AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add new Rome model (#536)

Browse Source 
[ROCm/rccl commit: 2151c79d14]
This commit is contained in:
Wenkai Du
2022-04-13 11:45:40 -07:00
committed by GitHub
parent f8023f2e07
commit 011447e4dc
4 changed files with 102 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/graph/rome_models.cc
+14
View File
@@ -513,6 +513,19 @@ static struct rcclRomeModel rome_model_73 = {
.options = "",
};
static struct rcclRomeModel rome_model_74 = {
.nGpus = 8, .nCpus = 4, .nNics = 4, .nLinks = 3,
.gpuIds = { 0xc1000, 0xc6000, 0xc9000, 0xce000, 0xd1000, 0xd6000, 0xd9000, 0xde000, },
.nicIds = { 0xc5000, 0xcd000, 0xd5000, 0xdd000, },
.gpuNuma = { 3, 3, 1, 1, 0, 0, 2, 2, },
.nicNuma = { 3, 1, 0, 2, },
.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 = { PATH_PHB, PATH_PHB, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_PHB, PATH_PHB, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_PHB, PATH_PHB, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_SYS, PATH_PHB, PATH_PHB, },
.pattern = "21212121",
.ringBase = "N0 0 1 5 4 6 7 3 2 N1|N1 2 3 7 6 4 5 1 0 N0|N3 7 6 0 1 3 2 4 5 N2|N2 5 4 2 3 1 0 6 7 N3|N0 0 1 5 4 6 7 3 2 N1|N1 2 3 7 6 4 5 1 0 N0|N3 7 6 0 1 3 2 4 5 N2|N2 5 4 2 3 1 0 6 7 N3",
.options = "tuning=3",
};
static struct rcclRomeModel romeTopoModels[] = {
rome_model_22,
rome_model_25,
@@ -550,6 +563,7 @@ static struct rcclRomeModel romeTopoModels[] = {
rome_model_71,
rome_model_72,
rome_model_73,
rome_model_74,
};
/* Parse user defined rings. Format is like :
projects/rccl/tools/scripts/topo_val.sh
+1 -1
View File
@@ -21,7 +21,7 @@
DIR="$(cd -P "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
for i in {0..73}
for i in {0..74}
do
if [[ $i -eq 50 ]] || [[ $i -eq 51 ]]
then
projects/rccl/tools/topo_expl/models/topo_8p1h_3.xml
+86
View File
@@ -0,0 +1,86 @@
<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 GT/s" link_width="16">
<gpu dev="0" sm="90" gcn="910" arch="38911" rank="0" gdr="1">
<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 GT/s" link_width="16">
<gpu dev="1" sm="90" gcn="910" arch="38911" rank="1" gdr="1">
<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>
<pci busid="0000:c5:00.0" class="0x020000" vendor="0x17db" device="0x0501" subsystem_vendor="0x17db" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="net0" dev="0" speed="200000" port="1" guid="0x0" maxconn="128" gdr="1"/>
</nic>
</pci>
</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 GT/s" link_width="16">
<gpu dev="2" sm="90" gcn="910" arch="38911" rank="2" gdr="1">
<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 GT/s" link_width="16">
<gpu dev="3" sm="90" gcn="910" arch="38911" rank="3" gdr="1">
<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>
<pci busid="0000:cd:00.0" class="0x020000" vendor="0x17db" device="0x0501" subsystem_vendor="0x17db" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="net1" dev="1" speed="200000" port="1" guid="0x1" maxconn="128" gdr="1"/>
</nic>
</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 GT/s" link_width="16">
<gpu dev="4" sm="90" gcn="910" arch="38911" rank="4" gdr="1">
<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 GT/s" link_width="16">
<gpu dev="5" sm="90" gcn="910" arch="38911" rank="5" gdr="1">
<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:d5:00.0" class="0x020000" vendor="0x17db" device="0x0501" subsystem_vendor="0x17db" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="net2" dev="2" speed="200000" port="1" guid="0x2" maxconn="128" gdr="1"/>
</nic>
</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 GT/s" link_width="16">
<gpu dev="6" sm="90" gcn="910" arch="38911" rank="6" gdr="1">
<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 GT/s" link_width="16">
<gpu dev="7" sm="90" gcn="910" arch="38911" rank="7" gdr="1">
<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>
<pci busid="0000:dd:00.0" class="0x020000" vendor="0x17db" device="0x0501" subsystem_vendor="0x17db" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<nic>
<net name="net3" dev="3" speed="200000" port="1" guid="0x3" maxconn="128" gdr="1"/>
</nic>
</pci>
</cpu>
</system>
projects/rccl/tools/topo_expl/topo_expl.cpp
+1
View File
@@ -143,6 +143,7 @@ NodeModelDesc model_descs[] = {
{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."},
{4, "topo_8p1h_3.xml", "4 nodes 8P1H Alt."},
};
int main(int argc,char* argv[])