portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-11 14:40:46 +01:00
Code Releases Activity

SVE_readme update, update Grid_vector_types.h

Browse Source
This commit is contained in:
nmeyer-ur 2020-04-25 13:48:26 +02:00
parent 499edc0636
commit 5abec5b8a9
2 changed files with 8 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
Grid/simd/Grid_vector_types.h
View File

@ -118,11 +118,10 @@ accelerator_inline Grid_half sfw_float_to_half(float ff) {
#ifdef GEN #ifdef GEN
#if defined(A64FX) // breakout A64FX SVE ACLE here #if defined(A64FX) // breakout A64FX SVE ACLE here
//#pragma message("building for A64FX / SVE ACLE") //#pragma message("building for A64FX / SVE ACLE")
#if defined(HOTFIX) #if defined(ARMCLANGCOMPAT)
#pragma message("applying armclang hotfix") #pragma message("applying armclang fix")
#define ARMCLANGHOTFIX // armclang 20.0 compiles, but binaries give wrong results without hotfix //#else
#else // #pragma message("not applying armclang fix")
#pragma message("not applying armclang hotfix")
#endif #endif
#include <arm_sve.h> #include <arm_sve.h>
#include "Grid_a64fx-2.h" #include "Grid_a64fx-2.h"
@ -241,7 +240,7 @@ public:
return sizeof(Vector_type) / sizeof(Scalar_type); return sizeof(Vector_type) / sizeof(Scalar_type);
} }
#ifdef ARMCLANGHOTFIX #ifdef ARMCLANGCOMPAT
accelerator_inline Grid_simd &operator=(const Grid_simd &&rhs) { accelerator_inline Grid_simd &operator=(const Grid_simd &&rhs) {
svint8_t tmp = svld1(svptrue_b8(), (int8_t*)&(rhs.v)); svint8_t tmp = svld1(svptrue_b8(), (int8_t*)&(rhs.v));
svst1(svptrue_b8(), (int8_t*)this, tmp); svst1(svptrue_b8(), (int8_t*)this, tmp);

6
SVE_README.txt
View File

@ -7,14 +7,14 @@ armclang 20.0 VLA
../configure --with-lime=/home/men04359/lime/c-lime --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp CXX=armclang++ CC=armclang CXXFLAGS="-std=c++11 -fno-unroll-loops -mllvm -vectorizer-min-trip-count=2 -march=armv8-a+sve -DARMCLANGCOMPAT -DA64FX -DA64FXASM -DDSLASHINTRIN" LDFLAGS=-static GRID_LDFLAGS=-static MPI_CXXLDFLAGS=-static ../configure --with-lime=/home/men04359/lime/c-lime --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp CXX=armclang++ CC=armclang CXXFLAGS="-std=c++11 -fno-unroll-loops -mllvm -vectorizer-min-trip-count=2 -march=armv8-a+sve -DARMCLANGCOMPAT -DA64FX -DA64FXASM -DDSLASHINTRIN" LDFLAGS=-static GRID_LDFLAGS=-static MPI_CXXLDFLAGS=-static
must use armclang 20.0 with ARMCLANGCOMPAT applied, otherwise Benchmark_wilson gives wrong result must use armclang 20.0 with ARMCLANGCOMPAT, otherwise Benchmark_wilson gives wrong result
armclang 20.1 VLA armclang 20.1 VLA
../configure --with-lime=/home/men04359/lime/c-lime --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp CXX=armclang++ CC=armclang CXXFLAGS="-std=c++11 -mcpu=a64fx -DA64FX -DA64FXASM -DDSLASHINTRIN" LDFLAGS=-static GRID_LDFLAGS=-static MPI_CXXLDFLAGS=-static ../configure --with-lime=/home/men04359/lime/c-lime --without-hdf5 --enable-gen-simd-width=64 --enable-simd=GEN --enable-precision=double --enable-comms=none --enable-openmp CXX=armclang++ CC=armclang CXXFLAGS="-std=c++11 -mcpu=a64fx -DARMCLANGCOMPAT -DA64FX -DA64FXASM -DDSLASHINTRIN" LDFLAGS=-static GRID_LDFLAGS=-static MPI_CXXLDFLAGS=-static
must use armclang 20.1 with ARMCLANGCOMPAT applied, otherwise Benchmark_wilson gives wrong result must use armclang 20.1 with ARMCLANGCOMPAT, otherwise Benchmark_wilson gives wrong result
Test_simd build error caused by -mcpu=a64fx ? Test_simd build error caused by -mcpu=a64fx ?