AI/rocm-systems
2
0
Derivar 0
Código Questões Pedidos de integração Operações Pacotes Planeamentos Lançamentos Wiki Trabalho

Add new Rome model (#535)

Ver a proveniência 
[ROCm/rccl commit: ba4c165bf3]
Este cometimento está contido em:
Wenkai Du
2022-04-12 13:27:32 -07:00
cometido por GitHub
ascendente be9374aa34
cometimento f8023f2e07
5 ficheiros modificados com 93 adições e 2 eliminações
Descarregar ficheiro patch Descarregar ficheiro diff Expandir todos os ficheiros Colapsar todos os ficheiros
projects/rccl/src/graph/rome_models.cc
+14
Ver ficheiro
@@ -500,6 +500,19 @@ static struct rcclRomeModel rome_model_72 = {
.options = "disableNumaMatching=1,tuning=2",
};
static struct rcclRomeModel rome_model_73 = {
.nGpus = 8, .nCpus = 4, .nNics = 0, .nLinks = 3,
.gpuIds = { 0xc1000, 0xc6000, 0xc9000, 0xce000, 0xd1000, 0xd6000, 0xd9000, 0xde000, },
.nicIds = { },
.gpuNuma = { 3, 3, 1, 1, 0, 0, 2, 2, },
.nicNuma = { },
.connMatrix = { 0, 4, 1, 0, 0, 0, 2, 0, 4, 0, 0, 1, 0, 1, 0, 0, 1, 0, 0, 4, 2, 0, 0, 0, 0, 1, 4, 0, 0, 0, 0, 1, 0, 0, 2, 0, 0, 4, 1, 0, 0, 1, 0, 0, 4, 0, 0, 1, 2, 0, 0, 0, 1, 0, 0, 4, 0, 0, 0, 1, 0, 1, 4, 0, },
.gdrLevel = { },
.pattern = "20202020",
.ringBase = "0 1 3 2 4 5 7 6|6 7 5 4 2 3 1 0|0 1 5 4 6 7 3 2|2 3 7 6 4 5 1 0",
.options = "",
};
static struct rcclRomeModel romeTopoModels[] = {
rome_model_22,
rome_model_25,
@@ -536,6 +549,7 @@ static struct rcclRomeModel romeTopoModels[] = {
rome_model_68,
rome_model_71,
rome_model_72,
rome_model_73,
};
/* Parse user defined rings. Format is like :
projects/rccl/tools/scripts/topo_val.sh
+1 -1
Ver ficheiro
@@ -21,7 +21,7 @@
DIR="$(cd -P "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
for i in {0..72}
for i in {0..73}
do
if [[ $i -eq 50 ]] || [[ $i -eq 51 ]]
then
projects/rccl/tools/topo_expl/Makefile
+1 -1
Ver ficheiro
@@ -6,7 +6,7 @@ endif
HIPCC = $(HIP_PATH)/bin/hipcc
EXE = topo_expl
CXXFLAGS = -g -O3 -Iinclude -I../../src -I../../src/include -I../../src/graph/ -I/opt/rocm/rocm_smi/include/ -DTOPO_EXPL -DENABLE_TRACE -lnuma
CXXFLAGS = -g -O3 -Iinclude -I../../src -I../../src/include -I../../src/graph/ -I/opt/rocm/rocm_smi/include/ -DTOPO_EXPL -DENABLE_TRACE
files = $(EXE).cpp model.cpp utils.cpp ../../src/graph/topo.cc ../../src/graph/rings.cc ../../src/graph/paths.cc ../../src/graph/trees.cc \
../../src/graph/search.cc ../../src/graph/connect.cc ../../src/graph/tuning.cc ../../src/graph/xml.cc ../../src/misc/nvmlwrap_stub.cc ../../src/graph/rome_models.cc
projects/rccl/tools/topo_expl/models/topo_8p1h_2.xml
+76
Ver ficheiro
@@ -0,0 +1,76 @@
<system version="2">
<cpu numaid="3" affinity="ffff0000,00000000,ffff0000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="175" modelid="48">
<pci busid="0000:c1:00.0" class="0x038000" vendor="0x1002" device="0x7408" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="16.0 GT/s PCIe" link_width="16">
<gpu dev="0" sm="90" gcn="910" arch="38911" rank="0" gdr="0">
<xgmi target="0000:c6:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:c9:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d9:00.0" count="2" tclass="0x038000"/>
</gpu>
</pci>
<pci busid="0000:c6:00.0" class="0x038000" vendor="0x1002" device="0x7408" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="16.0 GT/s PCIe" link_width="16">
<gpu dev="1" sm="90" gcn="910" arch="38911" rank="1" gdr="0">
<xgmi target="0000:c1:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:ce:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d6:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
<nic>
<net name="vethe2d5378" dev="1" speed="10000" port="0" guid="0x1" maxconn="65536" gdr="0"/>
</nic>
</cpu>
<cpu numaid="1" affinity="00000000,ffff0000,00000000,ffff0000" arch="x86_64" vendor="AuthenticAMD" familyid="175" modelid="48">
<pci busid="0000:c9:00.0" class="0x038000" vendor="0x1002" device="0x7408" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="16.0 GT/s PCIe" link_width="16">
<gpu dev="2" sm="90" gcn="910" arch="38911" rank="2" gdr="0">
<xgmi target="0000:c1:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:ce:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:d1:00.0" count="2" tclass="0x038000"/>
</gpu>
</pci>
<pci busid="0000:ce:00.0" class="0x038000" vendor="0x1002" device="0x7408" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="16.0 GT/s PCIe" link_width="16">
<gpu dev="3" sm="90" gcn="910" arch="38911" rank="3" gdr="0">
<xgmi target="0000:c6:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c9:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:de:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</cpu>
<cpu numaid="0" affinity="00000000,0000ffff,00000000,0000ffff" arch="x86_64" vendor="AuthenticAMD" familyid="175" modelid="48">
<pci busid="0000:d1:00.0" class="0x038000" vendor="0x1002" device="0x7408" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="16.0 GT/s PCIe" link_width="16">
<gpu dev="4" sm="90" gcn="910" arch="38911" rank="4" gdr="0">
<xgmi target="0000:c9:00.0" count="2" tclass="0x038000"/>
<xgmi target="0000:d6:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:d9:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
<pci busid="0000:d6:00.0" class="0x038000" vendor="0x1002" device="0x7408" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="16.0 GT/s PCIe" link_width="16">
<gpu dev="5" sm="90" gcn="910" arch="38911" rank="5" gdr="0">
<xgmi target="0000:c6:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d1:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:de:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
<pci busid="0000:91:00.0" class="0x060400" vendor="0x11f8" device="0x4028" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16.0 GT/s PCIe" link_width="4">
<pci busid="0000:94:00.0" class="0x020000" vendor="0x8086" device="0x1537" subsystem_vendor="0xffff" subsystem_device="0x0000" link_speed="2.5 GT/s PCIe" link_width="1">
<nic>
<net name="ens2" dev="0" speed="1000" port="0" guid="0x0" maxconn="65536" gdr="0"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="2" affinity="0000ffff,00000000,0000ffff,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="175" modelid="48">
<pci busid="0000:d9:00.0" class="0x038000" vendor="0x1002" device="0x7408" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="16.0 GT/s PCIe" link_width="16">
<gpu dev="6" sm="90" gcn="910" arch="38911" rank="6" gdr="0">
<xgmi target="0000:c1:00.0" count="2" tclass="0x038000"/>
<xgmi target="0000:d1:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:de:00.0" count="4" tclass="0x038000"/>
</gpu>
</pci>
<pci busid="0000:de:00.0" class="0x038000" vendor="0x1002" device="0x7408" subsystem_vendor="0x1002" subsystem_device="0x0b0c" link_speed="16.0 GT/s PCIe" link_width="16">
<gpu dev="7" sm="90" gcn="910" arch="38911" rank="7" gdr="0">
<xgmi target="0000:ce:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d6:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d9:00.0" count="4" tclass="0x038000"/>
</gpu>
</pci>
</cpu>
</system>
projects/rccl/tools/topo_expl/topo_expl.cpp
+1
Ver ficheiro
@@ -142,6 +142,7 @@ NodeModelDesc model_descs[] = {
{4, "topo_8p1h_n1.xml", "4 nodes 8P1H"},
{1, "topo_8p1h_1.xml", "single node 8P1H Alt."},
{4, "topo_8p1h_1.xml", "4 nodes 8P1H Alt."},
{1, "topo_8p1h_2.xml", "single node 8P1H Alt."},
};
int main(int argc,char* argv[])