Is there built-in support in the TPL for batching operations?
I was recently playing with a routine to carry out character replacement on a character array which required a lookup table i.e. transliteration:
for (int i = 0; i < chars.Length; i++)
{
char replaceChar;
if (lookup.TryGetValue(chars[i], out replaceChar))
{
chars[i] = replaceChar;
}
}
I could see that this could be trivially parallelized, so jumped in with a first stab which I knew would perform worse as the tasks were too fine-grained:
Parallel.For(0, chars.Length, i =>
{
char replaceChar;
if (lookup.TryGetValue(chars[i], out replaceChar))
{
chars[i] = replaceChar;
}
});
I then reworked the algorithm to use batching so that the work could be chunked onto different threads in less fine-grained batches. This made use of threads as expected and I got some near linear speed up.
I'm sure that there must be built-in support for batching in the TPL. What is the syntax, and how do I use it?
const int CharBatch = 100;
int charLen = chars.Length;
Parallel.For(0, ((charLen / CharBatch) + 1), i =>
{
int batchUpper = ((i + 1) * CharBatch);
for (int j = i * CharBatch; j < batchUpper && j < charLen; j++)
{
char replaceChar;
if (lookup.TryGetValue(chars[j], out replaceChar))
{
chars[j] = replaceChar;
}
}
});
