BendtorchGpu.tk_src: def of bendtorch-gpu@1.3.0.0
Import: import bendtorch-gpu@1.3.0.0/main.bend as BendtorchGpu
Lines 686-693 of bendtorch-gpu@1.3.0.0/main.bend: https://jend.leeeet.dev/read.txt?path=bendtorch-gpu@1.3.0.0/main.bend&start_line=686&end_line=693

import Base

# K and V transposed for the score cells (kᵀ: b·kv·hd·t, by sequence, key/value head, column c, position j), so
# that neighbouring lanes, which take neighbouring keys j, read neighbouring cells. dot0 reads kᵀ with stride t: the same
# numbers in the same order as reading k's row, so the same bits.
def tk_src(+e: U32, +t32: U32, +kv: U32, +hd32: U32, +kvw: U32) -> U32:
  +r = U32.div(e, t32)
  +bj = U32.div(r, hd32)
  +row = ((U32.div(bj, kv) * t32 : U32) + U32.mod(e, t32) : U32)
  ((row * kvw : U32) + ((U32.mod(bj, kv) * hd32 : U32) + U32.mod(r, hd32) : U32) : U32)

Uses (5):
  https://jend.leeeet.dev/definition.txt?path=Base/base.bend&name=U32
  https://jend.leeeet.dev/definition.txt?path=Base/base.bend&name=U32.add
  https://jend.leeeet.dev/definition.txt?path=Base/base.bend&name=U32.div
  https://jend.leeeet.dev/definition.txt?path=Base/base.bend&name=U32.mod
  https://jend.leeeet.dev/definition.txt?path=Base/base.bend&name=U32.mul