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

GGML_CUDA_FORCE_CUSTOM_MEMORY_POOL was added to force use only custom memory pool

Browse source
This commit is contained in:
Oleksii Maryshchenko 2023-11-03 15:06:40 +01:00
parent bd56886fd6
commit c42ca8f1b7
1 changed files with 6 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ggml-cuda.cu
View file

@ -108,6 +108,10 @@
#define CUDA_USE_TENSOR_CORES
#endif
#if !defined(GGML_CUDA_FORCE_CUSTOM_MEMORY_POOL)
#define CUDA_USE_MEMORY_POOL
#endif
// max batch size to use MMQ kernels when tensor cores are available
#define MMQ_MAX_BATCH_SIZE 32
@ -5845,7 +5849,7 @@ void ggml_init_cublas() {
cudaDeviceProp prop;
CUDA_CHECK(cudaGetDeviceProperties(&prop, id));
fprintf(stderr, " Device %d: %s, compute capability %d.%d", id, prop.name, prop.major, prop.minor);
#if defined(CUDA_USE_MEMORY_POOL)
// configure memory pool
if (prop.memoryPoolsSupported == 1) {
cudaError_t err = cudaDeviceGetMemPool(&g_cudaMemPools[id], id);
@ -5859,7 +5863,7 @@ void ggml_init_cublas() {
} else {
fprintf(stderr, ", CUDA memory pool is not supported\n");
}
#endif
g_tensor_split[id] = total_vram;
total_vram += prop.totalGlobalMem;
#if defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)