public void SpeedTest()
{
int length = 65536, ch = 256, batch = 4;
Shape shape = Shape.Map1D(ch, length, batch);
OverflowCheckedTensor v1 = new OverflowCheckedTensor(shape);
OverflowCheckedTensor v2 = new OverflowCheckedTensor(Shape.Vector(batch));
OverflowCheckedTensor v3 = new OverflowCheckedTensor(shape);
BatchwiseMul ope = new BatchwiseMul(shape);
Stopwatch sw = new Stopwatch();
sw.Start();
ope.Execute(v1, v2, v3);
ope.Execute(v3, v2, v1);
ope.Execute(v1, v2, v3);
ope.Execute(v3, v2, v1);
sw.Stop();
Console.WriteLine($"{sw.ElapsedMilliseconds / 4} msec");
}
public void ExecuteTest()
{
Random rd = new Random(1234);
foreach (int batch in new int[] { 1, 2, 3, 4, 5 })
{
for (int i = 0; i < 4; i++)
{
for (int length = i * 1024 - 2; length <= i * 1024 + 2; length++)
{
if (length < 1)
{
continue;
}
foreach (int ch in new int[] { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 })
{
float[] x1 = (new float[length * ch * batch]).Select((_) => (float)rd.NextDouble()).ToArray();
float[] x2 = (new float[batch]).Select((_) => (float)rd.NextDouble()).ToArray();
Shape shape = Shape.Map1D(ch, length, batch);
OverflowCheckedTensor v1 = new OverflowCheckedTensor(shape, x1);
OverflowCheckedTensor v2 = new OverflowCheckedTensor(Shape.Vector(shape.Batch), x2);
OverflowCheckedTensor v3 = new OverflowCheckedTensor(shape);
BatchwiseMul ope = new BatchwiseMul(shape);
ope.Execute(v1, v2, v3);
CollectionAssert.AreEqual(x1, v1.State);
CollectionAssert.AreEqual(x2, v2.State);
float[] y = v3.State;
for (int j = 0; j < length * ch; j++)
{
for (int k = 0; k < batch; k++)
{
Assert.AreEqual(x1[j + length * ch * k] * x2[k], y[j + length * ch * k], 1e-6f, $"length:{length} ch:{ch} batch:{batch}, idx:{j},{k}");
}
}
}
}
;
}
}
}
