Commits

Show all
Author Commit Message Labels Comments Date
Antonio Cuni
tweak the test to make it failing because of raw virtuals, and fix the code
Branches
virtual-raw-mallocs
Antonio Cuni
rpython fix: this can never happen
Branches
virtual-raw-mallocs
Antonio Cuni
implement bh_new_raw_buffer for ll cpus: this makes x86 virtualref tests passing
Branches
virtual-raw-mallocs
Antonio Cuni
move raw virtuals tests to VirtualMiscTests, so that they can be run also by the x86 backend tests
Branches
virtual-raw-mallocs
Antonio Cuni
tentative RPython fix: in ResumeDataDirectReader virtuals_int_cache contains real integers, while in ResumeDataBox reader it contains boxes, so we cannot initialize it by 0
Branches
virtual-raw-mallocs
Antonio Cuni
kill support for rawbuffer virtualstate: we do not need to have it for optimizing cffi, and it seems to cause bugs
Branches
virtual-raw-mallocs
Antonio Cuni
failing test and fix for forcing raw virtuals after GUARD_NOT_FORCE
Branches
virtual-raw-mallocs
Antonio Cuni
apply the raw virtuals optimization only if we are allocating an array of chars. With other types it's harder because the optimizer does not know the size of the items, and it's not useful for cffi anyway
Branches
virtual-raw-mallocs
Antonio Cuni
add a test for a virtual which survives one iteration of the loop, and fix it
Branches
virtual-raw-mallocs
Antonio Cuni
implement resume for virtual raw buffers
Branches
virtual-raw-mallocs
Antonio Cuni
dump RawBuffer logs when something goes wrong
Branches
virtual-raw-mallocs
Antonio Cuni
bah, it might happen that we have two arraydescrs which are equal but not identical: apply the optimization also in those cases
Branches
virtual-raw-mallocs
Antonio Cuni
make the official integral type for pointers unsigned. This is in line to what cffi does
Branches
virtual-raw-mallocs
Antonio Cuni
some refactoring to make sure that we always cast integer types to the correct signed/unsigned type when writing to cdata. It is hard to test because it does not have any visible result: the only change is that in the JIT trace the SETARRAYITEM_RAW will have the correct signed/unsigned descr, and it will allow the raw_malloc virtuals to work
Branches
virtual-raw-mallocs
Antonio Cuni
backout 71357aa32f93: it does not solve the problem, we will hack it differently
Branches
virtual-raw-mallocs
Antonio Cuni
hg merge default; this fixes the problem with raw_malloc and integers, they are now optimized as well
Branches
virtual-raw-mallocs
Antonio Cuni
use ldexp instead of pow, to make sure that the raw_malloc optimization works also with integers type. Failing so far
Branches
virtual-raw-mallocs
Antonio Cuni
improve the test to check that the raw_malloc has been virtualized
Branches
virtual-raw-mallocs
Antonio Cuni
rename this file to test_ffi, to contain all the ffi related tests (so not only _ffi, but also _cffi_backend)
Branches
virtual-raw-mallocs
Armin Rigo
Try to avoid writing with a type on which the sign was forgotten.
Antonio Cuni
factor out a lot of common code in virtualstate.py too
Branches
virtual-raw-mallocs
Antonio Cuni
refactor things to share lot of code between VArrayValue and VRawBufferValue
Branches
virtual-raw-mallocs
Armin Rigo
Add also the fence instructions
Armin Rigo
Add movnt to ignore.
Armin Rigo
Copy-paste this line from rctime.txt.
Armin Rigo
Merged in bdkearns/pypy (pull request #100: replace some obsolete use of time with rctime, more test_ioctl fixes):
Brian Kearns
additional fixes for test_ioctl
Brian Kearns
replace some obsolete use of time with rctime
Armin Rigo
Allow "usemodules=('time',)" to be satisfied also by the 'rctime' module.
Armin Rigo
Fix leak: calling .copy() would initialize a ctx and then overwrite it
  1. Prev
  2. Next