Repository navigation
HRX: pin our patched branch (IQ3_XXS, honest op claims), rebased on every bump - #29
Merged
Merged
Conversation
… on every bump - third_party/llama.cpp -> 1bit-MONSTER/llama.cpp 1bit/hrx-vulkan-patched (d2a9239f9): AMD's f1a0aca plus an IQ3_XXS matmul kernel and honest op claims, so ggml-hrx only takes nodes it can execute and the rest falls back. - bump-hrx.yml: our commits are rebased onto each new AMD pair; the previous tip is tagged patched-<sha> first, so every pinned commit stays reachable. A conflict fails the run for a hand rebase. - docs/hrx.md: "Our patches", with the Strix Halo results: test-backend-ops -b HRX0 791 OK / 0 failed; UD-Q4_K_XL, UD-Q2_K_XL and UD-IQ2_M perplexity on HRX0 now matches Vulkan0; Q4_K_M unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
bong-water-water-bong
pushed a commit
that referenced
this pull request
Sep 27, 2026
…bal barrier (#123, #140) Brings 1bit-MONSTER/llama.cpp#28 and #29: in every decode-split reduce_fused variant the barrier between the reduce's global output stores and pack_completed_q8's output loads was kernel.barrier<workgroup> (LDS only), so the next_q8 copy could be packed from stale output. It is now kernel.barrier<global> followed by the LDS barrier (5 sites, ops and qwen3_moe corpora; #29 restores the LDS fence #28 dropped and adds the missing global barrier to the undispatched standalone reduce_f32 kernels). The pin moves fa226f9 -> 00adc2b, fast-forward. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
bong-water-water-bong
added a commit
that referenced
this pull request
Sep 27, 2026
…bal barrier (#123, #140) (#176) Brings 1bit-MONSTER/llama.cpp#28 and #29: in every decode-split reduce_fused variant the barrier between the reduce's global output stores and pack_completed_q8's output loads was kernel.barrier<workgroup> (LDS only), so the next_q8 copy could be packed from stale output. It is now kernel.barrier<global> followed by the LDS barrier (5 sites, ops and qwen3_moe corpora; #29 restores the LDS fence #28 dropped and adds the missing global barrier to the undispatched standalone reduce_f32 kernels). The pin moves fa226f9 -> 00adc2b, fast-forward. Co-authored-by: bong-water-water-bong <bong-water-water-bong@1bit.gg> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This was referenced Sep 27, 2026
This was referenced Oct 7, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Moves
third_party/llama.cppfrom AMD's commit to1bit-MONSTER/llama.cpp1bit/hrx-vulkan-patched(d2a9239f9) = AMD'sf1a0aca+ 3 commits:kernels/hrx/mul_mat_vec_iq3xxs_f32.loomand matchercommon/dispatch-mul-mat-iq3-xxs.{h,cpp}(our files, full Apache notice), ported from thefeat/hrx-port-ae91949work onto AMD's current layout.unsupported HRX node, no CPU fallback). It now claims only what its dispatcher can execute (except nodes already in HRX memory, like KV views, which cannot move). AMD's IQ4_NL/IQ4_XS matmul and GET_ROWS for IQ4_XS / batched IQ3_S give wrong values, so those nodes are declined.Verified on Strix Halo (Qwen3-0.6B, perplexity 8 x 512 wikitext tokens):
test-backend-ops -b HRX0: 791 OK, 0 failed (about 1150 fail on AMD's commit).tests/serve_e2e.shpasses on hrx and vulkan; UD-Q2_K_XL serves on hrx.bump-hrx.ymlnow rebases our commits onto each new AMD pair, tagging the previous tippatched-<sha>first so every pinned commit stays reachable; a conflict fails the run. I dry-ran that logic against a local copy of the fork with a fake new AMD commit: the 3 commits replayed with authorship kept, the branch moved, the old tip stayed reachable by tag.🤖 Generated with Claude Code