Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[SYCL] Fix Q8_0 reorder: garbage on 2nd prompt + crash on full VRAM (#21638)

Browse Source 
* [SYCL] Fix Q8_0 reorder: add missing dequantize path for GEMM

The Q8_0 reorder optimization (#21527) was missing a reorder-aware
dequantizer for the GEMM code path used during prompt processing.
After token generation reordered Q8_0 weights (via DMMV/MMVQ), the
next prompt processing pass would read them with the standard
dequantizer, producing garbage output.

Add dequantize_block_q8_0_reorder() and wire it into both
ggml_get_to_fp16_sycl() and ggml_get_to_fp32_sycl(), matching the
pattern already used by Q4_0, Q4_K, and Q6_K.

Fixes #21589

AI (Claude) was used to assist with root cause investigation and
writing the kernel code. All code was human-reviewed and tested
on real hardware.

* SYCL: fix reorder crash when device memory is full

The reorder optimization allocates a temporary buffer the full size of
the weight tensor on the device. When VRAM is nearly full (large models
on a single GPU), this allocation fails and the subsequent memcpy crashes
on a NULL pointer.

Fix: try device allocation first, fall back to host memory if device
memory is full. The reorder kernel still works correctly reading from
host memory over PCIe. This is slower for the one-time reorder (~21 t/s
vs ~38 t/s on Intel Arc Pro B70), but the optimization is preserved for
all subsequent inference. If both device and host allocation fail, skip
the reorder and fall back to the unoptimized kernel path.

Also fixes a bug where opt_for_reorder() marked tensors as reordered
even when the reorder was skipped due to allocation failure. This caused
DMMV/MMVQ kernels to read the original AoS data as if it were SoA,
producing garbage output or NaN results.

Tested on Intel Arc Pro B70 (32GB) with Q8_0, Q4_K_M models. Coding was
AI-assisted (Claude), reviewed and tested on hardware by a human.

Fixes #20478

* SYCL: add RAII temp buffer class + macro guard for host fallback

Replace sycl_ext_malloc_with_fallback/sycl_ext_free_fallback free
functions with sycl_reorder_temp_buffer RAII class. The host_fallback
bool is now a private member, and cleanup happens automatically at
scope exit.

Add GGML_SYCL_HOST_MEM_FALLBACK cmake option (default ON) to guard
the host memory fallback code path. Device access to host memory
requires Linux kernel 6.8+ (Ubuntu 26.04+); users on older kernels
can set -DGGML_SYCL_HOST_MEM_FALLBACK=OFF to disable it.

Addresses arthw's review on PR #21638.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* SYCL: document GGML_SYCL_HOST_MEM_FALLBACK build option in SYCL.md

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* SYCL: add reorder-aware DMMV dequantizers for Q4_K and Q6_K

Q4_K and Q6_K had reorder support for MMVQ and GEMM paths but not
DMMV. When the DMMV path encountered reordered data it would abort.

Add DMMV kernels that read from the SOA reorder layout for both
types. Same math as the non-reorder versions, different memory
access pattern.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Katostrofik
2026-04-16 08:34:05 +03:00
committed by GitHub
co-authored by Claude Opus 4.6
parent 408225bb1a
commit b1be68e8ca
7 changed files with 466 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-sycl/convert.cpp
+31 -2
View File
@@ -151,6 +151,25 @@ static void dequantize_row_q4_0_sycl_reorder(const void *vx, dst_t *y, const int
}
template <typename dst_t>
static void dequantize_row_q8_0_sycl_reorder(const void *vx, dst_t *y, const int64_t k,
dpct::queue_ptr stream) {
dpct::has_capability_or_fail(stream->get_device(),
{sycl::aspect::fp16});
int constexpr WARP_K = WARP_SIZE * QK8_0;
const int n_warp = (k + WARP_K - 1) / WARP_K;
GGML_ASSERT(k % QK8_0 == 0);
stream->parallel_for(sycl::nd_range<3>(sycl::range<3>(1, 1, n_warp) *
sycl::range<3>(1, 1, WARP_SIZE),
sycl::range<3>(1, 1, WARP_SIZE)),
[=](sycl::nd_item<3> item_ct1) [[sycl::reqd_sub_group_size(WARP_SIZE)]]{
dequantize_block_q8_0_reorder(vx, y, k, item_ct1);
});
}
template <typename dst_t>
static void dequantize_row_q4_1_sycl(const void *vx, dst_t *y, const int64_t k,
dpct::queue_ptr stream) {
@@ -614,7 +633,12 @@ to_fp16_sycl_t ggml_get_to_fp16_sycl(ggml_type type, ggml_tensor * dst) {
case GGML_TYPE_Q5_1:
return dequantize_block_sycl<QK5_1, QR5_1, dequantize_q5_1>;
case GGML_TYPE_Q8_0:
return dequantize_block_sycl<QK8_0, QR8_0, dequantize_q8_0>;
if (dst->src[0]->extra &&
((ggml_tensor_extra_gpu *) dst->src[0]->extra)->optimized_feature.reorder) {
return dequantize_row_q8_0_sycl_reorder;
} else {
return dequantize_block_sycl<QK8_0, QR8_0, dequantize_q8_0>;
}
case GGML_TYPE_Q2_K:
return dequantize_row_q2_K_sycl;
case GGML_TYPE_Q3_K:
@@ -683,7 +707,12 @@ to_fp32_sycl_t ggml_get_to_fp32_sycl(ggml_type type, ggml_tensor *dst) {
case GGML_TYPE_Q5_1:
return dequantize_block_sycl<QK5_1, QR5_1, dequantize_q5_1>;
case GGML_TYPE_Q8_0:
return dequantize_block_sycl<QK8_0, QR8_0, dequantize_q8_0>;
if (dst->src[0]->extra &&
((ggml_tensor_extra_gpu*)dst->src[0]->extra)->optimized_feature.reorder) {
return dequantize_row_q8_0_sycl_reorder;
} else {
return dequantize_block_sycl<QK8_0, QR8_0, dequantize_q8_0>;
}
case GGML_TYPE_Q2_K:
return dequantize_row_q2_K_sycl;
case GGML_TYPE_Q3_K: