floraachy/OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
843 Commits 47 Branches 64 Tags
f331cb1a76fded2efbf5aec7fe3a53bb73dc12e9
Commit Graph

98 Commits

Author SHA1 Message Date
Olivier Grisel
2c556f093a Add cast to function pointer to remove warning 2014-02-25 11:08:32 +01:00
Olivier Grisel
3b027d2528 Do not reference pthread_atfork in non-SMP_SERVER mode 2014-02-25 11:08:32 +01:00
Olivier Grisel
49bd98f410 Do not reference pthread_atfork under windows 2014-02-19 19:25:48 +01:00
Olivier Grisel
138a841390 FIX #294: make OpenBLAS thread-pool resilient to fork via pthread_atfork 2014-02-19 19:01:15 +01:00
Olivier Grisel
046e4013cb Revert "Refs #294. Used pthread_atfork to avoid hang after a Unix fork." 
This reverts commit 3617c22a56.
 2014-02-19 18:32:54 +01:00
Zhang Xianyi
3617c22a56 Refs #294. Used pthread_atfork to avoid hang after a Unix fork. 
The problem is the mutex we used in blas_server. Thus, we must clear
the mutex before the fork and re-init them at parent and child process.

If you used OpenMP, GOMP has the same problem by now. Please try other OpenMP
implemantation.
 2014-02-18 15:36:04 +08:00
Zhang Xianyi
8c7687b419 Refs #338. Added OPENBLAS_VERBOSE environment variable on runtime 
By default, OpenBLAS doesn't output the warning message. You can set
OPENBLAS_VERBOSE (e.g. export OPENBLAS_VERBOSE=1) to enable the warning
message on runtime.
 2014-01-24 02:05:59 +08:00
Zhang Xianyi
ab69443bd4 Refs #332. Added addtional Intel Ivy Bridge and Haswell CPU-id. 2014-01-05 23:44:29 +08:00
Zhang Xianyi
b263e096af Refs #307. Delete debug printf. 2013-12-31 15:53:13 +08:00
wernsaar
8191efc420 changed level3.c 2013-12-01 13:52:08 +01:00
wernsaar
0b6e13b689 Merge remote branch 'origin/develop' into haswell 2013-12-01 13:38:11 +01:00
wernsaar
5c648a8984 Merge remote branch 'origin/develop' into haswell 2013-12-01 11:25:33 +01:00
Zhang Xianyi
5048a80032 Refs #283. Fixed the incorrect usage of long data type for Windows 64. 2013-11-14 13:46:42 +08:00
Zhang Xianyi
a2942456ef Refs #307. Fixed the hang bug when free OpenBLAS dll in Windows. 2013-11-13 10:00:18 +08:00
wernsaar
a77c71eaf5 added highly optimized dgemm_kernel for HASWELL 2013-10-28 10:23:47 +01:00
wernsaar
fe8c5666f9 optimized dgemm_kernel for HASWELL 2013-10-20 16:52:26 +02:00
Sébastien Villemot
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
Zhang Xianyi
2638370844 Init code base for Intel Haswell. 2013-08-13 00:54:59 +08:00
Zhang Xianyi
673e453b3f Enable bulldozer kernels. 2013-08-05 16:07:54 +08:00
Zhang Xianyi
143cca4dd5 Merge branch 'develop' into bulldozer 2013-08-05 15:51:53 +08:00
Zhang Xianyi
534c5ec919 Fixed #261. Use strncmp instead of a comparing trick. 2013-07-29 16:48:35 +08:00
Zhang Xianyi
5b504d6c23 Refs #263. Rollback bulldozer and piledriver kernels to barcelona kernels. 2013-07-28 17:39:24 +08:00
Zhang Xianyi
72b1edaf1b Merge branch 'develop' into bulldozer 
Conflicts:
	kernel/x86_64/KERNEL.BULLDOZER
 2013-07-28 06:38:25 +02:00
Zhang Xianyi
4471c77905 Fixed #261. Use strncmp instead of a comparing trick. 2013-07-26 23:43:54 +08:00
Zhang Xianyi
77b572fa0b Merge branch 'loongson3a' into develop 
Conflicts:
	Makefile.system
 2013-07-20 22:33:17 +08:00
Zhang Xianyi
2a7503e563 Refs #225. Fixed a bug in GEMM OpenMP threading. 2013-07-15 09:56:19 +08:00
grisuthedragon
c19a488af2 create openblas_get_parallel to retrieve information which 
parallelization model is used by OpenBLAS.
 2013-07-11 21:39:19 +08:00
Zhang Xianyi
32d2ca3035 Refs #214, #221, #246. Fixed the getrf overflow bug on Windows. 
I used a smaller threshold since the stack size is 1MB on windows.
 2013-07-11 03:20:02 +08:00
wernsaar
6f008abcef replaced defined(DOUBLE) by !defined(XDOUBLE) 2013-07-09 18:17:50 +02:00
Zhang Xianyi
f54f5bac9e Refs #248. Fixed the LSB compatiable issue for BLAS only. 
For example, make CC=lsbcc NO_LAPACK=1.
 2013-07-09 15:38:03 +08:00
Zhang Xianyi
5d3312142a Refs #221 #246. Fixed the overflowing stack bug in mutlithreading BLAS3. 
When NUM_THREADS(MAX_CPU_NUNBERS) is very large ,e.g. 256.

typedef struct {
  volatile BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
} job_t;

job_t          job[MAX_CPU_NUMBER];

The job array is equal 8MB.

Thus, We use malloc instead of stack allocation.
 2013-07-08 01:07:05 +08:00
Zhang Xianyi
886cbaf4e4 Support AMD Piledriver by bulldozer kernels. 2013-07-06 12:06:43 -03:00
Zhang Xianyi
32dbeb636d Refs #221. Set stack limit to 16MB to prevent a SEGFAULT bug on Mac OS X with DYNAMIC_ARCH=1 & NUM_THREADS=256. 2013-07-02 14:17:55 +08:00
Dan Luu
88ef307cef Refs #241. Add Haswell support (using sandybridge optimizations) 2013-06-30 22:35:14 +08:00
Zhang Xianyi
cd1d473ba0 Merge pull request #230 from wernsaar/develop 
Refs #230. New dgemm and sgemm Kernel for BULLDOZER
 2013-06-13 07:29:27 -07:00
wernsaar
25491e42f9 New dgemm kernel for BULLDOZER: dgemm_kernel_8x2_bulldozer.S 2013-06-08 09:40:17 +02:00
Zhang Xianyi
65ffead0cf Refs #124. Check XSAVE flag on x86 CPU. 2013-06-06 22:50:43 +08:00
Xianyi Zhang
6b01d58712 Disable the optimization of muli-threading gemm on the Loongson3A. 2013-03-30 20:12:43 +00:00
Zhang Xianyi
f1ce74ffdd Improved the print when OS don't support AVX. 2013-03-02 14:15:54 +08:00
Zhang Xianyi
d744c9590a In OpenMP threading, preallocate the thread buffer instead of allocating the buffer every time. This patch improved the performance slightly. 2013-03-01 14:36:47 +08:00
Zhang Xianyi
3cc6ae793e Refs #174. Return sb pointer when OpenMP or Windows. 2013-02-26 00:48:21 +08:00
Zhang Xianyi
5155e3f509 Refs #174. Fixed the overflowing buffer bug of multithreading hbmv and sbmv. 
Instead of using thread 0 buffer, each thread uses its own sb buffer.
Thus, it can avoid overflowing thread 0 buffer.
 2013-02-13 16:05:58 +08:00
Zhang Xianyi
5c8bf6ae0e Merge branch 'bulldozer' into develop 2013-02-10 01:19:42 +08:00
Zhang Xianyi
6ae2f868fd Set the affinity. Only use 1 core of each module on bulldozer. 2013-02-09 18:19:02 +01:00
Zhang Xianyi
299b5a44dc Merge branch 'develop' of github.com:xianyi/OpenBLAS into bulldozer 2013-02-09 16:22:04 +01:00
Zhang Xianyi
8cdb795438 Refs #187. Use binary code for xgetbv, which is compatible with old compiler. 2013-01-22 00:25:08 +08:00
Zhang Xianyi
a4ee6f3915 Fixed #172. Support Intel Xeon E7540. 2012-12-18 08:57:46 +08:00
Zhang Xianyi
fba6b590f2 Merge branch 'master' into develop 2012-12-15 22:49:37 +08:00
Julian Taylor
1138817dd2 add a sanity check on the detected cpu type 
if we have 64 bit pointers we can't have a 32 bit cpu, so fall back to
the 64bit cpu fallback (prescott)
E.g. the cpu detection fails in amd qemu64 emulation (family 6 model 2)
causing it to use the uninitialized gotoblas_ATHLON
 2012-12-15 13:29:46 +01:00
Zhang Xianyi
bdf8d9411e Refs #163. Obtain the build configure on runtime. 
openblas_get_config function returns the configure string.
So far, it supports USE64BITINT, NO_CBLAS, NO_LAPACK, NO_LAPACKE,
DYNAMIC_ARCH, NO_AFFINITY.

Example:
 #include <stdio.h>
extern char * openblas_get_config();
void main()
{
  printf("%s\n",openblas_get_config());
  return;
}
 2012-12-10 15:52:51 +08:00