trunk/36739e0856ebe768e44f9d411f535d68f51e38a5: [ROCm][Inductor][UT] Enable the FP8 scaled_mm swizzle v2 tests (#198600)
- PyTorch 近 90 天出现 1504 次
- PyTorch 近 90 天第 1485 次发版
- 上一次:同一天稍早 · ciflow/torchtitan/198697
发生了什么
Swizzled MX scales are not handled by any inductor template, so test_scaled_mm_v2_swizzle_compile checks that compile falls back to the ATen _scaled_mm_v2 kernel. That path was skipped on ROCm because SWIZZLE_32_4_4 has no hipBLASLt scale mode. gfx950 on ROCm 7.14 packs MX FP8 scales as SWIZZLE_32_8, so drop skipIfRocm and pick the layout with rocm_mx_swizzle. Older ROCm and non-gfx950 still skip. Test Plan: python…
摘要按规则整理自下方来源原文