leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 29 Packages Projects Releases Wiki Activity

CUTLASS 2.6.1 - functional and performance enhancements to strided DGRAD, fixes, and tuning

Browse Source 
* cutlass 2.6 update

* remove debug prints

* cutlass 2.6.1 (minor update)

* Updated CHANGELOG.

* Minor edit to readme to indicate patch version.

* Minor edit to readme.

Co-authored-by:  Haicheng Wu <haichengw@nvidia.com>, Andrew Kerr <akerr@nvidia.com>
This commit is contained in:
Manish Gupta
2021-09-03 10:26:15 -07:00
committed by GitHub
co-authored by Haicheng Wu <haichengw@nvidia.com>, Andrew Kerr <akerr@nvidia.com>
parent a01feb93d9
commit 6c2f8f2fb8
55 changed files with 317 additions and 315 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/unit/conv/device/conv2d_problems.h
+43
View File
@@ -281,6 +281,22 @@ struct TestbedConv2dProblemSizes {
{1, 1} // dilation (dilation_h, dilation_w)
));
conv2d_default_sizes.push_back(cutlass::conv::Conv2dProblemSize(
{1, 8, 8, 8}, // input size (NHWC)
{8, 3, 3, 8}, // filter size (KRSC)
{1, 1, 1, 1}, // padding (pad_h, _, pad_w, _)
{3, 3}, // stride (stride_h, stride_w)
{1, 1} // dilation (dilation_h, dilation_w)
));
conv2d_default_sizes.push_back(cutlass::conv::Conv2dProblemSize(
{1, 16, 16, 8}, // input size (NHWC)
{8, 3, 3, 8}, // filter size (KRSC)
{3, 3, 3, 3}, // padding (pad_h, _, pad_w, _)
{3, 3}, // stride (stride_h, stride_w)
{1, 1} // dilation (dilation_h, dilation_w)
));
////////////////////////////////////////////////////////////////////////////////////
// Medium input size (1x16x16x128), filter size (1x1, 2x2, 3x3, 5x5), stride (1, 1)
////////////////////////////////////////////////////////////////////////////////////
@@ -389,6 +405,25 @@ struct TestbedConv2dProblemSizes {
{1, 1} // dilation (dilation_h, dilation_w)
));
////////////////////////////////////////////////////////////////////////////////////
// Medium input size padding > stride, asymmetric filter, padding and striding
////////////////////////////////////////////////////////////////////////////////////
conv2d_default_sizes.push_back(cutlass::conv::Conv2dProblemSize(
{1, 27, 31, 256}, // input size (NHWC)
{512, 3, 3, 256}, // filter size (KRSC)
{5, 5, 7, 7}, // padding (pad_h, _, pad_w, _)
{3, 4}, // stride (stride_h, stride_w)
{1, 1} // dilation (dilation_h, dilation_w)
));
conv2d_default_sizes.push_back(cutlass::conv::Conv2dProblemSize(
{1, 27, 35, 256}, // input size (NHWC)
{512, 7, 5, 256}, // filter size (KRSC)
{11, 11, 7, 7}, // padding (pad_h, _, pad_w, _)
{3, 5}, // stride (stride_h, stride_w)
{1, 1} // dilation (dilation_h, dilation_w)
));
////////////////////////////////////////////////////////////////////////////////////
// Medium input size *mixed* stride (1, 2) and (2, 1),
// filter (3, 3), default padding
@@ -419,6 +454,14 @@ struct TestbedConv2dProblemSizes {
{2, 2}, // stride (stride_h, stride_w)
{1, 1} // dilation (dilation_h, dilation_w)
));
conv2d_default_sizes.push_back(cutlass::conv::Conv2dProblemSize(
{1, 32, 32, 16}, // input size (NHWC)
{32, 3, 3, 16}, // filter size (KRSC)
{1, 1, 1, 1}, // padding (pad_h, _, pad_w, _)
{6, 2}, // stride (stride_h, stride_w)
{1, 1} // dilation (dilation_h, dilation_w)
));
conv2d_default_sizes.push_back(cutlass::conv::Conv2dProblemSize(
{32, 32, 32, 32}, // input size (NHWC)