Carl Friedrich Bolz-Tereick pushed to branch branch/fix-intutils-ovf-bug at PyPy / pypy
Commits: 350a1e9c by Carl Friedrich Bolz-Tereick at 2022-10-15T15:24:27+02:00 int_neg also has a bug also add some consistency checks between the methods --HG-- branch : fix-intutils-ovf-bug - - - - - ca0788f3 by Carl Friedrich Bolz-Tereick at 2022-10-15T19:02:42+02:00 get rid of has_upper and has_lower completely Their meaning was basically never super obvious, and you need to be careful to understand it correctly everywhere. Instead, we now have: - methods that check whether overflow is known to not occur for certain operations (add/sub/.._bound_cannot_overflow) - the default add/sub/mul/..._bound methods will return result bounds that are correct whether overflow happens or not. - for int_add_ovf/guard_no_overflow there is another new method on the bounds, add_bound_no_overflow that returns the tighter range that applies if we know from the guard that overflow has not occurred. --HG-- branch : fix-intutils-ovf-bug - - - - - 92e91d51 by Carl Friedrich Bolz-Tereick at 2022-10-15T19:41:08+02:00 another incorrect test, about int_neg: the following implication is wrong in the presence of overflow: -x <= 0 ⇒ x >= 0 with the (only) counterexample of x = MININT the inverse one holds: -x >= 0 ⇒ x <= 0 in practice, int_neg is mostly preceded by int_eq(x, MINIT), guard_false, so make things work in that case --HG-- branch : fix-intutils-ovf-bug - - - - - 7 changed files: - rpython/jit/metainterp/optimizeopt/intbounds.py - rpython/jit/metainterp/optimizeopt/intutils.py - rpython/jit/metainterp/optimizeopt/rewrite.py - rpython/jit/metainterp/optimizeopt/test/test_intbound.py - rpython/jit/metainterp/optimizeopt/test/test_optimizebasic.py - rpython/jit/metainterp/optimizeopt/unroll.py - rpython/jit/metainterp/optimizeopt/vstring.py View it on Heptapod: https://foss.heptapod.net/pypy/pypy/-/compare/cba4ba4fcbdc602c7e50d91ab5e19faa554cf827...92e91d515a3234c03cf94fd7cff4693a3e61c037 -- View it on Heptapod: https://foss.heptapod.net/pypy/pypy/-/compare/cba4ba4fcbdc602c7e50d91ab5e19faa554cf827...92e91d515a3234c03cf94fd7cff4693a3e61c037 You're receiving this email because of your account on foss.heptapod.net.
_______________________________________________ pypy-commit mailing list -- pypy-commit@python.org To unsubscribe send an email to pypy-commit-le...@python.org https://mail.python.org/mailman3/lists/pypy-commit.python.org/ Member address: arch...@mail-archive.com
