rmuir commented on a change in pull request #18:
URL: https://github.com/apache/lucene/pull/18#discussion_r594337476
##########
File path: lucene/core/src/java/org/apache/lucene/util/VectorUtil.java
##########
@@ -30,66 +36,26 @@ public static float dotProduct(float[] a, float[] b) {
if (a.length != b.length) {
throw new IllegalArgumentException("vector dimensions differ: " +
a.length + "!=" + b.length);
}
- float res = 0f;
- /*
- * If length of vector is larger than 8, we use unrolled dot product to
accelerate the
- * calculation.
- */
- int i;
- for (i = 0; i < a.length % 8; i++) {
- res += b[i] * a[i];
- }
- if (a.length < 8) {
- return res;
- }
- for (; i + 31 < a.length; i += 32) {
- res +=
- b[i + 0] * a[i + 0]
- + b[i + 1] * a[i + 1]
- + b[i + 2] * a[i + 2]
- + b[i + 3] * a[i + 3]
- + b[i + 4] * a[i + 4]
- + b[i + 5] * a[i + 5]
- + b[i + 6] * a[i + 6]
- + b[i + 7] * a[i + 7];
- res +=
- b[i + 8] * a[i + 8]
- + b[i + 9] * a[i + 9]
- + b[i + 10] * a[i + 10]
- + b[i + 11] * a[i + 11]
- + b[i + 12] * a[i + 12]
- + b[i + 13] * a[i + 13]
- + b[i + 14] * a[i + 14]
- + b[i + 15] * a[i + 15];
- res +=
- b[i + 16] * a[i + 16]
- + b[i + 17] * a[i + 17]
- + b[i + 18] * a[i + 18]
- + b[i + 19] * a[i + 19]
- + b[i + 20] * a[i + 20]
- + b[i + 21] * a[i + 21]
- + b[i + 22] * a[i + 22]
- + b[i + 23] * a[i + 23];
- res +=
- b[i + 24] * a[i + 24]
- + b[i + 25] * a[i + 25]
- + b[i + 26] * a[i + 26]
- + b[i + 27] * a[i + 27]
- + b[i + 28] * a[i + 28]
- + b[i + 29] * a[i + 29]
- + b[i + 30] * a[i + 30]
- + b[i + 31] * a[i + 31];
+ int i = 0;
+ float res = 0;
+ // if the array size is large (2x platform vector size), its worth the
overhead to vectorize
+ // vector loop is unrolled a single time (2 accumulators in parallel)
+ if (a.length >= 2 * SPECIES.length()) {
Review comment:
@msokolov here's a few notes of troubles i ran into in case it helps for
the next time:
* start with "full-performance" example in the JEP/javadocs and then iterate
from there.
* don't use "single masked vector op" at the end. don't use masked ops at
all :)
* don't put kernel "last" like current code in master. it must be "first"
(alignment restrictions i think).
* loop structure is important, otherwise whole loop gets slow (bounds checks
I think). See the tricks i had to do with upperBound: that's the only way i was
able to get a fast loop.
I may revisit the existing scalar code in master and see if things can be
improved there, too.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
