portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 07:17:06 +01:00
Code Releases Activity

Merge branch 'develop' into feature/a64fx-3

Browse Source
This commit is contained in:
Nils Meyer
2020-12-19 02:37:11 +01:00
parent 909acd55cd 7adb253e25
commit 3f9ae6e7e7
50 changed files with 2456 additions and 354 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
Grid/simd/Grid_gpu_vec.h
View File

@ -38,12 +38,20 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
#ifdef GRID_HIP
#include <hip/hip_fp16.h>
#endif
#ifdef GRID_SYCL
namespace Grid {
typedef struct { uint16_t x;} half;
typedef struct { half x; half y;} half2;
typedef struct { float x; float y;} float2;
typedef struct { double x; double y;} double2;
}
#endif
namespace Grid {
#if (!defined(GRID_CUDA)) && (!defined(GRID_HIP))
typedef struct { uint16_t x;} half;
#endif
typedef struct Half2_t { half x; half y; } Half2;
#define COALESCE_GRANULARITY ( GEN_SIMD_WIDTH )
@ -156,7 +164,7 @@ accelerator_inline float half2float(half h)
f = __half2float(h);
#else
Grid_half hh;
hh.x = hr.x;
hh.x = h.x;
f= sfw_half_to_float(hh);
#endif
return f;

8
Grid/simd/Grid_vector_unops.h
View File

@ -125,14 +125,6 @@ accelerator_inline Grid_simd<S, V> sqrt(const Grid_simd<S, V> &r) {
return SimdApply(SqrtRealFunctor<S>(), r);
}
template <class S, class V>
accelerator_inline Grid_simd<S, V> rsqrt(const Grid_simd<S, V> &r) {
return SimdApply(RSqrtRealFunctor<S>(), r);
}
template <class Scalar>
accelerator_inline Scalar rsqrt(const Scalar &r) {
return (RSqrtRealFunctor<Scalar>(), r);
}
template <class S, class V>
accelerator_inline Grid_simd<S, V> cos(const Grid_simd<S, V> &r) {
return SimdApply(CosRealFunctor<S>(), r);
}