From 788fcbc5ddb16101e6388170bf84c0d10bd8c006 Mon Sep 17 00:00:00 2001 From: Katostrofik Date: Mon, 20 Apr 2026 01:39:45 -0400 Subject: [PATCH] [SYCL] Fix reorder MMVQ assert on unaligned vocab sizes (#22035) * [SYCL] Fix reorder MMVQ assert on unaligned vocab sizes The reorder mul_mat_vec_q dispatchers for Q4_0, Q8_0, Q4_K, and Q6_K asserted that block_num_y was a multiple of 16 subgroups. Models with a vocab size not divisible by 16 (for example HY-MT at 120818) aborted on model load when the output projection tripped the assert. I replaced the assert with padding: block_num_y now rounds up to a whole number of subgroup-sized workgroups. The kernel already has the row bounds check (`if (row >= nrows) return;`) so the extra padded threads early-exit cleanly. Row values are uniform across a subgroup so the collective reduce stays safe. For aligned vocab sizes the padded block_num_y equals the old value, so the kernel launch is identical and there is no regression. Thanks to @arthw for flagging the relationship to #21527. Fixes #22020. AI assisted coding, tested on Intel B70 hardware. * sycl: use WARP_SIZE for num_subgroups in reorder MMVQ launches Replaces the hardcoded 16 with WARP_SIZE in the four reorder_mul_mat_vec launch helpers (Q4_0, Q8_0, Q4_K, Q6_K). Compile-time no-op on the Intel target where WARP_SIZE is 16, but makes the relationship to subgroup size explicit. Per review by @NeoZhangJianyu on #22035. Assisted by Claude. --- ggml/src/ggml-sycl/mmvq.cpp | 24 ++++++++++++------------ 1 file changed, 12 insertions(+), 12 deletions(-) diff --git a/ggml/src/ggml-sycl/mmvq.cpp b/ggml/src/ggml-sycl/mmvq.cpp index af22b98ddd..3a4577ecbb 100644 --- a/ggml/src/ggml-sycl/mmvq.cpp +++ b/ggml/src/ggml-sycl/mmvq.cpp @@ -537,9 +537,9 @@ static void mul_mat_vec_q_iq4_xs_q8_1(const void *__restrict__ vx, static void reorder_mul_mat_vec_q4_0_q8_1_sycl(const void * vx, const void * vy, float * dst, const int ncols, const int nrows, dpct::queue_ptr stream) { GGML_ASSERT(ncols % QK4_0 == 0); - const int block_num_y = ceil_div(nrows, GGML_SYCL_MMV_Y); - constexpr size_t num_subgroups = 16; - GGML_ASSERT(block_num_y % num_subgroups == 0); + // Round up to a whole number of subgroup-sized workgroups; out-of-range rows are skipped inside the kernel. + constexpr size_t num_subgroups = WARP_SIZE; + const int block_num_y = ceil_div(nrows, GGML_SYCL_MMV_Y * (int) num_subgroups) * (int) num_subgroups; const sycl::range<3> global_size(1, GGML_SYCL_MMV_Y, (block_num_y * WARP_SIZE)); const sycl::range<3> workgroup_size(1, GGML_SYCL_MMV_Y, num_subgroups * WARP_SIZE); @@ -682,9 +682,9 @@ static void mul_mat_vec_q5_1_q8_1_sycl(const void *vx, const void *vy, static void reorder_mul_mat_vec_q8_0_q8_1_sycl(const void * vx, const void * vy, float * dst, const int ncols, const int nrows, dpct::queue_ptr stream) { GGML_ASSERT(ncols % QK8_0 == 0); - const int block_num_y = ceil_div(nrows, GGML_SYCL_MMV_Y); - constexpr size_t num_subgroups = 16; - GGML_ASSERT(block_num_y % num_subgroups == 0); + // Round up to a whole number of subgroup-sized workgroups; out-of-range rows are skipped inside the kernel. + constexpr size_t num_subgroups = WARP_SIZE; + const int block_num_y = ceil_div(nrows, GGML_SYCL_MMV_Y * (int) num_subgroups) * (int) num_subgroups; const sycl::range<3> global_size(1, GGML_SYCL_MMV_Y, (block_num_y * WARP_SIZE)); const sycl::range<3> workgroup_size(1, GGML_SYCL_MMV_Y, num_subgroups * WARP_SIZE); @@ -798,9 +798,9 @@ static void reorder_mul_mat_vec_q4_k_q8_1_sycl(const void * vx, const void * vy, const int nrows, dpct::queue_ptr stream) { GGML_ASSERT(ncols % QK_K == 0); - const int block_num_y = ceil_div(nrows, GGML_SYCL_MMV_Y); - constexpr size_t num_subgroups = 16; - GGML_ASSERT(block_num_y % num_subgroups == 0); + // Round up to a whole number of subgroup-sized workgroups; out-of-range rows are skipped inside the kernel. + constexpr size_t num_subgroups = WARP_SIZE; + const int block_num_y = ceil_div(nrows, GGML_SYCL_MMV_Y * (int) num_subgroups) * (int) num_subgroups; const sycl::range<3> global_size(1, GGML_SYCL_MMV_Y, block_num_y * WARP_SIZE); const sycl::range<3> workgroup_size(1, GGML_SYCL_MMV_Y, num_subgroups * WARP_SIZE); @@ -842,9 +842,9 @@ static void mul_mat_vec_q5_K_q8_1_sycl(const void *vx, const void *vy, static void reorder_mul_mat_vec_q6_k_q8_1_sycl(const void * vx, const void * vy, float * dst, const int ncols, const int nrows, dpct::queue_ptr stream) { GGML_ASSERT(ncols % QK_K == 0); - const int block_num_y = ceil_div(nrows, GGML_SYCL_MMV_Y); - constexpr size_t num_subgroups = 16; - GGML_ASSERT(block_num_y % num_subgroups == 0); + // Round up to a whole number of subgroup-sized workgroups; out-of-range rows are skipped inside the kernel. + constexpr size_t num_subgroups = WARP_SIZE; + const int block_num_y = ceil_div(nrows, GGML_SYCL_MMV_Y * (int) num_subgroups) * (int) num_subgroups; const sycl::range<3> global_size(1, GGML_SYCL_MMV_Y, block_num_y * WARP_SIZE); const sycl::range<3> workgroup_size(1, GGML_SYCL_MMV_Y, num_subgroups * WARP_SIZE);