batchlas::ops::trmm Namespace Reference

Functions

candidates()

template<class T>

auto batchlas::ops::trmm::candidates ( )

constexpr

Every compiled choice, in tie-break order (§6.3): triangular, expand, vendor.

Definition at line 22 of file choice.hh.

Types

TrmmChoice

using batchlas::ops::trmm::TrmmChoice = std::variant<Triangular, Expand, Vendor>

row tile (trmm_row_tile) derived; natives: CUDA GPU

Definition at line 19 of file choice.hh.

Variables

last_resort

std::array<std::string_view, 3> batchlas::ops::trmm::last_resort {"expand", "triangular", "vendor"}

inlineconstexpr

nothing runs

Definition at line 24 of file choice.hh.

spec

select::OpSpec batchlas::ops::trmm::spec {Op::trmm, select::Lib::level3, {last_resort}}

inlineconstexpr

op, vendor library, rules

Definition at line 26 of file choice.hh.

kMaxGridBatch

std::int64_t batchlas::ops::trmm::kMaxGridBatch = 65535

inlineconstexpr

both natives: batch in grid z; 65536 throws

Definition at line 28 of file choice.hh.

kMaxGridTiles

std::int64_t batchlas::ops::trmm::kMaxGridTiles = 65535

inlineconstexpr

triangular: row x column tiles in grid y

Definition at line 29 of file choice.hh.

kNativeWg

int batchlas::ops::trmm::kNativeWg = 256

inlineconstexpr

widest native work-group: the 128-row tile, the expansion

Definition at line 30 of file choice.hh.

key_names

std::array<std::string_view, 4> batchlas::ops::trmm::key_names {"side:exact", "order:log:2", "q:log", "batch:log"}

inlineconstexpr

Keys: side exact; order = A.rows(), weight 2 (work ~ order^2 q batch); q = B.cols() (Left) or B.rows() (Right).

Definition at line 33 of file choice.hh.

grid_order

std::array<int, 9> batchlas::ops::trmm::grid_order {1, 16, 32, 64, 128, 256, 512, 1024, 2048}

inlineconstexpr

order axis

Definition at line 36 of file choice.hh.

grid_q

std::array<int, 5> batchlas::ops::trmm::grid_q {1, 16, 128, 1024, 4096}

inlineconstexpr

q axis of the grid

Definition at line 37 of file choice.hh.

grid_batch

std::array<int, 4> batchlas::ops::trmm::grid_batch {1, 128, 1024, 32768}

inlineconstexpr

batch axis of the grid

Definition at line 38 of file choice.hh.

Classes
struct	Triangular
	trmm_triangular_tiles: Side::Left; tile grid <= 65535 More...
struct	Expand
	expand_triangular + the public gemm; scratch must fit More...
struct	Vendor
	backend::trmm_vendor (cublas?trmm loop) More...