| commit | e120ace48d8ff312e9eb66a2d6cfbf1920047d38 | |
| author | Brett Simmers <bsimmers@fb.com> | |
| Wed, 17 Dec 2014 11:14:38 +0000 (17 03:14 -0800) | ||
| committer | hhvm-bot <hhvm-bot@fb.com> | |
| Wed, 17 Dec 2014 11:30:33 +0000 (17 03:30 -0800) | ||
| tree | ad572e22871e4b8b10fcf22f72ba60ad91dccbc7 | treesnapshot (tar.gz zip) |
| parent | 6ccf272a9d098bdb9544b40803e0b42c5c741dfe | commitdiff |
Fix some repo-mode llvm tests
Summary: - trflags is sometimes nonzero for fallback, so properly implement it
- The comment in ratToAssertType about already knowing Cell and Ref in the jit
is only true for stack slots, not locals. Losing this type information was
giving us a Gen base for a minstr instruction, upsetting one of the
assertions. It was llvm-only because we change the length and shape of
tracelets by forcing a break after FCall, and this kind of bug tends to
depend on that.
Reviewed By: @jasone
Differential Revision: D1742487
Summary: - trflags is sometimes nonzero for fallback, so properly implement it
- The comment in ratToAssertType about already knowing Cell and Ref in the jit
is only true for stack slots, not locals. Losing this type information was
giving us a Gen base for a minstr instruction, upsetting one of the
assertions. It was llvm-only because we change the length and shape of
tracelets by forcing a break after FCall, and this kind of bug tends to
depend on that.
Reviewed By: @jasone
Differential Revision: D1742487
| hphp/runtime/vm/jit/irgen-types.cpp | diffblobblamehistory | |
| hphp/runtime/vm/jit/srcdb.cpp | diffblobblamehistory | |
| hphp/runtime/vm/jit/vasm-llvm.cpp | diffblobblamehistory |