From e538cd68153b1a259ac935e6c06a74c46e77196f Mon Sep 17 00:00:00 2001 From: bong-water-water-bong Date: Thu, 12 Nov 2026 19:23:45 -0400 Subject: [PATCH] llama.cpp: pin e6c80014e, a fused HIP matcher can claim the node it is rooted at (engine #679) e7796262e -> e6c80014e is one commit, llama.cpp #332: hip_declare_tensor_claim / hip_tensor_claimed, consulted by device_supports_op. With the private add-on (gpu-kernels #69) the QSA indexer's f16 gather now reaches hip.qsa_pool_blocks in Flash-Next decode. At 18.8k: 74 -> 50 splits per step, 0 tensors copied into CPU splits (12 x 4.85 MB before), greedy text identical. Without an add-on registration, every supports_op answer is unchanged. Pinned tree's own suite (gfx1151, e6c80014e, no add-on): hrx-backend-test host 85/85, device 13/13. Co-Authored-By: Claude Opus 5.5 --- registry/architectures.json | 2 +- third_party/llama.cpp | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/registry/architectures.json b/registry/architectures.json index 08f1c1d70..6a66bbb11 100644 --- a/registry/architectures.json +++ b/registry/architectures.json @@ -1,7 +1,7 @@ { "about": "HF architecture -> GGUF architecture and the backends whose code accepts it. Generated by tools/registry_build.py from the pinned sources; do not edit.", "sources": { - "llama.cpp (hrx)": "e7796262e2c4639eddb7a9a65f5a4abdc9797f71" + "llama.cpp (hrx)": "e6c80014eb1c00bd4c1900d18379752227303fbb" }, "counts": { "hrx": 298, diff --git a/third_party/llama.cpp b/third_party/llama.cpp index e7796262e..e6c80014e 160000 --- a/third_party/llama.cpp +++ b/third_party/llama.cpp @@ -1 +1 @@ -Subproject commit e7796262e2c4639eddb7a9a65f5a4abdc9797f71 +Subproject commit e6c80014eb1c00bd4c1900d18379752227303fbb