Skip to content

Refactor Storage system to use duck-typed allocators #1046

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
cliffburdick merged 15 commits into from
Sep 24, 2025
Merged

Refactor Storage system to use duck-typed allocators #1046

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
97ca646
Refactor Storage system to use duck-typed allocators
cliffburdick Aug 27, 2025
1d35822
Missing unit test
cliffburdick Aug 27, 2025
cd79f87
Block doxygen section
cliffburdick Aug 27, 2025
9edf5d7
Improved interface
cliffburdick Aug 27, 2025
ed1a788
Improved interface
cliffburdick Aug 27, 2025
6e32f04
Sparse fixes
cliffburdick Aug 28, 2025
9ce02c4
Sparse fixes
cliffburdick Aug 28, 2025
8d624f5
Sparse fixes
cliffburdick Aug 28, 2025
4f31e09
Fixes on sparse
cliffburdick Aug 29, 2025
aae829b
Size fix
cliffburdick Sep 8, 2025
fc53a65
Fixed size
cliffburdick Sep 8, 2025
79459a3
Storage tests
cliffburdick Sep 15, 2025
77fb588
Remove debug
cliffburdick Sep 15, 2025
8ca0783
Fixing ownership
cliffburdick Sep 22, 2025
88e57b5
Fixed sparse allocations
cliffburdick Sep 23, 2025
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
9 changes: 7 additions & 2 deletions docs_input/api/creation/tensors/make.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,15 +22,20 @@ Return by Value
.. doxygenfunction:: make_tensor( T *data, ShapeType &&shape, bool owning = false)
.. doxygenfunction:: make_tensor( TensorType &tensor, typename TensorType::value_type *data, typename TensorType::shape_container &&shape)
.. doxygenfunction:: make_tensor( TensorType &tensor, typename TensorType::value_type *ptr)
.. doxygenfunction:: make_tensor( Storage &&s, ShapeType &&shape)
.. doxygenfunction:: make_tensor( TensorType &tensor, typename TensorType::storage_type &&s, typename TensorType::shape_container &&shape)
.. doxygenfunction:: make_tensor( T* const data, D &&desc, bool owning = false)
.. doxygenfunction:: make_tensor( TensorType &tensor, typename TensorType::value_type* const data, typename TensorType::desc_type &&desc)
.. doxygenfunction:: make_tensor( D &&desc, matxMemorySpace_t space = MATX_MANAGED_MEMORY, cudaStream_t stream = 0)
.. doxygenfunction:: make_tensor( TensorType &&tensor, typename TensorType::desc_type &&desc, matxMemorySpace_t space = MATX_MANAGED_MEMORY, cudaStream_t stream = 0)
.. doxygenfunction:: make_tensor( T *const data, const index_t (&shape)[RANK], const index_t (&strides)[RANK], bool owning = false)
.. doxygenfunction:: make_tensor( TensorType &tensor, typename TensorType::value_type *const data, const index_t (&shape)[TensorType::Rank()], const index_t (&strides)[TensorType::Rank()])

Custom Allocator Support
~~~~~~~~~~~~~~~~~~~~~~~~
.. doxygenfunction:: make_tensor( const index_t (&shape)[RANK], Allocator&& alloc)
.. doxygenfunction:: make_tensor( ShapeType &&shape, Allocator&& alloc)
.. doxygenfunction:: make_tensor( TensorType &tensor, const index_t (&shape)[TensorType::Rank()], Allocator&& alloc)
.. doxygenfunction:: make_tensor( TensorType &tensor, ShapeType &&shape, Allocator&& alloc)

Return by Pointer
~~~~~~~~~~~~~~~~~
.. doxygenfunction:: make_tensor_p( const index_t (&shape)[RANK], matxMemorySpace_t space = MATX_MANAGED_MEMORY, cudaStream_t stream = 0)
Expand Down
100 changes: 50 additions & 50 deletions include/matx/core/make_sparse_tensor.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,6 +49,25 @@ __MATX_INLINE__ static void setZero(T *ptr, index_t sz,
}
}

// Helper to create a Storage<T> with zeros
template <typename T>
__MATX_INLINE__ Storage<T> makeZeroStorage(index_t sz, matxMemorySpace_t space) {
assert(sz > 0);
Storage<T> storage(sz, space);
if (storage.data() != nullptr) {
setZero(storage.data(), sz, space);
}
return storage;
}

// Helper to create an empty Storage<T>
template <typename T>
__MATX_INLINE__ Storage<T> makeEmptyStorage() {
return Storage<T>();
}



// Helper method to fill memory.
template <typename T>
__MATX_INLINE__ static void setVal(T *ptr, T val, matxMemorySpace_t space) {
Expand All @@ -59,25 +78,6 @@ __MATX_INLINE__ static void setVal(T *ptr, T val, matxMemorySpace_t space) {
}
}

// Helper method to create zero storage.
template <typename T>
__MATX_INLINE__ static auto
makeDefaultNonOwningZeroStorage(index_t sz, matxMemorySpace_t space) {
T *ptr = nullptr;
assert(sz > 0);
matxAlloc((void **)&ptr, sz * sizeof(T), space, /*stream=*/0);
setZero(ptr, sz, space);
raw_pointer_buffer<T, matx_allocator<T>> buf{ptr, sz * sizeof(T),
/*owning=*/false};
return basic_storage<decltype(buf)>{std::move(buf)};
}

// Helper method to create empty storage.
template <typename T>
__MATX_INLINE__ static auto makeDefaultNonOwningEmptyStorage() {
raw_pointer_buffer<T, matx_allocator<T>> buf{nullptr, 0, /*owning=*/false};
return basic_storage<decltype(buf)>{std::move(buf)};
}

//
// MatX implements a universal sparse tensor type that uses a tensor format
Expand Down Expand Up @@ -112,24 +112,24 @@ auto make_tensor_coo(ValTensor &val, CrdTensor &row, CrdTensor &col,
// compression should set up pos[0] = {0, nse}. This is done
// here, using the same memory space as the other data.
matxMemorySpace_t space = GetPointerKind(val.GetStorage().data());
auto tp = makeDefaultNonOwningZeroStorage<POS>(2, space);
Storage<POS> tp = makeZeroStorage<POS>(2, space);
setVal(tp.data() + 1, static_cast<POS>(val.Size(0)), space);
// Construct COO.
return sparse_tensor_t<VAL, CRD, POS, COO>(
shape, val.GetStorage(), {row.GetStorage(), col.GetStorage()},
{tp, makeDefaultNonOwningEmptyStorage<POS>()});
{std::move(tp), makeEmptyStorage<POS>()});
}

// Constructs a zero sparse matrix in COO format (viz. nse=0).
template <typename VAL, typename CRD, typename POS = index_t>
auto make_zero_tensor_coo(const index_t (&shape)[2],
matxMemorySpace_t space = MATX_MANAGED_MEMORY) {
return sparse_tensor_t<VAL, CRD, POS, COO>(
shape, makeDefaultNonOwningEmptyStorage<VAL>(),
{makeDefaultNonOwningEmptyStorage<CRD>(),
makeDefaultNonOwningEmptyStorage<CRD>()},
{makeDefaultNonOwningZeroStorage<POS>(2, space),
makeDefaultNonOwningEmptyStorage<POS>()});
shape, makeEmptyStorage<VAL>(),
{makeEmptyStorage<CRD>(),
makeEmptyStorage<CRD>()},
{makeZeroStorage<POS>(2, space),
makeEmptyStorage<POS>()});
}

// Constructs a sparse matrix in CSR format directly from the values, the
Expand All @@ -153,20 +153,20 @@ auto make_tensor_csr(ValTensor &val, PosTensor &rowp, CrdTensor &col,
// Construct CSR.
return sparse_tensor_t<VAL, CRD, POS, CSR>(
shape, val.GetStorage(),
{makeDefaultNonOwningEmptyStorage<CRD>(), col.GetStorage()},
{makeDefaultNonOwningEmptyStorage<POS>(), rowp.GetStorage()});
{makeEmptyStorage<CRD>(), col.GetStorage()},
{makeEmptyStorage<POS>(), rowp.GetStorage()});
}

// Constructs a zero sparse matrix in CSR format (viz. nse=0).
template <typename VAL, typename CRD, typename POS>
auto make_zero_tensor_csr(const index_t (&shape)[2],
matxMemorySpace_t space = MATX_MANAGED_MEMORY) {
return sparse_tensor_t<VAL, CRD, POS, CSR>(
shape, makeDefaultNonOwningEmptyStorage<VAL>(),
{makeDefaultNonOwningEmptyStorage<CRD>(),
makeDefaultNonOwningEmptyStorage<CRD>()},
{makeDefaultNonOwningEmptyStorage<POS>(),
makeDefaultNonOwningZeroStorage<POS>(shape[0] + 1, space)});
shape, makeEmptyStorage<VAL>(),
{makeEmptyStorage<CRD>(),
makeEmptyStorage<CRD>()},
{makeEmptyStorage<POS>(),
makeZeroStorage<POS>(shape[0] + 1, space)});
}

// Constructs a sparse matrix in CSC format directly from the values, the
Expand All @@ -190,20 +190,20 @@ auto make_tensor_csc(ValTensor &val, PosTensor &colp, CrdTensor &row,
// Construct CSC.
return sparse_tensor_t<VAL, CRD, POS, CSC>(
shape, val.GetStorage(),
{makeDefaultNonOwningEmptyStorage<CRD>(), row.GetStorage()},
{makeDefaultNonOwningEmptyStorage<POS>(), colp.GetStorage()});
{makeEmptyStorage<CRD>(), row.GetStorage()},
{makeEmptyStorage<POS>(), colp.GetStorage()});
}

// Constructs a zero sparse matrix in CSC format (viz. nse=0).
template <typename VAL, typename CRD, typename POS>
auto make_zero_tensor_csc(const index_t (&shape)[2],
matxMemorySpace_t space = MATX_MANAGED_MEMORY) {
return sparse_tensor_t<VAL, CRD, POS, CSC>(
shape, makeDefaultNonOwningEmptyStorage<VAL>(),
{makeDefaultNonOwningEmptyStorage<CRD>(),
makeDefaultNonOwningEmptyStorage<CRD>()},
{makeDefaultNonOwningEmptyStorage<POS>(),
makeDefaultNonOwningZeroStorage<POS>(shape[1] + 1, space)});
shape, makeEmptyStorage<VAL>(),
{makeEmptyStorage<CRD>(),
makeEmptyStorage<CRD>()},
{makeEmptyStorage<POS>(),
makeZeroStorage<POS>(shape[1] + 1, space)});
}

// Constructs a sparse matrix in DIA format directly from the values and the
Expand Down Expand Up @@ -235,14 +235,14 @@ auto make_tensor_dia(ValTensor &val, CrdTensor &off,
// compression should set up pos[0] = {0, #diags}. This is done
// here, using the same memory space as the other data.
matxMemorySpace_t space = GetPointerKind(val.GetStorage().data());
auto tp = makeDefaultNonOwningZeroStorage<POS>(2, space);
Storage<POS> tp = makeZeroStorage<POS>(2, space);
setVal(tp.data() + 1, static_cast<POS>(off.Size(0)), space);
// Construct DIA-I/J.
using DIA = std::conditional_t<std::is_same_v<IDX, DIA_INDEX_I>, DIAI, DIAJ>;
return sparse_tensor_t<VAL, CRD, POS, DIA>(
shape, val.GetStorage(),
{off.GetStorage(), makeDefaultNonOwningEmptyStorage<CRD>()},
{tp, makeDefaultNonOwningEmptyStorage<POS>()});
{off.GetStorage(), makeEmptyStorage<CRD>()},
{std::move(tp), makeEmptyStorage<POS>()});
}

// Constructs a sparse tensor in uniform batched DIA format directly from
Expand Down Expand Up @@ -275,15 +275,15 @@ auto make_tensor_uniform_batched_dia(ValTensor &val, CrdTensor &off,
// compression should set up pos[0] = {0, #diags}. This is done
// here, using the same memory space as the other data.
matxMemorySpace_t space = GetPointerKind(val.GetStorage().data());
auto tp = makeDefaultNonOwningZeroStorage<POS>(2, space);
Storage<POS> tp = makeZeroStorage<POS>(2, space);
setVal(tp.data() + 1, static_cast<POS>(off.Size(0)), space);
// Construct Batched DIA-I/J.
using DIA = std::conditional_t<std::is_same_v<IDX, DIA_INDEX_I>,
BatchedDIAIUniform, BatchedDIAJUniform>;
return sparse_tensor_t<VAL, CRD, POS, DIA>(
shape, val.GetStorage(),
{off.GetStorage(), makeDefaultNonOwningEmptyStorage<CRD>()},
{tp, makeDefaultNonOwningEmptyStorage<POS>()});
{off.GetStorage(), makeEmptyStorage<CRD>()},
{std::move(tp), makeEmptyStorage<POS>()});
}

// Convenience constructor for uniform batched tri-diagonal storage.
Expand All @@ -307,21 +307,21 @@ auto make_tensor_uniform_batched_tri_dia(ValTensor &val,
}
// Construct the off = { -1, 0, +1 } in values memory space.
matxMemorySpace_t space = GetPointerKind(val.GetStorage().data());
auto off = makeDefaultNonOwningZeroStorage<CRD>(3, space);
Storage<CRD> off = makeZeroStorage<CRD>(3, space);
setVal(off.data() + 0, static_cast<CRD>(-1), space);
setVal(off.data() + 2, static_cast<CRD>(+1), space);
// Note that the DIA API typically does not involve positions.
// However, under the formal DSL specifications, the top level
// compression should set up pos[0] = {0, #diags}. This is done
// here, using the same memory space as the other data.
auto tp = makeDefaultNonOwningZeroStorage<POS>(2, space);
Storage<POS> tp = makeZeroStorage<POS>(2, space);
setVal(tp.data() + 1, static_cast<POS>(3), space);
// Construct Batched DIA-I/J.
using DIA = std::conditional_t<std::is_same_v<IDX, DIA_INDEX_I>,
BatchedDIAIUniform, BatchedDIAJUniform>;
return sparse_tensor_t<VAL, CRD, POS, DIA>(
shape, val.GetStorage(), {off, makeDefaultNonOwningEmptyStorage<CRD>()},
{tp, makeDefaultNonOwningEmptyStorage<POS>()});
shape, val.GetStorage(), {std::move(off), makeEmptyStorage<CRD>()},
{std::move(tp), makeEmptyStorage<POS>()});
}

} // namespace experimental
Expand Down
Loading