K^T's B fragments of slice k (keys 16 k..), n-tile pair p (columns
16 p..), LDSM.T through pair p's lane: the registers the query pass's
sqsLoadFragments fills from its padded tile
2063def sqdLoadKeys = (lambda unrestricted base : Nat . (lambda unrestricted wait : Nat .
2064 (lambda unrestricted tail : (family SM86Program) .
2065 (saFor 16 (lambda unrestricted i : Nat .
2066 (let unrestricted k = (naturalDivideUnchecked i 4) in (let unrestricted pair = (naturalModuloUnchecked i 4) in
2067 (saOp (constructor SM86InstructionBody SM86LoadSharedMatrix (saR (saB k (naturalMultiply 2 pair))) (saR (sqdLaneKeys pair))
2068 (saU (naturalAdd base (naturalMultiply k (naturalMultiply 16 sqdRowBytes))))
2069 (constructor SM86SharedMatrixCount SM86SharedMatrix4)
2070 (constructor SM86SharedMatrixTranspose SM86SharedMatrixTransposed)
2071 (saCtl saSB1 (naturalSelect (naturalIsZero i) wait saWaitNone)))))))
2072 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.