x17: apply echo512 improvement

add a tiny 1% on x17, better than nothing...

x16/x16r.cu

@@ -491,8 +491,9 @@ extern "C" int scanhash_x16r(int thr_id, struct work* work, uint32_t max_nonce,
 			case ECHO:
 				if (use_compat_kernels[thr_id])
 					x11_echo512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
-				else
+				else {
 					x16_echo512_cpu_hash_64(thr_id, throughput, d_hash[thr_id]); order++;
+				}
 				TRACE("echo   :");
 				break;
 			case HAMSI:

x16/x16s.cu

@@ -489,8 +489,9 @@ extern "C" int scanhash_x16s(int thr_id, struct work* work, uint32_t max_nonce,
 			case ECHO:
 				if (use_compat_kernels[thr_id])
 					x11_echo512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
-				else
+				else {
 					x16_echo512_cpu_hash_64(thr_id, throughput, d_hash[thr_id]); order++;
+				}
 				TRACE("echo   :");
 				break;
 			case HAMSI:

x17/x17.cu

@@ -32,6 +32,8 @@ extern "C" {
 
 static uint32_t *d_hash[MAX_GPUS];
 
+extern void x16_echo512_cpu_hash_64(int thr_id, uint32_t threads, uint32_t *d_hash);
+
 extern void x13_hamsi512_cpu_init(int thr_id, uint32_t threads);
 extern void x13_hamsi512_cpu_hash_64(int thr_id, uint32_t threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);
 
@@ -151,12 +153,14 @@ extern "C" void x17hash(void *output, const void *input)
 }
 
 static bool init[MAX_GPUS] = { 0 };
+static bool use_compat_kernels[MAX_GPUS] = { 0 };
 
 extern "C" int scanhash_x17(int thr_id, struct work* work, uint32_t max_nonce, unsigned long *hashes_done)
 {
 	uint32_t *pdata = work->data;
 	uint32_t *ptarget = work->target;
 	const uint32_t first_nonce = pdata[19];
+	const int dev_id = device_map[thr_id];
 
 	uint32_t throughput =  cuda_default_throughput(thr_id, 1U << 19); // 19=256*256*8;
 	//if (init[thr_id]) throughput = min(throughput, max_nonce - first_nonce);
@@ -166,7 +170,7 @@ extern "C" int scanhash_x17(int thr_id, struct work* work, uint32_t max_nonce, u
 
 	if (!init[thr_id])
 	{
-		cudaSetDevice(device_map[thr_id]);
+		cudaSetDevice(dev_id);
 		if (opt_cudaschedule == -1 && gpu_threads == 1) {
 			cudaDeviceReset();
 			// reduce cpu usage
@@ -174,6 +178,11 @@ extern "C" int scanhash_x17(int thr_id, struct work* work, uint32_t max_nonce, u
 		}
 		gpulog(LOG_INFO, thr_id, "Intensity set to %g, %u cuda threads", throughput2intensity(throughput), throughput);
 
+		cuda_get_arch(thr_id);
+		use_compat_kernels[thr_id] = (cuda_arch[dev_id] < 500);
+		if (use_compat_kernels[thr_id])
+			x11_echo512_cpu_init(thr_id, throughput);
+
 		quark_blake512_cpu_init(thr_id, throughput);
 		quark_groestl512_cpu_init(thr_id, throughput);
 		quark_skein512_cpu_init(thr_id, throughput);
@@ -183,7 +192,6 @@ extern "C" int scanhash_x17(int thr_id, struct work* work, uint32_t max_nonce, u
 		x11_luffaCubehash512_cpu_init(thr_id, throughput);
 		x11_shavite512_cpu_init(thr_id, throughput);
 		x11_simd512_cpu_init(thr_id, throughput);
-		x11_echo512_cpu_init(thr_id, throughput);
 		x13_hamsi512_cpu_init(thr_id, throughput);
 		x13_fugue512_cpu_init(thr_id, throughput);
 		x14_shabal512_cpu_init(thr_id, throughput);
@@ -220,7 +228,11 @@ extern "C" int scanhash_x17(int thr_id, struct work* work, uint32_t max_nonce, u
 		x11_luffaCubehash512_cpu_hash_64(thr_id, throughput, d_hash[thr_id], order++);
 		x11_shavite512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
 		x11_simd512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
+		if (use_compat_kernels[thr_id])
 			x11_echo512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
+		else {
+			x16_echo512_cpu_hash_64(thr_id, throughput, d_hash[thr_id]); order++;
+		}
 		x13_hamsi512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
 		x13_fugue512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
 		x14_shabal512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);