Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-01 00:39:00 +01:00
Code Issues Packages Projects Releases Wiki Activity

fix NVCC version on Makefile, __halves2half2 -> make_half2

Browse Source
This commit is contained in:
slaren 2023-07-16 00:20:43 +02:00
parent 24cc6f008f
commit 33ab185dd1
2 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
Makefile
View File

@ -176,6 +176,7 @@ ifdef LLAMA_CUDA
OBJS += ggml-cuda.o
NVCC = nvcc
NVCCFLAGS = --forward-unknown-to-host-compiler
NVCCV := $(shell $(NVCC) --version | tail -n 1)
ifdef LLAMA_DEBUG
NVCCFLAGS += -lineinfo
endif # LLAMA_DEBUG

2
ggml-cuda-kern.h
View File

@ -14,7 +14,7 @@ template<> struct vec2_t_impl<float> { typedef float2 type; };
template<typename T> using vec2_t = typename vec2_t_impl<T>::type;
template<typename T> inline __host__ __device__ vec2_t<T> make_vec2_t(const T & x, const T & y);
template<> inline __host__ __device__ vec2_t<half> make_vec2_t(const half & x, const half & y) { return __halves2half2(x, y); }
template<> inline __host__ __device__ vec2_t<half> make_vec2_t(const half & x, const half & y) { return make_half2 (x, y); }
template<> inline __host__ __device__ vec2_t<float> make_vec2_t(const float & x, const float & y) { return make_float2(x, y); }
// the cuda headers define operators for half2, but not for float2