Skip to content

[MIOpen Downstream] Fix Reduction Kernel #34

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
asroy merged 14 commits into from
Oct 6, 2021
Merged

[MIOpen Downstream] Fix Reduction Kernel #34

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
14 commits
Select commit Hold shift + click to select a range
a18e648
Tiny fix in using data type template parameters in blockwise and dire…
qianfengz Sep 15, 2021
5a9f630
Fix with regard to implementing GetZeroVal() in both kernel and host
qianfengz Sep 15, 2021
eac1753
Avoid convert to compType from dstDataType before writting the output…
qianfengz Sep 15, 2021
f0019df
Add half_t support to NumericLimits and make constexpr GetZeroVal() o…
qianfengz Sep 17, 2021
92e1588
Add CONSTANT decorator for descriptor read buffer
qianfengz Sep 5, 2021
52ae56f
Use get_thread_local_1d_id() for thread local Id
qianfengz Sep 2, 2021
4fea425
Rename GetZeroVal() to GetReductionZeroVal() in the kernels
qianfengz Sep 17, 2021
7a7497f
Remove constexpr from initialized zeroVal and tiny fix in reduction_o…
qianfengz Sep 17, 2021
7218a2b
Occasional tiny simplification and update in the kernel files
qianfengz Sep 18, 2021
2bc1ce0
Update to re-order tensor dimensions on the host, split second_call k…
qianfengz Sep 18, 2021
e030a28
Update to remove OpenCL tidy checking failures
qianfengz Sep 18, 2021
1b9c39b
Update for better readability
qianfengz Sep 24, 2021
c36084f
Remove unused codes and not-needed template parameters in the kernel …
qianfengz Sep 27, 2021
f49d774
Merge remote-tracking branch 'origin/develop' into miopen_downstream-…
Oct 6, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
38 changes: 25 additions & 13 deletions composable_kernel/include/tensor_operation/gridwise_generic_2d_reduction_blockwise.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,7 +92,7 @@ struct GridwiseReduction_xy_to_x_blockwise
// LDS
__shared__ compType p_in_block_buffer[BlockBufferSize];

auto zeroVal = opReduce::GetZeroVal();
const auto zeroVal = opReduce::GetReductionZeroVal();

const auto src_global_buf = make_dynamic_buffer<AddressSpaceEnum_t::Global>(
p_src_global, src2dDesc.GetElementSpaceSize(), type_convert<srcDataType>{}(zeroVal));
Expand Down Expand Up @@ -180,6 +180,10 @@ struct GridwiseReduction_xy_to_x_blockwise
if(!float_equal_one{}(alpha))
accuValue_buf(I0) *= type_convert<compType>{}(alpha);

StaticBuffer<AddressSpaceEnum_t::Vgpr, dstDataType, 1, true> dstValue_buf;

dstValue_buf(I0) = type_convert<dstDataType>{}(accuValue_buf[I0]);

if(!float_equal_zero{}(beta))
{
auto threadwise_dst_load =
Expand All @@ -200,11 +204,11 @@ struct GridwiseReduction_xy_to_x_blockwise
threadwise_dst_load.Run(
dst1dDesc, dst_global_buf, ReducedDataDesc, make_tuple(I0), priorDstValue_buf);

accuValue_buf(I0) += type_convert<compType>{}(priorDstValue_buf[I0] * beta);
dstValue_buf(I0) += priorDstValue_buf[I0] * beta;
}

auto threadwise_dst_store =
ThreadwiseTensorSliceTransfer_v1r3<compType,
ThreadwiseTensorSliceTransfer_v1r3<dstDataType,
dstDataType,
decltype(ReducedDataDesc),
dst1dDescType,
Expand All @@ -218,7 +222,7 @@ struct GridwiseReduction_xy_to_x_blockwise
make_multi_index(block_global_1d_id));

threadwise_dst_store.Run(
ReducedDataDesc, make_tuple(I0), accuValue_buf, dst1dDesc, dst_global_buf);
ReducedDataDesc, make_tuple(I0), dstValue_buf, dst1dDesc, dst_global_buf);
}
};

Expand All @@ -239,7 +243,7 @@ struct GridwiseReduction_xy_to_x_blockwise
__shared__ compType p_in_block_buffer[BlockBufferSize];
__shared__ int block_indices_buffer[BlockBufferSize];

auto zeroVal = opReduce::GetZeroVal();
const auto zeroVal = opReduce::GetReductionZeroVal();

const auto src_global_buf = make_dynamic_buffer<AddressSpaceEnum_t::Global>(
p_src_global, src2dDesc.GetElementSpaceSize(), type_convert<srcDataType>{}(zeroVal));
Expand Down Expand Up @@ -281,7 +285,7 @@ struct GridwiseReduction_xy_to_x_blockwise
ThreadClusterLengths,
Sequence<0, 1>,
srcDataType,
dstDataType,
compType,
src2dDescType,
decltype(in_block_desc),
Sequence<0, 1>,
Expand Down Expand Up @@ -345,6 +349,10 @@ struct GridwiseReduction_xy_to_x_blockwise
if(!float_equal_one{}(alpha))
accuValue_buf(I0) *= type_convert<compType>{}(alpha);

StaticBuffer<AddressSpaceEnum_t::Vgpr, dstDataType, 1, true> dstValue_buf;

dstValue_buf(I0) = type_convert<dstDataType>{}(accuValue_buf[I0]);

if(!float_equal_zero{}(beta))
{
auto threadwise_dst_load =
Expand All @@ -368,11 +376,11 @@ struct GridwiseReduction_xy_to_x_blockwise
make_tuple(I0),
priorDstValue_buf);

accuValue_buf(I0) += type_convert<compType>{}(priorDstValue_buf[I0] * beta);
dstValue_buf(I0) += priorDstValue_buf[I0] * beta;
}

auto threadwise_dst_val_store =
ThreadwiseTensorSliceTransfer_v1r3<compType,
ThreadwiseTensorSliceTransfer_v1r3<dstDataType,
dstDataType,
decltype(ReducedDataDesc),
dst1dDescType,
Expand Down Expand Up @@ -400,7 +408,7 @@ struct GridwiseReduction_xy_to_x_blockwise
make_multi_index(block_global_1d_id));

threadwise_dst_val_store.Run(
ReducedDataDesc, make_tuple(I0), accuValue_buf, dst1dDesc, dst_global_val_buf);
ReducedDataDesc, make_tuple(I0), dstValue_buf, dst1dDesc, dst_global_val_buf);
threadwise_dst_idx_store.Run(
ReducedDataDesc, make_tuple(I0), accuIndex_buf, dst1dDesc, dst_global_idx_buf);
}
Expand All @@ -423,7 +431,7 @@ struct GridwiseReduction_xy_to_x_blockwise
__shared__ compType p_in_block_buffer[BlockBufferSize];
__shared__ int block_indices_buffer[BlockBufferSize];

auto zeroVal = opReduce::GetZeroVal();
const auto zeroVal = opReduce::GetReductionZeroVal();

const auto src_global_val_buf =
make_dynamic_buffer<AddressSpaceEnum_t::Global>(ws_values_global,
Expand Down Expand Up @@ -547,6 +555,10 @@ struct GridwiseReduction_xy_to_x_blockwise
if(!float_equal_one{}(alpha))
accuValue_buf(I0) *= type_convert<compType>{}(alpha);

StaticBuffer<AddressSpaceEnum_t::Vgpr, dstDataType, 1, true> dstValue_buf;

dstValue_buf(I0) = type_convert<dstDataType>{}(accuValue_buf[I0]);

if(!float_equal_zero{}(beta))
{
auto threadwise_dst_load =
Expand All @@ -570,11 +582,11 @@ struct GridwiseReduction_xy_to_x_blockwise
make_tuple(I0),
priorDstValue_buf);

accuValue_buf(I0) += type_convert<compType>{}(priorDstValue_buf[I0] * beta);
dstValue_buf(I0) += priorDstValue_buf[I0] * beta;
}

auto threadwise_dst_val_store =
ThreadwiseTensorSliceTransfer_v1r3<compType,
ThreadwiseTensorSliceTransfer_v1r3<dstDataType,
dstDataType,
decltype(ReducedDataDesc),
dst1dDescType,
Expand Down Expand Up @@ -602,7 +614,7 @@ struct GridwiseReduction_xy_to_x_blockwise
make_multi_index(block_global_1d_id));

threadwise_dst_val_store.Run(
ReducedDataDesc, make_tuple(I0), accuValue_buf, dst1dDesc, dst_global_val_buf);
ReducedDataDesc, make_tuple(I0), dstValue_buf, dst1dDesc, dst_global_val_buf);
threadwise_dst_idx_store.Run(
ReducedDataDesc, make_tuple(I0), accuIndex_buf, dst1dDesc, dst_global_idx_buf);
}
Expand Down
40 changes: 26 additions & 14 deletions ...sable_kernel/include/tensor_operation/gridwise_generic_2d_reduction_direct_threadwise.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -82,7 +82,7 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
(void)ws_indices_global;
(void)indices_global;

const auto zeroVal = opReduce::GetZeroVal();
const auto zeroVal = opReduce::GetReductionZeroVal();

const auto src_global_buf = make_dynamic_buffer<AddressSpaceEnum_t::Global>(
p_src_global, src2dDesc.GetElementSpaceSize(), type_convert<srcDataType>{}(zeroVal));
Expand Down Expand Up @@ -147,6 +147,10 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
if(!float_equal_one{}(alpha))
accuValue_buf(I0) *= type_convert<compType>{}(alpha);

StaticBuffer<AddressSpaceEnum_t::Vgpr, dstDataType, 1, true> dstValue_buf;

dstValue_buf(I0) = type_convert<dstDataType>{}(accuValue_buf[I0]);

if(!float_equal_zero{}(beta))
{
auto threadwise_dst_load = ThreadwiseTensorSliceTransfer_v2<dstDataType,
Expand All @@ -166,11 +170,11 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
threadwise_dst_load.Run(
dst1dDesc, dst_global_buf, ReducedDataDesc, make_tuple(I0), priorDstValue_buf);

accuValue_buf(I0) += type_convert<compType>{}(priorDstValue_buf[I0] * beta);
dstValue_buf(I0) += priorDstValue_buf[I0] * beta;
}

auto threadwise_dst_store =
ThreadwiseTensorSliceTransfer_v1r3<compType,
ThreadwiseTensorSliceTransfer_v1r3<dstDataType,
dstDataType,
decltype(ReducedDataDesc),
dst1dDescType,
Expand All @@ -184,7 +188,7 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
make_multi_index(thread_global_1d_id));

threadwise_dst_store.Run(
ReducedDataDesc, make_tuple(I0), accuValue_buf, dst1dDesc, dst_global_buf);
ReducedDataDesc, make_tuple(I0), dstValue_buf, dst1dDesc, dst_global_buf);
};

template <>
Expand All @@ -200,7 +204,7 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
{
(void)ws_indices_global;

const auto zeroVal = opReduce::GetZeroVal();
const auto zeroVal = opReduce::GetReductionZeroVal();

const auto src_global_buf = make_dynamic_buffer<AddressSpaceEnum_t::Global>(
p_src_global, src2dDesc.GetElementSpaceSize(), type_convert<srcDataType>{}(zeroVal));
Expand Down Expand Up @@ -232,7 +236,7 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
index_t thread_global_1d_id = get_block_1d_id() * BlockSize + get_thread_local_1d_id();

auto threadwise_src_load = ThreadwiseTensorSliceTransfer_v2<srcDataType,
dstDataType,
compType,
src2dDescType,
decltype(ThreadBufferDesc),
ThreadBufferLengths,
Expand Down Expand Up @@ -271,6 +275,10 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
if(!float_equal_one{}(alpha))
accuValue_buf(I0) *= type_convert<compType>{}(alpha);

StaticBuffer<AddressSpaceEnum_t::Vgpr, dstDataType, 1, true> dstValue_buf;

dstValue_buf(I0) = type_convert<dstDataType>{}(accuValue_buf[I0]);

if(!float_equal_zero{}(beta))
{
auto threadwise_dst_load = ThreadwiseTensorSliceTransfer_v2<dstDataType,
Expand All @@ -290,11 +298,11 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
threadwise_dst_load.Run(
dst1dDesc, dst_global_val_buf, ReducedDataDesc, make_tuple(I0), priorDstValue_buf);

accuValue_buf(I0) += type_convert<compType>{}(priorDstValue_buf[I0] * beta);
dstValue_buf(I0) += priorDstValue_buf[I0] * beta;
}

auto threadwise_dst_val_store =
ThreadwiseTensorSliceTransfer_v1r3<compType,
ThreadwiseTensorSliceTransfer_v1r3<dstDataType,
dstDataType,
decltype(ReducedDataDesc),
dst1dDescType,
Expand Down Expand Up @@ -322,7 +330,7 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
make_multi_index(thread_global_1d_id));

threadwise_dst_val_store.Run(
ReducedDataDesc, make_tuple(I0), accuValue_buf, dst1dDesc, dst_global_val_buf);
ReducedDataDesc, make_tuple(I0), dstValue_buf, dst1dDesc, dst_global_val_buf);
threadwise_dst_idx_store.Run(
ReducedDataDesc, make_tuple(I0), accuIndex_buf, dst1dDesc, dst_global_idx_buf);
};
Expand All @@ -340,7 +348,7 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
{
(void)origReduceLen;

const auto zeroVal = opReduce::GetZeroVal();
const auto zeroVal = opReduce::GetReductionZeroVal();

const auto src_global_val_buf =
make_dynamic_buffer<AddressSpaceEnum_t::Global>(ws_values_global,
Expand Down Expand Up @@ -377,7 +385,7 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
index_t thread_global_1d_id = get_block_1d_id() * BlockSize + get_thread_local_1d_id();

auto threadwise_src_val_load = ThreadwiseTensorSliceTransfer_v2<srcDataType,
dstDataType,
compType,
src2dDescType,
decltype(ThreadBufferDesc),
ThreadBufferLengths,
Expand Down Expand Up @@ -430,6 +438,10 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
if(!float_equal_one{}(alpha))
accuValue_buf(I0) *= type_convert<compType>{}(alpha);

StaticBuffer<AddressSpaceEnum_t::Vgpr, dstDataType, 1, true> dstValue_buf;

dstValue_buf(I0) = type_convert<dstDataType>{}(accuValue_buf[I0]);

if(!float_equal_zero{}(beta))
{
auto threadwise_dst_load = ThreadwiseTensorSliceTransfer_v2<dstDataType,
Expand All @@ -449,11 +461,11 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
threadwise_dst_load.Run(
dst1dDesc, dst_global_val_buf, ReducedDataDesc, make_tuple(I0), priorDstValue_buf);

accuValue_buf(I0) += type_convert<compType>{}(priorDstValue_buf[I0] * beta);
dstValue_buf(I0) += priorDstValue_buf[I0] * beta;
}

auto threadwise_dst_val_store =
ThreadwiseTensorSliceTransfer_v1r3<compType,
ThreadwiseTensorSliceTransfer_v1r3<dstDataType,
dstDataType,
decltype(ReducedDataDesc),
dst1dDescType,
Expand Down Expand Up @@ -481,7 +493,7 @@ struct GridwiseReduction_xy_to_x_direct_threadwise
make_multi_index(thread_global_1d_id));

threadwise_dst_val_store.Run(
ReducedDataDesc, make_tuple(I0), accuValue_buf, dst1dDesc, dst_global_val_buf);
ReducedDataDesc, make_tuple(I0), dstValue_buf, dst1dDesc, dst_global_val_buf);
threadwise_dst_idx_store.Run(
ReducedDataDesc, make_tuple(I0), accuIndex_buf, dst1dDesc, dst_global_idx_buf);
};
Expand Down
Loading