portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-09 21:50:45 +01:00
Code Releases Activity

GEN seems to defined by default -> some fixes applied

Browse Source
This commit is contained in:
nmeyer-ur 2020-05-08 10:47:49 +02:00
parent f45621109b
commit 48a340a9d1
2 changed files with 23 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
Grid/simd/Grid_a64fx-fixedsize.h
View File

@ -43,6 +43,9 @@
#pragma error "Missing SVE feature" #pragma error "Missing SVE feature"
#endif /* __ARM_FEATURE_SVE */ #endif /* __ARM_FEATURE_SVE */
// safety definition, not sure if it's important
#define GEN_SIMD_WIDTH 64u
// low-level API // low-level API
NAMESPACE_BEGIN(Grid); NAMESPACE_BEGIN(Grid);
NAMESPACE_BEGIN(Optimization); NAMESPACE_BEGIN(Optimization);

36
Grid/simd/Grid_vector_types.h
View File

@ -115,25 +115,29 @@ accelerator_inline Grid_half sfw_float_to_half(float ff) {
#include "Grid_generic.h" #include "Grid_generic.h"
#endif #endif
*/ */
#ifdef GEN
#if defined(A64FX) // breakout A64FX SVE ACLE here
#pragma message("building for A64FX / SVE ACLE VLA")
#if defined(ARMCLANGCOMPAT)
#pragma message("applying armclang fix")
//#else
// #pragma message("not applying armclang fix")
#endif
#include <arm_sve.h>
#include "Grid_a64fx-2.h"
#else
#pragma message("building for A64FX / GEN")
#include "Grid_generic.h"
#endif
#endif
// A64FX with gcc 10 // A64FX with gcc 10
#ifdef A64FXGCC #ifdef A64FXGCC
#include "Grid_a64fx-fixedsize.h" #include "Grid_a64fx-fixedsize.h"
#else
#ifdef GEN
#if defined(A64FX) // breakout A64FX SVE ACLE here
#pragma message("building for A64FX / SVE ACLE VLA")
#if defined(ARMCLANGCOMPAT)
#pragma message("applying armclang fix")
//#else
// #pragma message("not applying armclang fix")
#endif
#include <arm_sve.h>
#include "Grid_a64fx-2.h"
#else
#pragma message("building for A64FX / GEN")
#include "Grid_generic.h"
#endif
#else
#pragma error("Undefined architecture")
#endif
#endif #endif
#ifdef SSE4 #ifdef SSE4
#include "Grid_sse4.h" #include "Grid_sse4.h"
#endif #endif