\ K]c @`sdZddlmZmZmZddlZddlZddlZddlm Z m Z m Z m Z m Z mZmZddlZddlmZddlmZddlmZmZdd lmZmZmZejeZdZdZ d Z!d Z"d Z#e Z$e Z%d j&Z'x1e(e'D]#\Z)Z*e+ej,ee*e)q*WdZ-ej.Z/de0fdYZ1de0fdYZ2idd6dd6Z3e3e4j5dZ6ye1j7Z8Wn dGZ8nXe8dHkr dIdJdKdLdMdNdOdPdQdRf Z9n!dSdTdUdVdWdXdYdZd[f Z9dZ:dZ;dZ<de0fdYZ=d Z>d!Z?d"Z@d#ZAd$ZBd%ZCd&ZDejEd'ZFejEd(ZGejEd)ZHejEd*jId+d,ZJejEd-ZKd.d/d0d1d2d3d4d5d6d7d8d9h ZLejEd:ZMejEd;ZNejEd<ZOejEd=ZPejEd>ZQejEd?ZRejEd@ZSejEdAZTdBZUdCZVdDZWdEZXdFZYdS(\s( This is a direct translation of nvvm.h i(tprint_functiontabsolute_importtdivisionN(tc_void_ptc_inttPOINTERtc_char_ptc_size_ttbyreftc_char(tir(tconfigi(t NvvmErrortNvvmSupportError(t get_libdevicetopen_libdevicet open_cudalibiiis NVVM_SUCCESS NVVM_ERROR_OUT_OF_MEMORY NVVM_ERROR_PROGRAM_CREATION_FAILURE NVVM_ERROR_IR_VERSION_MISMATCH NVVM_ERROR_INVALID_INPUT NVVM_ERROR_INVALID_PROGRAM NVVM_ERROR_INVALID_IR NVVM_ERROR_INVALID_OPTION NVVM_ERROR_NO_MODULE_IN_PROGRAM NVVM_ERROR_COMPILATION cC`s(y tWntk rtSXtSdS(s( Return if libNVVM is available N(tNVVMR tFalsetTrue(((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyt is_available1s   RcB`seZdZi eeeeefd6eeefd6eeefd6eeeeefd6eeeeefd6eeeefd6eeefd6eeeefd6eeefd 6Z d Z d Z d Z ed ZRS(sProcess-wide singleton. t nvvmVersiontnvvmCreateProgramtnvvmDestroyProgramtnvvmAddModuleToProgramtnvvmCompileProgramtnvvmGetCompiledResultSizetnvvmGetCompiledResulttnvvmGetProgramLogSizetnvvmGetProgramLogc C`st|jdkrtj||_}ytddt|_Wn2tk ry}d|_d}t ||nXx\|j j D]H\}}t |j|}|d|_ |d|_t|||qWnWdQX|jS(Ntnvvmtcccs;libNVVM cannot be found. Do `conda install cudatoolkit`: %sii(t _nvvm_lockt_NVVM__INSTANCEtNonetobjectt__new__RRtdrivertOSErrorR t _PROTOTYPEStitemstgetattrtrestypetargtypestsetattr(tclstinstteterrmsgtnametprototfunc((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyR$js   cC`sPt}t}|jt|t|}|j|d|j|jfS(NsFailed to get version.(RRRt check_errortvalue(tselftmajortminorterr((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyt get_versions   cC`sF|rBt|t|}|r9t|tjdqB|ndS(Ni(R tRESULT_CODE_NAMEStprinttsystexit(R6terrortmsgR>texc((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyR4s  N(t__name__t __module__t__doc__t nvvm_resultRRt nvvm_programRRR'R"R!R$R:RR4(((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyR?s  tCompilationUnitcB`s>eZdZdZdZdZdZdZRS(cC`sJt|_t|_|jjt|j}|jj|ddS(NsFailed to create CU(RR%RFt_handleRRR4(R6R9((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyt__init__s  cC`s;t}|jt|j}|j|ddtdS(NsFailed to destroy CUR>(RRRRHR4R(R6R%R9((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyt__del__s cC`s;|jj|j|t|d}|jj|ddS(s Add a module level NVVM IR to a compilation unit. - The buffer should contain an NVVM module IR either in the bitcode representation (LLVM3.0) or in the text representation. sFailed to add moduleN(R%RRHtlenR"R4(R6tbufferR9((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyt add_modulesc K`sg}d|kr4|jdr4|jdq4n|jdr`|jd|jdn|jdr|jd|jdnd}x[|D]S}||krtt|j|}|jd |jd d |fqqW|r,djtt|j }t dj |nt t |g|D]}t |jd^q@}|jj|jt ||} |j| dt} |jj|jt| } |j| dt| j} |jj|j| } |j| d|j|_| S(sFPerform Compliation The valid compiler options are * - -g (enable generation of debugging information) * - -opt= * - 0 (disable optimizations) * - 3 (default, enable optimizations) * - -arch= * - compute_20 (default) * - compute_30 * - compute_35 * - -ftz= * - 0 (default, preserve denormal values, when performing * single-precision floating-point operations) * - 1 (flush denormal values to zero, when performing * single-precision floating-point operations) * - -prec-sqrt= * - 0 (use a faster approximation for single-precision * floating-point square root) * - 1 (default, use IEEE round-to-nearest mode for * single-precision floating-point square root) * - -prec-div= * - 0 (use a faster approximation for single-precision * floating-point division and reciprocals) * - 1 (default, use IEEE round-to-nearest mode for * single-precision floating-point division and reciprocals) * - -fma= * - 0 (disable FMA contraction) * - 1 (default, enable FMA contraction) * tdebugs-gtopts-opt=%dtarchs-arch=%stftzt prec_sqrttprec_divtfmas-%s=%dt_t-s, sunsupported option {0}tutf8sFailed to compile s&Failed to get size of compiled result.sFailed to get compiled result.(RQRRRSRT(tpoptappendtgettinttbooltreplacetjointmaptreprtkeysR tformatRRKtencodeR%RRHt _try_errorRRRR R5Rtget_logtlog( R6toptionstoptst other_optionstktvtoptstrtxtc_optsR9treslentptxbuf((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pytcompiles>#   *(! cC`s'|jj|d||jfdS(Ns%s %s(R%R4Re(R6R9R@((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyRdscC`st}|jj|jt|}|jj|d|jdkrt|j}|jj|j|}|jj|d|jj dSdS(Ns#Failed to get compilation log size.isFailed to get compilation log.RWt( RR%RRHRR4R5R Rtdecode(R6RoR9tlogbuf((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyRes (RBRCRIRJRMRqRdRe(((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyRGs    T se-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64i se-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64i@iiiicC`ssxhttD]Z\}}||kr)|S||kr |dkrXtd||qgt|dSq q WtdS(Nis@GPU compute capability %d.%d is not supported (requires >=%d.%d)ii(t enumeratet SUPPORTED_CCR (tmycctitcc((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyt _find_arch$s   cC`s/tjrtj}nt||f}d|S(s1Matches with the closest architecture option s compute_%d%d(R t FORCE_CUDA_CCRz(R7R8RP((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pytget_arch_option7s  sMissing libdevice file for {arch}. Please ensure you have package cudatoolkit >= 8. Install package by: conda install cudatoolkit t LibDevicecB`s;eZiZddddgZdZdZdZRS(t compute_20t compute_30t compute_35t compute_50cC`s~||jkra|j|}t|dkrKttjd|nt||j| val or dold is NaN (using ugt semantics) %lt = fcmp ugt {T} %dold, %val br i1 %lt, label %attempt, label %done attempt: ; Attempt to swap in the smaller value %iold = bitcast {T} %dold to {Ti} %iptr = bitcast {T}* %ptr to {Ti}* %ival = bitcast {T} %val to {Ti} %cas = cmpxchg volatile {Ti}* %iptr, {Ti} %iold, {Ti} %ival monotonic %dcas = bitcast {Ti} %cas to {T} br label %gt_check done: ; Return min %ret = phi {T} [ %ptrval, %entry ], [ %dold, %gt_check ] ret {T} %ret }} cC`sb|j}xFt|D]8\}}|jdrd}|jt||YRrs attributes #cs`s!|]}|tkr|VqdS(N(tsupported_attributes(t.0ta((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pys _ssgetelementptr s failed parsing getelementptr: %ssload scall s\1*\2RrRs @llvm.memsettdeclares N(RRR]Rtre_metadata_deftmatchR"tre_metadata_correct_usageRRR^tre_metadata_reftsubtre_unsupported_keywordstre_attributes_defRtsplittre_getelementptrRRtre_loadtre_calltre_rangetrstriptre_parenthesized_listt_replace_llvm_memset_usaget _replace_llvm_memset_declarationtre_annotationsRY( R RtbuftlinetassigpostlhstrhsRRtattrsR((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyR"sf   '         cC`srt|jdjd}tjd|djd}|jddj|dj|}dj|S( sNReplace `llvm.memset` usage for llvm7+. Used as functor for `re.sub. iRs align (\d+)iisi32 {}s, s({})(tlistRRtreRtinsertRbR^(Rtparamstaligntout((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyRs cC`sJt|jdjd}|jdddj|}dj|S(sTReplace `llvm.memset` declaration for llvm7+. Used as functor for `re.sub. iRiRs, s({})(RRRRR^Rb(RRR((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyRsc C`sddlm}m}m}m}|j}||j|d|j|jdf}|j||}|jd}|j |t j d} |j | d| d| d| dg} |j d| dS( Ni(tMetaDatatMetaDataStringtConstanttTypetkernelisnvvm.annotationsi isnvvmir.version(tllvmlite.llvmpy.coreRRRRtmoduleRZR[tget_or_insert_named_metadatataddR tIntTypet add_metadatatadd_named_metadata( tlfuncRRRRRtopstmdtnmdRtmd_ver((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pytset_cuda_kernels" - 3cC`s t|_dS(N(Rt data_layout(R((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pytfix_data_layouts(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ii(ZRDt __future__RRRR=tloggingRtctypesRRRRRRR t threadingtllvmliteR tnumbaR R?R R tlibsRRRt getLoggerRBtloggertADDRSPACE_GENERICtADDRSPACE_GLOBALtADDRSPACE_SHAREDtADDRSPACE_CONSTANTtADDRSPACE_LOCALRFRERR;RuRxRjR,tmodulesRtLockR R#RRGRttuplet __itemsize__RR:t NVVM_VERSIONRvRzR|RR}RRRRRRRRqRRRR]tre_metadata_debuginfoRRRRRRRRRRRRRRR(((s6lib/python2.7/site-packages/numba/cuda/cudadrv/nvvm.pyts$4   Q|   '!  & "    j