From 26c59fbf0f399950d843faba5a9a0fb5dc345df6 Mon Sep 17 00:00:00 2001
From: Con Kolivas <kernel@kolivas.org>
Date: Wed, 22 Feb 2012 16:59:28 +1100
Subject: [PATCH] Allow the worksize to be set per-device.

---
 cgminer.c    |  6 ++----
 device-gpu.c | 30 ++++++++++++++++++++++++++++++
 device-gpu.h |  1 +
 ocl.c        |  9 ++++-----
 4 files changed, 37 insertions(+), 9 deletions(-)

diff --git a/cgminer.c b/cgminer.c
index d0c1e2e2..b5965a0f 100644
--- a/cgminer.c
+++ b/cgminer.c
@@ -91,8 +91,6 @@ static int opt_fail_pause = 5;
 static int fail_pause = 5;
 int opt_log_interval = 5;
 static int opt_queue = 1;
-int opt_vectors;
-int opt_worksize;
 int opt_scantime = 60;
 int opt_expiry = 120;
 int opt_bench_algo = -1;
@@ -839,8 +837,8 @@ static struct opt_table opt_config_table[] = {
 			"Log verbose output to stderr as well as status output"),
 #ifdef HAVE_OPENCL
 	OPT_WITH_ARG("--worksize|-w",
-		     set_int_0_to_9999, opt_show_intval, &opt_worksize,
-		     "Override detected optimal worksize"),
+		     set_worksize, NULL, NULL,
+		     "Override detected optimal worksize - one value or comma separated list"),
 #endif
 	OPT_WITH_ARG("--userpass|-O",
 		     set_userpass, NULL, NULL,
diff --git a/device-gpu.c b/device-gpu.c
index 1c2ad64a..a60e80e8 100644
--- a/device-gpu.c
+++ b/device-gpu.c
@@ -93,6 +93,36 @@ char *set_vector(char *arg)
 
 	return NULL;
 }
+
+char *set_worksize(char *arg)
+{
+	int i, val = 0, device = 0;
+	char *nextptr;
+
+	nextptr = strtok(arg, ",");
+	if (nextptr == NULL)
+		return "Invalid parameters for set work size";
+	val = atoi(nextptr);
+	if (val < 1 || val > 9999)
+		return "Invalid value passed to set_worksize";
+
+	gpus[device++].work_size = val;
+
+	while ((nextptr = strtok(NULL, ",")) != NULL) {
+		val = atoi(nextptr);
+		if (val < 1 || val > 9999)
+			return "Invalid value passed to set_worksize";
+
+		gpus[device++].work_size = val;
+	}
+	if (device == 1) {
+		for (i = device; i < MAX_GPUDEVICES; i++)
+			gpus[i].work_size = gpus[0].work_size;
+	}
+
+	return NULL;
+}
+
 #endif
 
 #ifdef HAVE_ADL
diff --git a/device-gpu.h b/device-gpu.h
index 12af985e..2c8d1e32 100644
--- a/device-gpu.h
+++ b/device-gpu.h
@@ -16,6 +16,7 @@ extern char *set_temp_overheat(char *arg);
 extern char *set_temp_target(char *arg);
 extern char *set_intensity(char *arg);
 extern char *set_vector(char *arg);
+extern char *set_worksize(char *arg);
 void manage_gpu(void);
 extern void pause_dynamic_threads(int gpu);
 
diff --git a/ocl.c b/ocl.c
index 5869ce4d..54a09d3f 100644
--- a/ocl.c
+++ b/ocl.c
@@ -33,7 +33,6 @@
 #include "findnonce.h"
 #include "ocl.h"
 
-extern int opt_worksize;
 int opt_platform_id;
 
 char *file_contents(const char *filename, int *length)
@@ -349,12 +348,12 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
 	else
 		clState->vwidth = preferred_vwidth;
 
-	if (opt_worksize && opt_worksize <= (int)clState->max_work_size)
-		clState->work_size = opt_worksize;
+	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->work_size = 64;
+		clState->wsize = 64;
 	else
-		clState->work_size = (clState->max_work_size <= 256 ? clState->max_work_size : 256) / clState->vwidth;
+		clState->wsize = (clState->max_work_size <= 256 ? clState->max_work_size : 256) / clState->vwidth;
 
 	/* Create binary filename based on parameters passed to opencl
 	 * compiler to ensure we only load a binary that matches what would