From ed1678aad3a3ed12fd50dfd0a37bb9d52df5c766 Mon Sep 17 00:00:00 2001
From: seemingwang <seemingwang@users.noreply.github.com>
Date: Wed, 27 Apr 2022 15:32:37 +0800
Subject: [PATCH] fix test api problem (#42297)

* extract sub-graph

* graph-engine merging

* fix

* fix

* fix heter-ps config

* test performance

* test performance

* test performance

* test

* test

* update bfs

* change cmake

* test

* test gpu speed

* gpu_graph_engine optimization

* add dsm sample method

* add graph_neighbor_sample_v2

* Add graph_neighbor_sample_v2

* fix for loop

* add cpu sample interface

* fix kernel judgement

* add ssd layer to graph_engine

* fix allocation

* fix syntax error

* fix syntax error

* fix pscore class

* fix

* change index settings

* recover test

* recover test

* fix spelling

* recover

* fix

* move cudamemcpy after cuda stream sync

* fix linking problem

* remove comment

* add cpu test

* test

* add cpu test

* change comment

* combine feature table and graph table

* test

* test

* pybind

* test

* test

* test

* test

* pybind

* pybind

* fix cmake

* pybind

* fix

* fix

* add pybind

* add pybind

* optimize pybind

* test

* fix pybind

* fix

* pybind change

* remove file

Co-authored-by: DesmonDay <908660116@qq.com>
---
 .../fleet/heter_ps/.CMakeLists.txt.swp          | Bin 12288 -> 0 bytes
 .../fleet/heter_ps/graph_gpu_wrapper.cu         |  16 ++++++++--------
 2 files changed, 8 insertions(+), 8 deletions(-)
 delete mode 100644 paddle/fluid/framework/fleet/heter_ps/.CMakeLists.txt.swp

diff --git a/paddle/fluid/framework/fleet/heter_ps/.CMakeLists.txt.swp b/paddle/fluid/framework/fleet/heter_ps/.CMakeLists.txt.swp
deleted file mode 100644
index 7d3f69e7424d33094dfdd9a2da0d3110a4895c8d..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 12288
zcmeHNOOM+`9G?Q^QD{LF!G*&>ddQ;Y$*bamvX0q?=3&`M+dUb_nZzbe?Db=lP_0^t
z55O1TMh`t8@d>!WfeR;)ID$AJ&WJmId+c~(Cu}R|0Xmj`Yk9{1$Nc_}nP@$%UT<~q
zyXChKTrVKBTIj=1-}&&utFIvB+O{7KpLVT|s<o20@=AuOm!?;|(s8X?Ee-8uNvqf2
zE>-JSYGrr_RTkQOvNW~%%$!duvz6r96-gV7EK+gkZEOYRs4{l78Clrwv@*5`@hb-L
ziK{wZMd)Z#(l$NcEBnj)a_uzN`8;Rf5n|vxdb3`uh6^sd{u;jc;@}ZlJ6|eifHS}u
z;0$mEI0Kvk&H!hCGvGUC&~2dMv6za>@%>EZ9zXIMXMi)n8Q=_X1~>zp0nPwtfHS}u
z;0$mEI0OGd2DA}EKRu4n-6z0#`2T<M`~Q#U5&9nRIbaSL0=j^g0Y6_r=m&rYcnNTH
z9-#{0&*u>O1#lZs2h;%fo<-;);Ky?a-2i;|3__m)6u<%CeLw|p4sh>jggyfN{uDyr
z00`hL;O>(MeGK^aEJ9xcz5=`f_zisi5)k_SMd&}jI0Kx4{~ZHjx7qrjxQ8%YhUuu5
zIdnC5RZPAM!ip>?IC(1T0d9&?GN9YDSt{xi&o735RQh^2OU1`T^8=Tt$C?$8vawsm
zru9@ost}~J+UTT9JUei8Qr7j@-Svop4b3xrZD_5H>%kC@$g!ytRk!CpS^9ejc(~E(
zUsdIv(Ah&QF$KJIf($h~oJ&1br^KBTt2}kQf9mp#%v8s=t%SNux2}qPQI-@n48{o!
zfv#sf1iP8w>D!qfn-(d`Vz<!|@ugcrr*TbGg@FtseRaRpZODgevn8LbzZDC!;Qrpz
zscVq|p(h>Y#FXAZ7R0>zrl|D6a=+E<=44oxsQv=(_WEM+69c}1gYVy#!QOkgHO32q
zAxaLc4lhjKz&E9VdQG|<j*m>2=)Sdr7o62bdiyr>s7G~Cz(S~*FcSd|*Mz-bJT{ln
zOF8=gHbK_83E3JIR3@acsZd9pM`;en19a&diJ4@1B0|HTj6(LT>wrnlEqbDF0$JNk
z9x=nA!tATt(w$mR#D0-Y+nx1C!?am}WIi&-ahnQTRl(QX3GpH0=hHatP2$n+qeHD8
z1;kz9iK{t=I)P`$qfSJ!jb>97Z;Ha8-`H=9S)r(dt#v@IGm`UTHSKbqHghbZx|&aD
zp+%C+=$3AYqb`)EEf<+}u0i^<ti{EC8V8IQjIn>!`Zg9>vs>6mnFJz}UY`bO0pW6p
zOUeB8&Yj%MLnd@1y6X&JV!IxDNRFT|tl^cThY}l8LwSjZRCB)G7gEC5e9tMP1wq%v
zil`e~WLnxu+9Zu+qX9u#M!Tk@2t8S(5oRn>E%~iz$2npn1dIn!1IljvTTdc6+dKMj
zRv9(M_T|ioflt|nnL<k%aJJv(>B(?PgOU{s(=F@owp}ByKHHvS{n_CMGmve@xz3(x
zS<u2{Nnp<^VUCCG+i`6ucz;z)=J7$hx8G=EN$yFa+&{$01YO#;WP4z5;Ov)Z!^QL*
p(`=|huOqeEqO4vQWd)8NC%7#t3a-`4RVtBN>8>;6fi&1de*rIcBK80P

diff --git a/paddle/fluid/framework/fleet/heter_ps/graph_gpu_wrapper.cu b/paddle/fluid/framework/fleet/heter_ps/graph_gpu_wrapper.cu
index e99a0f4fe11..b0899b4a7f5 100644
--- a/paddle/fluid/framework/fleet/heter_ps/graph_gpu_wrapper.cu
+++ b/paddle/fluid/framework/fleet/heter_ps/graph_gpu_wrapper.cu
@@ -288,18 +288,18 @@ std::vector<int64_t> GraphGpuWrapper::graph_neighbor_sample(
   }
   /* VLOG(0) << "cumsum " << cumsum; */
 
-  std::vector<int64_t> res;
-  res.resize(cumsum * 2);
-  int count = 0;
+  std::vector<int64_t> cpu_key, res;
+  cpu_key.resize(key.size() * sample_size);
+
+  cudaMemcpy(cpu_key.data(), neighbor_sample_res.val,
+             key.size() * sample_size * sizeof(int64_t),
+             cudaMemcpyDeviceToHost);
   for (int i = 0; i < key.size(); i++) {
     for (int j = 0; j < actual_sample_size[i]; j++) {
-      res[count] = key[i];
-      count += 1;
+      res.push_back(key[i]);
+      res.push_back(cpu_key[i * sample_size + j]);
     }
   }
-
-  cudaMemcpy(res.data() + cumsum, neighbor_sample_res.val,
-             cumsum * sizeof(int64_t), cudaMemcpyDeviceToHost);
   /* for(int i = 0;i < res.size();i ++) { */
   /*     VLOG(0) << i << " " << res[i]; */
   /* } */
-- 
GitLab