Gcn arch name (#886)

We use CMake to determine if we're compiling against a version of ROCm that supports gcnArchName and handles architecture checking appropriately. It includes a few helper functions as drop ins for the functionality we used gcnArch for before; sometimes to enable flags, and sometimes to set frequencies.

tools/topo_expl/Makefile

@@ -9,7 +9,7 @@ EXE = topo_expl
 CXXFLAGS = -g -Iinclude -Ihipify_rccl/include -Ihipify_rccl/graph -I/opt/rocm/include/ -DTOPO_EXPL -DENABLE_TRACE -DNVTX_NO_IMPL
 
 files = $(EXE).cpp model.cpp utils.cpp hipify_rccl/graph/topo.cc hipify_rccl/graph/rings.cc hipify_rccl/graph/paths.cc hipify_rccl/graph/trees.cc ../../src/misc/param.cc \
-	hipify_rccl/graph/search.cc hipify_rccl/graph/connect.cc hipify_rccl/graph/tuning.cc hipify_rccl/graph/xml.cc ../../src/misc/nvmlwrap_stub.cc hipify_rccl/graph/rome_models.cc
+	hipify_rccl/graph/search.cc hipify_rccl/graph/connect.cc hipify_rccl/graph/tuning.cc hipify_rccl/graph/xml.cc ../../src/misc/nvmlwrap_stub.cc hipify_rccl/graph/rome_models.cc hipify_rccl/graph/archinfo.cc
 
 all: hipify $(EXE)
 
@@ -21,6 +21,7 @@ hipify:
 	mkdir -p hipify_rccl
 	cp -a ../../src/include/ hipify_rccl/
 	cp -a ../../src/graph/ hipify_rccl/
+	cp -ar ../../src/misc/archinfo.cc hipify_rccl/graph/
 	hipify-perl -inplace -quiet-warnings hipify_rccl/include/*.h
 	hipify-perl -inplace -quiet-warnings hipify_rccl/graph/*
 

tools/topo_expl/utils.cpp

@@ -862,17 +862,17 @@ ncclResult_t initTransportsRank_1(struct ncclComm* comm, struct allGatherInfo *a
   NCCLCHECK(ncclTopoIdToIndex(comm->topo, GPU, comm->busId, &idx));
   allGather3Data[rank].nc = 2;
   if (comm->topo->nodes[GPU].count == comm->topo->nRanks &&
-       comm->topo->nodes[GPU].nodes[idx].gpu.gcn == 906 && allXgmi)
+      IsArchMatch(comm->topo->nodes[GPU].nodes[idx].gpu.gcn, "gfx906") && allXgmi)
     allGather3Data[rank].nc = 4;
-  if (comm->topo->nodes[GPU].nodes[idx].gpu.gcn == 908)
+  if (IsArchMatch(comm->topo->nodes[GPU].nodes[idx].gpu.gcn, "gfx908"))
     allGather3Data[rank].nc = std::max(4/ringGraph.nChannels, 2);
   if (comm->topo->nodes[GPU].count == comm->topo->nRanks &&
        (comm->topo->type & RCCL_TOPO_CR8G))
     allGather3Data[rank].nc = 4;
   if (comm->topo->nodes[GPU].count == comm->topo->nRanks &&
-      comm->topo->nodes[GPU].nodes[idx].gpu.gcn == 910)
+      IsArchMatch(comm->topo->nodes[GPU].nodes[idx].gpu.gcn, "gfx90a"))
     allGather3Data[rank].nc = 4;
-  if (comm->topo->nodes[GPU].nodes[idx].gpu.gcn == 910)
+  if (IsArchMatch(comm->topo->nodes[GPU].nodes[idx].gpu.gcn, "gfx90a"))
     allGather3Data[rank].nc = std::max(allGather3Data[rank].nc, 4/ringGraph.nChannels);
   if (ringGraph.nChannels > MAXCHANNELS/2)
     allGather3Data[rank].nc = 1;