CUTLASS 3.6.0 (#1850)

* v3.6

* update changelog

* update readme

* fix typo

* fixing typos

* hopper gemm with weight prefetch

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
Co-authored-by: Haicheng Wu <haichengw@nvidia.com>
This commit is contained in:
Yujia Zhai
2024-10-09 15:33:27 -04:00
committed by GitHub
co-authored by yuzhai Haicheng Wu
parent 0837a2a00a
commit cc3c29a81a
354 changed files with 105937 additions and 8197 deletions
+13 -1
View File
@@ -718,7 +718,7 @@ bool TestAllConv2d(
}
// CUTLASS DGRAD's *strided* specialization does not support split-k mode
if ((ImplicitGemm::kConvolutionalOperator == cutlass::conv::Operator::kDgrad ||
if ((ImplicitGemm::kConvolutionalOperator == cutlass::conv::Operator::kDgrad ||
ImplicitGemm::kConvolutionalOperator == cutlass::conv::Operator::kDeconv) &&
(ImplicitGemm::UnderlyingKernel::Mma::IteratorA::kStrideSupport ==
cutlass::conv::StrideSupport::kStrided)) {
@@ -734,6 +734,18 @@ bool TestAllConv2d(
cutlass::from_real<typename ImplicitGemm::ElementCompute>(2.0),
cutlass::from_real<typename ImplicitGemm::ElementCompute>(2.0));
passed = testbed.run(
cutlass::conv::Conv2dProblemSize(
{1, 56, 56, 8}, // input size (NHWC)
{8, 1, 1, 8}, // filter size (KRSC)
{0, 0, 0, 0}, // padding (pad_h, _, pad_w, _)
{1, 1}, // stride (stride_h, stride_w)
{1, 1}) // dilation (dilation_h, dilation_w)
.reset_split_k_slices(2),
cutlass::conv::SplitKMode::kSerial,
cutlass::from_real<typename ImplicitGemm::ElementCompute>(2.0),
cutlass::from_real<typename ImplicitGemm::ElementCompute>(2.0));
if (!passed) {
return false;
}