From 128b0e70742110b8b52b0031f317799bec9f2368 Mon Sep 17 00:00:00 2001
From: Mustafa Abduljabbar <mustafa.abduljabbar@amd.com>
Date: Tue, 13 May 2025 17:07:03 -0500
Subject: [PATCH] Remove MSCCL single node AllGather XMLs (#1693)

* Remove MSCCL single node XMLs

* Remove comment on MSCCL AG single node support

[ROCm/rccl commit: d665547eef0790880564f5b62da21693f4acf5b5]
---
 projects/rccl/CHANGELOG.md                    |    2 +-
 .../allgather_8n_direct_0_8k_ll.xml           |  466 --------
 .../allgather_8n_direct_0_8k_ll_op.xml        |  514 ---------
 .../allgather_8n_direct_8k_1m_ll128.xml       |  914 ---------------
 .../allgather_8n_direct_8k_1m_ll128_op.xml    | 1010 -----------------
 5 files changed, 1 insertion(+), 2905 deletions(-)
 delete mode 100644 projects/rccl/tools/msccl-algorithms/allgather_8n_direct_0_8k_ll.xml
 delete mode 100644 projects/rccl/tools/msccl-algorithms/allgather_8n_direct_0_8k_ll_op.xml
 delete mode 100644 projects/rccl/tools/msccl-algorithms/allgather_8n_direct_8k_1m_ll128.xml
 delete mode 100644 projects/rccl/tools/msccl-algorithms/allgather_8n_direct_8k_1m_ll128_op.xml

diff --git a/projects/rccl/CHANGELOG.md b/projects/rccl/CHANGELOG.md
index c2f2d9c537..d64a89d399 100644
--- a/projects/rccl/CHANGELOG.md
+++ b/projects/rccl/CHANGELOG.md
@@ -19,7 +19,7 @@ Full documentation for RCCL is available at [https://rccl.readthedocs.io](https:
 
 ### Added
 
-* Added MSCCL support for AllGather single node and multinode (i.e., 8, 16 and 32 GPUs). To enable on multinode, set the
+* Added MSCCL support for multinode MI300/MI350 (i.e., 16 and 32 GPUs). To enable, set the
   environment variable `RCCL_MSCCL_FORCE_ENABLE=1`. Max message size for MSCCL AllGather usage is `12292 * sizeof(datatype) * nGPUs`.
 * Added synchronization before destroying proxy thread to fix a rare hang caused by early termination.
 
diff --git a/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_0_8k_ll.xml b/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_0_8k_ll.xml
deleted file mode 100644
index 5671a765c4..0000000000
--- a/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_0_8k_ll.xml
+++ /dev/null
@@ -1,466 +0,0 @@
-<algo name="allgather_allpairs" proto="LL" nchannels="2" nchunksperloop="16" ngpus="8" coll="allgather" inplace="1" outofplace="0" minBytes="0" maxBytes="8192">
-  <gpu id="0" i_chunks="0" o_chunks="16" 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="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="1" recv="1" chan="1">
-      <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="2" 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="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="2" recv="2" chan="1">
-      <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="4" 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="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="3" recv="3" chan="1">
-      <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>
-    <tb id="6" 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="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="4" recv="4" chan="1">
-      <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="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" 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="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="5" recv="5" chan="1">
-      <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="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" 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="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="6" recv="6" chan="1">
-      <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="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" 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="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="7" recv="7" chan="1">
-      <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="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="1" i_chunks="0" o_chunks="16" 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="0" recv="0" chan="1">
-      <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="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="3" send="2" recv="2" chan="1">
-      <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="4" 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="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="3" recv="3" chan="1">
-      <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>
-    <tb id="6" 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="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="4" recv="4" chan="1">
-      <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="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" 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="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="5" recv="5" chan="1">
-      <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="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" 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="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="6" recv="6" chan="1">
-      <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="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" 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="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="7" recv="7" chan="1">
-      <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="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="2" i_chunks="0" o_chunks="16" 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="0" recv="0" chan="1">
-      <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="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="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="1" recv="1" chan="1">
-      <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="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="6" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="3" recv="3" chan="1">
-      <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>
-    <tb id="6" send="4" recv="4" 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="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="4" recv="4" chan="1">
-      <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="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" 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="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="5" recv="5" chan="1">
-      <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="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" 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="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="6" recv="6" chan="1">
-      <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="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" 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="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="7" recv="7" chan="1">
-      <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="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="3" i_chunks="0" o_chunks="16" 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="0" recv="0" chan="1">
-      <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="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="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="1" recv="1" chan="1">
-      <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="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="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="2" recv="2" chan="1">
-      <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="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="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="4" recv="4" chan="1">
-      <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="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" 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="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="5" recv="5" chan="1">
-      <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="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="6" recv="6" 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="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="6" recv="6" chan="1">
-      <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="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" 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="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="7" recv="7" chan="1">
-      <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="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="4" i_chunks="0" o_chunks="16" s_chunks="0">
-    <tb id="0" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" 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="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" 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="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" 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="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" 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="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" 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="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" 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="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" 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="7" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" 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>
-    <tb id="8" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="5" i_chunks="0" o_chunks="16" s_chunks="0">
-    <tb id="0" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" 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="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" 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="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" 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="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" 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="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" 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="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" 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="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" 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="7" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" 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>
-    <tb id="8" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="6" i_chunks="0" o_chunks="16" s_chunks="0">
-    <tb id="0" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" 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="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" 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="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" 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="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" 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="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" 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="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" 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="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" 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="7" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" 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>
-    <tb id="8" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="7" i_chunks="0" o_chunks="16" s_chunks="0">
-    <tb id="0" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" 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="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" 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="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" 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="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" 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="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" 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="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" 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="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" 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="7" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" 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>
-    <tb id="8" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-</algo>
diff --git a/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_0_8k_ll_op.xml b/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_0_8k_ll_op.xml
deleted file mode 100644
index a9f7dc6553..0000000000
--- a/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_0_8k_ll_op.xml
+++ /dev/null
@@ -1,514 +0,0 @@
-<algo name="allgather_allpairs" proto="LL" nchannels="2" nchunksperloop="16" ngpus="8" coll="allgather" inplace="0" outofplace="1" minBytes="0" maxBytes="8192">
-  <gpu id="0" i_chunks="2" o_chunks="16" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="1" i_chunks="2" o_chunks="16" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="2" i_chunks="2" o_chunks="16" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="3" i_chunks="2" o_chunks="16" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="4" i_chunks="2" o_chunks="16" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="5" i_chunks="2" o_chunks="16" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="6" i_chunks="2" o_chunks="16" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="7" i_chunks="2" o_chunks="16" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-</algo>
diff --git a/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_8k_1m_ll128.xml b/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_8k_1m_ll128.xml
deleted file mode 100644
index a7c7a8f2e2..0000000000
--- a/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_8k_1m_ll128.xml
+++ /dev/null
@@ -1,914 +0,0 @@
-<algo name="allgather_allpairs" proto="LL128" nchannels="4" nchunksperloop="32" ngpus="8" coll="allgather" inplace="1" outofplace="0" minBytes="8192" maxBytes="1048575">
-  <gpu id="0" i_chunks="0" o_chunks="32" 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="4" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="1" recv="1" chan="1">
-      <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="2" send="1" recv="1" chan="2">
-      <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="3" send="1" recv="1" chan="3">
-      <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>
-    <tb id="4" 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="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="2" recv="2" chan="1">
-      <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="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="2" recv="2" chan="2">
-      <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="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="2" recv="2" chan="3">
-      <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="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" 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="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="3" recv="3" chan="1">
-      <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="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="3" recv="3" chan="2">
-      <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="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="3" recv="3" chan="3">
-      <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="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" 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="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="4" recv="4" chan="1">
-      <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="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="4" recv="4" chan="2">
-      <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="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="4" recv="4" chan="3">
-      <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="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" 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="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="5" recv="5" chan="1">
-      <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="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="5" recv="5" chan="2">
-      <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="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="5" recv="5" chan="3">
-      <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="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" 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="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="6" recv="6" chan="1">
-      <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="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="6" recv="6" chan="2">
-      <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="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="6" recv="6" chan="3">
-      <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="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" 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="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="7" recv="7" chan="1">
-      <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="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="7" recv="7" chan="2">
-      <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="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="7" recv="7" chan="3">
-      <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="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="1" i_chunks="0" o_chunks="32" 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="0" recv="0" chan="1">
-      <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="0" recv="0" chan="2">
-      <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="0" recv="0" chan="3">
-      <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="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="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="2" recv="2" chan="1">
-      <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="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="2" recv="2" chan="2">
-      <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="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="2" recv="2" chan="3">
-      <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="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" 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="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="3" recv="3" chan="1">
-      <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="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="3" recv="3" chan="2">
-      <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="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="3" recv="3" chan="3">
-      <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="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="4" recv="4" 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="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="4" recv="4" chan="1">
-      <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="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="4" recv="4" chan="2">
-      <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="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="4" recv="4" chan="3">
-      <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="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" 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="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="5" recv="5" chan="1">
-      <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="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="5" recv="5" chan="2">
-      <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="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="5" recv="5" chan="3">
-      <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="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" 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="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="6" recv="6" chan="1">
-      <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="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="6" recv="6" chan="2">
-      <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="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="6" recv="6" chan="3">
-      <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="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" 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="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="7" recv="7" chan="1">
-      <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="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="7" recv="7" chan="2">
-      <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="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="7" recv="7" chan="3">
-      <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="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="2" i_chunks="0" o_chunks="32" s_chunks="0">
-    <tb id="0" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" 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="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" 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="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" 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="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" 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="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" 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="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" 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="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" 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="7" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" 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>
-    <tb id="8" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="3" i_chunks="0" o_chunks="32" s_chunks="0">
-    <tb id="0" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" 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="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" 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="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" 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="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" 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="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" 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="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" 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="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" 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="7" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" 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>
-    <tb id="8" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="4" i_chunks="0" o_chunks="32" s_chunks="0">
-    <tb id="0" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" 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="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" 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="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" 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="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" 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="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" 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="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" 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="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" 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="7" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" 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>
-    <tb id="8" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="5" i_chunks="0" o_chunks="32" s_chunks="0">
-    <tb id="0" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" 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="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" 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="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" 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="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" 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="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" 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="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" 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="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" 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="7" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" 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>
-    <tb id="8" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="6" i_chunks="0" o_chunks="32" s_chunks="0">
-    <tb id="0" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" 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="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" 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="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" 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="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" 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="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" 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="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" 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="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" 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="7" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" 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>
-    <tb id="8" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="7" i_chunks="0" o_chunks="32" s_chunks="0">
-    <tb id="0" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" 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="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" 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="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" 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="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" 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="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" 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="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" 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="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" 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="7" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" 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>
-    <tb id="8" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="8" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="9" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="10" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="11" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="12" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="13" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="14" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="15" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="16" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="17" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="18" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="19" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="20" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="21" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="22" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="23" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="o" srcoff="28" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="24" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="o" srcoff="29" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="25" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="o" srcoff="30" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="26" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="o" srcoff="31" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="o" srcoff="27" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-</algo>
diff --git a/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_8k_1m_ll128_op.xml b/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_8k_1m_ll128_op.xml
deleted file mode 100644
index 17b6ea2bce..0000000000
--- a/projects/rccl/tools/msccl-algorithms/allgather_8n_direct_8k_1m_ll128_op.xml
+++ /dev/null
@@ -1,1010 +0,0 @@
-<algo name="allgather_allpairs" proto="LL128" nchannels="4" nchunksperloop="32" ngpus="8" coll="allgather" inplace="0" outofplace="1" minBytes="8192" maxBytes="1048575">
-  <gpu id="0" i_chunks="4" o_chunks="32" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="-1" recv="-1" chan="2">
-      <step s="0" type="cpy" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="-1" recv="-1" chan="3">
-      <step s="0" type="cpy" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="28" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="29" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="30" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="31" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="1" i_chunks="4" o_chunks="32" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="-1" recv="-1" chan="2">
-      <step s="0" type="cpy" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="-1" recv="-1" chan="3">
-      <step s="0" type="cpy" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="28" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="29" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="30" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="31" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="2" i_chunks="4" o_chunks="32" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="-1" recv="-1" chan="2">
-      <step s="0" type="cpy" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="-1" recv="-1" chan="3">
-      <step s="0" type="cpy" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="28" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="29" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="30" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="31" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="3" i_chunks="4" o_chunks="32" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="-1" recv="-1" chan="2">
-      <step s="0" type="cpy" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="-1" recv="-1" chan="3">
-      <step s="0" type="cpy" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="28" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="29" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="30" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="31" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="4" i_chunks="4" o_chunks="32" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="-1" recv="-1" chan="2">
-      <step s="0" type="cpy" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="-1" recv="-1" chan="3">
-      <step s="0" type="cpy" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="28" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="29" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="30" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="31" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="5" i_chunks="4" o_chunks="32" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="-1" recv="-1" chan="2">
-      <step s="0" type="cpy" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="-1" recv="-1" chan="3">
-      <step s="0" type="cpy" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="28" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="29" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="30" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="31" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="6" i_chunks="4" o_chunks="32" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="-1" recv="-1" chan="2">
-      <step s="0" type="cpy" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="-1" recv="-1" chan="3">
-      <step s="0" type="cpy" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="28" send="7" recv="7" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="29" send="7" recv="7" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="30" send="7" recv="7" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="31" send="7" recv="7" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-  <gpu id="7" i_chunks="4" o_chunks="32" s_chunks="0">
-    <tb id="0" send="-1" recv="-1" chan="0">
-      <step s="0" type="cpy" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="1" send="-1" recv="-1" chan="1">
-      <step s="0" type="cpy" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="2" send="-1" recv="-1" chan="2">
-      <step s="0" type="cpy" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="3" send="-1" recv="-1" chan="3">
-      <step s="0" type="cpy" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="4" send="0" recv="0" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="0" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="5" send="0" recv="0" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="1" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="6" send="0" recv="0" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="2" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="7" send="0" recv="0" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="3" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="8" send="1" recv="1" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="4" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="9" send="1" recv="1" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="5" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="10" send="1" recv="1" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="6" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="11" send="1" recv="1" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="7" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="12" send="2" recv="2" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="8" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="13" send="2" recv="2" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="9" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="14" send="2" recv="2" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="10" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="15" send="2" recv="2" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="11" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="16" send="3" recv="3" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="12" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="17" send="3" recv="3" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="13" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="18" send="3" recv="3" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="14" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="19" send="3" recv="3" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="15" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="20" send="4" recv="4" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="16" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="21" send="4" recv="4" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="17" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="22" send="4" recv="4" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="18" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="23" send="4" recv="4" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="19" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="24" send="5" recv="5" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="20" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="25" send="5" recv="5" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="21" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="26" send="5" recv="5" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="22" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="27" send="5" recv="5" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="23" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="28" send="6" recv="6" chan="0">
-      <step s="0" type="s" srcbuf="i" srcoff="0" dstbuf="o" dstoff="28" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="0" dstbuf="o" dstoff="24" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="29" send="6" recv="6" chan="1">
-      <step s="0" type="s" srcbuf="i" srcoff="1" dstbuf="o" dstoff="29" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="1" dstbuf="o" dstoff="25" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="30" send="6" recv="6" chan="2">
-      <step s="0" type="s" srcbuf="i" srcoff="2" dstbuf="o" dstoff="30" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="2" dstbuf="o" dstoff="26" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-    <tb id="31" send="6" recv="6" chan="3">
-      <step s="0" type="s" srcbuf="i" srcoff="3" dstbuf="o" dstoff="31" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-      <step s="1" type="r" srcbuf="i" srcoff="3" dstbuf="o" dstoff="27" cnt="1" depid="-1" deps="-1" hasdep="0"/>
-    </tb>
-  </gpu>
-</algo>