Source/Packages

Realization.Nvidia.SM86.StreamingAttentionSM86

packages/realizations/cooperative/nvidia-sm86/src/Realization/Nvidia/SM86/StreamingAttentionSM86.alpha

3,230 lines493 declarations193.3 KiBSHA-256 cc0366b3d857

def · lines 1163–1170

sfpFirstCopies

Full file
the prologue's copies: tile 0's K and V into buffers 0, K's tile 1 into K's buffer 1 (P1: a second tile); then K's offsets at tile 1, V's at 0. Tile 0's scores once its copies are visible.
1163def sfpFirstCopies = (lambda unrestricted tail : (family SM86Program) .
1164  (sfCopyTile 0 0
1165  (saFor 4 (lambda unrestricted i : Nat . (saAddImm (sfKeyOffset i) (sfKeyOffset i) (naturalMultiply saTile (naturalMultiply saHeadWidth saHalfBytes))))
1166  (saGreater saP1 saCount 1
1167  (sfpKeyCopy (sfpKeyBuffer 1) 1
1168  (sfWaitGroups 0 saWait5
1169  (sfBarrier
1170  (sfpScores 0 saS (sfpKeyBuffer 0) tail))))))))

The compiler supplied declaration spans and resolved links from this source snapshot. This page does not assert that this file belongs to a checked closure.