| Auto-Submit | +1 |
| Commit-Queue | +1 |
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
if (omitted_backedge && new_block->state()->is_resumable_loop()) {Just `new_block->state()->is_resumable_loop()` should work, no? (and revert the rest of the change)
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
| Auto-Submit | +1 |
Done, reverted the rest.
Confirmed it's equivalent: is_resumable_loop() already implies is_loop(), and I ran mjsunit with a temporary DCHECK_EQ(omitted_backedge, new_block->is_loop()) here, which held over 8848 tests — so the extra condition was indeed redundant.
if (omitted_backedge && new_block->state()->is_resumable_loop()) {Just `new_block->state()->is_resumable_loop()` should work, no? (and revert the rest of the change)
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
| Code-Review | +1 |
| Commit-Queue | +2 |
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
[maglev] Don't merge phi taggings across a resumable loop header
The phi-untagging snapshot table closes a loop-header merge phi by
making it its own backedge input, which assumes the header dominates its
backedge predecessor. A resumable loop is entered through the backedge
when the generator resumes, so the self-reference does not dominate;
report no tagging there instead and let it be recreated where needed.
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |