athas

let safescatter 't [m] [n] (dest: *[m]t) (is: [n]i32) (vs: [n]t): *[m]t =
  let dest' = scatter dest is vs
  let in_bounds = scatter (replicate 1 true)
                          (map (\i -> if i >= 0 && i < m then -1 else 0) is)
                          (replicate n false)
  in assert in_bounds[0] dest'

-- ==
-- entry: test_scatter test_safescatter
-- random input { 100 [1000000]i32 } auto output

athas

type Agent = { strategy_sum: []f32
             , regret_sum: []f32
             }

type Setting = {num_fields: i32,
                num_soldiers: i32
               }

let compare (x: i32) (y: i32) =
  i32.sgn (x-y)

athas

module FwdAD(T: real): {
  type r = T.t
  type t = (r, r)
  val inject: r -> t
  val set_deriv: t -> r -> t
  val get_deriv: t -> r
  val make_dual: r -> r -> t

  include from_prim with t = (r,r)
  include numeric with t = (r,r)

athas

-- ==
-- entry: direct im2col winograd mec
-- random input { [1024][1024]f32 [3][3]f32 }
-- random input { [512][512]f32 [3][3]f32 }

import "direct"
import "im2col"
import "mec"
import "winograd"

athas

  let interpretData [rows][cols]
                  (data: [rows][cols]f32): [][9]f32 =

    let res = 
      unsafe
      map (\flat ->
             unsafe
             let i = 1+flat / (cols-2)
             let j = 1+flat % (cols-2)
             in [data[i-1,j-1], data[i-1, j], data[i-1, j+1],

athas

""" Neural Network.

A 2-Hidden Layers Fully Connected Neural Network (a.k.a Multilayer Perceptron)
implementation with TensorFlow. This example is using the MNIST database
of handwritten digits (http://yann.lecun.com/exdb/mnist/).

Links:
[MNIST Dataset](http://yann.lecun.com/exdb/mnist/).

Author: Aymeric Damien

athas

#include <stdio.h>
#include <stdlib.h>
#include <stdbool.h>
#include <math.h>
#include <stdint.h>
#undef NDEBUG
#include <assert.h>
// Start of panic.h.

#include <stdarg.h>

athas

#ifdef cl_clang_storage_class_specifiers
#pragma OPENCL EXTENSION cl_clang_storage_class_specifiers : enable
#endif
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
#pragma OPENCL EXTENSION cl_khr_fp64 : enable
__kernel void dummy_kernel(__global unsigned char *dummy, int n)
{
    const int thread_gid = get_global_id(0);
    
    if (thread_gid >= n)

athas

let hood 'a (xs: []a) : [](a,a,a) =
  zip3 (rotate (-1) xs) xs (rotate 1 xs)

let hood_2d 'a (xss: [][]a) : [][]((a,a,a),
                                   (a,a,a),
                                   (a,a,a)) =
  map hood xss -- [][](a, a, a)
  |> hood -- []([](a, a, a), [](a, a, a), [](a, a, a))
  |> map (\(a,b,c) -> zip3 a b c)

athas

FHPNC Parallelism Challenge

At ICFP/FHPNC, Hans-Wolfgang Loidl suggested a joint "challenge" for the various projects and efforts within the FP/performance/parallelism intersection. The purpose is on one hand to demonstrate that there are actually some pretty solid projects going on, and on the other hand to investigate differences in approaches. Essentially something very similar to the old SICSA MultiCore Challenge.