Update to 2.10.

Biggest improvement over the 2.6'er we had before: automatic kernel ejection
for better and more flexible performance.

PR:		206120
Submitted by:	Johannes Dieterich <dieterich@ogolem.org> (maintainer)
This commit is contained in:
Raphael Kubo da Costa 2016-03-14 11:13:46 +00:00
parent f13db8aec5
commit bb27281dbc
Notes: svn2git 2021-03-31 03:12:20 +00:00
svn path=/head/; revision=411054
4 changed files with 74 additions and 6 deletions

View file

@ -2,7 +2,7 @@
# $FreeBSD$
PORTNAME= clblas
PORTVERSION= 2.6
PORTVERSION= 2.10
DISTVERSIONPREFIX= v
CATEGORIES= math
@ -13,7 +13,8 @@ LICENSE= APACHE20
LICENSE_FILE= ${WRKSRC}/LICENSE
BUILD_DEPENDS= opencl>=0:${PORTSDIR}/devel/opencl
LIB_DEPENDS= libOpenCL.so:${PORTSDIR}/devel/ocl-icd
LIB_DEPENDS= libOpenCL.so:${PORTSDIR}/devel/ocl-icd \
libboost_program_options.so:${PORTSDIR}/devel/boost-libs
RUN_DEPENDS= opencl>=0:${PORTSDIR}/devel/opencl
BROKEN_FreeBSD_9= clBLAS is only supported on FreeBSD 10.1 and newer
@ -23,10 +24,11 @@ GH_ACCOUNT= clMathLibraries
GH_PROJECT= clBLAS
USE_LDCONFIG= yes
USES= cmake
USES= cmake python
CMAKE_ARGS+= -DBUILD_TEST=OFF \
-DBUILD_KTEST=OFF \
-DCORR_TEST_WITH_ACML=OFF \
-DSUFFIX_LIB=""
CMAKE_SOURCE_PATH= ${WRKSRC}/src

View file

@ -1,2 +1,2 @@
SHA256 (clMathLibraries-clBLAS-v2.6_GH0.tar.gz) = 4607561a648949709bc7c368be4aaf7346174406e678454b643e31cfe861830c
SIZE (clMathLibraries-clBLAS-v2.6_GH0.tar.gz) = 954586
SHA256 (clMathLibraries-clBLAS-v2.10_GH0.tar.gz) = 72a46cb92f6057d10b503ce23d62db73d7332eb0645bb8bca3337f4c0558b429
SIZE (clMathLibraries-clBLAS-v2.10_GH0.tar.gz) = 1154117

View file

@ -0,0 +1,66 @@
--- src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_NN_B0_MX048_NX048_KX08_src.cpp.orig 2016-01-05 17:04:55 UTC
+++ src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_NN_B0_MX048_NX048_KX08_src.cpp
@@ -18,7 +18,7 @@ const unsigned int dgemm_Col_NN_B0_MX048
const unsigned int dgemm_Col_NN_B0_MX048_NX048_KX08_unroll = 8;
const char * const dgemm_Col_NN_B0_MX048_NX048_KX08_src = STRINGIFY(
-#pragma OPENCL EXTENSION cl_khr_fp64 : enable \n
+_Pragma("OPENCL EXTENSION cl_khr_fp64 : enable") \n
#define M6x6 \
rA[0] = lA[offA + 0];\
--- src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_NN_B1_MX048_NX048_KX08_src.cpp.orig 2016-01-05 17:04:55 UTC
+++ src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_NN_B1_MX048_NX048_KX08_src.cpp
@@ -18,7 +18,7 @@ const unsigned int dgemm_Col_NN_B1_MX048
const unsigned int dgemm_Col_NN_B1_MX048_NX048_KX08_unroll = 8;
const char * const dgemm_Col_NN_B1_MX048_NX048_KX08_src = STRINGIFY(
-#pragma OPENCL EXTENSION cl_khr_fp64 : enable \n
+_Pragma("OPENCL EXTENSION cl_khr_fp64 : enable") \n
#define M6x6 \
rA[0] = lA[offA + 0]; \
--- src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_NT_B0_MX048_NX048_KX08_src.cpp.orig 2016-01-05 17:04:55 UTC
+++ src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_NT_B0_MX048_NX048_KX08_src.cpp
@@ -18,7 +18,7 @@ const unsigned int dgemm_Col_NT_B0_MX048
const unsigned int dgemm_Col_NT_B0_MX048_NX048_KX08_unroll = 8;
const char * const dgemm_Col_NT_B0_MX048_NX048_KX08_src = STRINGIFY(
-#pragma OPENCL EXTENSION cl_khr_fp64 : enable \n
+_Pragma("OPENCL EXTENSION cl_khr_fp64 : enable") \n
\n
\ntypedef union _GPtr {
\n __global float *f;
--- src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_NT_B1_MX048_NX048_KX08_src.cpp.orig 2016-01-05 17:04:55 UTC
+++ src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_NT_B1_MX048_NX048_KX08_src.cpp
@@ -18,7 +18,7 @@ const unsigned int dgemm_Col_NT_B1_MX048
const unsigned int dgemm_Col_NT_B1_MX048_NX048_KX08_unroll = 8;
const char * const dgemm_Col_NT_B1_MX048_NX048_KX08_src = STRINGIFY(
-#pragma OPENCL EXTENSION cl_khr_fp64 : enable \n
+_Pragma("OPENCL EXTENSION cl_khr_fp64 : enable") \n
\n
\ntypedef union _GPtr {
\n __global float *f;
--- src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_TN_B0_MX048_NX048_KX08_src.cpp.orig 2016-01-05 17:04:55 UTC
+++ src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_TN_B0_MX048_NX048_KX08_src.cpp
@@ -18,7 +18,7 @@ const unsigned int dgemm_Col_TN_B0_MX048
const unsigned int dgemm_Col_TN_B0_MX048_NX048_KX08_unroll = 8;
const char * const dgemm_Col_TN_B0_MX048_NX048_KX08_src = STRINGIFY(
-#pragma OPENCL EXTENSION cl_khr_fp64 : enable \n
+_Pragma("OPENCL EXTENSION cl_khr_fp64 : enable") \n
__attribute__( (reqd_work_group_size(8, 8, 1)) )
__kernel void dgemm_Col_TN_B0_MX048_NX048_KX08_src (
--- src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_TN_B1_MX048_NX048_KX08_src.cpp.orig 2016-01-05 17:04:55 UTC
+++ src/library/blas/AutoGemm/UserGemmKernelSources/dgemm_Col_TN_B1_MX048_NX048_KX08_src.cpp
@@ -18,7 +18,7 @@ const unsigned int dgemm_Col_TN_B1_MX048
const unsigned int dgemm_Col_TN_B1_MX048_NX048_KX08_unroll = 8;
const char * const dgemm_Col_TN_B1_MX048_NX048_KX08_src = STRINGIFY(
-#pragma OPENCL EXTENSION cl_khr_fp64 : enable \n
+_Pragma("OPENCL EXTENSION cl_khr_fp64 : enable") \n
__attribute__( (reqd_work_group_size(8, 8, 1)) )
__kernel void dgemm_Col_TN_B1_MX048_NX048_KX08_src (

View file

@ -10,4 +10,4 @@ lib/cmake/clBLAS/clBLASTargets-%%CMAKE_BUILD_TYPE%%.cmake
lib/cmake/clBLAS/clBLASTargets.cmake
lib/libclBLAS.so
lib/libclBLAS.so.2
lib/libclBLAS.so.2.6.0
lib/libclBLAS.so.2.10.0