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

vulkan: Check maxStorageBufferRange in supports_op (#18709) 

* vulkan: Check maxStorageBufferRange in supports_op

* skip maxStorageBufferRange check when shader64BitIndexing is enabled
This commit is contained in:
Jeff Bolz 2026-01-14 03:59:05 -06:00 committed by GitHub
parent 47f9612492
commit 3e4bb29666
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 18 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -14413,13 +14413,29 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)dev->context; ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)dev->context;
const vk_device& device = ggml_vk_get_device(ctx->device); const vk_device& device = ggml_vk_get_device(ctx->device);
const bool uses_bda = (op->op == GGML_OP_IM2COL || op->op == GGML_OP_IM2COL_3D) &&
device->shader_int64 && device->buffer_device_address;
auto const & tensor_size_supported = [&](size_t tensor_size) {
if (tensor_size > device->max_buffer_size) {
return false;
}
// For im2col shaders using BDA, maxStorageBufferRange limit doesn't apply.
// If shader64BitIndexing is enabled, maxStorageBufferRange limit doesn't apply.
if (!uses_bda && !device->shader_64b_indexing) {
if (tensor_size > device->properties.limits.maxStorageBufferRange) {
return false;
}
}
return true;
};
// reject any tensors larger than the max buffer size // reject any tensors larger than the max buffer size
for (int i = 0; i < GGML_MAX_SRC; i++) { for (int i = 0; i < GGML_MAX_SRC; i++) {
if (op->src[i] && ggml_nbytes(op->src[i]) > device->max_buffer_size) { if (op->src[i] && !tensor_size_supported(ggml_nbytes(op->src[i]))) {
return false; return false;
} }
} }
if (ggml_nbytes(op) > device->max_buffer_size) { if (!tensor_size_supported(ggml_nbytes(op))) {
return false; return false;
} }