64-BitBrainstorm_2026/AMSS-NCKU
8
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Enable OpenMP task parallelism for C kernels

Browse Source
This commit is contained in:
CGH0S7
2026-03-30 20:34:34 +08:00
parent 8c1f4d8108
commit 5f664716ab
9 changed files with 137 additions and 79 deletions
Download Patch File Download Diff File Expand all files Collapse all files

125
AMSS_NCKU_source/bssn_rhs_c.C
View File

@@ -2,6 +2,24 @@
#include "bssn_rhs.h"
#include "share_func.h"
#include "tool.h"
#ifdef _OPENMP
#define BSSN_OMP_TASK_GROUP_BEGIN \
_Pragma("omp parallel") \
{ \
_Pragma("omp single nowait") \
{
#define BSSN_OMP_TASK_CALL(...) \
_Pragma("omp task") { __VA_ARGS__; }
#define BSSN_OMP_TASK_GROUP_END \
_Pragma("omp taskwait") \
} \
}
#else
#define BSSN_OMP_TASK_GROUP_BEGIN {
#define BSSN_OMP_TASK_CALL(...) { __VA_ARGS__; }
#define BSSN_OMP_TASK_GROUP_END }
#endif
// 0-based i,j,k
// #define IDX_F(i,j,k,nx,ny) ((i) + (j)*(nx) + (k)*(nx)*(ny))
// ex(1)=nx, ex(2)=ny, ex(3)=nz
@@ -108,18 +126,20 @@ int f_compute_rhs_bssn(int *ex, double &T,
chin1[i] = chi[i] + 1.0;
}
// 9ms //
fderivs(ex,betax,betaxx,betaxy,betaxz,X,Y,Z,ANTI, SYM, SYM,Symmetry,Lev);
fderivs(ex,betay,betayx,betayy,betayz,X,Y,Z, SYM,ANTI, SYM,Symmetry,Lev);
fderivs(ex,betaz,betazx,betazy,betazz,X,Y,Z, SYM, SYM,ANTI,Symmetry,Lev);
fderivs(ex,chi,chix,chiy,chiz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev);
fderivs(ex,dxx,gxxx,gxxy,gxxz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,Lev);
fderivs(ex,gxy,gxyx,gxyy,gxyz,X,Y,Z,ANTI,ANTI,SYM ,Symmetry,Lev);
fderivs(ex,gxz,gxzx,gxzy,gxzz,X,Y,Z,ANTI,SYM ,ANTI,Symmetry,Lev);
fderivs(ex,dyy,gyyx,gyyy,gyyz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,Lev);
fderivs(ex,gyz,gyzx,gyzy,gyzz,X,Y,Z,SYM ,ANTI,ANTI,Symmetry,Lev);
fderivs(ex,dzz,gzzx,gzzy,gzzz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,Lev);
fderivs(ex,Lap,Lapx,Lapy,Lapz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev);
fderivs(ex,trK,Kx,Ky,Kz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev);
BSSN_OMP_TASK_GROUP_BEGIN
BSSN_OMP_TASK_CALL(fderivs(ex,betax,betaxx,betaxy,betaxz,X,Y,Z,ANTI, SYM, SYM,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,betay,betayx,betayy,betayz,X,Y,Z, SYM,ANTI, SYM,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,betaz,betazx,betazy,betazz,X,Y,Z, SYM, SYM,ANTI,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,chi,chix,chiy,chiz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,dxx,gxxx,gxxy,gxxz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,gxy,gxyx,gxyy,gxyz,X,Y,Z,ANTI,ANTI,SYM ,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,gxz,gxzx,gxzy,gxzz,X,Y,Z,ANTI,SYM ,ANTI,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,dyy,gyyx,gyyy,gyyz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,gyz,gyzx,gyzy,gyzz,X,Y,Z,SYM ,ANTI,ANTI,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,dzz,gzzx,gzzy,gzzz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,Lap,Lapx,Lapy,Lapz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,trK,Kx,Ky,Kz,X,Y,Z,SYM,SYM,SYM,Symmetry,Lev))
BSSN_OMP_TASK_GROUP_END
// 3ms //
for(int i=0;i<all;i+=1){
@@ -316,15 +336,14 @@ int f_compute_rhs_bssn(int *ex, double &T,
);
}
// 22.3ms //
fdderivs(ex,betax,gxxx,gxyx,gxzx,gyyx,gyzx,gzzx,
X,Y,Z,ANTI,SYM, SYM ,Symmetry,Lev);
fdderivs(ex,betay,gxxy,gxyy,gxzy,gyyy,gyzy,gzzy,
X,Y,Z,SYM ,ANTI,SYM ,Symmetry,Lev);
fdderivs(ex,betaz,gxxz,gxyz,gxzz,gyyz,gyzz,gzzz,
X,Y,Z,SYM ,SYM, ANTI,Symmetry,Lev);
fderivs(ex,Gamx,Gamxx,Gamxy,Gamxz,X,Y,Z,ANTI,SYM ,SYM ,Symmetry,Lev);
fderivs(ex,Gamy,Gamyx,Gamyy,Gamyz,X,Y,Z,SYM ,ANTI,SYM ,Symmetry,Lev);
fderivs(ex,Gamz,Gamzx,Gamzy,Gamzz,X,Y,Z,SYM ,SYM ,ANTI,Symmetry,Lev);
BSSN_OMP_TASK_GROUP_BEGIN
BSSN_OMP_TASK_CALL(fdderivs(ex,betax,gxxx,gxyx,gxzx,gyyx,gyzx,gzzx,X,Y,Z,ANTI,SYM, SYM ,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fdderivs(ex,betay,gxxy,gxyy,gxzy,gyyy,gyzy,gzzy,X,Y,Z,SYM ,ANTI,SYM ,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fdderivs(ex,betaz,gxxz,gxyz,gxzz,gyyz,gyzz,gzzz,X,Y,Z,SYM ,SYM, ANTI,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,Gamx,Gamxx,Gamxy,Gamxz,X,Y,Z,ANTI,SYM ,SYM ,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,Gamy,Gamyx,Gamyy,Gamyz,X,Y,Z,SYM ,ANTI,SYM ,Symmetry,Lev))
BSSN_OMP_TASK_CALL(fderivs(ex,Gamz,Gamzx,Gamzy,Gamzz,X,Y,Z,SYM ,SYM ,ANTI,Symmetry,Lev))
BSSN_OMP_TASK_GROUP_END
// Fused: fxx/Gamxa + Gamma_rhs part 2 (2 loops -> 1)
for(int i=0;i<all;i+=1){
@@ -1063,30 +1082,32 @@ int f_compute_rhs_bssn(int *ex, double &T,
#endif
}
// advection + KO dissipation with shared symmetry buffer
lopsided_kodis(ex,X,Y,Z,dxx,gxx_rhs,betax,betay,betaz,Symmetry,SSS,eps);
lopsided_kodis(ex,X,Y,Z,Gamz,Gamz_rhs,betax,betay,betaz,Symmetry,SSA,eps);
lopsided_kodis(ex,X,Y,Z,gxy,gxy_rhs,betax,betay,betaz,Symmetry,AAS,eps);
lopsided_kodis(ex,X,Y,Z,Lap,Lap_rhs,betax,betay,betaz,Symmetry,SSS,eps);
lopsided_kodis(ex,X,Y,Z,gxz,gxz_rhs,betax,betay,betaz,Symmetry,ASA,eps);
lopsided_kodis(ex,X,Y,Z,betax,betax_rhs,betax,betay,betaz,Symmetry,ASS,eps);
lopsided_kodis(ex,X,Y,Z,dyy,gyy_rhs,betax,betay,betaz,Symmetry,SSS,eps);
lopsided_kodis(ex,X,Y,Z,betay,betay_rhs,betax,betay,betaz,Symmetry,SAS,eps);
lopsided_kodis(ex,X,Y,Z,gyz,gyz_rhs,betax,betay,betaz,Symmetry,SAA,eps);
lopsided_kodis(ex,X,Y,Z,betaz,betaz_rhs,betax,betay,betaz,Symmetry,SSA,eps);
lopsided_kodis(ex,X,Y,Z,dzz,gzz_rhs,betax,betay,betaz,Symmetry,SSS,eps);
lopsided_kodis(ex,X,Y,Z,dtSfx,dtSfx_rhs,betax,betay,betaz,Symmetry,ASS,eps);
lopsided_kodis(ex,X,Y,Z,Axx,Axx_rhs,betax,betay,betaz,Symmetry,SSS,eps);
lopsided_kodis(ex,X,Y,Z,dtSfy,dtSfy_rhs,betax,betay,betaz,Symmetry,SAS,eps);
lopsided_kodis(ex,X,Y,Z,Axy,Axy_rhs,betax,betay,betaz,Symmetry,AAS,eps);
lopsided_kodis(ex,X,Y,Z,dtSfz,dtSfz_rhs,betax,betay,betaz,Symmetry,SSA,eps);
lopsided_kodis(ex,X,Y,Z,Axz,Axz_rhs,betax,betay,betaz,Symmetry,ASA,eps);
lopsided_kodis(ex,X,Y,Z,Ayy,Ayy_rhs,betax,betay,betaz,Symmetry,SSS,eps);
lopsided_kodis(ex,X,Y,Z,Ayz,Ayz_rhs,betax,betay,betaz,Symmetry,SAA,eps);
lopsided_kodis(ex,X,Y,Z,Azz,Azz_rhs,betax,betay,betaz,Symmetry,SSS,eps);
lopsided_kodis(ex,X,Y,Z,chi,chi_rhs,betax,betay,betaz,Symmetry,SSS,eps);
lopsided_kodis(ex,X,Y,Z,trK,trK_rhs,betax,betay,betaz,Symmetry,SSS,eps);
lopsided_kodis(ex,X,Y,Z,Gamx,Gamx_rhs,betax,betay,betaz,Symmetry,ASS,eps);
lopsided_kodis(ex,X,Y,Z,Gamy,Gamy_rhs,betax,betay,betaz,Symmetry,SAS,eps);
BSSN_OMP_TASK_GROUP_BEGIN
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,dxx,gxx_rhs,betax,betay,betaz,Symmetry,SSS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,Gamz,Gamz_rhs,betax,betay,betaz,Symmetry,SSA,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,gxy,gxy_rhs,betax,betay,betaz,Symmetry,AAS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,Lap,Lap_rhs,betax,betay,betaz,Symmetry,SSS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,gxz,gxz_rhs,betax,betay,betaz,Symmetry,ASA,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,betax,betax_rhs,betax,betay,betaz,Symmetry,ASS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,dyy,gyy_rhs,betax,betay,betaz,Symmetry,SSS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,betay,betay_rhs,betax,betay,betaz,Symmetry,SAS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,gyz,gyz_rhs,betax,betay,betaz,Symmetry,SAA,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,betaz,betaz_rhs,betax,betay,betaz,Symmetry,SSA,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,dzz,gzz_rhs,betax,betay,betaz,Symmetry,SSS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,dtSfx,dtSfx_rhs,betax,betay,betaz,Symmetry,ASS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,Axx,Axx_rhs,betax,betay,betaz,Symmetry,SSS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,dtSfy,dtSfy_rhs,betax,betay,betaz,Symmetry,SAS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,Axy,Axy_rhs,betax,betay,betaz,Symmetry,AAS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,dtSfz,dtSfz_rhs,betax,betay,betaz,Symmetry,SSA,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,Axz,Axz_rhs,betax,betay,betaz,Symmetry,ASA,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,Ayy,Ayy_rhs,betax,betay,betaz,Symmetry,SSS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,Ayz,Ayz_rhs,betax,betay,betaz,Symmetry,SAA,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,Azz,Azz_rhs,betax,betay,betaz,Symmetry,SSS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,chi,chi_rhs,betax,betay,betaz,Symmetry,SSS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,trK,trK_rhs,betax,betay,betaz,Symmetry,SSS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,Gamx,Gamx_rhs,betax,betay,betaz,Symmetry,ASS,eps))
BSSN_OMP_TASK_CALL(lopsided_kodis(ex,X,Y,Z,Gamy,Gamy_rhs,betax,betay,betaz,Symmetry,SAS,eps))
BSSN_OMP_TASK_GROUP_END
// 2ms //
if(co==0){
for (int i=0;i<all;i+=1) {
@@ -1133,12 +1154,14 @@ int f_compute_rhs_bssn(int *ex, double &T,
}
// 1ms //
fderivs(ex,Axx,gxxx,gxxy,gxxz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,0);
fderivs(ex,Axy,gxyx,gxyy,gxyz,X,Y,Z,ANTI,ANTI,SYM ,Symmetry,0);
fderivs(ex,Axz,gxzx,gxzy,gxzz,X,Y,Z,ANTI,SYM ,ANTI,Symmetry,0);
fderivs(ex,Ayy,gyyx,gyyy,gyyz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,0);
fderivs(ex,Ayz,gyzx,gyzy,gyzz,X,Y,Z,SYM ,ANTI,ANTI,Symmetry,0);
fderivs(ex,Azz,gzzx,gzzy,gzzz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,0);
BSSN_OMP_TASK_GROUP_BEGIN
BSSN_OMP_TASK_CALL(fderivs(ex,Axx,gxxx,gxxy,gxxz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,0))
BSSN_OMP_TASK_CALL(fderivs(ex,Axy,gxyx,gxyy,gxyz,X,Y,Z,ANTI,ANTI,SYM ,Symmetry,0))
BSSN_OMP_TASK_CALL(fderivs(ex,Axz,gxzx,gxzy,gxzz,X,Y,Z,ANTI,SYM ,ANTI,Symmetry,0))
BSSN_OMP_TASK_CALL(fderivs(ex,Ayy,gyyx,gyyy,gyyz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,0))
BSSN_OMP_TASK_CALL(fderivs(ex,Ayz,gyzx,gyzy,gyzz,X,Y,Z,SYM ,ANTI,ANTI,Symmetry,0))
BSSN_OMP_TASK_CALL(fderivs(ex,Azz,gzzx,gzzy,gzzz,X,Y,Z,SYM ,SYM ,SYM ,Symmetry,0))
BSSN_OMP_TASK_GROUP_END
// 7ms //
for (int i=0;i<all;i+=1) {
gxxx[i] = gxxx[i] - ( Gamxxx[i] * Axx[i] + Gamyxx[i] * Axy[i] + Gamzxx[i] * Axz[i]

4
AMSS_NCKU_source/fdderivs_c.C
View File

@@ -41,8 +41,8 @@ void fdderivs(const int ex[3],
const size_t nz = (size_t)ex3 + 2;
const size_t fh_size = nx * ny * nz;
static double *fh = NULL;
static size_t cap = 0;
static thread_local double *fh = NULL;
static thread_local size_t cap = 0;
if (fh_size > cap) {
free(fh);

4
AMSS_NCKU_source/fderivs_c.C
View File

@@ -50,8 +50,8 @@ void fderivs(const int ex[3],
const size_t ny = (size_t)ex2 + 2;
const size_t nz = (size_t)ex3 + 2;
const size_t fh_size = nx * ny * nz;
static double *fh = NULL;
static size_t cap = 0;
static thread_local double *fh = NULL;
static thread_local size_t cap = 0;
if (fh_size > cap) {
free(fh);

10
AMSS_NCKU_source/lopsided_kodis_c.C
View File

@@ -43,7 +43,13 @@ void lopsided_kodis(const int ex[3],
const size_t nz = (size_t)ex3 + 3;
const size_t fh_size = nx * ny * nz;
double *fh = (double*)malloc(fh_size * sizeof(double));
static thread_local double *fh = NULL;
static thread_local size_t cap = 0;
if (fh_size > cap) {
free(fh);
fh = (double*)aligned_alloc(64, fh_size * sizeof(double));
cap = fh_size;
}
if (!fh) return;
symmetry_bd(3, ex, f, fh, SoA);
@@ -243,6 +249,4 @@ void lopsided_kodis(const int ex[3],
}
}
}
free(fh);
}

4
AMSS_NCKU_source/makefile
View File

@@ -16,7 +16,7 @@ PROFDATA = /home/$(shell whoami)/AMSS-NCKU/pgo_profile/default.profdata
ifeq ($(PGO_MODE),instrument)
## Phase 1: instrumentation — omit -ipo/-fp-model fast=2 for faster build and numerical stability
CXXAPPFLAGS = -O3 -xHost -fma -fprofile-instr-generate -ipo \
-Dfortran3 -Dnewc -I${MKLROOT}/include $(INTERP_LB_FLAGS)
-Dfortran3 -Dnewc -I${MKLROOT}/include $(INTERP_LB_FLAGS) $(OPENMP_FLAGS)
f90appflags = -O3 -xHost -fma -fprofile-instr-generate -ipo \
-align array64byte -fpp -I${MKLROOT}/include $(POLINT6_FLAG)
else
@@ -26,7 +26,7 @@ else
CXXAPPFLAGS = -O3 -xHost -fp-model fast=2 -fma -ipo \
-Dfortran3 -Dnewc -I${MKLROOT}/include $(INTERP_LB_FLAGS)
-Dfortran3 -Dnewc -I${MKLROOT}/include $(INTERP_LB_FLAGS) $(OPENMP_FLAGS)
f90appflags = -O3 -xHost -fp-model fast=2 -fma -ipo \
-align array64byte -fpp -I${MKLROOT}/include $(POLINT6_FLAG)
endif

3
AMSS_NCKU_source/makefile.inc
View File

@@ -29,6 +29,9 @@ endif
## instrument : PGO Phase 1 instrumentation to collect fresh profile data
PGO_MODE ?= opt
## OpenMP switch for C/C++ kernels
OPENMP_FLAGS ?= -qopenmp
## Interp_Points load balance profiling mode
## off : (default) no load balance instrumentation
## profile : Pass 1 — instrument Interp_Points to collect timing profile