+ j@^RIt^RIt^RIt^RIt^RIHt^RIHt^RIHt^RI H t H t H t H t ^RIHt^RIHt]P$P']P$P)]44t]P$P-]R4.tRsRsRsRsRsR t]P<!4R 4t!R R ] 4t!R t"Rt#Rt$Rt%Rt&!RR] 4t'!RR] 4t(R#)N)Path)knobs) GPUTarget) GPUDriverdecompose_descriptorexpand_signaturewrap_handle_tensordesc_impl) _allocation)compile_module_from_srcincludecaa a a a^RIpVP4R8wdR#^RIo ^RIHpHpHpHo Ho Hp!V V 3RlRS P4pS PW5!V4V!V4V!V44pS PR4PpTS .TnY8nRoS P!S^,4p T TT3Rlp T!T!T 4T 4'd&\"P$!S P'T 44#R# \dR#i;i) rNLinux)c_charc_intc_size_tc_void_pc_char_pPOINTERc,<]tRt^ tRZ3RZ3.tRtR#)8_find_already_mmapped_dylib_on_linux..DlPhdrInfo dlpi_addr dlpi_nameN)__name__ __module__ __qualname____firstlineno___fields___static_attributes__)rrs[/home/luca/screen-checking/.venv/lib/python3.14/site-packages/triton/backends/amd/driver.py DlPhdrInfor s ( # ( # r!z libc.so.6ic <VPPp\\P!V44pSVP 9d(SP W#\S\V444^#^#)) contentsrrosfsdecodenamememmoveminlen)infosizedatarpctypeslib_namemax_path_lengths&&& r callback6_find_already_mmapped_dylib_on_linux..callback7sPMM++ Y' ( qvv  NN4CY,P Qr")platformsystemr0rrrrrr Structure CFUNCTYPECDLLdl_iterate_phdr Exceptionargtypesrestypecreate_string_bufferr&r' string_at)r1r5rrrrr! callback_tr:pathr3rrr0r2sf @@@@r $_find_already_mmapped_dylib_on_linuxrBsG# KK  V%% !!%)OQXY_Q`aJ ++k2BB!+H5O#O  & &': ;Dz(+T22{{6++D122 + s4C<< D  D c Rp\PP;p'dPVPV4'd(\P P V4'dV#\RV RV 24h\V4pV'd9\P P V4'dV#\RV RV 24h.p\P P\P P\4RV4p\P P V4'dV#VPV4^RI pVP4pVP4pVP 'd V.V,pVF_p\P PVRRV4p\P P V4'dVu#VPV4Ka \P"!R4p V 'drV P%R 4F\p \P PW4p \P P V 4'dV u#VPV 4K^ \P"!R 4p V 'dZ\P PV RV4p \P P V 4'dV #VPV 4\&P(!R R .4P+4P-4pV'dZ\P PVRV4p \P P V 4'dV #VPV 4\P"!R 4pT'dZ\P PTRT4p\P P T4'dT#TPT4\&P(!RR.4P+RR7pTP34Uu.uFApTP-4PT4'gK*TP%4R,NKC ppTF=p\P P T4'dTu#TPT4K? \P PRT4p\P P T4'dT#TPT4\RT RT 24h \&P.\03dELi;iuupi)zlibamdhip64.sozTRITON_LIBHIP_PATH 'z' does not point to a valid zmemory mapped 'z'' in process does not point to a valid libNtorchLD_LIBRARY_PATH:HIP_PATH hipconfigz--path ROCM_PATHz/sbin/ldconfigz-pignore)errorsz/opt/rocm/lib/zcannot locate z after attempted paths )ramd libhip_pathendswithr&rAexists RuntimeErrorrBjoindirname__file__appendsitegetsitepackagesgetusersitepackagesENABLE_USER_SITEgetenvsplit subprocess check_outputdecodestripCalledProcessErrorFileNotFoundError splitlines)r1env_libhip_path mmapped_pathpaths local_librW site_packages user_siterAenv_ld_library_pathdf env_hip_path hip_lib_pathhip_root env_rocm_path rocm_lib_pathlibslinelocsloccommon_install_paths r _get_path_to_hip_runtime_dylibrwEsH ))////  # #H - -"''..2Q2Q" "1/1BB^_g^hijj8AL 77>>, ' ' _\N:abjaklmm E RWW__X6xHI ww~~i   LL((*M((*I " m3 ww||D'5(; 77>>$  K T ))$56$**3/A Q)Aww~~a   LLO 099Z(Lww||L%B 77>>, ' '  \"  **K+BCJJLRRT 77<<%BLww~~l++## LL & IIk*M ]E8D 77>>- ( (  ]#  " "$4d#; < C C8 C TD*.): ^):djjl>S>ST\>] DJJL  ):D ^ 77>>#  J S '',,'7B ww~~)**"" LL$% z1HP QQ;  ) )+< =   " _s1';T"#AT")T"+T"-'UU"UUc8aa]tRt^toV3RltRtRtVtV;t#)HIPUtilscp<\VR4'g\\V`V4VnVP#)instance)hasattrsuperry__new__r{)cls __class__s&r r~HIPUtils.__new__s-sJ'' 37%>"%(%>%>"jj (+(D(D% --oo )) ^^ MM r")rrrrrr) rrrrr~rr__classdictcell__ __classcell__r __classdict__s@@r ryrys ""r"rycVPR4'dR#VR8XdR#RRRRRRR R R R R RRRRRRRRRRRRRRRRRRR/V,#)*hipDeviceptr_t tensordesc TDMDescriptori1int8_ti8i16int16_ti32int32_ti64int64_tu1uint8_tu8u16uint16_tu32uint32_tu64uint64_tfp16doublebf16fp32f32fp64) startswith)tys&r ty_to_cpprs~ }}S \ h h y y y i i z z z x   !  r"caV3Rlo.pVF pS!W!4K VUu.uF q3R8wgK VNK pp\PPPPP V4#uupi)z^ Creates a kernel signature in C to be able to efficiently extract arguments in the launcher. c|<\V\4'dVF pS!W!4K R#VPV4R#N) isinstancetuplerV)sigoutputx_flatten_signatures&& r r1make_kernel_signature.._flatten_signatures0 c5 ! !"1- MM# r" constexpr)tritonruntimedriveractiveutilsr) signatureflat_signaturerrkernel_signaturers& @r make_kernel_signaturersk N3/#1F>a+5E>F >> ' ' - - F FGW XXGs A1A1c0.pVFp\V\4'd,VP\\ V4\ R74KDVR8wd#VP\R\ R74KmVP\R\R74K V#)z This recreates the signature with annotations as C objects which can then be used to efficiently flatten tuples, and remove constexpr in the launcher. ) nested_tupletyperN)rrrVrannotate_argumentsrrr)rannotated_argumentsrs& r rrsx  c5 ! !  & & ASTWAX_h(i k K   & &{:'V W  & &{='Y Z  r"c PVf \V4#V^,pVPpVPpVPP 4pRV9dRV9gQhVR,pVR,p^^rVP R.4p V 'd5\ V 4^8Xd\ V ^,4^8XgQhV ^,wrV^,p \PPPp \V \4'gQhV PPWxWWWV4pV.VOVO#)ai Translate a tensor descriptor argument into the appropriate list of kernel arguments. If `tensordesc_metadata` is provided, we will create a TDMDescriptor object. Otherwise, we decompose the tensor descriptor into base pointer, shape, strides, and padding flag. In both cases, we append the shape and strides at the end to match the expected kernel signature. elem_bits block_sizeinterval_padding_pairs)rshapestridesbasedata_ptrgetr+rrrrr HIPDriverrr)argtensordesc_metadata base_argskernel_metadatarrrrr pad_interval pad_amountr num_warpsrdescs&&& r make_tensordesc_argrs$"#C((lO IIEkkG 88   D - -,BU2UU U#K0I$\2J !*0445MrR)*a/C8Nq8Q4RVW4WWW#9!#< "I ^^ " " ) )F fi ( (( ( << - -iY^h.5 =D  #5 #7 ##r"c$\WV\4#)a Wrap a kernel launcher function to handle tensor descriptor arguments. Use the provided `tensordesc_metadata` to determine whether to create TDMDescriptor objects or decompose the tensor descriptors. Args: launcher (callable): The original kernel launcher function. signature (Dict[int, str]): The kernel signature mapping argument indices to types. tensordesc_metadata (List[Dict] or None): The list of tensor descriptor metadata, following the order of tensor descriptor arguments. If None, decompose tensor descriptors. Returns: launcher (callable): The wrapped kernel launcher function. )r r)launcherrtensordesc_metas&&&r wrap_handle_tensordescr)s 'xOM` aar"c,a]tRtRtoRtRtRtVtR#) HIPLauncheri;c6a\SR4'd SPM \4pV3RlpVP4UUu/uFwrVV!V4VbK pppSPP4UUu/uFwrVWVbK ppp\ VRR4p\ PPPPPp \VP4VR4p \V 4Vn\!V 4Vn\%WV4Vn VP&VnVP(VnVP&'d~\ PPPp \+V \,4'gQhV P/4p V PP1V 4p V R,'gQR4hVP2VnVP4VnVP6VnVP8VnR#uuppiuuppi) constantsc<\V\4'd'SPPP V43#T#r)rstrfn arg_namesindex)rrs&r &HIPLauncher.__init__..?s2Z3=O=OSVV--33A69VUVVr"rNrcooperativeLaunchz8Cooperative launch requested but not supported by device)r|rdictitemsrgetattrrrrrrrrvaluesrarg_annotationsrrrlaunch_cooperative_grid warp_sizerrget_current_devicerglobal_scratch_sizeglobal_scratch_alignprofile_scratch_sizeprofile_scratch_align)rrmetadatararg_idxidxvaluerrrexpanded_signaturerdevicedevice_propertiess&f& r rHIPLauncher.__init__=s%,S+%>%>CMMDF V;D??;LM;LZSWS\5(;L M25--2E2E2GH2GJCSZ2G H!(,=tD>>((//55<<-i.>.>.@/S_`12DE 56H I,X/R '/'G'G$!++  ' ' '^^**11Ffi00 00..0F & B B6 J $%899 KJ K9#+#?#? $,$A$A!$,$A$A!%-%C%C"+NHs H7 Hc *aaaaa\PPPoVVVV3Rlp VVVVV3Rlp V !VPVP \ P4p \ P!4'd.V !VPVP\ P4pMV !VPVP4pVPVPSSSSW]WWxV VPVPVP V 4R#)cx<V^8d2SS,S,pW0,pVP4pV!WAS 4#R#rN)r) r-align allocator grid_size alloc_sizealloc_fngridXgridYgridZstreams &&& r allocate_scratch.HIPLauncher.__call__..allocate_scratch[s;ax!EME1 &- $==? 6::r"cj<V^8d+SS,S,pW ,pSPW1S4#R#r ) allocate_default_profile_scratch) r-rrr active_driverrrrrs && r r>HIPLauncher.__call__..allocate_default_profile_scratchcs7ax!EME1 &- $EEjY_``r"N)rrrrrrr _allocatorhas_profile_allocatorrr_profile_allocatorrrrrr)rrrrrfunctionrlaunch_metadatalaunch_enter_hooklaunch_exit_hookargsrrglobal_scratchprofile_scratchrs&ffff&&&&&* @r __call__HIPLauncher.__call__Ws--44     *$*B*BDD]D]_j_u_uv  , , . ..t/H/H$JdJd/:/M/MOO?t?X?XZ^ZtZtuO D00%vx#oZjNND$8$8$:O:OQU Wr") rrrrrrrrrN)rrrrrr'rr)rs@r rr;sD4WWr"rcxaa]tRtRtoV3RltRt]R4tV3RlRltRt Rt R t R t R t R tVtV;t#) rivcX<\SV`4\4Vn\VnR#r)r}rryrr launcher_cls)rrs&r rHIPDriver.__init__xs Z 'r"c"^RIpVP#r )rEcudarrEs& r get_device_interfaceHIPDriver.get_device_interface}szzr"c^RIpVPP4;'dVPPRJ# \ dR#i;i)rNF)rEr. is_availableversionhip ImportError)rEs r is_activeHIPDriver.is_activesF  ::**,PP%--2C2C42O P  s$AA AAc&<V^8dQhRS[RS[/#)rreturn)r)formatrs"r __annotate__HIPDriver.__annotate__sr"c\V4#r)r)rrs&&r map_python_to_cpp_type HIPDriver.map_python_to_cpp_types }r"cVP4pVPPV4p\PP ;'g VR,pVR,p\ RVPR4^,V4#)archwarpSizer5rG)rrrrr override_archrr\)rrr rCrs& r get_current_targetHIPDriver.get_current_targetsh((* JJ<>r"cJ^RIpVPRVP44#)rNr.)rErrr/s& r get_active_torch_device!HIPDriver.get_active_torch_devices||FD$;$;$=>>r"c^RIHpV#)r)do_bench)triton.testingrL)rrLs& r get_benchmarkerHIPDriver.get_benchmarkers +r"cj^RIpRpVP\V^,4VPRR7#)rNr.)dtyperi)rEemptyint)rrE cache_sizes& r get_empty_cache_for_benchmark'HIPDriver.get_empty_cache_for_benchmarks.' {{3zQ/uyy{PPr"c&VP4R#r)zero_)rcaches&&r clear_cacheHIPDriver.clear_caches  r")r+r)rrrrrr0 staticmethodr7r@rFrIrNrUrZrrrrs@@r rrvsL( ?? Qr"r)) functoolsr&r]rpathlibrrtriton.backends.compilerrtriton.backends.driverrrrr triton.runtimer triton.runtime.buildr rArTrealpathrUrSrrrrrrrB lru_cacherwobjectryrrrrrrrrr"r rfs  .qq&8 ''//"''**84 5 Wi01     -` \R\R~"v"D 0Y,  !$Hb$8W&8Wv. .r"