AI/rocm-systems
2
0
Atdalīts 0
Kods Problēmas Izmaiņu pieprasījumi Darbības Pakotnes Projekti Laidieni Vikivietne Aktivitāte

Add Rome model and improve search (#305)

Pārlūkot izejas kodu 
[ROCm/rccl commit: 975b14dffa]
Šī revīzija ir iekļauta:
Wenkai Du
2020-11-17 14:55:06 -08:00
revīziju iesūtīja GitHub
vecāks 32fdfc93fc
revīzija b68ff1ebba
5 mainīti faili ar 123 papildinājumiem un 2 dzēšanām
Lejupielādēt ielāpa failu Lejupielādēt izmaiņu failu Izvērst visus failus Savērst visus failus
projects/rccl/src/graph/rome_models.h
+24
Parādīt failu
@@ -213,6 +213,28 @@ static struct rcclRomeModel rome_model_44 = {
.ringBase = "5 4 7 6 2 1 3 0|5 6 7 4 1 0 2 3",
};
static struct rcclRomeModel rome_model_45 = {
.nGpus = 8, .nCpus = 7, .nNics = 0, .nLinks = 3,
.gpuIds = { 0x43000, 0x23000, 0x26000, 0x3000, 0xc3000, 0xc6000, 0xa3000, 0x83000, },
.nicIds = { },
.gpuNuma = { 1, 2, 2, 3, 5, 5, 6, 7, },
.nicNuma = { },
.connMatrix = { 0, 1, 1, 1, 0, 0, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 0, 0, 1, 1, 1, 0, },
.pattern = "00102010002010",
.ringBase = "0 1 2 3 4 5 6 7|0 2 5 7 4 6 1 3|0 3 1 6 4 7 5 2|0 7 6 5 4 3 2 1",
};
static struct rcclRomeModel rome_model_46 = {
.nGpus = 8, .nCpus = 7, .nNics = 1, .nLinks = 3,
.gpuIds = { 0x43000, 0x23000, 0x26000, 0x3000, 0xc3000, 0xc6000, 0xa3000, 0x83000, },
.nicIds = { 0xe1000, },
.gpuNuma = { 1, 2, 2, 3, 5, 5, 6, 7, },
.nicNuma = { 4, },
.connMatrix = { 0, 1, 1, 1, 0, 0, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 0, 0, 1, 1, 1, 0, },
.pattern = "00102010012010",
.ringBase = "6 5 7 4 1 2 3 0|7 4 6 5 1 0 3 2",
};
static struct rcclRomeModel romeTopoModels[] = {
rome_model_22,
rome_model_25,
@@ -230,4 +252,6 @@ static struct rcclRomeModel romeTopoModels[] = {
rome_model_40,
rome_model_42,
rome_model_44,
rome_model_45,
rome_model_46,
};
projects/rccl/src/graph/search.cc
+3 -1
Parādīt failu
@@ -973,8 +973,10 @@ static bool permuteGpuIds(int *g, int n, int last, struct rcclRomeModel* ref, st
if (i < ref->nGpus) return false;
// match XGMI connection
for (i = 0; i < ref->nGpus; i++) {
for (j = 0; j < ref->nGpus; j++)
for (j = 0; j < ref->nGpus; j++) {
if (ref->connMatrix[i*ref->nGpus+j] != topo->connMatrix[g[i]*ref->nGpus+g[j]]) break;
if ((ref->gpuIds[i]-ref->gpuIds[j])*(topo->gpuIds[g[i]]-topo->gpuIds[g[j]]) < 0) break;
}
if (j < ref->nGpus) break;
}
if (i < ref->nGpus) return false;
projects/rccl/tools/scripts/topo_val.sh
+1 -1
Parādīt failu
@@ -21,7 +21,7 @@
DIR="$(cd -P "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
for i in {0..44}
for i in {0..46}
do
$DIR/../topo_expl/topo_expl -m $i > "topo_m$i.log"
$DIR/../TopoVisual/topo_visual.sh -i "topo_m$i.log"
projects/rccl/tools/topo_expl/models/topo_8p_rome_n4_1.xml
+93
Parādīt failu
@@ -0,0 +1,93 @@
<system version="2">
<cpu numaid="1" affinity="00000000,00000000,00000000,ffff0000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:41:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:43:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="0" sm="90" gcn="908" arch="38911" rank="0" gdr="1">
<xgmi target="0000:23:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:26:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="2" affinity="00000000,00000000,0000ffff,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:21:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:23:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="1" sm="90" gcn="908" arch="38911" rank="1" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:26:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:24:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:26:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="2" sm="90" gcn="908" arch="38911" rank="2" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:23:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:03:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="3" affinity="00000000,00000000,ffff0000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:01:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:03:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="3" sm="90" gcn="908" arch="38911" rank="3" gdr="1">
<xgmi target="0000:43:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:23:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:26:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="5" affinity="00000000,ffff0000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:c1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="4" sm="90" gcn="908" arch="38911" rank="4" gdr="1">
<xgmi target="0000:c6:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:c4:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:c6:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="5" sm="90" gcn="908" arch="38911" rank="5" gdr="1">
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="6" affinity="0000ffff,00000000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:a1:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:a3:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="6" sm="90" gcn="908" arch="38911" rank="6" gdr="1">
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c6:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:83:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="7" affinity="ffff0000,00000000,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:81:00.0" class="0x060400" link_speed="16 GT/s" link_width="16">
<pci busid="0000:83:00.0" class="0x038000" link_speed="16 GT/s" link_width="16">
<gpu dev="7" sm="90" gcn="908" arch="38911" rank="7" gdr="1">
<xgmi target="0000:c3:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:c6:00.0" count="1" tclass="0x038000"/>
<xgmi target="0000:a3:00.0" count="1" tclass="0x038000"/>
</gpu>
</pci>
</pci>
</cpu>
<cpu numaid="4" affinity="00000000,0000ffff,00000000,00000000" arch="x86_64" vendor="AuthenticAMD" familyid="143" modelid="49">
<pci busid="0000:e1:00.0" class="0x020700" link_speed="16 GT/s" link_width="16">
<nic>
<net name="mlx5_0" dev="0" speed="200000" port="1" guid="0x42fc9f00039b0398" maxconn="262144" gdr="1"/>
</nic>
</pci>
</cpu>
</system>
projects/rccl/tools/topo_expl/topo_expl.cpp
+2
Parādīt failu
@@ -114,6 +114,8 @@ NodeModelDesc model_descs[] = {
{4, "topo_4p3l_n4.xml", "4 nodes 8 gfx908 Rome NPS=4"},
{1, "topo_4p3l_n2_1.xml", "single node 8 gfx908 Rome"},
{4, "topo_4p3l_n2_1.xml", "4 nodes 8 gfx908 Rome"},
{1, "topo_8p_rome_n4_1.xml", "single node 8 gfx908 Rome NPS=4"},
{4, "topo_8p_rome_n4_1.xml", "4 nodes node 8 gfx908 Rome NPS=4"},
};
int main(int argc,char* argv[])