mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-14 02:02:46 +00:00
Unified conv3D API + support for all data types. (#133)
* Convolution ND
* Code unification across dimensions for generating tensor descriptors.
* Example
* Instances
* Move convnd f32 instance file to comply with repo structure.
* Conv 1D tensor layouts.
* Formatting and use ReferenceConv
* Reference ConvFwd supporting 1D and 2D convolution.
* Debug printing TensorLayout name.
* Conv fwd 1D instance f32
* Refactor conv ND example.
Needed to support various conv dimensio.
Needed to support various conv dimensions
* Rename conv nd example director to prevent conflicts.
* Refactor some common utility to single file.
Plus some tests.
* Refactor GetHostTensorDescriptor + UT.
* Add 1D test case.
* Test reference convolution 1d/2d
* Remove some leftovers.
* Fix convolution example error for 1D
* Refactor test check errors utility function.
* Test Conv2D Fwd XDL
* More UT for 1D case.
* Parameterize input & weight initializers.
* Rename example to prevent conflicts.
* Split convnd instance into separate files for 1d/2d
* Address review comments.
* Fix data type for flops/gbytes calculations.
* Assign example number 11.
* 3D cases for convolution utility functions.
* 3D reference convolution.
* Add support for 3D convolution.
* Check for inputs bigger than 2GB.
* Formatting
* Support for bf16/f16/f32/i8 - conv instances + UT.
* Use check_err from test_util.hpp.
* Split convnd test into separate files for each dim.
* Fix data generation and use proper instances.
* Formatting
* Skip tensor initialization if not necessary.
* Fix CMakefiles.
* Remove redundant conv2d_fwd test.
* Lower problem size for conv3D UT.
* 3D case for convnd example.
* Remove leftovers after merge.
* Add Conv Specialization string to GetTypeString
* Skip instance causing numerical errors.
* Small fixes.
* Remove redundant includes.
* Fix namespace name error.
* Script for automatic testing and logging convolution fwd UTs
* Comment out numactl cmd.
* Refine weights initalization and relax rtol for fp16
* Fix weights initialization for int8.
* Add type_convert when store output in ref conv 1D.
* Get back old conv2d_fwd_xdl operation.
* Silence conv debug print.
* format
* clean
* clean
* Fix merge.
* Fix namespace for check_err
Co-authored-by: Adam Osewski <aosewski@amd.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>
[ROCm/composable_kernel commit: f91579aab6]
This commit is contained in:
@@ -84,6 +84,9 @@ DeviceConvFwdBasePtr GetConvInstance(int num_dim_spatial)
|
||||
{
|
||||
switch(num_dim_spatial)
|
||||
{
|
||||
case 3: {
|
||||
return std::make_unique<DeviceConvNDFwdInstance<3>>();
|
||||
}
|
||||
case 2: {
|
||||
return std::make_unique<DeviceConvNDFwdInstance<2>>();
|
||||
}
|
||||
@@ -173,6 +176,9 @@ HostTensorDescriptor GetOutputHostTensorDescriptor(const std::vector<std::size_t
|
||||
|
||||
switch(num_dim_spatial)
|
||||
{
|
||||
case 3: {
|
||||
return ck::conv_util::GetHostTensorDescriptor(dims, tl::NDHWK{});
|
||||
}
|
||||
case 2: {
|
||||
return ck::conv_util::GetHostTensorDescriptor(dims, tl::NHWK{});
|
||||
}
|
||||
@@ -192,6 +198,9 @@ HostTensorDescriptor GetFiltersHostTensorDescriptor(const std::vector<std::size_
|
||||
|
||||
switch(num_dim_spatial)
|
||||
{
|
||||
case 3: {
|
||||
return ck::conv_util::GetHostTensorDescriptor(dims, tl::KZYXC{});
|
||||
}
|
||||
case 2: {
|
||||
return ck::conv_util::GetHostTensorDescriptor(dims, tl::KYXC{});
|
||||
}
|
||||
@@ -211,6 +220,9 @@ HostTensorDescriptor GetInputHostTensorDescriptor(const std::vector<std::size_t>
|
||||
|
||||
switch(num_dim_spatial)
|
||||
{
|
||||
case 3: {
|
||||
return ck::conv_util::GetHostTensorDescriptor(dims, tl::NDHWC{});
|
||||
}
|
||||
case 2: {
|
||||
return ck::conv_util::GetHostTensorDescriptor(dims, tl::NHWC{});
|
||||
}
|
||||
@@ -360,6 +372,11 @@ int main(int argc, char* argv[])
|
||||
|
||||
switch(num_dim_spatial)
|
||||
{
|
||||
case 3: {
|
||||
auto ref_conv = ReferenceConvNDFwdInstance<3>();
|
||||
verify_f(ref_conv);
|
||||
break;
|
||||
}
|
||||
case 2: {
|
||||
auto ref_conv = ReferenceConvNDFwdInstance<2>();
|
||||
verify_f(ref_conv);
|
||||
|
||||
@@ -157,6 +157,12 @@
|
||||
#define CK_WORKAROUND_SWDEV_325164 1
|
||||
#endif
|
||||
|
||||
// workaround for verification failure ConvNd forward
|
||||
// https://github.com/ROCmSoftwarePlatform/composable_kernel/issues/135
|
||||
#ifndef CK_WORKAROUND_GITHUB_135
|
||||
#define CK_WORKAROUND_GITHUB_135 1
|
||||
#endif
|
||||
|
||||
namespace ck {
|
||||
|
||||
enum InMemoryDataOperationEnum_t
|
||||
|
||||
@@ -186,6 +186,28 @@ HostTensorDescriptor GetHostTensorDescriptor(const std::vector<std::size_t>& dim
|
||||
return HostTensorDescriptor(
|
||||
dims, std::vector<std::size_t>{C * dims[2] * dims[3], 1, dims[3] * C, C});
|
||||
}
|
||||
// 3D
|
||||
else if constexpr(std::is_same<TensorLayout, ck::tensor_layout::convolution::NCDHW>::value ||
|
||||
std::is_same<TensorLayout, ck::tensor_layout::convolution::KCZYX>::value ||
|
||||
std::is_same<TensorLayout, ck::tensor_layout::convolution::NKDHW>::value)
|
||||
{
|
||||
|
||||
return HostTensorDescriptor(dims,
|
||||
std::vector<std::size_t>{C * dims[2] * dims[3] * dims[4],
|
||||
dims[2] * dims[3] * dims[4],
|
||||
dims[3] * dims[4],
|
||||
dims[4],
|
||||
1});
|
||||
}
|
||||
else if constexpr(std::is_same<TensorLayout, ck::tensor_layout::convolution::NDHWC>::value ||
|
||||
std::is_same<TensorLayout, ck::tensor_layout::convolution::KZYXC>::value ||
|
||||
std::is_same<TensorLayout, ck::tensor_layout::convolution::NDHWK>::value)
|
||||
{
|
||||
return HostTensorDescriptor(
|
||||
dims,
|
||||
std::vector<std::size_t>{
|
||||
C * dims[2] * dims[3] * dims[4], 1, C * dims[3] * dims[4], C * dims[4], C});
|
||||
}
|
||||
|
||||
std::stringstream err_msg;
|
||||
err_msg << "Unsupported data layout provided: " << layout << "!";
|
||||
|
||||
@@ -1,6 +1,8 @@
|
||||
#ifndef CONVOLUTION_FORWARD_SPECIALIZATION
|
||||
#define CONVOLUTION_FORWARD_SPECIALIZATION
|
||||
|
||||
#include <string>
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
@@ -13,6 +15,18 @@ enum ConvolutionForwardSpecialization_t
|
||||
OddC,
|
||||
};
|
||||
|
||||
inline std::string getConvFwdSpecializationStr(const ConvolutionForwardSpecialization_t& s)
|
||||
{
|
||||
switch(s)
|
||||
{
|
||||
case Default: return "Default";
|
||||
case Filter1x1Pad0: return "Filter1x1Pad0";
|
||||
case Filter1x1Stride1Pad0: return "Filter1x1Stride1Pad0";
|
||||
case OddC: return "OddC";
|
||||
default: return "Unrecognized specialization!";
|
||||
}
|
||||
}
|
||||
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
|
||||
@@ -875,7 +875,8 @@ struct DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_W
|
||||
<< BlockSize << ", "
|
||||
<< MPerBlock << ", "
|
||||
<< NPerBlock << ", "
|
||||
<< K0PerBlock
|
||||
<< K0PerBlock << ", "
|
||||
<< getConvFwdSpecializationStr(ConvForwardSpecialization)
|
||||
<< ">";
|
||||
// clang-format on
|
||||
|
||||
|
||||
@@ -466,7 +466,6 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
|
||||
<< arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
|
||||
}
|
||||
#endif
|
||||
|
||||
if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
|
||||
arg.b_grid_desc_k0_n_k1_,
|
||||
arg.c_grid_desc_m_n_,
|
||||
@@ -708,7 +707,8 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
|
||||
<< BlockSize << ", "
|
||||
<< MPerBlock << ", "
|
||||
<< NPerBlock << ", "
|
||||
<< K0PerBlock
|
||||
<< K0PerBlock << ", "
|
||||
<< getConvFwdSpecializationStr(ConvForwardSpecialization)
|
||||
<< ">";
|
||||
// clang-format on
|
||||
|
||||
|
||||
@@ -367,6 +367,155 @@ struct DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
|
||||
}
|
||||
}
|
||||
|
||||
template <ck::index_t NDim, typename std::enable_if<NDim == 3, bool>::type = false>
|
||||
static auto GetInputTensorDescriptor(ck::index_t N,
|
||||
ck::index_t C,
|
||||
ck::index_t gemm_m,
|
||||
ck::index_t gemm_k,
|
||||
ck::index_t gemm_m_pad,
|
||||
const std::vector<ck::index_t>& input_spatial_lengths,
|
||||
const std::vector<ck::index_t>& filter_spatial_lengths,
|
||||
const std::vector<ck::index_t>& output_spatial_lengths,
|
||||
const std::vector<ck::index_t>& conv_filter_strides,
|
||||
const std::vector<ck::index_t>& conv_filter_dilations,
|
||||
const std::vector<ck::index_t>& input_left_pads,
|
||||
const std::vector<ck::index_t>& input_right_pads)
|
||||
{
|
||||
const ck::index_t gemm_k0 = gemm_k / GemmK1Number;
|
||||
const index_t Di = input_spatial_lengths[0];
|
||||
const index_t Hi = input_spatial_lengths[1];
|
||||
const index_t Wi = input_spatial_lengths[2];
|
||||
|
||||
const index_t Do = output_spatial_lengths[0];
|
||||
const index_t Ho = output_spatial_lengths[1];
|
||||
const index_t Wo = output_spatial_lengths[2];
|
||||
|
||||
const index_t ConvStrideD = conv_filter_strides[0];
|
||||
const index_t ConvStrideH = conv_filter_strides[1];
|
||||
const index_t ConvStrideW = conv_filter_strides[2];
|
||||
|
||||
if constexpr(ConvForwardSpecialization ==
|
||||
ConvolutionForwardSpecialization_t::Filter1x1Stride1Pad0)
|
||||
{
|
||||
const auto in_gemmmraw_gemmk_grid_desc =
|
||||
make_naive_tensor_descriptor_packed(make_tuple(gemm_m, gemm_k));
|
||||
|
||||
// in_gemmk0_gemmm_gemmk1_grid_desc
|
||||
return transform_tensor_descriptor(
|
||||
in_gemmmraw_gemmk_grid_desc,
|
||||
make_tuple(make_unmerge_transform(make_tuple(gemm_k0, GemmK1Number)),
|
||||
make_right_pad_transform(gemm_m, gemm_m_pad)),
|
||||
make_tuple(Sequence<1>{}, Sequence<0>{}),
|
||||
make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
|
||||
}
|
||||
else if constexpr(ConvForwardSpecialization ==
|
||||
ConvolutionForwardSpecialization_t::Filter1x1Pad0)
|
||||
{
|
||||
const auto in_n_di_hi_wi_c_grid_desc =
|
||||
make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
|
||||
|
||||
const auto in_n_do_ho_wo_c_grid_desc = transform_tensor_descriptor(
|
||||
in_n_di_hi_wi_c_grid_desc,
|
||||
make_tuple(make_pass_through_transform(N),
|
||||
make_embed_transform(make_tuple(Do), make_tuple(ConvStrideD)),
|
||||
make_embed_transform(make_tuple(Ho), make_tuple(ConvStrideH)),
|
||||
make_embed_transform(make_tuple(Wo), make_tuple(ConvStrideW)),
|
||||
make_pass_through_transform(C)),
|
||||
make_tuple(
|
||||
Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
|
||||
make_tuple(
|
||||
Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
|
||||
|
||||
const auto in_gemmk0_gemmmraw_gemmk1_grid_desc = transform_tensor_descriptor(
|
||||
in_n_do_ho_wo_c_grid_desc,
|
||||
make_tuple(make_unmerge_transform(make_tuple(gemm_k0, GemmK1Number)),
|
||||
make_merge_transform(make_tuple(N, Do, Ho, Wo))),
|
||||
make_tuple(Sequence<4>{}, Sequence<0, 1, 2, 3>{}),
|
||||
make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
|
||||
|
||||
// in_gemmk0_gemmm_gemmk1_grid_desc
|
||||
return transform_tensor_descriptor(
|
||||
in_gemmk0_gemmmraw_gemmk1_grid_desc,
|
||||
make_tuple(make_pass_through_transform(gemm_k0),
|
||||
make_right_pad_transform(gemm_m, gemm_m_pad),
|
||||
make_pass_through_transform(GemmK1Number)),
|
||||
make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
|
||||
make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
|
||||
}
|
||||
else
|
||||
{
|
||||
const index_t Z = filter_spatial_lengths[0];
|
||||
const index_t Y = filter_spatial_lengths[1];
|
||||
const index_t X = filter_spatial_lengths[2];
|
||||
|
||||
const index_t ConvDilationD = conv_filter_dilations[0];
|
||||
const index_t ConvDilationH = conv_filter_dilations[1];
|
||||
const index_t ConvDilationW = conv_filter_dilations[2];
|
||||
|
||||
const index_t InLeftPadD = input_left_pads[0];
|
||||
const index_t InLeftPadH = input_left_pads[1];
|
||||
const index_t InLeftPadW = input_left_pads[2];
|
||||
|
||||
const index_t InRightPadD = input_right_pads[0];
|
||||
const index_t InRightPadH = input_right_pads[1];
|
||||
const index_t InRightPadW = input_right_pads[2];
|
||||
|
||||
const auto in_n_di_hi_wi_c_grid_desc =
|
||||
make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
|
||||
|
||||
const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
|
||||
in_n_di_hi_wi_c_grid_desc,
|
||||
make_tuple(make_pass_through_transform(N),
|
||||
make_pad_transform(Di, InLeftPadD, InRightPadD),
|
||||
make_pad_transform(Hi, InLeftPadH, InRightPadH),
|
||||
make_pad_transform(Wi, InLeftPadW, InRightPadW),
|
||||
make_pass_through_transform(C)),
|
||||
make_tuple(
|
||||
Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
|
||||
make_tuple(
|
||||
Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
|
||||
|
||||
const auto in_n_z_do_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
|
||||
in_n_hip_wip_c_grid_desc,
|
||||
make_tuple(
|
||||
make_pass_through_transform(N),
|
||||
make_embed_transform(make_tuple(Z, Do), make_tuple(ConvDilationD, ConvStrideD)),
|
||||
make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
|
||||
make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
|
||||
make_pass_through_transform(C)),
|
||||
make_tuple(
|
||||
Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
|
||||
make_tuple(Sequence<0>{},
|
||||
Sequence<1, 2>{},
|
||||
Sequence<3, 4>{},
|
||||
Sequence<5, 6>{},
|
||||
Sequence<7>{}));
|
||||
|
||||
const auto in_gemmk_gemmmraw_grid_desc = transform_tensor_descriptor(
|
||||
in_n_z_do_y_ho_x_wo_c_grid_desc,
|
||||
make_tuple(make_merge_transform(make_tuple(Z, Y, X, C)),
|
||||
make_merge_transform(make_tuple(N, Do, Ho, Wo))),
|
||||
make_tuple(Sequence<1, 3, 5, 7>{}, Sequence<0, 2, 4, 6>{}),
|
||||
make_tuple(Sequence<0>{}, Sequence<1>{}));
|
||||
|
||||
const auto in_gemmk0_gemmmraw_gemmk1_grid_desc = transform_tensor_descriptor(
|
||||
in_gemmk_gemmmraw_grid_desc,
|
||||
make_tuple(make_unmerge_transform(make_tuple(gemm_k0, GemmK1Number)),
|
||||
make_pass_through_transform(gemm_m)),
|
||||
make_tuple(Sequence<0>{}, Sequence<1>{}),
|
||||
make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
|
||||
|
||||
// in_gemmk0_gemmm_gemmk1_grid_desc
|
||||
return transform_tensor_descriptor(
|
||||
in_gemmk0_gemmmraw_gemmk1_grid_desc,
|
||||
make_tuple(make_pass_through_transform(gemm_k0),
|
||||
make_right_pad_transform(gemm_m, gemm_m_pad),
|
||||
make_pass_through_transform(GemmK1Number)),
|
||||
make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
|
||||
make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
|
||||
}
|
||||
}
|
||||
|
||||
static index_t GetGemmMRaw(ck::index_t N,
|
||||
const std::vector<ck::index_t>& output_spatial_lengths)
|
||||
{
|
||||
@@ -445,6 +594,13 @@ struct DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
|
||||
1, 1, 1, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1});
|
||||
}
|
||||
|
||||
template <ck::index_t NDim, typename std::enable_if<NDim == 3, bool>::type = false>
|
||||
static auto GetABCGridDesc()
|
||||
{
|
||||
return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
|
||||
1, 1, 1, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1});
|
||||
}
|
||||
|
||||
using ABCGridDescs = decltype(GetABCGridDesc<NumDimSpatial>());
|
||||
|
||||
using AGridDesc_K0_M_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I0])>;
|
||||
@@ -593,6 +749,7 @@ struct DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
|
||||
|
||||
float Run(const Argument& arg, int nrepeat = 1)
|
||||
{
|
||||
#if 0
|
||||
{
|
||||
std::cout << "arg.a_grid_desc_k0_m_k1_{" << arg.a_grid_desc_k0_m_k1_.GetLength(I0)
|
||||
<< ", " << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
|
||||
@@ -605,7 +762,7 @@ struct DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
|
||||
std::cout << "arg.c_grid_desc_m_n_{ " << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
|
||||
<< arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
|
||||
}
|
||||
|
||||
#endif
|
||||
if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
|
||||
arg.b_grid_desc_k0_n_k1_,
|
||||
arg.c_grid_desc_m_n_,
|
||||
@@ -704,6 +861,22 @@ struct DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
|
||||
|
||||
static bool IsSupportedArgument(const Argument& arg)
|
||||
{
|
||||
// Input tensors can't be bigger than 2GB each.
|
||||
constexpr std::size_t GB2 = 2 * 1e9;
|
||||
|
||||
if(arg.a_grid_desc_k0_m_k1_.GetElementSpaceSize() > GB2)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
if(arg.b_grid_desc_k0_n_k1_.GetElementSpaceSize() > GB2)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
if(arg.c_grid_desc_m_n_.GetElementSpaceSize() > GB2)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
|
||||
if constexpr(ConvForwardSpecialization ==
|
||||
ConvolutionForwardSpecialization_t::Filter1x1Stride1Pad0)
|
||||
{
|
||||
@@ -851,7 +1024,8 @@ struct DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
|
||||
<< BlockSize << ", "
|
||||
<< MPerBlock << ", "
|
||||
<< NPerBlock << ", "
|
||||
<< K0PerBlock
|
||||
<< K0PerBlock << ", "
|
||||
<< getConvFwdSpecializationStr(ConvForwardSpecialization)
|
||||
<< ">";
|
||||
// clang-format on
|
||||
|
||||
|
||||
@@ -85,6 +85,7 @@ struct NKHW : public BaseTensorLayout
|
||||
static constexpr const char* name = "NKHW";
|
||||
};
|
||||
|
||||
// 3D Conv
|
||||
struct NDHWC : public BaseTensorLayout
|
||||
{
|
||||
static constexpr const char* name = "NDHWC";
|
||||
@@ -100,6 +101,21 @@ struct NDHWK : public BaseTensorLayout
|
||||
static constexpr const char* name = "NDHWK";
|
||||
};
|
||||
|
||||
struct NCDHW : public BaseTensorLayout
|
||||
{
|
||||
static constexpr const char* name = "NCDHW";
|
||||
};
|
||||
|
||||
struct KCZYX : public BaseTensorLayout
|
||||
{
|
||||
static constexpr const char* name = "KCZYX";
|
||||
};
|
||||
|
||||
struct NKDHW : public BaseTensorLayout
|
||||
{
|
||||
static constexpr const char* name = "NKDHW";
|
||||
};
|
||||
|
||||
} // namespace convolution
|
||||
|
||||
template <
|
||||
|
||||
@@ -14,9 +14,9 @@ namespace host {
|
||||
//
|
||||
// @brief Reference implementation for forward convolution.
|
||||
//
|
||||
// @paragraph Supported tensor layouts. Input tensor supports NCHiWi data layout.
|
||||
// Weights tensor supports KCYX data layout. Output tensor supports
|
||||
// NKHoWo data layout.
|
||||
// @paragraph Supports both NCHW as well as NHWC formats (and their respective
|
||||
// counterparts for weight and output) as long as tensor descriptor
|
||||
// lengths is in NCHW.
|
||||
//
|
||||
// @tparam InDataType Input tensor data type.
|
||||
// @tparam WeiDataType Weights tensor data type.
|
||||
@@ -100,9 +100,9 @@ struct ReferenceConvFwd : public device::BaseOperator
|
||||
float v_wei;
|
||||
|
||||
arg.in_element_op_(v_in,
|
||||
static_cast<const float>(arg.input_(n, c, wi)));
|
||||
ck::type_convert<float>(arg.input_(n, c, wi)));
|
||||
arg.wei_element_op_(v_wei,
|
||||
static_cast<const float>(arg.weight_(k, c, x)));
|
||||
ck::type_convert<float>(arg.weight_(k, c, x)));
|
||||
|
||||
v_acc += v_in * v_wei;
|
||||
}
|
||||
@@ -112,7 +112,7 @@ struct ReferenceConvFwd : public device::BaseOperator
|
||||
float v_out;
|
||||
|
||||
arg.out_element_op_(v_out, v_acc);
|
||||
arg.output_(n, k, wo) = v_out;
|
||||
arg.output_(n, k, wo) = ck::type_convert<OutDataType>(v_out);
|
||||
};
|
||||
|
||||
make_ParallelTensorFunctor(f_ncw,
|
||||
@@ -169,6 +169,61 @@ struct ReferenceConvFwd : public device::BaseOperator
|
||||
|
||||
return 0;
|
||||
}
|
||||
else if constexpr(NumDimSpatial == 3)
|
||||
{
|
||||
auto f_nchw = [&](auto n, auto k, auto d_o, auto ho, auto wo) {
|
||||
float v_acc = 0;
|
||||
|
||||
for(int c = 0; c < arg.weight_.mDesc.GetLengths()[1]; ++c)
|
||||
{
|
||||
for(int z = 0; z < arg.weight_.mDesc.GetLengths()[2]; ++z)
|
||||
{
|
||||
int di = d_o * arg.conv_strides_[0] + z * arg.conv_dilations_[0] -
|
||||
arg.in_left_pads_[0];
|
||||
for(int y = 0; y < arg.weight_.mDesc.GetLengths()[3]; ++y)
|
||||
{
|
||||
int hi = ho * arg.conv_strides_[1] + y * arg.conv_dilations_[1] -
|
||||
arg.in_left_pads_[1];
|
||||
for(int x = 0; x < arg.weight_.mDesc.GetLengths()[4]; ++x)
|
||||
{
|
||||
int wi = wo * arg.conv_strides_[2] +
|
||||
x * arg.conv_dilations_[2] - arg.in_left_pads_[2];
|
||||
if(di >= 0 && di < arg.input_.mDesc.GetLengths()[2] &&
|
||||
hi >= 0 && hi < arg.input_.mDesc.GetLengths()[3] &&
|
||||
wi >= 0 && wi < arg.input_.mDesc.GetLengths()[4])
|
||||
{
|
||||
float v_in;
|
||||
float v_wei;
|
||||
|
||||
arg.in_element_op_(
|
||||
v_in,
|
||||
ck::type_convert<float>(arg.input_(n, c, di, hi, wi)));
|
||||
arg.wei_element_op_(
|
||||
v_wei,
|
||||
ck::type_convert<float>(arg.weight_(k, c, z, y, x)));
|
||||
v_acc += v_in * v_wei;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
float v_out;
|
||||
|
||||
arg.out_element_op_(v_out, v_acc);
|
||||
arg.output_(n, k, d_o, ho, wo) = ck::type_convert<OutDataType>(v_out);
|
||||
};
|
||||
|
||||
make_ParallelTensorFunctor(f_nchw,
|
||||
arg.output_.mDesc.GetLengths()[0],
|
||||
arg.output_.mDesc.GetLengths()[1],
|
||||
arg.output_.mDesc.GetLengths()[2],
|
||||
arg.output_.mDesc.GetLengths()[3],
|
||||
arg.output_.mDesc.GetLengths()[4])(
|
||||
std::thread::hardware_concurrency());
|
||||
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
|
||||
float Run(const device::BaseArgument* p_arg, int) override
|
||||
|
||||
@@ -23,6 +23,7 @@ add_subdirectory(gemm_bias_relu_add)
|
||||
add_subdirectory(batched_gemm)
|
||||
add_subdirectory(conv1d_fwd)
|
||||
add_subdirectory(conv2d_fwd)
|
||||
add_subdirectory(conv3d_fwd)
|
||||
add_subdirectory(conv2d_fwd_bias_relu)
|
||||
add_subdirectory(conv2d_fwd_bias_relu_add)
|
||||
add_subdirectory(conv2d_fwd_bias_relu_atomic_add)
|
||||
|
||||
@@ -1,6 +1,9 @@
|
||||
# device_conv1d_fwd_instance
|
||||
set(DEVICE_CONV1D_FWD_INSTANCE_SOURCE
|
||||
device_conv1d_fwd_xdl_nwc_kxc_nwk_bf16_instance.cpp;
|
||||
device_conv1d_fwd_xdl_nwc_kxc_nwk_f16_instance.cpp;
|
||||
device_conv1d_fwd_xdl_nwc_kxc_nwk_f32_instance.cpp;
|
||||
device_conv1d_fwd_xdl_nwc_kxc_nwk_int8_instance.cpp;
|
||||
)
|
||||
|
||||
add_library(device_conv1d_fwd_instance SHARED ${DEVICE_CONV1D_FWD_INSTANCE_SOURCE})
|
||||
|
||||
@@ -0,0 +1,112 @@
|
||||
#include <stdlib.h>
|
||||
#include "config.hpp"
|
||||
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv1d_fwd_instance {
|
||||
|
||||
using F32 = float;
|
||||
using BF16 = bhalf_t;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
static constexpr auto ConvFwd1x1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Pad0;
|
||||
|
||||
static constexpr auto ConvFwd1x1S1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Stride1Pad0;
|
||||
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv1d_fwd_xdl_nwc_kxc_nwk_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
#if !CK_WORKAROUND_GITHUB_135
|
||||
// FIXME: this instance causes numerical errors.
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
#endif
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
void add_device_conv1d_fwd_xdl_nwc_kxc_nwk_bf16_instances(
|
||||
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
|
||||
{
|
||||
add_device_operation_instances(instances, device_conv1d_fwd_xdl_nwc_kxc_nwk_bf16_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_bf16_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_bf16_instances{});
|
||||
}
|
||||
|
||||
} // namespace device_conv1d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -0,0 +1,109 @@
|
||||
#include <stdlib.h>
|
||||
#include "config.hpp"
|
||||
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv1d_fwd_instance {
|
||||
|
||||
using F16 = ck::half_t;
|
||||
using F32 = float;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
static constexpr auto ConvFwd1x1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Pad0;
|
||||
|
||||
static constexpr auto ConvFwd1x1S1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Stride1Pad0;
|
||||
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv1d_fwd_xdl_nwc_kxc_nwk_f16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_f16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
void add_device_conv1d_fwd_xdl_nwc_kxc_nwk_f16_instances(
|
||||
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
|
||||
{
|
||||
add_device_operation_instances(instances, device_conv1d_fwd_xdl_nwc_kxc_nwk_f16_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_f16_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_f16_instances{});
|
||||
}
|
||||
|
||||
} // namespace device_conv1d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -0,0 +1,111 @@
|
||||
#include <stdlib.h>
|
||||
#include "config.hpp"
|
||||
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv1d_fwd_instance {
|
||||
|
||||
using F32 = float;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
static constexpr auto ConvFwd1x1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Pad0;
|
||||
|
||||
static constexpr auto ConvFwd1x1S1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Stride1Pad0;
|
||||
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv1d_fwd_xdl_nwc_kxc_nwk_int8_instances =
|
||||
std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 1, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_int8_instances =
|
||||
std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 1, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_int8_instances =
|
||||
std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 1, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
void add_device_conv1d_fwd_xdl_nwc_kxc_nwk_int8_instances(
|
||||
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
|
||||
{
|
||||
add_device_operation_instances(instances, device_conv1d_fwd_xdl_nwc_kxc_nwk_int8_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_p0_int8_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv1d_fwd_xdl_nwc_kxc_nwk_1x1_s1_p0_int8_instances{});
|
||||
}
|
||||
|
||||
} // namespace device_conv1d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -29,67 +29,67 @@ static constexpr auto ConvFwd1x1S1P0 =
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################|InData|WeiData|OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_p0_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################|InData|WeiData|OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################|InData|WeiData|OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
|
||||
@@ -1,6 +1,6 @@
|
||||
#include <stdlib.h>
|
||||
#include "config.hpp"
|
||||
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "device_operation_instance.hpp"
|
||||
|
||||
@@ -28,67 +28,67 @@ static constexpr auto ConvFwd1x1S1P0 =
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 2, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_p0_f32_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 2, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f32_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 2, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
|
||||
@@ -28,67 +28,67 @@ static constexpr auto ConvFwd1x1S1P0 =
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_p0_int8_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_int8_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
|
||||
@@ -0,0 +1,13 @@
|
||||
# device_conv3d_fwd_instance
|
||||
set(DEVICE_CONV3D_FWD_INSTANCE_SOURCE
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp;
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp;
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp;
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp;
|
||||
)
|
||||
add_library(device_conv3d_fwd_instance SHARED ${DEVICE_CONV3D_FWD_INSTANCE_SOURCE})
|
||||
target_compile_features(device_conv3d_fwd_instance PUBLIC)
|
||||
set_target_properties(device_conv3d_fwd_instance PROPERTIES POSITION_INDEPENDENT_CODE ON)
|
||||
install(TARGETS device_conv3d_fwd_instance LIBRARY DESTINATION lib)
|
||||
|
||||
clang_tidy_check(device_conv3d_fwd_instance)
|
||||
@@ -0,0 +1,113 @@
|
||||
#include <stdlib.h>
|
||||
#include "config.hpp"
|
||||
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv3d_fwd_instance {
|
||||
|
||||
using F32 = float;
|
||||
using BF16 = bhalf_t;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
static constexpr auto ConvFwd1x1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Pad0;
|
||||
|
||||
static constexpr auto ConvFwd1x1S1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Stride1Pad0;
|
||||
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
#if !CK_WORKAROUND_GITHUB_135
|
||||
// FIXME: this instance causes numerical errors.
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
#endif
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_p0_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_bf16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< BF16, BF16, BF16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
void add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
|
||||
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
|
||||
{
|
||||
add_device_operation_instances(instances,
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_bf16_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_p0_bf16_instances{});
|
||||
add_device_operation_instances(
|
||||
instances, device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_bf16_instances{});
|
||||
}
|
||||
|
||||
} // namespace device_conv3d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -0,0 +1,110 @@
|
||||
#include <stdlib.h>
|
||||
#include "config.hpp"
|
||||
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv3d_fwd_instance {
|
||||
|
||||
using F16 = ck::half_t;
|
||||
using F32 = float;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
static constexpr auto ConvFwd1x1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Pad0;
|
||||
|
||||
static constexpr auto ConvFwd1x1S1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Stride1Pad0;
|
||||
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_p0_f16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f16_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F16, F16, F16, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
void add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
|
||||
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
|
||||
{
|
||||
add_device_operation_instances(instances,
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f16_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_p0_f16_instances{});
|
||||
add_device_operation_instances(
|
||||
instances, device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f16_instances{});
|
||||
}
|
||||
|
||||
} // namespace device_conv3d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -0,0 +1,109 @@
|
||||
#include <stdlib.h>
|
||||
#include "config.hpp"
|
||||
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv3d_fwd_instance {
|
||||
|
||||
using F32 = float;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
static constexpr auto ConvFwd1x1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Pad0;
|
||||
|
||||
static constexpr auto ConvFwd1x1S1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Stride1Pad0;
|
||||
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f32_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_p0_f32_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f32_instances = std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 256, 128, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 256, 4, 4, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 128, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 128, 4, 4, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 64, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 64, 128, 4, 4, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 64, 64, 4, 4, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 64, 4, 4, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 64, 128, 4, 4, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 32, 4, 4, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 32, 128, 4, 4, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 64, 32, 4, 4, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< F32, F32, F32, F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 32, 64, 4, 4, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
void add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
|
||||
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
|
||||
{
|
||||
add_device_operation_instances(instances,
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f32_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_p0_f32_instances{});
|
||||
add_device_operation_instances(
|
||||
instances, device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_f32_instances{});
|
||||
}
|
||||
|
||||
} // namespace device_conv3d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -0,0 +1,112 @@
|
||||
#include <stdlib.h>
|
||||
#include "config.hpp"
|
||||
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "device_operation_instance.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv3d_fwd_instance {
|
||||
|
||||
using F32 = float;
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
static constexpr auto ConvFwd1x1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Pad0;
|
||||
|
||||
static constexpr auto ConvFwd1x1S1P0 =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Filter1x1Stride1Pad0;
|
||||
|
||||
// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_int8_instances =
|
||||
std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, 3, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_p0_int8_instances =
|
||||
std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1P0, 3, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
using device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_int8_instances =
|
||||
std::tuple<
|
||||
// clang-format off
|
||||
//################################################################| InData| WeiData| OutData| AccData| In| Wei| Out| ConvForward| NumDim| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
|
||||
//################################################################| Type| Type| Type| Type| Elementwise| Elementwise| Elementwise| Specialization|Spatial| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
|
||||
//################################################################| | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
|
||||
//################################################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 256, 128, 4, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 256, 4, 16, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 128, 4, 16, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 128, 4, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 64, 128, 4, 16, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 64, 64, 4, 16, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 128, 64, 4, 16, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 256, 64, 128, 4, 16, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 128, 32, 4, 16, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 128, 32, 128, 4, 16, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 64, 32, 4, 16, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>,
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< int8_t, int8_t, int8_t, int32_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, 3, 64, 32, 64, 4, 16, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, true, 7, 1>
|
||||
// clang-format on
|
||||
>;
|
||||
|
||||
void add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_int8_instances(
|
||||
std::vector<DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>>& instances)
|
||||
{
|
||||
add_device_operation_instances(instances,
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_int8_instances{});
|
||||
add_device_operation_instances(instances,
|
||||
device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_p0_int8_instances{});
|
||||
add_device_operation_instances(
|
||||
instances, device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_1x1_s1_p0_int8_instances{});
|
||||
}
|
||||
|
||||
} // namespace device_conv3d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
@@ -67,8 +67,8 @@ Best Perf: 1.1933 ms, 107.977 TFlops, 79.0848 GB/s
|
||||
#arg8: print matrix value (0=no, 1=yes)
|
||||
#arg9: run kernel # of times (>1)
|
||||
#arg10 to 24: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx
|
||||
##################### op datatype in_layout wei_layout out_layout verify init log repeat N__ K___ C___ Y X Hi__ Wi__ Strides Dilations LeftPads RightPads
|
||||
./profiler/ckProfiler conv 1 1 1 1 1 1 0 5 128 256 192 3 3 71 71 2 2 1 1 1 1 1 1
|
||||
##################### op datatype in_layout wei_layout out_layout verify init log repeat N__ K___ C___ Y X Hi__ Wi__ Strides Dilations LeftPads RightPads
|
||||
./profiler/ckProfiler conv_fwd 1 1 1 1 1 1 0 5 128 256 192 3 3 71 71 2 2 1 1 1 1 1 1
|
||||
```
|
||||
|
||||
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
|
||||
|
||||
110
script/test_convnd_fwd.sh
Normal file
110
script/test_convnd_fwd.sh
Normal file
@@ -0,0 +1,110 @@
|
||||
#!/usr/bin/env bash
|
||||
|
||||
# set -e
|
||||
|
||||
DIM1=False
|
||||
DIM2=True
|
||||
DIM3=False
|
||||
DATE=220317
|
||||
GIT_HASH=4e6dfda
|
||||
LOG_DIR=${DATE}_${GIT_HASH}
|
||||
SUFFIX=${GIT_HASH}
|
||||
|
||||
|
||||
#--------------------------------------------------------------------------
|
||||
# Commandline arguments parsing
|
||||
# like: cmd -key[--key] value
|
||||
#--------------------------------------------------------------------------
|
||||
|
||||
POSITIONAL=()
|
||||
while [[ $# -gt 0 ]]
|
||||
do
|
||||
key="$1"
|
||||
|
||||
case $key in
|
||||
-d1|--d1)
|
||||
DIM1=True
|
||||
echo DIM1: "${DIM1}"
|
||||
shift # past argument
|
||||
;;
|
||||
-d2|--d2)
|
||||
DIM2=True
|
||||
echo DIM2: "${DIM2}"
|
||||
shift # past argument
|
||||
;;
|
||||
-d3|--d3)
|
||||
DIM3=True
|
||||
echo DIM3: "${DIM3}"
|
||||
shift # past argument
|
||||
;;
|
||||
-all|--all)
|
||||
DIM1=True
|
||||
DIM2=True
|
||||
DIM3=True
|
||||
echo DIM1: "${DIM1}"
|
||||
echo DIM2: "${DIM2}"
|
||||
echo DIM3: "${DIM3}"
|
||||
shift # past argument
|
||||
;;
|
||||
-s|--suffix)
|
||||
SUFFIX=${SUFFIX}_"$2"
|
||||
echo SUFFIX: "${SUFFIX}"
|
||||
shift # past argument
|
||||
shift # past value
|
||||
;;
|
||||
*) # unknown option
|
||||
POSITIONAL+=("$1") # save it in an array for later
|
||||
shift # past argument
|
||||
;;
|
||||
esac
|
||||
done
|
||||
set -- "${POSITIONAL[@]}" # restore positional parameters
|
||||
|
||||
#--------------------------------------------------------------------------
|
||||
|
||||
# NUMACTL="numactl --cpunodebind=1 --membind=1"
|
||||
NUMACTL=
|
||||
# ENV_CONF=
|
||||
GPU=mi100
|
||||
PROF_ITER_COUNT=10000
|
||||
LOG_DIR_PATH=../log/${LOG_DIR}
|
||||
set -x
|
||||
|
||||
#-------------------------------------------------------------------------------
|
||||
# 1D
|
||||
#-------------------------------------------------------------------------------
|
||||
|
||||
if [[ "${DIM1}" == "True" ]]; then
|
||||
mkdir -p ${LOG_DIR_PATH}
|
||||
echo ">>>>>>>> RUN test conv1d nwc <<<<<<<<<<"
|
||||
CMD="./../build/bin/test_conv1d_fwd"
|
||||
${NUMACTL} ${CMD} 2>&1 \
|
||||
| tee ${LOG_DIR_PATH}/test_conv1d_fwd_nwc_${SUFFIX}_${GPU}.log
|
||||
|
||||
fi
|
||||
|
||||
#-------------------------------------------------------------------------------
|
||||
# 2D
|
||||
#-------------------------------------------------------------------------------
|
||||
|
||||
if [[ "${DIM2}" == "True" ]]; then
|
||||
mkdir -p ${LOG_DIR_PATH}
|
||||
echo ">>>>>>>> RUN test conv2d nhwc <<<<<<<<<<"
|
||||
CMD="./../build/bin/test_conv2d_fwd"
|
||||
${NUMACTL} ${CMD} 2>&1 \
|
||||
| tee ${LOG_DIR_PATH}/test_conv2d_fwd_nhwc_${SUFFIX}_${GPU}.log
|
||||
|
||||
fi
|
||||
|
||||
#-------------------------------------------------------------------------------
|
||||
# 3D
|
||||
#-------------------------------------------------------------------------------
|
||||
|
||||
if [[ "${DIM3}" == "True" ]]; then
|
||||
mkdir -p ${LOG_DIR_PATH}
|
||||
echo ">>>>>>>> RUN test conv3d ndhwc <<<<<<<<<<"
|
||||
CMD="./../build/bin/test_conv3d_fwd"
|
||||
${NUMACTL} ${CMD} 2>&1 \
|
||||
| tee ${LOG_DIR_PATH}/test_conv3d_fwd_ndhwc_${SUFFIX}_${GPU}.log
|
||||
|
||||
fi
|
||||
@@ -37,7 +37,6 @@ add_subdirectory(reference_conv_fwd)
|
||||
add_subdirectory(gemm)
|
||||
add_subdirectory(grouped_gemm)
|
||||
add_subdirectory(gemm_split_k)
|
||||
add_subdirectory(conv2d_fwd)
|
||||
add_subdirectory(convnd_fwd)
|
||||
add_subdirectory(conv2d_bwd_data)
|
||||
add_subdirectory(batched_gemm)
|
||||
|
||||
@@ -109,7 +109,7 @@ bool TestBatchedGemm(const std::size_t batch_count, DeviceBatchedGemmPtr& gemmPt
|
||||
gemmPtr, params, a, b, c_device, a_element_op, b_element_op, c_element_op);
|
||||
|
||||
// Assert
|
||||
// bool res = test_util::check_err(
|
||||
// bool res = test::check_err(
|
||||
// c_device.mData, c_host.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
|
||||
bool res = check_error(c_device, c_host) < 0.007815f;
|
||||
|
||||
|
||||
@@ -1,3 +0,0 @@
|
||||
add_test_executable(test_conv2d_fwd conv2d_fwd.cpp)
|
||||
target_link_libraries(test_conv2d_fwd PRIVATE host_tensor)
|
||||
target_link_libraries(test_conv2d_fwd PRIVATE device_conv2d_fwd_instance)
|
||||
@@ -1,308 +0,0 @@
|
||||
#include "config.hpp"
|
||||
#include "device.hpp"
|
||||
#include "host_tensor.hpp"
|
||||
#include "host_tensor_generator.hpp"
|
||||
#include "host_conv.hpp"
|
||||
#include "tensor_layout.hpp"
|
||||
#include "device_tensor.hpp"
|
||||
#include "device_conv_fwd.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "reference_conv_fwd.hpp"
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv2d_fwd_instance {
|
||||
|
||||
using DeviceConvFwdNoOpPtr = DeviceConvFwdPtr<ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough>;
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
void add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(
|
||||
std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
} // namespace device_conv2d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
|
||||
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
template <typename T>
|
||||
static bool check_out(const Tensor<T>& ref, const Tensor<T>& result)
|
||||
{
|
||||
float max_diff = 1e-6;
|
||||
|
||||
for(int i = 0; i < ref.mData.size(); ++i)
|
||||
{
|
||||
float diff = std::abs(double(ref.mData[i]) - double(result.mData[i]));
|
||||
if(max_diff < diff)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
int main(int argc, char* argv[])
|
||||
{
|
||||
int data_type = 0;
|
||||
int init_method = 0;
|
||||
|
||||
// Conv shape
|
||||
ck::index_t N = 128;
|
||||
ck::index_t K = 256;
|
||||
ck::index_t C = 192;
|
||||
ck::index_t Y = 3;
|
||||
ck::index_t X = 3;
|
||||
ck::index_t Hi = 71;
|
||||
ck::index_t Wi = 71;
|
||||
ck::index_t conv_stride_h = 2;
|
||||
ck::index_t conv_stride_w = 2;
|
||||
ck::index_t conv_dilation_h = 1;
|
||||
ck::index_t conv_dilation_w = 1;
|
||||
ck::index_t in_left_pad_h = 1;
|
||||
ck::index_t in_left_pad_w = 1;
|
||||
ck::index_t in_right_pad_h = 1;
|
||||
ck::index_t in_right_pad_w = 1;
|
||||
if(argc == 1)
|
||||
{
|
||||
data_type = 1;
|
||||
init_method = 1;
|
||||
}
|
||||
else if(argc == 3)
|
||||
{
|
||||
data_type = std::stoi(argv[1]);
|
||||
init_method = std::stoi(argv[2]);
|
||||
}
|
||||
else if(argc == 18)
|
||||
{
|
||||
data_type = std::stoi(argv[1]);
|
||||
init_method = std::stoi(argv[2]);
|
||||
|
||||
N = std::stoi(argv[3]);
|
||||
K = std::stoi(argv[4]);
|
||||
C = std::stoi(argv[5]);
|
||||
Y = std::stoi(argv[6]);
|
||||
X = std::stoi(argv[7]);
|
||||
Hi = std::stoi(argv[8]);
|
||||
Wi = std::stoi(argv[9]);
|
||||
conv_stride_h = std::stoi(argv[10]);
|
||||
conv_stride_w = std::stoi(argv[11]);
|
||||
conv_dilation_h = std::stoi(argv[12]);
|
||||
conv_dilation_w = std::stoi(argv[13]);
|
||||
in_left_pad_h = std::stoi(argv[14]);
|
||||
in_left_pad_w = std::stoi(argv[15]);
|
||||
in_right_pad_h = std::stoi(argv[16]);
|
||||
in_right_pad_w = std::stoi(argv[17]);
|
||||
}
|
||||
else
|
||||
{
|
||||
printf("arg1: data type (0=fp32, 1=fp16, 2= bfp16, 3= int8_t )\n");
|
||||
printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
|
||||
printf("arg3 to 17: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
|
||||
"RightPx\n");
|
||||
exit(1);
|
||||
}
|
||||
|
||||
auto Run = [&](auto input_type, auto wei_type, auto out_type) {
|
||||
using InDataType = decltype(input_type);
|
||||
using WeiDataType = decltype(wei_type);
|
||||
using OutDataType = decltype(out_type);
|
||||
|
||||
using ReferenceConvFwdInstance = ck::tensor_operation::host::ReferenceConvFwd<InDataType,
|
||||
WeiDataType,
|
||||
OutDataType,
|
||||
InElementOp,
|
||||
WeiElementOp,
|
||||
OutElementOp>;
|
||||
|
||||
const ck::index_t YEff = (Y - 1) * conv_dilation_h + 1;
|
||||
const ck::index_t XEff = (X - 1) * conv_dilation_w + 1;
|
||||
|
||||
const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - YEff) / conv_stride_h + 1;
|
||||
const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
|
||||
|
||||
const std::vector<ck::index_t> input_spatial_lengths{Hi, Wi};
|
||||
const std::vector<ck::index_t> filter_spatial_lengths{Y, X};
|
||||
const std::vector<ck::index_t> output_spatial_lengths{Ho, Wo};
|
||||
const std::vector<ck::index_t> conv_filter_strides{conv_stride_h, conv_stride_w};
|
||||
const std::vector<ck::index_t> conv_filter_dilations{conv_dilation_h, conv_dilation_w};
|
||||
const std::vector<ck::index_t> input_left_pads{in_left_pad_h, in_left_pad_w};
|
||||
const std::vector<ck::index_t> input_right_pads{in_right_pad_h, in_right_pad_w};
|
||||
|
||||
auto f_host_tensor_descriptor =
|
||||
[](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W) {
|
||||
return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
|
||||
std::vector<std::size_t>({C_ * H * W, 1, W * C_, C_}));
|
||||
};
|
||||
|
||||
Tensor<InDataType> in_n_c_hi_wi(f_host_tensor_descriptor(N, C, Hi, Wi));
|
||||
Tensor<WeiDataType> wei_k_c_y_x(f_host_tensor_descriptor(K, C, Y, X));
|
||||
Tensor<OutDataType> out_n_k_ho_wo_host_result(f_host_tensor_descriptor(N, K, Ho, Wo));
|
||||
Tensor<OutDataType> out_n_k_ho_wo_device_result(f_host_tensor_descriptor(N, K, Ho, Wo));
|
||||
|
||||
std::cout << "in_n_c_hi_wi: " << in_n_c_hi_wi.mDesc << std::endl;
|
||||
std::cout << "wei_k_c_y_x: " << wei_k_c_y_x.mDesc << std::endl;
|
||||
std::cout << "out_n_k_ho_wo: " << out_n_k_ho_wo_host_result.mDesc << std::endl;
|
||||
|
||||
switch(init_method)
|
||||
{
|
||||
case 0: break;
|
||||
case 1:
|
||||
in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5});
|
||||
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
|
||||
break;
|
||||
default:
|
||||
in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{0, 1});
|
||||
wei_k_c_y_x.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-1, 1});
|
||||
}
|
||||
|
||||
DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_hi_wi.mDesc.GetElementSpace());
|
||||
DeviceMem wei_device_buf(sizeof(WeiDataType) * wei_k_c_y_x.mDesc.GetElementSpace());
|
||||
DeviceMem out_device_buf(sizeof(OutDataType) *
|
||||
out_n_k_ho_wo_device_result.mDesc.GetElementSpace());
|
||||
|
||||
in_device_buf.ToDevice(in_n_c_hi_wi.mData.data());
|
||||
wei_device_buf.ToDevice(wei_k_c_y_x.mData.data());
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
using DeviceConvFwdNoOpPtr =
|
||||
ck::tensor_operation::device::DeviceConvFwdPtr<PassThrough, PassThrough, PassThrough>;
|
||||
|
||||
// add device Conv instances
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
|
||||
if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, float> &&
|
||||
ck::is_same_v<ck::remove_cv_t<WeiDataType>, float> &&
|
||||
ck::is_same_v<ck::remove_cv_t<OutDataType>, float>)
|
||||
{
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(conv_ptrs);
|
||||
}
|
||||
else if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, ck::half_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<WeiDataType>, ck::half_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<OutDataType>, ck::half_t>)
|
||||
{
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
|
||||
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
|
||||
}
|
||||
else if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, ck::bhalf_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<WeiDataType>, ck::bhalf_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<OutDataType>, ck::bhalf_t>)
|
||||
{
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(conv_ptrs);
|
||||
}
|
||||
else if constexpr(ck::is_same_v<ck::remove_cv_t<InDataType>, int8_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<WeiDataType>, int8_t> &&
|
||||
ck::is_same_v<ck::remove_cv_t<OutDataType>, int8_t>)
|
||||
{
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(conv_ptrs);
|
||||
}
|
||||
|
||||
if(conv_ptrs.size() <= 0)
|
||||
{
|
||||
throw std::runtime_error("wrong! no device Conv instance found");
|
||||
}
|
||||
|
||||
auto ref_conv = ReferenceConvFwdInstance{};
|
||||
auto ref_invoker = ref_conv.MakeInvoker();
|
||||
|
||||
auto ref_argument = ref_conv.MakeArgument(in_n_c_hi_wi,
|
||||
wei_k_c_y_x,
|
||||
out_n_k_ho_wo_host_result,
|
||||
conv_filter_strides,
|
||||
conv_filter_dilations,
|
||||
input_left_pads,
|
||||
input_right_pads,
|
||||
InElementOp{},
|
||||
WeiElementOp{},
|
||||
OutElementOp{});
|
||||
|
||||
ref_invoker.Run(ref_argument);
|
||||
|
||||
// profile device Conv instances
|
||||
bool success = false;
|
||||
for(auto& conv_ptr : conv_ptrs)
|
||||
{
|
||||
auto argument_ptr = conv_ptr->MakeArgumentPointer(
|
||||
static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
|
||||
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
|
||||
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
|
||||
N,
|
||||
K,
|
||||
C,
|
||||
input_spatial_lengths,
|
||||
filter_spatial_lengths,
|
||||
output_spatial_lengths,
|
||||
conv_filter_strides,
|
||||
conv_filter_dilations,
|
||||
input_left_pads,
|
||||
input_right_pads,
|
||||
PassThrough{},
|
||||
PassThrough{},
|
||||
PassThrough{});
|
||||
|
||||
auto invoker_ptr = conv_ptr->MakeInvokerPointer();
|
||||
|
||||
if(conv_ptr->IsSupportedArgument(argument_ptr.get()))
|
||||
{
|
||||
invoker_ptr->Run(argument_ptr.get(), 0);
|
||||
|
||||
out_device_buf.FromDevice(out_n_k_ho_wo_device_result.mData.data());
|
||||
if(!check_out(out_n_k_ho_wo_host_result, out_n_k_ho_wo_device_result))
|
||||
{
|
||||
success = false;
|
||||
break;
|
||||
}
|
||||
success = true;
|
||||
}
|
||||
}
|
||||
|
||||
if(success)
|
||||
{
|
||||
std::cout << "test conv2d fwd : Pass" << std::endl;
|
||||
return 0;
|
||||
}
|
||||
else
|
||||
{
|
||||
std::cout << "test conv2d fwd: Fail " << std::endl;
|
||||
return -1;
|
||||
}
|
||||
};
|
||||
int res = -1;
|
||||
if(data_type == 0)
|
||||
{
|
||||
res = Run(float(), float(), float());
|
||||
}
|
||||
else if(data_type == 1)
|
||||
{
|
||||
res = Run(ck::half_t(), ck::half_t(), ck::half_t());
|
||||
}
|
||||
else if(data_type == 2)
|
||||
{
|
||||
Run(ck::bhalf_t(), ck::bhalf_t(), ck::bhalf_t());
|
||||
}
|
||||
else if(data_type == 3)
|
||||
{
|
||||
res = Run(int8_t(), int8_t(), int8_t());
|
||||
}
|
||||
|
||||
return res;
|
||||
}
|
||||
@@ -5,33 +5,10 @@
|
||||
#include "config.hpp"
|
||||
#include "conv_utils.hpp"
|
||||
#include "tensor_layout.hpp"
|
||||
#include "test_util.hpp"
|
||||
|
||||
namespace {
|
||||
|
||||
template <typename T>
|
||||
bool cmp_vec(const std::vector<T>& out, const std::vector<T>& ref, const std::string& msg)
|
||||
{
|
||||
if(out.size() != ref.size())
|
||||
{
|
||||
std::cout << "out.size() != ref.size(), :" << out.size() << " != " << ref.size()
|
||||
<< std::endl
|
||||
<< msg << std::endl;
|
||||
return false;
|
||||
}
|
||||
|
||||
for(std::size_t i = 0; i < ref.size(); ++i)
|
||||
{
|
||||
if(out[i] != ref[i])
|
||||
{
|
||||
std::cout << "out[" << i << "] != ref[" << i << "]: " << out[i] << "!=" << ref[i]
|
||||
<< std::endl
|
||||
<< msg << std::endl;
|
||||
return false;
|
||||
}
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
bool TestConvParams_GetOutputSpatialLengths()
|
||||
{
|
||||
bool res{true};
|
||||
@@ -43,26 +20,26 @@ bool TestConvParams_GetOutputSpatialLengths()
|
||||
// padding {{1,1}, {1,1}}
|
||||
ck::conv_util::ConvParams conv_params;
|
||||
std::vector<ck::index_t> out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = cmp_vec(out_spatial_len,
|
||||
std::vector<ck::index_t>{36, 36},
|
||||
"Error: ConvParams 2D default constructor.");
|
||||
res = test::check_err(out_spatial_len,
|
||||
std::vector<ck::index_t>{36, 36},
|
||||
"Error: ConvParams 2D default constructor.");
|
||||
|
||||
conv_params.conv_filter_strides = std::vector<ck::index_t>{1, 1};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = cmp_vec(
|
||||
res = test::check_err(
|
||||
out_spatial_len, std::vector<ck::index_t>{71, 71}, "Error: ConvParams 2D stride {1,1}.");
|
||||
|
||||
conv_params.conv_filter_strides = std::vector<ck::index_t>{2, 2};
|
||||
conv_params.input_left_pads = std::vector<ck::index_t>{2, 2};
|
||||
conv_params.input_right_pads = std::vector<ck::index_t>{2, 2};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = cmp_vec(out_spatial_len,
|
||||
std::vector<ck::index_t>{37, 37},
|
||||
"Error: ConvParams 2D padding left/right {2,2}.");
|
||||
res = test::check_err(out_spatial_len,
|
||||
std::vector<ck::index_t>{37, 37},
|
||||
"Error: ConvParams 2D padding left/right {2,2}.");
|
||||
|
||||
conv_params.conv_filter_dilations = std::vector<ck::index_t>{2, 2};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = cmp_vec(
|
||||
res = test::check_err(
|
||||
out_spatial_len, std::vector<ck::index_t>{36, 36}, "Error: ConvParams 2D dilation {2,2}.");
|
||||
|
||||
conv_params.conv_filter_strides = std::vector<ck::index_t>{3, 3};
|
||||
@@ -70,9 +47,9 @@ bool TestConvParams_GetOutputSpatialLengths()
|
||||
conv_params.input_right_pads = std::vector<ck::index_t>{1, 1};
|
||||
conv_params.conv_filter_dilations = std::vector<ck::index_t>{2, 2};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = cmp_vec(out_spatial_len,
|
||||
std::vector<ck::index_t>{23, 23},
|
||||
"Error: ConvParams 2D strides{3,3}, padding {1,1}, dilations {2,2}.");
|
||||
res = test::check_err(out_spatial_len,
|
||||
std::vector<ck::index_t>{23, 23},
|
||||
"Error: ConvParams 2D strides{3,3}, padding {1,1}, dilations {2,2}.");
|
||||
|
||||
// -------------------------- 1D ------------------------------------
|
||||
conv_params.num_dim_spatial = 1;
|
||||
@@ -84,25 +61,24 @@ bool TestConvParams_GetOutputSpatialLengths()
|
||||
conv_params.input_right_pads = std::vector<ck::index_t>{1};
|
||||
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = cmp_vec(
|
||||
out_spatial_len, std::vector<ck::index_t>{36}, "Error: ConvParams 1D default constructor.");
|
||||
res = test::check_err(out_spatial_len, std::vector<ck::index_t>{36}, "Error: ConvParams 1D.");
|
||||
|
||||
conv_params.conv_filter_strides = std::vector<ck::index_t>{1, 1};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res =
|
||||
cmp_vec(out_spatial_len, std::vector<ck::index_t>{71}, "Error: ConvParams 1D stride {1}.");
|
||||
res = test::check_err(
|
||||
out_spatial_len, std::vector<ck::index_t>{71}, "Error: ConvParams 1D stride {1}.");
|
||||
|
||||
conv_params.conv_filter_strides = std::vector<ck::index_t>{2};
|
||||
conv_params.input_left_pads = std::vector<ck::index_t>{2};
|
||||
conv_params.input_right_pads = std::vector<ck::index_t>{2};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = cmp_vec(out_spatial_len,
|
||||
std::vector<ck::index_t>{37},
|
||||
"Error: ConvParams 1D padding left/right {2}.");
|
||||
res = test::check_err(out_spatial_len,
|
||||
std::vector<ck::index_t>{37},
|
||||
"Error: ConvParams 1D padding left/right {2}.");
|
||||
|
||||
conv_params.conv_filter_dilations = std::vector<ck::index_t>{2};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = cmp_vec(
|
||||
res = test::check_err(
|
||||
out_spatial_len, std::vector<ck::index_t>{36}, "Error: ConvParams 1D dilation {2}.");
|
||||
|
||||
conv_params.conv_filter_strides = std::vector<ck::index_t>{3};
|
||||
@@ -110,9 +86,52 @@ bool TestConvParams_GetOutputSpatialLengths()
|
||||
conv_params.input_right_pads = std::vector<ck::index_t>{1};
|
||||
conv_params.conv_filter_dilations = std::vector<ck::index_t>{2};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = cmp_vec(out_spatial_len,
|
||||
std::vector<ck::index_t>{23},
|
||||
"Error: ConvParams 1D strides{3}, padding {1}, dilations {2}.");
|
||||
res = test::check_err(out_spatial_len,
|
||||
std::vector<ck::index_t>{23},
|
||||
"Error: ConvParams 1D strides{3}, padding {1}, dilations {2}.");
|
||||
|
||||
// -------------------------- 3D ------------------------------------
|
||||
conv_params.num_dim_spatial = 3;
|
||||
conv_params.filter_spatial_lengths = std::vector<ck::index_t>{3, 3, 3};
|
||||
conv_params.input_spatial_lengths = std::vector<ck::index_t>{71, 71, 71};
|
||||
conv_params.conv_filter_strides = std::vector<ck::index_t>{2, 2, 2};
|
||||
conv_params.conv_filter_dilations = std::vector<ck::index_t>{1, 1, 1};
|
||||
conv_params.input_left_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
conv_params.input_right_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = test::check_err(
|
||||
out_spatial_len, std::vector<ck::index_t>{36, 36, 36}, "Error: ConvParams 3D.");
|
||||
|
||||
conv_params.conv_filter_strides = std::vector<ck::index_t>{1, 1, 1};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = test::check_err(out_spatial_len,
|
||||
std::vector<ck::index_t>{71, 71, 71},
|
||||
"Error: ConvParams 3D stride {1, 1, 1}.");
|
||||
|
||||
conv_params.conv_filter_strides = std::vector<ck::index_t>{2, 2, 2};
|
||||
conv_params.input_left_pads = std::vector<ck::index_t>{2, 2, 2};
|
||||
conv_params.input_right_pads = std::vector<ck::index_t>{2, 2, 2};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = test::check_err(out_spatial_len,
|
||||
std::vector<ck::index_t>{37, 37, 37},
|
||||
"Error: ConvParams 3D padding left/right {2, 2, 2}.");
|
||||
|
||||
conv_params.conv_filter_dilations = std::vector<ck::index_t>{2, 2, 2};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = test::check_err(out_spatial_len,
|
||||
std::vector<ck::index_t>{36, 36, 36},
|
||||
"Error: ConvParams 3D dilation {2, 2, 2}.");
|
||||
|
||||
conv_params.conv_filter_strides = std::vector<ck::index_t>{3, 3, 3};
|
||||
conv_params.input_left_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
conv_params.input_right_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
conv_params.conv_filter_dilations = std::vector<ck::index_t>{2, 2, 2};
|
||||
out_spatial_len = conv_params.GetOutputSpatialLengths();
|
||||
res = test::check_err(
|
||||
out_spatial_len,
|
||||
std::vector<ck::index_t>{23, 23, 23},
|
||||
"Error: ConvParams 3D strides{3, 3, 3}, padding {1, 1, 1}, dilations {2, 2, 2}.");
|
||||
|
||||
return res;
|
||||
}
|
||||
@@ -123,23 +142,44 @@ bool TestGetHostTensorDescriptor()
|
||||
namespace tl = ck::tensor_layout::convolution;
|
||||
std::vector<std::size_t> dims{2, 3, 4, 5};
|
||||
HostTensorDescriptor h = ck::conv_util::GetHostTensorDescriptor(dims, tl::NHWC{});
|
||||
res = cmp_vec(h.GetLengths(), {2, 3, 4, 5}, "Error: wrong NHWC dimensions lengths!");
|
||||
res =
|
||||
cmp_vec(h.GetStrides(), {3 * 4 * 5, 1, 3 * 5, 3}, "Error: wrong NHWC dimensions strides!");
|
||||
res = test::check_err(h.GetLengths(), {2, 3, 4, 5}, "Error: wrong NHWC dimensions lengths!");
|
||||
res = test::check_err(
|
||||
h.GetStrides(), {3 * 4 * 5, 1, 3 * 5, 3}, "Error: wrong NHWC dimensions strides!");
|
||||
|
||||
h = ck::conv_util::GetHostTensorDescriptor(dims, tl::NCHW{});
|
||||
res = cmp_vec(h.GetLengths(), {2, 3, 4, 5}, "Error: wrong NCHW dimensions lengths!");
|
||||
res =
|
||||
cmp_vec(h.GetStrides(), {3 * 4 * 5, 4 * 5, 5, 1}, "Error: wrong NCHW dimensions strides!");
|
||||
res = test::check_err(h.GetLengths(), {2, 3, 4, 5}, "Error: wrong NCHW dimensions lengths!");
|
||||
res = test::check_err(
|
||||
h.GetStrides(), {3 * 4 * 5, 4 * 5, 5, 1}, "Error: wrong NCHW dimensions strides!");
|
||||
|
||||
dims = std::vector<std::size_t>{2, 3, 4};
|
||||
h = ck::conv_util::GetHostTensorDescriptor(dims, tl::NWC{});
|
||||
res = cmp_vec(h.GetLengths(), {2, 3, 4}, "Error: wrong NWC dimensions lengths!");
|
||||
res = cmp_vec(h.GetStrides(), {3 * 4, 1, 3}, "Error: wrong NWC dimensions strides!");
|
||||
res = test::check_err(h.GetLengths(), {2, 3, 4}, "Error: wrong NWC dimensions lengths!");
|
||||
res = test::check_err(h.GetStrides(), {3 * 4, 1, 3}, "Error: wrong NWC dimensions strides!");
|
||||
|
||||
h = ck::conv_util::GetHostTensorDescriptor(dims, tl::NCW{});
|
||||
res = cmp_vec(h.GetLengths(), {2, 3, 4}, "Error: wrong NCW dimensions lengths!");
|
||||
res = cmp_vec(h.GetStrides(), {3 * 4, 4, 1}, "Error: wrong NCW dimensions strides!");
|
||||
res = test::check_err(h.GetLengths(), {2, 3, 4}, "Error: wrong NCW dimensions lengths!");
|
||||
res = test::check_err(h.GetStrides(), {3 * 4, 4, 1}, "Error: wrong NCW dimensions strides!");
|
||||
|
||||
dims = std::vector<std::size_t>{2, 3, 4, 5, 6};
|
||||
h = ck::conv_util::GetHostTensorDescriptor(dims, tl::NDHWC{});
|
||||
res = test::check_err(h.GetLengths(), dims, "Error: wrong NDHWC dimensions lengths!");
|
||||
res = test::check_err(h.GetStrides(),
|
||||
{3 * 4 * 5 * 6, // N
|
||||
1, // C
|
||||
3 * 5 * 6, // D
|
||||
3 * 6, // H
|
||||
3}, // W
|
||||
"Error: wrong NDHWC dimensions strides!");
|
||||
|
||||
h = ck::conv_util::GetHostTensorDescriptor(dims, tl::NCDHW{});
|
||||
res = test::check_err(h.GetLengths(), dims, "Error: wrong NCDHW dimensions lengths!");
|
||||
res = test::check_err(h.GetStrides(),
|
||||
{3 * 4 * 5 * 6, // N
|
||||
4 * 5 * 6, // C
|
||||
5 * 6, // D
|
||||
6, // H
|
||||
1}, // W
|
||||
"Error: wrong NCDHW dimensions strides!");
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
@@ -1,2 +1,17 @@
|
||||
add_test_executable(test_convnd_fwd convnd_fwd.cpp)
|
||||
target_link_libraries(test_convnd_fwd PRIVATE host_tensor)
|
||||
add_custom_target(test_convnd_fwd)
|
||||
|
||||
add_test_executable(test_conv1d_fwd conv1d_fwd.cpp)
|
||||
target_link_libraries(test_conv1d_fwd PRIVATE host_tensor)
|
||||
target_link_libraries(test_conv1d_fwd PRIVATE device_conv1d_fwd_instance)
|
||||
add_dependencies(test_convnd_fwd test_conv1d_fwd)
|
||||
|
||||
add_test_executable(test_conv2d_fwd conv2d_fwd.cpp)
|
||||
target_link_libraries(test_conv2d_fwd PRIVATE host_tensor)
|
||||
target_link_libraries(test_conv2d_fwd PRIVATE device_conv2d_fwd_instance)
|
||||
add_dependencies(test_convnd_fwd test_conv2d_fwd)
|
||||
|
||||
add_test_executable(test_conv3d_fwd conv3d_fwd.cpp)
|
||||
target_link_libraries(test_conv3d_fwd PRIVATE host_tensor)
|
||||
target_link_libraries(test_conv3d_fwd PRIVATE device_conv3d_fwd_instance)
|
||||
add_dependencies(test_convnd_fwd test_conv3d_fwd)
|
||||
|
||||
|
||||
149
test/convnd_fwd/conv1d_fwd.cpp
Normal file
149
test/convnd_fwd/conv1d_fwd.cpp
Normal file
@@ -0,0 +1,149 @@
|
||||
#include <iostream>
|
||||
#include <stdexcept>
|
||||
#include <tuple>
|
||||
#include <vector>
|
||||
|
||||
#include "data_type.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "conv_test_util.hpp"
|
||||
#include "host_tensor.hpp"
|
||||
#include "tensor_layout.hpp"
|
||||
#include "test_util.hpp"
|
||||
|
||||
// Forward declarations for conv instances.
|
||||
|
||||
using DeviceConvFwdNoOpPtr =
|
||||
ck::tensor_operation::device::DeviceConvFwdPtr<ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough>;
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv1d_fwd_instance {
|
||||
|
||||
void add_device_conv1d_fwd_xdl_nwc_kxc_nwk_bf16_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
void add_device_conv1d_fwd_xdl_nwc_kxc_nwk_f16_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
void add_device_conv1d_fwd_xdl_nwc_kxc_nwk_f32_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
void add_device_conv1d_fwd_xdl_nwc_kxc_nwk_int8_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
} // namespace device_conv1d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
|
||||
namespace {
|
||||
|
||||
bool TestConv1DNWC()
|
||||
{
|
||||
bool res{true};
|
||||
ck::conv_util::ConvParams params;
|
||||
params.num_dim_spatial = 1;
|
||||
params.N = 2;
|
||||
params.K = 16;
|
||||
params.C = 4;
|
||||
params.filter_spatial_lengths = std::vector<ck::index_t>{3};
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{16};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{1};
|
||||
params.conv_filter_dilations = std::vector<ck::index_t>{1};
|
||||
params.input_left_pads = std::vector<ck::index_t>{1};
|
||||
params.input_right_pads = std::vector<ck::index_t>{1};
|
||||
|
||||
auto host_tensors = test::conv::GetHostTensors<float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::convolution::NWC,
|
||||
ck::tensor_layout::convolution::KXC,
|
||||
ck::tensor_layout::convolution::NWK>(params);
|
||||
const Tensor<float>& input = std::get<0>(host_tensors);
|
||||
const Tensor<float>& weights = std::get<1>(host_tensors);
|
||||
Tensor<float>& host_output = std::get<2>(host_tensors);
|
||||
Tensor<float>& device_output = std::get<3>(host_tensors);
|
||||
|
||||
test::conv::RunReferenceConv<1>(params, input, weights, host_output);
|
||||
test::conv::RunConv<1>(params, input, weights, device_output);
|
||||
res = res &&
|
||||
test::check_err(
|
||||
device_output.mData, host_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
bool TestConv1DNWCInstances(const std::vector<DeviceConvFwdNoOpPtr>& conv_ptrs)
|
||||
{
|
||||
ck::conv_util::ConvParams params;
|
||||
params.num_dim_spatial = 1;
|
||||
params.filter_spatial_lengths = std::vector<ck::index_t>{3};
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{71};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{2};
|
||||
params.conv_filter_dilations = std::vector<ck::index_t>{1};
|
||||
params.input_left_pads = std::vector<ck::index_t>{1};
|
||||
params.input_right_pads = std::vector<ck::index_t>{1};
|
||||
|
||||
auto host_tensors = test::conv::GetHostTensors<T,
|
||||
T,
|
||||
T,
|
||||
ck::tensor_layout::convolution::NWC,
|
||||
ck::tensor_layout::convolution::KXC,
|
||||
ck::tensor_layout::convolution::NWK>(params);
|
||||
const Tensor<T>& input = std::get<0>(host_tensors);
|
||||
const Tensor<T>& weights = std::get<1>(host_tensors);
|
||||
Tensor<T>& host_output = std::get<2>(host_tensors);
|
||||
Tensor<T>& device_output = std::get<3>(host_tensors);
|
||||
|
||||
test::conv::RunReferenceConv<1>(params, input, weights, host_output);
|
||||
return test::conv::RunConvInstances<1>(
|
||||
params, conv_ptrs, input, weights, device_output, host_output);
|
||||
}
|
||||
bool TestConv1DNWCBF16Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv1d_fwd_instance::
|
||||
add_device_conv1d_fwd_xdl_nwc_kxc_nwk_bf16_instances(conv_ptrs);
|
||||
return TestConv1DNWCInstances<ck::bhalf_t>(conv_ptrs);
|
||||
}
|
||||
|
||||
bool TestConv1DNWCF16Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv1d_fwd_instance::
|
||||
add_device_conv1d_fwd_xdl_nwc_kxc_nwk_f16_instances(conv_ptrs);
|
||||
return TestConv1DNWCInstances<ck::half_t>(conv_ptrs);
|
||||
}
|
||||
|
||||
bool TestConv1DNWCF32Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv1d_fwd_instance::
|
||||
add_device_conv1d_fwd_xdl_nwc_kxc_nwk_f32_instances(conv_ptrs);
|
||||
return TestConv1DNWCInstances<float>(conv_ptrs);
|
||||
}
|
||||
|
||||
bool TestConv1DNWCInt8Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv1d_fwd_instance::
|
||||
add_device_conv1d_fwd_xdl_nwc_kxc_nwk_int8_instances(conv_ptrs);
|
||||
return TestConv1DNWCInstances<int8_t>(conv_ptrs);
|
||||
}
|
||||
|
||||
} // anonymous namespace
|
||||
|
||||
int main()
|
||||
{
|
||||
bool res{true};
|
||||
res = TestConv1DNWC();
|
||||
std::cout << "TestConv1DNWC ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
|
||||
res = TestConv1DNWCBF16Instances();
|
||||
std::cout << "\nTestConv1DNWCBF16Instances ..... " << (res ? "SUCCESS" : "FAILURE")
|
||||
<< std::endl;
|
||||
res = TestConv1DNWCF16Instances();
|
||||
std::cout << "\nTestConv1DNWCF16Instances ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
res = TestConv1DNWCF32Instances();
|
||||
std::cout << "\nTestConv1DNWCF32Instances ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
res = TestConv1DNWCInt8Instances();
|
||||
std::cout << "\nTestConv1DNWCInt8Instances ..... " << (res ? "SUCCESS" : "FAILURE")
|
||||
<< std::endl;
|
||||
}
|
||||
147
test/convnd_fwd/conv2d_fwd.cpp
Normal file
147
test/convnd_fwd/conv2d_fwd.cpp
Normal file
@@ -0,0 +1,147 @@
|
||||
#include <half.hpp>
|
||||
#include <iostream>
|
||||
#include <stdexcept>
|
||||
#include <tuple>
|
||||
#include <vector>
|
||||
|
||||
#include "data_type.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "conv_test_util.hpp"
|
||||
#include "host_tensor.hpp"
|
||||
#include "tensor_layout.hpp"
|
||||
#include "test_util.hpp"
|
||||
|
||||
// Forward declarations for conv instances.
|
||||
using DeviceConvFwdNoOpPtr =
|
||||
ck::tensor_operation::device::DeviceConvFwdPtr<ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough>;
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv2d_fwd_instance {
|
||||
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
void add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(
|
||||
std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
} // namespace device_conv2d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
|
||||
namespace {
|
||||
|
||||
bool TestConv2DNHWC()
|
||||
{
|
||||
bool res{true};
|
||||
ck::conv_util::ConvParams params;
|
||||
params.N = 2;
|
||||
params.K = 16;
|
||||
params.C = 4;
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{16, 16};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{1, 1};
|
||||
|
||||
auto host_tensors = test::conv::GetHostTensors(params);
|
||||
const Tensor<float>& input = std::get<0>(host_tensors);
|
||||
const Tensor<float>& weights = std::get<1>(host_tensors);
|
||||
Tensor<float>& host_output = std::get<2>(host_tensors);
|
||||
Tensor<float>& device_output = std::get<3>(host_tensors);
|
||||
|
||||
test::conv::RunReferenceConv<2>(params, input, weights, host_output);
|
||||
test::conv::RunConv<2>(params, input, weights, device_output);
|
||||
res = res &&
|
||||
test::check_err(
|
||||
device_output.mData, host_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
bool TestConv2DNHWCInstances(const std::vector<DeviceConvFwdNoOpPtr>& conv_ptrs)
|
||||
{
|
||||
ck::conv_util::ConvParams params;
|
||||
params.num_dim_spatial = 2;
|
||||
params.filter_spatial_lengths = std::vector<ck::index_t>{3, 3};
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{71, 71};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{2, 2};
|
||||
params.conv_filter_dilations = std::vector<ck::index_t>{1, 1};
|
||||
params.input_left_pads = std::vector<ck::index_t>{1, 1};
|
||||
params.input_right_pads = std::vector<ck::index_t>{1, 1};
|
||||
|
||||
auto host_tensors = test::conv::GetHostTensors<T,
|
||||
T,
|
||||
T,
|
||||
ck::tensor_layout::convolution::NHWC,
|
||||
ck::tensor_layout::convolution::KYXC,
|
||||
ck::tensor_layout::convolution::NHWK>(params);
|
||||
const Tensor<T>& input = std::get<0>(host_tensors);
|
||||
const Tensor<T>& weights = std::get<1>(host_tensors);
|
||||
Tensor<T>& host_output = std::get<2>(host_tensors);
|
||||
Tensor<T>& device_output = std::get<3>(host_tensors);
|
||||
|
||||
test::conv::RunReferenceConv<2>(params, input, weights, host_output);
|
||||
return test::conv::RunConvInstances<2>(
|
||||
params, conv_ptrs, input, weights, device_output, host_output);
|
||||
}
|
||||
|
||||
bool TestConv2DNHWCBF16Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(conv_ptrs);
|
||||
return TestConv2DNHWCInstances<ck::bhalf_t>(conv_ptrs);
|
||||
}
|
||||
|
||||
bool TestConv2DNHWCF16Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
|
||||
return TestConv2DNHWCInstances<ck::half_t>(conv_ptrs);
|
||||
}
|
||||
|
||||
bool TestConv2DNHWCF32Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(conv_ptrs);
|
||||
return TestConv2DNHWCInstances<float>(conv_ptrs);
|
||||
}
|
||||
|
||||
bool TestConv2DNHWCInt8Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv2d_fwd_instance::
|
||||
add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(conv_ptrs);
|
||||
return TestConv2DNHWCInstances<int8_t>(conv_ptrs);
|
||||
}
|
||||
|
||||
} // anonymous namespace
|
||||
|
||||
int main()
|
||||
{
|
||||
bool res{true};
|
||||
res = TestConv2DNHWC();
|
||||
std::cout << "TestConv2DNHWC ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
|
||||
res = TestConv2DNHWCBF16Instances();
|
||||
std::cout << "\nTestConv2DNHWCBF16Instances ..... " << (res ? "SUCCESS" : "FAILURE")
|
||||
<< std::endl;
|
||||
res = TestConv2DNHWCF16Instances();
|
||||
std::cout << "\nTestConv2DNHWCF16Instances ....." << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
res = TestConv2DNHWCF32Instances();
|
||||
std::cout << "\nTestConv2DNHWCF32Instances ..... " << (res ? "SUCCESS" : "FAILURE")
|
||||
<< std::endl;
|
||||
res = TestConv2DNHWCInt8Instances();
|
||||
std::cout << "\nTestConv2DNHWCInt8Instances ..... " << (res ? "SUCCESS" : "FAILURE")
|
||||
<< std::endl;
|
||||
|
||||
return 0;
|
||||
}
|
||||
294
test/convnd_fwd/conv3d_fwd.cpp
Normal file
294
test/convnd_fwd/conv3d_fwd.cpp
Normal file
@@ -0,0 +1,294 @@
|
||||
#include <half.hpp>
|
||||
#include <iostream>
|
||||
#include <stdexcept>
|
||||
#include <tuple>
|
||||
#include <vector>
|
||||
|
||||
#include "data_type.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "conv_test_util.hpp"
|
||||
#include "host_tensor.hpp"
|
||||
#include "tensor_layout.hpp"
|
||||
#include "test_util.hpp"
|
||||
|
||||
// Forward declarations for conv instances.
|
||||
using DeviceConvFwdNoOpPtr =
|
||||
ck::tensor_operation::device::DeviceConvFwdPtr<ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough>;
|
||||
|
||||
namespace ck {
|
||||
namespace tensor_operation {
|
||||
namespace device {
|
||||
namespace device_conv3d_fwd_instance {
|
||||
|
||||
void add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
void add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f16_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
void add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f32_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
void add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_int8_instances(std::vector<DeviceConvFwdNoOpPtr>&);
|
||||
|
||||
} // namespace device_conv3d_fwd_instance
|
||||
} // namespace device
|
||||
} // namespace tensor_operation
|
||||
} // namespace ck
|
||||
|
||||
namespace {
|
||||
|
||||
bool TestConv3DNDHWC()
|
||||
{
|
||||
bool res{true};
|
||||
ck::conv_util::ConvParams params;
|
||||
params.num_dim_spatial = 3;
|
||||
params.N = 2;
|
||||
params.K = 16;
|
||||
params.C = 4;
|
||||
params.filter_spatial_lengths = std::vector<ck::index_t>{3, 3, 3};
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{16, 16, 16};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.conv_filter_dilations = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_left_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_right_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
|
||||
auto host_tensors = test::conv::GetHostTensors<float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::convolution::NDHWC,
|
||||
ck::tensor_layout::convolution::KZYXC,
|
||||
ck::tensor_layout::convolution::NDHWK>(params);
|
||||
const Tensor<float>& input = std::get<0>(host_tensors);
|
||||
const Tensor<float>& weights = std::get<1>(host_tensors);
|
||||
Tensor<float>& host_output = std::get<2>(host_tensors);
|
||||
Tensor<float>& device_output = std::get<3>(host_tensors);
|
||||
|
||||
test::conv::RunReferenceConv<3>(params, input, weights, host_output);
|
||||
test::conv::RunConv<3>(params, input, weights, device_output);
|
||||
res = res &&
|
||||
test::check_err(
|
||||
device_output.mData, host_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
bool TestConv3DNDHWC2GBInput()
|
||||
{
|
||||
// >2GB Input
|
||||
ck::conv_util::ConvParams params;
|
||||
params.num_dim_spatial = 3;
|
||||
params.N = 2;
|
||||
params.K = 16;
|
||||
params.C = 32;
|
||||
params.filter_spatial_lengths = std::vector<ck::index_t>{3, 3, 3};
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{32, 1000, 1000};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.conv_filter_dilations = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_left_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_right_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
|
||||
auto host_tensors =
|
||||
test::conv::GetHostTensors<float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::convolution::NDHWC,
|
||||
ck::tensor_layout::convolution::KZYXC,
|
||||
ck::tensor_layout::convolution::NDHWK>(params, false);
|
||||
const Tensor<float>& input = std::get<0>(host_tensors);
|
||||
const Tensor<float>& weights = std::get<1>(host_tensors);
|
||||
Tensor<float>& device_output = std::get<3>(host_tensors);
|
||||
|
||||
try
|
||||
{
|
||||
test::conv::RunConv<3>(params, input, weights, device_output);
|
||||
}
|
||||
catch(const std::runtime_error& err)
|
||||
{
|
||||
std::string err_msg{"Error! device_conv with the specified compilation parameters does "
|
||||
"not support this Conv problem"};
|
||||
if(err.what() != err_msg)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
}
|
||||
std::cout << "Error: Failure checking oversized tensor!" << std::endl;
|
||||
return false;
|
||||
}
|
||||
|
||||
bool TestConv3DNDHWC2GBFilters()
|
||||
{
|
||||
// >2GB Filters
|
||||
ck::conv_util::ConvParams params;
|
||||
params.num_dim_spatial = 3;
|
||||
params.N = 2;
|
||||
params.K = 16;
|
||||
params.C = 32;
|
||||
params.filter_spatial_lengths = std::vector<ck::index_t>{4, 1000, 1000};
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{16, 16, 16};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.conv_filter_dilations = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_left_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_right_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
|
||||
auto host_tensors =
|
||||
test::conv::GetHostTensors<float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::convolution::NDHWC,
|
||||
ck::tensor_layout::convolution::KZYXC,
|
||||
ck::tensor_layout::convolution::NDHWK>(params, false);
|
||||
const Tensor<float>& input = std::get<0>(host_tensors);
|
||||
const Tensor<float>& weights = std::get<1>(host_tensors);
|
||||
Tensor<float>& device_output = std::get<3>(host_tensors);
|
||||
|
||||
try
|
||||
{
|
||||
test::conv::RunConv<3>(params, input, weights, device_output);
|
||||
}
|
||||
catch(const std::runtime_error& err)
|
||||
{
|
||||
std::string err_msg{"Error! device_conv with the specified compilation parameters does "
|
||||
"not support this Conv problem"};
|
||||
if(err.what() != err_msg)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
}
|
||||
std::cout << "Error: Failure checking oversized tensor!" << std::endl;
|
||||
return false;
|
||||
}
|
||||
|
||||
bool TestConv3DNDHWC2GBOutput()
|
||||
{
|
||||
// >2GB Output
|
||||
ck::conv_util::ConvParams params;
|
||||
params.num_dim_spatial = 3;
|
||||
params.N = 2;
|
||||
params.K = 16;
|
||||
params.C = 2;
|
||||
params.filter_spatial_lengths = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{1000, 1000, 30};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.conv_filter_dilations = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_left_pads = std::vector<ck::index_t>{2, 2, 2};
|
||||
params.input_right_pads = std::vector<ck::index_t>{2, 2, 2};
|
||||
|
||||
auto host_tensors =
|
||||
test::conv::GetHostTensors<float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::convolution::NDHWC,
|
||||
ck::tensor_layout::convolution::KZYXC,
|
||||
ck::tensor_layout::convolution::NDHWK>(params, false);
|
||||
const Tensor<float>& input = std::get<0>(host_tensors);
|
||||
const Tensor<float>& weights = std::get<1>(host_tensors);
|
||||
Tensor<float>& device_output = std::get<3>(host_tensors);
|
||||
|
||||
try
|
||||
{
|
||||
test::conv::RunConv<3>(params, input, weights, device_output);
|
||||
}
|
||||
catch(const std::runtime_error& err)
|
||||
{
|
||||
std::string err_msg{"Error! device_conv with the specified compilation parameters does "
|
||||
"not support this Conv problem"};
|
||||
if(err.what() != err_msg)
|
||||
{
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
}
|
||||
std::cout << "Error: Failure checking oversized tensor!" << std::endl;
|
||||
return false;
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
bool TestConv3DNDHWCInstances(const std::vector<DeviceConvFwdNoOpPtr>& conv_ptrs)
|
||||
{
|
||||
ck::conv_util::ConvParams params;
|
||||
params.N = 64;
|
||||
params.num_dim_spatial = 3;
|
||||
params.filter_spatial_lengths = std::vector<ck::index_t>{3, 3, 2};
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{32, 32, 2};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{2, 2, 2};
|
||||
params.conv_filter_dilations = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_left_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_right_pads = std::vector<ck::index_t>{1, 1, 1};
|
||||
|
||||
auto host_tensors = test::conv::GetHostTensors<T,
|
||||
T,
|
||||
T,
|
||||
ck::tensor_layout::convolution::NDHWC,
|
||||
ck::tensor_layout::convolution::KZYXC,
|
||||
ck::tensor_layout::convolution::NDHWK>(params);
|
||||
const Tensor<T>& input = std::get<0>(host_tensors);
|
||||
const Tensor<T>& weights = std::get<1>(host_tensors);
|
||||
Tensor<T>& host_output = std::get<2>(host_tensors);
|
||||
Tensor<T>& device_output = std::get<3>(host_tensors);
|
||||
|
||||
test::conv::RunReferenceConv<3>(params, input, weights, host_output);
|
||||
return test::conv::RunConvInstances<3>(
|
||||
params, conv_ptrs, input, weights, device_output, host_output);
|
||||
}
|
||||
|
||||
bool TestConv3DNDHWCBF16Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv3d_fwd_instance::
|
||||
add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(conv_ptrs);
|
||||
return TestConv3DNDHWCInstances<ck::bhalf_t>(conv_ptrs);
|
||||
}
|
||||
|
||||
bool TestConv3DNDHWCF16Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv3d_fwd_instance::
|
||||
add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f16_instances(conv_ptrs);
|
||||
return TestConv3DNDHWCInstances<ck::half_t>(conv_ptrs);
|
||||
}
|
||||
|
||||
bool TestConv3DNDHWCF32Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv3d_fwd_instance::
|
||||
add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_f32_instances(conv_ptrs);
|
||||
return TestConv3DNDHWCInstances<float>(conv_ptrs);
|
||||
}
|
||||
|
||||
bool TestConv3DNDHWCInt8Instances()
|
||||
{
|
||||
std::vector<DeviceConvFwdNoOpPtr> conv_ptrs;
|
||||
ck::tensor_operation::device::device_conv3d_fwd_instance::
|
||||
add_device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk_int8_instances(conv_ptrs);
|
||||
return TestConv3DNDHWCInstances<int8_t>(conv_ptrs);
|
||||
}
|
||||
|
||||
} // anonymous namespace
|
||||
|
||||
int main()
|
||||
{
|
||||
bool res{true};
|
||||
res = TestConv3DNDHWC();
|
||||
std::cout << "TestConv3DNDHWC ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
|
||||
res = TestConv3DNDHWC2GBInput();
|
||||
std::cout << "\nTestConv3DNDHWC2GBInput ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
res = TestConv3DNDHWC2GBFilters();
|
||||
std::cout << "\nTestConv3DNDHWC2GBFilters ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
res = TestConv3DNDHWC2GBOutput();
|
||||
std::cout << "\nTestConv3DNDHWC2GBOutput ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
|
||||
res = TestConv3DNDHWCBF16Instances();
|
||||
std::cout << "\nTestConv3DNDHWCBF16Instances ..... " << (res ? "SUCCESS" : "FAILURE")
|
||||
<< std::endl;
|
||||
res = TestConv3DNDHWCF16Instances();
|
||||
std::cout << "\nTestConv3DNDHWCF16Instances ..... " << (res ? "SUCCESS" : "FAILURE")
|
||||
<< std::endl;
|
||||
res = TestConv3DNDHWCF32Instances();
|
||||
std::cout << "\nTestConv3DNDHWCF32Instances ..... " << (res ? "SUCCESS" : "FAILURE")
|
||||
<< std::endl;
|
||||
res = TestConv3DNDHWCInt8Instances();
|
||||
std::cout << "\nTestConv3DNDHWCInt8Instances ..... " << (res ? "SUCCESS" : "FAILURE")
|
||||
<< std::endl;
|
||||
|
||||
return 0;
|
||||
}
|
||||
@@ -202,19 +202,19 @@ struct TestGemm
|
||||
bool res = false;
|
||||
if(std::is_same<CDataType, float>::value)
|
||||
{
|
||||
res = test_util::check_err(c_device.mData, c_host.mData, "Error: incorrect results!");
|
||||
res = test::check_err(c_device.mData, c_host.mData, "Error: incorrect results!");
|
||||
|
||||
std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
}
|
||||
else if(std::is_same<CDataType, ck::half_t>::value)
|
||||
{
|
||||
res = test_util::check_err(c_device.mData, c_host.mData, "Error: incorrect results!");
|
||||
res = test::check_err(c_device.mData, c_host.mData, "Error: incorrect results!");
|
||||
|
||||
std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
}
|
||||
else if(std::is_same<CDataType, int8_t>::value)
|
||||
{
|
||||
res = test_util::check_err(c_device.mData, c_host.mData, "Error: incorrect results!");
|
||||
res = test::check_err(c_device.mData, c_host.mData, "Error: incorrect results!");
|
||||
|
||||
std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
}
|
||||
@@ -330,7 +330,7 @@ struct TestGemmBF16
|
||||
bf16_to_f32_(c_device_bf16, c_device_fp32);
|
||||
|
||||
// Assert
|
||||
bool res = test_util::check_err(
|
||||
bool res = test::check_err(
|
||||
c_device_fp32.mData, c_host_fp32.mData, "Error: incorrect results!", 1e-2f, 1e-3f);
|
||||
|
||||
std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
|
||||
@@ -1,262 +1,289 @@
|
||||
#include <algorithm>
|
||||
#include <cstdlib>
|
||||
#include <half.hpp>
|
||||
#include <iostream>
|
||||
#include <numeric>
|
||||
#include <tuple>
|
||||
#include <vector>
|
||||
|
||||
#include "config.hpp"
|
||||
#include "conv_utils.hpp"
|
||||
#include "device.hpp"
|
||||
#include "device_tensor.hpp"
|
||||
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "host_tensor.hpp"
|
||||
#include "reference_conv_fwd.hpp"
|
||||
#include "tensor_layout.hpp"
|
||||
#include "test_util.hpp"
|
||||
|
||||
namespace {
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
template <ck::index_t SpatialDims, typename InDataType, typename WeiDataType, typename OutDataType>
|
||||
using DeviceConvNDFwdInstance = ck::tensor_operation::device::
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
|
||||
// clang-format off
|
||||
InDataType, //
|
||||
WeiDataType, //
|
||||
OutDataType, //
|
||||
InDataType, //
|
||||
InElementOp, // Input Elementwise Operation
|
||||
WeiElementOp, // Weights Elementwise Operation
|
||||
OutElementOp, // Output Elementwise Operation
|
||||
ConvFwdDefault, // ConvForwardSpecialization
|
||||
SpatialDims, // SptialDims
|
||||
64, // BlockSize
|
||||
16, // MPerBlock
|
||||
16, // NPerBlock
|
||||
4, // K0PerBlock
|
||||
1, // K1
|
||||
16, // MPerXDL
|
||||
16, // NPerXDL
|
||||
1, // MXdlPerWave
|
||||
1, // NXdlPerWave
|
||||
S<1, 16, 1>, // ABlockTransferThreadClusterLengths_K0_M_K1
|
||||
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // ABlockTransferSrcAccessOrder
|
||||
2, // ABlockTransferSrcVectorDim
|
||||
1, // ABlockTransferSrcScalarPerVector
|
||||
1, // ABlockTransferDstScalarPerVector_K1
|
||||
true, // ABlockLdsAddExtraM
|
||||
S<1, 16, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1
|
||||
S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // BBlockTransferSrcAccessOrder
|
||||
2, // BBlockTransferSrcVectorDim
|
||||
1, // BBlockTransferSrcScalarPerVector
|
||||
1, // BBlockTransferDstScalarPerVector_K1
|
||||
true, // BBlockTransferAddExtraN
|
||||
7, // CThreadTransferSrcDstVectorDim
|
||||
1>; // CThreadTransferDstScalarPerVector
|
||||
// clang-format on
|
||||
|
||||
template <typename InDataType = float,
|
||||
typename WeiDataType = float,
|
||||
typename OutDataType = float,
|
||||
typename InLayout = ck::tensor_layout::convolution::NHWC,
|
||||
typename WeiLayout = ck::tensor_layout::convolution::KYXC,
|
||||
typename OutLayout = ck::tensor_layout::convolution::NHWK>
|
||||
auto GetHostTensors(const ck::conv_util::ConvParams& params)
|
||||
{
|
||||
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N),
|
||||
static_cast<std::size_t>(params.C)};
|
||||
input_dims.insert(std::end(input_dims),
|
||||
std::begin(params.input_spatial_lengths),
|
||||
std::end(params.input_spatial_lengths));
|
||||
|
||||
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K),
|
||||
static_cast<std::size_t>(params.C)};
|
||||
filter_dims.insert(std::end(filter_dims),
|
||||
std::begin(params.filter_spatial_lengths),
|
||||
std::end(params.filter_spatial_lengths));
|
||||
|
||||
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
|
||||
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N),
|
||||
static_cast<std::size_t>(params.K)};
|
||||
output_dims.insert(std::end(output_dims),
|
||||
std::begin(output_spatial_lengths),
|
||||
std::end(output_spatial_lengths));
|
||||
|
||||
Tensor<InDataType> input(ck::conv_util::GetHostTensorDescriptor(input_dims, InLayout{}));
|
||||
Tensor<WeiDataType> weights(ck::conv_util::GetHostTensorDescriptor(filter_dims, WeiLayout{}));
|
||||
Tensor<OutDataType> host_output(
|
||||
ck::conv_util::GetHostTensorDescriptor(output_dims, OutLayout{}));
|
||||
Tensor<OutDataType> device_output(
|
||||
ck::conv_util::GetHostTensorDescriptor(output_dims, OutLayout{}));
|
||||
|
||||
std::generate(input.begin(), input.end(), [n = 0]() mutable {
|
||||
return InDataType(n++) * InDataType(0.1f);
|
||||
});
|
||||
std::fill(weights.begin(), weights.end(), WeiDataType(0.5f));
|
||||
std::fill(host_output.begin(), host_output.end(), OutDataType(0.f));
|
||||
std::fill(device_output.begin(), device_output.end(), OutDataType(0.f));
|
||||
|
||||
return std::make_tuple(input, weights, host_output, device_output);
|
||||
}
|
||||
|
||||
template <ck::index_t NDim,
|
||||
typename InDataType = float,
|
||||
typename WeiDataType = float,
|
||||
typename OutDataType = float>
|
||||
void RunReferenceConv(const ck::conv_util::ConvParams& params,
|
||||
const Tensor<InDataType>& input,
|
||||
const Tensor<WeiDataType>& weights,
|
||||
Tensor<OutDataType>& output)
|
||||
{
|
||||
auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<InDataType,
|
||||
WeiDataType,
|
||||
OutDataType,
|
||||
InElementOp,
|
||||
WeiElementOp,
|
||||
OutElementOp,
|
||||
NDim>();
|
||||
auto ref_invoker = ref_conv.MakeInvoker();
|
||||
auto ref_argument = ref_conv.MakeArgument(input,
|
||||
weights,
|
||||
output,
|
||||
params.conv_filter_strides,
|
||||
params.conv_filter_dilations,
|
||||
params.input_left_pads,
|
||||
params.input_right_pads,
|
||||
InElementOp{},
|
||||
WeiElementOp{},
|
||||
OutElementOp{});
|
||||
|
||||
ref_invoker.Run(ref_argument);
|
||||
}
|
||||
|
||||
template <ck::index_t NDim,
|
||||
typename InDataType = float,
|
||||
typename WeiDataType = float,
|
||||
typename OutDataType = float>
|
||||
void RunConv(const ck::conv_util::ConvParams& params,
|
||||
const Tensor<InDataType>& input,
|
||||
const Tensor<WeiDataType>& weights,
|
||||
Tensor<OutDataType>& output)
|
||||
{
|
||||
DeviceMem in_device_buf(sizeof(InDataType) * input.mDesc.GetElementSpace());
|
||||
DeviceMem wei_device_buf(sizeof(WeiDataType) * weights.mDesc.GetElementSpace());
|
||||
DeviceMem out_device_buf(sizeof(OutDataType) * output.mDesc.GetElementSpace());
|
||||
|
||||
in_device_buf.ToDevice(input.mData.data());
|
||||
wei_device_buf.ToDevice(weights.mData.data());
|
||||
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
|
||||
|
||||
auto conv = DeviceConvNDFwdInstance<NDim, InDataType, WeiDataType, OutDataType>();
|
||||
auto invoker = conv.MakeInvoker();
|
||||
auto argument = conv.MakeArgument(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
|
||||
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
|
||||
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
|
||||
params.N,
|
||||
params.K,
|
||||
params.C,
|
||||
params.input_spatial_lengths,
|
||||
params.filter_spatial_lengths,
|
||||
output_spatial_lengths,
|
||||
params.conv_filter_strides,
|
||||
params.conv_filter_dilations,
|
||||
params.input_left_pads,
|
||||
params.input_right_pads,
|
||||
InElementOp{},
|
||||
WeiElementOp{},
|
||||
OutElementOp{});
|
||||
|
||||
if(!conv.IsSupportedArgument(argument))
|
||||
{
|
||||
throw std::runtime_error(
|
||||
"Error! device_conv with the specified compilation parameters does "
|
||||
"not support this Conv problem");
|
||||
}
|
||||
|
||||
invoker.Run(argument);
|
||||
out_device_buf.FromDevice(output.mData.data());
|
||||
}
|
||||
|
||||
bool TestConv2DNHWC()
|
||||
{
|
||||
bool res{true};
|
||||
ck::conv_util::ConvParams params;
|
||||
params.N = 2;
|
||||
params.K = 16;
|
||||
params.C = 4;
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{16, 16};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{1, 1};
|
||||
|
||||
auto host_tensors = GetHostTensors(params);
|
||||
const Tensor<float>& input = std::get<0>(host_tensors);
|
||||
const Tensor<float>& weights = std::get<1>(host_tensors);
|
||||
Tensor<float>& host_output = std::get<2>(host_tensors);
|
||||
Tensor<float>& device_output = std::get<3>(host_tensors);
|
||||
|
||||
RunReferenceConv<2>(params, input, weights, host_output);
|
||||
RunConv<2>(params, input, weights, device_output);
|
||||
res = res &&
|
||||
test_util::check_err(
|
||||
device_output.mData, host_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
bool TestConv1DNWC()
|
||||
{
|
||||
bool res{true};
|
||||
ck::conv_util::ConvParams params;
|
||||
params.num_dim_spatial = 1;
|
||||
params.N = 2;
|
||||
params.K = 16;
|
||||
params.C = 4;
|
||||
params.filter_spatial_lengths = std::vector<ck::index_t>{3};
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{16};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{1};
|
||||
params.conv_filter_dilations = std::vector<ck::index_t>{1};
|
||||
params.input_left_pads = std::vector<ck::index_t>{1};
|
||||
params.input_right_pads = std::vector<ck::index_t>{1};
|
||||
|
||||
auto host_tensors = GetHostTensors<float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::convolution::NWC,
|
||||
ck::tensor_layout::convolution::KXC,
|
||||
ck::tensor_layout::convolution::NWK>(params);
|
||||
const Tensor<float>& input = std::get<0>(host_tensors);
|
||||
const Tensor<float>& weights = std::get<1>(host_tensors);
|
||||
Tensor<float>& host_output = std::get<2>(host_tensors);
|
||||
Tensor<float>& device_output = std::get<3>(host_tensors);
|
||||
|
||||
RunReferenceConv<1>(params, input, weights, host_output);
|
||||
RunConv<1>(params, input, weights, device_output);
|
||||
res = res &&
|
||||
test_util::check_err(
|
||||
device_output.mData, host_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
} // anonymous namespace
|
||||
|
||||
int main()
|
||||
{
|
||||
bool res{true};
|
||||
res = TestConv1DNWC();
|
||||
std::cout << "TestConv1DNWC ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
res = TestConv2DNHWC();
|
||||
std::cout << "TestConv2DNHWC ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
}
|
||||
#ifndef TEST_CONV_UTIL_HPP
|
||||
#define TEST_CONV_UTIL_HPP
|
||||
|
||||
#include <algorithm>
|
||||
#include <cstdlib>
|
||||
#include <numeric>
|
||||
#include <random>
|
||||
#include <stdexcept>
|
||||
#include <tuple>
|
||||
#include <type_traits>
|
||||
#include <vector>
|
||||
|
||||
#include "config.hpp"
|
||||
#include "conv_utils.hpp"
|
||||
#include "device.hpp"
|
||||
#include "device_tensor.hpp"
|
||||
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "host_tensor.hpp"
|
||||
#include "reference_conv_fwd.hpp"
|
||||
#include "tensor_layout.hpp"
|
||||
#include "test_util.hpp"
|
||||
|
||||
namespace {
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
static constexpr auto ConvFwdDefault =
|
||||
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default;
|
||||
|
||||
template <ck::index_t SpatialDims, typename InDataType, typename WeiDataType, typename OutDataType>
|
||||
using DeviceConvNDFwdInstance = ck::tensor_operation::device::
|
||||
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
|
||||
// clang-format off
|
||||
InDataType, //
|
||||
WeiDataType, //
|
||||
OutDataType, //
|
||||
InDataType, //
|
||||
InElementOp, // Input Elementwise Operation
|
||||
WeiElementOp, // Weights Elementwise Operation
|
||||
OutElementOp, // Output Elementwise Operation
|
||||
ConvFwdDefault, // ConvForwardSpecialization
|
||||
SpatialDims, // SptialDims
|
||||
64, // BlockSize
|
||||
16, // MPerBlock
|
||||
16, // NPerBlock
|
||||
4, // K0PerBlock
|
||||
1, // K1
|
||||
16, // MPerXDL
|
||||
16, // NPerXDL
|
||||
1, // MXdlPerWave
|
||||
1, // NXdlPerWave
|
||||
S<1, 16, 1>, // ABlockTransferThreadClusterLengths_K0_M_K1
|
||||
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // ABlockTransferSrcAccessOrder
|
||||
2, // ABlockTransferSrcVectorDim
|
||||
1, // ABlockTransferSrcScalarPerVector
|
||||
1, // ABlockTransferDstScalarPerVector_K1
|
||||
true, // ABlockLdsAddExtraM
|
||||
S<1, 16, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1
|
||||
S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // BBlockTransferSrcAccessOrder
|
||||
2, // BBlockTransferSrcVectorDim
|
||||
1, // BBlockTransferSrcScalarPerVector
|
||||
1, // BBlockTransferDstScalarPerVector_K1
|
||||
true, // BBlockTransferAddExtraN
|
||||
7, // CThreadTransferSrcDstVectorDim
|
||||
1>; // CThreadTransferDstScalarPerVector
|
||||
// clang-format on
|
||||
|
||||
} // namespace
|
||||
|
||||
namespace test {
|
||||
namespace conv {
|
||||
|
||||
using DeviceConvFwdNoOpPtr =
|
||||
ck::tensor_operation::device::DeviceConvFwdPtr<ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough,
|
||||
ck::tensor_operation::element_wise::PassThrough>;
|
||||
|
||||
template <typename InDataType = float,
|
||||
typename WeiDataType = float,
|
||||
typename OutDataType = float,
|
||||
typename InLayout = ck::tensor_layout::convolution::NHWC,
|
||||
typename WeiLayout = ck::tensor_layout::convolution::KYXC,
|
||||
typename OutLayout = ck::tensor_layout::convolution::NHWK>
|
||||
auto GetHostTensors(const ck::conv_util::ConvParams& params, bool init = true)
|
||||
{
|
||||
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N),
|
||||
static_cast<std::size_t>(params.C)};
|
||||
input_dims.insert(std::end(input_dims),
|
||||
std::begin(params.input_spatial_lengths),
|
||||
std::end(params.input_spatial_lengths));
|
||||
|
||||
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K),
|
||||
static_cast<std::size_t>(params.C)};
|
||||
filter_dims.insert(std::end(filter_dims),
|
||||
std::begin(params.filter_spatial_lengths),
|
||||
std::end(params.filter_spatial_lengths));
|
||||
|
||||
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
|
||||
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N),
|
||||
static_cast<std::size_t>(params.K)};
|
||||
output_dims.insert(std::end(output_dims),
|
||||
std::begin(output_spatial_lengths),
|
||||
std::end(output_spatial_lengths));
|
||||
|
||||
Tensor<InDataType> input(ck::conv_util::GetHostTensorDescriptor(input_dims, InLayout{}));
|
||||
Tensor<WeiDataType> weights(ck::conv_util::GetHostTensorDescriptor(filter_dims, WeiLayout{}));
|
||||
Tensor<OutDataType> host_output(
|
||||
ck::conv_util::GetHostTensorDescriptor(output_dims, OutLayout{}));
|
||||
Tensor<OutDataType> device_output(
|
||||
ck::conv_util::GetHostTensorDescriptor(output_dims, OutLayout{}));
|
||||
|
||||
if(init)
|
||||
{
|
||||
std::mt19937 gen(11939);
|
||||
if constexpr(std::is_same<InDataType, uint8_t>::value)
|
||||
{
|
||||
std::uniform_int_distribution<> dis(-5, 5);
|
||||
std::generate(
|
||||
input.begin(), input.end(), [&dis, &gen]() { return InDataType(dis(gen)); });
|
||||
std::generate(
|
||||
weights.begin(), weights.end(), [&dis, &gen]() { return WeiDataType(dis(gen)); });
|
||||
}
|
||||
else
|
||||
{
|
||||
std::uniform_real_distribution<> dis(0.f, 1.f);
|
||||
std::generate(
|
||||
input.begin(), input.end(), [&dis, &gen]() { return InDataType(dis(gen)); });
|
||||
std::generate(
|
||||
weights.begin(), weights.end(), [&dis, &gen]() { return WeiDataType(dis(gen)); });
|
||||
}
|
||||
std::fill(host_output.begin(), host_output.end(), OutDataType(0.f));
|
||||
std::fill(device_output.begin(), device_output.end(), OutDataType(0.f));
|
||||
}
|
||||
|
||||
return std::make_tuple(input, weights, host_output, device_output);
|
||||
}
|
||||
|
||||
template <ck::index_t NDim,
|
||||
typename InDataType = float,
|
||||
typename WeiDataType = float,
|
||||
typename OutDataType = float>
|
||||
void RunReferenceConv(const ck::conv_util::ConvParams& params,
|
||||
const Tensor<InDataType>& input,
|
||||
const Tensor<WeiDataType>& weights,
|
||||
Tensor<OutDataType>& output)
|
||||
{
|
||||
auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<InDataType,
|
||||
WeiDataType,
|
||||
OutDataType,
|
||||
InElementOp,
|
||||
WeiElementOp,
|
||||
OutElementOp,
|
||||
NDim>();
|
||||
auto ref_invoker = ref_conv.MakeInvoker();
|
||||
auto ref_argument = ref_conv.MakeArgument(input,
|
||||
weights,
|
||||
output,
|
||||
params.conv_filter_strides,
|
||||
params.conv_filter_dilations,
|
||||
params.input_left_pads,
|
||||
params.input_right_pads,
|
||||
InElementOp{},
|
||||
WeiElementOp{},
|
||||
OutElementOp{});
|
||||
|
||||
ref_invoker.Run(ref_argument);
|
||||
}
|
||||
|
||||
template <ck::index_t NDim,
|
||||
typename InDataType = float,
|
||||
typename WeiDataType = float,
|
||||
typename OutDataType = float>
|
||||
void RunConv(const ck::conv_util::ConvParams& params,
|
||||
const Tensor<InDataType>& input,
|
||||
const Tensor<WeiDataType>& weights,
|
||||
Tensor<OutDataType>& output)
|
||||
{
|
||||
DeviceMem in_device_buf(sizeof(InDataType) * input.mDesc.GetElementSpace());
|
||||
DeviceMem wei_device_buf(sizeof(WeiDataType) * weights.mDesc.GetElementSpace());
|
||||
DeviceMem out_device_buf(sizeof(OutDataType) * output.mDesc.GetElementSpace());
|
||||
|
||||
in_device_buf.ToDevice(input.mData.data());
|
||||
wei_device_buf.ToDevice(weights.mData.data());
|
||||
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
|
||||
|
||||
auto conv = DeviceConvNDFwdInstance<NDim, InDataType, WeiDataType, OutDataType>();
|
||||
auto invoker = conv.MakeInvoker();
|
||||
auto argument = conv.MakeArgument(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
|
||||
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
|
||||
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
|
||||
params.N,
|
||||
params.K,
|
||||
params.C,
|
||||
params.input_spatial_lengths,
|
||||
params.filter_spatial_lengths,
|
||||
output_spatial_lengths,
|
||||
params.conv_filter_strides,
|
||||
params.conv_filter_dilations,
|
||||
params.input_left_pads,
|
||||
params.input_right_pads,
|
||||
InElementOp{},
|
||||
WeiElementOp{},
|
||||
OutElementOp{});
|
||||
|
||||
if(!conv.IsSupportedArgument(argument))
|
||||
{
|
||||
throw std::runtime_error(
|
||||
"Error! device_conv with the specified compilation parameters does "
|
||||
"not support this Conv problem");
|
||||
}
|
||||
|
||||
invoker.Run(argument);
|
||||
out_device_buf.FromDevice(output.mData.data());
|
||||
}
|
||||
|
||||
template <ck::index_t NDim,
|
||||
typename InDataType = float,
|
||||
typename WeiDataType = float,
|
||||
typename OutDataType = float>
|
||||
bool RunConvInstances(const ck::conv_util::ConvParams& params,
|
||||
const std::vector<DeviceConvFwdNoOpPtr>& conv_ptrs,
|
||||
const Tensor<InDataType>& input,
|
||||
const Tensor<WeiDataType>& weights,
|
||||
Tensor<OutDataType>& output,
|
||||
const Tensor<OutDataType>& host_output)
|
||||
{
|
||||
DeviceMem in_device_buf(sizeof(InDataType) * input.mDesc.GetElementSpace());
|
||||
DeviceMem wei_device_buf(sizeof(WeiDataType) * weights.mDesc.GetElementSpace());
|
||||
DeviceMem out_device_buf(sizeof(OutDataType) * output.mDesc.GetElementSpace());
|
||||
|
||||
in_device_buf.ToDevice(input.mData.data());
|
||||
wei_device_buf.ToDevice(weights.mData.data());
|
||||
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
|
||||
|
||||
bool res{true};
|
||||
for(auto& conv_ptr : conv_ptrs)
|
||||
{
|
||||
auto invoker = conv_ptr->MakeInvokerPointer();
|
||||
auto argument = conv_ptr->MakeArgumentPointer(
|
||||
static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
|
||||
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
|
||||
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
|
||||
params.N,
|
||||
params.K,
|
||||
params.C,
|
||||
params.input_spatial_lengths,
|
||||
params.filter_spatial_lengths,
|
||||
output_spatial_lengths,
|
||||
params.conv_filter_strides,
|
||||
params.conv_filter_dilations,
|
||||
params.input_left_pads,
|
||||
params.input_right_pads,
|
||||
InElementOp{},
|
||||
WeiElementOp{},
|
||||
OutElementOp{});
|
||||
|
||||
if(conv_ptr->IsSupportedArgument(argument.get()))
|
||||
{
|
||||
float atol{1e-5f};
|
||||
float rtol{1e-4f};
|
||||
if constexpr(std::is_same_v<InDataType, ck::half_t>)
|
||||
{
|
||||
atol = 1e-4f;
|
||||
rtol = 2.5e-3f;
|
||||
}
|
||||
invoker->Run(argument.get());
|
||||
out_device_buf.FromDevice(output.mData.data());
|
||||
res = res &&
|
||||
test::check_err(
|
||||
output.mData, host_output.mData, "Error: incorrect results!", atol, rtol);
|
||||
hipGetErrorString(
|
||||
hipMemset(out_device_buf.GetDeviceBuffer(), 0, out_device_buf.mMemSize));
|
||||
}
|
||||
}
|
||||
return res;
|
||||
}
|
||||
|
||||
} // namespace conv
|
||||
} // namespace test
|
||||
|
||||
#endif
|
||||
@@ -1,23 +1,28 @@
|
||||
#ifndef TEST_UTIL_HPP
|
||||
#define TEST_UTIL_HPP
|
||||
|
||||
#include <algorithm>
|
||||
#include <cmath>
|
||||
#include <cstdlib>
|
||||
#include <iostream>
|
||||
#include <iomanip>
|
||||
#include <iterator>
|
||||
#include <limits>
|
||||
#include <type_traits>
|
||||
#include <vector>
|
||||
|
||||
namespace test_util {
|
||||
#include "data_type.hpp"
|
||||
|
||||
namespace test {
|
||||
|
||||
template <typename T>
|
||||
typename std::enable_if<std::is_floating_point<T>::value, bool>::type
|
||||
typename std::enable_if<std::is_floating_point<T>::value && !std::is_same<T, ck::half_t>::value,
|
||||
bool>::type
|
||||
check_err(const std::vector<T>& out,
|
||||
const std::vector<T>& ref,
|
||||
const std::string& msg,
|
||||
T rtol = static_cast<T>(1e-5),
|
||||
T atol = static_cast<T>(1e-8))
|
||||
double rtol = 1e-5,
|
||||
double atol = 1e-8)
|
||||
{
|
||||
if(out.size() != ref.size())
|
||||
{
|
||||
@@ -28,9 +33,9 @@ check_err(const std::vector<T>& out,
|
||||
}
|
||||
|
||||
bool res{true};
|
||||
int err_count = 0;
|
||||
T err = 0;
|
||||
T max_err = std::numeric_limits<T>::min();
|
||||
int err_count = 0;
|
||||
double err = 0;
|
||||
double max_err = std::numeric_limits<double>::min();
|
||||
for(std::size_t i = 0; i < ref.size(); ++i)
|
||||
{
|
||||
err = std::abs(out[i] - ref[i]);
|
||||
@@ -41,7 +46,53 @@ check_err(const std::vector<T>& out,
|
||||
if(err_count < 5)
|
||||
{
|
||||
std::cout << std::setw(12) << std::setprecision(7) << "out[" << i << "] != ref["
|
||||
<< i << "]: " << out[i] << "!=" << ref[i] << std::endl
|
||||
<< i << "]: " << out[i] << " != " << ref[i] << std::endl
|
||||
<< msg << std::endl;
|
||||
}
|
||||
res = false;
|
||||
}
|
||||
}
|
||||
if(!res)
|
||||
{
|
||||
std::cout << std::setw(12) << std::setprecision(7) << "max err: " << max_err << std::endl;
|
||||
}
|
||||
return res;
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
typename std::enable_if<std::is_same<T, ck::bhalf_t>::value || std::is_same<T, ck::half_t>::value,
|
||||
bool>::type
|
||||
check_err(const std::vector<T>& out,
|
||||
const std::vector<T>& ref,
|
||||
const std::string& msg,
|
||||
double rtol = 1e-5,
|
||||
double atol = 1e-8)
|
||||
{
|
||||
if(out.size() != ref.size())
|
||||
{
|
||||
std::cout << "out.size() != ref.size(), :" << out.size() << " != " << ref.size()
|
||||
<< std::endl
|
||||
<< msg << std::endl;
|
||||
return false;
|
||||
}
|
||||
|
||||
bool res{true};
|
||||
int err_count = 0;
|
||||
double err = 0;
|
||||
double max_err = ck::type_convert<float>(ck::NumericLimits<T>::Min());
|
||||
for(std::size_t i = 0; i < ref.size(); ++i)
|
||||
{
|
||||
float o = ck::type_convert<float>(out[i]);
|
||||
float r = ck::type_convert<float>(ref[i]);
|
||||
err = std::abs(o - r);
|
||||
if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
|
||||
{
|
||||
max_err = err > max_err ? err : max_err;
|
||||
err_count++;
|
||||
if(err_count < 5)
|
||||
{
|
||||
std::cout << std::setw(12) << std::setprecision(7) << "out[" << i << "] != ref["
|
||||
<< i << "]: " << o << " != " << r << std::endl
|
||||
<< msg << std::endl;
|
||||
}
|
||||
res = false;
|
||||
@@ -98,8 +149,13 @@ bool check_err(const std::vector<_Float16>& out,
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
typename std::enable_if<std::is_integral<T>::value, bool>::type check_err(
|
||||
const std::vector<T>& out, const std::vector<T>& ref, const std::string& msg, T = 0, T = 0)
|
||||
typename std::enable_if<std::is_integral<T>::value && !std::is_same<T, ck::bhalf_t>::value,
|
||||
bool>::type
|
||||
check_err(const std::vector<T>& out,
|
||||
const std::vector<T>& ref,
|
||||
const std::string& msg,
|
||||
double = 0,
|
||||
double = 0)
|
||||
{
|
||||
if(out.size() != ref.size())
|
||||
{
|
||||
@@ -113,7 +169,7 @@ typename std::enable_if<std::is_integral<T>::value, bool>::type check_err(
|
||||
{
|
||||
if(out[i] != ref[i])
|
||||
{
|
||||
std::cout << "out[" << i << "] != ref[" << i << "]: " << out[i] << "!=" << ref[i]
|
||||
std::cout << "out[" << i << "] != ref[" << i << "]: " << out[i] << " != " << ref[i]
|
||||
<< std::endl
|
||||
<< msg << std::endl;
|
||||
return false;
|
||||
@@ -122,6 +178,13 @@ typename std::enable_if<std::is_integral<T>::value, bool>::type check_err(
|
||||
return true;
|
||||
}
|
||||
|
||||
} // namespace test_util
|
||||
} // namespace test
|
||||
|
||||
template <typename T>
|
||||
std::ostream& operator<<(std::ostream& os, const std::vector<T>& v)
|
||||
{
|
||||
std::copy(std::begin(v), std::end(v), std::ostream_iterator<T>(os, " "));
|
||||
return os;
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
@@ -289,13 +289,13 @@ bool test_reduce_no_index_impl(int init_method,
|
||||
{
|
||||
reduce_util::to_f32_vector(out, out_fp32);
|
||||
reduce_util::to_f32_vector(out_ref, out_ref_fp32);
|
||||
single_result = test_util::check_err(
|
||||
single_result = test::check_err(
|
||||
out_fp32.mData, out_ref_fp32.mData, "Error: incorrect data result!");
|
||||
}
|
||||
else
|
||||
{
|
||||
single_result =
|
||||
test_util::check_err(out.mData, out_ref.mData, "Error: incorrect data result!");
|
||||
test::check_err(out.mData, out_ref.mData, "Error: incorrect data result!");
|
||||
};
|
||||
|
||||
if(!single_result)
|
||||
@@ -376,13 +376,13 @@ bool test_reduce_no_index_impl(int init_method,
|
||||
{
|
||||
reduce_util::to_f32_vector(out, out_fp32);
|
||||
reduce_util::to_f32_vector(out_ref, out_ref_fp32);
|
||||
single_result = test_util::check_err(
|
||||
single_result = test::check_err(
|
||||
out_fp32.mData, out_ref_fp32.mData, "Error: incorrect data result!");
|
||||
}
|
||||
else
|
||||
{
|
||||
single_result =
|
||||
test_util::check_err(out.mData, out_ref.mData, "Error: incorrect data result!");
|
||||
test::check_err(out.mData, out_ref.mData, "Error: incorrect data result!");
|
||||
};
|
||||
|
||||
if(!single_result)
|
||||
|
||||
@@ -273,21 +273,21 @@ bool test_reduce_with_index_impl(int init_method,
|
||||
{
|
||||
reduce_util::to_f32_vector(out, out_fp32);
|
||||
reduce_util::to_f32_vector(out_ref, out_ref_fp32);
|
||||
single_result = test_util::check_err(
|
||||
single_result = test::check_err(
|
||||
out_fp32.mData, out_ref_fp32.mData, "Error: incorrect data result!");
|
||||
}
|
||||
else
|
||||
{
|
||||
single_result =
|
||||
test_util::check_err(out.mData, out_ref.mData, "Error: incorrect data result!");
|
||||
test::check_err(out.mData, out_ref.mData, "Error: incorrect data result!");
|
||||
};
|
||||
|
||||
if(NeedIndices)
|
||||
{
|
||||
out_indices_dev.FromDevice(out_indices.mData.data());
|
||||
single_result = single_result && test_util::check_err(out_indices_ref.mData,
|
||||
out_indices.mData,
|
||||
"Error: incorrect index result!");
|
||||
single_result = single_result && test::check_err(out_indices_ref.mData,
|
||||
out_indices.mData,
|
||||
"Error: incorrect index result!");
|
||||
};
|
||||
|
||||
if(!single_result)
|
||||
@@ -370,22 +370,21 @@ bool test_reduce_with_index_impl(int init_method,
|
||||
{
|
||||
reduce_util::to_f32_vector(out, out_fp32);
|
||||
reduce_util::to_f32_vector(out_ref, out_ref_fp32);
|
||||
single_result = test_util::check_err(
|
||||
single_result = test::check_err(
|
||||
out_fp32.mData, out_ref_fp32.mData, "Error: incorrect data result!");
|
||||
}
|
||||
else
|
||||
{
|
||||
single_result =
|
||||
test_util::check_err(out.mData, out_ref.mData, "Error: incorrect data result!");
|
||||
test::check_err(out.mData, out_ref.mData, "Error: incorrect data result!");
|
||||
};
|
||||
|
||||
if(NeedIndices)
|
||||
{
|
||||
out_indices_dev.FromDevice(out_indices.mData.data());
|
||||
single_result =
|
||||
single_result && test_util::check_err(out_indices_ref.mData,
|
||||
out_indices.mData,
|
||||
"Error: incorrect index result!");
|
||||
single_result = single_result && test::check_err(out_indices_ref.mData,
|
||||
out_indices.mData,
|
||||
"Error: incorrect index result!");
|
||||
};
|
||||
|
||||
if(!single_result)
|
||||
|
||||
@@ -23,11 +23,16 @@ template <typename T>
|
||||
struct FillMonotonicSeq
|
||||
{
|
||||
T m_init_value{0};
|
||||
T m_step{1};
|
||||
|
||||
template <typename ForwardIter>
|
||||
void operator()(ForwardIter first, ForwardIter last) const
|
||||
{
|
||||
std::iota(first, last, m_init_value);
|
||||
std::generate(first, last, [=, n = m_init_value]() mutable {
|
||||
auto tmp = n;
|
||||
n += m_step;
|
||||
return tmp;
|
||||
});
|
||||
}
|
||||
};
|
||||
|
||||
@@ -53,7 +58,7 @@ template <ck::index_t NDim,
|
||||
typename FillInputOp = FillMonotonicSeq<InDataType>,
|
||||
typename FillWeightsOp = FillConstant<WeiDataType>>
|
||||
Tensor<OutDataType> RunReferenceConv(const ck::conv_util::ConvParams& params,
|
||||
const FillInputOp& fill_input_op = FillInputOp{0},
|
||||
const FillInputOp& fill_input_op = FillInputOp{},
|
||||
const FillWeightsOp& fill_weights_op = FillWeightsOp{0.5f})
|
||||
{
|
||||
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N),
|
||||
@@ -84,6 +89,9 @@ Tensor<OutDataType> RunReferenceConv(const ck::conv_util::ConvParams& params,
|
||||
fill_weights_op(weights.begin(), weights.end());
|
||||
std::fill(host_output.begin(), host_output.end(), OutDataType(0.f));
|
||||
|
||||
// std::cout <<"input: " << input.mDesc << std::endl << input.mData << std::endl;
|
||||
// std::cout <<"weight: " << weights.mDesc << std::endl << weights.mData << std::endl;
|
||||
|
||||
auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<InDataType,
|
||||
WeiDataType,
|
||||
OutDataType,
|
||||
@@ -104,6 +112,7 @@ Tensor<OutDataType> RunReferenceConv(const ck::conv_util::ConvParams& params,
|
||||
OutElementOp{});
|
||||
|
||||
ref_invoker.Run(ref_argument);
|
||||
// std::cout <<"output: " << host_output.mDesc << std::endl << host_output.mData << std::endl;
|
||||
return host_output;
|
||||
}
|
||||
|
||||
@@ -139,10 +148,10 @@ bool TestConv2DNHWC()
|
||||
472.5,
|
||||
490.5,
|
||||
508.5};
|
||||
res = res && test_util::check_err(out_tensor.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error: wrong output tensor dimensions!");
|
||||
res = res && test_util::check_err(out_tensor.mData, ref_data, "Error: incorrect results!");
|
||||
res = res && test::check_err(out_tensor.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error: wrong output tensor dimensions!");
|
||||
res = res && test::check_err(out_tensor.mData, ref_data, "Error: incorrect results!");
|
||||
|
||||
params.N = 1;
|
||||
params.K = 2;
|
||||
@@ -162,10 +171,10 @@ bool TestConv2DNHWC()
|
||||
747., 747., 1138.5, 1138.5, 1174.5, 1174.5, 1210.5, 1210.5, 1246.5, 1246.5,
|
||||
1035., 1035., 1570.5, 1570.5, 1606.5, 1606.5, 1642.5, 1642.5, 1678.5, 1678.5,
|
||||
1323., 1323., 2002.5, 2002.5, 2038.5, 2038.5, 2074.5, 2074.5, 2110.5, 2110.5};
|
||||
res = res && test_util::check_err(out_tensor.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error: wrong output tensor dimensions!");
|
||||
res = res && test_util::check_err(out_tensor.mData, ref_data, "Error: incorrect results!");
|
||||
res = res && test::check_err(out_tensor.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error: wrong output tensor dimensions!");
|
||||
res = res && test::check_err(out_tensor.mData, ref_data, "Error: incorrect results!");
|
||||
|
||||
return res;
|
||||
}
|
||||
@@ -194,10 +203,10 @@ bool TestConv1DNWC()
|
||||
ck::tensor_layout::convolution::NWK>(params);
|
||||
std::vector<std::size_t> ref_dims{1, 1, 4};
|
||||
std::vector<float> ref_data{7.5, 13.5, 19.5, 25.5};
|
||||
res = res && test_util::check_err(out_tensor.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error: wrong output tensor dimensions!");
|
||||
res = res && test_util::check_err(out_tensor.mData, ref_data, "Error: incorrect results!");
|
||||
res = res && test::check_err(out_tensor.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error: wrong output tensor dimensions!");
|
||||
res = res && test::check_err(out_tensor.mData, ref_data, "Error: incorrect results!");
|
||||
|
||||
params.num_dim_spatial = 1;
|
||||
params.N = 1;
|
||||
@@ -219,10 +228,10 @@ bool TestConv1DNWC()
|
||||
ck::tensor_layout::convolution::NWK>(params);
|
||||
ref_dims = std::vector<std::size_t>{1, 2, 5};
|
||||
ref_data = std::vector<float>{9., 9., 19.5, 19.5, 31.5, 31.5, 43.5, 43.5, 55.5, 55.5};
|
||||
res = res && test_util::check_err(out_tensor.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error: wrong output tensor dimensions!");
|
||||
res = res && test_util::check_err(out_tensor.mData, ref_data, "Error: incorrect results!");
|
||||
res = res && test::check_err(out_tensor.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error: wrong output tensor dimensions!");
|
||||
res = res && test::check_err(out_tensor.mData, ref_data, "Error: incorrect results!");
|
||||
|
||||
params.num_dim_spatial = 1;
|
||||
params.N = 2;
|
||||
@@ -235,16 +244,14 @@ bool TestConv1DNWC()
|
||||
params.input_left_pads = std::vector<ck::index_t>{1};
|
||||
params.input_right_pads = std::vector<ck::index_t>{1};
|
||||
|
||||
auto out_tensor2 =
|
||||
RunReferenceConv<1,
|
||||
float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::convolution::NWC,
|
||||
ck::tensor_layout::convolution::KXC,
|
||||
ck::tensor_layout::convolution::NWK>(params, [](auto first, auto last) {
|
||||
std::generate(first, last, [n = 0]() mutable { return float(n++) * float(0.1f); });
|
||||
});
|
||||
auto out_tensor2 = RunReferenceConv<1,
|
||||
float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::convolution::NWC,
|
||||
ck::tensor_layout::convolution::KXC,
|
||||
ck::tensor_layout::convolution::NWK>(
|
||||
params, FillMonotonicSeq<float>{0.f, 0.1f});
|
||||
|
||||
ref_dims = std::vector<std::size_t>{2, 16, 16};
|
||||
ref_data = std::vector<float>{
|
||||
@@ -312,10 +319,94 @@ bool TestConv1DNWC()
|
||||
72.9, 72.9, 72.9, 72.9, 72.9, 72.9, 72.9, 72.9,
|
||||
49.4, 49.4, 49.4, 49.4, 49.4, 49.4, 49.4, 49.4,
|
||||
49.4, 49.4, 49.4, 49.4, 49.4, 49.4, 49.4, 49.4};
|
||||
res = res && test_util::check_err(out_tensor2.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error: wrong output tensor dimensions!");
|
||||
res = res && test_util::check_err(out_tensor2.mData, ref_data, "Error: incorrect results!");
|
||||
res = res && test::check_err(out_tensor2.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error: wrong output tensor dimensions!");
|
||||
res = res && test::check_err(out_tensor2.mData, ref_data, "Error: incorrect results!");
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
bool TestConv3DNCDHW()
|
||||
{
|
||||
bool res{true};
|
||||
ck::conv_util::ConvParams params;
|
||||
params.num_dim_spatial = 3;
|
||||
params.N = 1;
|
||||
params.K = 1;
|
||||
params.C = 2;
|
||||
params.filter_spatial_lengths = std::vector<ck::index_t>{3, 3, 3};
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{6, 6, 6};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.conv_filter_dilations = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_left_pads = std::vector<ck::index_t>{0, 0, 0};
|
||||
params.input_right_pads = std::vector<ck::index_t>{0, 0, 0};
|
||||
|
||||
auto out_tensor = RunReferenceConv<3,
|
||||
float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::convolution::NCDHW,
|
||||
ck::tensor_layout::convolution::KCZYX,
|
||||
ck::tensor_layout::convolution::NKDHW>(
|
||||
params, FillMonotonicSeq<float>{0.f, 0.1f});
|
||||
std::vector<std::size_t> ref_dims{1, 1, 4, 4, 4};
|
||||
std::vector<float> ref_data{
|
||||
407.7, 410.40002, 413.09998, 415.80002, 423.90002, 426.6, 429.30002, 432.,
|
||||
440.1, 442.80002, 445.5, 448.2, 456.30002, 459., 461.7, 464.40002,
|
||||
504.90002, 507.6, 510.30002, 513., 521.1, 523.8, 526.5, 529.2001,
|
||||
537.3, 540., 542.7001, 545.4, 553.5, 556.2001, 558.9, 561.6,
|
||||
602.10004, 604.8, 607.5, 610.2, 618.3, 621., 623.7, 626.4,
|
||||
634.5, 637.2, 639.9, 642.60004, 650.7, 653.4, 656.10004, 658.8,
|
||||
699.3, 702., 704.7, 707.4, 715.5, 718.2, 720.9, 723.60004,
|
||||
731.7, 734.4001, 737.10004, 739.8, 747.9001, 750.60004, 753.3, 756.};
|
||||
res = res && test::check_err(out_tensor.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error [case 1]: wrong output tensor dimensions!");
|
||||
res = res && test::check_err(out_tensor.mData, ref_data, "Error [case 1]: incorrect results!");
|
||||
|
||||
params.N = 1;
|
||||
params.K = 2;
|
||||
params.C = 2;
|
||||
params.filter_spatial_lengths = std::vector<ck::index_t>{3, 3, 3};
|
||||
params.input_spatial_lengths = std::vector<ck::index_t>{12, 12, 12};
|
||||
params.conv_filter_strides = std::vector<ck::index_t>{3, 3, 3};
|
||||
params.conv_filter_dilations = std::vector<ck::index_t>{1, 1, 1};
|
||||
params.input_left_pads = std::vector<ck::index_t>{0, 0, 0};
|
||||
params.input_right_pads = std::vector<ck::index_t>{0, 0, 0};
|
||||
|
||||
out_tensor = RunReferenceConv<3,
|
||||
float,
|
||||
float,
|
||||
float,
|
||||
ck::tensor_layout::convolution::NCDHW,
|
||||
ck::tensor_layout::convolution::KCZYX,
|
||||
ck::tensor_layout::convolution::NKDHW>(
|
||||
params, FillMonotonicSeq<float>{0.f, 0.1f});
|
||||
ref_dims = std::vector<std::size_t>{1, 2, 4, 4, 4};
|
||||
ref_data = std::vector<float>{
|
||||
2756.7002, 2764.7998, 2772.9001, 2781., 2853.9001, 2862., 2870.1, 2878.2002,
|
||||
2951.1, 2959.2002, 2967.2998, 2975.4001, 3048.2998, 3056.4001, 3064.5, 3072.6,
|
||||
3923.1, 3931.2, 3939.2998, 3947.4, 4020.2998, 4028.4001, 4036.5002, 4044.5999,
|
||||
4117.5, 4125.6, 4133.7, 4141.8, 4214.7, 4222.8, 4230.9004, 4239.,
|
||||
5089.5, 5097.5996, 5105.7, 5113.8, 5186.7, 5194.8, 5202.9, 5211.,
|
||||
5283.9004, 5292., 5300.0996, 5308.2, 5381.0996, 5389.2, 5397.3, 5405.4004,
|
||||
6255.9004, 6264.0005, 6272.1, 6280.2, 6353.1, 6361.2, 6369.301, 6377.4,
|
||||
6450.301, 6458.4, 6466.5, 6474.6, 6547.5, 6555.6, 6563.699, 6571.801,
|
||||
2756.7002, 2764.7998, 2772.9001, 2781., 2853.9001, 2862., 2870.1, 2878.2002,
|
||||
2951.1, 2959.2002, 2967.2998, 2975.4001, 3048.2998, 3056.4001, 3064.5, 3072.6,
|
||||
3923.1, 3931.2, 3939.2998, 3947.4, 4020.2998, 4028.4001, 4036.5002, 4044.5999,
|
||||
4117.5, 4125.6, 4133.7, 4141.8, 4214.7, 4222.8, 4230.9004, 4239.,
|
||||
5089.5, 5097.5996, 5105.7, 5113.8, 5186.7, 5194.8, 5202.9, 5211.,
|
||||
5283.9004, 5292., 5300.0996, 5308.2, 5381.0996, 5389.2, 5397.3, 5405.4004,
|
||||
6255.9004, 6264.0005, 6272.1, 6280.2, 6353.1, 6361.2, 6369.301, 6377.4,
|
||||
6450.301, 6458.4, 6466.5, 6474.6, 6547.5, 6555.6, 6563.699, 6571.801};
|
||||
res = res && test::check_err(out_tensor.mDesc.GetLengths(),
|
||||
ref_dims,
|
||||
"Error [case 2]: wrong output tensor dimensions!");
|
||||
res =
|
||||
res && test::check_err(
|
||||
out_tensor.mData, ref_data, "Error [case 2]: incorrect results!", 1e-4f, 1e-6f);
|
||||
|
||||
return res;
|
||||
}
|
||||
@@ -329,5 +420,7 @@ int main(void)
|
||||
std::cout << "TestConv2DNHWC ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
res = TestConv1DNWC();
|
||||
std::cout << "TestConv1DNHWC ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
res = TestConv3DNCDHW();
|
||||
std::cout << "TestConv3DNCDHW ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
||||
return 0;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user