Skip to content

Faster optimized frozen dictionary creation (5/n) #87960

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
stephentoub merged 2 commits into from
Jun 23, 2023
Merged

Faster optimized frozen dictionary creation (5/n) #87960

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
db6d1f0
address code review feedback from previous PR: move the duplicated lo…
adamsitnik Jun 22, 2023
1e071d0
Based on our observations, in more than 99.5% of cases the number of …
adamsitnik Jun 22, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
63 changes: 34 additions & 29 deletions src/libraries/System.Collections.Immutable/src/System/Collections/Frozen/FrozenHashTable.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -178,12 +178,16 @@ private static int CalcNumBuckets(ReadOnlySpan<int> hashCodes, bool optimizeForR
}
Debug.Assert(uniqueCodesCount != 0);

// Based on our observations, in more than 99.5% of cases the number of buckets that meets our criteria is
// at least twice as big as the number of unique hash codes.
int minNumBuckets = uniqueCodesCount * 2;

// In our precomputed primes table, find the index of the smallest prime that's at least as large as our number of
// hash codes. If there are more codes than in our precomputed primes table, which accommodates millions of values,
// give up and just use the next prime.
ReadOnlySpan<int> primes = HashHelpers.Primes;
int minPrimeIndexInclusive = 0;
while ((uint)minPrimeIndexInclusive < (uint)primes.Length && uniqueCodesCount > primes[minPrimeIndexInclusive])
while ((uint)minPrimeIndexInclusive < (uint)primes.Length && minNumBuckets > primes[minPrimeIndexInclusive])
{
minPrimeIndexInclusive++;
}
Expand Down Expand Up @@ -217,37 +221,27 @@ private static int CalcNumBuckets(ReadOnlySpan<int> hashCodes, bool optimizeForR

int bestNumBuckets = maxNumBuckets;
int bestNumCollisions = uniqueCodesCount;
int numBuckets = 0, numCollisions = 0;

// Iterate through each available prime between the min and max discovered. For each, compute
// the collision ratio.
for (int primeIndex = minPrimeIndexInclusive; primeIndex < maxPrimeIndexExclusive; primeIndex++)
{
// Get the number of buckets to try, and clear our seen bucket bitmap.
int numBuckets = primes[primeIndex];
numBuckets = primes[primeIndex];
Array.Clear(seenBuckets, 0, Math.Min(numBuckets, seenBuckets.Length));

// Determine the bucket for each hash code and mark it as seen. If it was already seen,
// track it as a collision.
int numCollisions = 0;
numCollisions = 0;

if (codes is not null && uniqueCodesCount != hashCodes.Length)
{
foreach (int code in codes)
{
uint bucketNum = (uint)code % (uint)numBuckets;
if ((seenBuckets[bucketNum / BitsPerInt32] & (1 << (int)bucketNum)) != 0)
{
numCollisions++;
if (numCollisions >= bestNumCollisions)
{
// If we've already hit the previously known best number of collisions,
// there's no point in continuing as worst case we'd just use that.
break;
}
}
else
if (!IsBucketFirstVisit(code))
{
seenBuckets[bucketNum / BitsPerInt32] |= 1 << (int)bucketNum;
break;
}
}
}
Expand All @@ -256,20 +250,9 @@ private static int CalcNumBuckets(ReadOnlySpan<int> hashCodes, bool optimizeForR
// All of the hash codes in hashCodes are unique. In such scenario, it's faster to iterate over a span.
foreach (int code in hashCodes)
{
uint bucketNum = (uint)code % (uint)numBuckets;
if ((seenBuckets[bucketNum / BitsPerInt32] & (1 << (int)bucketNum)) != 0)
{
numCollisions++;
if (numCollisions >= bestNumCollisions)
{
// If we've already hit the previously known best number of collisions,
// there's no point in continuing as worst case we'd just use that.
break;
}
}
else
if (!IsBucketFirstVisit(code))
{
seenBuckets[bucketNum / BitsPerInt32] |= 1 << (int)bucketNum;
break;
}
}
}
Expand All @@ -292,6 +275,28 @@ private static int CalcNumBuckets(ReadOnlySpan<int> hashCodes, bool optimizeForR
ArrayPool<int>.Shared.Return(seenBuckets);

return bestNumBuckets;

[MethodImpl(MethodImplOptions.AggressiveInlining)]
bool IsBucketFirstVisit(int code)
Comment thread
stephentoub marked this conversation as resolved.
{
uint bucketNum = (uint)code % (uint)numBuckets;
if ((seenBuckets[bucketNum / BitsPerInt32] & (1 << (int)bucketNum)) != 0)
{
numCollisions++;
if (numCollisions >= bestNumCollisions)
{
// If we've already hit the previously known best number of collisions,
// there's no point in continuing as worst case we'd just use that.
return false;
}
}
else
{
seenBuckets[bucketNum / BitsPerInt32] |= 1 << (int)bucketNum;
}

return true;
}
}

private readonly struct Bucket
Expand Down