apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

init (#6642)

Browse Source 
Co-authored-by: gongweibao <gognweibao@baidu.com>
This commit is contained in:
gongweibao
2026-03-04 21:55:31 +08:00
committed by GitHub
parent 5c8f5184d9
commit ddb06ff83f
306 changed files with 40627 additions and 34418 deletions
Download Patch File Download Diff File Expand all files Collapse all files
custom_ops/gpu_ops/get_output_msg_with_topk.cc
+60 -58
View File
@@ -27,10 +27,10 @@
#define K 20
struct msgdata {
long mtype;
int mtext[MAX_BSZ * (K + 1) + 2]; // stop_flag, bsz, tokens
float mtext_f[MAX_BSZ * (K + 1)]; // score
int mtext_ranks[MAX_BSZ]; // ranks
long mtype;
int mtext[MAX_BSZ * (K + 1) + 2]; // stop_flag, bsz, tokens
float mtext_f[MAX_BSZ * (K + 1)]; // score
int mtext_ranks[MAX_BSZ]; // ranks
};
void GetOutputTopK(const paddle::Tensor& x,
@@ -39,72 +39,74 @@ void GetOutputTopK(const paddle::Tensor& x,
int k,
int64_t rank_id,
bool wait_flag) {
static struct msgdata msg_rcv;
int msg_queue_id = 1;
static struct msgdata msg_rcv;
int msg_queue_id = 1;
if (const char* inference_msg_queue_id_env_p =
std::getenv("INFERENCE_MSG_QUEUE_ID")) {
std::string inference_msg_queue_id_env_str(
inference_msg_queue_id_env_p);
int inference_msg_queue_id_from_env =
std::stoi(inference_msg_queue_id_env_str);
if (const char* inference_msg_queue_id_env_p =
std::getenv("INFERENCE_MSG_QUEUE_ID")) {
std::string inference_msg_queue_id_env_str(inference_msg_queue_id_env_p);
int inference_msg_queue_id_from_env =
std::stoi(inference_msg_queue_id_env_str);
#ifdef GET_OUTPUT_DEBUG
std::cout << "Your INFERENCE_MSG_QUEUE_ID is: "
<< inference_msg_queue_id_from_env << std::endl;
std::cout << "Your INFERENCE_MSG_QUEUE_ID is: "
<< inference_msg_queue_id_from_env << std::endl;
#endif
msg_queue_id = inference_msg_queue_id_from_env;
}
static key_t key = ftok("/dev/shm", msg_queue_id);
msg_queue_id = inference_msg_queue_id_from_env;
}
static key_t key = ftok("/dev/shm", msg_queue_id);
static int msgid = msgget(key, IPC_CREAT | 0666);
static int msgid = msgget(key, IPC_CREAT | 0666);
#ifdef GET_OUTPUT_DEBUG
std::cout << "get_output_key: " << key << std::endl;
std::cout << "get_output msgid: " << msgid << std::endl;
std::cout << "get_output_key: " << key << std::endl;
std::cout << "get_output msgid: " << msgid << std::endl;
#endif
int64_t* out_data = const_cast<int64_t*>(x.data<int64_t>());
float* scores_data = const_cast<float*>(scores.data<float>());
int64_t* ranks_data = const_cast<int64_t*>(ranks.data<int64_t>());
int ret = -1;
if (!wait_flag) {
ret = msgrcv(msgid,
&msg_rcv,
(MAX_BSZ * (K + 1) + 2) * 4 + MAX_BSZ * (K + 1) * 4 + MAX_BSZ * 4,
0,
IPC_NOWAIT);
} else {
ret = msgrcv(msgid,
&msg_rcv,
(MAX_BSZ * (K + 1) + 2) * 4 + MAX_BSZ * (K + 1) * 4 + MAX_BSZ * 4,
0,
0);
}
if (ret == -1) {
// read none
out_data[0] = -2;
out_data[1] = 0;
return;
}
int bsz = msg_rcv.mtext[1];
out_data[0] = (int64_t)msg_rcv.mtext[0];
out_data[1] = (int64_t)msg_rcv.mtext[1];
for (int i = 0; i < bsz; i++) {
for (int j = 0; j < k + 1; j++) {
const int64_t offset = i * (K + 1) + j;
out_data[offset + 2] = (int64_t)msg_rcv.mtext[offset + 2];
scores_data[offset] = msg_rcv.mtext_f[offset];
}
ranks_data[i] = (int64_t)msg_rcv.mtext_ranks[i];
}
int64_t* out_data = const_cast<int64_t*>(x.data<int64_t>());
float* scores_data = const_cast<float*>(scores.data<float>());
int64_t* ranks_data = const_cast<int64_t*>(ranks.data<int64_t>());
int ret = -1;
if (!wait_flag) {
ret = msgrcv(
msgid,
&msg_rcv,
(MAX_BSZ * (K + 1) + 2) * 4 + MAX_BSZ * (K + 1) * 4 + MAX_BSZ * 4,
0,
IPC_NOWAIT);
} else {
ret = msgrcv(
msgid,
&msg_rcv,
(MAX_BSZ * (K + 1) + 2) * 4 + MAX_BSZ * (K + 1) * 4 + MAX_BSZ * 4,
0,
0);
}
if (ret == -1) {
// read none
out_data[0] = -2;
out_data[1] = 0;
return;
}
int bsz = msg_rcv.mtext[1];
out_data[0] = (int64_t)msg_rcv.mtext[0];
out_data[1] = (int64_t)msg_rcv.mtext[1];
for (int i = 0; i < bsz; i++) {
for (int j = 0; j < k + 1; j++) {
const int64_t offset = i * (K + 1) + j;
out_data[offset + 2] = (int64_t)msg_rcv.mtext[offset + 2];
scores_data[offset] = msg_rcv.mtext_f[offset];
}
ranks_data[i] = (int64_t)msg_rcv.mtext_ranks[i];
}
return;
}
PD_BUILD_STATIC_OP(get_output_topk)
.Inputs({"x", "scores", "ranks"})
.Attrs({"k: int", "rank_id: int64_t", "wait_flag: bool"})
.Outputs({"x_out", "scores_out", "ranks_out"})
.SetInplaceMap({{"x", "x_out"}, {"scores", "scores_out"}, {"ranks", "ranks_out"}})
.SetInplaceMap({{"x", "x_out"},
{"scores", "scores_out"},
{"ranks", "ranks_out"}})
.SetKernelFn(PD_KERNEL(GetOutputTopK));