Merge branch 'kernels'

AUTHORS.md

@@ -1,6 +1,6 @@
 # Authors
 
-## Current
+## Core
 
 * Scrypt-only refactor and maintenance: Noel Maersk <veox@wemakethings.net> LfxRFgXzA13TSTxgFGUFEtumv5ShGzAmLw
 * Core: Martin Danielsen <kalroth@gmail.com> 1DNBcSEENBwDKrcTyTW61ezWhzsPy5imkn
@@ -9,6 +9,13 @@
 * API: Andrew Smith <kan0i {at} kano-kun [dot] net> 1Jjk2LmktEQKnv8r2cZ9MvLiZwZ9gxabKm
 
 
+## OpenCL kernels
+
+* scrypt: Colin Percival
+* zuikkis: Zuikkis LeXck7EYgxyjw13zNDxZFmmgmWffFvhmSh
+* alexkarnew/alexkarold: Alexey Karimov LMqRcHdwnZtTMH6c2kWoxSoKM5KySfaP5C
+
+
 ## Bug fixes
 
 * Gabriel Devenyi

Makefile.am

@@ -37,7 +37,7 @@ sgminer_SOURCES += ocl.c ocl.h
 sgminer_SOURCES += findnonce.c findnonce.h
 sgminer_SOURCES += adl.c adl.h adl_functions.h
 sgminer_SOURCES += scrypt.c scrypt.h
-sgminer_SOURCES += *.cl
+sgminer_SOURCES += kernel/*.cl
 
-bin_SCRIPTS	= $(top_srcdir)/*.cl
+bin_SCRIPTS	= $(top_srcdir)/kernel/*.cl
 

README.md

@@ -18,9 +18,10 @@ License: GPLv3.  See `COPYING` for details.
 Documentation is available in directory `doc`. For details on several topics, see:
 
 * `API` for the RPC API specification;
-* `SCRYPT` for how to find the right balance in GPU configuration to mine Scrypt-based coins effectively;
 * `FAQ` for frequently asked questions;
 * `GPU` for semi-obsolete information on GPU configuration options and mining SHA256d-based coins;
+* `KERNEL.md` for OpenCL kernel-related information;
+* `MINING` for how to find the right balance in GPU configuration to mine Scrypt-based coins effectively;
 * `windows-build.txt` for information on how to build on Windows.
 
 Note that **most of the documentation is outdated**. If you want to contribute, fork this repository, update as needed, and submit a pull request.

configure.ac

@@ -317,8 +317,10 @@ fi
 
 AC_DEFINE_UNQUOTED([SGMINER_PREFIX], ["$prefix/bin"], [Path to sgminer install])
 
-AC_DEFINE_UNQUOTED([SCRYPT_KERNNAME], ["scrypt130511"], [Filename for scrypt kernel])
-
+AC_DEFINE_UNQUOTED([ALEXKARNEW_KERNNAME], ["alexkarnew"], [Filename for Alexey Karimov's optimised kernel for Catalyst >=13.4])
+AC_DEFINE_UNQUOTED([ALEXKAROLD_KERNNAME], ["alexkarold"], [Filename for Alexey Karimov's optimised kernel for Catalyst <13.4])
+AC_DEFINE_UNQUOTED([CKOLIVAS_KERNNAME], ["ckolivas"], [Filename for original scrypt kernel])
+AC_DEFINE_UNQUOTED([ZUIKKIS_KERNNAME], ["zuikkis"], [Filename for Zuikkis' optimised kernel])
 
 AC_SUBST(OPENCL_LIBS)
 AC_SUBST(OPENCL_FLAGS)

doc/KERNEL.md

@@ -0,0 +1,86 @@
+# Kernels
+
+## Available OpenCL kernels
+
+See directory `kernel`.
+
+
+## Parameter configuration
+
+### Common
+
+In general, switching kernels requires reconfiguring mining parameters,
+such as (but not necessarily limited to) `thread-concurrency`, `intensity`,
+`gpu-engine` and `gpu-memclock`.
+
+A description of how to do this is available in `doc/MINING`.
+
+
+### alexkarnew
+
+Alaxey Karimov's optimised kernel, based on `ckolivas`. For Catalyst >=13.4.
+
+Only supports `vectors=1`.
+
+[Announcement](https://litecointalk.org/index.php?topic=4082.0).
+
+
+### alexkarold
+
+Alaxey Karimov's optimised kernel, based on `ckolivas`. For Catalyst <13.4.
+
+Only supports `vectors=1`.
+
+[Announcement](https://litecointalk.org/index.php?topic=4082.0).
+
+
+### ckolivas
+
+The original Colin Percival `scrypt` kernel, maintained for a long time by
+Con Kolivas in `cgminer` and renamed to reflect the fact.
+
+Only supports `vectors=1`.
+
+
+### zuikkis
+
+Zuikkis' optimised kernel, based on `ckolivas`.
+
+Only supports `vectors=1`, `lookup-gap=2` and `worksize=256`.
+
+[Announcement](https://litecointalk.org/index.php?topic=6058.msg90873#msg90873).
+
+
+## Submitting new kernels
+
+### Requirements
+
+* OpenCL source code only, licenced under GPLv3 (or later).
+* Not hard-coded for a specific GPU model or manufacturer.
+* Known limitations and any specific configuration quirks must be mentioned.
+
+
+### Procedure
+
+1. Copy the kernel you wish to modify and commit it verbatim.
+
+This way, it is easy to verify that there are no hidden changes. Note in
+the commit message which kernel is used as a base.
+
+2. Make changes to the kernel. Commit them.
+
+This allows to produce a diff that makes sense.
+
+3. Search for KL_CKOLIVAS and CKOLIVAS_KERNNAME in the top-level source
+directory and make additions to the listed files in order to integrate
+the new kernel.
+
+Now it can be selected when starting via the `--kernel` argument or
+`kernel` configuration option.
+
+4. Add yourself to the "kernels" section in `AUTHORS.md`. Keep it short.
+
+5. Submit a pull request on GitHub, or file it at the issue tracker,
+outlining the changes made, known limitations, and tested GPUs. List
+your git repository and branch name. The current repository and issue
+tracker links should be in `README.md`.

driver-opencl.c

@@ -196,8 +196,14 @@ char *set_thread_concurrency(char *arg)
 
 static enum cl_kernels select_kernel(char *arg)
 {
-	if (!strcmp(arg, "scrypt"))
-		return KL_SCRYPT;
+	if (!strcmp(arg, ALEXKARNEW_KERNNAME))
+		return KL_ALEXKARNEW;
+	if (!strcmp(arg, ALEXKAROLD_KERNNAME))
+		return KL_ALEXKAROLD;
+	if (!strcmp(arg, CKOLIVAS_KERNNAME))
+		return KL_CKOLIVAS;
+	if (!strcmp(arg, ZUIKKIS_KERNNAME))
+		return KL_ZUIKKIS;
 
 	return KL_NONE;
 }
@@ -1291,8 +1297,17 @@ static bool opencl_thread_prepare(struct thr_info *thr)
 	if (!cgpu->kname)
 	{
 		switch (clStates[i]->chosen_kernel) {
-			case KL_SCRYPT:
-				cgpu->kname = "scrypt";
+			case KL_ALEXKARNEW:
+				cgpu->kname = ALEXKARNEW_KERNNAME;
+				break;
+			case KL_ALEXKAROLD:
+				cgpu->kname = ALEXKAROLD_KERNNAME;
+				break;
+			case KL_CKOLIVAS:
+				cgpu->kname = CKOLIVAS_KERNNAME;
+				break;
+			case KL_ZUIKKIS:
+				cgpu->kname = ZUIKKIS_KERNNAME;
 				break;
 			default:
 				break;
@@ -1322,11 +1337,15 @@ static bool opencl_thread_init(struct thr_info *thr)
 	}
 
 	switch (clState->chosen_kernel) {
-		case KL_SCRYPT:
-			thrdata->queue_kernel_parameters = &queue_scrypt_kernel;
-			break;
-		default:
-			break;
+	case KL_ALEXKARNEW:
+	case KL_ALEXKAROLD:
+	case KL_CKOLIVAS:
+	case KL_ZUIKKIS:
+		thrdata->queue_kernel_parameters = &queue_scrypt_kernel;
+		break;
+	default:
+		applog(LOG_ERR, "Failed to choose kernel in opencl_thread_init");
+		break;
 	}
 
 	thrdata->res = calloc(buffersize, 1);