bend-ml-tensor-gpu
Search uses release 0.1.5.1 · 147 definitions
Definitions (147)
- type Ethe sizes and strides of one product: C[i][j] = sum_p A[oa + i*arow + p*sa] * B[ob + p*sb + j*bcol], n x m, k steps (oa, ob: 0 except for matmul_precise's blocks)
- def pka product that counts only when its step is inside k (the last turn's steps past k add +0.0)
- type GRthe values of one turn: A's 4 rows and B's 4 columns at four consecutive steps, read through the arrays in order
- def gr_31
- def gr_30
- def gr_29
- def gr_28
- def gr_27
- def gr_26
- def gr_25
- def gr_24
- def gr_23
- def gr_22
- def gr_21
- def gr_20
- def gr_19
- def gr_18
- def gr_17
- def gr_16
- def gr_15
- def gr_14
- def gr_13
- def gr_12
- def gr_11
- def gr_10
- def gr_9
- def gr_8
- def gr_7
- def gr_6
- def gr_5
- def gr_4
- def gr_3
- def gr_2
- def gr_1
- def gr_0
- def turnthe turn whose A values start at ia and B values at ib
- type Tthe tile's cells after the four sums are combined, row by row
- def mm
- def wrwrites cell (i, j) of C when it is inside n x m
- def put
- def tile
- def leaf_if
- def leaf
- def tree
- def tree_new
- def bang_new
- def bang
- type ABC
- def run_e
- def e_ofothe product's sizes and strides
- def e_of
- def tiles
- def gemm_new
- def gemm
- def mm_fin
- def mm_nt_fin
- def mm_tn_fin
- def matmul
- def matmul_acc
- def matmul_nt
- def matmul_nt_acc
- def matmul_tn
- def matmul_tn_acc
- def depth2
- def depthleaves for n cells: 2^14 when there are that many, else the fewest powers of two that cover them
- def sharehow many of the cells t, t + l, t + 2l, ..
- def nonean array of one cell, for an unused input
- type O2
- type O3
- type O4
- type R1
- type R2
- type R3
- type R4
- def join2
- def join3
- def join4
- def tree1
- def root1
- def root1_new
- def go1
- def go1_new
- def run1
- def run1_new
- def tree2
- def root2
- def root2_new
- def go2
- def go2_new
- def run2
- def run2_new
- def tree3
- def root3
- def root3_new
- def go3
- def go3_new
- def run3
- def run3_new
- def tree4
- def root4