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

clarify granularity

Browse source
This commit is contained in:
slaren 2023-12-23 13:42:18 +01:00
parent 9452d0d54b
commit 20860daee2
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml-cuda.cu
View file

@ -6669,10 +6669,10 @@ static void * ggml_cuda_pool_malloc_vmm(size_t size, size_t * actual_size) {
prop.location.id = id; prop.location.id = id;
// get the minimum allocation granularity for this device // get the minimum allocation granularity for this device
size_t granularity = 0; size_t granularity;
CU_CHECK(cuMemGetAllocationGranularity(&granularity, &prop, CU_MEM_ALLOC_GRANULARITY_MINIMUM)); CU_CHECK(cuMemGetAllocationGranularity(&granularity, &prop, CU_MEM_ALLOC_GRANULARITY_MINIMUM));
// round up to the nearest granularity // round up to the next multiple of the granularity
reserve_size = granularity * ((reserve_size + granularity - 1) / granularity); reserve_size = granularity * ((reserve_size + granularity - 1) / granularity);
GGML_ASSERT(g_cuda_pool_size[id] + reserve_size <= CUDA_POOL_VMM_MAX_SIZE); GGML_ASSERT(g_cuda_pool_size[id] + reserve_size <= CUDA_POOL_VMM_MAX_SIZE);