portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2026-08-27 12:59:36 +01:00
Code Releases Activity

Improvements

Browse Source
This commit is contained in:
Peter Boyle
2026-08-23 02:38:23 -04:00
parent 06fa2fc160
commit 14d2f2c06e
1 changed files with 9 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Grid/algorithms/multigrid/RecursiveSchurInverse.h
+9 -2
View File
@@ -120,6 +120,7 @@ public:
uint64_t bytesAllreduce; uint64_t bytesAllreduce;
uint64_t nAllreduce; // panel collectives uint64_t nAllreduce; // panel collectives
uint64_t nGatherGemm; // GatherGemm calls uint64_t nGatherGemm; // GatherGemm calls
uint64_t nGather; // AllGatherV collectives (debug gate)
// Persistent grow-only device panel; assembly and collectives are // Persistent grow-only device panel; assembly and collectives are
// device-resident. Device builds require GPU-aware MPI. // device-resident. Device builds require GPU-aware MPI.
@@ -390,8 +391,13 @@ public:
for(int r=0;r<P;r++){ csum += counts[r]; if ( counts[r] ) nz++; } for(int r=0;r<P;r++){ csum += counts[r]; if ( counts[r] ) nz++; }
GRID_ASSERT( csum == (int64_t)panelWords ); GRID_ASSERT( csum == (int64_t)panelWords );
GRID_ASSERT( displs[rB1-1] + counts[rB1-1] == (int)panelWords ); GRID_ASSERT( displs[rB1-1] + counts[rB1-1] == (int)panelWords );
if ( gatherDebug && ((int)nAllreduce < gatherDebug) && (me==0) ) { // Gate on the count of GATHERS, not of all collectives. nAllreduce
std::cout << GridLogMessage << "GATHER["<<nAllreduce<<"]" // is already in the hundreds by the first gather -- the recursion is
// depth first and everything below the size threshold takes the
// allreduce path -- so gating on it prints nothing at any sane value.
nGather++;
if ( gatherDebug && ((int)nGather <= gatherDebug) && (me==0) ) {
std::cout << GridLogMessage << "GATHER["<<nGather-1<<"]"
<< " ranks ["<<rB0<<","<<rB1<<")" << " ranks ["<<rB0<<","<<rB1<<")"
<< " k "<<k<<" n "<<n<<" nchunk "<<nchunk << " k "<<k<<" n "<<n<<" nchunk "<<nchunk
<< " panelWords "<<panelWords << " panelWords "<<panelWords
@@ -733,6 +739,7 @@ public:
bytesAllreduce = 0; bytesAllreduce = 0;
nAllreduce = 0; nAllreduce = 0;
nGatherGemm = 0; nGatherGemm = 0;
nGather = 0;
SchurNode(0, P, 0, N, Arows); SchurNode(0, P, 0, N, Arows);