batchlas::ops::potrf Namespace Reference

Functions

candidates()

template<class T>

auto batchlas::ops::potrf::candidates ( )

constexpr

Every compiled choice, once, in tie-break order (§6.3): simpler first, vendor last.

Definition at line 34 of file choice.hh.

Types

PotrfChoice

using batchlas::ops::potrf::PotrfChoice = std::variant<Tiny, Cta, Lpanel, Blocked, Vendor>

natives need a GPU, sub-group 32, n >= 1

Definition at line 30 of file choice.hh.

Variables

spec

select::OpSpec batchlas::ops::potrf::spec {Op::potrf, select::Lib::solver}

inlineconstexpr

default last resort: blocked, vendor

Definition at line 41 of file choice.hh.

key_names

std::array<std::string_view, 3> batchlas::ops::potrf::key_names {"uplo:exact", "n:log:3", "batch:log"}

inlineconstexpr

Table keys: uplo exact (L|U); n log-distance, weight 3 (work ~ n^3); batch log, weight 1.

evidence: docs/design/flat-kernel-selection.md (§5.4)

Definition at line 45 of file choice.hh.

grid_n

std::array<int, 34> batchlas::ops::potrf::grid_n

inlineconstexpr

Initial value:

{1, 2, 3, 4, 6, 8, 12, 16, 20, 24, 28, 32, 36, 40, 48,
    56, 64, 80, 96, 112, 128, 160, 192, 224, 256, 288, 320, 384, 448, 512, 640, 768, 1024, 1280}

The tuner's coarse grid (§6.2): today's sweep grid, so converted and tuned tables line up.

Definition at line 48 of file choice.hh.

grid_batch

std::array<int, 5> batchlas::ops::potrf::grid_batch {128, 512, 2048, 8192, 32768}

inlineconstexpr

batch axis of the grid

Definition at line 50 of file choice.hh.

Classes
struct	Tiny
	potrf_tiny_dispatch: registers; n <= potrf_tiny_max_n, max_wg >= 64 More...
struct	Cta
	potrf_cta_dispatch: whole matrix in SLM; n <= the CTA SLM ceiling More...
struct	Lpanel
	potrf_lpanel_dispatch: SLM-resident panels; Lower only, n <= the LPanel SLM ceiling for panel. More...
struct	Blocked
	potrf_blocked_dispatch + gemm/trsm, Lower; nb/W: PotrfBlockedConst + env More...
struct	Vendor
	backend::potrf_vendor; needs the solver library More...