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

Update Rome models matching (#376)

Browse Source
This commit is contained in:
Wenkai Du
2021-05-25 10:12:40 -07:00
committed by GitHub
parent 8e817ecd6d
commit 4c83adb75c
4 changed files with 168 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
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..52}
for i in {0..53}
do
if [[ $i -eq 50 ]] || [[ $i -eq 51 ]]
then
tools/topo_expl/models/topo_8p_rome_4n_1.xml
+126
View File
@@ -0,0 +1,126 @@
<system version="2">
<cpu numaid="1" affinity="00000000,00000000,00000000,ffff0000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:43:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:46:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:48:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:4a:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="0" sm="90" gcn="908" arch="38911" rank="0" gdr="1">
<xgmi target="0000:50:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0f:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:4c:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:4e:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:50:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="90" gcn="908" arch="38911" rank="1" gdr="1">
<xgmi target="0000:4a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0f:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:45:00.0" class="0x020700" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0009" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_1" dev="0" speed="200000" port="1" guid="0x30654a0003a1420c" maxconn="262144" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="3" affinity="00000000,00000000,ffff0000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:03:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:05:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:08:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:0a:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="90" gcn="908" arch="38911" rank="2" gdr="1">
<xgmi target="0000:4a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:50:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0f:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:0b:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:0d:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:0f:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="3" sm="90" gcn="908" arch="38911" rank="3" gdr="1">
<xgmi target="0000:4a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:50:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:0a:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:13:00.0" class="0x020700" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0009" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_3" dev="1" speed="200000" port="1" guid="0x906b4a0003a1420c" maxconn="262144" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="5" affinity="00000000,ffff0000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:c4:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c7:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c9:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:cb:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="4" sm="90" gcn="908" arch="38911" rank="4" gdr="1">
<xgmi target="0000:d1:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:8a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:90:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:cd:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:cf:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:d1:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="90" gcn="908" arch="38911" rank="5" gdr="1">
<xgmi target="0000:cb:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:8a:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:90:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:c6:00.0" class="0x020700" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0009" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_5" dev="2" speed="200000" port="1" guid="0xb0f3b100039f59b8" maxconn="262144" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="7" affinity="ffff0000,00000000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:83:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:86:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:88:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:8a:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="6" sm="90" gcn="908" arch="38911" rank="6" gdr="1">
<xgmi target="0000:cb:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d1:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:90:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:8c:00.0" class="0x060400" vendor="0x11f8" device="0x4000" subsystem_vendor="0x11f8" subsystem_device="0xbeef" link_speed="16 GT/s" link_width="16">
<pci busid="0000:8e:00.0" class="0x060400" vendor="0x1002" device="0x14a0" subsystem_vendor="0x0000" subsystem_device="0x0000" link_speed="16 GT/s" link_width="16">
<pci busid="0000:90:00.0" class="0x038000" vendor="0x1002" device="0x738c" subsystem_vendor="0x1002" subsystem_device="0x0c34" link_speed="16 GT/s" link_width="16">
<gpu dev="7" sm="90" gcn="908" arch="38911" rank="7" gdr="1">
<xgmi target="0000:cb:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:d1:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:8a:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:85:00.0" class="0x020700" vendor="0x15b3" device="0x101b" subsystem_vendor="0x15b3" subsystem_device="0x0009" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_7" dev="3" speed="200000" port="1" guid="0x50815600039f59b8" maxconn="262144" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
</system>
tools/topo_expl/topo_expl.cpp
+1
View File
@@ -122,6 +122,7 @@ NodeModelDesc model_descs[] = {
{4, "topo_collnet_n1.xml", "4 nodes collnet 1 NICs"},
{4, "topo_collnet_n4.xml", "4 nodes collnet 4 NICs"},
{1, "topo_8p_90a.xml", "single node gfx90a"},
{4, "topo_8p_rome_4n_1.xml", "4 nodes node 8 gfx908 Rome 4 NICs NPS=4"},
};
int main(int argc,char* argv[])