U hdQ@sddlZddlZddlZddlZddlZddlmZmZm Z m Z m Z m Z ddl mZmZddlmZddlmZddlmZddlmZmZddlmZdd lmZdd lmZmZdd l m!Z!dd l"m#Z#dd l$m%Z%ddl&m'Z'm(Z(ddl)m Z*ddl+m,Z,ddl+m-Z-ddl.m/Z/dddddddddddddd d!d"gZ0Gd#d$d$ej1Z2Gd%d&d&e3Z4Gd'd(d(Z5Gd)d*d*eZ6Gd+d,d,eZ7Gd-d.d.eej1Z8dS)/N)config serializesigutilstypestypingutils)Cache CacheImpl)global_compiler_lock) Dispatcher)NumbaPerformanceWarning)Purposetypeof)get_current_device)wrap_arg) compile_cuda CUDACompiler)driver) get_context) cuda_target)missing_launch_config_msgnormalize_kernel_dimensions)rcuda) _dispatcher)warnZhsinZhcosZhlogZhlog10Zhlog2ZhexpZhexp10Zhexp2ZhsqrtZhrsqrtZhfloorZhceilZhrcpZhrintZhtruncZhdivc seZdZdZed0fdd ZeddZed d Zd d Z ed dZ eddZ e fddZ ddZddZeddZeddZeddZeddZedd Zd!d"Zd#d$Zd%d&Zd1d'd(Zd2d*d+Zd3d,d-Zd.d/ZZS)4_Kernelz CUDA Kernel specialized for a given set of argument types. When called, this object launches the kernel on the device. NFTc  s| r tdtd|_d|_||_||_||_||_|p@g|_ || rNdndd} t j } t |jt j|j|j|||| | d }|j}|jj}|j}|j}||j|j||| ||| \}|sg}dk|_|jrd_fd d tD}|r>tjsd }d |}t|tjtj t!}tj"|d }|#||D]}$|qB|j%|_&|j'|_'|j(|_)|_*|j+|_+||_|j|_|j,|_,g|_-g|_.g|_/dS)Nz,Cannot compile a device function as a kernelFr)fastmathoptdebuglineinfoinliner nvvm_optionsccZcudaCGGetIntrinsicHandleTcs"g|]}d|kr|qS)Z__numba_wrapper_) get_asm_str).0fnlibN/home/sam/Atlas/atlas_env/lib/python3.8/site-packages/numba/cuda/dispatcher.py lsz$_Kernel.__init__..z9https://numba.readthedocs.io/en/stable/cuda/bindings.htmlzUse of float16 requires the use of the NVIDIA CUDA bindings and setting the NUMBA_CUDA_USE_NVIDIA_BINDING environment variable to 1. Relevant documentation is available here: zcpp_function_wrappers.cu)0 RuntimeErrorsuper__init__Z objectmode entry_pointpy_funcargtypesr!r" extensionsrcompute_capabilityrrvoidtarget_context__code__ co_filenameco_firstlinenoZprepare_cuda_kernellibraryfndescr& cooperativeZneeds_cudadevrtcuda_fp16_math_funcsrZCUDA_USE_NVIDIA_BINDINGNotImplementedErrorospathdirnameabspath__file__joinappendZadd_linking_filename entry_name signaturetype_annotation_type_annotation _codelibrary call_helper environment_referenced_environmentsZliftedZ reload_init)selfr2r3linkr!r"r#rr4Z max_registersrdevicer$r%cresZtgt_ctxcodefilenameZlinenumkernelressmsgZbasedirZfunctions_cu_pathfilepath __class__r)r,r0.s~      z_Kernel.__init__cCs|jSN)rLrPr+r+r,r;sz_Kernel.librarycCs|jSr])rKr^r+r+r,rJsz_Kernel.type_annotationcCs|jSr])rOr^r+r+r,_find_referenced_environmentssz%_Kernel._find_referenced_environmentscCs |jSr])r7codegenr^r+r+r,r`sz_Kernel.codegencCs t|jjSr])tuplerIargsr^r+r+r,argument_typessz_Kernel.argument_typesc sX||} t|| d| _|| _|| _|| _d| _|| _|| _ || _ || _ || _ | S)& Rebuild an instance. N) __new__r/r0r1r=rHrIrKrLr!r"rMr4) clsr=rGrI codelibraryr!r"rMr4instancer[r+r,_rebuilds z_Kernel._rebuildc Cs(t|j|j|j|j|j|j|j|jdS)a Reduce the instance for serialization. Compiled definitions are serialized in PTX form. Type annotation are discarded. Thread, block and shared memory configuration are serialized. Stream information is discarded. )r=rGrIrgr!r"rMr4) dictr=rHrIrLr!r"rMr4r^r+r+r,_reduce_statess z_Kernel._reduce_statescCs|jdS)z7 Force binding to current CUDA context N)rL get_cufuncr^r+r+r,bindsz _Kernel.bindcCs|jjjS)zN The number of registers used by each thread for this kernel. )rLrlattrsregsr^r+r+r,regs_per_threadsz_Kernel.regs_per_threadcCs|jjjS)zD The amount of constant memory used by this kernel. )rLrlrnconstr^r+r+r,const_mem_sizesz_Kernel.const_mem_sizecCs|jjjS)zM The amount of shared memory used per block for this kernel. )rLrlrnZsharedr^r+r+r,shared_mem_per_blocksz_Kernel.shared_mem_per_blockcCs|jjjS)z: The maximum allowable threads per block. )rLrlrnZ maxthreadsr^r+r+r,max_threads_per_blocksz_Kernel.max_threads_per_blockcCs|jjjS)zM The amount of local memory used per thread for this kernel. )rLrlrnlocalr^r+r+r,local_mem_per_threadsz_Kernel.local_mem_per_threadcCs |jS)z6 Returns the LLVM IR for this kernel. )rL get_llvm_strr^r+r+r, inspect_llvmsz_Kernel.inspect_llvmcCs|jj|dS)z7 Returns the PTX code for this kernel. r%)rLr&)rPr%r+r+r, inspect_asmsz_Kernel.inspect_asmcCs |jS)zp Returns the SASS code for this kernel. Requires nvdisasm to be available on the PATH. )rLZget_sassr^r+r+r, inspect_sasssz_Kernel.inspect_sasscCsb|jdkrtd|dkr tj}td|j|jf|dtd|dt|j|dtd|ddS) Produce a dump of the Python source of this function annotated with the corresponding Numba IR and type information. The dump is written to *file*, or *sys.stdout* if *file* is *None*. Nz Type annotation is not availablez%s %sfilezP--------------------------------------------------------------------------------zP================================================================================)rK ValueErrorsysstdoutprintrHrc)rPr~r+r+r, inspect_typess  z_Kernel.inspect_typesrcCsHt}|j}t|tr*tdd|}||||}|jj }||S)a Calculates the maximum number of blocks that can be launched for this kernel in a cooperative grid in the current context, for the given block and dynamic shared memory sizes. :param blockdim: Block dimensions, either as a scalar for a 1D block, or a tuple for 2D or 3D blocks. :param dynsmemsize: Dynamic shared memory size in bytes. :return: The maximum number of blocks in the grid. cSs||Sr]r+)xyr+r+r,(z5_Kernel.max_cooperative_grid_blocks..) rrLrl isinstancera functoolsreduceZ$get_active_blocks_per_multiprocessorrRZMULTIPROCESSOR_COUNT)rPblockdimZ dynsmemsizectxcufuncZ active_per_smZsm_countr+r+r,max_cooperative_grid_blockss   z#_Kernel.max_cooperative_grid_blockscs|j|jrTjd}j|\}}|ttjks>t t} |j d|dg} g} t |j |D]\} } | | | || | qhtjrtjd}nd}|r|jp|}tjjf||||| fd|ji|jrtt| ||| jdkrfddfddd D}fd dd D}| j}|j|\}}}|dkrNd }n$|\}}}tj|}d |||f}d |||f}|rd||dff|dd}n|f}||| D] }|qdS)NZ __errcode__r)streamr=cs<jdj|f\}}t}tt||||jS)Nz%s__%s__) moduleget_global_symbolrGctypesc_intrdevice_to_host addressofvalue)rGZmemszval)rr+r, load_symbolUs z#_Kernel.launch..load_symbolcsg|]}d|qS)tidr+r'irr+r,r-]sz"_Kernel.launch..Zzyxcsg|]}d|qS)ctaidr+rrr+r,r-^sz"In function %r, file %s, line %s, z%stid=%s ctaid=%sz%s: %s)rLrlr!rGrrrZsizeofrAssertionErrorZmemsetziprc _prepare_argsrUSE_NV_BINDINGZbindingZCUstreamhandleZ launch_kernelr=rrrrMZ get_exceptionr@rArC)rPrbgriddimrr sharedmemexcnameZexcmemZexcszZexcvalretr kernelargstvZ zero_streamZ stream_handlerrrTZexcclsZexc_argslocZlocinfosymrZlinenoprefixwbr+)rrr,launch/sb         z_Kernel.launchcCsht|jD]}|j||||d\}}q t|tjrt|||}tj }t d} t d} ||j } ||j j } t|} tjrt| } t | }|| || || || ||t|jD]}|||j|qt|jD]}|||j|qnPt|tjrBttd||}||n"|tjkrttt|tj}||n|tjkrt|}||n|tj krt!|}||n|tj"krt#t|}||n|tj$kr|t!|j%|t!|j&nL|tj'krL|t|j%|t|j&nt|tj(tj)frz|t*|tj+nt|tj,rt|||}|j-} tjrt t| } || nt|tj.rt/|t/|kst0t1||D]\}}|2|||||qnVt|tj3rZz|2|j |j4|||Wn t5k rVt5||YnXn t5||dS)zF Convert arguments to ctypes and append to kernelargs )rrrzc_%sN)6reversedr4Z prepare_argsrrZArrayrZ to_devicerZ c_ssize_tZc_void_psizeZdtypeitemsizerZdevice_pointerrintrFrangendimshapestridesIntegergetattrZfloat16Zc_uint16npviewZuint16Zfloat64Zc_doubleZfloat32Zc_floatbooleanZc_uint8Z complex64realimagZ complex128Z NPDatetimeZ NPTimedeltaZc_int64Zint64ZRecordZdevice_ctypes_pointerZ BaseTuplelenrrrZ EnumMemberrr?)rPtyrrrr extensionZdevaryZc_intpZmeminfoparentZnitemsrZptrdataZaxcvalZdevrecrrr+r+r,rws                     z_Kernel._prepare_args) NFFFFNNTF)N)r)rr)__name__ __module__ __qualname____doc__r r0propertyr;rJr_r`rc classmethodrirkrmrprrrsrtrvrxrzr{rrrr __classcell__r+r+r[r,r(sRf             Hrc@s$eZdZddZddZddZdS)ForAllcCs6|dkrtd|||_||_||_||_||_dS)Nrz0Can't create ForAll with negative task count: %s)r dispatcherntasksthread_per_blockrr)rPrrtpbrrr+r+r,r0szForAll.__init__cGs^|jdkrdS|jjr|j}n |jj|}||}|j|d|}||||j|jf|S)Nrr)rr specialized specialize_compute_thread_per_blockrr)rPrbrrrr+r+r,__call__s    zForAll.__call__cCsZ|j}|dkr|St}tt|j}t|jd|j dd}|j f|\}}|SdS)Nri)funcZb2d_funcZmemsizeZblocksizelimit) rrnextiter overloadsvaluesrjrLrlrZget_max_potential_block_size)rPrrrrVkwargs_r+r+r,rsz ForAll._compute_thread_per_blockN)rrrr0rrr+r+r+r,rs rc@seZdZddZddZdS)_LaunchConfigurationc Csd||_||_||_||_||_tjr`d}|d|d|d}||kr`d|d}tt|dS)Nrrz Grid size zB will likely result in GPU under-utilization due to low occupancy.) rrrrrrZCUDA_LOW_OCCUPANCY_WARNINGSrr ) rPrrrrrZ min_grid_sizeZ grid_sizerYr+r+r,r0s  z_LaunchConfiguration.__init__cGs|j||j|j|j|jSr])rcallrrrrrPrbr+r+r,rsz_LaunchConfiguration.__call__N)rrrr0rr+r+r+r,rsrc@s$eZdZddZddZddZdS) CUDACacheImplcCs|Sr])rk)rPrVr+r+r,r"szCUDACacheImpl.reducecCs tjf|Sr])rri)rPr7payloadr+r+r,rebuild%szCUDACacheImpl.rebuildcCsdS)NTr+)rPrSr+r+r,check_cachable(szCUDACacheImpl.check_cachableN)rrrrrrr+r+r+r,r!src@seZdZdZeZdS) CUDACachezS Implements a cache that saves and loads CUDA kernels and compile results. N)rrrrrZ _impl_classr+r+r+r,r3srcs8eZdZdZdZeZeffdd Ze ddZ ddZ e j d d d?d d ZddZd@ddZe ddZddZddZddZddZddZe ddZdAd!d"ZdBd#d$ZdCd%d&ZdDd'd(ZdEd)d*Zd+d,ZdFd-d.Zd/d0Zd1d2Z dGd3d4Z!dHd5d6Z"dId7d8Z#dJd9d:Z$e%d;d<Z&d=d>Z'Z(S)KCUDADispatchera CUDA Dispatcher object. When configured and called, the dispatcher will specialize itself for the given arguments (if no suitable specialized version already exists) & compute capability, and launch on the device associated with the current context. Dispatcher objects are not to be constructed by the user, but instead are created using the :func:`numba.cuda.jit` decorator. Fcs"tj|||dd|_i|_dS)N) targetoptionspipeline_classF)r/r0 _specializedspecializations)rPr2rrr[r+r,r0Ls   zCUDADispatcher.__init__cCs t|Sr]) cuda_typesrr^r+r+r, _numba_type_\szCUDADispatcher._numba_type_cCst|j|_dSr])rr2_cacher^r+r+r,enable_caching`szCUDADispatcher.enable_cachingr)maxsizercCst||\}}t|||||Sr])rr)rPrrrrr+r+r, configurecszCUDADispatcher.configurecCst|dkrtd|j|S)N)rrz.must specify at least the griddim and blockdim)rrrrr+r+r, __getitem__hs zCUDADispatcher.__getitem__cCst|||||dS)a3Returns a 1D-configured dispatcher for a given number of tasks. This assumes that: - the kernel maps the Global Thread ID ``cuda.grid(1)`` to tasks on a 1-1 basis. - the kernel checks that the Global Thread ID is upper-bounded by ``ntasks``, and does nothing if it is not. :param ntasks: The number of tasks. :param tpb: The size of a block. An appropriate value is chosen if this parameter is not supplied. :param stream: The stream on which the configured dispatcher will be launched. :param sharedmem: The number of bytes of dynamic shared memory required by the kernel. :return: A configured dispatcher, ready to launch on a set of arguments.)rrr)r)rPrrrrr+r+r,forallmszCUDADispatcher.forallcCs |jdS)aS A list of objects that must have a `prepare_args` function. When a specialized kernel is called, each argument will be passed through to the `prepare_args` (from the last object in this list to the first). The arguments to `prepare_args` are: - `ty` the numba type of the argument - `val` the argument value itself - `stream` the CUDA stream used for the current call to the kernel - `retr` a list of zero-arg functions that you may want to append post-call cleanup work to. The `prepare_args` function must return a tuple `(ty, val)`, which will be passed in turn to the next right-most `extension`. After all the extensions have been called, the resulting `(ty, val)` will be passed into Numba's default argument marshalling logic. r4)rgetr^r+r+r,r4szCUDADispatcher.extensionscOs ttdSr])rr)rPrbrr+r+r,rszCUDADispatcher.__call__cCsB|jrtt|j}ntjj|f|}||||||dS)zJ Compile if necessary and invoke this kernel with *args*. N) rrrrrrr Z _cuda_callr)rPrbrrrrrVr+r+r,rszCUDADispatcher.callcs(|rtfdd|D}t|S)Ncsg|]}|qSr+) typeof_pyvalr'ar^r+r,r-sz4CUDADispatcher._compile_for_args..)rcompilera)rPrbkwsr3r+r^r,_compile_for_argssz CUDADispatcher._compile_for_argscCsNzt|tjWStk rHt|rBttj|ddtjYSYnXdS)NF)sync)rr argumentrrZ is_cuda_arrayZ as_cuda_array)rPrr+r+r,rs zCUDADispatcher.typeof_pyvalcstj}tfdd|D}jr,tdj||f}|rD|Sj}tj |d}| || d|_ |j||f<|S)zd Create a new instance of this dispatcher specialized for the given *args*. csg|]}j|qSr+)Z typingctxZresolve_argument_typerr^r+r,r-sz-CUDADispatcher.specialize..zDispatcher already specialized)rT) rr5rarr.rrrrr2rZdisable_compiler)rPrbr%r3Zspecializationrr+r^r,rs$ zCUDADispatcher.specializecCs|jS)z> True if the Dispatcher has been specialized. )rr^r+r+r,rszCUDADispatcher.specializedNcCsH|dk r|j|jjS|jr0tt|jjSdd|jDSdS)a Returns the number of registers used by each thread in this kernel for the device in the current context. :param signature: The signature of the compiled kernel to get register usage for. This may be omitted for a specialized kernel. :return: The number of registers used by the compiled variant of the kernel for the given signature and current device. NcSsi|]\}}||jqSr+)rpr'sigoverloadr+r+r, sz6CUDADispatcher.get_regs_per_thread..)rrbrprrrritemsrPrIr+r+r,get_regs_per_threads z"CUDADispatcher.get_regs_per_threadcCsH|dk r|j|jjS|jr0tt|jjSdd|jDSdS)a Returns the size in bytes of constant memory used by this kernel for the device in the current context. :param signature: The signature of the compiled kernel to get constant memory usage for. This may be omitted for a specialized kernel. :return: The size in bytes of constant memory allocated by the compiled variant of the kernel for the given signature and current device. NcSsi|]\}}||jqSr+)rrrr+r+r,rsz5CUDADispatcher.get_const_mem_size..)rrbrrrrrrrrr+r+r,get_const_mem_sizes z!CUDADispatcher.get_const_mem_sizecCsH|dk r|j|jjS|jr0tt|jjSdd|jDSdS)a Returns the size in bytes of statically allocated shared memory for this kernel. :param signature: The signature of the compiled kernel to get shared memory usage for. This may be omitted for a specialized kernel. :return: The amount of shared memory allocated by the compiled variant of the kernel for the given signature and current device. NcSsi|]\}}||jqSr+)rsrr+r+r,rsz;CUDADispatcher.get_shared_mem_per_block..)rrbrsrrrrrrr+r+r,get_shared_mem_per_blocks z'CUDADispatcher.get_shared_mem_per_blockcCsH|dk r|j|jjS|jr0tt|jjSdd|jDSdS)a( Returns the maximum allowable number of threads per block for this kernel. Exceeding this threshold will result in the kernel failing to launch. :param signature: The signature of the compiled kernel to get the max threads per block for. This may be omitted for a specialized kernel. :return: The maximum allowable threads per block for the compiled variant of the kernel for the given signature and current device. NcSsi|]\}}||jqSr+)rtrr+r+r,r&sz.)rrbrtrrrrrrr+r+r,get_max_threads_per_blocks z(CUDADispatcher.get_max_threads_per_blockcCsH|dk r|j|jjS|jr0tt|jjSdd|jDSdS)a Returns the size in bytes of local memory per thread for this kernel. :param signature: The signature of the compiled kernel to get local memory usage for. This may be omitted for a specialized kernel. :return: The amount of local memory allocated by the compiled variant of the kernel for the given signature and current device. NcSsi|]\}}||jqSr+)rvrr+r+r,r9sz;CUDADispatcher.get_local_mem_per_thread..)rrbrvrrrrrrr+r+r,get_local_mem_per_thread)s z'CUDADispatcher.get_local_mem_per_threadcCsP|jr|t||jj}d|}tj|||jd}t |j}||||fS)z Get a typing.ConcreteTemplate for this dispatcher and the given *args* and *kws* types. This allows resolution of the return type. A (template, pysig, args, kws) tuple is returned. zCallTemplate({0}))keyZ signatures) _can_compilecompile_devicerar2rformatrZmake_concrete_templateZnopython_signaturesrZ pysignature)rPrbr func_namerGZ call_templateZpysigr+r+r,get_call_template<s  z CUDADispatcher.get_call_templatec Cs||jkr|j|jd}|jd}|jd}|jd}|jdrRdnd|d}tj}t|j||||||||d } | |j|<| j | j | j | j gW5QRXn |j|} | S) zCompile the device function for the given argument types. Each signature is compiled once by caching the compiled function inside this object. Returns the `CompileResult`. r!r"r#rrrr)rrr ) rZ_compiling_counterrrrr5rr2r7Zinsert_user_functionr1r<r;) rPrb return_typer!r"r#rr$r%rSr+r+r,r Ws2         zCUDADispatcher.compile_devicecCs,dd|D}|j||dd||j|<dS)NcSsg|] }|jqSr+)_coderr+r+r,r-sz/CUDADispatcher.add_overload..Tr)Z_insertr)rPrVr3Zc_sigr+r+r, add_overload~szCUDADispatcher.add_overloadcCst|\}}|dks$|tjks$t|jr.cSsi|]\}}||qSr+)rxrr+r+r,rs)rrrr;rwrxrrPrIrRr+r+r,rxs zCUDADispatcher.inspect_llvmcs|tj|jd}|dk rD|r2|j|jS|j|Sn4|r`fdd|jDSfdd|jDSdS)a+ Return this kernel's PTX assembly code for for the device in the current context. :param signature: A tuple of argument types. :return: The PTX code for the given signature, or a dict of PTX codes for all previously-encountered signatures. rRNcsi|]\}}||jqSr+)r;r&rryr+r,rsz.CUDADispatcher.inspect_asm..csi|]\}}||qSr+)rzrryr+r,rs) rr5rrrr;r&rzrrr+ryr,rzs    zCUDADispatcher.inspect_asmcCsB|jdrtd|dk r*|j|Sdd|jDSdS)a Return this kernel's SASS assembly code for for the device in the current context. :param signature: A tuple of argument types. :return: The SASS code for the given signature, or a dict of SASS codes for all previously-encountered signatures. SASS for the device in the current context is returned. Requires nvdisasm to be available on the PATH. rRz(Cannot inspect SASS of a device functionNcSsi|]\}}||qSr+)r{)r'rdefnr+r+r,rsz/CUDADispatcher.inspect_sass..)rrr.rr{rrr+r+r,r{s zCUDADispatcher.inspect_sasscCs2|dkrtj}|jD]\}}|j|dqdS)r|Nr})rrrrr)rPr~rrr+r+r,rszCUDADispatcher.inspect_typescCs|||}|S)rdr+)rfr2rrhr+r+r,ris zCUDADispatcher._rebuildcCst|j|jdS)zd Reduce the instance for serialization. Compiled definitions are discarded. )r2r)rjr2rr^r+r+r,rkszCUDADispatcher._reduce_states)rr)rrr)N)N)N)N)N)N)N)N)N)N))rrrrZ _fold_argsrZ targetdescrrr0rrrr lru_cacherrrr4rrrrrrrrrr r rr rrrxrzr{rrrirkrr+r+r[r,r:sH              '$     r)9numpyrr@rrrZ numba.corerrrrrrZnumba.core.cachingrr Znumba.core.compiler_lockr Znumba.core.dispatcherr Znumba.core.errorsr Znumba.core.typing.typeofr rZnumba.cuda.apirZnumba.cuda.argsrZnumba.cuda.compilerrrZnumba.cuda.cudadrvrZnumba.cuda.cudadrv.devicesrZnumba.cuda.descriptorrZnumba.cuda.errorsrrZ numba.cudarZnumbarrwarningsrr>Z ReduceMixinrobjectrrrrrr+r+r+r,s\              1.