diff --git a/Makefile.am b/Makefile.am
index 9c8c2944..347b4af4 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -17,7 +17,7 @@ INCLUDES	= $(PTHREAD_FLAGS) -fno-strict-aliasing $(JANSSON_INCLUDES)
 
 bin_PROGRAMS	= cgminer
 
-bin_SCRIPTS	= phatk120203.cl poclbm120203.cl
+bin_SCRIPTS	= *.cl
 
 cgminer_LDFLAGS	= $(PTHREAD_FLAGS)
 cgminer_LDADD	= $(DLOPEN_FLAGS) @LIBCURL_LIBS@ @JANSSON_LIBS@ @PTHREAD_LIBS@ \
@@ -44,7 +44,7 @@ cgminer_SOURCES += device-gpu.h device-gpu.c
 # the original GPU related sources, unchanged
 cgminer_SOURCES += ocl.c ocl.h findnonce.c findnonce.h
 cgminer_SOURCES += adl.c adl.h adl_functions.h
-cgminer_SOURCES += phatk120203.cl poclbm120203.cl
+cgminer_SOURCES += *.cl
 
 if HAS_CPUMINE
 # original CPU related sources, unchanged
diff --git a/configure.ac b/configure.ac
index 9dcd9b7a..b8a26b39 100644
--- a/configure.ac
+++ b/configure.ac
@@ -293,6 +293,11 @@ fi
 
 AC_DEFINE_UNQUOTED([CGMINER_PREFIX], ["$prefix/bin"], [Path to cgminer install])
 
+AC_DEFINE_UNQUOTED([PHATK_KERNNAME], ["phatk120203"], [Filename for phatk kernel])
+AC_DEFINE_UNQUOTED([POCLBM_KERNNAME], ["poclbm120203"], [Filename for poclbm kernel])
+AC_DEFINE_UNQUOTED([DIAKGCN_KERNNAME], ["diakgcn120208"], [Filename for diakgcn kernel])
+
+
 AC_SUBST(OPENCL_LIBS)
 AC_SUBST(OPENCL_FLAGS)
 AC_SUBST(JANSSON_LIBS)
diff --git a/device-gpu.c b/device-gpu.c
index e05f5546..d40db48c 100644
--- a/device-gpu.c
+++ b/device-gpu.c
@@ -754,6 +754,43 @@ static cl_int queue_phatk_kernel(_clState *clState, dev_blk_ctx *blk)
 	return status;
 }
 
+static cl_int queue_diakgcn_kernel(_clState *clState, dev_blk_ctx *blk)
+{
+	cl_kernel *kernel = &clState->kernel;
+	cl_int status = 0;
+	int num = 0;
+
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->ctx_a);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->ctx_b);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->ctx_c);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->ctx_d);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->ctx_e);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->ctx_f);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->ctx_g);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->ctx_h);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->cty_b);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->cty_c);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->cty_d);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->cty_f);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->cty_g);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->cty_h);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->nonce);
+
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->fW0);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->fW1);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->fW2);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->fW3);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->fW15);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->fW01r);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->fcty_e);
+	status |= clSetKernelArg(*kernel, num++, sizeof(uint), (void *)&blk->fcty_e2);
+
+	status |= clSetKernelArg(*kernel, num++, sizeof(clState->outputBuffer),
+				 (void *)&clState->outputBuffer);
+
+	return status;
+}
+
 static void set_threads_hashes(unsigned int vectors, unsigned int *threads,
 			       unsigned int *hashes, size_t *globalThreads,
 			       unsigned int minthreads, int intensity)
@@ -905,9 +942,11 @@ static void opencl_detect()
 		return;
 
 	if (opt_kernel) {
-		if (strcmp(opt_kernel, "poclbm") && strcmp(opt_kernel, "phatk"))
-			quit(1, "Invalid kernel name specified - must be poclbm or phatk");
-		if (!strcmp(opt_kernel, "poclbm"))
+		if (strcmp(opt_kernel, "poclbm") && strcmp(opt_kernel, "phatk") && strcmp(opt_kernel, "diakgcn"))
+			quit(1, "Invalid kernel name specified - must be poclbm, phatk or diakgcn");
+		if (!strcmp(opt_kernel, "diakgcn"))
+			chosen_kernel = KL_DIAKGCN;
+		else if (!strcmp(opt_kernel, "poclbm"))
 			chosen_kernel = KL_POCLBM;
 		else
 			chosen_kernel = KL_PHATK;
@@ -1042,6 +1081,9 @@ static bool opencl_thread_init(struct thr_info *thr)
 		default:
 			thrdata->queue_kernel_parameters = &queue_phatk_kernel;
 			break;
+		case KL_DIAKGCN:
+			thrdata->queue_kernel_parameters = &queue_diakgcn_kernel;
+			break;
 	}
 
 	thrdata->res = calloc(BUFFERSIZE, 1);
diff --git a/miner.h b/miner.h
index 44e6a537..adedaaac 100644
--- a/miner.h
+++ b/miner.h
@@ -675,6 +675,7 @@ enum cl_kernel {
 	KL_NONE,
 	KL_POCLBM,
 	KL_PHATK,
+	KL_DIAKGCN,
 };
 
 extern void get_datestamp(char *, struct timeval *);
diff --git a/ocl.c b/ocl.c
index 76ceb605..53115576 100644
--- a/ocl.c
+++ b/ocl.c
@@ -348,21 +348,27 @@ _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
 	char filename[16];
 
 	if (chosen_kernel == KL_NONE) {
-		if (!clState->hasBitAlign || strstr(name, "Tahiti"))
+		if (strstr(name, "Tahiti"))
+			chosen_kernel = KL_DIAKGCN;
+		else if (!clState->hasBitAlign)
 			chosen_kernel = KL_POCLBM;
 		else
 			chosen_kernel = KL_PHATK;
 	}
 
 	switch (chosen_kernel) {
+		case KL_DIAKGCN:
+			strcpy(filename, DIAKGCN_KERNNAME".cl");
+			strcpy(binaryfilename, DIAKGCN_KERNNAME);
+			break;
 		case KL_POCLBM:
-			strcpy(filename, "poclbm120203.cl");
-			strcpy(binaryfilename, "poclbm120203");
+			strcpy(filename, POCLBM_KERNNAME".cl");
+			strcpy(binaryfilename, POCLBM_KERNNAME);
 			break;
 		case KL_NONE: /* Shouldn't happen */
 		case KL_PHATK:
-			strcpy(filename, "phatk120203.cl");
-			strcpy(binaryfilename, "phatk120203");
+			strcpy(filename, PHATK_KERNNAME".cl");
+			strcpy(binaryfilename, PHATK_KERNNAME);
 			break;
 	}