opencl: Add a simple "fullheader" kernel interface

driver-opencl.c

@@ -1244,6 +1244,34 @@ cl_int queue_scrypt_kernel(const struct opencl_kernel_info * const kinfo, _clSta
 }
 #endif
 
+#ifdef USE_OPENCL_FULLHEADER
+static
+cl_int queue_fullheader_kernel(const struct opencl_kernel_info * const kinfo, _clState * const clState, struct work * const work, __maybe_unused const cl_uint threads)
+{
+	const struct mining_algorithm * const malgo = work_mining_algorithm(work);
+	const cl_kernel * const kernel = &kinfo->kernel;
+	unsigned int num = 0;
+	cl_int status = 0;
+	uint8_t blkheader[80];
+	
+	work->nonce_diff = malgo->opencl_min_nonce_diff;
+	
+	if (!kinfo->goffset)
+	{
+		cl_uint nonce_base = work->blk.nonce;
+		CL_SET_ARG(nonce_base);
+	}
+	
+	swap32yes(blkheader, work->data, 80/4);
+	status = clEnqueueWriteBuffer(clState->commandQueue, clState->CLbuffer0, CL_TRUE, 0, sizeof(blkheader), blkheader, 0, NULL, NULL);
+	
+	CL_SET_ARG(clState->CLbuffer0);
+	CL_SET_ARG(clState->outputBuffer);
+	
+	return status;
+}
+#endif
+
 
 static
 struct opencl_kernel_interface kernel_interfaces[] = {
@@ -1254,6 +1282,9 @@ struct opencl_kernel_interface kernel_interfaces[] = {
 	{"diakgcn", queue_diakgcn_kernel},
 	{"diablo",  queue_diablo_kernel },
 #endif
+#ifdef USE_OPENCL_FULLHEADER
+	{"fullheader", queue_fullheader_kernel },
+#endif
 #ifdef USE_SCRYPT
 	{"scrypt",  queue_scrypt_kernel },
 #endif

findnonce.c

@@ -168,6 +168,10 @@ static void *postcalc_hash(void *userdata)
 
 	for (entry = 0; entry < pcd->res[found]; entry++) {
 		uint32_t nonce = pcd->res[entry];
+#ifdef USE_OPENCL_FULLHEADER
+		if (pcd->kinterface == KL_FULLHEADER)
+			nonce = swab32(nonce);
+#endif
 
 		applog(LOG_DEBUG, "OCL NONCE %u found in slot %d", nonce, entry);
 		submit_nonce(thr, &pcd->work, nonce);

miner.h

@@ -381,6 +381,9 @@ enum cl_kernels {
 	KL_DIAKGCN,
 	KL_DIABLO,
 #endif
+#ifdef USE_OPENCL_FULLHEADER
+	KL_FULLHEADER,
+#endif
 #ifdef USE_SCRYPT
 	KL_SCRYPT,
 #endif

ocl.c

@@ -785,6 +785,11 @@ bool opencl_load_kernel(struct cgpu_info * const cgpu, _clState * const clState,
 			case KL_PHATK:
 				kernel_goffset_support = 0;
 				break;
+#endif
+#ifdef USE_OPENCL_FULLHEADER
+			case KL_FULLHEADER:
+				kernel_goffset_support = 1;
+				break;
 #endif
 			case KL_NONE: case OPENCL_KERNEL_INTERFACE_COUNT:
 #ifdef USE_SCRYPT
@@ -1156,36 +1161,51 @@ built:
 	free((void*)cgpu->kname);
 	cgpu->kname = strdup(kernel_file);
 
-#ifdef USE_SCRYPT
-	if (kernelinfo->interface == KL_SCRYPT && !clState->padbufsize)
+#ifdef MAX_CLBUFFER0_SZ
+	switch (kernelinfo->interface)
 	{
-		size_t ipt = (1024 / data->lookup_gap + (1024 % data->lookup_gap > 0));
-		size_t bufsize = 128 * ipt * data->thread_concurrency;
-
-		/* Use the max alloc value which has been rounded to a power of
-		 * 2 greater >= required amount earlier */
-		if (bufsize > data->max_alloc) {
-			applog(LOG_WARNING, "Maximum buffer memory device %d supports says %lu", gpu, (unsigned long)data->max_alloc);
-			applog(LOG_WARNING, "Your scrypt settings come to %lu", (unsigned long)bufsize);
-		}
-		applog(LOG_DEBUG, "Creating scrypt buffer sized %lu", (unsigned long)bufsize);
-		clState->padbufsize = bufsize;
-
-		/* This buffer is weird and might work to some degree even if
-		 * the create buffer call has apparently failed, so check if we
-		 * get anything back before we call it a failure. */
-		clState->padbuffer8 = NULL;
-		clState->padbuffer8 = clCreateBuffer(clState->context, CL_MEM_READ_WRITE, bufsize, NULL, &status);
-		if (status != CL_SUCCESS && !clState->padbuffer8) {
-			applog(LOG_ERR, "Error %d: clCreateBuffer (padbuffer8), decrease TC or increase LG", status);
-			return false;
-		}
-
-		clState->CLbuffer0 = clCreateBuffer(clState->context, CL_MEM_READ_ONLY, 128, NULL, &status);
-		if (status != CL_SUCCESS) {
-			applog(LOG_ERR, "Error %d: clCreateBuffer (CLbuffer0)", status);
-			return false;
-		}
+#ifdef USE_SCRYPT
+		case KL_SCRYPT:
+			if (!clState->padbufsize)
+			{
+				size_t ipt = (1024 / data->lookup_gap + (1024 % data->lookup_gap > 0));
+				size_t bufsize = 128 * ipt * data->thread_concurrency;
+
+				/* Use the max alloc value which has been rounded to a power of
+				 * 2 greater >= required amount earlier */
+				if (bufsize > data->max_alloc) {
+					applog(LOG_WARNING, "Maximum buffer memory device %d supports says %lu", gpu, (unsigned long)data->max_alloc);
+					applog(LOG_WARNING, "Your scrypt settings come to %lu", (unsigned long)bufsize);
+				}
+				applog(LOG_DEBUG, "Creating scrypt buffer sized %lu", (unsigned long)bufsize);
+				clState->padbufsize = bufsize;
+
+				/* This buffer is weird and might work to some degree even if
+				 * the create buffer call has apparently failed, so check if we
+				 * get anything back before we call it a failure. */
+				clState->padbuffer8 = NULL;
+				clState->padbuffer8 = clCreateBuffer(clState->context, CL_MEM_READ_WRITE, bufsize, NULL, &status);
+				if (status != CL_SUCCESS && !clState->padbuffer8) {
+					applog(LOG_ERR, "Error %d: clCreateBuffer (padbuffer8), decrease TC or increase LG", status);
+					return false;
+				}
+			}
+			// NOTE: fallthru
+#endif
+#ifdef USE_OPENCL_FULLHEADER
+		case KL_FULLHEADER:
+#endif
+			if (!clState->CLbuffer0)
+			{
+				clState->CLbuffer0 = clCreateBuffer(clState->context, CL_MEM_READ_ONLY, MAX_CLBUFFER0_SZ, NULL, &status);
+				if (status != CL_SUCCESS) {
+					applog(LOG_ERR, "Error %d: clCreateBuffer (CLbuffer0)", status);
+					return false;
+				}
+			}
+			break;
+		default:
+			break;
 	}
 #endif
 

ocl.h

@@ -8,6 +8,14 @@
 
 #include "miner.h"
 
+#define SCRYPT_CLBUFFER0_SZ      (128)
+#define FULLHEADER_CLBUFFER0_SZ  ( 80)
+#ifdef USE_SCRYPT
+#	define MAX_CLBUFFER0_SZ  SCRYPT_CLBUFFER0_SZ
+#elif USE_OPENCL_FULLHEADER
+#	define MAX_CLBUFFER0_SZ  FULLHEADER_CLBUFFER0_SZ
+#endif
+
 struct mining_algorithm;
 struct opencl_kernel_info;
 typedef struct _clState _clState;
@@ -21,8 +29,10 @@ struct _clState {
 	cl_command_queue commandQueue;
 	
 	cl_mem outputBuffer;
-#ifdef USE_SCRYPT
+#ifdef MAX_CLBUFFER0_SZ
 	cl_mem CLbuffer0;
+#endif
+#ifdef USE_SCRYPT
 	cl_mem padbuffer8;
 	size_t padbufsize;
 	void * cldata;