Skip to content

N
NiuTrans.Tensor
Commit 99097e41 by huchi
Options

add support for greedy search

parent bfa6fc90
正在显示 包含 245 行增加 和 245 行删除
source/Main.cpp
......@@ -19,6 +19,10 @@
* $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2018-07-10
*/
//#define CRTDBG_MAP_ALLOC
//#include <stdlib.h>
//#include <crtdbg.h>
#include <stdio.h>
#include "./network/XNet.h"
#include "./tensor/XUtility.h"
......@@ -27,9 +31,7 @@
#include "./sample/fnnlm/FNNLM.h"
#include "./sample/transformer/Transformer.h"
//#define CRTDBG_MAP_ALLOC
//#include <stdlib.h>
//#include <crtdbg.h>
using namespace nts;
using namespace fnnlm;
......@@ -37,19 +39,10 @@ using namespace transformer;
int main( int argc, const char ** argv )
{
//_CrtSetDbgFlag(_CrtSetDbgFlag(_CRTDBG_REPORT_FLAG) | _CRTDBG_LEAK_CHECK_DF);
//_CrtSetBreakAlloc(2708);
/*_CrtSetDbgFlag(_CrtSetDbgFlag(_CRTDBG_REPORT_FLAG) | _CRTDBG_LEAK_CHECK_DF);
_CrtSetBreakAlloc(2708);*/
TransformerMain(argc - 1, argv + 1);
/*XTensor x;
InitTensor2D(&x, 2, 2);
float d[]{ 1,2,3,4 };
x.SetData(d, 4);
XTensor y;
y = ReduceSum(x, 0);
y.Dump(stderr);*/
//_CrtDumpMemoryLeaks();
return 0;
......
source/sample/transformer/T2TAttention.cpp
......@@ -62,7 +62,7 @@ void T2TAttention::InitModel(int argc, char** argv,
float minmax = 0;
LoadParamInt(argc, argv, "nhead", &nhead, 8);
LoadParamInt(argc, argv, "nhead", &nhead, 4);
LoadParamInt(argc, argv, "d", &dk, DEFAULT_EMBEDDING_SIZE);
LoadParamInt(argc, argv, "d", &dv, DEFAULT_EMBEDDING_SIZE);
LoadParamInt(argc, argv, "d", &d, DEFAULT_EMBEDDING_SIZE);
......@@ -70,15 +70,15 @@ void T2TAttention::InitModel(int argc, char** argv,
LoadParamFloat(argc, argv, "attminmax", &minmax, 0.1F);
LoadParamFloat(argc, argv, "dropoutatt", &dropoutP, 0);
InitTensor2D(&wq, d, d, X_FLOAT, devID);
InitTensor1D(&bq, d, X_FLOAT, devID);
InitTensor2D(&wk, d, d, X_FLOAT, devID);
InitTensor1D(&bk, d, X_FLOAT, devID);
InitTensor2D(&wv, d, d, X_FLOAT, devID);
InitTensor1D(&bv, d, X_FLOAT, devID);
InitTensor2D(&rp_embedding_k, max_relative_position * 2 + 1, d/nhead, X_FLOAT, devID);
InitTensor2D(&wa, d, d, X_FLOAT, devID);
InitTensor1D(&ba, d, X_FLOAT, devID);
InitTensor2DV2(&wq, d, d, X_FLOAT, devID);
InitTensor1DV2(&bq, d, X_FLOAT, devID);
InitTensor2DV2(&wk, d, d, X_FLOAT, devID);
InitTensor1DV2(&bk, d, X_FLOAT, devID);
InitTensor2DV2(&wv, d, d, X_FLOAT, devID);
InitTensor1DV2(&bv, d, X_FLOAT, devID);
InitTensor2DV2(&rp_embedding_k, max_relative_position * 2 + 1, d/nhead, X_FLOAT, devID);
InitTensor2DV2(&wo, d, d, X_FLOAT, devID);
InitTensor1DV2(&bo, d, X_FLOAT, devID);
}
/*
......@@ -94,24 +94,27 @@ make the network
>> cacheType - which type that cache is
<< return - multi-attention result
*/
XTensor T2TAttention::Make( XTensor& k, XTensor& q, XTensor& v, XTensor* mask, bool isTraining, Cache* cache, int cacheType)
XTensor T2TAttention::Make(XTensor& k, XTensor& q, XTensor& v, XTensor* mask, bool isTraining, Cache* cache, int cacheType)
{
const bool isEnc = (!cache) ? true : false;
/* linear transformation before self-attention */
XTensor q2, k2, v2;
q2 = MatrixMul(q, X_NOTRANS, wq, X_TRANS) + bq;
q2 = MatrixMul(q, wq) + bq;
if (!cache) {
/* self attention for encoder layers */
k2 = MatrixMul(k, X_NOTRANS, wk, X_TRANS) + bk;
v2 = MatrixMul(v, X_NOTRANS, wv, X_TRANS) + bv;
k2 = MatrixMul(k, wk) + bk;
v2 = MatrixMul(v, wv) + bv;
return MakeRPRAttention(k2, q2, v2, mask, isTraining, isEnc);
}
else {
if (cacheType == SELF_ATT) {
k2 = MatrixMul(k, X_NOTRANS, wk, X_TRANS) + bk;
v2 = MatrixMul(v, X_NOTRANS, wv, X_TRANS) + bv;
k2 = MatrixMul(k, wk) + bk;
v2 = MatrixMul(v, wv) + bv;
/* if hit, we only concat the cache with the new token */
if (!cache->miss) {
......@@ -121,12 +124,13 @@ XTensor T2TAttention::Make( XTensor& k, XTensor& q, XTensor& v, XTensor* mask,
cache->key = k2;
cache->value = v2;
cache->miss = false;
return MakeRPRAttention(cache->key, q2, cache->value, mask, isTraining, isEnc);
}
else if (cacheType == EN_DE_ATT) {
if (cache->miss) {
cache->key = MatrixMul(k, X_NOTRANS, wk, X_TRANS) + bk;
cache->value = MatrixMul(v, X_NOTRANS, wv, X_TRANS) + bv;
cache->key = MatrixMul(k, wk) + bk;
cache->value = MatrixMul(v, wv) + bv;
cache->miss = false;
}
return MakeAttention(cache->key, q2, cache->value, mask, isTraining, isEnc);
......@@ -145,7 +149,7 @@ make the attention network given keys, queries and values (after linear transfor
>> mask - as it is
>> isTraining - indicates whether the model is used for training
*/
XTensor T2TAttention::MakeAttention(XTensor &k, XTensor& q, XTensor& v, XTensor* mask, bool isTraining, bool is_encoder)
XTensor T2TAttention::MakeAttention(XTensor& k, XTensor& q, XTensor& v, XTensor* mask, bool isTraining, bool is_encoder)
{
XTensor kheads;
XTensor qheads;
......@@ -163,21 +167,20 @@ XTensor T2TAttention::MakeAttention(XTensor &k, XTensor& q, XTensor& v, XTensor*
/* scalar = softmax(Q * K^T / sqrt(dk)) * V */
dot = BMMul(qheads, X_NOTRANS, kheads, X_TRANS);
/*if (isMasked && mask) {
_SumMe(&dot, mask);
}*/
/*if (isMasked && mask)
_SumMe(&dot, mask);*/
dot = Linear(dot, 1.0F / (float)sqrt((float)dk / nhead));
scalar = Softmax(dot, -1);
/*if(isTraining && dropoutP > 0)
scalar = Dropout(scalar, dropoutP);*/
if(isTraining && dropoutP > 0)
scalar = Dropout(scalar, dropoutP);
att = BMMul(scalar, vheads);
/* concatenate the heads */
return MulAndShift(Merge(att, att.order - 1), X_NOTRANS, wa, X_TRANS, ba);
return MulAndShift(Merge(att, att.order - 1), wo, bo);
}
/*
......@@ -215,14 +218,12 @@ XTensor T2TAttention::MakeRPRAttention(XTensor& k, XTensor& q, XTensor& v, XTens
InitTensor4DV2(&dot, nhead, batch_size, len_q, len_kv, X_FLOAT, q.devID);
/* generate the relative emb index (L_q, L_kv) */
GetRPEmbedding(&emb_matrix, len_q, len_kv, max_relative_position, q.devID,is_encoder);
GetRPEmbedding(&emb_matrix, len_q, len_kv, max_relative_position, q.devID, is_encoder);
/* generate the relative key from the rp_embedding_k (L_q, L_kv, H/K) */
_Gather(&rp_embedding_k, &relative_key, &emb_matrix);
/* RPR dot product (K, B, L_q, L_kv)*/
qheads = qheads / float(nhead);
RPDotProduct(&qheads, &kheads, &relative_key, &dot, true);
......@@ -230,19 +231,19 @@ XTensor T2TAttention::MakeRPRAttention(XTensor& k, XTensor& q, XTensor& v, XTens
_SumMe(&dot, mask);*/
/* scale the dot result */
//dot = Linear(dot, 1.0F / (float)sqrt((float)dk / nhead));
dot = Linear(dot, 1.0F / (float)sqrt((float)dk / nhead));
/* softmax */
scalar = Softmax(dot, -1);
/*if (isTraining && dropoutP > 0)
scalar = Dropout(scalar, dropoutP);*/
if (isTraining && dropoutP > 0)
scalar = Dropout(scalar, dropoutP);
/* generate the relative attention output (K, B, L_q, H/K) */
att = BMMul(scalar, vheads);
/* concatenate the heads */
return MulAndShift(Merge(att, att.order - 1), X_NOTRANS, wa, X_TRANS, ba);
return MulAndShift(Merge(att, att.order - 1), wo, bo);
}
void T2TAttention::GetRPEmbedding(XTensor* emb_matrix, const int len_q, const int len_kv, const int max_relative_length, const int devID, const bool is_encoder)
......@@ -251,10 +252,11 @@ void T2TAttention::GetRPEmbedding(XTensor* emb_matrix, const int len_q, const in
XTensor range;
InitTensor1DV2(&range, len_kv, X_INT, devID);
int* index = new int[len_kv];
// for encoder self-attention which the L_q = L_kv
if (is_encoder)
{
for (int i = 0; i <len_kv; i++)
for (int i = 0; i < len_kv; i++)
index[i] = i;
range.SetData(index, len_kv);
XTensor range_2D, range_2D_t;
......@@ -267,7 +269,7 @@ void T2TAttention::GetRPEmbedding(XTensor* emb_matrix, const int len_q, const in
// for decoder self-attention which the L_q != L_kv, and L_q is 1
else
{
for (int i = 0; i <len_kv; i++)
for (int i = 0; i < len_kv; i++)
index[i] = -len_kv + i + 1;
range.SetData(index, len_kv);
_Unsqueeze(&range, emb_matrix, 0, len_q);
......@@ -299,7 +301,6 @@ void T2TAttention::RPDotProduct(XTensor* x, XTensor* y, XTensor* z, XTensor* att
XTensor context;
InitTensor4DV2(&context, head_num, batch_size, len_q, last_dim, X_FLOAT, x->devID);
_MatrixMulBatched(x, X_NOTRANS, y, transpose_flag, &context);
//if (profiler_) profiler_->FinishTimer("RPDotPro-BMM");
// reshape and transpose x to (L_q, K*B, H/K or L_kv)
int merge_dims[] = { head_num * batch_size, len_q, x->dimSize[3] };
......@@ -323,5 +324,6 @@ void T2TAttention::RPDotProduct(XTensor* x, XTensor* y, XTensor* z, XTensor* att
relative_t.Reshape(4, split_dims);
_Sum(&context, &relative_t, attention);
}
}
source/sample/transformer/T2TAttention.h
......@@ -90,14 +90,18 @@ public:
/* bias for V */
XTensor bv;
XTensor wBig;
XTensor bBig;
/* RPR emb */
XTensor rp_embedding_k;
/* transformation after dot-product attention */
XTensor wa;
XTensor wo;
/* bias after dot-product attention */
XTensor ba;
XTensor bo;
/* size of transformed Q and K */
int dk;
......
source/sample/transformer/T2TDecoder.cpp
......@@ -31,27 +31,27 @@ namespace transformer
/* constructor */
AttDecoder::AttDecoder()
{
attentions = NULL;
selfAtt = NULL;
fnns = NULL;
attLayerNorms = NULL;
attentionsEnde = NULL;
attEndeLayerNorms = NULL;
decodeLayerNorm = NULL;
selfCache = NULL;
contextCache = NULL;
selfAttLayerNorms = NULL;
enDeAtt = NULL;
enDeAttLayerNorms = NULL;
decoderLayerNorm = NULL;
selfAttCache = NULL;
enDeAttCache = NULL;
}
/* de-constructor */
AttDecoder::~AttDecoder()
{
delete[] selfCache;
delete[] contextCache;
delete[] attentions;
delete[] selfAttCache;
delete[] enDeAttCache;
delete[] selfAtt;
delete[] fnns;
delete[] attLayerNorms;
delete[] attentionsEnde;
delete[] attEndeLayerNorms;
delete decodeLayerNorm;
delete[] selfAttLayerNorms;
delete[] enDeAtt;
delete[] enDeAttLayerNorms;
delete decoderLayerNorm;
}
/*
......@@ -71,7 +71,7 @@ void AttDecoder::InitModel(int argc, char ** argv,
devID = myDevID;
ignored = myIgnored;
LoadParamInt(argc, argv, "nlayer", &nlayer, 3);
LoadParamInt(argc, argv, "nlayer", &nlayer, 4);
LoadParamInt(argc, argv, "hsize", &hSize, DEFAULT_EMBEDDING_SIZE);
LoadParamInt(argc, argv, "esize", &eSize, DEFAULT_EMBEDDING_SIZE);
LoadParamInt(argc, argv, "vsizetgt", &vSize, 34040);
......@@ -83,24 +83,24 @@ void AttDecoder::InitModel(int argc, char ** argv,
/* embedding model */
embedder.InitModel(argc, argv, devID, false);
attentions = new T2TAttention[nlayer];
selfAtt = new T2TAttention[nlayer];
fnns = new T2TFNN[nlayer];
attLayerNorms = new T2TLN[nlayer];
attentionsEnde = new T2TAttention[nlayer];
attEndeLayerNorms = new T2TLN[nlayer];
decodeLayerNorm = new T2TLN;
selfCache = new Cache[nlayer];
contextCache = new Cache[nlayer];
selfAttLayerNorms = new T2TLN[nlayer];
enDeAtt = new T2TAttention[nlayer];
enDeAttLayerNorms = new T2TLN[nlayer];
decoderLayerNorm = new T2TLN;
selfAttCache = new Cache[nlayer];
enDeAttCache = new Cache[nlayer];
/* initialize the stacked layers */
for (int i = 0; i < nlayer; i++) {
attentions[i].InitModel(argc, argv, myIsMasked, myIgnored, myDevID);
selfAtt[i].InitModel(argc, argv, myIsMasked, myIgnored, myDevID);
fnns[i].InitModel(argc, argv, myDevID);
attLayerNorms[i].InitModel(argc, argv, myDevID);
attentionsEnde[i].InitModel(argc, argv, true, myIgnored, myDevID);
attEndeLayerNorms[i].InitModel(argc, argv, myDevID);
selfAttLayerNorms[i].InitModel(argc, argv, myDevID);
enDeAtt[i].InitModel(argc, argv, true, myIgnored, myDevID);
enDeAttLayerNorms[i].InitModel(argc, argv, myDevID);
}
decodeLayerNorm->InitModel(argc, argv, myDevID);
decoderLayerNorm->InitModel(argc, argv, myDevID);
}
/*
......@@ -131,48 +131,38 @@ XTensor AttDecoder::Make(XTensor &inputDec, XTensor &outputEnc, XTensor *mask, X
XTensor attNorm;
/* layer normalization */
inputNorm = attLayerNorms[i].Make(x);
//inputNorm.Dump(stderr, "inputNorm", 10);
inputNorm = selfAttLayerNorms[i].Make(x);
/******************/
/* self attention */
att = attentions[i].Make(inputNorm, inputNorm, inputNorm, NULL, isTraining, &selfCache[i], SELF_ATT);
att = selfAtt[i].Make(inputNorm, inputNorm, inputNorm, NULL, isTraining, &selfAttCache[i], SELF_ATT);
/* dropout */
if(isTraining && dropoutP > 0)
att = Dropout(att, dropoutP);
/* residual connection */
_SumMe(&att, &x);
//att.Dump(stderr, "Sum(att, x)", 10);
att = att + x;
/* layer normalization */
attNorm = attEndeLayerNorms[i].Make(att);
//attNorm.Dump(stderr, "attNorm", 10);
attNorm = enDeAttLayerNorms[i].Make(att);
/* encoder-decoder attention */
ende = attentionsEnde[i].Make(outputEnc, attNorm, outputEnc, &maskEncDec, isTraining, &contextCache[i], EN_DE_ATT);
//ende.Dump(stderr, "ende atten", 10);
ende = enDeAtt[i].Make(outputEnc, attNorm, outputEnc, &maskEncDec, isTraining, &enDeAttCache[i], EN_DE_ATT);
/* dropout */
if(isTraining && dropoutP > 0)
ende = Dropout(ende, dropoutP);
/* residual connection */
_SumMe(&ende, &att);
//res.Dump(stderr, "Sum(ende, att)", 10);
ende = ende + att;
/* fnn */
x = fnns[i].Make(ende, isTraining);
//x.Dump(stderr, "fnns[i]", 10);
}
x = decodeLayerNorm->Make(x);
//x.Dump(stderr, "decodeLayerNorm", 10);
x.SetName(DECODING_NAME);
x = decoderLayerNorm->Make(x);
return x;
}
......
source/sample/transformer/T2TDecoder.h
......@@ -63,13 +63,13 @@ public:
T2TFNN * fnns;
/* attention model of each layer */
T2TAttention * attentions;
T2TAttention * selfAtt;
/* layer normalization for attention */
T2TLN * attLayerNorms;
T2TLN * selfAttLayerNorms;
/* layer normalization for decoder */
T2TLN * decodeLayerNorm;
T2TLN * decoderLayerNorm;
/* input tensor of the encoder */
XTensor * input;
......@@ -78,16 +78,16 @@ public:
XTensor * output;
/* encoder-decoder attention model of each layer */
T2TAttention * attentionsEnde;
T2TAttention * enDeAtt;
/* layer normalization for encoder-decoder attention */
T2TLN * attEndeLayerNorms;
T2TLN * enDeAttLayerNorms;
/* layer cache list */
Cache* selfCache;
Cache* selfAttCache;
/* layer cache list */
Cache* contextCache;
Cache* enDeAttCache;
public:
/* constructor */
......
source/sample/transformer/T2TEmbedding.cpp
......@@ -62,7 +62,7 @@ void T2TEmbedder::InitModel(int argc, char ** argv, int myDevID, bool isEnc)
LoadParamInt(argc, argv, "d", &d, DEFAULT_EMBEDDING_SIZE);
LoadParamInt(argc, argv, "pad", &padIdx, 1);
InitTensor2D(&w, vSize, eSize, X_FLOAT, devID);
InitTensor2DV2(&w, vSize, eSize, X_FLOAT, devID);
maxLength = maxLength + 1 + 1;
DTYPE v = 1.0F/(float)sqrt((float)eSize);
......@@ -80,7 +80,7 @@ make positional embeddings (of size eSize * length)
*/
void T2TEmbedder::MakePosEmbedding(int eSize, int d, int length, int padIdx)
{
InitTensor2D(&posEmbeddingBase, length, eSize, X_FLOAT, devID);
InitTensor2DV2(&posEmbeddingBase, length, eSize, X_FLOAT, devID);
float * data = new float[posEmbeddingBase.unitNum];
......@@ -113,47 +113,47 @@ make the network
*/
XTensor T2TEmbedder::Make(XTensor &input, int prevLen)
{
/* assert padding index is 1 */
///* assert padding index is 1 */
CheckNTErrors(input.order > 1, "Wrong input tensor size!");
CheckNTErrors(input.dimSize[input.order - 1] < maxLength, "The sequence is too long!");
CheckNTErrors(vSize > 0, "set vocabulary size by \"-vsize\"");
CheckNTErrors(eSize > 0, "set embedding size by \"-esize\"");
//CheckNTErrors(input.order > 1, "Wrong input tensor size!");
//CheckNTErrors(input.dimSize[input.order - 1] < maxLength, "The sequence is too long!");
//CheckNTErrors(vSize > 0, "set vocabulary size by \"-vsize\"");
//CheckNTErrors(eSize > 0, "set embedding size by \"-esize\"");
//
//XTensor wordEmbedding, position, posEmbedding;
//InitTensor(&position, &input);
XTensor wordEmbedding, position, posEmbedding;
InitTensor(&position, &input);
//int* posData = new int[input.unitNum];
int* posData = new int[input.unitNum];
//XTensor inputCPU;
//InitTensorOnCPU(&inputCPU, &input);
//_CopyValues(&input, &inputCPU);
XTensor inputCPU;
InitTensorOnCPU(&inputCPU, &input);
_CopyValues(&input, &inputCPU);
//for (int i = 0; i < inputCPU.GetDim(0); i++) {
// int startNoPad = 2 + prevLen - 1;
// int* p = ((int*)inputCPU.data) + i * inputCPU.GetDim(1);
// for (int j = 0; j < inputCPU.GetDim(1); j++) {
// if (p[j] == 1) {
// posData[i * inputCPU.GetDim(1) + j] = 1;
// }
// else {
// posData[i * inputCPU.GetDim(1) + j] = startNoPad++;
// }
// }
//}
for (int i = 0; i < inputCPU.GetDim(0); i++) {
int startNoPad = 2 + prevLen - 1;
int* p = ((int*)inputCPU.data) + i * inputCPU.GetDim(1);
for (int j = 0; j < inputCPU.GetDim(1); j++) {
if (p[j] == 1) {
posData[i * inputCPU.GetDim(1) + j] = 1;
}
else {
posData[i * inputCPU.GetDim(1) + j] = startNoPad++;
}
}
}
//position.SetData(posData, position.unitNum);
//delete[] posData;
position.SetData(posData, position.unitNum);
delete[] posData;
/* we make positional embeddings first */
if(true){
posEmbedding = Gather(posEmbeddingBase, position);
}
///* we make positional embeddings first */
//if(true){
// posEmbedding = Gather(posEmbeddingBase, position);
//}
/* then we make word embeddings */
XTensor wordEmbedding;
wordEmbedding = Gather(w, input);
wordEmbedding = Linear(wordEmbedding, (float)sqrt((float)eSize));
......
source/sample/transformer/T2TEmbedding.h
......@@ -29,7 +29,7 @@ using namespace nts;
namespace transformer
{
#define DEFAULT_EMBEDDING_SIZE 512
#define DEFAULT_EMBEDDING_SIZE 128
/*
embedding (of word at position i):
......
source/sample/transformer/T2TEncoder.cpp
......@@ -34,7 +34,7 @@ AttEncoder::AttEncoder()
attentions = NULL;
fnns = NULL;
attLayerNorms = NULL;
encodeLayerNorm = NULL;
encoderLayerNorm = NULL;
}
/* de-constructor */
......@@ -43,7 +43,7 @@ AttEncoder::~AttEncoder()
delete[] attentions;
delete[] fnns;
delete[] attLayerNorms;
delete encodeLayerNorm;
delete encoderLayerNorm;
}
/*
......@@ -61,7 +61,7 @@ void AttEncoder::InitModel(int argc, char ** argv,
devID = myDevID;
ignored = myIgnored;
LoadParamInt(argc, argv, "nlayer", &nlayer, 35);
LoadParamInt(argc, argv, "nlayer", &nlayer, 20);
LoadParamInt(argc, argv, "hsize", &hSize, DEFAULT_EMBEDDING_SIZE);
LoadParamInt(argc, argv, "esize", &eSize, DEFAULT_EMBEDDING_SIZE);
LoadParamInt(argc, argv, "vsize", &vSize, 34040);
......@@ -76,7 +76,7 @@ void AttEncoder::InitModel(int argc, char ** argv,
attentions = new T2TAttention[nlayer];
fnns = new T2TFNN[nlayer];
attLayerNorms = new T2TLN[nlayer];
encodeLayerNorm = new T2TLN;
encoderLayerNorm = new T2TLN;
/* initialize the stacked layers */
for(int i = 0; i < nlayer; i++){
......@@ -84,7 +84,7 @@ void AttEncoder::InitModel(int argc, char ** argv,
fnns[i].InitModel(argc, argv, myDevID);
attLayerNorms[i].InitModel(argc, argv, myDevID);
}
encodeLayerNorm->InitModel(argc, argv, myDevID);
encoderLayerNorm->InitModel(argc, argv, myDevID);
}
/*
......@@ -123,13 +123,9 @@ XTensor AttEncoder::Make(XTensor &input, XTensor *mask, XTensor &maskEncDec, boo
/* fnn */
x = fnns[i].Make(res, isTraining);
}
x = encodeLayerNorm->Make(x);
x.SetName(ENCODING_NAME);
input.SetName(ENCODING_INPUT_NAME);
x = encoderLayerNorm->Make(x);
return x;
}
......
source/sample/transformer/T2TEncoder.h
......@@ -93,11 +93,11 @@ public:
/* attention model of each layer */
T2TAttention * attentions;
/* layer normalization for attention */
/* layer normalizations for attention */
T2TLN * attLayerNorms;
/* layer normalization for encoder */
T2TLN * encodeLayerNorm;
T2TLN * encoderLayerNorm;
/* input tensor of the encoder */
XTensor * input;
......
source/sample/transformer/T2TFNN.cpp
......@@ -15,7 +15,7 @@
* limitations under the License.
*/
/*
/*
* $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2018-07-31
*/
......@@ -48,7 +48,7 @@ initialize the model
>> argv - list of pointers to the arguments
>> myDevID - device id
*/
void T2TFNN::InitModel(int argc, char ** argv, int myDevID)
void T2TFNN::InitModel(int argc, char** argv, int myDevID)
{
devID = myDevID;
......@@ -56,14 +56,14 @@ void T2TFNN::InitModel(int argc, char ** argv, int myDevID)
LoadParamInt(argc, argv, "d", &inSize, DEFAULT_EMBEDDING_SIZE);
LoadParamInt(argc, argv, "d", &outSize, DEFAULT_EMBEDDING_SIZE);
LoadParamInt(argc, argv, "fnnh", &hSize, outSize * 4);
LoadParamInt(argc, argv, "fnnh", &hSize, outSize * 8);
LoadParamFloat(argc, argv, "fnnminmax", &minmax, 0.1F);
LoadParamFloat(argc, argv, "dropoutfnn", &dropoutP, 0);
InitTensor2DV2(&w1, hSize, inSize, X_FLOAT, devID);
InitTensor2DV2(&w1, inSize, hSize, X_FLOAT, devID);
InitTensor1DV2(&b1, hSize, X_FLOAT, devID);
InitTensor2DV2(&w2, outSize, hSize, X_FLOAT, devID);
InitTensor2DV2(&w2, hSize, outSize, X_FLOAT, devID);
InitTensor1DV2(&b2, outSize, X_FLOAT, devID);
fnnLayerNorm.InitModel(argc, argv, myDevID);
......@@ -84,19 +84,19 @@ y = max(0, x * w1 + b1) * w2 + b2
>> input - the input tensor
>> return - the output tensor
*/
XTensor T2TFNN::Make(XTensor &input, bool isTraining)
XTensor T2TFNN::Make(XTensor& input, bool isTraining)
{
XTensor t1;
/* t1 = max(0, x * w1 + b1) */
t1 = Rectify(MulAndShift(fnnLayerNorm.Make(input), X_NOTRANS, w1, X_TRANS, b1));
t1 = Rectify(MulAndShift(fnnLayerNorm.Make(input), w1, b1));
if(isTraining && dropoutP > 0)
if (isTraining && dropoutP > 0)
t1 = Dropout(t1, dropoutP);
/* result = t1 * w2 + b2 */
XTensor res;
res = MulAndShift(t1, X_NOTRANS, w2, X_TRANS, b2);
res = MulAndShift(t1, w2, b2);
_SumMe(&res, &input);
return res;
}
......
source/sample/transformer/T2TLayerNormal.cpp
......@@ -53,8 +53,8 @@ void T2TLN::InitModel(int argc, char ** argv, int myDevID)
d = 0;
LoadParamInt(argc, argv, "d", &d, DEFAULT_EMBEDDING_SIZE);
InitTensor1D(&w, d, X_FLOAT, devID);
InitTensor1D(&b, d, X_FLOAT, devID);
InitTensor1DV2(&w, d, X_FLOAT, devID);
InitTensor1DV2(&b, d, X_FLOAT, devID);
}
/*
......@@ -78,7 +78,7 @@ XTensor T2TLN::Make(XTensor &input)
mean = ReduceMean(x, x.order - 1);
/* \sigma = (sum_i (x_i - \mu)^2)/m */
variance = ReduceVariance(x, x.order - 1, mean);
variance = ReduceVariance(x, x.order - 1, mean) + 1e-5F;
/* standard = sqrt(variance) */
standard = Power(variance, 0.5F);
......
source/sample/transformer/T2TModel.h
......@@ -103,7 +103,7 @@ public:
/* read the parameters */
void Read(const char * fn);
};
void FastRead(XTensor* x, FILE* f);
}
#endif
source/sample/transformer/T2TOutput.cpp
......@@ -56,13 +56,11 @@ void T2TOutput::InitModel(int argc, char ** argv, int myDevID)
LoadParamInt(argc, argv, "vsizetgt", &vSize, -1);
LoadParamInt(argc, argv, "d", &inSize, DEFAULT_EMBEDDING_SIZE);
LoadParamInt(argc, argv, "d", &hSize, DEFAULT_EMBEDDING_SIZE);
LoadParamFloat(argc, argv, "outputminmax", &minmax, 0.08F);
InitTensor2D(&w, hSize, vSize, X_FLOAT, devID);
InitTensor2DV2(&w, vSize, hSize, X_FLOAT, devID);
}
/*
make the network (redefined output tensor)
>> input - input tensor
......@@ -72,9 +70,7 @@ void T2TOutput::Make(XTensor &input, XTensor &output)
{
XTensor &x = input;
output = LogSoftmax(MMul(x, X_NOTRANS, w, X_NOTRANS), -1);
output.SetName(OUTPUT_NAME);
output = LogSoftmax(MMul(x, X_NOTRANS, w, X_TRANS), -1);
}
}
source/sample/transformer/T2TPredictor.cpp
......@@ -15,7 +15,7 @@
* limitations under the License.
*/
/*
/*
* $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2019-03-13
*/
......@@ -38,7 +38,7 @@ T2TStateBundle::T2TStateBundle()
/* de-constructor */
T2TStateBundle::~T2TStateBundle()
{
if(states != NULL)
if (states != NULL)
delete[] states;
}
......@@ -50,12 +50,12 @@ void T2TStateBundle::MakeStates(int num)
{
CheckNTErrors(num > 0, "invalid number");
if(states != NULL)
if (states != NULL)
delete[] states;
states = new T2TState[num];
for(int i = 0; i < num; i++){
for (int i = 0; i < num; i++) {
states[i].prediction = -1;
states[i].pid = T2T_PID_EMPTY;
states[i].isEnd = false;
......@@ -74,7 +74,7 @@ void T2TStateBundle::MakeStates(int num)
/* constructor */
T2TPredictor::T2TPredictor()
{
startSymbol = -1;
startSymbol = 2;
}
/* de-constructor */
......@@ -90,29 +90,36 @@ create an initial state
>> beamSize - beam size
>> state - the state to be initialized
*/
void T2TPredictor::Create(T2TModel * model, XTensor * top, const XTensor * input, int beamSize, T2TStateBundle * state)
void T2TPredictor::Create(T2TModel* model, XTensor* top, const XTensor* input, int beamSize, T2TStateBundle* state)
{
int dims[MAX_TENSOR_DIM_NUM];
for (int i = 0; i < input->order - 1; i++)
dims[i] = input->GetDim(i);
dims[input->order - 1] = beamSize;
InitTensor(&state->probPath, input->order, dims, X_FLOAT, input->devID);
InitTensor(&state->nstep, input->order, dims, X_FLOAT, input->devID);
InitTensor(&state->endMark, input->order, dims, X_INT, input->devID);
InitTensorV2(&state->probPath, input->order, dims, X_FLOAT, 1.0F, input->devID);
InitTensorV2(&state->nstep, input->order, dims, X_FLOAT, 1.0F, input->devID);
InitTensorV2(&state->endMark, input->order, dims, X_INT, 1.0F, input->devID);
float* data = new float[state->probPath.unitNum];
/*float* data = new float[state->probPath.unitNum];
for (int i = 0; i < state->probPath.unitNum; ++i) {
data[i] = -1e20F;
if (i % beamSize == 0)
data[i] = 0;
}
state->probPath.SetData(data, state->probPath.unitNum);
delete[] data;*/
SetDataFixed(state->probPath, -1e9F);
for (int i = 0; i < state->probPath.unitNum; ++i) {
if (i % beamSize == 0)
state->probPath.Set(0.0F, i);
}
state->nstep.SetZeroAll();
state->endMark.SetZeroAll();
delete[] data;
state->stateNum = 0;
}
......@@ -133,7 +140,7 @@ read a state
2) probablities of hypotheses
3) parts of the network for expanding toward the next state
*/
void T2TPredictor::Read(T2TModel * model, T2TStateBundle * state)
void T2TPredictor::Read(T2TModel* model, T2TStateBundle* state)
{
m = model;
s = state;
......@@ -147,8 +154,7 @@ predict the next state
>> paddingEnc - padding of the encoder
>>> isStart - is the start or not
*/
void T2TPredictor::Predict(T2TStateBundle * next, XTensor * encoding,
XTensor * inputEnc, XTensor * paddingEnc, bool isStart)
void T2TPredictor::Predict(T2TStateBundle* next, XTensor* encoding, XTensor* inputEnc, XTensor* paddingEnc, bool isStart)
{
int dims[MAX_TENSOR_DIM_NUM];
......@@ -159,40 +165,41 @@ void T2TPredictor::Predict(T2TStateBundle * next, XTensor * encoding,
XTensor first;
CheckNTErrors(inputEnc->order >= 2, "Wrong order of the tensor!");
for(int i = 0; i < inputEnc->order - 1; i++)
for (int i = 0; i < inputEnc->order - 1; i++)
dims[i] = inputEnc->GetDim(i);
dims[inputEnc->order - 1] = 1;
InitTensor(&first, inputEnc->order, dims, X_INT, inputEnc->devID);
InitTensorV2(&first, inputEnc->order, dims, X_INT, 1.0F, inputEnc->devID);
SetDataFixedInt(first, startSymbol);
/* add a new word into the input sequence of the decoder side */
if (isStart) {
inputDec = Identity(first);
}
else{
else {
/* only pass one step to the decoder */
inputDec = GetLastPrediction(s);
inputDec.SetDevice(inputEnc->devID);
}
/* prediction probabilities */
XTensor &output = next->prob;
XTensor& output = next->prob;
XTensor decoding;
for(int i = 0; i < inputDec.order - 1; i++)
for (int i = 0; i < inputDec.order - 1; i++)
dims[i] = inputDec.GetDim(i);
dims[inputDec.order - 1] = inputDec.GetDim(-1);
XTensor paddingDec;
InitTensor(&paddingDec, inputDec.order, dims, X_INT, paddingEnc->devID);
InitTensorV2(&paddingDec, inputDec.order, dims, X_INT, 1.0F, paddingEnc->devID);
SetDataFixedInt(paddingDec, 1);
XTensor maskDec;
XTensor maskEncDec;
/* decoder mask */
m->MakeMTMaskDec(*inputEnc, inputDec, *paddingEnc, paddingDec, maskDec, maskEncDec, 0);
//m->MakeMTMaskDec(*inputEnc, inputDec, *paddingEnc, paddingDec, maskDec, maskEncDec, 0);
/* make the decoding network */
decoding = m->decoder->Make(inputDec, *encoding, NULL, maskEncDec, false);
......@@ -207,34 +214,34 @@ void T2TPredictor::Predict(T2TStateBundle * next, XTensor * encoding,
generate paths up to the states of the current step
>> state - state bundle of the current step
*/
XTensor T2TPredictor::GeneratePaths(T2TStateBundle * state)
XTensor T2TPredictor::GeneratePaths(T2TStateBundle* state)
{
CheckNTErrors(state->stateNum >= 0, "Illegal state!");
int distance = -1;
for(int i = 0; i < state->stateNum; i++){
T2TState * cur = state->states + i;
for (int i = 0; i < state->stateNum; i++) {
T2TState* cur = state->states + i;
int nsteps = 0;
while(cur != NULL){
while (cur != NULL) {
nsteps++;
cur = cur->last;
}
if(nsteps > distance)
if (nsteps > distance)
distance = nsteps;
}
XTensor path;
InitTensor2D(&path, state->stateNum, distance, X_INT);
InitTensor2DV2(&path, state->stateNum, distance, X_INT);
path.SetZeroAll();
for(int i = 0; i < state->stateNum; i++){
T2TState * cur = state->states + i;
for (int i = 0; i < state->stateNum; i++) {
T2TState* cur = state->states + i;
int nsteps = 0;
while(cur != NULL){
while (cur != NULL) {
nsteps++;
path.Set2DInt(cur->prediction, i, distance - nsteps);
cur = cur->last;
......@@ -253,7 +260,7 @@ XTensor T2TPredictor::GetLastPrediction(T2TStateBundle* state)
CheckNTErrors(state->stateNum >= 0, "Illegal state!");
XTensor lastPred;
InitTensor2D(&lastPred, state->stateNum, 1, X_INT);
InitTensor2DV2(&lastPred, state->stateNum, 1, X_INT);
for (int i = 0; i < state->stateNum; i++) {
T2TState* cur = state->states + i;
......
source/sample/transformer/T2TPredictor.h
......@@ -15,7 +15,7 @@
* limitations under the License.
*/
/*
/*
* $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2019-03-13
* This is the first source file I create in 2019 - new start!
*/
......@@ -66,7 +66,7 @@ public:
int nstep;
/* pointer to the previous state */
T2TState * last;
T2TState* last;
};
/* a bundle of states */
......@@ -95,7 +95,7 @@ public:
XTensor nstep;
/* list of states */
T2TState * states;
T2TState* states;
/* number of states */
int stateNum;
......@@ -123,10 +123,10 @@ class T2TPredictor
{
private:
/* pointer to the transformer model */
T2TModel * m;
T2TModel* m;
/* current state */
T2TStateBundle * s;
T2TStateBundle* s;
/* start symbol */
int startSymbol;
......@@ -139,19 +139,19 @@ public:
~T2TPredictor();
/* create an initial state */
void Create(T2TModel * model, XTensor * top, const XTensor * input, int beamSize, T2TStateBundle * state);
void Create(T2TModel* model, XTensor* top, const XTensor* input, int beamSize, T2TStateBundle* state);
/* set the start symbol */
void SetStartSymbol(int symbol);
/* read a state */
void Read(T2TModel * model, T2TStateBundle * state);
void Read(T2TModel* model, T2TStateBundle* state);
/* predict the next state */
void Predict(T2TStateBundle * next, XTensor * encoding, XTensor * inputEnc, XTensor * paddingEnc, bool isStart);
void Predict(T2TStateBundle* next, XTensor* encoding, XTensor* inputEnc, XTensor* paddingEnc, bool isStart);
/* generate paths up to the states of the current step */
XTensor GeneratePaths(T2TStateBundle * state);
XTensor GeneratePaths(T2TStateBundle* state);
/* get the predictions of the previous step */
XTensor GetLastPrediction(T2TStateBundle* state);
......
source/sample/transformer/T2TSearch.h
......@@ -15,7 +15,7 @@
* limitations under the License.
*/
/*
/*
* $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2019-03-27
*/
......@@ -51,10 +51,10 @@ private:
int batchSize;
/* we keep the final hypotheses in a heap for each sentence in the batch. */
XHeap<MIN_HEAP, float> * fullHypos;
XHeap<MIN_HEAP, float>* fullHypos;
/* array of the end symbols */
int * endSymbols;
int* endSymbols;
/* number of the end symbols */
int endSymbolNum;
......@@ -70,40 +70,40 @@ public:
~T2TSearch();
/* initialize the model */
void Init(int argc, char ** argv);
void Init(int argc, char** argv);
/* search for the most promising states */
void Search(T2TModel * model, XTensor * input, XTensor * padding, XTensor * output);
void Search(T2TModel* model, XTensor* input, XTensor* padding, XTensor* output);
/* preparation */
void Prepare(int myBatchSize,int myBeamSize);
void Prepare(int myBatchSize, int myBeamSize);
/* compute the model score for each hypothesis */
void Score(T2TStateBundle * prev, T2TStateBundle * beam);
void Score(T2TStateBundle* prev, T2TStateBundle* beam);
/* generate token indices via beam pruning */
void Generate(T2TStateBundle * beam);
void Generate(T2TStateBundle* beam);
/* expand the search graph */
void Expand(T2TStateBundle * prev, T2TStateBundle * beam);
void Expand(T2TStateBundle* prev, T2TStateBundle* beam);
/* collect hypotheses with ending symbol */
void Collect(T2TStateBundle * beam);
void Collect(T2TStateBundle* beam);
/* fill the hypotheis heap with incomplete hypothses */
void FillHeap(T2TStateBundle * beam);
void FillHeap(T2TStateBundle* beam);
/* save the output sequences in a tensor */
void Dump(XTensor * output);
void Dump(XTensor* output);
/* check if the token is an end symbol */
bool IsEnd(int token);
/* set end symbols for search */
void SetEnd(const int * tokens, const int tokenNum);
void SetEnd(const int* tokens, const int tokenNum);
/* make a mask to prevent duplicated entries in beam expansion for the first position */
XTensor MakeFirstMask(T2TStateBundle * beam);
XTensor MakeFirstMask(T2TStateBundle* beam);
};
}
......
source/sample/transformer/T2TTester.cpp
......@@ -15,7 +15,7 @@
* limitations under the License.
*/
/*
/*
* $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2019-03-27
*/
......@@ -44,7 +44,7 @@ T2TTester::~T2TTester()
}
/* initialize the model */
void T2TTester::Init(int argc, char ** argv)
void T2TTester::Init(int argc, char** argv)
{
LoadParamInt(argc, argv, "vsize", &vSize, 34040);
LoadParamInt(argc, argv, "vsizetgt", &vSizeTgt, vSize);
......@@ -60,7 +60,7 @@ test the model
>> ofn - output data file
>> model - model that is trained
*/
void T2TTester::Test(const char * fn, const char * ofn, T2TModel * model)
void T2TTester::Test(const char* fn, const char* ofn, T2TModel* model)
{
int wc = 0;
int wordCount = 0;
......@@ -94,8 +94,8 @@ void T2TTester::Test(const char * fn, const char * ofn, T2TModel * model)
count++;
wordCount = 0;
for (int i = 0; i < model->decoder->nlayer; ++i) {
model->decoder->selfCache[i].miss = true;
model->decoder->contextCache[i].miss = true;
model->decoder->selfAttCache[i].miss = true;
model->decoder->enDeAttCache[i].miss = true;
}
vector<int> indices = batchLoader.LoadBatch(&batchEnc, &paddingEnc, sentBatch, devID);
......@@ -103,14 +103,14 @@ void T2TTester::Test(const char * fn, const char * ofn, T2TModel * model)
XTensor output;
seacher.Search(model, &batchEnc, &paddingEnc, &output);
output.Dump(stderr);
for (int i = 0; i < indices.size(); ++i) {
Result res;
XTensor sent, srcIdx, tgtIdx;
InitTensor1D(&srcIdx, 1, X_INT, output.devID);
int idx[]{i};
InitTensor1DV2(&srcIdx, 1, X_INT, output.devID);
int idx[]{ i };
srcIdx.SetData(idx, 1);
InitTensor(&tgtIdx, &srcIdx);
InitTensorV2(&tgtIdx, &srcIdx);
SetAscendingOrder(tgtIdx, 0);
sent = CopyIndexed(output, 0, srcIdx, tgtIdx);
......@@ -153,7 +153,7 @@ dump the result into the file
>> file - data file
>> output - output tensor
*/
void T2TTester::Dump(FILE * file, XTensor * output)
void T2TTester::Dump(FILE* file, XTensor* output)
{
int seqLength = output->GetDim(-1);
......
source/sample/transformer/T2TTester.h
......@@ -15,7 +15,7 @@
* limitations under the License.
*/
/*
/*
* $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2019-03-27
* A week with no trips :)
*/
......@@ -56,13 +56,13 @@ public:
~T2TTester();
/* initialize the model */
void Init(int argc, char ** argv);
void Init(int argc, char** argv);
/* test the model */
void Test(const char * fn, const char * ofn, T2TModel * model);
void Test(const char* fn, const char* ofn, T2TModel* model);
/* dump the result into the file */
void Dump(FILE * file, XTensor * output);
void Dump(FILE* file, XTensor* output);
};
}
......
source/sample/transformer/Transformer.h
......@@ -38,7 +38,7 @@ namespace transformer
{
/* entrance of the program */
int TransformerMain(int argc, const char ** argv);
int TransformerMain(int argc, const char** argv);
}
......
source/tensor/XList.cpp
......@@ -28,6 +28,7 @@
#include "XList.h"
#include "XGlobal.h"
/* the nts (NiuTrans.Tensor) namespace */
namespace nts {
......@@ -363,6 +364,8 @@ template struct TensorListBase<long>;
template struct TensorListBase<float>;
template struct TensorListBase<short>;
template struct TensorListBase<XTensor*>;
template struct TensorListBase<uint64_t>;
template struct TensorListBase<void*>;
} /* end of the nts (NiuTrans.Tensor) namespace */
\ No newline at end of file
source/tensor/XList.h
......@@ -26,6 +26,8 @@
#include "XMem.h"
#include "XGlobal.h"
#include <cstdint>
#ifndef __TensorList_H__
#define __TensorList_H__
......@@ -118,7 +120,14 @@ public:
void Shuffle(int nround = 10, int beg = -1, int len = 0);
/* short */
T& operator[] (int i) { return GetItem(i); };
T& operator[] (int i) {
CheckNTErrors(i >= -count && i < count, "Index of a list item is out of scope!");
CheckNTErrors(count > 0, "Cannt index the item in an empty list!");
if (i < 0)
return items[count + i];
else
return items[i];
};
T& Get(int i) { return GetItem(i); };
void Set(int i, T item) { SetItem(i, item); };
};
......@@ -132,7 +141,7 @@ typedef TensorListBase<char*> StrList;
typedef TensorListBase<long> LongList;
typedef TensorListBase<float> FloatList;
typedef TensorListBase<short> ShortList;
typedef TensorListBase<uint64_t> UInt64List;
typedef TensorListBase<XTensor*> TensorList;
} /* end of the nts (NiuTrans.Tensor) namespace */
......
source/tensor/core/reduce/ReduceMax.cpp
......@@ -86,7 +86,7 @@ void _funcCPUName(const XTensor * input, XTensor * output, int dim)
vecBuf[j] = VectorBuffer::loadu((DTYPE*)(ip)+j * vecBufLength); \
} \
for (int j = 1; j < strideNum / 32; j++) { \
const DTYPE* ptr = (DTYPE*)(ip + j * vecBufLength); \
const DTYPE* ptr = (DTYPE*)(ip + j * 4 * vecBufLength); \
vecBuf[0] = vecBuf[0]._vectorOp(VectorBuffer::loadu(ptr + 0 * vecBufLength)); \
vecBuf[1] = vecBuf[1]._vectorOp(VectorBuffer::loadu(ptr + 1 * vecBufLength)); \
vecBuf[2] = vecBuf[2]._vectorOp(VectorBuffer::loadu(ptr + 2 * vecBufLength)); \
......@@ -106,7 +106,7 @@ void _funcCPUName(const XTensor * input, XTensor * output, int dim)
else { \
/* data is separated */ \
for(int i = 0; i < blockNum; i++){ \
for(int j = 0; j < input->dimSize[input->order - 1] / 32; j++){ \
for(int j = 0; j < stride / 32; j++){ \
DTYPE * ip = (DTYPE*)input->data + blockSize * i; \
DTYPE * op = (DTYPE*)output->data + stride * i; \
VectorBuffer vecBuf[4]; \
......
source/tensor/core/reduce/ReduceMean.cpp
......@@ -42,7 +42,7 @@ void _ReduceMean(const XTensor * input, XTensor * output, int dim)
int num = input->dimSize[dim];
_ReduceSum(input, output, dim);
_ScaleAndShiftMe(output, (DTYPE)1/num, 0);
_ScaleAndShiftMe(output, 1.0F/(DTYPE)(num), 0);
}
/*
......
source/tensor/core/reduce/ReduceSum.cpp
......@@ -105,7 +105,7 @@ void _ReduceSum(const XTensor * input, XTensor * output, int dim, const XTensor
vecBuf[j] = VectorBuffer::loadu((DTYPE*)(ip) + j * vecBufLength, isExp, power, bias);
}
for(int j = 1; j < strideNum / 32; j++){
const DTYPE* ptr = (DTYPE*)(ip + j * vecBufLength);
const DTYPE* ptr = (DTYPE*)(ip + (j * 4) * vecBufLength);
vecBuf[0] = vecBuf[0] + VectorBuffer::loadu(ptr + 0 * vecBufLength, isExp, power, bias);
vecBuf[1] = vecBuf[1] + VectorBuffer::loadu(ptr + 1 * vecBufLength, isExp, power, bias);
vecBuf[2] = vecBuf[2] + VectorBuffer::loadu(ptr + 2 * vecBufLength, isExp, power, bias);
......@@ -122,7 +122,7 @@ void _ReduceSum(const XTensor * input, XTensor * output, int dim, const XTensor
} else{
//data is separated
for(int i = 0; i < blockNum; i++){
for(int j = 0; j < input->dimSize[input->order - 1] / 32; j++){
for(int j = 0; j < stride / 32; j++){
DTYPE * ip = (DTYPE*)input->data + blockSize * i;
DTYPE * op = (DTYPE*)output->data + stride * i;
DTYPE * sp = shift != NULL ? (DTYPE*)shift->data + stride * i : NULL;
......@@ -133,8 +133,7 @@ void _ReduceSum(const XTensor * input, XTensor * output, int dim, const XTensor
}
VectorBuffer vecBuf[4];
for(int k = 0; k < 4; k++){
vecBuf[k] = VectorBuffer::loadu((DTYPE*)(ip) + (j * 4 + k) * 32 / sizeof(DTYPE), isExp, power, bias + j * 32 / sizeof(DTYPE));
vecBuf[k] = VectorBuffer::loadu((DTYPE*)(ip) + (j * 4 + k) * 32 / sizeof(DTYPE), isExp, power, bias + k * 32 / sizeof(DTYPE));
}
for(int k = 1; k < strideNum; k++){
DTYPE * ptr = ip + k * stride + (j * 4) * vecBufLength;
......
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论