From 3567b69e5eaae571948205be00a3996c4b86b6a0 Mon Sep 17 00:00:00 2001
From: Con Kolivas <kernel@kolivas.org>
Date: Fri, 26 Aug 2011 10:20:02 +1000
Subject: [PATCH] Remove fragile source patching for bitalign, vectors et. al
 and simply pass it with the compiler options.

---
 ocl.c           | 80 ++++++++++++++-----------------------------------
 phatk110817.cl  |  4 ---
 poclbm110817.cl |  6 ----
 3 files changed, 23 insertions(+), 67 deletions(-)

diff --git a/ocl.c b/ocl.c
index 6ababb64..60375a26 100644
--- a/ocl.c
+++ b/ocl.c
@@ -372,18 +372,12 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
 	size_t *binary_sizes;
 	char **binaries;
 	int pl;
-	char *source, *rawsource = file_contents(filename, &pl);
+	char *source = file_contents(filename, &pl);
 	size_t sourceSize[] = {(size_t)pl};
 
-	if (!rawsource)
+	if (!source)
 		return NULL;
 
-	source = malloc(pl);
-	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");
@@ -456,69 +450,42 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
 	/////////////////////////////////////////////////////////////////
 
 build:
-	memcpy(source, rawsource, pl);
-
-	/* Patch the source file with the preferred_vwidth */
-	if (clState->preferred_vwidth > 1) {
-		char *find = strstr(source, "VECTORSX");
+	clState->program = clCreateProgramWithSource(clState->context, 1, (const char **)&source, sourceSize, &status);
+	if (status != CL_SUCCESS) {
+		applog(LOG_ERR, "Error: Loading Binary into cl_program (clCreateProgramWithSource)");
+		return NULL;
+	}
 
-		if (unlikely(!find)) {
-			applog(LOG_ERR, "Unable to find VECTORSX in source");
-			return NULL;
-		}
-		find += 7; // "VECTORS"
-		if (clState->preferred_vwidth == 2)
-			strncpy(find, "2", 1);
-		else
-			strncpy(find, "4", 1);
-		if (opt_debug)
-			applog(LOG_DEBUG, "Patched source to suit %d vectors", clState->preferred_vwidth);
+	clRetainProgram(clState->program);
+	if (status != CL_SUCCESS) {
+		applog(LOG_ERR, "Error: Retaining Program (clRetainProgram)");
+		return NULL;
 	}
 
-	/* Patch the source file defining BITALIGN */
-	if (clState->hasBitAlign) {
-		char *find = strstr(source, "BITALIGNX");
+	/* create a cl program executable for all the devices specified */
+	char CompilerOptions[256];
 
-		if (unlikely(!find)) {
-			applog(LOG_ERR, "Unable to find BITALIGNX in source");
-			return NULL;
-		}
-		find += 8; // "BITALIGN"
-		strncpy(find, " ", 1);
+	sprintf(CompilerOptions, "-DWORKSIZE=%d -DVECTORS%d",
+		(int)clState->work_size, clState->preferred_vwidth);
+	if (opt_debug)
+		applog(LOG_DEBUG, "Setting worksize to %d", clState->work_size);
+	if (clState->preferred_vwidth > 1 && opt_debug)
+		applog(LOG_DEBUG, "Patched source to suit %d vectors", clState->preferred_vwidth);
+
+	if (clState->hasBitAlign) {
+		strcat(CompilerOptions, " -DBITALIGN");
 		if (opt_debug)
 			applog(LOG_DEBUG, "cl_amd_media_ops found, patched source with BITALIGN");
 	} else if (opt_debug)
 		applog(LOG_DEBUG, "cl_amd_media_ops not found, will not BITALIGN patch");
 
 	if (patchbfi) {
-		char *find = strstr(source, "BFI_INTX");
-
-		if (unlikely(!find)) {
-			applog(LOG_ERR, "Unable to find BFI_INTX in source");
-			return NULL;
-		}
-		find += 7; // "BFI_INT"
-		strncpy(find, " ", 1);
+		strcat(CompilerOptions, " -DBFI_INT");
 		if (opt_debug)
 			applog(LOG_DEBUG, "cl_amd_media_ops found, patched source with BFI_INT");
 	} else if (opt_debug)
 		applog(LOG_DEBUG, "cl_amd_media_ops not found, will not BFI_INT patch");
 
-	clState->program = clCreateProgramWithSource(clState->context, 1, (const char **)&source, sourceSize, &status);
-	if (status != CL_SUCCESS) {
-		applog(LOG_ERR, "Error: Loading Binary into cl_program (clCreateProgramWithSource)");
-		return NULL;
-	}
-
-	clRetainProgram(clState->program);
-	if (status != CL_SUCCESS) {
-		applog(LOG_ERR, "Error: Retaining Program (clRetainProgram)");
-		return NULL;
-	}
-
-	/* create a cl program executable for all the devices specified */
-	char CompilerOptions[256];
-	sprintf(CompilerOptions, "%s%i", "-DWORKSIZE=", (int)clState->work_size);
 	//int n = 1000;
 	//while(n--)
 	//	printf("%s", CompilerOptions);
@@ -610,7 +577,6 @@ build:
 	}
 
 	free(source);
-	free(rawsource);
 
 	/* Save the binary to be loaded next time */
 	binaryfile = fopen(binaryfilename, "wb");
diff --git a/phatk110817.cl b/phatk110817.cl
index 1bc30b01..20da40ae 100644
--- a/phatk110817.cl
+++ b/phatk110817.cl
@@ -2,10 +2,6 @@
 // I have therefore decided to keep it public-domain.
 
 
-#define VECTORSX
-#define BFI_INTX
-#define BITALIGNX
-
 #ifdef VECTORS4
 	typedef uint4 u;
 #else 
diff --git a/poclbm110817.cl b/poclbm110817.cl
index 693be63b..0db94584 100644
--- a/poclbm110817.cl
+++ b/poclbm110817.cl
@@ -5,9 +5,6 @@
 // This file is taken and modified from the public-domain poclbm project, and
 // we have therefore decided to keep it public-domain in Phoenix.
 
-// The X is a placeholder for patching to suit hardware
-#define VECTORSX
-
 #ifdef VECTORS4
 	typedef uint4 u;
 #elif defined VECTORS2
@@ -36,9 +33,6 @@ __constant uint K[64] = {
 // detected, use it for ch. Otherwise, construct ch out of simpler logical
 // primitives.
 
-#define BFI_INTX
-#define BITALIGNX
-
 #ifdef BFI_INT
 	// Well, slight problem... It turns out BFI_INT isn't actually exposed to
 	// OpenCL (or CAL IL for that matter) in any way. However, there is