vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

vulkan: fix diag_mask_inf (#11323)

Browse source 
With robustbufferaccess disabled, this shader was showing OOB stores. There
is a bounds check in the code, but the workgrouop dimensions were reversed vs
CUDA and it was running the wrong number of threads. So fix the workgroup
dimensions and disable robustness for this pipeline.
This commit is contained in:
Jeff Bolz 2025-01-23 01:01:17 -06:00 • committed by GitHub
parent 6152129d05
commit 5245729e33
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View file

@ -2012,7 +2012,7 @@ static void ggml_vk_load_shaders(vk_device& device) {
ggml_vk_create_pipeline(device, device->pipeline_leaky_relu_f32, "leaky_relu_f32", leaky_relu_f32_len, leaky_relu_f32_data, "main", 2, sizeof(vk_op_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_tanh_f32, "tanh_f32", tanh_f32_len, tanh_f32_data, "main", 2, sizeof(vk_op_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_diag_mask_inf_f32, "diag_mask_inf_f32", diag_mask_inf_f32_len, diag_mask_inf_f32_data, "main", 2, sizeof(vk_op_diag_mask_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_diag_mask_inf_f32, "diag_mask_inf_f32", diag_mask_inf_f32_len, diag_mask_inf_f32_data, "main", 2, sizeof(vk_op_diag_mask_push_constants), {1, 512, 1}, {}, 1, true);
ggml_vk_create_pipeline(device, device->pipeline_soft_max_f32, "soft_max_f32", soft_max_f32_len, soft_max_f32_data, "main", 3, sizeof(vk_op_soft_max_push_constants), {1, 1, 1}, { device->subgroup_size }, 1);
ggml_vk_create_pipeline(device, device->pipeline_soft_max_f32_wg512, "soft_max_f32_wg512", soft_max_f32_len, soft_max_f32_data, "main", 3, sizeof(vk_op_soft_max_push_constants), {1, 1, 1}, { 512 }, 1);