Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | add ascii fast paths to the tolower/toupper functions of the unicode dbs too | Carl Friedrich Bolz-Tereick | 2021-03-02 | 10 | -570/+162 |
| | | | | | (the part of the diff around special casing is due to c6650aed42e0 not regenerating *all* unicodedb versions, only unicodedb_5_2_0) | ||||
* | Implement vmprof support for aarch64 | fijal | 2021-03-01 | 7 | -11/+76 |
|\ | |||||
| * | hopefully add vmprof support to pypy on aarch64 | Maciej Fijalkowski | 2021-03-01 | 4 | -12/+36 |
| | | |||||
| * | use the real way to do it | fijal | 2021-03-01 | 1 | -1/+1 |
| | | |||||
| * | seems sometimes we have mc sometimes we dont | fijal | 2021-03-01 | 1 | -5/+5 |
| | | |||||
| * | an attempt to get vmprof going on aarch64 | fijal | 2021-03-01 | 1 | -0/+36 |
| | | |||||
| * | actually enable vmprof on aarch64 | Maciej Fijalkowski | 2021-02-23 | 1 | -2/+4 |
| | | |||||
| * | Start a new branch to get support for vmprof on aarch64 | Maciej Fijalkowski | 2021-02-23 | 2 | -2/+5 |
| | | |||||
* | | remove pdb | Carl Friedrich Bolz-Tereick | 2021-02-26 | 1 | -2/+0 |
| | | |||||
* | | maxsplit -> maxcount for replace functions | Carl Friedrich Bolz-Tereick | 2021-02-26 | 1 | -29/+29 |
| | | |||||
* | | remove code duplication with rstr by having the real implementation of search | Carl Friedrich Bolz-Tereick | 2021-02-26 | 2 | -110/+19 |
| | | | | | | | | only live in rlib/rstring.py | ||||
* | | follow what cpython is doing more systematically: | Carl Friedrich Bolz-Tereick | 2021-02-26 | 2 | -29/+117 |
| | | | | | | | | | | add similar cases, stop using StringBuilder, make a correctly sized llstr directly. needs a refactoring | ||||
* | | add a random test for finding | Carl Friedrich Bolz-Tereick | 2021-02-25 | 1 | -1/+21 |
| | | |||||
* | | fix a tiny performance bug in our string search that we ported from cpython. | Carl Friedrich Bolz-Tereick | 2021-02-25 | 3 | -4/+6 |
| | | | | | | | | | | | | | | | | | | | | | | | | | | the condition is a bit complicated: - we need a last character that is unique in the string - we are at a position in the string that matches the last character, but a previous char is a mismatch - the next char in the haystack is in the bloom filter if all this is met, we want to skip a whole needle length, not len(needle) - 1 this was pointed out by Tim Peters here: https://bugs.python.org/msg378301 | ||||
* | | second optimization: have a fast path in replace for single character strings | Carl Friedrich Bolz-Tereick | 2021-02-25 | 1 | -0/+30 |
| | | | | | | | | here too | ||||
* | | copy an optimization from CPython: when the search string of str.replace and | Carl Friedrich Bolz-Tereick | 2021-02-25 | 2 | -1/+18 |
|/ | | | | str.split doesn't occur in the string, don't create a copy but just reuse self. | ||||
* | update the files for vmprof-python | Maciej Fijalkowski | 2021-02-23 | 2 | -1/+5 |
| | |||||
* | merge intbount-improvements-3 | Carl Friedrich Bolz-Tereick | 2021-02-21 | 6 | -84/+251 |
|\ | | | | | | | | | refactor the intbounds analyzer a little bit, add some optimizations for int_neg and int_invert | ||||
| * | Backed out changeset f1c2880a6b49 | Carl Friedrich Bolz-Tereick | 2021-02-15 | 3 | -141/+5 |
| | | | | | | | | (some test fails weirdly, trying to see what caused it) | ||||
| * | fix type-unsafe test | Carl Friedrich Bolz-Tereick | 2020-04-28 | 1 | -5/+6 |
| | | |||||
| * | better reasoning about upper bounds of or and xor, and about lower bounds of or | Carl Friedrich Bolz-Tereick | 2020-04-29 | 3 | -5/+141 |
| | | |||||
| * | make the JIT reason about int_invert and int_neg | Carl Friedrich Bolz-Tereick | 2020-04-26 | 5 | -0/+163 |
| | | |||||
| * | kill unnecessary imports | Carl Friedrich Bolz-Tereick | 2021-02-04 | 2 | -4/+2 |
| | | |||||
| * | more of same | Carl Friedrich Bolz-Tereick | 2021-02-04 | 3 | -19/+37 |
| | | |||||
| * | call more appropriate methods a bit everywhere | Carl Friedrich Bolz-Tereick | 2021-02-04 | 3 | -58/+45 |
| | | |||||
* | | workaround for a crash when running test_recursive_pickle in test_functools on | Carl Friedrich Bolz-Tereick | 2021-02-19 | 1 | -2/+19 |
| | | | | | | | | | | | | | | | | | | the py3.7 branch The test produces stack overflows intentionally, and if we blackhole at the wrong stack depth then an vmprof frame is not popped correctly for some reason. unfortunately I was not able to find the reason, but this mitigation at least prevents the vm from crashing. | ||||
* | | randomly fix broken target | Carl Friedrich Bolz-Tereick | 2021-02-19 | 1 | -6/+7 |
| | | |||||
* | | fix C warning, lltype.Signed is a long, not an int | Carl Friedrich Bolz-Tereick | 2021-02-17 | 1 | -2/+2 |
|/ | |||||
* | use low-level interface where applicable, add at least a minimal hypothesis ↵ | Carl Friedrich Bolz-Tereick | 2021-02-04 | 2 | -13/+21 |
| | | | | test for make_... | ||||
* | remove unreachable code(?) | Carl Friedrich Bolz-Tereick | 2021-02-03 | 1 | -4/+0 |
| | |||||
* | add a comment about something that I have to re-learn every time I read this ↵ | Carl Friedrich Bolz-Tereick | 2020-04-29 | 1 | -0/+6 |
| | | | | code | ||||
* | remove copy-pasted comment | Carl Friedrich Bolz-Tereick | 2020-04-29 | 1 | -2/+0 |
| | |||||
* | express the make_le/ge/gt/lt methods in terms of their _const variants | Carl Friedrich Bolz-Tereick | 2020-04-29 | 1 | -8/+32 |
| | |||||
* | merge default | Carl Friedrich Bolz-Tereick | 2021-02-01 | 8 | -15/+75 |
|\ | |||||
| * | merge heads | Matti Picus | 2021-02-01 | 8 | -15/+75 |
| |\ | |||||
| | * | remerge fixes from rpython-never-allocate | Matti Picus | 2021-01-28 | 4 | -12/+12 |
| | |\ | |||||
| | | * | unify 'hints' and '_hints' to '_hints' | Matti Picus | 2021-01-28 | 4 | -12/+12 |
| | | | | |||||
| | * | | re-apply backed out changes from rpython-never-allocate | Matti Picus | 2021-01-28 | 4 | -3/+63 |
| | | | | |||||
* | | | | merge default | Carl Friedrich Bolz-Tereick | 2021-01-31 | 2 | -0/+30 |
|\| | | | |||||
| * | | | optimize chains of longlong2float(float2longlong(x)) and vice versa | Carl Friedrich Bolz-Tereick | 2021-01-31 | 2 | -0/+30 |
| |/ / | |||||
| * | | Backed out changes from rpython-never-allocate since it is not ready yet | Matti Picus | 2021-01-22 | 4 | -63/+3 |
| | | | |||||
| * | | Merge branch 'branch/rpython-never-allocate' into 'branch/default' | Antonio Cuni | 2021-01-21 | 4 | -3/+63 |
| |\| | | | | | | | | | | | | | | | | introduce a @never_allocate class decorator See merge request pypy/pypy!764 | ||||
| | * | improve the test to check that if we constant-fold the MyClass malloc(), ↵ | Antonio Cuni | 2020-10-02 | 1 | -6/+19 |
| | | | | | | | | | | | | everything works | ||||
| | * | introduce a @never_allocate class decorator, which ensure that a certain ↵ | Antonio Cuni | 2020-10-02 | 4 | -3/+50 |
| | | | | | | | | | | | | RPython class is never actually instantiated at runtime. Useful to ensure that e.g. it's always constant-folded away | ||||
* | | | merge default | Carl Friedrich Bolz-Tereick | 2021-01-19 | 6 | -16/+33 |
|\| | | |||||
| * | | fix a bug in the heapcache around nonstandard virtualizables (it doesn't occur | Carl Friedrich Bolz-Tereick | 2021-01-18 | 3 | -11/+21 |
| | | | | | | | | | | | | | | | | | | | | | | | | in pypy, but I had it in another jit recently): in some rare weird circumstances the nonstandard virtualizable can be a constant! also rename is_nonstandard_virtualizable to the more precise is_known_nonstandard_virtualizable | ||||
| * | | add str2constcharp for tests | Matti Picus | 2021-01-04 | 1 | -0/+7 |
| | | | |||||
| * | | test, fix parsing "inf" and friends in PyOS_string_to_double (issue 3375) | Matti Picus | 2021-01-03 | 1 | -2/+2 |
| | | | | | | | | | | | | also fix "const char *" in dg_strtod | ||||
| * | | happy new year | Matti Picus | 2021-01-01 | 1 | -3/+3 |
| | | | |||||
* | | | store everything erased | Carl Friedrich Bolz-Tereick | 2020-12-29 | 1 | -0/+1 |
|/ / |