GOSTSec/sgminer
1
0
Fork 0
mirror of https://github.com/GOSTSec/sgminer synced 2025-01-30 08:24:26 +00:00
Code Issues Releases Wiki Activity

Use diablo kernel on all future SDKs for Tahiti and set preferred vector width to 1 on poclbm kernel only.

Browse Source
This commit is contained in:
Con Kolivas 2012-02-23 20:24:32 +11:00
parent 66a19f8294
commit 709c4cd8e1
1 changed files with 27 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

56
ocl.c
View File

@ -332,32 +332,6 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
} }
applog(LOG_DEBUG, "Max work group size reported %d", clState->max_work_size); applog(LOG_DEBUG, "Max work group size reported %d", clState->max_work_size);
/* For some reason 2 vectors is still better even if the card says
* otherwise, and many cards lie about their max so use 256 as max
* unless explicitly set on the command line. 79x0 cards perform
* better without vectors */
if (preferred_vwidth > 1) {
if (strstr(name, "Tahiti"))
preferred_vwidth = 1;
else
preferred_vwidth = 2;
}
if (gpus[gpu].vwidth)
clState->vwidth = gpus[gpu].vwidth;
else {
clState->vwidth = preferred_vwidth;
gpus[gpu].vwidth = preferred_vwidth;
}
if (gpus[gpu].work_size && gpus[gpu].work_size <= clState->max_work_size)
clState->wsize = gpus[gpu].work_size;
else if (strstr(name, "Tahiti"))
clState->wsize = 64;
else
clState->wsize = (clState->max_work_size <= 256 ? clState->max_work_size : 256) / clState->vwidth;
gpus[gpu].work_size = clState->wsize;
/* Create binary filename based on parameters passed to opencl /* Create binary filename based on parameters passed to opencl
* compiler to ensure we only load a binary that matches what would * compiler to ensure we only load a binary that matches what would
* have otherwise created. The filename is: * have otherwise created. The filename is:
@ -378,9 +352,10 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
applog(LOG_INFO, "Selecting diablo kernel"); applog(LOG_INFO, "Selecting diablo kernel");
clState->chosen_kernel = KL_DIABLO; clState->chosen_kernel = KL_DIABLO;
} }
} else if (strstr(vbuff, "898.1")) { // Windows 64 bit 12.2 driver } else if (strstr(vbuff, "898.1") || // Windows 64 bit 12.2 driver
applog(LOG_INFO, "Selecting diablo kernel"); strstr(name, "Tahiti")) { // All non SDK 2.6 79x0
clState->chosen_kernel = KL_DIABLO; applog(LOG_INFO, "Selecting diablo kernel");
clState->chosen_kernel = KL_DIABLO;
} else if (clState->hasBitAlign) { } else if (clState->hasBitAlign) {
applog(LOG_INFO, "Selecting phatk kernel"); applog(LOG_INFO, "Selecting phatk kernel");
clState->chosen_kernel = KL_PHATK; clState->chosen_kernel = KL_PHATK;
@ -393,10 +368,18 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
} else } else
clState->chosen_kernel = gpus[gpu].kernel; clState->chosen_kernel = gpus[gpu].kernel;
/* For some reason 2 vectors is still better even if the card says
* otherwise, and many cards lie about their max so use 256 as max
* unless explicitly set on the command line. */
if (preferred_vwidth > 2)
preferred_vwidth = 2;
switch (clState->chosen_kernel) { switch (clState->chosen_kernel) {
case KL_POCLBM: case KL_POCLBM:
strcpy(filename, POCLBM_KERNNAME".cl"); strcpy(filename, POCLBM_KERNNAME".cl");
strcpy(binaryfilename, POCLBM_KERNNAME); strcpy(binaryfilename, POCLBM_KERNNAME);
/* This kernel prefers to not use vectors */
preferred_vwidth = 1;
break; break;
case KL_PHATK: case KL_PHATK:
strcpy(filename, PHATK_KERNNAME".cl"); strcpy(filename, PHATK_KERNNAME".cl");
@ -413,6 +396,21 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
break; break;
} }
if (gpus[gpu].vwidth)
clState->vwidth = gpus[gpu].vwidth;
else {
clState->vwidth = preferred_vwidth;
gpus[gpu].vwidth = preferred_vwidth;
}
if (gpus[gpu].work_size && gpus[gpu].work_size <= clState->max_work_size)
clState->wsize = gpus[gpu].work_size;
else if (strstr(name, "Tahiti"))
clState->wsize = 64;
else
clState->wsize = (clState->max_work_size <= 256 ? clState->max_work_size : 256) / clState->vwidth;
gpus[gpu].work_size = clState->wsize;
FILE *binaryfile; FILE *binaryfile;
size_t *binary_sizes; size_t *binary_sizes;
char **binaries; char **binaries;