mratsim

import macros


type Conv2DLayer[T] = object
  data: T

type Context[T] = ref object
  data: T

####################################

mratsim

import macros, tables, hashes

import ../src/tensor/backend/metadataArray

type TopoShape = DynamicStackArray[int]
  ## TopoShape is compatible but conceptually different from a MetadataArray
  ## TopoShape does not store the batch size information for tensors
  ## and is used to describe layers shape

mratsim

import ../src/arraymancer, random

randomize(42)

let
  ctx = newContext Tensor[float32] # Autograd/neural network graph
  n = 32                           # Batch size

let
  x_train = read_mnist_images("build/train-images.idx3-ubyte").astype(float32) / 255'f32

mratsim

type
  TrainableLayer = object {.inheritable.}
    ## Layer with trainable parameters
    weight: int
    bias: int

  Conv2DLayer{.final.} = object of TrainableLayer
  LinearLayer{.final.} = object of TrainableLayer

mratsim

import random, sequtils, times

type
  Op = enum
    Halt # = 0x0000
    Inc  # = 0x0100
    Dec  # = 0x0110
    Mul2 # = 0x0230
    Div2 # = 0x0240
    Add7 # = 0x0307

mratsim

macro myzip(x: ForLoopStmt): untyped =
  x.expectKind nnkForStmt
  result = newStmtList()
  echo x.treeRepr
  echo "\n##################"

  # Count the number of idents:
  var ident_count = 1
  while x[ident_count + 1].kind == nnkIdent:
    inc ident_count

mratsim

import macros, ../src/arraymancer

# Transforms

dumpTree:
  network model_name:
    context: ctx
    input_shape: [1, 28, 28]              # Real shape [N, 1, 28, 28]
    layers:

mratsim

import macros, typetraits

#######################################################

static: echo "########################################"
static: echo "### zip Tree ###"
dumpTree:
  iterator zip(a: seq[int], b: seq[bool]): (int,bool) {.noSideEffect.} =
    let len = min(a.len, b.len)

mratsim

# Mpint
# Copyright 2018 Status Research & Development GmbH
# Licensed under either of
#
#  * Apache License, version 2.0, ([LICENSE-APACHE](LICENSE-APACHE) or http://www.apache.org/licenses/LICENSE-2.0)
#  * MIT license ([LICENSE-MIT](LICENSE-MIT) or http://opensource.org/licenses/MIT)
#
# at your option. This file may not be copied, modified, or distributed except according to those terms.

import  ./uint_type, macros

mratsim

func `shl`*(x: MpUintImpl, y: SomeInteger): MpUintImpl {.inline.}=
  ## Compute the `shift left` operation of x and y
  # Note: inlining this poses codegen/aliasing issue when doing `x = x shl 1`

  # TODO: would it be better to reimplement this using an array of bytes/uint64
  # That opens up to endianness issues.

  const halfSize = size_mpuintimpl(x) div 2
  let defect = halfSize - int(y)