AI/rocm-systems
2
0
Форкнуть 0
Код Задачи Запросы на слияние Действия Пакеты Проекты Релизы Вики Активность

Introduce allgather for MSCCL on 8 sockets up to 320KB. (#931)

Просмотр исходного кода
Этот коммит содержится в:
Wen-Heng (Jack) Chung
2023-10-24 18:41:12 -05:00
коммит произвёл GitHub
родитель 3f9ffe4788
Коммит bfb8642450
1 изменённых файлов: 242 добавлений и 0 удалений
Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы
tools/msccl-algorithms/allgather-allpairs-8n-ll.xml
+242
Просмотреть файл
@@ -0,0 +1,242 @@
<algo name="allgather_allpairs" proto="LL" nchannels="1" nchunksperloop="8" ngpus="8" coll="allgather" inplace="1" outofplace="0" minBytes="1" maxBytes="327680">
<gpu id="0" i_chunks="0" o_chunks="8" s_chunks="0">
<tb id="0" send="1" recv="1" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="1" send="2" recv="2" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="2" send="3" recv="3" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="3" send="4" recv="4" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="4" send="5" recv="5" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="5" send="6" recv="6" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="6" send="7" recv="7" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
</gpu>
<gpu id="1" i_chunks="0" o_chunks="8" s_chunks="0">
<tb id="0" send="0" recv="0" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="1" send="2" recv="2" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="2" send="3" recv="3" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="3" send="4" recv="4" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="4" send="5" recv="5" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="5" send="6" recv="6" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="6" send="7" recv="7" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
</gpu>
<gpu id="2" i_chunks="0" o_chunks="8" s_chunks="0">
<tb id="0" send="0" recv="0" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="1" send="1" recv="1" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="2" send="3" recv="3" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="3" send="4" recv="4" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="4" send="5" recv="5" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="5" send="6" recv="6" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="6" send="7" recv="7" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
</gpu>
<gpu id="3" i_chunks="0" o_chunks="8" s_chunks="0">
<tb id="0" send="0" recv="0" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="1" send="1" recv="1" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="2" send="2" recv="2" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="3" send="4" recv="4" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="4" send="5" recv="5" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="5" send="6" recv="6" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="6" send="7" recv="7" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
</gpu>
<gpu id="4" i_chunks="0" o_chunks="8" s_chunks="0">
<tb id="0" send="0" recv="0" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="1" send="1" recv="1" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="2" send="2" recv="2" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="3" send="3" recv="3" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="4" send="5" recv="5" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="5" send="6" recv="6" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="6" send="7" recv="7" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
</gpu>
<gpu id="5" i_chunks="0" o_chunks="8" s_chunks="0">
<tb id="0" send="0" recv="0" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="1" send="1" recv="1" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="2" send="2" recv="2" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="3" send="3" recv="3" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="4" send="4" recv="4" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="5" send="6" recv="6" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="6" send="7" recv="7" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
</gpu>
<gpu id="6" i_chunks="0" o_chunks="8" s_chunks="0">
<tb id="0" send="0" recv="0" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="1" send="1" recv="1" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="2" send="2" recv="2" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="3" send="3" recv="3" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="4" send="4" recv="4" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="5" send="5" recv="5" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="6" send="7" recv="7" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
</gpu>
<gpu id="7" i_chunks="0" o_chunks="8" s_chunks="0">
<tb id="0" send="0" recv="0" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="1" send="1" recv="1" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="2" send="2" recv="2" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="3" send="3" recv="3" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="4" send="4" recv="4" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="5" send="5" recv="5" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="5" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
<tb id="6" send="6" recv="6" chan="0">
<step s="0" type="s" srcbuf="o" srcoff="7" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
<step s="1" type="r" srcbuf="o" srcoff="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
</tb>
</gpu>
</algo>