trunk/129e6f1d40be5e428a664436d97803f70aa747c0: [inductor] Enable MXFP8 nested-reduction test on XPU (#194864)
- PyTorch 近 90 天出现 2586 次
- PyTorch 近 90 天第 2551 次发版
- 上一次：同一天稍早 · viable/strict/1788342684: Keep saved-activation strides symbolic under dynamic shapes (#194420)
发生了什么
I reviewed this AI-assisted description and confirmed it reflects the final implementation. Summary test_rmsnorm_mxfp8_scale_swizzle_kernel_form was disabled on XPU because its helper directly called the private, SM100-only cvt_e8m0_rceil prim. This change keeps that prim and its lowering unchanged. The test helper now: uses Inline ASM HOP for the SM100 E8M0 conversion on CUDA; uses ordinary tensor bit operations fo…
摘要按规则整理自下方来源原文