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

CUDA: fix __builtin_assume for CUDA < 11.2

Browse source
This commit is contained in:
JohannesGaessler 2023-08-19 17:17:55 +02:00
parent 1f0bccb279
commit 2faad208ae
1 changed files with 8 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

9
ggml-cuda.cu
View file

@ -51,7 +51,14 @@ static_assert(sizeof(half) == sizeof(ggml_fp16_t), "wrong fp16 size");
exit(1); \ exit(1); \
} \ } \
} while (0) } while (0)
#endif // CUDART_VERSION >= 11 #endif // CUDART_VERSION >= 12000
// define nop for old CUDA versions to fix compilation issues
#if CUDART_VERSION < 11020
__device__ void __builtin_assume(bool exp) {
(void) exp;
}
#endif // CUDART_VERSION < 11020
#ifdef GGML_CUDA_F16 #ifdef GGML_CUDA_F16
typedef half dfloat; // dequantize float typedef half dfloat; // dequantize float