bend-tensors

Search uses release 0.0.0.2 · 306 definitions

Releases (2)

0.0.0.2 indexed
0.0.0.1 unindexed

Definitions (306)

bend_tensors.bend

def F32.max.ofBase's F32.max and F32.show take + parameters, which a template slot does not accept; these take plain ones.
def F32.show.of
type Blk
type Vec4
def Vec4.full
def Vec4.zeros
def Vec4.arangex, x + 1, x + 2, x + 3
def Vec4.map
def Vec4.zip
def Vec4.add
def Vec4.sub
def Vec4.mul
def Vec4.scale
def Vec4.sumthe sum, as (x0 + x1) + (x2 + x3): every reduction in this file pairs its terms the same way, so laws can equate them
def Vec4.dot
def Vec4.max
def Vec4.min
def Vec4.argmax.pickthe index (0-3) and value of the largest element; the first wins ties
def Vec4.argmax.two
def Vec4.argmax
def Vec4.to_list
def Vec4.from_listthe first four of xs, padded with 0.0, and the rest
def Vec4.show
def Blk.full
def Blk.zeros
def Blk.eye
def Blk.map
def Blk.zip
def Blk.add
def Blk.sub
def Blk.mul
def Blk.scale
def Blk.transpose
def Blk.matmulm * n: 64 multiply-adds, each cell as (a0 b0 + a1 b1) + (a2 b2 + a3 b3)
def Blk.matvecm * v: each row dotted with v
def Blk.sum_rowsrow i's sum, as (a_i0 + a_i1) + (a_i2 + a_i3)
def Blk.sumthe sum of every cell, as the sum of the row sums
def Blk.sum_colscolumn j's sum, as (a_0j + a_1j) + (a_2j + a_3j): the row sums of the transpose
def Blk.max_rows
def Blk.add_rowv added to every row
def Blk.sub_rowv subtracted from every row
def Blk.sub_colrow i subtracted by v_i (the column vector v broadcast along rows)
def Blk.scale_rowsrow i scaled by v_i
def Blk.to_listthe 16 cells in row-major order, before acc
def Blk.from_vec4sa block from its four rows
def Blk.from_rowsa block from four rows of four; a short row pads with 0.0
def Blk.row
def Vec4.maskelements at index >= k set to 0.0
def Vec4.max_uptothe largest of the first k elements
def Vec4.min_uptothe smallest of the first k elements
def Vec4.argmax_uptothe index and value of the largest of the first k elements
def Blk.mask_rowsrows at index >= k set to 0.0
def Blk.mask_colscolumns at index >= k set to 0.0
def Blk.maskrows at index >= ki and columns at index >= kj set to 0.0
def Blk.max_rows_uptoeach row's largest among its first k columns
law VecT
type VecT.Leaf
type VecT.Node
def VecT.sizethe element count of a VecT(n): 4 * 2^n
def VecT.full
def VecT.zeros
def VecT.ones
def VecT.arange.fromx, x + 1, x + 2, ..
def VecT.arange0, 1, 2, ..
def VecT.map
def VecT.zip
def VecT.add
def VecT.sub
def VecT.mulelementwise product
def VecT.exp
def VecT.scale
def VecT.shiftk added to every element
def VecT.sumthe sum, left half plus right half
def VecT.dot
def VecT.max
def VecT.min
def VecT.argmax.twothe two halves' winners, the right one's index offset by the half's size
def VecT.argmaxthe index and value of the largest element; the first wins ties
def VecT.to_list.gothe elements in order, before acc the tree is read borrowed (+v, matched in place), never consumed: a reader that consumes a shared tree makes the compiler copy it, and bend 2.0.16 miscompiles that copy when the depth is a literal (the bias of a model: memory faults in the loop; issues/shared_vec_loop_*)
def VecT.to_list
def VecT.from_listthe first 4 * 2^n elements of xs, padded with 0.0
def VecT.softmax.fin
def VecT.softmaxexp(v - max v) / sum
def VecT.show
law MatT
type MatT.Leaf
type MatT.Rows
type MatT.Cols
type MatT.Quad
def MatT.full
def MatT.zeros
def MatT.ones
def MatT.eyethe identity of MatT(n, n)
def MatT.map
def MatT.zip
def MatT.add
def MatT.sub
def MatT.mulelementwise (Hadamard) product; the matrix product is MatT.matmul
def MatT.exp
def MatT.scale