ó àÆ÷Xc @`sÆddlmZmZmZddlZddlZddlZddlmZddl Z ddl Z ddl Z ddl jZeedddddd„Zd„Zd „Zed d ƒZejd d dddddeddƒejddddddeddƒejddddddddddddƒejdd dddd!dddddd"ƒejd#d$dddd%dddddd&ƒejd'dddd(dddddd)ƒejd*dddd+dddd,ƒejd-d.dddd/dd0dddd1ƒed2krÂejejƒ\ZZeedƒrgeejƒejdƒnej s}ed3ƒnej!dkr˜ej"Z!n ej!Z!ej#dkr¼ej"Z#n ej#Z#ej$dkràej"Z$n ej$Z$eej% ej de!d!e#d%e$d4ej&d+ej'ƒ\Z(Z)ej%r6qÂej rLee(ƒqÂeƒed5ej&d6d7ƒed8d6d7ƒed9e!e#e#e$fƒeƒed:e(e)fƒeƒed;ƒndS(<i(tabsolute_importtprint_functiontdivisionN(t OptionParseriÐi tCc C`sy|r“tdƒtdtjjjƒtdtjjƒtdtjjƒtdtjjƒtdƒtdtj ƒtdtj ƒtd tj ƒtd ƒtd t j d ƒƒtd t j dƒƒtdt j dƒƒtƒtdƒtjƒtdtjjƒtdtjƒtdtjƒtjjjdƒs]tjjjdƒr“tdƒtjtjjjjdfƒtƒq“ntjtj||fdtjjd|ƒƒ}tjtj||fdtjjd|ƒƒ}tjtj||fdtjjd|ƒƒ} tjgd| d| dtj||ƒfgƒ} t g| j!j"j#ƒD]} | j$j%j&dk^qoƒrgt'| j(j)| j(j*ƒD]0\} } | j$j%j&dkr¯t+| dƒ^q¯}t,|ƒdksýt-‚|d rd!}q~d"}net g| j!j"j#ƒD]} | j$j%j&d#k^q/ƒr\d$}n"d%}|t.| j!j"j#ƒƒ7}d }d&}| ƒ|rkt+td'ƒoÍt+tjd(ƒoÍt/| tjjj0ƒ}t+td)ƒoñt/| tj1j2ƒ}t3j3ƒ}xt4|ƒD] }| ƒq W|r7tjjj5ƒn|r\| j6d*t7d+t7ƒj8ƒnt3j3ƒ}n|||fS(,sn :param execute: If True, execute a Theano function that should call gemm. :param verbose: If True, will print some Theano flags and env variables. :param M,N,K: The M,N,K size used by gemm. :param iters: The number of calls to gemm to do. :return: a tuple (execution time, str that represents the implementation used) sSome Theano flags:s blas.ldflags=s compiledir=s floatX=s device=sSome OS information:s sys.platform=s sys.version=s sys.prefix=sSome environment variables:s MKL_NUM_THREADS=tMKL_NUM_THREADSs OMP_NUM_THREADS=tOMP_NUM_THREADSs GOTO_NUM_THREADS=tGOTO_NUM_THREADSsANumpy config: (used when the Theano flag "blas.ldflags" is empty)sNumpy dot module:sNumpy location:sNumpy version:tgpus nvcc version:s --versiontdtypetordertupdatesgš™™™™™Ù?gš™™™™™é?tGemmtcthunkiis(CPU (with direct Theano binding to blas)sPCPU (without direct Theano binding to blas but with numpy/scipy binding to blas)tGpuGemmtGPUs9ERROR, unable to tell if Theano used the cpu or the gpu: iÿÿÿÿtsandboxtcudatgpuarraytborrowtreturn_internal_type(9tprintttheanotconfigtblastldflagst compiledirtfloatXtdevicetsystplatformtversiontprefixtostgetenvtnpt show_configtdott __module__t__file__t __version__t startswithtinit_gpu_devicet subprocesstcallRRt nvcc_compilert nvcc_pathtsharedtonestfunctiontTtanytmakertfgraphttoposorttopt __class__t__name__tziptfntnodestthunksthasattrtlentAssertionErrortstrt isinstancetCudaNdarraySharedVariableRtGpuArraySharedVariablettimetranget synchronizet get_valuetTruetsync(texecutetverbosetMtNtKtitersR tatbtctftxtnodetthunktc_impltimpltt0tt1RJtsync2ti((s6/tmp/pip-build-X4mzal/theano/theano/misc/check_blas.pyRKs‚         $ $ $ 64%*   4   cC`stƒ|jS(N(RKtCOMPLETE(tstatetchannel((s6/tmp/pip-build-X4mzal/theano/theano/misc/check_blas.pyt jobman_jobkscC`stƒS(N(RK(((s6/tmp/pip-build-X4mzal/theano/theano/misc/check_blas.pyttestpstusagesq%prog Compute time needed to perform BLAS gemm computations between matrices of size (M, N) and (N, K).s-qs--quiettactiont store_truetdesttquiettdefaultthelps=If true, do not print the comparison table and config optionss --print_onlyt print_onlys)If true, do not perform gemm computationss-Ms--MtstoreRMttypetintsThe M size to gemms-Ns--NRNsThe N size to gemms-Ks--KROsThe K size to gemms--itertitersThe number of calls to gemms--orderR sŸThe numpy memory layout parameter used when creating the numpy.ndarray objects. It accepts 'C' for C memory order and 'F' for Fortran order (for all matrices).s-Bs--BtBiˆsThe M, N, and K for big gemmt__main__s% Some results that you can compare against. They were 10 executions of gemm in float64 with matrices of shape 2000x2000 (M=N=K=2000). All memory layout was in C order. CPU tested: Xeon E5345(2.33Ghz, 8M L2 cache, 1333Mhz FSB), Xeon E5430(2.66Ghz, 12M L2 cache, 1333Mhz FSB), Xeon E5450(3Ghz, 12M L2 cache, 1333Mhz FSB), Xeon X5560(2.8Ghz, 12M L2 cache, hyper-threads?) Core 2 E8500, Core i7 930(2.8Ghz, hyper-threads enabled), Core i7 950(3.07GHz, hyper-threads enabled) Xeon X5550(2.67GHz, 8M l2 cache?, hyper-threads enabled) Libraries tested: * numpy with ATLAS from distribution (FC9) package (1 thread) * manually compiled numpy and ATLAS with 2 threads * goto 1.26 with 1, 2, 4 and 8 threads * goto2 1.13 compiled with multiple threads enabled Xeon Xeon Xeon Core2 i7 i7 Xeon Xeon lib/nb threads E5345 E5430 E5450 E8500 930 950 X5560 X5550 numpy 1.3.0 blas 775.92s numpy_FC9_atlas/1 39.2s 35.0s 30.7s 29.6s 21.5s 19.60s goto/1 18.7s 16.1s 14.2s 13.7s 16.1s 14.67s numpy_MAN_atlas/2 12.0s 11.6s 10.2s 9.2s 9.0s goto/2 9.5s 8.1s 7.1s 7.3s 8.1s 7.4s goto/4 4.9s 4.4s 3.7s - 4.1s 3.8s goto/8 2.7s 2.4s 2.0s - 4.1s 3.8s openblas/1 14.04s openblas/2 7.16s openblas/4 3.71s openblas/8 3.70s mkl 11.0.083/1 7.97s mkl 10.2.2.025/1 13.7s mkl 10.2.2.025/2 7.6s mkl 10.2.2.025/4 4.0s mkl 10.2.2.025/8 2.0s goto2 1.13/1 14.37s goto2 1.13/2 7.26s goto2 1.13/4 3.70s goto2 1.13/8 1.94s goto2 1.13/16 3.16s Test time in float32 cuda version 6.5 6.0 5.5 5.0 4.2 4.1 4.0 3.2 3.0 # note gpu K6000/NOECC 0.06s 0.06s K40 0.07s K20m/ECC 0.08s 0.08s 0.07s K20/NOECC 0.07s M2090 0.19s C2075 0.25s M2075 0.25s M2070 0.25s 0.27s 0.32s M2070-Q 0.48s 0.27s 0.32s M2050(Amazon) 0.25s C1060 0.46s K600 1.04s GTX Titan Black 0.05s GTX Titan(D15U-50) 0.06s 0.06s don't work GTX 780 0.06s GTX 980 0.06s GTX 970 0.08s GTX 680 0.11s 0.12s 0.154s 0.218s GRID K520 0.14s GTX 580 0.16s 0.16s 0.164s 0.203s GTX 480 0.19s 0.19s 0.192s 0.237s 0.27s GTX 750 Ti 0.20s GTX 470 0.23s 0.23s 0.238s 0.297s 0.34s GTX 660 0.18s 0.20s 0.23s GTX 560 0.30s GTX 650 Ti 0.27s GTX 765M 0.27s GTX 460 0.37s 0.45s GTX 285 0.42s 0.452s 0.452s 0.40s # cuda 3.0 seems faster? driver version? 750M 0.49s GT 610 2.38s GTX 550 Ti 0.57s GT 520 2.68s 3.06s GT 520M 2.44s 3.19s # with bumblebee on Ubuntu 12.04 GT 220 3.80s GT 210 6.35s 8500 GT 10.68s Results for larger matrices. There were 10 executions of gemm in float32 with matrices of shape 5000x5000 (M=N=K=5000). All memory layout was in C order. cuda version 7.5 7.0 6.5 gpu M40 0.47s k80 0.96s K6000/NOECC 0.69s K40 0.88s K20m/ECC K20/NOECC M2090 C2075 M2075 M2070 M2070-Q M2050(Amazon) C1060 K600 GTX Titan X 0.45s 0.47s GTX Titan Black 0.64s 0.64s GTX Titan(D15U-50) GTX 780 GTX 980 Ti 0.41s GTX 980 GTX 970 0.66s GTX 680 1.57s GRID K520 GTX 750 Ti 2.01s 2.01s GTX 750 2.46s 2.37s GTX 660 2.32s 2.32s GTX 580 2.42s 2.47s GTX 480 2.87s 2.88s TX1 7.6s (float32 storage and computation) GT 610 33.5s RPs We executedtendt s-calls to gemm with a and b matrices of shapess(%d, %d) and (%d, %d).s"Total execution time: %.2fs on %s.s—Try to run this script a few times. Experience shows that the first time is not as fast as followings calls. The difference is not big, but consistent.(*t __future__RRRR!RREtoptparseRR+tnumpyR#Rt theano.tensorttensorR2RIRKRaRbtparsert add_optiontFalseR9t parse_argstargvtoptionst argumentsR>RRitexitRgRMRoRNRORjRnR ttRY(((s6/tmp/pip-build-X4mzal/theano/theano/misc/check_blas.pytsŒ      U           ~