ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-11 08:50:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Make buffer coherence configurable in tensor view (#2041)

Browse Source 
* Make buffer coherence configurable in tensor view

* Fix clang-format for tensor_view.hpp
This commit is contained in:
MHYang-gh
2025-04-09 06:34:11 +08:00
committed by GitHub
parent 2c563fecf7
commit 03ce8729fd
1 changed files with 9 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
include/ck_tile/core/tensor/tensor_view.hpp
View File

@@ -411,18 +411,21 @@ struct null_tensor_view
};
template <address_space_enum BufferAddressSpace = address_space_enum::generic,
amd_buffer_coherence_enum Coherence = amd_buffer_coherence_enum::coherence_default,
typename DataType,
typename... Ts>
CK_TILE_HOST_DEVICE constexpr auto make_tensor_view(DataType* p,
const tensor_descriptor<Ts...>& desc)
{
auto buffer_view = make_buffer_view<BufferAddressSpace>(p, desc.get_element_space_size());
auto buffer_view =
make_buffer_view<BufferAddressSpace, Coherence>(p, desc.get_element_space_size());
return tensor_view<decltype(buffer_view), decltype(desc)>{buffer_view, desc};
}
template <address_space_enum BufferAddressSpace = address_space_enum::generic,
memory_operation_enum DstInMemOp = memory_operation_enum::set,
amd_buffer_coherence_enum Coherence = amd_buffer_coherence_enum::coherence_default,
typename DataType,
typename... Lengths,
typename... Strides,
@@ -441,12 +444,14 @@ make_naive_tensor_view(DataType* p,
number<GuaranteedLastDimensionVectorLength>{},
number<GuaranteedLastDimensionVectorStride>{});
auto buffer_view = make_buffer_view<BufferAddressSpace>(p, desc.get_element_space_size());
auto buffer_view =
make_buffer_view<BufferAddressSpace, Coherence>(p, desc.get_element_space_size());
return tensor_view<decltype(buffer_view), decltype(desc), DstInMemOp>{buffer_view, desc};
}
template <address_space_enum BufferAddressSpace = address_space_enum::generic,
amd_buffer_coherence_enum Coherence = amd_buffer_coherence_enum::coherence_default,
typename DataType,
typename... Lengths,
index_t GuaranteedLastDimensionVectorLength = -1>
@@ -458,7 +463,8 @@ make_naive_tensor_view_packed(DataType* p,
auto desc =
make_naive_tensor_descriptor_packed(lengths, number<GuaranteedLastDimensionVectorLength>{});
auto buffer_view = make_buffer_view<BufferAddressSpace>(p, desc.get_element_space_size());
auto buffer_view =
make_buffer_view<BufferAddressSpace, Coherence>(p, desc.get_element_space_size());
return tensor_view<decltype(buffer_view), decltype(desc)>{buffer_view, desc};
}