portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

testing gcc 10.0.1: build errors in Exchange1 using -DA64FX and in Lattice_base.h building Dslash only

Browse Source
This commit is contained in:
nils meyer 2020-04-19 01:25:40 +02:00
parent 6fdce60492
commit 64b72fc17f
11 changed files with 53 additions and 37 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
Grid/qcd/action/fermion/implementation/WilsonKernelsAsmA64FX.h
View File

@ -32,7 +32,12 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#if defined(A64FXASM)
#pragma message("invoking A64FX Dslash")
// include here if A64FX was not defined
#ifndef A64FX
#include <arm_sve.h>
#endif
#pragma message("specialize A64FX Dslash")
// undefine everything
#include <simd/Fujitsu_A64FX_undef.h>

2
Grid/qcd/action/fermion/implementation/WilsonKernelsAsmBody.h
View File

@ -34,6 +34,8 @@
#define DIR7_RECON TP_RECON_ACCUM
#endif
#pragma message("this should not happen")
////////////////////////////////////////////////////////////////////////////////
// Comms then compute kernel
////////////////////////////////////////////////////////////////////////////////

2
Grid/simd/Fujitsu_A64FX_asm_double.h
View File

@ -38,8 +38,6 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define LOCK_GAUGE(A)
#define UNLOCK_GAUGE(A)
#define MASK_REGS DECLARATIONS_A64FXd
#define COMPLEX_SIGNS(A)
#define LOAD64(A,B)
#define SAVE_RESULT(A,B) RESULT_A64FXd(A); PREFETCH_RESULT_L2_STORE(B)
#define MULT_2SPIN_1(Dir) MULT_2SPIN_1_A64FXd(Dir)
#define MULT_2SPIN_2 MULT_2SPIN_2_A64FXd

2
Grid/simd/Fujitsu_A64FX_asm_single.h
View File

@ -38,8 +38,6 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define LOCK_GAUGE(A)
#define UNLOCK_GAUGE(A)
#define MASK_REGS DECLARATIONS_A64FXf
#define COMPLEX_SIGNS(A)
#define LOAD64(A,B)
#define SAVE_RESULT(A,B) RESULT_A64FXf(A); PREFETCH_RESULT_L2_STORE(B)
#define MULT_2SPIN_1(Dir) MULT_2SPIN_1_A64FXf(Dir)
#define MULT_2SPIN_2 MULT_2SPIN_2_A64FXf

28
Grid/simd/Fujitsu_A64FX_intrin_double.h
View File

@ -38,8 +38,6 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define LOCK_GAUGE(A)
#define UNLOCK_GAUGE(A)
#define MASK_REGS DECLARATIONS_A64FXd
#define COMPLEX_SIGNS(A)
#define LOAD64(A,B)
#define SAVE_RESULT(A,B) RESULT_A64FXd(A); PREFETCH_RESULT_L2_STORE(B)
#define MULT_2SPIN_1(Dir) MULT_2SPIN_1_A64FXd(Dir)
#define MULT_2SPIN_2 MULT_2SPIN_2_A64FXd
@ -111,7 +109,7 @@ Author: Nils Meyer <nils.meyer@ur.de>
pg1 = svptrue_b64(); \
svuint64_t table0; \
svfloat64_t zero0; \
zero0 = __svzero(zero0);
zero0 = svdup_f64(0.);
#define Chimu_00 Chi_00
#define Chimu_01 Chi_01
@ -559,18 +557,18 @@ Author: Nils Meyer <nils.meyer@ur.de>
// ZERO_PSI
#define ZERO_PSI_A64FXd \
result_00 = __svzero(result_00); \
result_01 = __svzero(result_01); \
result_02 = __svzero(result_02); \
result_10 = __svzero(result_10); \
result_11 = __svzero(result_11); \
result_12 = __svzero(result_12); \
result_20 = __svzero(result_20); \
result_21 = __svzero(result_21); \
result_22 = __svzero(result_22); \
result_30 = __svzero(result_30); \
result_31 = __svzero(result_31); \
result_32 = __svzero(result_32);
result_00 = svdup_f64(0.); \
result_01 = svdup_f64(0.); \
result_02 = svdup_f64(0.); \
result_10 = svdup_f64(0.); \
result_11 = svdup_f64(0.); \
result_12 = svdup_f64(0.); \
result_20 = svdup_f64(0.); \
result_21 = svdup_f64(0.); \
result_22 = svdup_f64(0.); \
result_30 = svdup_f64(0.); \
result_31 = svdup_f64(0.); \
result_32 = svdup_f64(0.);
// PREFETCH_RESULT_L2_STORE (prefetch store to L2)
#define PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXd(base) \

28
Grid/simd/Fujitsu_A64FX_intrin_single.h
View File

@ -38,8 +38,6 @@ Author: Nils Meyer <nils.meyer@ur.de>
#define LOCK_GAUGE(A)
#define UNLOCK_GAUGE(A)
#define MASK_REGS DECLARATIONS_A64FXf
#define COMPLEX_SIGNS(A)
#define LOAD64(A,B)
#define SAVE_RESULT(A,B) RESULT_A64FXf(A); PREFETCH_RESULT_L2_STORE(B)
#define MULT_2SPIN_1(Dir) MULT_2SPIN_1_A64FXf(Dir)
#define MULT_2SPIN_2 MULT_2SPIN_2_A64FXf
@ -111,7 +109,7 @@ Author: Nils Meyer <nils.meyer@ur.de>
pg1 = svptrue_b32(); \
svuint32_t table0; \
svfloat32_t zero0; \
zero0 = __svzero(zero0);
zero0 = svdup_f32(0.);
#define Chimu_00 Chi_00
#define Chimu_01 Chi_01
@ -559,18 +557,18 @@ Author: Nils Meyer <nils.meyer@ur.de>
// ZERO_PSI
#define ZERO_PSI_A64FXf \
result_00 = __svzero(result_00); \
result_01 = __svzero(result_01); \
result_02 = __svzero(result_02); \
result_10 = __svzero(result_10); \
result_11 = __svzero(result_11); \
result_12 = __svzero(result_12); \
result_20 = __svzero(result_20); \
result_21 = __svzero(result_21); \
result_22 = __svzero(result_22); \
result_30 = __svzero(result_30); \
result_31 = __svzero(result_31); \
result_32 = __svzero(result_32);
result_00 = svdup_f32(0.); \
result_01 = svdup_f32(0.); \
result_02 = svdup_f32(0.); \
result_10 = svdup_f32(0.); \
result_11 = svdup_f32(0.); \
result_12 = svdup_f32(0.); \
result_20 = svdup_f32(0.); \
result_21 = svdup_f32(0.); \
result_22 = svdup_f32(0.); \
result_30 = svdup_f32(0.); \
result_31 = svdup_f32(0.); \
result_32 = svdup_f32(0.);
// PREFETCH_RESULT_L2_STORE (prefetch store to L2)
#define PREFETCH_RESULT_L2_STORE_INTERNAL_A64FXf(base) \

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

@ -385,7 +385,7 @@ struct MultComplex{
svbool_t pg1 = acle<T>::pg1();
typename acle<T>::vt a_v = svld1(pg1, a.v);
typename acle<T>::vt b_v = svld1(pg1, b.v);
typename acle<T>::vt z_v = __svzero(z_v);
typename acle<T>::vt z_v = acle<T>::zero();
// using FCMLA
typename acle<T>::vt r_v = svcmla_x(pg1, z_v, a_v, b_v, 90);

5
Grid/simd/Grid_vector_types.h
View File

@ -118,7 +118,10 @@ accelerator_inline Grid_half sfw_float_to_half(float ff) {
#ifdef GEN
#if defined(A64FX) // breakout A64FX SVE ACLE here
//#pragma message("building for A64FX / SVE ACLE")
#define ARMCLANGHOTFIX
#if defined(clang)
#define ARMCLANGHOTFIX // armclang 20.0 compiles, but binaries give wrong results without hotfix
#endif
#include <arm_sve.h>
#include "Grid_a64fx-2.h"
#else
#include "Grid_generic.h"

8
benchmarks/Benchmark_dwf.cc
View File

@ -201,12 +201,20 @@ int main (int argc, char ** argv)
double volume=Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
double flops=single_site_flops*volume*ncall;
// RF/L1: 4d Wilson
double data_L1 = (volume * 180 * 64 / 4 * ncall) / (1024.*1024.*1024.);
// L2 throughput
double data_L2 = (volume * 9 * 12 * 64 / 4 * ncall + (volume/Ls) * 8*9 * 64/4) / (1024.*1024.*1024.);
std::cout<<GridLogMessage << "Called Dw "<<ncall<<" times in "<<t1-t0<<" us"<<std::endl;
// std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
// std::cout<<GridLogMessage << "norm ref "<< norm2(ref)<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl;
std::cout<<GridLogMessage << "mflop/s per rank = "<< flops/(t1-t0)/NP<<std::endl;
std::cout<<GridLogMessage << "mflop/s per node = "<< flops/(t1-t0)/NN<<std::endl;
std::cout<<GridLogMessage << "RF/L1 GiB/s (base 2) = "<< 1000000. * data_L1/((t1-t0))<<std::endl;
std::cout<<GridLogMessage << "L2 GiB/s (base 2) = "<< 1000000. * data_L2/((t1-t0))<<std::endl;
err = ref-result;
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;
//exit(0);

3
benchmarks/Benchmark_wilson.cc
View File

@ -152,6 +152,7 @@ int main (int argc, char ** argv)
std::cout<<GridLogMessage << "Calling Dw"<<std::endl;
int ncall=1000;
//int ncall=1;
double t0=usecond();
for(int i=0;i<ncall;i++){
Dw.Dhop(src,result,0);
@ -173,12 +174,14 @@ int main (int argc, char ** argv)
}
double data = (volume * 180 * 64 / 4 * ncall) / (1024.*1024.*1024.);
std::cout<<GridLogMessage << "Called Dw"<<std::endl;
std::cout<<GridLogMessage << "flops per site " << single_site_flops << std::endl;
std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
std::cout<<GridLogMessage << "norm ref "<< norm2(ref)<<std::endl;
std::cout<<GridLogMessage << "mflop/s = "<< flops/(t1-t0)<<std::endl;
std::cout<<GridLogMessage << "GiB/s (base 2) = "<< 1000000. * data/((t1-t0))<<std::endl;
err = ref-result;
std::cout<<GridLogMessage << "norm diff "<< norm2(err)<<std::endl;

3
benchmarks/Benchmark_wilson_sweep.cc
View File

@ -132,7 +132,10 @@ void bench_wilson (
for(int i=0; i<ncall; i++) { Dw.Dhop(src,result,dag); }
double t1 = usecond();
double flops = single_site_flops * volume * ncall;
double data_tp = (volume * 180 * 64 * ncall) / 1000.; // / (1024.*1024.*1024.);
//std::cout << flops/(t1-t0) << " (" << data_tp/(t1-t0) << " MB/s) \t";
std::cout << flops/(t1-t0) << "\t\t";
}
void bench_wilson_eo (