Delaunay

# I setup my docker runner to cache those directories
# So the environments are only updated when different
variables:
  XDG_CACHE_HOME: "/cache/pip"
  PIP_CACHE_DIR:  "/cache/pip"
  OLYMPUS_DATA_PATH: "/cache/data"
  ENV_CACHE_DIR: "/cache/env"

stages:
  - tests

Delaunay

HIP_DB=3 HCC_SERIALIZE_KERNEL=0x6 python test/test_cuda.py TestCuda.test_multinomial_invalid_probs_cuda > log.out
HIP_DB=0x3 [api+sync]
  hip-api pid:22173 tid:1:HIP initialized short_tid#1 (maps to full_tid: 0x7fd84c378740)
  hip-sync pid:22173 tid:1:locking criticalData=0x7fd7ee7da798 for ctx:0x7fd7ee7da780.dev:0..
  hip-sync pid:22173 tid:1: addStream: stream:0.0
  hip-sync pid:22173 tid:1:created ctx with defaultStream=0x7fd7eee38670 (stream:0.0)
  hip-sync pid:22173 tid:1:auto-unlocking criticalData=0x7fd7ee7da798 for ctx:0x7fd7ee7da780.dev:0...
  hip-sync pid:22173 tid:1:pid=22173 <ihipInit>                     g_numLogicalThreads=4
<<hip-api pid:22173 tid:1.1 hipGetDeviceCount (0x7ffcaee0a1c4) @370289587147796
  hip-api pid:22173 tid:1.1 hipGetDeviceCount              ret= 0 (hipSuccess)>> +3657 ns

Delaunay

import torchvision
import torchvision.transforms as transforms
import time
import torch
import os
import h5py
import numpy as np

from MixedPrecision.tools.stats import StatStream

Delaunay

#include <thread>
#include <mutex>
#include <tuple>
#include <list>
#include <memory>

namespace pool{

namespace internal {
    template<typename T>