProgramInfo( clState->program, CL_PROGRAM_NUM_DEVICES, sizeof(nDevices), &nDevices, NULL );
+ binary_sizes = (size_t *)malloc( sizeof(size_t)*nDevices );
+ err = clGetProgramInfo( clState->program, CL_PROGRAM_BINARY_SIZES, sizeof(size_t)*nDevices, binary_sizes, NULL );
/* copy over all of the generated binaries. */
+ binaries = (char **)malloc( sizeof(char *)*nDevices );
if (opt_debug)
applog(LOG_DEBUG, "binary size %d : %d", gpu, binary_sizes[gpu]);
binaries[gpu] = (char *)malloc( sizeof(char)*binary_sizes[gpu] );
- status = clGetProgramInfo( clState->program, CL_PROGRAM_BINARIES, sizeof(char *)*nDevices, binaries, NULL );
- if (unlikely(status != CL_SUCCESS))
- {
- applog(LOG_ERR, "Error: Getting program info. (clGetPlatformInfo)");
- return NULL;
- }
+ err = clGetProgramInfo( clState->program, CL_PROGRAM_BINARIES, sizeof(char *)*nDevices, binaries, NULL );
unsigned remaining = binary_sizes[gpu];
char *w = binaries[gpu];
@@ -516,7 +437,7 @@ build:
* back and find the 2nd incidence of \x7ELF (rewind by one
* from ELF) and then patch the opcocdes */
if (!advance(&w, &remaining, ".text"))
- {patchbfi = 0; goto build;}
+ {patchbfi = 0; goto retry;}
w++; remaining--;
if (!advance(&w, &remaining, ".text")) {
/* 32 bit builds only one ELF */
@@ -526,7 +447,7 @@ build:
memcpy(&length, w + 289, 4);
w = binaries[gpu]; remaining = binary_sizes[gpu];
if (!advance(&w, &remaining, "ELF"))
- {patchbfi = 0; goto build;}
+ {patchbfi = 0; goto retry;}
w++; remaining--;
if (!advance(&w, &remaining, "ELF")) {
/* 32 bit builds only one ELF */
@@ -557,23 +478,6 @@ build:
free(source);
free(rawsource);
- /* Save the binary to be loaded next time */
- binaryfile = fopen(binaryfilename, "w");
- if (!binaryfile) {
- /* Not a fatal problem, just means we build it again next time */
- if (opt_debug)
- applog(LOG_DEBUG, "Unable to create file %s", binaryfilename);
- } else {
- if (unlikely(fwrite(binaries[gpu], 1, binary_sizes[gpu], binaryfile) != binary_sizes[gpu])) {
- applog(LOG_ERR, "Unable to fwrite to binaryfile");
- return NULL;
- }
- fclose(binaryfile);
- }
-built:
- free(binaries);
- free(binary_sizes);
-
applog(LOG_INFO, "Initialising kernel %s with%s BFI_INT patching, %d vectors and worksize %d",
filename, patchbfi ? "" : "out", clState->preferred_vwidth, clState->work_size);