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

Deformable batch works

Browse Source 
Signed-off-by: Davide Sapienza <sapienza.dav@gmail.com>
This commit is contained in:
dsapienza
2020-04-25 02:47:31 +02:00
parent 13c9dc6620
commit 6e5ab031c4
4 changed files with 35 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/DeformConv2d.cpp
+2 -2
View File
@@ -102,13 +102,13 @@ dnnType* DeformConv2d::infer(dataDim_t &dim, dnnType* srcData) {
dcnV2CudaForward(stat, handle,
srcData, this->data_d,
this->bias2_d, ones_d1,
offset, mask,
offset, mask,
dstData, ones_d2,
this->kernelH, this->kernelW,
this->strideH, this->strideW,
this->paddingH, this->paddingW,
1, 1,
this->deformableGroup,
this->deformableGroup, 0, //batch_id for cudnn is set to 0 (no batch)
preconv->input_dim.n, preconv->input_dim.c, preconv->input_dim.h, preconv->input_dim.w,
this->output_dim.n, this->output_dim.c, this->output_dim.h, this->output_dim.w,
chunk_dim);
src/kernels/deformable_conv.cu
+7 -5
View File
@@ -241,12 +241,13 @@ void dcnV2CudaForward(cublasStatus_t stat, cublasHandle_t handle,
const int stride_h, const int stride_w,
const int pad_h, const int pad_w,
const int dilation_h, const int dilation_w,
const int deformable_group,
const int deformable_group, const int batch_id,
const int in_n, const int in_c, const int in_h, const int in_w,
const int out_n, const int out_c, const int out_h, const int out_w,
const int chunk_dim, cudaStream_t stream)
{
// stat and handle have be moved out to preserve 2 - 6 milliseconds every 100.
const int batch = batch_id;
const int channels = in_c;
const int height = in_h;
const int width = in_w;
@@ -265,13 +266,14 @@ void dcnV2CudaForward(cublasStatus_t stat, cublasHandle_t handle,
stat = cublasSgemm(handle, CUBLAS_OP_T, CUBLAS_OP_N,
n, m, k, &alpha,
ones, k, bias, k,
&beta, output, n);
&beta, output + batch * out_c * out_h * out_w, n);
if (stat != CUBLAS_STATUS_SUCCESS)
FatalError("CUBLAS initialization failed\n");
modulatedDeformableIm2colCuda(stream,
input, offset,
mask,
input + batch * channels * height * width,
offset,// + b * 2 * int((float)chunk_dim / batch),
mask,// + b * int((float)chunk_dim / batch),
1, channels, height, width,
height_out, width_out, deformable_group, columns);
// modulatedDeformableIm2colCudaGeneralVersion(stream,
@@ -290,7 +292,7 @@ void dcnV2CudaForward(cublasStatus_t stat, cublasHandle_t handle,
stat = cublasSgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N,
n, m, k, &alpha,
columns, n, weight, k,
&beta, output, n);
&beta, output + batch * out_c * out_h * out_w, n);
if (stat != CUBLAS_STATUS_SUCCESS)
FatalError("CUBLAS initialization failed\n");