B \<+@sNddlmZmZddlZddlZddlZddlZddlZddl Z ddl Z ddl m Z m Z ddlZddlmZmZddlmZmZddlmZddlmZmZmZeejjd Zd d d d Zej ej gej!ej!ggej"ej"gej#ej#gej$gej%gej&gej'ggej(gej)ggggggej*ej*gej+gej,gggggej-ej-gggggej.ej.gej/ej/gggggggej0ej0gej1ej1gej2ej2gej3ej3ggd*Z4dddddddgZ5dde46DZ4dde46DZ7dde46DZ8d+ddZ9ddZ:d d!Z;eGd"d#d#eZd&d'Z?eGd(d)d)eZ@eAd*krJeBdS),)divisionprint_functionN)chain combinations)prangeunittest_support)compile_isolatedFlags)exec_)TestCasetagoverride_env_configz"SVML tests need SVML to be presentZnehalemZhaswellzskylake-avx512))*sincospowexplogacosacoshasinasinhZatan2atanatanhZcbrtZcdfnormZ cdfnorminvZceilZcosdcosherferfcZerfcinvZerfinvZexp10Zexp2expm1ZfloorZfmodZhypotZinvsqrtlog10log1pZlog2ZlogbZ nearbyintZrintroundZsindsinhsqrttantanhZtruncr%r&r!r r"r'rcCs"i|]\}}t|dkr||qS)r)len).0kvr,4lib/python3.7/site-packages/numba/tests/test_svml.py Qsr.cCs&g|]\}}ddd|Dkr|qS)zTsz.r,)r)fr+r,r,r-r3Ssr3cCs&g|]\}}ddd|Dkr|qS)z .r,)r)r4r+r,r,r-r3Us cCs|dkr.ddd|D}d||||f} np|dkrhd|||d|f} | |dr`d |nd 7} n6|d ks||d ks|tdd d|D}djft} |dkp|dk} | r|dn|} | r|dn|} t|ddrdnd} tjjrd| nd| }d| | | f}|dkr"|g}d|g}n>|g}g}|dkr`| sF|dkr`|dd| | d| fg7}|dkr|dkrdg}||g}n|dkrd g}||g}| ||fS)!z For a given function and its usage modes, returns python code and assembly patterns it should and should not generate scalar,cSsg|] }|dqS)z[0]r,)r)ar,r,r-r3asz!func_patterns..z%s%s[0] += math.%s(%s) numpyz%s%s += np.%s(%s)intz.astype(np.%s)  rangercSsg|] }|dqS)z[i]r,)r)r8r,r,r-r3hszP{pad}for i in {mode}({res}.size): {pad}{pad}{res}[i] += math.{func}({arg_list}) float32 complex64r4rfastmathFZ_haz$_$z__svml_%s%d%s,Z__svml_rint32z%zmmr%ZsqrtsZvsqrtp) join startswithAssertionErrorformatlocalsgetattrnumbaconfigIS_OSX)funcargsZresdtypemodevlenflagsZpadarg_listbodyZis_f32r4r+Z prec_suffZ scalar_funcZ svml_funccontainsavoidsr,r,r- func_patternsYs>      rVcCsdjftS)z, Returns pretty name for given set of modes z%{dtype}_{mode}{vlen}_{flags.__name__})rFrG)rNrOrPrQr,r,r- usecase_namesrWc Cst||||}djft}t|dkr*tnt}|drD|t}t}t}xD|D]<} t | dgd||||\} } } | | || 7}| | qVW|d7}i} t |t | || |_ | |||fS)z? Combine multiple function calls under single umbrella usecase z\def {name}(n): x = np.empty(n*8, dtype=np.{dtype}) ret = np.empty_like(x) r9complexxretz return ret)rWrFrGset numpy_funcs other_funcsrD differencecomplex_funcs_excluderVupdater globals__doc__)rNrOrPrQnamerSZfuncsrTrUr4bcr8Zldictr,r,r-combo_svml_usecases"     rfc@sFeZdZdZdZeddgee Z e ddZ e ddZ d S) TestSVMLGenerationz; Tests all SVML-generating functions produce desired calls F|z \$[a-z_]\w+,csl|dr|dkrdS|do&dk}||ft| dfdd}tdt|dS) NrXr9r:rzNot implementedc st\}}}tdtNtdd8yt|tjfd}Wntd|jYnXWdQRXWdQRX|j fdd|D}fdd|D}| | o| d t |t |d fd d d D|jfdS) NNUMBA_CPU_NAMENUMBA_CPU_FEATURESr@)rQzraised while compiling csg|]}|kr|qSr,r,)r)pattern)asmr,r-r3szJTestSVMLGeneration._inject_test..test_template..csg|]}|kr|qSr,r,)r)rk)rlr,r-r3sz?While expecting %s and no %s, it contains: %s when compiling %sr;cs$g|]}j|rd|kr|qS)") asm_filtersearch)r)line)clsr,r-r3s)rfrvlen2cpurrIZint64 Exceptionrblibrary get_asm_strZ assertTruer0rCr1)selffnrTrUZjitZmissedfound)rMrqrQrP)rlr- test_templates ( z6TestSVMLGeneration._inject_test..test_templateZtest_)rDunittest skipUnlesssetattrrW)rqrNrOrPrQZskippedryr,)rMrqrQrPr- _inject_tests   zTestSVMLGeneration._inject_testc sdgtfddttdDgg}xZD]R}t}|d|ddd|d |_x|D]}||qjW||q4WxFd D]>}x8tD]0}x*|D]"}xd D]}| ||||qWqWqWqWt j d d kr x&dD]}t ||t dt||qWdS)Nr?csg|]}tt|qSr,)listr)r)rY) test_flagsr,r-r3sz3TestSVMLGeneration.autogenerate..r nrtZ error_modelr9_)Zusecase)r>float64r=rB)r6r<rr9rr)Ztest_int32_range4_usecase important)sumr<r(r r[rC__name__appendrrr}sys version_infor|r rH) rqZ flag_listZftrQr4rNrPrOnr,)rr- autogenerates(          zTestSVMLGeneration.autogenerateN)r __module__ __qualname__rb_numba_parallel_test_recompilerCr~ svml_funcsrn classmethodr}rr,r,r,r-rgs  !rgcCs t|S)N)mathr)rYr,r,r-math_sin_scalarsrcCs:tj|tjd}x$t|D]}tt|||<qW|S)N)rN)npemptyrr<rr)rrZrYr,r,r- math_sin_loopsrcsdeZdZdZdZfddZddZddZd d Zd d Z d dZ e dddZ ddZ ZS)TestSVMLz Tests SVML behaves as expected FcsHt|_|jdt|_|jd|jdtt|j|dS)Nrr?)r rQr[ fastflagssuperr__init__)rvrM) __class__r,r-rs    zTestSVML.__init__cOsB|rttdd|D}t|||jd}t|||jd}||fS)NcSsg|]}t|qSr,)rIZtypeof)r)rYr,r,r-r3sz$TestSVML.compile..)rQ)rEtuplerrQr)rvrLrMkwargsZsigstdfastr,r,r-rs zTestSVML.compilecGs~|s tSg}xf|D]^}t|tjr6||dqt|tjrR||qt|tjrj||qt dqWt|S)Nr*z%Unsupported argument type encountered) r isinstancerZndarrayrcopyZnumbernumbersNumber ValueError)rvrMZnew_argsrYr,r,r- copy_argss      zTestSVML.copy_argsc Os|j|f|\}}|dd}|dd}|dd}||j|} |j|j|} |j|j|} tjj| | f|tjj| | f|td|Ptdd:|j|f|\}}|r||||r|||WdQRXWdQRXdS)N std_pattern fast_patterncpu_namezskylake-avx512rirjr@) rpoprZ entry_pointrZtestingZassert_almost_equalrcheck_svml_presence) rvZpyfuncrMrZjitstdZjitfastrrrZ py_expectedZ jitstd_resultZjitfast_resultr,r,r-check*s       zTestSVML.checkcCs|j}|||dS)N)rtruZassertIn)rvrLrkrlr,r,r-rIs zTestSVML.check_svml_presencecCs4tjjr dnd}|jtd|d|jtd|ddS)Nz$_sinz$sing@)r)r)rIrJrKrr)rvZpatr,r,r-test_scalar_contextMszTestSVML.test_scalar_contextrcCsd}d}|jtd||ddS)Nz__svml_sin8_ha,z __svml_sin8, )rr)rr)rvrrr,r,r- test_svmlSszTestSVML.test_svmlcCsPd}tjtjd|gtjtjd}|\}}|jdkrLtd|j|fdS)Naif 1: import os import numpy as np import math def math_sin_loop(n): ret = np.empty(n, dtype=np.float64) for x in range(n): ret[x] = math.sin(np.float64(x)) return ret def check_no_svml(): try: # ban the use of SVML os.environ['NUMBA_DISABLE_INTEL_SVML'] = '1' # delay numba imports to account for env change as # numba.__init__ picks up SVML and it is too late by # then to override using `numba.config` import numba from numba import config from numba.tests.support import override_env_config from numba.compiler import compile_isolated, Flags # compile for overridden CPU, with and without fastmath with override_env_config('NUMBA_CPU_NAME', 'skylake-avx512'), override_env_config('NUMBA_CPU_FEATURES', ''): sig = (numba.int32,) f = Flags() f.set('nrt') std = compile_isolated(math_sin_loop, sig, flags=f) f.set('fastmath') fast = compile_isolated(math_sin_loop, sig, flags=f) fns = std, fast # assert no SVML call is present in the asm for fn in fns: asm = fn.library.get_asm_str() assert '__svml_sin' not in asm finally: # not really needed as process is separate os.environ['NUMBA_DISABLE_INTEL_SVML'] = '0' config.reload_config() check_no_svml() z-c)stdoutstderrrz/process failed with code %s: stderr follows %s ) subprocessPopenr executablePIPEZ communicate returncoderEdecode)rvcodepopenouterrr,r,r-test_svml_disabled[s-   zTestSVML.test_svml_disabled)rrrrbrrrrrrrr rr __classcell__r,r,)rr-rs  r__main__)r5)CZ __future__rrrr9rrr importlibrr itertoolsrrrIrrrzZnumba.compilerrr Z numba.sixr Zsupportr r rr{rJZ USING_SVMLZ needs_svmlrrrrrrrrrrrrrrrr r!r"r$r%r&r'rr_itemsr\r]rVrWrfrgrrrrrmainr,r,r,r-s               4 E