portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-09-20 09:15:38 +01:00
Code Releases Activity

fix build error enabling fcmla/mac in vector types for VLA

Browse Source
This commit is contained in:
nmeyer-ur 2020-05-21 21:21:03 +02:00
parent 8c5a5fdfce
commit 6ddcef1bca
2 changed files with 7 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
Grid/simd/Grid_a64fx-2.h
View File

@ -396,7 +396,7 @@ struct MultComplex{
struct MultAddComplex{
// Complex a*b+c
template <typename T>
inline void mac(const vec<T> &a, const vec<T> b, const vec<T> c){
inline vec<T> operator()(vec<T> a, vec<T> b, vec<T> c){
vec<T> out;
svbool_t pg1 = acle<T>::pg1();
typename acle<T>::vt a_v = svld1(pg1, a.v);
@ -406,8 +406,9 @@ struct MultAddComplex{
// using FCMLA
typename acle<T>::vt r_v = svcmla_x(pg1, c_v, a_v, b_v, 0);
r_v = svcmla_x(pg1, r_v, a_v, b_v, 90);
svst1(pg1, out.v, r_v);
svst1(pg1, a.v, r_v);
return out;
}
};

10
Grid/simd/Grid_vector_types.h
View File

@ -299,8 +299,7 @@ public:
// FIXME -- alias this to an accelerator_inline MAC struct.
// FIXME VLA build error
//#if defined(A64FX) || defined(A64FXFIXEDSIZE) // VLA only: build error
#if defined(A64FXFIXEDSIZE)
#if defined(A64FX) || defined(A64FXFIXEDSIZE)
friend accelerator_inline void mac(Grid_simd *__restrict__ y,
const Grid_simd *__restrict__ a,
const Grid_simd *__restrict__ x) {
@ -791,10 +790,9 @@ accelerator_inline Grid_simd<S, V> operator*(Grid_simd<S, V> a, Grid_simd<S, V>
return ret;
};
// ----------------A64FX MAC ---------------------
// ---------------- A64FX MAC -------------------
// Distinguish between complex types and others
//#if defined(A64FX) || defined(A64FXFIXEDSIZE) // VLA only: build error
#if defined(A64FXFIXEDSIZE)
#if defined(A64FX) || defined(A64FXFIXEDSIZE)
template <class S, class V, IfComplex<S> = 0>
accelerator_inline Grid_simd<S, V> fxmac(Grid_simd<S, V> a, Grid_simd<S, V> b, Grid_simd<S, V> c) {
Grid_simd<S, V> ret;
@ -810,7 +808,7 @@ accelerator_inline Grid_simd<S, V> fxmac(Grid_simd<S, V> a, Grid_simd<S, V> b, G
return ret;
};
#endif
// -------------------------------------
// ----------------------------------------------
// Distinguish between complex types and others