mVhOddlZddlZddlZddlZddlZddlmZddlmZddl m Z ddl m Z ddl mZejj!ejj#eZejj'edgZdZej,d Zd ZGd d eZd ZdZGddeZGddeZy)N)Path)_build)get_cache_manager) GPUTarget) GPUDriverincludec ddl}|jdk7ryddl ddlm}m}m}m m m}G fdd j} j|||||||} jdj}| g|_||_d j!dz} fd } ||| | r$t#j$ j'| Sy#t$rYywxYw) NrLinux)c_charc_intc_size_tc_void_pc_char_pPOINTERc"eZdZdWfdWfgZy)8_find_already_mmapped_dylib_on_linux..DlPhdrInfo dlpi_addr dlpi_nameN)__name__ __module__ __qualname___fields_)rrsJ/home/dcms/DCMS/lib/python3.12/site-packages/triton/backends/amd/driver.py DlPhdrInfors ( # ( # rz libc.so.6ic |jj}ttj|}|j vr'j ||tt|yy)Nrr) contentsrrosfsdecodenamememmoveminlen)infosizedatarpctypeslib_namemax_path_lengths rcallbackz6_find_already_mmapped_dylib_on_linux..callback2sRMM++ Y' ( qvv  NN4CY,P Qr)platformsystemr)r r r rrr Structure CFUNCTYPECDLLdl_iterate_phdr Exceptionargtypesrestypecreate_string_bufferrr string_at)r*r-r r r rr callback_tr2pathr,rrr)r+s` @@@@r$_find_already_mmapped_dylib_on_linuxr:sG# KK V%% !!%)OQXY_Q`aJ ++k2BB!+H5O#OO  & &': ;Dz(+T2{{6++D122 + s1C-- C98C9cd}tjd}|rC|j|r!tjj |r|St d|d|t |}|r2tjj |r|St d|d|g}ddl}|j}|j}|jr|g|z}|D]X}tjj|dd |}tjj |r|cS|j|Ztjd }|rj|jd D]V} tjj| |} tjj | r| cS|j| Xtjd d gj!} | j#D cgc]5} | j%j|s#| jd7} } | D]6}tjj |r|cS|j|8tjjd|}tjj |r|S|j|t d|d|cc} w)Nzlibamdhip64.soTRITON_LIBHIP_PATHzTRITON_LIBHIP_PATH 'z' does not point to a valid zmemory mapped 'z'' in process does not point to a valid rtorchlibLD_LIBRARY_PATH:z/sbin/ldconfigz-pz/opt/rocm/lib/zcannot locate z after attempted paths )rgetenvendswithr9exists RuntimeErrorr:sitegetsitepackagesgetusersitepackagesENABLE_USER_SITEjoinappendsplit subprocess check_outputdecode splitlinesstrip)r*env_libhip_path mmapped_pathpathsrF site_packages user_siter9env_ld_library_pathdflibslinelocsloccommon_install_paths r_get_path_to_hip_runtime_dylibr_@shHii 45O  # #H -"''..2Q" "1/1BB^_g^hijj8AL 77>>, ' _\N:abjaklmm E((*M((*I " m3 ww||D'5(; 77>>$ K T ))$56$**3/ A Q)Aww~~a  LLO    " "$4d#; < C C ED*.): ^djjl>S>ST\>]DJJL  ^D ^ 77>># J S '',,'7B ww~~)*"" LL$% z1HP QQ _s )$KKc tj|jdj}t |}|j |d}|t j5}tjj|d}t|d5}|j|dddt|||gtg}t|d5}|j|j!|dd}ddddddddl} | j$j'||} | j$j)| } | j*j-| | S#1swYxYw#1swYuxYw#1swYyxYw) Nzutf-8z.sozmain.cwrbT)binaryr)hashlibsha256encode hexdigestrget_filetempfileTemporaryDirectoryrr9rJopenwriter include_dirputreadimportlib.utilutilspec_from_file_locationmodule_from_specloader exec_module) srcr!keycache cache_pathtmpdirsrc_pathrYso importlibspecmods rcompile_module_from_srcrsK ..G, - 7 7 9C c "E4& -J  ( ( * Lfww||FH5Hh$   hKDBb$ L1"YYqvvxD6dYK  L  L >> 1 1$ CD .. ) )$ /CKKC J   L L  L Ls<(-E5E'(E5&E)5E5E& "E5)E2 .E55E>c$eZdZfdZdZxZS)HIPUtilscdt|dstt|||_|jS)Ninstance)hasattrsuperr__new__r)cls __class__s rrzHIPUtils.__new__s*sJ' 37%>"r)rrrrr __classcell__rs@rrrs  ?rrc>|ddk(ryddddddd d dd d d d d d d|S)Nr*hipDeviceptr_tint32_tint8_tint16_tint64_tuint32_tuint8_tuint16_tuint64_tfloatdouble)i1i8i16i32i64u1u8u16u32u64fp16bf16fp32f32fp64)tys r ty_to_cpprsQ !u|   !  rc fdfdfddj|jDcgc] }| c}}d|z}djt|j}tt t |j d}t|Dcic]\}}|| }}}t|dkDr)ddjd |jDznd}djd |jD} g} |jD]=\}}|dd k(r| jd |d $|dk7s*| jd|?t} ttt|} |jDcgc]\}}|dk7s d|} }}| jdd| dt| dkDrd| zndddj| d|d|ddj|jDcgc]\}}|d|dc}}d|d|ddj|jDcgc]\}}|dd k(rd|d|d|d |d! nd c}}d"t| dkDrddj| zndd#} | Scc}wcc}}wcc}}wcc}}wcc}}w)$Nc^t|trdjt|S|S)N,) isinstancetuplerJmap)sig_serialize_signatures rrz+make_launcher.._serialize_signatures) c5 !88C 4c:; ; rct|tr!djt|}d|dS|ddk(ry|dvryt |S)Nr[]rrz PyObject* constexprrrrJrr)rval_extracted_types rrz&make_launcher.._extracted_typesS b% ((334Cse1:  a5C< + }rc t|tr!djt|}d|dS|ddk(ry|dvrydd d d d d dddddd t |S)N()rrOrrYrXlbhiLBHIK) rrlongrrrrrrrrr)rr format_ofs rrz make_launcher..format_ofs b% ''#i,-Cse1:  a5C< +    B-  rr piiiKKOOOOrrz, c3,K|] \}}d|yw)z&_argNr.0rrs r z make_launcher..s LB5 Lsc3NK|]\}}|dk7s t|d|yw)rz argN)rrs rrz make_launcher..s,h2VX\gVgYr]O4s3hs %%rptr_infoz.dev_ptrr_argz&argz&global_scratcha; #define __HIP_PLATFORM_AMD__ #include #include #include #include #include // The list of paths to search for the HIP runtime library. The caller Python // code should substitute the search path placeholder. static const char *hipLibSearchPaths[] = {"a"}; // The list of HIP dynamic library symbols and their signature we are interested // in this file. #define HIP_SYMBOL_LIST(FOR_EACH_ERR_FN, FOR_EACH_STR_FN) \ FOR_EACH_STR_FN(hipGetErrorString, hipError_t hipError) \ FOR_EACH_ERR_FN(hipModuleLaunchKernel, hipFunction_t f, \ unsigned int gridDimX, unsigned int gridDimY, \ unsigned int gridDimZ, unsigned int blockDimX, \ unsigned int blockDimY, unsigned int blockDimZ, \ unsigned int sharedMemBytes, hipStream_t stream, \ void **kernelParams, void **extra) \ FOR_EACH_ERR_FN(hipModuleLaunchCooperativeKernel, hipFunction_t f, \ unsigned int gridDimX, unsigned int gridDimY, \ unsigned int gridDimZ, unsigned int blockDimX, \ unsigned int blockDimY, unsigned int blockDimZ, \ unsigned int sharedMemBytes, hipStream_t stream, \ void **kernelParams, void **extra) \ FOR_EACH_ERR_FN(hipPointerGetAttribute, void *data, \ hipPointer_attribute attribute, hipDeviceptr_t ptr) // The HIP symbol table for holding resolved dynamic library symbols. struct HIPSymbolTable { #define DEFINE_EACH_ERR_FIELD(hipSymbolName, ...) \ hipError_t (*hipSymbolName)(__VA_ARGS__); #define DEFINE_EACH_STR_FIELD(hipSymbolName, ...) \ const char *(*hipSymbolName)(__VA_ARGS__); HIP_SYMBOL_LIST(DEFINE_EACH_ERR_FIELD, DEFINE_EACH_STR_FIELD) }; static struct HIPSymbolTable hipSymbolTable; bool initSymbolTable() { // Use the HIP runtime library loaded into the existing process if it exits. void *lib = dlopen("libamdhip64.so", RTLD_NOLOAD); if (lib) { // printf("[triton] chosen loaded libamdhip64.so in the process\n"); } // Otherwise, go through the list of search paths to dlopen the first HIP // driver library. if (!lib) { int n = sizeof(hipLibSearchPaths) / sizeof(hipLibSearchPaths[0]); for (int i = 0; i < n; ++i) { void *handle = dlopen(hipLibSearchPaths[i], RTLD_LAZY | RTLD_LOCAL); if (handle) { lib = handle; // printf("[triton] chosen %s\n", hipLibSearchPaths[i]); } } } if (!lib) { PyErr_SetString(PyExc_RuntimeError, "cannot open libamdhip64.so"); return false; } // Resolve all symbols we are interested in. dlerror(); // Clear existing errors const char *error = NULL; #define QUERY_EACH_FN(hipSymbolName, ...) \ *(void **)&hipSymbolTable.hipSymbolName = dlsym(lib, #hipSymbolName); \ error = dlerror(); \ if (error) { \ PyErr_SetString(PyExc_RuntimeError, \ "cannot query " #hipSymbolName " from libamdhip64.so"); \ dlclose(lib); \ return false; \ } HIP_SYMBOL_LIST(QUERY_EACH_FN, QUERY_EACH_FN) return true; } static inline void gpuAssert(hipError_t code, const char *file, int line) { if (code != HIP_SUCCESS) { const char* prefix = "Triton Error [HIP]: "; const char* str = hipSymbolTable.hipGetErrorString(code); char err[1024] = {0}; snprintf(err, 1024, "%s Code: %d, Messsage: %s", prefix, code, str ); PyErr_SetString(PyExc_RuntimeError, err); } } #define HIP_CHECK(ans) { gpuAssert((ans), __FILE__, __LINE__); } static void _launch(int gridX, int gridY, int gridZ, int num_warps, int num_ctas, int launch_cooperative_grid, int clusterDimX, int clusterDimY, int clusterDimZ, int shared_memory, hipStream_t stream, hipFunction_t functionzc) { // printf("_launch hip kernel\n"); hipDeviceptr_t global_scratch = 0; void *params[] = { z }; if (gridX*gridY*gridZ > 0 && launch_cooperative_grid) { HIP_CHECK(hipSymbolTable.hipModuleLaunchCooperativeKernel(function, gridX, gridY, gridZ, z*num_warps, 1, 1, shared_memory, stream, params, 0)); return; } if (gridX*gridY*gridZ > 0) { HIP_CHECK(hipSymbolTable.hipModuleLaunchKernel(function, gridX, gridY, gridZ, a*num_warps, 1, 1, shared_memory, stream, params, 0)); } } typedef struct _DevicePtrInfo { hipDeviceptr_t dev_ptr; bool valid; } DevicePtrInfo; static inline DevicePtrInfo getPointer(PyObject *obj, int idx) { DevicePtrInfo ptr_info; ptr_info.dev_ptr = 0; ptr_info.valid = true; if (PyLong_Check(obj)) { ptr_info.dev_ptr = (hipDeviceptr_t)PyLong_AsUnsignedLongLong(obj); return ptr_info; } if (obj == Py_None) { // valid nullptr return ptr_info; } PyObject *ptr = PyObject_GetAttrString(obj, "data_ptr"); if(ptr){ PyObject *empty_tuple = PyTuple_New(0); PyObject *ret = PyObject_Call(ptr, empty_tuple, NULL); Py_DECREF(empty_tuple); Py_DECREF(ptr); if (!PyLong_Check(ret)) { PyErr_SetString(PyExc_TypeError, "data_ptr method of Pointer object must return 64-bit int"); ptr_info.valid = false; return ptr_info; } ptr_info.dev_ptr = (hipDeviceptr_t)PyLong_AsUnsignedLongLong(ret); if(!ptr_info.dev_ptr) return ptr_info; uint64_t dev_ptr; hipError_t status = hipSymbolTable.hipPointerGetAttribute(&dev_ptr, HIP_POINTER_ATTRIBUTE_DEVICE_POINTER, ptr_info.dev_ptr); if (status == hipErrorInvalidValue) { PyErr_Format(PyExc_ValueError, "Pointer argument (at %d) cannot be accessed from Triton (cpu tensor?)", idx); ptr_info.valid = false; } ptr_info.dev_ptr = (hipDeviceptr_t)dev_ptr; Py_DECREF(ret); return ptr_info; } PyErr_SetString(PyExc_TypeError, "Pointer argument must be either uint64 or have data_ptr method"); return ptr_info; } static PyObject* launch(PyObject* self, PyObject* args) { // printf("launch\n"); int gridX, gridY, gridZ; uint64_t _stream; uint64_t _function; int launch_cooperative_grid; PyObject *launch_enter_hook = NULL; PyObject *launch_exit_hook = NULL; PyObject *kernel_metadata = NULL; PyObject *launch_metadata = NULL;  z _argz; z if(!PyArg_ParseTuple(args, "a", &launch_cooperative_grid, &gridX, &gridY, &gridZ, &_stream, &_function, &kernel_metadata, &launch_metadata, &launch_enter_hook, &launch_exit_hook a=)) { return NULL; } // extract kernel metadata int num_warps, num_ctas, shared_memory, clusterDimX, clusterDimY, clusterDimZ; if (!PyArg_ParseTuple(kernel_metadata, "iiiiii", &num_warps, &num_ctas, &shared_memory, &clusterDimX, &clusterDimY, &clusterDimZ)) { return NULL; } // extract launch metadata if (launch_enter_hook != Py_None){ PyObject* args = Py_BuildValue("(O)", launch_metadata); PyObject* ret = PyObject_CallObject(launch_enter_hook, args); Py_DECREF(args); if (!ret) return NULL; } // raise exception asap zDevicePtrInfo ptr_infoz = getPointer(_argz); if (!ptr_infoz.valid) return NULL;z; _launch(gridX, gridY, gridZ, num_warps, num_ctas, launch_cooperative_grid, clusterDimX, clusterDimY, clusterDimZ, shared_memory, (hipStream_t)_stream, (hipFunction_t)_functionan); if(launch_exit_hook != Py_None){ PyObject* args = Py_BuildValue("(O)", launch_metadata); PyObject* ret = PyObject_CallObject(launch_exit_hook, args); Py_DECREF(args); if (!ret) return NULL; } if(PyErr_Occurred()) { return NULL; } // return None Py_INCREF(Py_None); return Py_None; } static PyMethodDef ModuleMethods[] = { {"launch", launch, METH_VARARGS, "Entry point for all kernels with this signature"}, {NULL, NULL, 0, NULL} // sentinel }; static struct PyModuleDef ModuleDef = { PyModuleDef_HEAD_INIT, "__triton_launcher", NULL, //documentation -1, //size ModuleMethods }; PyMODINIT_FUNC PyInit___triton_launcher(void) { if (!initSymbolTable()) { return NULL; } PyObject *m = PyModule_Create(&ModuleDef); if(m == NULL) { return NULL; } PyModule_AddFunctions(m, ModuleMethods); return m; } ) rJvaluesrlistfilterboolrL enumerater$itemsrKr_range) constants signature warp_sizer args_formatformatrs args_list arg_declsinternal_args_listrparamsrvrrrs @@@r make_launcherrs ,''93C3C3EFR9R=FGK K 'F193C3C3EFGIVD)//#"678I"+I"67$!QA7I7PST]P^abPbtyy L)//:K LLLhjI hARhhI"22 a5C<  % %8&< = ;   % %QCj 1 2 12K%I' (F&/oo&7 MUQ2;LQCj MF M MM#$ -.9MY:`ruxyBuCFGuGaehqaqMO`PPyy()*^^g]hiST]R]<^x88Y__=N OEAr#$E!B / OPQR &x(RS\Q\](99R[RaRaRcdINIJLNoqrsotx{o{&qc);A3bCSTUSVVjkBDDdeffr[^_q[ruv[vswz~zCzCDVzWsW|~r*Ck CX JCG8NT P0ds#J8* J=% K3KK ##K,ceZdZdZdZy) HIPLauncherctdr jn t}fd}|jDcic]\}}|||}}}jjDcic]\}}|| }}}t |||j td}|j|_|j|_ ycc}}wcc}}w)Nrctt|tr&jjj |fS|SN)rstrfn arg_namesindex)xrvs rz&HIPLauncher.__init__..s-Z3=OSVV--33A69UVr__triton_launcher) rrdictrrrrrlaunchlaunch_cooperative_grid) rrvmetadatararg_idxidxvaluerrs ` rrzHIPLauncher.__init__s%,S+%>CMMDF V;D??;LMZS%WS\5(M M25--2E2E2GHJCS%ZH HIy(2D2DE%c+>?jj '/'G'G$ NHs C/ C c>|j|jg|yr)rr)rargss r__call__zHIPLauncher.__call__s D00848rN)rrrrrrrrrrs H9rrcReZdZfdZdZedZdZdZdZ dZ dZ xZ S) HIPDrivercVt|t|_t|_yr)rrrutilsr launcher_cls)rrs rrzHIPDriver.__init__s Z 'rc"ddl}|jS)Nr)r=cudarr=s rget_device_interfacezHIPDriver.get_device_interfaceszzrcZ ddl}|jjduS#t$rYywxYw)NrF)r=versionhip ImportError)r=s r is_activezHIPDriver.is_actives1  ==$$D0 0  s  **c|j}|jj|}|d}|d}td|j dd|S)NarchwarpSizerr@r)get_current_devicerrrrL)rdevicedevice_propertiesrrs rget_current_targetzHIPDriver.get_current_target sU((* JJ<>rcJddl}|jd|jS)Nrr )r=rrr s rget_active_torch_devicez!HIPDriver.get_active_torch_devices||FD$;$;$=>>rcddlm}|S)Nr)do_bench)triton.testingr)rrs rget_benchmarkerzHIPDriver.get_benchmarkers +rcbddl}d}|jt|dz|jdS)Nrir )dtyper)r=emptyint)rr= cache_sizes rget_empty_cache_for_benchmarkz'HIPDriver.get_empty_cache_for_benchmarks.' {{3zQ/uyy{PPrc$|jyr)zero_)rrxs r clear_cachezHIPDriver.clear_cache!s  r) rrrrr  staticmethodrrrrr&r)rrs@rrrs;( ?? Qrr) functoolsrrdrMripathlibrtriton.runtime.buildrtriton.runtime.cachertriton.backends.compilerrtriton.backends.driverrr9rrealpath__file__rJrmr: lru_cacher_robjectrrrrrrrrr5s '2., ''//"''**84 5ww||GY/0 -`;R;R|&?v?( ,hV 9& 9 + +r