portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-27 14:15:55 +01:00
Code Releases Activity

some changes in breaking out A64FX: use -DA64FXFIXEDSIZE for fixed size, but also define GEN

Browse Source
This commit is contained in:
nmeyer-ur 2020-05-08 12:05:31 +02:00
parent 39f3ae5b1d
commit fbed02690d
2 changed files with 15 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
Grid/simd/Grid_a64fx-fixedsize.h
View File

@ -36,6 +36,7 @@
//static_assert(GEN_SIMD_WIDTH % 64u == 0, "A64FX SIMD vector size is 64 bytes"); //static_assert(GEN_SIMD_WIDTH % 64u == 0, "A64FX SIMD vector size is 64 bytes");
/*
#ifdef __ARM_FEATURE_SVE #ifdef __ARM_FEATURE_SVE
#pragma message("Yes, we have SVE feature") #pragma message("Yes, we have SVE feature")
#include <arm_sve.h> #include <arm_sve.h>
@ -43,6 +44,7 @@
#pragma error "Missing SVE feature" #pragma error "Missing SVE feature"
#endif /* __ARM_FEATURE_SVE */ #endif /* __ARM_FEATURE_SVE */
// gcc 10 features // gcc 10 features
#if __ARM_FEATURE_SVE_BITS==512 #if __ARM_FEATURE_SVE_BITS==512
#pragma message("Fixed-size SVE ACLE") #pragma message("Fixed-size SVE ACLE")
@ -58,7 +60,7 @@ typedef svuint64_t lutd __attribute__((arm_sve_vector_bits(512))); // LUTs for
#endif /* __ARM_FEATURE_SVE_BITS */ #endif /* __ARM_FEATURE_SVE_BITS */
// safety definition, not sure if it's necessary // safety definition, not sure if it's necessary
#define GEN_SIMD_WIDTH 64u //#define GEN_SIMD_WIDTH 64u
// low-level API // low-level API
NAMESPACE_BEGIN(Grid); NAMESPACE_BEGIN(Grid);

20
Grid/simd/Grid_vector_types.h
View File

@ -119,25 +119,29 @@ accelerator_inline Grid_half sfw_float_to_half(float ff) {
#ifdef A64FXGCC #ifdef A64FXGCC
#include "Grid_a64fx-fixedsize.h" #include "Grid_a64fx-fixedsize.h"
#else #else
#ifdef GEN
#if defined(A64FX) // breakout A64FX SVE ACLE here #ifdef GEN
#if defined(A64FX) || defined(A64FXGCC) // breakout A64FX SVE ACLE here
#include <arm_sve.h>
if defined(A64FX) // VLA
#pragma message("building for A64FX / SVE ACLE VLA") #pragma message("building for A64FX / SVE ACLE VLA")
#if defined(ARMCLANGCOMPAT) #if defined(ARMCLANGCOMPAT)
#pragma message("applying armclang fix") #pragma message("applying armclang fix")
//#else //#else
// #pragma message("not applying armclang fix") //#pragma message("not applying armclang fix")
#endif #endif
#include <arm_sve.h>
#include "Grid_a64fx-2.h" #include "Grid_a64fx-2.h"
#endif
#if defined(A64FXFIXEDSIZE) // fixed size data types
#include "Grid_a64fx-fixedsize.h"
#endif
#else #else
#pragma message("building for A64FX / GEN") #pragma message("building for GEN") // generic
#include "Grid_generic.h" #include "Grid_generic.h"
#endif #endif
#else
#pragma error("Undefined architecture")
#endif
#endif #endif
#ifdef SSE4 #ifdef SSE4
#include "Grid_sse4.h" #include "Grid_sse4.h"
#endif #endif