by

Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
Grigory Starkov
@Gregstrq
I have i7 with 8 logical cores in hyperthreaded mode.
Yes. In all cases the loop is single-threaded.
Andrew
@brada4
Exact CPU trade name would be beneficial. New processors clock down to half with all cores employed + at the same moment it could activate some thermal throttling in CPU. It will require you to do accirate CPU temperature monitoring while GEMV-s run (try all 1-2-3-4)
I have some suspicion regarding thread affinity . Can you open an issue in github + code in julia triggering it?
Grigory Starkov
@Gregstrq

I have Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz.
Also I've checked temperature with sensors and CPU frequency with lscpu.
Temperature of idle cores is 45, freq is about 800 MHz.
For one process the temp of active core is 94, other cores --- 65. Freq is about 3.5GHz.
For 4 parallel processes temp of all cores is about 55, freq is kept at about 1500 MHz.

If I use 3 processes will clocking down be disabled? Is there a way to allow system to use fans actively and to disable thermal throttling?

avocadochicken
@avocadochicken
Hello everyone! I cloned the git repo and started compiling OpenBLAS (x86_64 AMD, kernel: 4.4.0, gcc: 5.4.0). My cpu is working (1 core) on 'TEST 2/2 fork:safety' since SEVERAL HOURS! I assume this is a bug. How to troubleshoot?
Its in folder OpenBLAS/utest, compiled utest_main.c, utest_amax.c and test_fork.c. And I think it was linked into openblas_utest, which is running now for ages...
'TEST 1/2 amax:samax [OK]' was working fine
avocadochicken
@avocadochicken
Some else just opened an issue (=> 1456).
Andrew
@brada4
@Gregstrq you need to check/fix your cooling and power setup. Parking cores is short-term workaround to prevent CPU from frying (or throttiling down), Intel ARK has details on power draw (which all is emitted as heat under heavy load)
shivaraj karthikeyan
@shivaraj1996_twitter
Hello everyone! I want to set different cpu affinity for each thread created by openblas_set_num_threads(8) method. Could anyone explain me how to do that?
Ethan Smith
@ethanhs
Has anyone tried building with libcxx on Windows with clang-cl?
Andrew
@brada4
OpenBLAS is written in C, which means DLL does not require C++ libraries.
Ethan Smith
@ethanhs
Oh great!
Prajwal B V
@VisitantP
Hello sir
Trying to build opencv with contrib along with openblas
but I couldn't locate the library file in the package of openblas
I am from Electrical Engineering and not very proficient in coding... any help is greatly appreciated
Jacob Bogers
@Jacob_Bogers_twitter
hi, i read the remark 'openblas is written in c', can someone, i thought openblas was written in purely machine langauge,
oh my bad, that was "goblas"
oops
Andrew
@brada4
It is written in C that calls assembler fragments aka kernels that implement BLAS function on something like 16x16 matrix blocks, in parallel on multiple CPU cores.
Also there is whole LAPACK library (except for BLAS functions described above), written in fortran , added to project , so that full reference LAPACK API is provided.
There are parts that interact with operating system, like threading, where assembly is not an option.
You can examine code in interface/ - parallel dispatchers
driver/ - the actual parallel code in c
kernel/ - assembly kernels, and optimized C kernels, later like handling batches of 2^n floats, so that compiler elects for SIMD instructions on particular CPU
Andrew
@brada4
It is very similar to gotoblas, adding some performance tuning and support for newer CPUs, like AVX ISA and above.
Jacob Bogers
@Jacob_Bogers_twitter
gotoblas, yes,
thanks for explenation,
how much faster is openblas compared to the blas written in fortran? (the reference implementation)
thanks for your answers
Should be 5-20 times faster per core, with some sub-optimality parallelizing, but still very close to linear parallelism speedup for huge enough in/out samples.
It will be slightly slower in lapack part than MKL, as MKL includes significant algebraical improvements over reference LAPACK, but will be on par with BLAS.
... BLAS part of LAPACK vs BLAS part of MKL.
Jacob Bogers
@Jacob_Bogers_twitter
@brada4 what is every bar verticly, there are no units
and reference is the fortran lib?
do you have other per stats other then dgemm?
i mean there are approx 149 functions in BLAS (including the ones dealing with complex matrices)
Andrew
@brada4
Blas level1 is bound to memory speed - new RAM generation, better spped. level2 complexity is approximately size of data, typically also bound to memory speed, though needing more CPU power. Level 3 is bound to procesing speed. Probably web search yields you better comparisons. You can make your own benchmark, like based on benchmarks/* programs/scripts if you encounter sort of regression vs other BLAS implementation.
Units feature prominently over top of graph - sandybridge CPU core means 20Gflops with openblas or MKL.
idrigo
@idrigo
Guys, hello! I’m seeking for help.
Was trying to compile some fortran code and the exceptions cannot find -llapack and cannot find -lblas were thrown. So, I downloaded and compiled OpenBIAS code. The question is: where can I locate these libraries?
Thanks a lot for help!
Andrew
@brada4
You are looking for library names provided by reference BLAS, probably available via your system's package manager. Probably seek in FAQs if you intent to use openblas instead.
Jacob Bogers
@Jacob_Bogers_twitter
netlib has a pdf cheat sheet of all blas functions
for people who are interestedin mobile blas
finished somemonth ago
Özgür Akkurt
@oezgurmakkurt
hi, how can I see what blocksize openblas uses on my cpu
Jacob Bogers
@Jacob_Bogers_twitter
what do you mean by "blocksize"?
what is your cpu?
Plain Wu
@plainwuatdeepforce_gitlab
Hi, I got a bunch of errors when building OpenBlas on nVidia Nano (CORTEXA57):
OpenBLAS/Source/kernel/arm64/cgemm_kernel_8x4.S:1455: Error: unknown mnemonic op_ri' --op_ri v19.4s,v2.4s,v9.s[0]' ... lots of op_ir, op_rr those unknowns. Any one came across this issue?
Plain Wu
@plainwuatdeepforce_gitlab
My bad. I tried on latest commit, and it goes well. The error occurred on tag v0.3.2.
Ruchika
@ruchikabgosain
Hi, im trying to test my patch from github on SUSE linuxone server. I have successfully ssh'ed and got the terminal access for the same. Can i know how to proceed?