Commits

Author Commit Message Labels Comments Date
Maciej Fijalkowski
Add a harmless assert
Branches
direct-assembler-call
Maciej Fijalkowski
Seems that we're a bit out of luck in places :( I cannot possible imagine how the virtualizable might not already be on the stack, however, if this is true this will crash something. Since I have no clue, I can't test it a bit, maybe deeper thinking is required.
Branches
direct-assembler-call
Maciej Fijalkowski
add a passing test. At least we test certain path of code
Branches
direct-assembler-call
Maciej Fijalkowski
Write direct tests for assembler_call support in the backend. Implement missing cases in x86 backend
Branches
direct-assembler-call
Maciej Fijalkowski
This is supposed to be a non-resizable list
Branches
direct-assembler-call
Maciej Fijalkowski
RPythonization. While we're at it, store portal_calldescr directly on cpu instead of on LoopToken (we can have only one anyway)
Branches
direct-assembler-call
Maciej Fijalkowski
turn on by default list_comprehenstion_operations. Saves some annoying warnings
Branches
direct-assembler-call
Maciej Fijalkowski
implement guard recovery code, forgot that we have a guard to handle as well
Branches
direct-assembler-call
Maciej Fijalkowski
Pass the first virtualizable test. Seems the other one explodes with "already forced" assertion error.
Branches
direct-assembler-call
Maciej Fijalkowski
Do the correct thing. Add an additional _x86_direct_bootstrap_code to LoopToken which represents a compiled portal that has convinient semantics for calling.
Branches
direct-assembler-call
Maciej Fijalkowski
(pedronis, fijal) IN-PROGRESS start implementing CALL_ASSEMBLER on x86 backend, passes 3 tests so far
Branches
direct-assembler-call
Maciej Fijalkowski
(pedronis, fijal) Attach a calldescr to a token
Branches
direct-assembler-call
Maciej Fijalkowski
(pedronis, fijal) Hack a bit to make forcing work as expected.
Branches
direct-assembler-call
Maciej Fijalkowski
A test that is supposed to be forcing virtualizable, except it's not
Branches
direct-assembler-call
Maciej Fijalkowski
(pedronis, fijal) A bit of obscure interface, but pass the virtualizable test of direct assembler call.
Branches
direct-assembler-call
Maciej Fijalkowski
(pedronis, fijal) Consider CALL_ASSEMBLER a call with effectinfo=None
Branches
direct-assembler-call
Maciej Fijalkowski
(pedronis, fijal) Directly test assembler_call_helper
Branches
direct-assembler-call
Maciej Fijalkowski
Call portal directly, instead of setting future values. Should sort-of-work, needs more tests
Branches
direct-assembler-call
Samuele Pedroni
XXX remarking a confusion point
Branches
direct-assembler-call
Maciej Fijalkowski
Surprisingly, this seems to be enough to cover all cases for direct assembler call helper.
Branches
direct-assembler-call
Maciej Fijalkowski
(pedronis, fijal) Do the correct thing about raising an exception from a direct assembler call
Branches
direct-assembler-call
Maciej Fijalkowski
(pedronis, fijal) Pass the next. We grow helper in warmspot.py that is used in case direct assembler call fails.
Branches
direct-assembler-call
Maciej Fijalkowski
Reformatting
Branches
direct-assembler-call
Maciej Fijalkowski
(pedronis, fijal) Pass the first test. Trace like a residual call to portal, but instead substitute residual call with a direct assembler call later on. Break translation of llgraph backend for now.
Branches
direct-assembler-call
Maciej Fijalkowski
(pedronis, fijal) Merge 69283 from the previous branch, (arigo, fijal) IN-PROGRESS start thinking about direct calls
Branches
direct-assembler-call
Maciej Fijalkowski
(pedronis, fijal) Branch to implement direct assembler calls
Branches
direct-assembler-call
Carl Friedrich Bolz
Finally I found a good reason to have folding of pure ops with identical arguments: When using generators, the residual code of the function using the generator is handling the generator frame. Doing this produces many oois instructions, that check again and again whether the generator frame is the virtualizable or not.
Carl Friedrich Bolz
fix comment, importing lives somewhere else now.
Armin Rigo
Merge branch/c-traceback. It adds to all crashing RPython programs a traceback capability that should let it display where the RPython exception comes from. (Only implemented in the C backend; I except the oo backends to be already able to give such tracebacks.)
Benjamin Peterson
allow __get__ to raise an AttributeError triggering the use of __getattr__
  1. Prev
  2. Next