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

change to only use graphics queue if overridden with env var GGML_VK_ALLOW_GRAPHICS_QUEUE

This commit is contained in:
Ruben Ortlam 2026-03-15 20:34:43 +01:00
parent c68b2428b3
commit 824aa149ca
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -4981,9 +4981,9 @@ static vk_device ggml_vk_get_device(size_t idx) {
std::vector<vk::QueueFamilyProperties> queue_family_props = device->physical_device.getQueueFamilyProperties();
// Try to find a non-graphics compute queue and transfer-focused queues
// On AMD GPUs with RADV driver, the graphics queue seems to be faster, so don't avoid it
// Avoid on small GPUs to prevent interfering with graphics tasks
const vk::QueueFlagBits graphics_flag = (device->vendor_id == VK_VENDOR_ID_AMD && device->driver_id == vk::DriverId::eMesaRadv && device->shader_core_count > 16) ? (vk::QueueFlagBits)0 : vk::QueueFlagBits::eGraphics;
// Allow overriding avoiding the graphics queue because it can increase performance on RADV
const bool allow_graphics_queue = (getenv("GGML_VK_ALLOW_GRAPHICS_QUEUE") != nullptr);
const vk::QueueFlagBits graphics_flag = allow_graphics_queue ? (vk::QueueFlagBits)0 : vk::QueueFlagBits::eGraphics;
const uint32_t compute_queue_family_index = ggml_vk_find_queue_family_index(queue_family_props, vk::QueueFlagBits::eCompute, graphics_flag, -1, 1);
const uint32_t transfer_queue_family_index = ggml_vk_find_queue_family_index(queue_family_props, vk::QueueFlagBits::eTransfer, vk::QueueFlagBits::eCompute | graphics_flag, compute_queue_family_index, 1);