mirror of
https://github.com/LostRuins/koboldcpp.git
synced 2026-08-10 08:55:44 +00:00
d6b61ac0d3
* sycl: fix use-after-return of the SDPA scale in the oneDNN flash-attention path The scale was uploaded with an async memcpy sourced from a stack local. On the in-order queue that copy is ordered behind the K/V staging kernels; once n_kv is large enough (>= ~26k observed on Arc Pro B70) the staging outlives the host stack frame and the copy reads recycled memory, feeding the SDPA a garbage scale. Output then collapses to a single repeated token and the KV cache is poisoned for the rest of the session. Short contexts win the race by accident, and test-backend-ops caps FLASH_ATTN_EXT at kv=1024, which is why CI never caught it. The previous device_count > 1 wait_and_throw() gate (and reverting it, PR #25741) fixes the symptom only by keeping the frame alive across the copy at the cost of a host sync on every FA call. Fix: cache one device scalar per (device, value) -- the scale is constant per model -- and upload it synchronously once. The single-device fast path (no per-call host sync) is then safe: every device-side hazard already serializes on the in-order queue. The multi-GPU conservative wait is kept unchanged. Also: - GGML_SYCL_FA_ONEDNN_MAX_KV env (0 = unlimited): optional n_kv ceiling that routes very long sequences to the native FA kernel. - test-backend-ops: FLASH_ATTN_EXT F16 cases up to kv=65536 (Qwen3.6-27B geometry hsk=hsv=256 GQA 6, and hsk=128 GQA 4), closing the kv=1024 blind spot. Note the race itself needs a live multi-op pipeline to reproduce; single-op runs pass even on broken builds. Verified on Arc Pro B70 (bmg_g31), Qwen3.6-27B Q4_K, -c 131072: output byte-identical at temp 0 to the native FA path through 32k-deep prefill, with prefill depth-flat at 820-840 t/s (vs 340-350 native at 32k depth). Assisted-by: Claude Fable 5 * sycl: handle GGML_SYCL_FA_ONEDNN_MAX_KV like the other runtime env vars and document it Review feedback on #25880: - read the variable once at backend init into g_ggml_sycl_fa_onednn_max_kv via ggml_sycl_get_env, and print it in the startup env listing (-lv 4 shows it) - document GGML_SYCL_FA_ONEDNN and GGML_SYCL_FA_ONEDNN_MAX_KV in the SYCL.md runtime table Also trim the added FLASH_ATTN_EXT cases to kv={4096,16384}: the 32768/65536 shapes exceed the legacy NMSE threshold on both the oneDNN and native kernels (long-sequence fp16 accumulation drift, present before this PR) and would fail CI for an unrelated reason. Assisted-by: Claude Fable 5 * sycl: clarify GGML_SYCL_FA_ONEDNN_MAX_KV default is disabled Assisted-by: Claude Fable 5 * sycl: state default behavior of GGML_SYCL_FA_ONEDNN_MAX_KV explicitly Assisted-by: Claude Fable 5 * Update ggml/src/ggml-sycl/fattn-onednn.cpp Co-authored-by: Neo Zhang <zhang.jianyu@outlook.com> * sycl: write the SDPA scale from a kernel instead of caching it The per-(device, value) scale cache was a function-local static unordered_map with no synchronization, so concurrent backend instances could access and rehash it at the same time. Write the scalar with a single_task instead. The value is captured into the command, so no host memory has to outlive the call -- which is what the use-after-return fix needed in the first place. That removes the shared container, the leaked device allocation and the string key, and it also closes the remaining async-memcpy-from-a-stack-local on the first flash-attention call. Ordering does not rely on timing: the queue is created with sycl::property::queue::in_order and the dnnl stream wraps that same queue, so the write completes before the SDPA reads the scalar. The multi-GPU wait_and_throw() branch is unchanged. Also drop the <cstdlib> include, which is unused. Assisted-by: Claude Opus 5 --------- Co-authored-by: Neo Zhang <zhang.jianyu@outlook.com> |
||
|---|---|---|
| .. | ||
| peg-parser | ||
| snapshots | ||
| .gitignore | ||
| CMakeLists.txt | ||
| get-model.cpp | ||
| get-model.h | ||
| gguf-model-data.cpp | ||
| gguf-model-data.h | ||
| test-alloc.cpp | ||
| test-arg-parser.cpp | ||
| test-autorelease.cpp | ||
| test-backend-ops.cpp | ||
| test-backend-sampler.cpp | ||
| test-barrier.cpp | ||
| test-batch-alloc.cpp | ||
| test-c.c | ||
| test-chat-auto-parser.cpp | ||
| test-chat-peg-parser.cpp | ||
| test-chat-template.cpp | ||
| test-chat.cpp | ||
| test-col2im-1d.cpp | ||
| test-double-float.cpp | ||
| test-export-graph-ops.cpp | ||
| test-gbnf-validator.cpp | ||
| test-gguf-model-data.cpp | ||
| test-gguf.cpp | ||
| test-grammar-integration.cpp | ||
| test-grammar-llguidance.cpp | ||
| test-grammar-parser.cpp | ||
| test-jinja.cpp | ||
| test-json-schema-to-grammar.cpp | ||
| test-llama-archs.cpp | ||
| test-llama-grammar.cpp | ||
| test-log.cpp | ||
| test-lora-conversion-inference.sh | ||
| test-model-load-cancel.cpp | ||
| test-mtmd-c-api.c | ||
| test-opt.cpp | ||
| test-peg-parser.cpp | ||
| test-quant-type-selection.cpp | ||
| test-quantize-fns.cpp | ||
| test-quantize-perf.cpp | ||
| test-quantize-stats.cpp | ||
| test-reasoning-budget.cpp | ||
| test-recurrent-state-rollback.cpp | ||
| test-rope.cpp | ||
| test-sampling.cpp | ||
| test-save-load-state.cpp | ||
| test-state-restore-fragmented.cpp | ||
| test-thread-safety.cpp | ||
| test-tokenizer-0.cpp | ||
| test-tokenizer-0.py | ||
| test-tokenizer-0.sh | ||
| test-tokenizer-1-bpe.cpp | ||
| test-tokenizer-1-spm.cpp | ||
| test-tokenizer-random.py | ||
| test-tokenizers-repo.sh | ||
| testing.h | ||