Skip to content

Small optimization in OptimizedScalarQuantizer by using mul instead of div #132397

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
merged 4 commits into from
Aug 4, 2025
Merged

Small optimization in OptimizedScalarQuantizer by using mul instead of div #132397

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
d571325
Small optimization in OptimizedScalarQuantizer by using mul instead o…
iverase Aug 4, 2025
c26ff2c
iter
iverase Aug 4, 2025
ee0006a
[CI] Auto commit changes from spotless
Aug 4, 2025
b0b4577
iter
iverase Aug 4, 2025
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
3 changes: 2 additions & 1 deletion ...ain/java/org/elasticsearch/simdvec/internal/vectorization/DefaultESVectorUtilSupport.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -80,10 +80,11 @@ public void calculateOSQGridPoints(float[] target, int[] quantize, int points, f
float dbb = 0;
float dax = 0;
float dbx = 0;
float invPmOnes = 1f / (points - 1f);
for (int i = 0; i < target.length; ++i) {
float v = target[i];
float k = quantize[i];
float s = k / (points - 1);
float s = k * invPmOnes;
float ms = 1f - s;
daa = fma(ms, ms, daa);
dab = fma(ms, s, dab);
Expand Down
20 changes: 11 additions & 9 deletions ...in21/java/org/elasticsearch/simdvec/internal/vectorization/PanamaESVectorUtilSupport.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -132,7 +132,7 @@ public void centerAndCalculateOSQStatsEuclidean(float[] vector, float[] centroid
FloatVector centeredVec = v.sub(c);
FloatVector deltaVec = centeredVec.sub(vecMeanVec);
norm2Vec = fma(centeredVec, centeredVec, norm2Vec);
vecMeanVec = vecMeanVec.add(deltaVec.div(count));
vecMeanVec = vecMeanVec.add(deltaVec.mul(1f / count));
FloatVector delta2Vec = centeredVec.sub(vecMeanVec);
m2Vec = fma(deltaVec, delta2Vec, m2Vec);
minVec = minVec.min(centeredVec);
Expand Down Expand Up @@ -214,7 +214,7 @@ public void centerAndCalculateOSQStatsDp(float[] vector, float[] centroid, float
FloatVector centeredVec = v.sub(c);
FloatVector deltaVec = centeredVec.sub(vecMeanVec);
norm2Vec = fma(centeredVec, centeredVec, norm2Vec);
vecMeanVec = vecMeanVec.add(deltaVec.div(count));
vecMeanVec = vecMeanVec.add(deltaVec.mul(1f / count));
FloatVector delta2Vec = centeredVec.sub(vecMeanVec);
m2Vec = fma(deltaVec, delta2Vec, m2Vec);
minVec = minVec.min(centeredVec);
Expand Down Expand Up @@ -278,6 +278,7 @@ public void calculateOSQGridPoints(float[] target, int[] quantize, int points, f
float dbb = 0;
float dax = 0;
float dbx = 0;
float invPmOnes = 1f / (points - 1f);
// if the array size is large (> 2x platform vector size), it's worth the overhead to vectorize
if (target.length > 2 * FLOAT_SPECIES.length()) {
FloatVector daaVec = FloatVector.zero(FLOAT_SPECIES);
Expand All @@ -286,11 +287,11 @@ public void calculateOSQGridPoints(float[] target, int[] quantize, int points, f
FloatVector daxVec = FloatVector.zero(FLOAT_SPECIES);
FloatVector dbxVec = FloatVector.zero(FLOAT_SPECIES);
FloatVector ones = FloatVector.broadcast(FLOAT_SPECIES, 1f);
FloatVector pmOnes = FloatVector.broadcast(FLOAT_SPECIES, points - 1f);
FloatVector invPmOnesVec = FloatVector.broadcast(FLOAT_SPECIES, invPmOnes);
for (; i < FLOAT_SPECIES.loopBound(target.length); i += FLOAT_SPECIES.length()) {
FloatVector v = FloatVector.fromArray(FLOAT_SPECIES, target, i);
FloatVector oVec = IntVector.fromArray(INTEGER_SPECIES, quantize, i).convert(VectorOperators.I2F, 0).reinterpretAsFloats();
FloatVector sVec = oVec.div(pmOnes);
FloatVector sVec = oVec.mul(invPmOnesVec);
FloatVector smVec = ones.sub(sVec);
daaVec = fma(smVec, smVec, daaVec);
dabVec = fma(smVec, sVec, dabVec);
Expand All @@ -307,7 +308,7 @@ public void calculateOSQGridPoints(float[] target, int[] quantize, int points, f

for (; i < target.length; i++) {
float k = quantize[i];
float s = k / (points - 1);
float s = k * invPmOnes;
float ms = 1f - s;
daa = fma(ms, ms, daa);
dab = fma(ms, s, dab);
Expand Down Expand Up @@ -798,25 +799,26 @@ public static float ipFloatByteImpl(float[] q, byte[] d) {
@Override
public int quantizeVectorWithIntervals(float[] vector, int[] destination, float lowInterval, float upperInterval, byte bits) {
float nSteps = ((1 << bits) - 1);
float step = (upperInterval - lowInterval) / nSteps;
float invStep = nSteps / (upperInterval - lowInterval);
int sumQuery = 0;
int i = 0;
if (vector.length > 2 * FLOAT_SPECIES.length()) {
int limit = FLOAT_SPECIES.loopBound(vector.length);
FloatVector lowVec = FloatVector.broadcast(FLOAT_SPECIES, lowInterval);
FloatVector upperVec = FloatVector.broadcast(FLOAT_SPECIES, upperInterval);
FloatVector stepVec = FloatVector.broadcast(FLOAT_SPECIES, step);
FloatVector invStepVec = FloatVector.broadcast(FLOAT_SPECIES, invStep);
for (; i < limit; i += FLOAT_SPECIES.length()) {
FloatVector v = FloatVector.fromArray(FLOAT_SPECIES, vector, i);
FloatVector xi = v.max(lowVec).min(upperVec); // clamp
IntVector assignment = xi.sub(lowVec).div(stepVec).add(0.5f).convert(VectorOperators.F2I, 0).reinterpretAsInts(); // round
// round
IntVector assignment = xi.sub(lowVec).mul(invStepVec).add(0.5f).convert(VectorOperators.F2I, 0).reinterpretAsInts();
sumQuery += assignment.reduceLanes(ADD);
assignment.intoArray(destination, i);
}
}
for (; i < vector.length; i++) {
float xi = Math.min(Math.max(vector[i], lowInterval), upperInterval);
int assignment = Math.round((xi - lowInterval) / step);
int assignment = Math.round((xi - lowInterval) * invStep);
sumQuery += assignment;
destination[i] = assignment;
}
Expand Down