CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-17 11:37:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: use fastdiv in set-rows (#16834)

Browse Source 
* CUDA: use fastdiv in set-rows

* add assert about value fitting in u32
This commit is contained in:
Aman Gupta
2025-10-29 21:11:53 +08:00
committed by GitHub
parent 144a4ce824
commit e41bcce8f0
2 changed files with 106 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
ggml/src/ggml-cuda/common.cuh
View File

@@ -625,8 +625,11 @@ static __device__ __forceinline__ float ggml_cuda_e8m0_to_fp32(uint8_t x) {
// and a shift:
//
// n/d = (mulhi(n, mp) + n) >> L;
static const uint3 init_fastdiv_values(uint32_t d) {
GGML_ASSERT(d != 0);
static const uint3 init_fastdiv_values(uint64_t d_64) {
GGML_ASSERT(d_64 != 0);
GGML_ASSERT(d_64 <= std::numeric_limits<uint32_t>::max());
uint32_t d = (uint32_t)d_64;
// compute L = ceil(log2(d));
uint32_t L = 0;