Add another Rome model (#553)

* Add another Rome model

* Add option to force enable intranet on single node

* Limit p2p channels to number of ranks

* Refine p2p channels handling
This commit is contained in:
Wenkai Du
2022-05-31 11:31:30 -07:00
committed by GitHub
parent a0a686e74c
commit ef499c4810
7 changed files with 160 additions and 5 deletions
+1 -1
View File
@@ -21,7 +21,7 @@
DIR="$(cd -P "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
for i in {0..76}
for i in {0..78}
do
if [[ $i -eq 50 ]] || [[ $i -eq 51 ]]
then
+146
View File
@@ -0,0 +1,146 @@
<system version="2">
<cpu numaid="0" affinity="00000000,00000000,ffffffff,ffffffff,00000000,00000000,ffffffff,ffffffff" arch="x86_64" vendor="AuthenticAMD" familyid="175" modelid="1">
<pci busid="0000:22:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:2e:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:30:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x1022" subsystem_device="0x14c7" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:32:00.0" class="0x038000" vendor="0x1002" device="0x740c" 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="1">
<xgmi target="0000:35:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:11:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:8e:00.0" count="2" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:33:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x1022" subsystem_device="0x14c7" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:35:00.0" class="0x038000" vendor="0x1002" device="0x740c" 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="1">
<xgmi target="0000:32:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:14:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:b3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:24:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:26:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0007" link_speed="16.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_1" dev="0" speed="200000" port="1" latency="0.000000" guid="0x203d2b0003fd7010" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
<pci busid="0000:2b:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:2d:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0007" link_speed="16.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_2" dev="1" speed="200000" port="1" latency="0.000000" guid="0x503d2b0003fd7010" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</pci>
<pci busid="0000:01:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:0d:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:0f:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x1022" subsystem_device="0x14c7" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:11:00.0" class="0x038000" vendor="0x1002" device="0x740c" 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="1">
<xgmi target="0000:32:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:14:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:ae:00.0" count="2" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:12:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x1022" subsystem_device="0x14c7" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:14:00.0" class="0x038000" vendor="0x1002" device="0x740c" 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="1">
<xgmi target="0000:35:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:11:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:93:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:03:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:05:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0007" link_speed="16.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_3" dev="2" speed="200000" port="1" latency="0.000000" guid="0x143d2b0003fd7010" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
<pci busid="0000:0a:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:0c:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0007" link_speed="16.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_4" dev="3" speed="200000" port="1" latency="0.000000" guid="0x243d2b0003fd7010" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</pci>
</cpu>
<cpu numaid="1" affinity="ffffffff,ffffffff,00000000,00000000,ffffffff,ffffffff,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="175" modelid="1">
<pci busid="0000:a1:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:a9:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:ac:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x1022" subsystem_device="0x14c7" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:ae:00.0" class="0x038000" vendor="0x1002" device="0x740c" 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="1">
<xgmi target="0000:11:00.0" count="2" tclass="0x038000"/>
<xgmi target="0000:b3:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:8e:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:ab:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0007" link_speed="16.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_6" dev="4" speed="200000" port="1" latency="0.000000" guid="0x883a2b0003fd7010" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
<pci busid="0000:af:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:b1:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x1022" subsystem_device="0x14c7" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:b3:00.0" class="0x038000" vendor="0x1002" device="0x740c" 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="1">
<xgmi target="0000:35:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:ae:00.0" count="4" tclass="0x038000"/>
<xgmi target="0000:93:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:b4:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0007" link_speed="16.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_7" dev="5" speed="200000" port="1" latency="0.000000" guid="0xf83c2b0003fd7010" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</pci>
<pci busid="0000:81:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:89:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:8c:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x1022" subsystem_device="0x14c7" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:8e:00.0" class="0x038000" vendor="0x1002" device="0x740c" 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="1">
<xgmi target="0000:32:00.0" count="2" tclass="0x038000"/>
<xgmi target="0000:ae:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:93:00.0" count="4" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:8b:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0007" link_speed="16.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_8" dev="6" speed="200000" port="1" latency="0.000000" guid="0x2c3d2b0003fd7010" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
<pci busid="0000:8f:00.0" class="0x060400" vendor="0x1000" device="0xc010" subsystem_vendor="0x1000" subsystem_device="0xa096" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:91:00.0" class="0x060400" vendor="0x1022" device="0x14c7" subsystem_vendor="0x1022" subsystem_device="0x14c7" link_speed="16.0 GT/s PCIe" link_width="16">
<pci busid="0000:93:00.0" class="0x038000" vendor="0x1002" device="0x740c" 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="1">
<xgmi target="0000:14:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:b3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:8e:00.0" count="4" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:94:00.0" class="0x020000" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0007" link_speed="16.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_9" dev="7" speed="200000" port="1" latency="0.000000" guid="0x283d2b0003fd7010" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</pci>
</cpu>
</system>
+2
View File
@@ -147,6 +147,8 @@ NodeModelDesc model_descs[] = {
{4, "topo_8p1h_3.xml", "4 nodes 8P1H Alt."},
{1, "topo_8p1h_4.xml", "Single node 8P1H Alt."},
{2, "topo_8p1h_4.xml", "2 nodes 8P1H Alt."},
{1, "topo_8p1h_5.xml", "Single node 8P1H Alt."},
{2, "topo_8p1h_5.xml", "2 nodes 8P1H Alt."},
};
int main(int argc,char* argv[])