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

perf(RestrictProlong): 用 Restrict_cached/OutBdLow2Hi_cached 替换非缓存版本，Sync_finish 改为渐进式解包

Browse Source 
- RestrictProlong/RestrictProlong_aux 中的 Restrict() 和 OutBdLow2Hi() 替换为 _cached 版本，
  复用 gridseg 列表和 MPI 缓冲区，避免每次调用重新分配
- 新增 sync_cache_restrict/sync_cache_outbd 两组 per-level 缓存
- Sync_finish 从 MPI_Waitall 改为 MPI_Waitsome 渐进式解包，降低尾延迟
- AsyncSyncState 扩展 req_node/req_is_recv/pending_recv 字段支持渐进解包

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
CGH0S7
2026-03-02 20:48:38 +08:00
parent b3c367f15b
commit 4012e9d068
4 changed files with 527 additions and 483 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
AMSS_NCKU_source/bssn_class.C
View File

@@ -736,6 +736,8 @@ void bssn_class::Initialize()
sync_cache_cor = new Parallel::SyncCache[GH->levels];
sync_cache_rp_coarse = new Parallel::SyncCache[GH->levels];
sync_cache_rp_fine = new Parallel::SyncCache[GH->levels];
sync_cache_restrict = new Parallel::SyncCache[GH->levels];
sync_cache_outbd = new Parallel::SyncCache[GH->levels];
}
//================================================================================================
@@ -2213,7 +2215,7 @@ void bssn_class::Evolve(int Steps)
GH->Regrid(Symmetry, BH_num, Porgbr, Porg0,
SynchList_cor, OldStateList, StateList, SynchList_pre,
fgt(PhysTime - dT_mon, StartTime, dT_mon / 2), ErrorMonitor);
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); sync_cache_restrict[il].invalidate(); sync_cache_outbd[il].invalidate(); }
#endif
#if (REGLEV == 0 && (PSTR == 1 || PSTR == 2))
@@ -2429,7 +2431,7 @@ void bssn_class::RecursiveStep(int lev)
if (GH->Regrid_Onelevel(lev, Symmetry, BH_num, Porgbr, Porg0,
SynchList_cor, OldStateList, StateList, SynchList_pre,
fgt(PhysTime - dT_lev, StartTime, dT_lev / 2), ErrorMonitor))
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); sync_cache_restrict[il].invalidate(); sync_cache_outbd[il].invalidate(); }
#endif
}
@@ -2608,7 +2610,7 @@ void bssn_class::ParallelStep()
if (GH->Regrid_Onelevel(GH->mylev, Symmetry, BH_num, Porgbr, Porg0,
SynchList_cor, OldStateList, StateList, SynchList_pre,
fgt(PhysTime - dT_lev, StartTime, dT_lev / 2), ErrorMonitor))
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); sync_cache_restrict[il].invalidate(); sync_cache_outbd[il].invalidate(); }
#endif
}
@@ -2775,7 +2777,7 @@ void bssn_class::ParallelStep()
if (GH->Regrid_Onelevel(lev + 1, Symmetry, BH_num, Porgbr, Porg0,
SynchList_cor, OldStateList, StateList, SynchList_pre,
fgt(PhysTime - dT_levp1, StartTime, dT_levp1 / 2), ErrorMonitor))
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); sync_cache_restrict[il].invalidate(); sync_cache_outbd[il].invalidate(); }
// a_stream.clear();
// a_stream.str("");
@@ -2790,7 +2792,7 @@ void bssn_class::ParallelStep()
if (GH->Regrid_Onelevel(lev, Symmetry, BH_num, Porgbr, Porg0,
SynchList_cor, OldStateList, StateList, SynchList_pre,
fgt(PhysTime - dT_lev, StartTime, dT_lev / 2), ErrorMonitor))
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); sync_cache_restrict[il].invalidate(); sync_cache_outbd[il].invalidate(); }
// a_stream.clear();
// a_stream.str("");
@@ -2809,7 +2811,7 @@ void bssn_class::ParallelStep()
if (GH->Regrid_Onelevel(lev - 1, Symmetry, BH_num, Porgbr, Porg0,
SynchList_cor, OldStateList, StateList, SynchList_pre,
fgt(PhysTime - dT_lev, StartTime, dT_levm1 / 2), ErrorMonitor))
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); sync_cache_restrict[il].invalidate(); sync_cache_outbd[il].invalidate(); }
// a_stream.clear();
// a_stream.str("");
@@ -2825,7 +2827,7 @@ void bssn_class::ParallelStep()
if (GH->Regrid_Onelevel(lev - 1, Symmetry, BH_num, Porgbr, Porg0,
SynchList_cor, OldStateList, StateList, SynchList_pre,
fgt(PhysTime - dT_lev, StartTime, dT_levm1 / 2), ErrorMonitor))
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); }
for (int il = 0; il < GH->levels; il++) { sync_cache_pre[il].invalidate(); sync_cache_cor[il].invalidate(); sync_cache_rp_coarse[il].invalidate(); sync_cache_rp_fine[il].invalidate(); sync_cache_restrict[il].invalidate(); sync_cache_outbd[il].invalidate(); }
// a_stream.clear();
// a_stream.str("");
@@ -5796,7 +5798,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB,
#endif
#if (RPB == 0)
Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, Symmetry);
Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, Symmetry, sync_cache_restrict[lev]);
#elif (RPB == 1)
// Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SL,SynchList_pre,Symmetry);
Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, GH->rsul[lev], Symmetry);
@@ -5820,7 +5822,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB,
#if (RPB == 0)
#if (MIXOUTB == 0)
Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry);
Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry, sync_cache_outbd[lev]);
#elif (MIXOUTB == 1)
Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry);
#endif
@@ -5847,7 +5849,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB,
#endif
#if (RPB == 0)
Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry, sync_cache_restrict[lev]);
#elif (RPB == 1)
// Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SL,SL,Symmetry);
Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, GH->rsul[lev], Symmetry);
@@ -5871,7 +5873,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB,
#if (RPB == 0)
#if (MIXOUTB == 0)
Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry, sync_cache_outbd[lev]);
#elif (MIXOUTB == 1)
Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
#endif
@@ -5940,7 +5942,7 @@ void bssn_class::RestrictProlong_aux(int lev, int YN, bool BB,
}
#if (RPB == 0)
Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, Symmetry);
Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, Symmetry, sync_cache_restrict[lev]);
#elif (RPB == 1)
// Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SL,SynchList_pre,Symmetry);
Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SL, SynchList_pre, GH->rsul[lev], Symmetry);
@@ -5950,7 +5952,7 @@ void bssn_class::RestrictProlong_aux(int lev, int YN, bool BB,
#if (RPB == 0)
#if (MIXOUTB == 0)
Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry);
Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry, sync_cache_outbd[lev]);
#elif (MIXOUTB == 1)
Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SL, Symmetry);
#endif
@@ -5962,7 +5964,7 @@ void bssn_class::RestrictProlong_aux(int lev, int YN, bool BB,
else // no time refinement levels and for all same time levels
{
#if (RPB == 0)
Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry, sync_cache_restrict[lev]);
#elif (RPB == 1)
// Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SL,SL,Symmetry);
Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, GH->rsul[lev], Symmetry);
@@ -5972,7 +5974,7 @@ void bssn_class::RestrictProlong_aux(int lev, int YN, bool BB,
#if (RPB == 0)
#if (MIXOUTB == 0)
Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry, sync_cache_outbd[lev]);
#elif (MIXOUTB == 1)
Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SL, SL, Symmetry);
#endif
@@ -6027,7 +6029,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB)
}
#if (RPB == 0)
Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, SynchList_pre, Symmetry);
Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, SynchList_pre, Symmetry, sync_cache_restrict[lev]);
#elif (RPB == 1)
// Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SynchList_cor,SynchList_pre,Symmetry);
Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, SynchList_pre, GH->rsul[lev], Symmetry);
@@ -6037,7 +6039,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB)
#if (RPB == 0)
#if (MIXOUTB == 0)
Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry);
Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry, sync_cache_outbd[lev]);
#elif (MIXOUTB == 1)
Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry);
#endif
@@ -6051,7 +6053,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB)
if (myrank == 0)
cout << "===: " << GH->Lt[lev - 1] << "," << GH->Lt[lev] + dT_lev << endl;
#if (RPB == 0)
Parallel::Restrict(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, StateList, Symmetry);
Parallel::Restrict_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, StateList, Symmetry, sync_cache_restrict[lev]);
#elif (RPB == 1)
// Parallel::Restrict_bam(GH->PatL[lev-1],GH->PatL[lev],SynchList_cor,StateList,Symmetry);
Parallel::Restrict_bam(GH->PatL[lev - 1], GH->PatL[lev], SynchList_cor, StateList, GH->rsul[lev], Symmetry);
@@ -6061,7 +6063,7 @@ void bssn_class::RestrictProlong(int lev, int YN, bool BB)
#if (RPB == 0)
#if (MIXOUTB == 0)
Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry);
Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry, sync_cache_outbd[lev]);
#elif (MIXOUTB == 1)
Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry);
#endif
@@ -6102,7 +6104,7 @@ void bssn_class::ProlongRestrict(int lev, int YN, bool BB)
#if (RPB == 0)
#if (MIXOUTB == 0)
Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry);
Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry, sync_cache_outbd[lev]);
#elif (MIXOUTB == 1)
Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], SynchList_pre, SynchList_cor, Symmetry);
#endif
@@ -6115,7 +6117,7 @@ void bssn_class::ProlongRestrict(int lev, int YN, bool BB)
{
#if (RPB == 0)
#if (MIXOUTB == 0)
Parallel::OutBdLow2Hi(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry);
Parallel::OutBdLow2Hi_cached(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry, sync_cache_outbd[lev]);
#elif (MIXOUTB == 1)
Parallel::OutBdLow2Himix(GH->PatL[lev - 1], GH->PatL[lev], StateList, SynchList_cor, Symmetry);
#endif