pub const DECODE_GQA_PACK_WIDTH: u32 = 6;Expand description
2026-09-25: Query heads one packed CTA carries: the Rust copy of the
kernels’ #define PD_GQA, which sizes their q_reg/o_reg register
arrays at compile time. cuda_sources_declare_the_pack_width_rust_dispatches_on
fails if the two differ.
6 is Qwen3.8-27B’s ratio (q_heads = 24, kv_heads = 4 in
kernels/hopper/qwen3.8-27b/MODEL.toml). gqa_pack_shape_ok refuses any
other ratio, which keeps the unpacked kernel.