fe8c5666f9
optimized dgemm_kernel for HASWELL
2013-10-20 16:52:26 +02:00
f6b50057e2
corrected and testet FMA3 Code
2013-10-19 10:52:20 +02:00
2840d56aeb
added dgemm_kernel for Piledriver
2013-10-19 09:47:15 +02:00
2d49db2f5b
moved compiler flags from Makefile.rule to Makefile.arm
2013-10-16 19:04:42 +02:00
04391e6d9c
optimized param.h
2013-10-16 18:04:34 +02:00
85484a42df
added kernels for cgemm, ctrmm, zgemm and ztrmm
2013-10-16 18:00:41 +02:00
3983011f0b
added sgemm- and strmm_kernel
2013-10-14 08:22:27 +02:00
2a1515c9dd
added dgemm_ncopy_4_vfpv3.S
2013-10-12 16:48:29 +02:00
31f51e78bc
minor optimizations on dgemm_kernel
2013-10-12 09:42:18 +02:00
beffee7d91
Fixed buffer overflow bug in kernel/x86_64/dgemv_t.S file.
2013-10-11 03:20:20 +08:00
a35f4343fa
Merge pull request #301 from yieldthought/develop
...
Remove -Wl,--retain-symbols-file from dynamic link line to fix tool support
2013-10-09 00:46:49 -07:00
ce5626a384
Remove -Wl,--retain-symbols-file from dynamic library linking to fix tool support
...
The aim is to restrict the symbols exported from openBLAS, but for dynamic libraries --retain-symbols-file has the opposite effect intended. It removes symbols from the .symtab section of the library, which is used by tools such as debuggers, profilers, objdump etc. but does not remove them from .dynsym, which is the section used by the runtime when loading the binary.
In short, using --retain-symbols-file like this doesn't prevent symbol leakage but does prevent tools from analyzing applications linked with openBLAS.
This patch improves tool support for openBLAS applications but does not address symbol leakage through .dynsym.
2013-10-08 16:37:17 +02:00
e0b968c3a7
Changed kernels for dgemm and dtrmm
2013-10-05 12:59:44 +02:00
93f1074dd4
changed some values for arm
2013-09-30 18:03:56 +02:00
1c63180bb6
updated dgemm_kernel_8x2_vfpv3.S
2013-09-30 17:31:23 +02:00
22a8fcc4b7
add modified c_check perl program
2013-09-29 19:42:33 +02:00
9965d48005
added Makefile.arm
2013-09-29 18:55:21 +02:00
4a474ea7dc
changed dgemm_kernel to use fused multiply add
2013-09-29 17:46:23 +02:00
69ce737cc5
modified Makefile.L3 for ARM
2013-09-28 19:13:47 +02:00
d13788d1b4
common files modified for ARM
2013-09-28 19:10:32 +02:00
70411af888
initial checkin of kernel/arm
2013-09-28 19:02:25 +02:00
16eb780e13
Refs #262 . Fixed compatibility issues of GNU stack markings with PathScale EKOPath(tm) Compiler Suite: Version 4.0.12.1
2013-09-22 09:37:59 +08:00
a746724e84
Added backers.
2013-09-05 15:39:45 +08:00
3f7b0cd994
Merge pull request #290 from larsmans/missing-threshold
...
check if GEMM_MULTITHREAD_THRESHOLD defined in gemm.c
Set a fallback value.
2013-08-29 00:33:55 +08:00
cc6db2ecfe
Merge pull request #291 from larsmans/fix-makefile-prefix
...
fix default prefix handling in makefiles
2013-08-28 09:26:16 -07:00
3175be4b3d
Merge pull request #289 from larsmans/no-noconst
...
get rid of the generated cblas_noconst.h file
2013-08-28 09:25:23 -07:00
a29e6592da
fix default prefix handling in makefiles
...
PREFIX wasn't communicated to Makefile.install (where it matters)
by Makefile. The result is that the default PREFIX is empty and
OpenBLAS was being installed in /lib.
2013-08-28 17:41:31 +02:00
212463dce9
get rid of the generated cblas_noconst.h file
2013-08-28 16:52:24 +02:00
037bd82bef
Merge pull request #288 from sebastien-villemot/develop
...
Avoid failure on qemu guests declaring an Athlon CPU without 3dnow!
2013-08-28 06:26:37 -07:00
eae4cfa3f6
Avoid failure on qemu guests declaring an Athlon CPU without 3dnow!
...
The present patch verifies that, on machines declaring an Athlon CPU model and
family, the 3dnow and 3dnowext feature flags are indeed present. If they are
not, it fallbacks on the most generic x86 kernel. This prevents crashes due to
illegal instruction on qemu guests with a weird configuration.
Closes #272
2013-08-28 14:29:42 +02:00
6c4a7d0828
Import AMD Piledriver DGEMM kernel generated by AUGEM.
...
So far, this kernel doesn't deal with edge.
AUGEM: Automatically Generate High Performance Dense Linear Algebra
Kernels on x86 CPUs.
Qian Wang, Xianyi Zhang, Yunquan Zhang, and Qing Yi. In the
International Conference for High Performance Computing, Networking,
Storage and Analysis (SC'13). Denver, CO. Nov, 2013.
2013-08-25 10:16:01 -03:00
fe98de2f68
Merge branch 'bulldozer' into develop
2013-08-24 11:46:18 -03:00
db389b5915
Refs #281 . Detect __CYGWIN__ macro for Cygwin x86_64.
...
Signed-off-by: Zhang Xianyi <traits.zhang@gmail.com>
2013-08-24 14:50:17 +08:00
52f587db7f
Refs #281 . Detect _WIN32 macro for Windows API.
...
http://www.mail-archive.com/bug-gnulib@gnu.org/msg05722.html
2013-08-24 01:10:02 +08:00
067e8417fd
removed unnessesary instructions from zgemm_kernel_2x2_bulldozer.S
2013-08-23 22:22:43 +08:00
a82da3d069
removed unnessesary instructions
2013-08-23 22:22:27 +08:00
1569bf14f8
Refs #282 . Fixed zgemv_n typo bug on Win64.
2013-08-23 16:27:17 +08:00
df554aebd2
Merge pull request #280 from ViralBShah/develop
...
Patch LAPACK XLASD4.f as discussed in JuliaLang/julia#2340
2013-08-21 08:21:51 -07:00
eae6920f2d
Patch LAPACK XLASD4.f as discussed in JuliaLang/julia#2340
2013-08-21 19:14:07 +05:30
c92ae012a6
Refs #279 . Provide ONLY_CBLAS flag. If you only need CBLAS without
...
a fortran compiler, please try make ONLY_CBLAS=1.
This mode only compiler CBLAS without BLAS fortran interface and LAPACK.
2013-08-21 00:03:25 +08:00
f51a849d91
Merge pull request #278 from wernsaar/haswell
...
Merge wernsaar's Haswell gemm kernels.
2013-08-17 08:24:37 -07:00
44ef70420c
added cgemm_kernel_8x2_haswell.S
2013-08-16 18:54:56 +02:00
d488b1b1aa
added zgemm_kernel_4x2_haswell.S
2013-08-16 10:29:47 +02:00
4070d9a123
added dgemm_kernel_16x2_haswell.S
2013-08-15 19:17:20 +02:00
0b90c0ec64
added sgemm_kernel_16x4_haswell.S
2013-08-15 18:46:14 +02:00
2b8ab8f55b
sgemm_kernel_16x4_haswell.S minor changes
2013-08-14 01:44:41 +02:00
1cb9579cd0
added zgemm_kernel_4x2_haswell.S and fixed a bug in sgemm_kernel_16x4_haswell.S
2013-08-14 01:23:15 +02:00
2638370844
Init code base for Intel Haswell.
2013-08-13 00:54:59 +08:00
89637f87c8
added sgemm- and dgemm-kernel for HASWELL processor
2013-08-12 18:04:10 +02:00
c0b1e41bec
Merge branch 'bulldozer' into develop
2013-08-12 23:22:10 +08:00
wernsaar
wangqian
Zhang Xianyi
yieldthought
Lars Buitinck
Sébastien Villemot
Viral B. Shah