AlphaGitHub
Source/Packages

Realization.Nvidia.SM86.AdamWHalfSM86

packages/realizations/cooperative/nvidia-sm86/src/Realization/Nvidia/SM86/AdamWHalfSM86.alpha

428 lines73 declarations26.3 KiBSHA-256 9692515bc485
Download exact fileView at revisionbcc11a4e7647d31628

def · lines 362–371

ahLeadStore

Full file
thread T 0 stores the step's scalars and 1 (pending) to the learner state (argument 11's pointer): the update the next forward's side warps finish
362def ahLeadStore = (lambda unrestricted tail : (family SM86Program) .
363  (saGreater (constructor SM86Predicate SM86Predicate2) 2 0
364  (saMovConst 16 (saArgument adamWHalfSM86StateArgument)
365  (saMovConst 17 (naturalAdd (saArgument adamWHalfSM86StateArgument) 4)
366  (saMovImm 27 1
367  (saUnless (constructor SM86Predicate SM86Predicate2)
368    (constructor SM86InstructionBody SM86StoreGlobal64 (saR 16) (saR 24) (saU 0) ahStoreControl)
369  (saUnless (constructor SM86Predicate SM86Predicate2)
370    (constructor SM86InstructionBody SM86StoreGlobal (saR 16) (saR 27) (saU 8) ahStoreControl)
371    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.