happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

vulkan: use a fixed 1KB buffer for the add_rms_fusion opt (#17514)

This commit is contained in:
Jeff Bolz 2025-11-26 23:32:30 -06:00 committed by GitHub
parent e509411cf1
commit 142df17c9c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -5289,7 +5289,8 @@ static void ggml_vk_init(ggml_backend_vk_context * ctx, size_t idx) {
ctx->prealloc_size_x = 0;
ctx->prealloc_size_y = 0;
ctx->prealloc_size_split_k = 0;
ctx->prealloc_size_add_rms_partials = 0;
// Fixed size of 1KB, for deterministic behavior
ctx->prealloc_size_add_rms_partials = 1024;
ctx->fence = ctx->device->device.createFence({});
ctx->almost_ready_fence = ctx->device->device.createFence({});
@ -13095,7 +13096,6 @@ static ggml_status ggml_backend_vk_graph_compute(ggml_backend_t backend, ggml_cg
ctx->fused_ops_write_mask = 0;
}
ctx->prealloc_size_add_rms_partials = std::max(ctx->prealloc_size_add_rms_partials, ctx->prealloc_size_add_rms_partials_offset);
ctx->last_total_mul_mat_bytes = total_mul_mat_bytes;
if (vk_perf_logger_enabled) {