[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13060829#comment-13060829 ] Uwe Schindler commented on LUCENE-3179: --- Committed long versions and additional tests: rev 1143558 (trunk), rev 1143560 (3.x). I did not commit the cutover to Long.numberOfLeadingZeroes, because it was not performance tested. Also from the use-case, on machines without intrinsics, the JDK-given methods are slower (see comments in BitUtils.ntz, as in most cases the bits are shifted away (in nextSetBit), so the faster algorithm is to inverse the algorithm when calculating ntz. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Assignee: Paul Elschot Priority: Minor Fix For: 3.3, 4.0 Attachments: LUCENE-3179-fix.patch, LUCENE-3179-fix.patch, LUCENE-3179-long-ntz.patch, LUCENE-3179-long-ntz.patch, LUCENE-3179.patch, LUCENE-3179.patch, LUCENE-3179.patch, TestBitUtil.java, TestOpenBitSet.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13057980#comment-13057980 ] Uwe Schindler commented on LUCENE-3179: --- Any other comments/microbenchmarks from other committers? Dawid and Paul? I would like to commit this if nobody objects! What should we do with the then obsolete BitUtils methods? OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Assignee: Paul Elschot Priority: Minor Fix For: 3.3, 4.0 Attachments: LUCENE-3179-fix.patch, LUCENE-3179-fix.patch, LUCENE-3179-long-ntz.patch, LUCENE-3179-long-ntz.patch, LUCENE-3179.patch, LUCENE-3179.patch, LUCENE-3179.patch, TestBitUtil.java, TestOpenBitSet.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13055506#comment-13055506 ] Michael McCandless commented on LUCENE-3179: Patch looks good Uwe -- thanks! OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Assignee: Paul Elschot Priority: Minor Fix For: 3.3, 4.0 Attachments: LUCENE-3179-fix.patch, LUCENE-3179-fix.patch, LUCENE-3179-long-ntz.patch, LUCENE-3179-long-ntz.patch, LUCENE-3179.patch, LUCENE-3179.patch, LUCENE-3179.patch, TestBitUtil.java, TestOpenBitSet.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13055063#comment-13055063 ] Michael McCandless commented on LUCENE-3179: Thanks for fixing these Uwe! I actually don't like how generic OBS has become... ie, that all methods have an int and long version, that the OBS doesn't know how many bits it holds (I added this field recently, but only for assertions), that some methods grow the number of bits and others don't, some methods accept out-of-bounds indices (negative and numBits), etc. I think it's grown to accommodate too many users but I'm not sure what we should do to fix this. Maybe factor out (yet another) bit set impl that doesn't grow, knows its number of bits, has these fast getNext/getPrev set bit methods, operates only on int indices, etc. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Assignee: Paul Elschot Priority: Minor Fix For: 3.3, 4.0 Attachments: LUCENE-3179-fix.patch, LUCENE-3179-fix.patch, LUCENE-3179.patch, LUCENE-3179.patch, LUCENE-3179.patch, TestBitUtil.java, TestOpenBitSet.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13055064#comment-13055064 ] Michael McCandless commented on LUCENE-3179: bq. One more comment: When working on the code, the symmetry all other methods have between long and int is broken here. For consistency we should add the long method, too. I just don't like the missing consistency. I think we should add the long version, for consistency. bq. Also: OpenBitSet.nextSetBit() does not use Long.numberOfTrailingZeroes() but the new prevSetBit() does. As both methods have intrinsics, why only use one of them? Yonik? Good question! In testing on this issue, above, Dawid and Paul found the intrinsics were faster on modern JREs... seems like nextSetBit should cutover too? OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Assignee: Paul Elschot Priority: Minor Fix For: 3.3, 4.0 Attachments: LUCENE-3179-fix.patch, LUCENE-3179-fix.patch, LUCENE-3179.patch, LUCENE-3179.patch, LUCENE-3179.patch, TestBitUtil.java, TestOpenBitSet.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[
https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13054578#comment-13054578
]
Uwe Schindler commented on LUCENE-3179:
---
The check for negative indexes must be done to make the following loop work
(which is standard to iterate backwards from startBit on all bits):
{code:java}
for (int i = bs.prevSetBit(startBit); i = 0; i = bs.prevSetBit(i-1)) {
// operate on index i here
}
{code}
This would fail with AIOOBE when i=0 on the last iteration (happens if 0th bit
is set), because bs.prevSetBit(i-1) has negative parameter. The exit condition
is checked later, so -1 must be allowed.
OpenBitSet.prevSetBit()
---
Key: LUCENE-3179
URL: https://issues.apache.org/jira/browse/LUCENE-3179
Project: Lucene - Java
Issue Type: Improvement
Reporter: Paul Elschot
Priority: Minor
Fix For: 3.3, 4.0
Attachments: LUCENE-3179-fix.patch, LUCENE-3179.patch,
LUCENE-3179.patch, LUCENE-3179.patch, TestBitUtil.java, TestOpenBitSet.patch
Find a previous set bit in an OpenBitSet.
Useful for parent testing in nested document query execution LUCENE-2454 .
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
-
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13054606#comment-13054606 ] Paul Elschot commented on LUCENE-3179: -- The 3179-fix patch looks good to me. I remember I had some doubts about which bit was actually the last one, and stopped worrying about it when the tests passed. This patch makes it very clear what the last bit is. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3, 4.0 Attachments: LUCENE-3179-fix.patch, LUCENE-3179-fix.patch, LUCENE-3179.patch, LUCENE-3179.patch, LUCENE-3179.patch, TestBitUtil.java, TestOpenBitSet.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13054607#comment-13054607 ] Yonik Seeley commented on LUCENE-3179: -- +1, patch looks good Uwe. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3, 4.0 Attachments: LUCENE-3179-fix.patch, LUCENE-3179-fix.patch, LUCENE-3179.patch, LUCENE-3179.patch, LUCENE-3179.patch, TestBitUtil.java, TestOpenBitSet.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13054681#comment-13054681 ] Uwe Schindler commented on LUCENE-3179: --- One more comment: When working on the code, the symmetry all other methods have between long and int is broken here. For consistency we should add the long method, too. I just don't like the missing consistency. Also: OpenBitSet.nextSetBit() does not use Long.numberOfTrailingZeroes() but the new prevSetBit() does. As both methods have intrinsics, why only use one of them? Yonik? Any comments? OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Assignee: Paul Elschot Priority: Minor Fix For: 3.3, 4.0 Attachments: LUCENE-3179-fix.patch, LUCENE-3179-fix.patch, LUCENE-3179.patch, LUCENE-3179.patch, LUCENE-3179.patch, TestBitUtil.java, TestOpenBitSet.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13046641#comment-13046641 ] Michael McCandless commented on LUCENE-3179: I think we should just commit this? It's a useful API. LUCENE-3171 (alternative nested docs impl w/ single pass collector) also could use this. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch, LUCENE-3179.patch, TestBitUtil.java Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[
https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13046140#comment-13046140
]
Paul Elschot commented on LUCENE-3179:
--
I did a bit of performance testing (sun java 1.6.0_xx, not the very latest one).
This is a typical output on my machine (the dummy can be ignored, it is only
there to make sure that nothing is optimized away):
{noformat}
BitUtil nlz time: 5664 picosec/call, dummy: 11572915728
Longnlz time: 8464 picosec/call, dummy: 7715277152
{noformat}
That means that the nlz code in the patch is definitely faster than
Long.numberOfLeadingZeros for the test arguments used.
The test arguments are divided roughly evenly for the possible numbers of
leading zero bits.
OpenBitSet.prevSetBit()
---
Key: LUCENE-3179
URL: https://issues.apache.org/jira/browse/LUCENE-3179
Project: Lucene - Java
Issue Type: Improvement
Reporter: Paul Elschot
Priority: Minor
Fix For: 3.3
Attachments: LUCENE-3179.patch
Find a previous set bit in an OpenBitSet.
Useful for parent testing in nested document query execution LUCENE-2454 .
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
-
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13046169#comment-13046169 ] Dawid Weiss commented on LUCENE-3179: - You're not providing the required contex -- what exact JVM and what exact processor did you test on? I've just ran your test on my machine with the following result: BitUtil nlz time: 3109 picosec/call, dummy: 20252602524 Longnlz time: 1279 picosec/call, dummy: 48220482200 I'm guessing yours didn't use the intrinsic inline at all (for whatever reason). My machine is a fairly old Intel I7 860 running 64-Bit server hotspot 1.6.0_24-b07. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch, TestBitUtil.java Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13046227#comment-13046227 ] Paul Elschot commented on LUCENE-3179: -- The java.vm.version value 1.6.0_03-b05, java.vm.info value is mixed mode. The processor is an Athlon II X3 450 at 800 MHz. Since the Long time is about 2.5 times faster than the BitUtil with a 64 bit processor, I'll change the patch to use Long. When the hardware allows better performance, it should be used. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch, TestBitUtil.java Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045576#comment-13045576 ] Yonik Seeley commented on LUCENE-3179: -- Hey Paul, did you try this implementation against Long.numberOfLeadingZeros? The later Oracle Java6 implementations have instrinsified this method, so it might be faster: http://bugs.sun.com/view_bug.do?bug_id=6823354 OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3197.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045587#comment-13045587 ] Paul Elschot commented on LUCENE-3179: -- I did not try this against Long.numberOfLeadingZeros, but in case that is faster we should use that of course. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045655#comment-13045655 ] Uwe Schindler commented on LUCENE-3179: --- If it's faster, should we not replace it completely in Lucene? The impl in Java 5 (Sun JDK) is identical to ours from BitUtils, so why replicate? If it gets intrinsic, it can only get faster. I assume its a relict from pre-Java-1.5 times like Lucene 2.9. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045656#comment-13045656 ] Uwe Schindler commented on LUCENE-3179: --- With the previous comment I also refer to nextSetBit(). OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045659#comment-13045659 ] Dawid Weiss commented on LUCENE-3179: - I posted the benchmarks of intrinsic vs. manual (OpenBitSet) performance of nlz and pop (bitcount) methods a while ago -- they should still be around JIRA somewhere. If I recall right, the difference was significant, although not like an order of magnitude or something... and on CPUs without intrinsic instructions the implementation handcrafted by Yonik was actually faster than the one in the standard library. Of course these days most CPUs will have popcnt/ nlz instructions, so it makes sense to switch. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045662#comment-13045662 ] Dawid Weiss commented on LUCENE-3179: - I think it's the 1.6 that adds these intrinsics -- I don't know if they've been backported to updates to 1.5, but this should be relatively easy to verify empirically. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045667#comment-13045667 ] Dawid Weiss commented on LUCENE-3179: - Intrinsics are implemented/added at the hotspot (jit) level, you won't see them in src.jar -- all calls to specific methods in Long.* or Integer.* are replaced by handcrafted assembly (usually process-specific instructions that do what a given method should do). If you're interested, check out openjdk code of hotspot and scan for intrinsics (or popcnt). OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045669#comment-13045669 ] Uwe Schindler commented on LUCENE-3179: --- You misunderstood me, i know what intrinsics are. My confusion was related to that: bq. and on CPUs without intrinsic instructions the implementation handcrafted by Yonik was actually faster than the one in the standard library And the so called hand crafted method is identical in src.jar and Yonik's code. So without intrinsics, the standard library and Yoniks code should be identical in performance, as it was same code, the last time I looked into it. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045674#comment-13045674 ] Yonik Seeley commented on LUCENE-3179: -- bq. And the so called hand crafted method is identical in src.jar and Yonik's code. For pop, yes. But not for ntz or pop_array and friends. BitUtil.pop exists because this was originally written to work with java1.4 which didn't have Long.bitCount() http://markmail.org/message/5ay4m2thsvsahk3c OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045676#comment-13045676 ] Paul Elschot commented on LUCENE-3179: -- The micro benchmarks for ntz() and pop() are at LUCENE-2221 OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045679#comment-13045679 ] Dawid Weiss commented on LUCENE-3179: - Oh, ok -- clear. So, my comment was related to the various methods of doing bitcounts and other bit-fiddling on arrays of long values (for example pop_array) -- these are HD derived implementations; I compared them to naive loops using intrinsics and naive loops on cpus (and jvms) without intrinsics -- in that case simple loops with intrinsics was faster than Lucene's code, but Lucene's code was faster than simple loops without intrinsics (effectively using whatever was in the std. library). OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045682#comment-13045682 ] Uwe Schindler commented on LUCENE-3179: --- OK, so we can sefely remove BitUtil.pop and replace by the Java 5 method (maybe review again the code in src.jar also for ntz). And if this one is an intrinsic in Java 6 its even faster. Now we talk the same language :-) OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-3179) OpenBitSet.prevSetBit()
[ https://issues.apache.org/jira/browse/LUCENE-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13045685#comment-13045685 ] Paul Elschot commented on LUCENE-3179: -- As to the performance, the current patch at LUCENE-2454 has a bitwise linear search to do this. OpenBitSet.prevSetBit() --- Key: LUCENE-3179 URL: https://issues.apache.org/jira/browse/LUCENE-3179 Project: Lucene - Java Issue Type: Improvement Reporter: Paul Elschot Priority: Minor Fix For: 3.3 Attachments: LUCENE-3179.patch Find a previous set bit in an OpenBitSet. Useful for parent testing in nested document query execution LUCENE-2454 . -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
