ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 06:49:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Regulate reduction accumulator operations and Element-wise operations (#274)

Browse Source 
* Remove template from Reducton operation classes and add template to their operator() and GetIdentityValue() interfaces

* Change to unary elementwise operators and the reduce_unary_operator (class for mapping) and dependent variations in all host layers

* Remove the data type template parameter from reduce_binary_operator (class for mapping) and dependent variations in host layers

* Add InMemoryDataOperatonSupportedOnDataType to check the matching between data type and InMemoryDataOperation

* Use struct-scope operator template instantiation for binary and unary element-wise operations

* Change a few more elementwise operations to use template for operator()

* Tiny correction in Normalize operator

* Add static_assert to check the data type appliability for some reduction accumulator and element-wise operatons

* Correction in some examples with regard to using ReduceAccDataType

* Use static_assert for UnaryDivide

* Update to merged codes to use Element-wise operations and Reduction Accumulator operations correctly

* Tiny fix with regard to SetWorkSpacePointer()
This commit is contained in:
Qianfeng
2022-06-18 04:10:25 +08:00
committed by GitHub
parent 63cdd92398
commit 1f543bfa79
48 changed files with 891 additions and 837 deletions
Download Patch File Download Diff File Expand all files Collapse all files

49
example/12_reduce/reduce_blockwise.cpp
View File

@@ -33,11 +33,11 @@ constexpr ReduceTensorOp ReduceOpId = ReduceTensorOp::NORM2;
constexpr bool PropagateNan = true;
constexpr bool OutputIndex = false;
using ReduceOperation = typename reduce_binary_operator<AccDataType, ReduceOpId>::opType;
using ReduceOperation = typename reduce_binary_operator<ReduceOpId>::opType;
using InElementwiseOperation =
typename reduce_unary_operator<AccDataType, ReduceOpId, true, true>::InElementwiseOperation;
typename reduce_unary_operator<ReduceOpId, true, true>::InElementwiseOperation;
using AccElementwiseOperation =
typename reduce_unary_operator<AccDataType, ReduceOpId, true, true>::AccElementwiseOperation;
typename reduce_unary_operator<ReduceOpId, true, true>::AccElementwiseOperation;
using DeviceReduceInstance = DeviceReduceMultiBlock<InDataType,
AccDataType,
@@ -247,6 +247,13 @@ int main(int argc, char* argv[])
DeviceMem out_index_dev(indicesSizeInBytes);
InElementwiseOperation in_elementwise_op;
AccElementwiseOperation acc_elementwise_op;
std::tie(in_elementwise_op, acc_elementwise_op) =
reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
static_cast<int32_t>(reduce_total_length));
if(args.do_verification)
{
ReductionHost<InDataType,
@@ -261,8 +268,13 @@ int main(int argc, char* argv[])
OutputIndex>
hostReduce(in.mDesc, out_ref.mDesc, invariantDims, reduceDims);
hostReduce.Run(
alpha, in.mData.data(), beta, out_ref.mData.data(), out_indices_ref.mData.data());
hostReduce.Run(alpha,
in.mData.data(),
beta,
out_ref.mData.data(),
out_indices_ref.mData.data(),
in_elementwise_op,
acc_elementwise_op);
};
std::vector<ck::index_t> i_inLengths;
@@ -277,20 +289,19 @@ int main(int argc, char* argv[])
auto reduce = DeviceReduceInstance{};
auto argument_ptr = reduce.MakeArgumentPointer(
i_inLengths,
i_inStrides,
i_outLengths,
i_outStrides,
reduceDims,
alpha,
beta,
in_dev.GetDeviceBuffer(),
nullptr,
out_dev.GetDeviceBuffer(),
out_index_dev.GetDeviceBuffer(),
InElementwiseOperation{static_cast<int32_t>(reduce_total_length)},
AccElementwiseOperation{static_cast<int32_t>(reduce_total_length)});
auto argument_ptr = reduce.MakeArgumentPointer(i_inLengths,
i_inStrides,
i_outLengths,
i_outStrides,
reduceDims,
alpha,
beta,
in_dev.GetDeviceBuffer(),
nullptr,
out_dev.GetDeviceBuffer(),
out_index_dev.GetDeviceBuffer(),
in_elementwise_op,
acc_elementwise_op);
if(!reduce.IsSupportedArgument(argument_ptr.get()))
{