batchlas::ops::syrk Namespace Reference

Functions

candidates()

template<class T>

auto batchlas::ops::syrk::candidates ( )

constexpr

Every compiled choice in tie-break order (§6.3); triangular is float only (R7).

Definition at line 25 of file choice.hh.

form_of()

std::string_view batchlas::ops::syrk::form_of ( std::int64_t a, std::int64_t b )

inline

Definition at line 38 of file choice.hh.

triangular_groups()

std::int64_t batchlas::ops::syrk::triangular_groups ( std::int64_t n, std::int64_t tile )

inlineconstexpr

Definition at line 57 of file choice.hh.

gram_tile()

int batchlas::ops::syrk::gram_tile ( std::int64_t n )

inlineconstexpr

Definition at line 63 of file choice.hh.

gram_threads()

int batchlas::ops::syrk::gram_threads ( std::int64_t n )

inlineconstexpr

Definition at line 64 of file choice.hh.

gram_slm_bytes()

template<class T>

std::int64_t batchlas::ops::syrk::gram_slm_bytes ( std::int64_t n )

constexpr

Definition at line 69 of file choice.hh.

herk_gram_pinned()

bool batchlas::ops::syrk::herk_gram_pinned ( )

inline

herk's opt-in to the conjugating gram kernel: syrk's pin is exactly gram.

Definition at line 74 of file choice.hh.

Types

SyrkChoice

using batchlas::ops::syrk::SyrkChoice = std::variant<Gram, Triangular, Vendor>

fieldless (tile, leg derived); natives: CUDA GPU

Definition at line 22 of file choice.hh.

Variables

last_resort

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

inlineconstexpr

none runs

Definition at line 30 of file choice.hh.

spec

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

inlineconstexpr

op, vendor library, rules

Definition at line 31 of file choice.hh.

key_names

std::array<std::string_view, 5> batchlas::ops::syrk::key_names

inlineconstexpr

Initial value:

{"form:exact", "trans:exact", "n:log:2", "k:log",
                                                           "batch:log"}

Keys: form (transcribed squareish ratio of n, k) and trans N|T|C exact; n weight 2 (work ~ n^2 k batch); k = op(A)'s inner.

Definition at line 35 of file choice.hh.

grid_n

std::array<int, 30> batchlas::ops::syrk::grid_n

inlineconstexpr

Initial value:

{1,   2,   4,   8,   16,  32,   64,   128,  129,  256,
                                            257, 384, 385, 512, 513, 640,  641,  768,  769,  896,
                                            897, 1024, 1025, 1152, 1153, 1536, 1537, 2176, 2177, 4096}

n axis

Definition at line 44 of file choice.hh.

grid_trans

std::array<std::string_view, 3> batchlas::ops::syrk::grid_trans {"N", "T", "C"}

inlineconstexpr

trans axis of the grid

Definition at line 48 of file choice.hh.

grid_k

std::array<int, 6> batchlas::ops::syrk::grid_k {1, 7, 8, 64, 512, 4096}

inlineconstexpr

k axis of the grid

Definition at line 49 of file choice.hh.

grid_batch

std::array<int, 17> batchlas::ops::syrk::grid_batch {1, 2, 3, 4, 5, 6, 7, 8, 10, 11, 15, 16, 26, 27, 128, 1024, 32768}

inlineconstexpr

Definition at line 50 of file choice.hh.

kMaxGridBatch

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

inlineconstexpr

batch ceiling of both tile kernels

Definition at line 54 of file choice.hh.

kMaxGridTiles

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

inlineconstexpr

triangular's tile-count ceiling

Definition at line 55 of file choice.hh.

kTriangularWg

int batchlas::ops::syrk::kTriangularWg = 256

inlineconstexpr

syrk_triangular_tiles' 16 x 16 work-group

Definition at line 56 of file choice.hh.

Classes
struct	Gram
	syrk_gram_tiles: one tile covers C; n <= 128, tile SLM fits More...
struct	Triangular
	syrk_triangular_tiles<float>: 128-wide tiles of one half More...
struct	Vendor
	backend::syrk_vendor (per-item ?syrk loop) More...