`TA::retile` support for `DistArray` with tensor-of-tensors tiles#474

Merged

evaleev merged 50 commits intomasterfrom gaudel/feature/retile_tot_arrays

Oct 29, 2024

+98-13

Commits on Sep 16, 2024

[ci skip][wip] TA::retile support for DistArray with tensor-of-tensor tiles.
bimalgaudel
committed

Commits on Sep 18, 2024

Completes TA::retile support for DistArray with tensor-of-tensor tiles.
bimalgaudel
committed

Commits on Sep 25, 2024

TiledRange1: can construct using a range of tile hashmarks

evaleev
authored and
bimalgaudel
committed
<-> Eigen::{Vector,Matrix,Tensor} conversions can handle DistArrays with non-zero base Ranges

evaleev
authored and
bimalgaudel
committed
[skip ci] typo

evaleev
authored and
bimalgaudel
committed
remove duplicate vlock copy in concat

evaleev
authored and
bimalgaudel
committed
introduced tile_ranges_match_trange(DistArray) for validating tile ranges against trange

evaleev
authored and
bimalgaudel
committed
SizeArray is a viewable range

evaleev
authored and
bimalgaudel
committed
fixed assignment to block expression from an expression with nonzero base

evaleev
authored and
bimalgaudel
committed
[skip ci] to_container.hpp -> range/conversion.hpp

evaleev
authored and
bimalgaudel
committed
[skip ci] svd dox fixup

evaleev
authored and
bimalgaudel
committed
TiledRange1{int x} constructs an empty element range at [x,x)

evaleev
authored and
bimalgaudel
committed
make Range1 printable and shiftable

evaleev
authored and
bimalgaudel
committed
make TiledRange1 shiftable

evaleev
authored and
bimalgaudel
committed
TiledRange1 printer reimplemented in terms of Range1 printer

evaleev
authored and
bimalgaudel
committed
[skip ci] dox++

evaleev
authored and
bimalgaudel
committed
Range::shift is const

evaleev
authored and
bimalgaudel
committed
Range::shift is nodiscard

evaleev
authored and
bimalgaudel
committed
TiledRange is shiftable

evaleev
authored and
bimalgaudel
committed
introduced BlkTsrExpr::{{set_,}trange_lobound,preserve_lobound}() that allow to use block tensor expressions even with DistArrays that have non-zero lobound

evaleev
authored and
bimalgaudel
committed
dox fixup [skip ci]

evaleev
authored and
bimalgaudel
committed
introduced TiledRange::is_congruent

evaleev
authored and
bimalgaudel
committed
like contraction, reduction expression and binary expression can ignore absolute positions if ignore_tile_position() is on

evaleev
authored and
bimalgaudel
committed
introduced TiledRange1::{lo,up}bound which feel to have unambiguous meaning

evaleev
authored and
bimalgaudel
committed
more TsrExpr::block variants tagged by preserve_lobound_t

evaleev
authored and
bimalgaudel
committed
TA::expressions::preserve_lobound_t -> TA::preserve_lobound_t

evaleev
authored and
bimalgaudel
committed
btas <-> ta tensor conversions work for non-0-lobound

evaleev
authored and
bimalgaudel
committed
introduced member versions of TiledRange1::make_uniform

evaleev
authored and
bimalgaudel
committed
TiledRange1::make_uniform(rng) with empty range preserves its lobound

evaleev
authored and
bimalgaudel
committed
heig: work around the n=0 corner case

evaleev
authored and
bimalgaudel
committed
bump MAD tag to pull in m-a-d-n-e-s-s/madness#547

evaleev
authored and
bimalgaudel
committed
BinaryExpr: account for ignore_tile_position when checking preconditions

evaleev
authored and
bimalgaudel
committed
[python] simplify make_trange by using TiledRange1::make_uniform

evaleev
authored and
bimalgaudel
committed
[ci] greatly reduce the gitlab matrix, replace rel/deb builds with relwithdebinfo

evaleev
authored and
bimalgaudel
committed
TA::host_allocator is serializable, so that btas::Tensor can be used as a tile again

evaleev
authored and
bimalgaudel
committed
introduced TA::Tile::at_ordinal + strengthen disambiguation checks for potential at_ordinal uses

evaleev
authored and
bimalgaudel
committed
hostEnv -> host::Env + fixup to make f294db3 build

evaleev
authored and
bimalgaudel
committed
[ci] do not use gcc toolchain on macos, instead try linux + enable ccache

evaleev
authored and
bimalgaudel
committed
[ci] try symlinking libscalapack-openmpi.so to libscalapack.so to help out FindReferenceSCALAPACK

evaleev
authored and
bimalgaudel
committed
[unit] [cuda] another disambiguation via at_ordinal

evaleev
authored and
bimalgaudel
committed
[ci] disable Gitlab jobs except CUDA

evaleev
authored and
bimalgaudel
committed
introduce {Tensor,Tile}::c{begin,end}

evaleev
authored and
bimalgaudel
committed
TensorInterface is a range

evaleev
authored and
bimalgaudel
committed
dox fixup

evaleev
authored and
bimalgaudel
committed
btas::Tensor can be copied into from Tensor and TensorInterface ... this allows TA::retile on DistArrays of btas::Tensors

evaleev
authored and
bimalgaudel
committed
[unit] btas_suite/tensor_ctor: initialize ta_tensor properly

evaleev
authored and
bimalgaudel
committed
fixup TensorInterface::Iterator::deference()

evaleev
authored and
bimalgaudel
committed
pull in ValeevGroup/BTAS#179

evaleev
authored and
bimalgaudel
committed

Commits on Oct 15, 2024