mirror of
https://github.com/odin-lang/Odin.git
synced 2026-09-02 10:13:35 +00:00
55da0990e7
The tests said one thing and the table another: WHILELT and the three SVE scatter stores. The tests were right and had been right for a while -- the table edit that went with them was never committed, and the branch looked green only because the tests read a stale generated blob that still carried the fix. Regenerating the blobs (which merging upstream forced, since its 35 new mnemonics overflowed the old ones) made the disagreement visible. WHILELT mask 0xFF20FC10 -> 0xFFE0FC10 ST1B scatter Zt.S 0xE4008000 -> 0xE4408000, mask 0xFFA0E000 -> 0xFFE0A000 ST1W scatter Zt.S 0xE5008000 -> 0xE5408000, mask 0xFFA0E000 -> 0xFFE0A000 ST1D scatter Zt.D bits unchanged, mask 0xFFA0E000 -> 0xFFE0A000 ST1B and ST1W carried the SAME bits for their Zt.S and Zt.D forms, so the 32-bit scatter was encoding as the 64-bit one; the S rows now differ from the D rows beside them, which is what tells them apart. Values taken from the committed test expectations and independently corroborated by the pre-regeneration blob -- two sources that agree. arm64: 330 passed 0 failed, 133 passed 0 failed. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Riok9vMpkLmo78wsVKJHhz