leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 27 Packages Projects Releases Wiki Activity

CUTLASS 3.6.0 (#1850)

Browse Source 
* 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
Download Patch File Download Diff File Expand all files Collapse all files
test/unit/conv/cache_testbed_output.h
+1 -1
View File
@@ -609,7 +609,7 @@ inline CachedTestKey CreateCachedGemmTestKey(
ElementCompute alpha,
ElementCompute beta,
cutlass::TensorView<ElementA, LayoutA> A,
cutlass::TensorView<ElementA, LayoutB> B,
cutlass::TensorView<ElementB, LayoutB> B,
cutlass::TensorView<ElementC, LayoutC> C
) {
test/unit/conv/device/conv2d_testbed.h
+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;
}
test/unit/conv/device/conv3d_testbed.h
+7 -1
View File
@@ -246,12 +246,18 @@ public:
split_k_mode
);
cutlass::Status status = conv3d_op.can_implement(conv3d_args);
if (status != cutlass::Status::kSuccess) {
std::cerr << "can_implement failed for the given problem_size: \n";
return false;
}
// find workspace requirement for parallel split-k reduction
size_t workspace_size = Conv3d::get_workspace_size(conv3d_args);
cutlass::device_memory::allocation<uint8_t> workspace(workspace_size);
cutlass::Status status = conv3d_op.initialize(conv3d_args, workspace.get());
status = conv3d_op.initialize(conv3d_args, workspace.get());
if (status != cutlass::Status::kSuccess) {
cudaError_t error = cudaGetLastError();
test/unit/conv/device_3x/conv_problem_sizes.hpp
+84 -61
View File
@@ -308,8 +308,8 @@ get_conv_problem_vector<2, cutlass::conv::Operator::kFprop>() {
// 2x5 filter, asymmetric padding 1,0/1,0, w/ dilation
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 16, 16, 64},
{256, 2, 5, 64},
{2, 16, 16, 64},
{256, 2, 5, 64},
{1, 1},
{0, 0},
{1, 1},
@@ -319,8 +319,8 @@ get_conv_problem_vector<2, cutlass::conv::Operator::kFprop>() {
// 2x5 filter, asymmetric padding 1,0/1,0, w/ stride, w/ dilation
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 16, 15, 64},
{256, 2, 5, 64},
{2, 16, 15, 64},
{256, 2, 5, 64},
{1, 1},
{0, 0},
{2, 3},
@@ -441,7 +441,7 @@ get_conv_problem_vector<3, cutlass::conv::Operator::kFprop>() {
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 16, 10, 16, 64},
{96, 3, 4, 5, 64},
{96, 3, 4, 5, 64},
{1, 0, 1},
{0, 2, 0},
{1, 1, 1},
@@ -452,7 +452,7 @@ get_conv_problem_vector<3, cutlass::conv::Operator::kFprop>() {
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 16, 10, 16, 64},
{96, 3, 4, 5, 64},
{96, 3, 4, 5, 64},
{1, 0, 1},
{0, 2, 0},
{2, 2, 3},
@@ -462,6 +462,7 @@ get_conv_problem_vector<3, cutlass::conv::Operator::kFprop>() {
return problem_shapes;
}
/////////////////////////////////////////////////////////////////////////////////////////////////
// Wgrad
/////////////////////////////////////////////////////////////////////////////////////////////////
@@ -570,6 +571,28 @@ get_conv_problem_vector<1, cutlass::conv::Operator::kWgrad>() {
{2},
1
});
// To test streamk, equals to gemm-MxNxK size 128x640x2048
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 1024, 128},
{640, 1, 128},
{0},
{0},
{1},
{1},
1
});
// To test streamk, equals to gemm-MxNxK size 128x640x2080
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 1040, 128},
{640, 1, 128},
{0},
{0},
{1},
{1},
1
});
return problem_shapes;
}
@@ -659,7 +682,7 @@ get_conv_problem_vector<2, cutlass::conv::Operator::kWgrad>() {
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 15, 16, 32},
{256, 2, 5, 32},
{256, 2, 5, 32},
{1, 1},
{0, 0},
{2, 3},
@@ -670,7 +693,7 @@ get_conv_problem_vector<2, cutlass::conv::Operator::kWgrad>() {
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 16, 16, 32},
{256, 2, 5, 32},
{256, 2, 5, 32},
{1, 1},
{0, 0},
{1, 1},
@@ -681,7 +704,7 @@ get_conv_problem_vector<2, cutlass::conv::Operator::kWgrad>() {
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 16, 15, 32},
{256, 2, 5, 32},
{256, 2, 5, 32},
{1, 1},
{0, 0},
{2, 3},
@@ -690,26 +713,26 @@ get_conv_problem_vector<2, cutlass::conv::Operator::kWgrad>() {
});
// To test streamk, equals to gemm-MxNxK size 128x640x2048
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 64, 16, 128}, // nhwc
{640, 1, 1, 128}, // krsc
{0, 0}, // padding lower (pad_h, pad_w)
{0, 0}, // padding upper (pad_h, pad_w)
{1, 1}, // stride (stride_h, stride_w)
{1, 1}, // dilation (dilation_h, dilation_w)
1 // group
});
cutlass::conv::Mode::kCrossCorrelation,
{2, 64, 16, 128},
{640, 1, 1, 128},
{0, 0},
{0, 0},
{1, 1},
{1, 1},
1
});
// To test streamk, equals to gemm-MxNxK size 128x640x2080
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 65, 16, 128}, // nhwc
{640, 1, 1, 128}, // krsc
{0, 0}, // padding lower (pad_h, pad_w)
{0, 0}, // padding upper (pad_h, pad_w)
{1, 1}, // stride (stride_h, stride_w)
{1, 1}, // dilation (dilation_h, dilation_w)
1 // group
});
cutlass::conv::Mode::kCrossCorrelation,
{2, 65, 16, 128},
{640, 1, 1, 128},
{0, 0},
{0, 0},
{1, 1},
{1, 1},
1
});
return problem_shapes;
}
@@ -755,7 +778,7 @@ get_conv_problem_vector<3, cutlass::conv::Operator::kWgrad>() {
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 16, 10, 16, 32},
{96, 3, 4, 5, 32},
{96, 3, 4, 5, 32},
{1, 0, 1},
{0, 2, 0},
{2, 2, 3},
@@ -766,7 +789,7 @@ get_conv_problem_vector<3, cutlass::conv::Operator::kWgrad>() {
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 16, 10, 16, 32},
{96, 3, 4, 5, 32},
{96, 3, 4, 5, 32},
{1, 0, 1},
{0, 2, 0},
{1, 1, 1},
@@ -775,26 +798,26 @@ get_conv_problem_vector<3, cutlass::conv::Operator::kWgrad>() {
});
// To test streamk, equals to gemm-MxNxK size 128x640x2048
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 1, 64, 16, 128}, // ndhwc
{640, 1, 1, 1, 128}, // ktrsc
{0, 0, 0}, // padding lower (pad_d, pad_h, pad_w)
{0, 0, 0}, // padding upper (pad_d, pad_h, pad_w)
{1, 1, 1}, // stride (stride_d, stride_h, stride_w)
{1, 1, 1}, // dilation (dilation_d, dilation_h, dilation_w)
1 // group
});
cutlass::conv::Mode::kCrossCorrelation,
{2, 1, 64, 16, 128},
{640, 1, 1, 1, 128},
{0, 0, 0},
{0, 0, 0},
{1, 1, 1},
{1, 1, 1},
1
});
// To test streamk, equals to gemm-MxNxK size 128x640x2080
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 1, 65, 16, 128}, // ndhwc
{640, 1, 1, 1, 128}, // ktrsc
{0, 0, 0}, // padding lower (pad_d, pad_h, pad_w)
{0, 0, 0}, // padding upper (pad_d, pad_h, pad_w)
{1, 1, 1}, // stride (stride_d, stride_h, stride_w)
{1, 1, 1}, // dilation (dilation_d, dilation_h, dilation_w)
1 // group
});
cutlass::conv::Mode::kCrossCorrelation,
{2, 1, 65, 16, 128},
{640, 1, 1, 1, 128},
{0, 0, 0},
{0, 0, 0},
{1, 1, 1},
{1, 1, 1},
1
});
return problem_shapes;
}
@@ -848,8 +871,8 @@ get_conv_problem_vector<1, cutlass::conv::Operator::kDgrad, false>() {
// Filter-K = 16 for predication
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{1, 8, 16},
{64,1, 16},
{1, 8, 16},
{64, 1, 16},
{0},
{0},
{1},
@@ -870,7 +893,7 @@ get_conv_problem_vector<1, cutlass::conv::Operator::kDgrad, false>() {
// N = 7 and K = 256 for a even larger grid
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{7, 8, 256},
{7, 8, 256},
{64, 1, 256},
{0},
{0},
@@ -881,7 +904,7 @@ get_conv_problem_vector<1, cutlass::conv::Operator::kDgrad, false>() {
// 3 filter, no padding
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 8, 256},
{2, 8, 256},
{64, 3, 256},
{0},
{0},
@@ -892,7 +915,7 @@ get_conv_problem_vector<1, cutlass::conv::Operator::kDgrad, false>() {
// 3 filter, symmetric padding with k % cta_k !=0
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 8, 256},
{2, 8, 256},
{32, 3, 256},
{1},
{1},
@@ -903,7 +926,7 @@ get_conv_problem_vector<1, cutlass::conv::Operator::kDgrad, false>() {
// 4 filter, asymmetric padding
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 8, 256},
{2, 8, 256},
{64, 4, 256},
{0},
{1},
@@ -915,7 +938,7 @@ get_conv_problem_vector<1, cutlass::conv::Operator::kDgrad, false>() {
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 16, 64},
{256, 3, 64},
{256, 3, 64},
{0},
{1},
{1},
@@ -993,7 +1016,7 @@ get_conv_problem_vector<2, cutlass::conv::Operator::kDgrad, false>() {
// N = 7 and K = 256 for a even larger grid
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{7, 8, 8, 256},
{7, 8, 8, 256},
{64, 1, 1, 256},
{0, 0},
{0, 0},
@@ -1004,7 +1027,7 @@ get_conv_problem_vector<2, cutlass::conv::Operator::kDgrad, false>() {
// 3x3 filter, no padding
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 8, 8, 256},
{2, 8, 8, 256},
{64, 3, 3, 256},
{0, 0},
{0, 0},
@@ -1015,7 +1038,7 @@ get_conv_problem_vector<2, cutlass::conv::Operator::kDgrad, false>() {
// 3x3 filter, symmetric padding with k % cta_k !=0
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 8, 8, 256},
{2, 8, 8, 256},
{32, 3, 3, 256},
{1, 1},
{1, 1},
@@ -1026,7 +1049,7 @@ get_conv_problem_vector<2, cutlass::conv::Operator::kDgrad, false>() {
// 2x5 filter, asymmetric padding 1,0/1,0
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 8, 8, 256},
{2, 8, 8, 256},
{64, 2, 5, 256},
{1, 1},
{0, 0},
@@ -1038,7 +1061,7 @@ get_conv_problem_vector<2, cutlass::conv::Operator::kDgrad, false>() {
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 16, 16, 64},
{256, 2, 5, 64},
{256, 2, 5, 64},
{1, 1},
{0, 0},
{1, 1},
@@ -1116,7 +1139,7 @@ get_conv_problem_vector<3, cutlass::conv::Operator::kDgrad, false>() {
problem_shapes.push_back({
cutlass::conv::Mode::kCrossCorrelation,
{2, 16, 10, 16, 64},
{64, 3, 4, 5, 96},
{64, 3, 4, 5, 96},
{1, 0, 1},
{0, 2, 0},
{1, 1, 1},
@@ -1247,7 +1270,7 @@ get_conv_problem_vector<3, cutlass::conv::Operator::kDgrad, true>() {
{64, 3, 4, 5, 96},
{1, 0, 1},
{0, 2, 0},
{2, 4, 2},
{2, 1, 2},
{4, 2, 3},
1
});
test/unit/conv/device_3x/dgrad/sm90_conv1d_dgrad_implicit_gemm_f16_f16_f32_tensorop_f16.cu
+16
View File
@@ -87,7 +87,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -130,7 +132,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -216,7 +222,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -264,7 +272,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -307,7 +317,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -350,7 +362,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -393,7 +407,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/dgrad/sm90_conv1d_dgrad_implicit_gemm_f16_f16_f32_tensorop_f32.cu
+16
View File
@@ -87,7 +87,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -130,7 +132,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -216,7 +222,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -263,7 +271,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -306,7 +316,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -349,7 +361,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -392,7 +406,9 @@ TEST(SM90_device_conv1d_dgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/dgrad/sm90_conv2d_dgrad_implicit_gemm_f16_f16_f32_tensorop_f16.cu
+16
View File
@@ -88,7 +88,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -132,7 +134,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -176,7 +180,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -220,7 +226,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -268,7 +276,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -312,7 +322,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -356,7 +368,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -400,7 +414,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/dgrad/sm90_conv2d_dgrad_implicit_gemm_f16_f16_f32_tensorop_f32.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -129,7 +131,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -217,7 +223,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -265,7 +273,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -309,7 +319,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -353,7 +365,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -397,7 +411,9 @@ TEST(SM90_device_conv2d_dgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/dgrad/sm90_conv3d_dgrad_implicit_gemm_f16_f16_f32_tensorop_f16.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -130,7 +132,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -174,7 +178,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -218,7 +224,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -266,7 +274,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -310,7 +320,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -354,7 +366,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -398,7 +412,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/dgrad/sm90_conv3d_dgrad_implicit_gemm_f16_f16_f32_tensorop_f32.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -129,7 +131,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -217,7 +223,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -265,7 +273,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -309,7 +319,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -353,7 +365,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -397,7 +411,9 @@ TEST(SM90_device_conv3d_dgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv1d_fprop_implicit_gemm_f16_f16_f32_tensorop_f16.cu
+17 -1
View File
@@ -84,7 +84,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -127,7 +129,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -169,8 +173,10 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::StageCountAutoCarveout<static_cast<int>(sizeof(typename CollectiveEpilogue::SharedStorage))>,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -213,7 +219,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -261,7 +269,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -304,7 +314,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -347,7 +359,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -390,7 +404,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv1d_fprop_implicit_gemm_f16_f16_f32_tensorop_f32.cu
+19
View File
@@ -84,7 +84,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -127,7 +129,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -170,7 +174,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -213,7 +219,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -260,7 +268,10 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -303,7 +314,10 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -346,7 +360,10 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -389,7 +406,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv1d_fprop_implicit_gemm_s8_s8_s32_tensorop_s32.cu
+21 -1
View File
@@ -55,6 +55,7 @@ using namespace cute;
// Cluster 1x1x1
//
TEST(SM90_device_conv1d_fprop_implicitgemm_s8nwc_s8nwc_s32nwc_tensor_op_s32, 64x64x64_1x1x1) {
using ElementAct = int8_t;
using ElementFlt = int8_t;
using ElementOut = int32_t;
@@ -84,10 +85,13 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_s8nwc_s8nwc_s32nwc_tensor_op_s32, 64x
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape = cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
>;
using Conv = cutlass::conv::device::ConvUniversalAdapter<ConvKernel>;
@@ -128,7 +132,10 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_s8nwc_s8nwc_s32nwc_tensor_op_s32, 64x
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape = cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -171,7 +178,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_s8nwc_s8nwc_s32nwc_tensor_op_s32, 64x
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape = cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -214,7 +223,10 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_s8nwc_s8nwc_s32nwc_tensor_op_s32, 64x
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape = cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -261,7 +273,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_s8nwc_s8nwc_s32nwc_tensor_op_s32, 128
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape = cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -304,7 +318,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_s8nwc_s8nwc_s32nwc_tensor_op_s32, 128
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape = cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -347,7 +363,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_s8nwc_s8nwc_s32nwc_tensor_op_s32, 128
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape = cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -390,7 +408,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_s8nwc_s8nwc_s32nwc_tensor_op_s32, 128
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape = cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv1d_fprop_implicit_gemm_tf32_tf32_f32_tensorop_f32.cu
+17
View File
@@ -84,7 +84,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_tf32nwc_tf32nwc_f32nwc_tensor_op_f32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -127,7 +129,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_tf32nwc_tf32nwc_f32nwc_tensor_op_f32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -170,7 +174,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_tf32nwc_tf32nwc_f32nwc_tensor_op_f32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -213,7 +219,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_tf32nwc_tf32nwc_f32nwc_tensor_op_f32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -260,7 +268,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_tf32nwc_tf32nwc_f32nwc_tensor_op_f32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -303,7 +313,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_tf32nwc_tf32nwc_f32nwc_tensor_op_f32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -346,7 +358,9 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_tf32nwc_tf32nwc_f32nwc_tensor_op_f32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -389,7 +403,10 @@ TEST(SM90_device_conv1d_fprop_implicitgemm_tf32nwc_tf32nwc_f32nwc_tensor_op_f32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv2d_fprop_implicit_gemm_f16_f16_f32_tensorop_f16.cu
+17
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -130,7 +132,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -175,7 +179,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -220,7 +226,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -269,7 +277,10 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -314,7 +325,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -359,7 +372,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -404,7 +419,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv2d_fprop_implicit_gemm_f16_f16_f32_tensorop_f32.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -130,7 +132,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -175,7 +179,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -220,7 +226,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -269,7 +277,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -314,7 +324,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -359,7 +371,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -404,7 +418,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv2d_fprop_implicit_gemm_s8_s8_s32_tensorop_s32.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_s8nhwc_s8nhwc_s32nhwc_tensor_op_s32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -129,7 +131,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_s8nhwc_s8nhwc_s32nhwc_tensor_op_s32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_s8nhwc_s8nhwc_s32nhwc_tensor_op_s32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -217,7 +223,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_s8nhwc_s8nhwc_s32nhwc_tensor_op_s32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -265,7 +273,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_s8nhwc_s8nhwc_s32nhwc_tensor_op_s32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -309,7 +319,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_s8nhwc_s8nhwc_s32nhwc_tensor_op_s32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -353,7 +365,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_s8nhwc_s8nhwc_s32nhwc_tensor_op_s32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -397,7 +411,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_s8nhwc_s8nhwc_s32nhwc_tensor_op_s32,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv2d_fprop_implicit_gemm_tf32_tf32_f32_tensorop_f32.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -130,7 +132,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -175,7 +179,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -220,7 +226,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -269,7 +277,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -314,7 +324,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -359,7 +371,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -404,7 +418,9 @@ TEST(SM90_device_conv2d_fprop_implicitgemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv3d_fprop_implicit_gemm_f16_f16_f32_tensorop_f16.cu
+17 -1
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -128,8 +130,10 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::StageCountAutoCarveout<static_cast<int>(sizeof(typename CollectiveEpilogue::SharedStorage))>,
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -217,7 +223,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -265,7 +273,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -309,7 +319,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -353,7 +365,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -397,7 +411,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv3d_fprop_implicit_gemm_f16_f16_f32_tensorop_f32.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -129,7 +131,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -217,7 +223,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -265,7 +273,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -309,7 +319,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -353,7 +365,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -397,7 +411,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv3d_fprop_implicit_gemm_s8_s8_s32_tensorop_s32.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_s8ndhwc_s8ndhwc_s32ndhwc_tensor_op_s3
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -129,7 +131,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_s8ndhwc_s8ndhwc_s32ndhwc_tensor_op_s3
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_s8ndhwc_s8ndhwc_s32ndhwc_tensor_op_s3
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -217,7 +223,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_s8ndhwc_s8ndhwc_s32ndhwc_tensor_op_s3
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -265,7 +273,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_s8ndhwc_s8ndhwc_s32ndhwc_tensor_op_s3
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -309,7 +319,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_s8ndhwc_s8ndhwc_s32ndhwc_tensor_op_s3
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -353,7 +365,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_s8ndhwc_s8ndhwc_s32ndhwc_tensor_op_s3
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -397,7 +411,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_s8ndhwc_s8ndhwc_s32ndhwc_tensor_op_s3
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/fprop/sm90_conv3d_fprop_implicit_gemm_tf32_tf32_f32_tensorop_f32.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_tf32ndhwc_tf32ndhwc_f32ndhwc_tensor_o
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -129,7 +131,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_tf32ndhwc_tf32ndhwc_f32ndhwc_tensor_o
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_tf32ndhwc_tf32ndhwc_f32ndhwc_tensor_o
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -217,7 +223,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_tf32ndhwc_tf32ndhwc_f32ndhwc_tensor_o
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -265,7 +273,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_tf32ndhwc_tf32ndhwc_f32ndhwc_tensor_o
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -309,7 +319,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_tf32ndhwc_tf32ndhwc_f32ndhwc_tensor_o
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -353,7 +365,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_tf32ndhwc_tf32ndhwc_f32ndhwc_tensor_o
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -397,7 +411,9 @@ TEST(SM90_device_conv3d_fprop_implicitgemm_tf32ndhwc_tf32ndhwc_f32ndhwc_tensor_o
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/testbed_conv.hpp
+9 -6
View File
@@ -120,12 +120,11 @@ struct DenseConvParams {
// get the default arguments without sparse data
auto get_mainloop_arguments(
ProblemShape const& problem_shape,
[[maybe_unused]] ProblemShape const& problem_shape,
thrust::universal_vector<ElementA>& tensor_A,
thrust::universal_vector<ElementB>& tensor_B
) {
auto args = typename Conv::ConvKernel::MainloopArguments {
problem_shape,
tensor_A.data().get(),
tensor_B.data().get(),
};
@@ -298,11 +297,12 @@ struct ConvTestbed {
using DecompositionMode = typename cutlass::gemm::kernel::detail::PersistentTileSchedulerSm90StreamKParams::DecompositionMode;
typename Conv::ConvKernel::TileScheduler::Arguments scheduler_args{};
if constexpr (cute::is_same_v<typename Conv::ConvKernel::TileScheduler::Arguments, cutlass::gemm::StreamKScheduler>) {
if constexpr (cute::is_same_v<typename Conv::ConvKernel::TileSchedulerTag, cutlass::gemm::StreamKScheduler>) {
scheduler_args = { static_cast<int>(splits), static_cast<int>(max_swizzle), raster_order, decomposition_mode };
}
auto mainloop_args = params.get_mainloop_arguments(problem_shape, tensor_A, tensor_B);
auto mainloop_args = params.get_mainloop_arguments(problem_shape, tensor_A, tensor_B);
auto epilogue_args = typename Conv::ConvKernel::EpilogueArguments {
{},
tensor_C.data().get(),
@@ -312,6 +312,7 @@ struct ConvTestbed {
};
auto args = typename Conv::Arguments {
problem_shape,
mainloop_args, // MainloopArguments
epilogue_args, // EpilogueArguments
hw_info,
@@ -615,8 +616,10 @@ bool TestAllConv(double alpha = 1.0, double beta = 0.0, float epsilon = 0.0f
#endif
for (DecompositionMode decomp_mode : decomposition_modes) {
std::vector problem_splits = {Splits{1}};
if (decomp_mode == DecompositionMode::Heuristic || decomp_mode == DecompositionMode::SplitK) {
problem_splits.push_back(Splits{2});
if constexpr (UsesStreamKScheduler) {
if (decomp_mode == DecompositionMode::Heuristic || decomp_mode == DecompositionMode::SplitK) {
problem_splits.push_back(Splits{2});
}
}
for (auto splits : problem_splits) {
test/unit/conv/device_3x/wgrad/sm90_conv1d_wgrad_implicit_gemm_f16_f16_f32_tensorop_f16.cu
+16
View File
@@ -84,7 +84,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -127,7 +129,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -170,7 +174,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -213,7 +219,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -261,7 +269,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -304,7 +314,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -347,7 +359,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -390,7 +404,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f16, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/wgrad/sm90_conv1d_wgrad_implicit_gemm_f16_f16_f32_tensorop_f32.cu
+16
View File
@@ -84,7 +84,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -127,7 +129,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -170,7 +174,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -213,7 +219,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 6
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -260,7 +268,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -303,7 +313,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -346,7 +358,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -389,7 +403,9 @@ TEST(SM90_device_conv1d_wgrad_implicitgemm_f16nwc_f16nwc_f32nwc_tensor_op_f32, 1
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/wgrad/sm90_conv2d_wgrad_implicit_gemm_f16_f16_f32_tensorop_f16.cu
+17
View File
@@ -85,7 +85,10 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -129,7 +132,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +178,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -217,7 +224,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -265,7 +274,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -309,7 +320,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -353,7 +366,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -397,7 +412,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f16
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/wgrad/sm90_conv2d_wgrad_implicit_gemm_f16_f16_f32_tensorop_f32.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -129,7 +131,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -217,7 +223,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -265,7 +273,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -309,7 +319,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -353,7 +365,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -397,7 +411,9 @@ TEST(SM90_device_conv2d_wgrad_implicitgemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/wgrad/sm90_conv3d_wgrad_implicit_gemm_f16_f16_f32_tensorop_f16.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -129,7 +131,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -218,7 +224,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -266,7 +274,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -310,7 +320,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -354,7 +366,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -398,7 +412,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
test/unit/conv/device_3x/wgrad/sm90_conv3d_wgrad_implicit_gemm_f16_f16_f32_tensorop_f32.cu
+16
View File
@@ -85,7 +85,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -129,7 +131,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -173,7 +177,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -217,7 +223,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -265,7 +273,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -309,7 +319,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -353,7 +365,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;
@@ -397,7 +411,9 @@ TEST(SM90_device_conv3d_wgrad_implicitgemm_f16ndhwc_f16ndhwc_f32ndhwc_tensor_op_
cutlass::conv::collective::KernelScheduleAuto
>::CollectiveOp;
using ProblemShape=cutlass::conv::ConvProblemShape<CollectiveMainloop::DispatchPolicy::ConvOp, CollectiveMainloop::DispatchPolicy::NumSpatialDimensions>;
using ConvKernel = cutlass::conv::kernel::ConvUniversal<
ProblemShape,
CollectiveMainloop,
CollectiveEpilogue
>;