marfrit/rocky-vulkan-llama
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

Commit Graph

  • 636fa430cd Champion shader ported to July build (29x over stock GPU) BUT ties CPU prefill / loses decode -> no faster-than-CPU win on Mali (VA squeeze). Honest correction. master Markus Fritsche 2026-07-12 07:05:03 +02:00
  • a6467cc359 RESULTS: gemma4 renders CORRECTLY on Mali GPU (panvk, plain 7.0.0-rc3) -> ' Paris.'; stock ~1 t/s prefill (needs champion shader). Correctness gate PASSED. Markus Fritsche 2026-07-12 06:28:14 +02:00
  • d32c653c9a Kernel bring-up: A1 hangs panvk compute, plain 7.0.0-rc3 runs it; rebuild-vs-A1 disproven (uABI byte-identical); both panthor 1.7.0 Markus Fritsche 2026-07-12 06:17:47 +02:00
  • 107ccddbb1 Clean up for public release: README rewrite, LICENSE (MIT), patches reorganized Markus Fritsche 2026-07-11 20:35:44 +02:00
  • 5b541d596e Shader CHAMPION converged: 28 t/s (4.8x), correctness-gated; f16acc refused; ~15% left is quantize-dispatch (profile, not tile) Claude (noether) 2026-07-11 19:11:46 +02:00
  • 6ee32814b2 Shader iter6: BK_STEP=1 regressed (barrier crossover); BK_STEP=2 is sweet spot, reverted to iter5 champion Claude (noether) 2026-07-11 18:36:21 +02:00
  • 220a3b1667 Shader iter5 CHAMPION: scoped BK_STEP (spec const 11) -> L=28 t/s, M restored to 20.6 (clean) Claude (noether) 2026-07-11 18:27:38 +02:00
  • 1e23ce0cd2 Shader iter4: BK_STEP=2 -> L path 28 t/s (+9%), but M regressed (global #define); needs L-scoping Claude (noether) 2026-07-11 18:15:53 +02:00
  • 05f240ef6c Shader iter3 CHAMPION: BLOCK 256 same-registers -> ub128 25.8 t/s (+10%), spill-free Claude (noether) 2026-07-11 17:58:10 +02:00
  • 2600465c3c Shader iter2: 32-acc tile REGRESSED 2.4x -> occupancy-bound, not AI-bound; iter1 champion Claude (noether) 2026-07-11 17:44:52 +02:00
  • 272604b10d Shader iter1: Mali L-warptile (WM=WN=16) -> ub128 4x (5.8->23.4 t/s), spill 5328->96 Claude (noether) 2026-07-11 17:29:55 +02:00
  • dbf97a0fc1 FIX: panvk priv-heap exhaustion solved (shared TLS + WLS reuse); crash gone ub64->1024 Claude (noether) 2026-07-11 16:30:55 +02:00
  • bec204e375 Phase 1: consumer-side levers falsified; accumulation is in-flight cmdbufs x 304MB TLS Claude (noether) 2026-07-11 16:00:09 +02:00
  • 881b118158 Add fix plan: cleanup_frequency=10 is the accumulation trigger; consumer-side bounding is the primary lever Claude (noether) 2026-07-11 14:58:50 +02:00
  • 97629e935c Fully characterize the ub64 cliff: TLS-stack register spill is the dominant term Claude (noether) 2026-07-11 14:31:27 +02:00
  • e74bae69fe panvk WLS fix v1 + confirmed mechanism (partial: reuse halves but doesn't eliminate) Claude (noether) 2026-07-11 14:06:47 +02:00
  • 16b96537fd rocky-vulkan-llama: skeleton — deps, root-cause diagnosis, llama.cpp submit patch Claude (noether) 2026-07-11 13:33:00 +02:00