ɯwg5ddlZddlZddlZddlZddlmZmZddlZddlm Z ddl m Z ddl m Z ddlmZddlmZdd lmZGd d ZGd d ZGddZeZy)N)OptionalUnion) free_symbols unimplemented)NewCellVariable)ConstantVariable)ClosureVariable)SymNodeVariableceZdZdZddZdZdZdZddee de e e jffd Z d Zd Zd Zd ZdZdZdefdZy) ComptimeVara A ComptimeVar represents a Python value, at some particular point in time, in the Python code we are symbolically evaluating with torchdynamo. This must be distinguished from a runtime value, as at compile-time there are some properties of the variable we do not know (for example, if the ComptimeVar represents a Tensor, we only know metadata about the tensor; we do NOT know what the actual data in the Tensor is.) returnNc||_yN_ComptimeVar__variable)selfvs [/home/mcse/projects/flask/flask-venv/lib/python3.12/site-packages/torch/_dynamo/comptime.py__init__zComptimeVar.__init__$s c6|jjS)a Returns an fx.Proxy (or tuple/list of fx.Proxy) representing this variable in the FX graph we are assembling to pass to the user compiler. This method only works for variables we actually track in the FX graph, aka Tensors (and ints, if you are compiling with dynamic shapes). In particular, if you have a list or tuple of tensors, you will get a list/tuple of proxies (not a single proxy representing the entire list/tuple). )ras_proxyrs rrzComptimeVar.as_proxy's''))rc6|jjS)z; Returns True if as_proxy() would succeed. )ris_proxyrs rrzComptimeVar.is_proxy5s''))rcd|jjjjdS)ax Returns a "fake" value (either a FakeTensor or a SymInt) representing the variable in question. This only works for variables that denote Tensor or int. You can use this to query metadata; e.g., v.as_fake().size(0) will tell you the compile-time known size of the tensor. WARNING: Do NOT mutate the returned tensor. example_value)rrnodemetars ras_fakezComptimeVar.as_fake;s('')..33ODDrdimc@|jj|S)z Returns the size of the tensor (if dim is None) or the size at the dimension dim. The returned size may be a SymInt. )r"size)rr#s rr%zComptimeVar.sizeGs ||~""3''rc6|jjS)zd Returns what type(v) would have returned for the variable at compile time. )r python_typers rr'zComptimeVar.python_typeNs **,,rc6|jjS)a Returns the Python value this variable would have, but only if it is completely known at compile-time (e.g., it is constant). WARNING: Do NOT mutate the returned constant. The returned constant may or may not correspond to the actual value this variable may take on at runtime; for example, if the variable in question is a constant list, we may return a copy of that list. )ras_python_constantrs rr)zComptimeVar.as_python_constantUs1133rc6|jjS)zC Returns True if as_python_constant would succeed. )ris_python_constantrs rr+zComptimeVar.is_python_constantas1133rct|jtr*t|jj}t |Sy)NF) isinstancerr rsym_numbool)rfss r is_dynamiczComptimeVar.is_dynamicgs0 doo 7doo556B8Orct|jtr|jjyt|jtryt d|jdt |jd)zW Forces that a value is static, inducing a guard on its specific value z cannot force z (z) staticN)r-rr evaluate_exprr AssertionErrortypers r force_staticzComptimeVar.force_staticmsb doo 7 OO ) ) + )9 :  043H2IR rc|jS)z Returns the internal data structure VariableTracker that Dynamo uses to represent variables at compile time. There are no BC guarantees on this API and WE RESERVE THE RIGHT TO BREAK YOUR CODE if you rely on it. rrs r1_i_will_not_complain_if_bc_breaks_VariableTrackerz=ComptimeVar._i_will_not_complain_if_bc_breaks_VariableTracker{src6|jjSr)r debug_reprrs r__repr__zComptimeVar.__repr__s))++rrNr)__name__ __module__ __qualname____doc__rrrr"rintrtorchSymIntr%r'r)r+r1r6r8strr;rrrrsk ** E( (sELL7H1I(- 44  ,#,rrceZdZdZddZdddedefdZdd Zd Z d Z d dd dZ dZ dZ dddZddddZddddZddddZddddZdddZdZdZy)ComptimeContextz This context class provides access to a public API for Dynamo's internals. If there is something here you would find useful that is missing, please file a feature request at https://github.com/pytorch/pytorch/ rNc||_yr_ComptimeContext__tx)rtxs rrzComptimeContext.__init__s  rr stacklevelnamec|j|}t|dr||jvr|j|}t|tr6t |j jj|jSt |j jj|S|j|}t|tr.t |j jj|St |S)zL Retrieve the compile-time known information about a local. closure_cells) _ComptimeContext__get_txhasattrrPr-r routputroot_txsymbolic_localsrN side_effects load_cellr )rrNrMrKcellrs r get_localzComptimeContext.get_locals]]: & 2 'DB4D4D,D##D)D$0"299#4#4#D#DTYY#OPP"299#9#9#C#CD#IJJ""4(A!_-"299#9#9#C#CA#FGG"1~%rct|y)z0 Manually trigger a graph break Nr)rmsgs r graph_breakzComptimeContext.graph_breaks  crcB|jjjS)z Retrieve the partially constructed FX graph that would be passed to the user compiler after compilation. )rJrSgraphrs rr_zComptimeContext.graphs yy%%%rc2|jrJdy)zV Asserts that the int is static (and not dynamic, per dynamic shapes) zKexpected static but got dynamic (run with TORCH_LOGS=dynamic for more info)N)r1)rvals r assert_staticzComptimeContext.assert_statics#   Y X Y  rT)verbosefilect|jjjj d|j |y)z Print the partially constructed FX graph that would be passed to the user compiler after compilation. r)rcrdN)printrJrSr_ python_codesrc)rrcrds r print_graphzComptimeContext.print_graphs8  II   " " . .vw . G K KRV rc@t|jjSr)rGrJparentrs rrlzComptimeContext.parentstyy//00rcV|j}t|D]}|j}|Sr)rJrangerl)rrMrK_s r__get_txzComptimeContext.__get_txs- YYz" AB  rrfc0tt||y)Nrf)rgrepr)rrards rrgzComptimeContext.prints d3id#r)rdrMc|j|}ttj|j|j |j jj|y)z Print the current series of opcodes being executed (not including parent frames), including where you are in the particular opcode stream. )current_offsetrfN)rQrgdisBytecodef_code instructionsinstruction_pointeroffset)rrdrMrKs r print_disaszComptimeContext.print_disassQ ]]: & LL !r/E/EFMM ce  rc|j|}|jD] }td|j|"y)a Print the current Python value stack. Note that this is NOT the same as the traceback; use print_bt() to print that. Note that at stacklevel=0, this will typically be empty, as comptime cannot currently be used in an expression context where there would be intermediates on the stack. If you would find this useful, please file a bug at https://github.com/pytorch/pytorch/ NB: Stack grows downwards in our print z- rfN)rQstackrgr:)rrdrMrKss rprint_value_stackz!ComptimeContext.print_value_stacks=]]: & 4A Bq||~&'d 3 4rc|j|}|jjD]%\}}t|d|j |'y)z Print all of the locals available in the current context. By default this view is very limited; you can get more information about any individual local using get_local(). z = rfN)rQrUitemsrgr:)rrdrMrKkrs r print_localszComptimeContext.print_localssP ]]: &&&,,. 8DAq QCs1<<>*+$ 7 8rc ,g}|j|}|/|j|jt|dd}|/t dj t jjt|j|y)a Print the user code backtrace, starting at the beginning of the frame Dynamo started evaluating. Note that this MAY NOT go all the way to the torch.compile invocation, as we may have done a graph break and are compiling an intermediate frame as the starting point. If you think the other behavior would be better, file a bug at https://github.com/pytorch/pytorch/ Nrlrf) rQappend frame_summarygetattrrgjoin traceback StackSummary from_listreversedformat)rrdrMr}rKs rprint_btzComptimeContext.print_bts} ]]: &n LL))+ ,Xt,Bn  GGI**44Xe_ELLN O rctdjdt|jjj D|y)z Print the currently installed guards for the Dynamo context. This does NOT include guards associated with variables that may or may not be installed in the future if those variables are used.  c34K|]}t|ywr)rr).0guards r z/ComptimeContext.print_guards..sT5e &TsrfN)rgrsortedrJrSguards)rrds r print_guardszComptimeContext.print_guards s6  IITF499;K;K;R;R4ST T rc|jS)a Returns the internal data structure InstructionTranslator that Dynamo uses to track state of symbolic evaluation. There are no BC guarantees on this API and WE RESERVE THE RIGHT TO BREAK YOUR CODE if you rely on it. rIrs r7_i_will_not_complain_if_bc_breaks_InstructionTranslatorzGComptimeContext._i_will_not_complain_if_bc_breaks_InstructionTranslatorsyyrc.tj|yr)timesleep)rsecs rrzComptimeContext.sleep$s  3rr<)zComptimeContext.graph_break)r=r>r?r@rrDrrZr]r_rbrjrlrQrgr{rrrrrrrErrrGrGs 23&c&{&( &Y&* 1 "&$#'1  )- 4$(A8 $ &$(  rrGceZdZedfdZedZedZedZedddZeddd Z eddd Z eddd Z eddd Z ed Z edZedZedZedZy) _ComptimecyrrErErrz_Comptime.*src|y)zJfn gets called at compile time in TorchDynamo, calls fallback_fn otherwiseNrE)fn fallback_fns r__call__z_Comptime.__call__)s   rctdy)Nc"|jSr)r]ctxs rrz'_Comptime.graph_break..2 S__.rcomptimerErrr]z_Comptime.graph_break0 ./rc&tdfdy)NcB|j|jdS)Ne)rgrZrs rrz!_Comptime.print..6sSYYs}}S'9:rctSr)rgrsrrz!_Comptime.print..6s E!Hrrrs`rrgz_Comptime.print4s:.:rrrrErrrjz_Comptime.print_graph8rrrrLctdy)Ncf|j|jdjdzSNrMrrL)r{rZr)rs rrz'_Comptime.print_disas..?s.==6IIKaO(rrrLs rr{z_Comptime.print_disas<  rctdy)Ncf|j|jdjdzSrrrZr)rs rrz-_Comptime.print_value_stack..G0--==6IIKaO.rrrLs rrz_Comptime.print_value_stackDrrctd|S)Ncf|j|jdjdzSrrrs rrz8_Comptime.print_value_stack_and_return..Rrrr)rrMs rprint_value_stack_and_returnz&_Comptime.print_value_stack_and_returnOs  rctdy)Ncf|j|jdjdzSr)rrZr)rs rrz(_Comptime.print_locals..[s0((==6IIKaO)rrrLs rrz_Comptime.print_localsXrrctdy)Ncf|j|jdjdzSr)rrZr)rs rrz$_Comptime.print_bt..cs. ==6IIKaO%rrrLs rrz_Comptime.print_bt`rrctdy)Nc"|jSr)rrs rrz(_Comptime.print_guards..jsS--/rrrErrrz_Comptime.print_guardshs /0rctdy)NcB|j|jdSNra)rbrZrs rrz)_Comptime.assert_static..nsS..s}}U/CDrrras rrbz_Comptime.assert_staticls DErctdy)Nc@|jdjSr)rZr6rs rrz(_Comptime.force_static..rsS]]51>>@rrrs rr6z_Comptime.force_staticps @Arc d}t|y)a Like pdb breakpoint(), but drop into pdb whenever this line of code is compiled by dynamo. Use it by putting this in your model code:: from torch._dynamo.comptime import comptime comptime.breakpoint() And then, inside pdb, you can access 'ctx' to query things about the compilation context:: (Pdb) !ctx.print_bt() (Pdb) !ctx.print_locals() (Pdb) p ctx.get_local("attention").as_fake() cL|j}tjyr)rlbuiltins breakpoint) inner_ctxrs rinnerz#_Comptime.breakpoint..inners""$C    !rNr)rs rrz_Comptime.breakpointts$ " rctdy)Nc^|j|jdjS)Nr)rrZr)rs rrz!_Comptime.sleep..s!SYYs}}U';'N'N'PQrr)rs rrz_Comptime.sleeps QRrN)r=r>r? staticmethodrr]rgrjr{rrrrrrbr6rrrErrrr(s2!- 00NN00"#  ()  67#$     11FFBB.SSrr)rrurrtypingrrrB%torch.fx.experimental.symbolic_shapesrexcr variablesr variables.constantr variables.miscr variables.tensorr rrGrrrErrrs] " >&0+-l,l,b[[|fSfSR ;r