Oct 10, 2025 agenda

Author: DrTimothyAldenDavis

minutes/2025-10-08.md

@@ -0,0 +1,118 @@
+Minutes, Oct 8, 2025
+
+    * agenda:
+
+        - assign one or more algorithms or utilities to everyone for review
+            (see below)
+
+        - new GraphBLAS features (review GxB API):
+
+            - printing of user-defined types 
+
+            - GPU: getting close to production for a few internal kernels
+
+            - C<C> = ... some uses require M=C to be copied, then do C<M>=...
+                kernels that can be done without the copy, where C is aliased
+                with M or A:
+
+                (1) C<C,struct> = scalar, in SSGrB for a long time;
+                        C becomes iso-valued.
+
+                (2) C<C,struct> += A, now takes O(nnz(A)*log(n)) time, at most.
+                        C can have zombies.  No new entries added to C and
+                        no entries deleted.  To appear in v10.2.0.
+
+            - possible new features, to exploit C==M and/or C==A aliasing:
+
+                These 2 kernels are in FalkorDB which I don't have a special
+                case:
+
+                    (3) C<!M,struct,replace> = C, which I think only does
+                        deletions; my current assign kernels might be OK.
+                        Always followed by a GrB_wait.
+
+                    (4) C<M,struct>=M, which only does additions in FalkorDB
+                        but could do updates to existing entries in C in
+                        general.  my current assign kernels might be OK.
+                        Always followed by a GrB_wait.
+
+                other possible aliases to exploit:
+
+                    (5) C<C,struct> = A, can lead to updates or deletions
+
+                    (6) C<A,struct> += A, possibly with replace
+
+                    (7) C<!M, struct> += M, possibly with replace
+
+                Are these cases important to LAGraph, python-graphblas,
+                FalkorDB, etc?
+
+--------------------------------------------------------------------------------
+
+    consider these methods for src.  Minor revisions needed in some of them,
+    including API revisions:
+
+        LAGr_BreadthFirstSearch_Extended.c: uses LG_:
+            LG_BreadthFirstSearch_SSGrB_Extended.c
+            LG_BreadthFirstSearch_vanilla_Extended.c
+        LAGr_PeerPressureClustering.c
+        LAGraph_AllKCore.c: need both basic & adv methods; this is mixed
+        LAGraph_AllKTruss.c
+        LAGraph_HITS.c
+        LAGraph_Hdip_Fiedler.c: needs lots of polish
+        LAGraph_KCore.c: need both basic & adv
+        LAGraph_KTruss.c
+        LAGraph_MaximalIndependentSet.c: rename LAGr_, need a basic method
+        LAGraph_SquareClustering.c
+        LAGraph_VertexCentrality_Triangle.c
+        LAGraph_coloring_MIS.c
+        LAGraph_coloring_independent_set.c
+        LAGraph_lcc.c
+        LAGraph_cdlp.c: use Container methods
+
+    abandon the need for vanilla methods and just return GrB_NOT_IMPLEMENTED
+    if not using SuiteSparse/GraphBLAS.h.
+
+        LAGr_EdgeBetweennessCentrality.c
+        LAGr_MarkovClustering.c (but see latest issue posted)
+        LAGr_Modularity.c
+        LAGr_PartitionQuality.c
+        LAGr_SwapEdges.c
+        LAGraph_SwapEdges.c
+        LAGraph_CFL_reachability.c
+        LAGraph_EstimateDiameter.c
+        LAGraph_ExactDiameter.c
+        LAGraph_FastGraphletTransform.c
+        LAGraph_KCoreDecompose.c
+        LAGraph_MultiSourceBFS.c
+        LAGraph_RegularPathQuery.c
+        LAGraph_RichClubCoefficient.c
+        LAGraph_scc.c
+        LG_CC_FastSV5.c
+        LG_CC_FastSV7_FA.c
+
+    Algorithms need work:
+
+        LAGr_MaxFlow.c (need to provide output R matrix?  And flow_matrix is
+                expensive to compute; revise)
+        LAGraph_argminmax.c
+        LAGraph_msf.c
+        LAGraph_BF_*.c: need just one variant
+
+    need new LAGraph_Graph enums:
+
+        LAGraph_MaximalMatching.c: need LAGraph_Graph incidence matrix
+        LAGraph_Coarsen_Matching.c: need LAGraph_Graph incidence matrix,
+            and vanilla method
+        LAGr_MaximumMatching.c:  need to switch to Container methods;
+            need LAGraph_Graph bipartite graph, and vanilla method
+        LAGraph_cc_lacc.c: unsure
+
+    do not consider:
+
+        LAGr_PageRankGX.c: for LDBC benchmarks only
+        LAGr_TriangleCount_GPU.c: for testing draft GPU kernels only
+        LAGraph_dnn.c: for an IEEE HPEC Graph Challenge only
+        LAGraph_HelloWorld.c: will always be experimental
+        LAGraph_cdlp_withsort.c: OK, but slow, see LAGraph_cdlp instead
+