use subgroup min and max to check for gcn (requires https://github.com/ggerganov/llama.cpp/pull/10721)
This commit is contained in:
parent
c7bc42cea0
commit
9af9e80163
1 changed files with 1 additions and 1 deletions
|
@ -1812,7 +1812,7 @@ static void ggml_vk_load_shaders(vk_device& device) {
|
|||
|
||||
// AMD GCN graphics cards perform best when the number of rows per shader is doubled
|
||||
uint32_t rm = 1;
|
||||
if ((device->vendor_id == VK_VENDOR_ID_AMD) && (device->subgroup_size == 64))
|
||||
if ((device->subgroup_size_control) && (device->vendor_id == VK_VENDOR_ID_AMD) && (device->subgroup_min_size == 64) && (device->subgroup_max_size == 64))
|
||||
rm = 2;
|
||||
|
||||
// computing additional rows per workgroup is a benefit for Q4_0 -> Q5_1, but not for Q8_0.
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue