mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 32 Packages Projects Releases Wiki Activity

vulkan: fix non-contig rope (#19299)

Browse Source
This commit is contained in:
Jeff Bolz
2026-02-05 01:38:59 -06:00
committed by GitHub
parent af252d0758
commit c342c3b93d
8 changed files with 100 additions and 95 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-vulkan/ggml-vulkan.cpp
+24 -8
View File
@@ -1263,25 +1263,30 @@ struct vk_op_diag_mask_push_constants {
struct vk_op_rope_push_constants {
uint32_t rope_mode;
uint32_t ncols;
uint32_t nrows;
uint32_t n_dims;
float freq_scale;
uint32_t p_delta_rows;
float freq_base;
float ext_factor;
float attn_factor;
float corr_dims[2];
float theta_scale;
uint32_t has_ff;
uint32_t ne02;
uint32_t s1;
uint32_t s2;
int32_t sections[4];
uint32_t is_imrope;
uint32_t is_back;
uint32_t set_rows_stride;
uint32_t ne00;
uint32_t ne01;
uint32_t ne02;
uint32_t nb01;
uint32_t nb02;
uint32_t nb03;
uint32_t nb11;
uint32_t nb12;
uint32_t nb13;
};
static_assert(sizeof(vk_op_rope_push_constants) <= 128, "sizeof(vk_op_rope_push_constants) must be <= 128");
// For fused rms_norm+mul+rope(+view+set_rows)
struct vk_op_rms_norm_mul_rope_push_constants {
@@ -10405,12 +10410,22 @@ static vk_op_rope_push_constants ggml_vk_make_rope_constants(const ggml_tensor *
uint32_t nb01 = src0->nb[1] / ggml_type_size(src0->type);
uint32_t nb02 = src0->nb[2] / ggml_type_size(src0->type);
uint32_t nb03 = src0->nb[3] / ggml_type_size(src0->type);
uint32_t nb11 = dst->nb[1] / ggml_type_size(dst->type);
uint32_t nb12 = dst->nb[2] / ggml_type_size(dst->type);
uint32_t nb13 = dst->nb[3] / ggml_type_size(dst->type);
vk_op_rope_push_constants rope {
(uint32_t)mode, (uint32_t)src0->ne[0], (uint32_t)ggml_nrows(src0), (uint32_t)n_dims, freq_scale, (uint32_t)src0->ne[1],
freq_base, ext_factor, attn_factor, {corr_dims[0], corr_dims[1]}, theta_scale,
has_ff, (uint32_t)src0->ne[2], nb01, nb02,
(uint32_t)mode, (uint32_t)ggml_nrows(src0), (uint32_t)n_dims, freq_scale,
freq_base, ext_factor, attn_factor, {corr_dims[0], corr_dims[1]}, theta_scale, has_ff,
{ sections[0], sections[1], sections[2], sections[3] }, is_imrope, backprop, set_rows_stride,
(uint32_t)src0->ne[0],
(uint32_t)src0->ne[1],
(uint32_t)src0->ne[2],
nb01, nb02, nb03,
nb11, nb12, nb13,
};
return rope;
@@ -14798,6 +14813,7 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
case GGML_OP_REPEAT_BACK:
return op->type == GGML_TYPE_F32 && op->src[0]->type == GGML_TYPE_F32;
case GGML_OP_ROPE:
return ggml_is_contiguous_rows(op) && ggml_is_contiguous_rows(op->src[0]);
case GGML_OP_ROPE_BACK:
case GGML_OP_NONE:
case GGML_OP_RESHAPE: