Skip to content

feat(GFX1100-TG200): lever-C fused norm-quant epilogue #2791

Description

@ghazni101

Row: `GFX1100-TG200`

Lever-C adds an opt-in fused norm-quant epilogue (`VT_NORM_QUANT_FUSED=1`): `RmsNormRowKernel` emits the row's Q8_K blocks alongside its normal output, and `MatmulBTQuant`'s K-quant branch skips the standalone `QuantizeQ8KK` when the consuming activation matches the producer token. Byte-identical to the standalone path by construction (shared `QuantQ8KSBlock` body).

New files: `rocm_act_quant.h` (shared Q8_K quant-block body), `rocm_norm_quant_bridge.h` (producer-consumer token contract).

Depends on #2782 (F1) and #2790 (T4a).

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions