ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-11 17:00:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

Adding Instances and Examples for FP8-based Scaled Convolution with ReLU Activation and AMAX Reduction. (#1469)

Browse Source 
* Enable CMakePresets build

* Verify Convolution, Scaling and ReLU algorithms.

* Add tensor element-wise scale and type cast operation.

* Reduction implemented but does not work.

* Exploration of Reduction functionality.

* Completed example for Convolution scaled with ReLu activation and AMAX reduction.

* WIP: Add required instances for convolution.

* WIP: Create client example. Implement convolution stage.

* Add elementwise instances.

* Add elementwise scale + convert example.

* Add reduction instances.

* WIP: Client example for AMAX reduction.

* WIP: Add instances for multistage reduction.

* WIP: Implementation of multistage reduction.

* Refactoring.

* Clean up.

* Guard off FP8 instances when the data type is not available.

* Improve output readability.

* Addressing reviewer's comments.
This commit is contained in:
Andriy Roshchenko
2024-08-20 09:30:56 -06:00
committed by GitHub
parent f48529b511
commit a94113a941
17 changed files with 1891 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_outelementop_instance.hpp
View File

@@ -184,6 +184,43 @@ using device_grouped_conv_fwd_xdl_outelementop_bf8_f8_instances = std::tuple<
// clang-format on
>;
template <index_t NDimSpatial,
typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
ConvolutionForwardSpecialization ConvSpec,
typename OutElementOp>
using device_grouped_conv_fwd_xdl_outelementop_f8_f8_f32_instances = std::tuple<
// clang-format off
//########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| Compute| Compute|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| TypeA| TypeB|
//########################################| | | | | | | | | | | | Operation| Operation| Operation| | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| | |
//########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
#ifdef CK_ENABLE_FP8
// generic instance
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, F8, F8>,
// instances for small conv.K and conv.C
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, F8, F8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F8, F8, F32, F32, Tuple<>, F32, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, F8, F8>
#endif
// clang-format on
>;
} // namespace instance
} // namespace device
} // namespace tensor_operation

84
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_convscale_relu.hpp
View File

@@ -8,6 +8,7 @@
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_abd.hpp"
#include "ck/tensor_operation/gpu/element/combined_element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -99,6 +100,89 @@ struct DeviceOperationInstanceFactory<
}
};
namespace ew = ck::tensor_operation::element_wise;
using CombConvScaleRelu = ew::UnaryCombinedOp<ew::Scale, ew::Scale, ew::Relu>;
#ifdef CK_ENABLE_FP8
void add_device_grouped_conv3d_fwd_xdl_combconvscale_relu_ndhwgc_gkzyxc_ndhwgk_f8_f8_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
F8,
F8,
ck::Tuple<>,
F32,
PassThrough,
PassThrough,
CombConvScaleRelu,
F8,
F8>>>& instances);
#endif
template <ck::index_t NumDimSpatial,
typename InLayout,
typename WeiLayout,
typename DLayouts,
typename OutLayout,
typename InDataType,
typename WeiDataType,
typename DDataTypes,
typename OutDataType,
typename AComputeType,
typename BComputeType>
struct DeviceOperationInstanceFactory<
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
InLayout,
WeiLayout,
DLayouts,
OutLayout,
InDataType,
WeiDataType,
DDataTypes,
OutDataType,
PassThrough,
PassThrough,
CombConvScaleRelu,
AComputeType,
BComputeType>>
{
using DeviceOp = DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
InLayout,
WeiLayout,
DLayouts,
OutLayout,
InDataType,
WeiDataType,
DDataTypes,
OutDataType,
PassThrough,
PassThrough,
CombConvScaleRelu,
AComputeType,
BComputeType>;
static auto GetInstances()
{
std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
{
#ifdef CK_ENABLE_FP8
if constexpr(is_same_v<InDataType, f8_t> && is_same_v<WeiDataType, f8_t> &&
is_same_v<OutDataType, F32> && is_same_v<AComputeType, f8_t> &&
is_same_v<BComputeType, f8_t>)
{
add_device_grouped_conv3d_fwd_xdl_combconvscale_relu_ndhwgc_gkzyxc_ndhwgk_f8_f8_f32_instances(
op_ptrs);
}
#endif
}
return op_ptrs;
}
};
} // namespace instance
} // namespace device
} // namespace tensor_operation

13
library/include/ck/library/tensor_operation_instance/gpu/permute_scale.hpp
View File

@@ -70,6 +70,12 @@ void add_device_permute_scale_6d_f32_instances(
DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, element_wise::Scale, 6>>>&);
#endif
#ifdef CK_ENABLE_FP8
void add_device_permute_scale_6d_f32_f8_instances(
std::vector<std::unique_ptr<
DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F8>, element_wise::Scale, 6>>>&);
#endif
template <typename InDataTypeTuple,
typename OutDataTypeTuple,
typename ElementwiseOperation,
@@ -184,6 +190,13 @@ struct DeviceOperationInstanceFactory<
{
add_device_permute_scale_6d_f16_instances(op_ptrs);
}
#endif
#ifdef CK_ENABLE_FP8
if constexpr(is_same_v<InDataTypeTuple, ck::Tuple<F32>> &&
is_same_v<OutDataTypeTuple, ck::Tuple<F8>>)
{
add_device_permute_scale_6d_f32_f8_instances(op_ptrs);
}
#endif
}
return op_ptrs;

46
library/include/ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp
View File

@@ -10,6 +10,7 @@ namespace tensor_operation {
namespace device {
namespace instance {
using F8 = ck::f8_t;
using F16 = ck::half_t;
using F32 = float;
@@ -183,6 +184,51 @@ using device_permute_scale_f32_instances = std::tuple<
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F32>, ElementwiseOp, NDims, 32, 32, 16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F32>, ElementwiseOp, NDims, 32, 16, 32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>
>;
#ifdef CK_ENABLE_FP8
template <index_t NDims,
typename ElementwiseOp>
using device_permute_scale_f32_f8_instances = std::tuple<
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 256, 64, 64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 256, 128, 32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 256, 32, 128, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 64, 32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 32, 64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 16, 128, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 128, 16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 64, 32, 32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 64, 16, 64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 64, 64, 16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 32, 32, 16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 32, 16, 32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<4>, ck::Sequence<4>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 256, 128, 128, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 256, 256, 64, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 256, 64, 256, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 128, 64, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 64, 128, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 32, 256, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 256, 32, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 64, 64, 64, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 64, 32, 128, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 64, 128, 32, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 32, 64, 32, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 32, 32, 64, 8, 8, ck::Sequence<1, 0>, ck::Sequence<8>, ck::Sequence<8>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 256, 64, 64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 256, 128, 32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 256, 32, 128, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 64, 32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 32, 64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 16, 128, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 128, 128, 16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 64, 32, 32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 64, 16, 64, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 64, 64, 16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 32, 32, 16, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>,
DeviceElementwiseImpl<ck::Tuple<F32>, ck::Tuple<F8>, ElementwiseOp, NDims, 32, 16, 32, 4, 4, ck::Sequence<1, 0>, ck::Sequence<1>, ck::Sequence<1>>
>;
#endif
// clang-format on
} // namespace instance

27
library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.hpp
View File

@@ -14,15 +14,24 @@ namespace device {
namespace instance {
// clang-format off
// InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex
extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
// InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 6, 6, ReduceAMax, UnaryAbs, PassThrough, true, false>(std::vector<DeviceReducePtr<F32, F32, F32, 6, 6, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 5, 5, ReduceAMax, UnaryAbs, PassThrough, true, false>(std::vector<DeviceReducePtr<F32, F32, F32, 5, 5, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, true, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 6, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>(std::vector<DeviceReducePtr<F32, F32, F32, 6, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 5, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>(std::vector<DeviceReducePtr<F32, F32, F32, 5, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, true, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 3, 3, ReduceAMax, PassThrough, PassThrough, true, false>(std::vector<DeviceReducePtr<F32, F32, F32, 3, 3, ReduceAMax, PassThrough, PassThrough, true, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 2, 2, ReduceAMax, PassThrough, PassThrough, true, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 2, ReduceAMax, PassThrough, PassThrough, true, false>>&);
extern template void add_device_reduce_instance_blockwise< F32, F32, F32, 1, 1, ReduceAMax, PassThrough, PassThrough, true, false>(std::vector<DeviceReducePtr<F32, F32, F32, 1, 1, ReduceAMax, PassThrough, PassThrough, true, false>>&);
// clang-format on
} // namespace instance