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

fix cmake build

Browse source
This commit is contained in:
Johannes Gäßler 2024-04-02 11:58:59 +02:00 • committed by Georgi Gerganov
parent bb0d51accd
commit c63dfdf765
2 changed files with 24 additions and 40 deletions
Download patch file Download diff file Expand all files Collapse all files

26
ggml-cuda/common.cuh
View file

@ -271,7 +271,6 @@ static __device__ __forceinline__ float2 warp_reduce_sum(float2 a) {
return a;
}
#ifdef GGML_CUDA_F16
static __device__ __forceinline__ half2 warp_reduce_sum(half2 a) {
#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL
#pragma unroll
@ -284,7 +283,6 @@ static __device__ __forceinline__ half2 warp_reduce_sum(half2 a) {
NO_DEVICE_CODE;
#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL
}
#endif // GGML_CUDA_F16
static __device__ __forceinline__ float warp_reduce_max(float x) {
#pragma unroll
@ -294,18 +292,18 @@ static __device__ __forceinline__ float warp_reduce_max(float x) {
return x;
}
//static __device__ __forceinline__ half2 warp_reduce_max(half2 x) {
//#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL && CUDART_VERSION >= CUDART_HMAX
//#pragma unroll
// for (int mask = 16; mask > 0; mask >>= 1) {
// x = __hmax2(x, __shfl_xor_sync(0xffffffff, x, mask, 32));
// }
// return x;
//#else
// GGML_UNUSED(x);
// NO_DEVICE_CODE;
//#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL && CUDART_VERSION >= CUDART_HMAX
//}
static __device__ __forceinline__ half2 warp_reduce_max(half2 x) {
#if !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL && CUDART_VERSION >= CUDART_HMAX
#pragma unroll
for (int mask = 16; mask > 0; mask >>= 1) {
x = __hmax2(x, __shfl_xor_sync(0xffffffff, x, mask, 32));
}
return x;
#else
GGML_UNUSED(x);
NO_DEVICE_CODE;
#endif // !(defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= CC_PASCAL && CUDART_VERSION >= CUDART_HMAX
}
#if defined(GGML_USE_HIPBLAS)