Commits

Show all
Author Commit Message Labels Comments Date
Remi Meier
fix
Branches
stmgc-c8-gcc
Remi Meier
improve information supplied to 'perf'
Branches
stmgc-c8-gcc
Remi Meier
enable 'perf' to display loop names for jit-trace profiling Add a script that extracts a perf-PID.map file from a pypylog. This allows 'perf report' to show the name of traces instead of code addresses. Still needs some work, but basic functionality is there.
Branches
stmgc-c8-gcc
Armin Rigo
Add another test, which I'm unable to fix either
Armin Rigo
(fijal, arigo) Don't use TypeError in RPython please (grafted from b2013723e0ecaefaf2f35f063c74c873bfd57667)
Armin Rigo
Issue #2100: massively improve the performance of map() with more than one sequence argument
Remi Meier
fix for interleaved jit-backend entries in pypylog
Branches
stmgc-c8-gcc
Remi Meier
import stmgc
Branches
stmgc-c8-gcc
Armin Rigo
Fixes
Branches
fix-strbuf
Armin Rigo
If unicodedata_handler happens to be None here, don't interpret \N sequences at all (grafted from 522b1566ead2b0fbdbe6952232268a07ad39eb3b)
Armin Rigo
Issue #2097: corner case in _sqlite3
Armin Rigo
(fijal, arigo) Adding the W_UnicodeBufferObject
Branches
fix-strbuf
Armin Rigo
If unicodedata_handler happens to be None here, don't interpret \N sequences at all
Branches
fix-strbuf
Remi Meier
import stmgc
Branches
stmgc-c8-gcc
Richard Plangger
finished impl of float_ne with guards
Branches
vecopt
Richard Plangger
added test to check if cast single float to bool is working properly aswell
Richard Plangger
#2101 ported fix from vecopt to default
Richard Plangger
bool(0.1) return True, in the jit bool(0.1) transforms to cast_float_to_int which is wrong, updated the jtransform.py to emit float_is_true instead (similar to the issue with cast int -> bool added implementation for both float_ne, float_eq as vector operations
Branches
vecopt
Remi Meier
import stmgc
Branches
stmgc-c8-gcc
Richard Plangger
moved vec_guard_false out of the assembler into the vector_ext file, adapted test_micronumpy to use int_is_ture instead of int_and(X, 255) (which is wrong) refactored blend unused slots (out of _guard_true/false helper for vector arguments)
Branches
vecopt
Richard Plangger
pblendw_xxi -> only tests byte for the i argument last two tests in test_vectorize
Branches
vecopt
Ronan Lamy
Support multiple axes in ufunc.reduce()
Branches
ufunc-reduce
Richard Plangger
stripping sse4 of the model returned from host platform in the test to pass it (c compiler will never append sse4) considering that descr can be a basicfaildescr in the tests
Branches
vecopt
Ronan Lamy
Refactor do_axis_reduce() so that it takes the axis_flags list instead of a single axis
Branches
ufunc-reduce
Richard Plangger
all but 2 vectoriztion tests passing again. the scheduling that prefers pure operations messes up these test cases
Branches
vecopt
Richard Plangger
restricted the test environment, loop versioning only works resumedescr (not basicfaildescr) rawstart is now saved to the loop token to be more easily reconstructed when mc and datablockwrappers are removed from the assembler
Branches
vecopt
Remi Meier
tiny bit of cleanup
Branches
stmgc-c8-gcc
Remi Meier
don't register around_extcall things *twice* for every thread start (it is still done in rthread too often)
Branches
stmgc-c8-gcc
Remi Meier
small improvement to generated code
Branches
stmgc-c8-gcc
mattip
document merged branch
  1. Prev
  2. Next