mmr/tkDNN
Archived
1
0
Fork 0
Code Issues 46 Pull Requests 12 Releases 4 Activity

batch seems ok in yolo3_berkely

Browse Source 
layers to be checked:
DeformableConvRT
FlattenConcatRT
ReshapeRT
RouteRT (dont know why but seems working)
This commit is contained in:
Francesco Gatti
2020-04-21 19:41:40 +00:00
parent 7c81c5a43c
commit 3d940a9fa2
10 changed files with 51 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/tkDNN/NetworkRT.h
+8
View File
@@ -63,6 +63,7 @@ public:
const static int MAX_BUFFERS_RT = 10;
void* buffersRT[MAX_BUFFERS_RT];
dataDim_t buffersDIM[MAX_BUFFERS_RT];
int buf_input_idx, buf_output_idx;
dataDim_t input_dim, output_dim;
@@ -81,6 +82,13 @@ public:
return 0;
}
int getBuffersN() {
if(engineRT != nullptr)
return engineRT->getNbBindings();
else
return 0;
}
/**
Do inferece
*/
include/tkDNN/pluginsRT/ActivationLeakyRT.h
+1 -1
View File
@@ -42,7 +42,7 @@ public:
virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override {
activationLEAKYForward((dnnType*)reinterpret_cast<const dnnType*>(inputs[0]),
reinterpret_cast<dnnType*>(outputs[0]), size, stream);
reinterpret_cast<dnnType*>(outputs[0]), batchSize*size, stream);
return 0;
}
include/tkDNN/pluginsRT/ActivationReLUCeilingRT.h
+1 -1
View File
@@ -41,7 +41,7 @@ public:
virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override {
activationReLUCeilingForward((dnnType*)reinterpret_cast<const dnnType*>(inputs[0]),
reinterpret_cast<dnnType*>(outputs[0]), size, ceiling, stream);
reinterpret_cast<dnnType*>(outputs[0]), batchSize*size, ceiling, stream);
return 0;
}
include/tkDNN/pluginsRT/ActivationSigmoidRT.h
+1 -1
View File
@@ -42,7 +42,7 @@ public:
virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override {
activationSIGMOIDForward((dnnType*)reinterpret_cast<const dnnType*>(inputs[0]),
reinterpret_cast<dnnType*>(outputs[0]), size, stream);
reinterpret_cast<dnnType*>(outputs[0]), batchSize*size, stream);
return 0;
}
include/tkDNN/pluginsRT/MaxPoolingFixedSizeRT.h
+1 -1
View File
@@ -45,7 +45,7 @@ public:
//std::cout<<this->n<<" "<<this->c<<" "<<this->h<<" "<<this->w<<" "<<this->stride_H<<" "<<this->stride_W<<" "<<this->winSize<<" "<<this->padding<<std::endl;
dnnType *srcData = (dnnType*)reinterpret_cast<const dnnType*>(inputs[0]);
dnnType *dstData = reinterpret_cast<dnnType*>(outputs[0]);
MaxPoolingForward(srcData, dstData, this->n, this->c, this->h, this->w, this->stride_H, this->stride_W, this->winSize, this->padding);
MaxPoolingForward(srcData, dstData, batchSize, this->c, this->h, this->w, this->stride_H, this->stride_W, this->winSize, this->padding);
return 0;
}
include/tkDNN/pluginsRT/RegionRT.h
+5 -5
View File
@@ -50,16 +50,16 @@ public:
for (int b = 0; b < batchSize; ++b){
for(int n = 0; n < num; ++n){
int index = entry_index(b, n*w*h, 0, batchSize);
int index = entry_index(b, n*w*h, 0);
activationLOGISTICForward(srcData + index, dstData + index, 2*w*h, stream);
index = entry_index(b, n*w*h, coords, batchSize);
index = entry_index(b, n*w*h, coords);
activationLOGISTICForward(srcData + index, dstData + index, w*h, stream);
}
}
//softmax start
int index = entry_index(0, 0, coords + 1, batchSize);
int index = entry_index(0, 0, coords + 1);
softmaxForward( srcData + index, classes, batchSize*num,
(batchSize*c*h*w)/num,
w*h, 1, w*h, 1, dstData + index, stream);
@@ -85,10 +85,10 @@ public:
int c, h, w;
int classes, coords, num;
int entry_index(int batch, int location, int entry, int batchSize) {
int entry_index(int batch, int location, int entry) {
int n = location / (w*h);
int loc = location % (w*h);
return batch*c*h*w*batchSize + n*w*h*(coords+classes+1) + entry*w*h + loc;
return batch*c*h*w + n*w*h*(coords+classes+1) + entry*w*h + loc;
}
};
include/tkDNN/pluginsRT/YoloRT.h
+4 -4
View File
@@ -62,10 +62,10 @@ public:
for (int b = 0; b < batchSize; ++b){
for(int n = 0; n < n_masks; ++n){
int index = entry_index(b, n*w*h, 0, batchSize);
int index = entry_index(b, n*w*h, 0);
activationLOGISTICForward(srcData + index, dstData + index, 2*w*h, stream);
index = entry_index(b, n*w*h, 4, batchSize);
index = entry_index(b, n*w*h, 4);
activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*w*h, stream);
}
}
@@ -109,10 +109,10 @@ public:
dnnType *mask;
dnnType *bias;
int entry_index(int batch, int location, int entry, int batchSize) {
int entry_index(int batch, int location, int entry) {
int n = location / (w*h);
int loc = location % (w*h);
return batch*c*h*w*batchSize + n*w*h*(4+classes+1) + entry*w*h + loc;
return batch*c*h*w + n*w*h*(4+classes+1) + entry*w*h + loc;
}
};