Skip to content

Commit 33ab185

Browse files
committed
fix NVCC version on Makefile, __halves2half2 -> make_half2
1 parent 24cc6f0 commit 33ab185

2 files changed

Lines changed: 2 additions & 1 deletion

File tree

Makefile

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -176,6 +176,7 @@ ifdef LLAMA_CUDA
176176
OBJS += ggml-cuda.o
177177
NVCC = nvcc
178178
NVCCFLAGS = --forward-unknown-to-host-compiler
179+
NVCCV := $(shell $(NVCC) --version | tail -n 1)
179180
ifdef LLAMA_DEBUG
180181
NVCCFLAGS += -lineinfo
181182
endif # LLAMA_DEBUG

ggml-cuda-kern.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -14,7 +14,7 @@ template<> struct vec2_t_impl<float> { typedef float2 type; };
1414
template<typename T> using vec2_t = typename vec2_t_impl<T>::type;
1515

1616
template<typename T> inline __host__ __device__ vec2_t<T> make_vec2_t(const T & x, const T & y);
17-
template<> inline __host__ __device__ vec2_t<half> make_vec2_t(const half & x, const half & y) { return __halves2half2(x, y); }
17+
template<> inline __host__ __device__ vec2_t<half> make_vec2_t(const half & x, const half & y) { return make_half2 (x, y); }
1818
template<> inline __host__ __device__ vec2_t<float> make_vec2_t(const float & x, const float & y) { return make_float2(x, y); }
1919

2020
// the cuda headers define operators for half2, but not for float2

0 commit comments

Comments
 (0)