Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | avoid deadlock when using compile cache with blocking executors | Håvard Pettersen | 2020-06-14 | 1 | -1/+1 |
| | | | | | | | | - avoid taking cache lock in compile task destructor - avoid holding cache lock while posting compile task - add unit test trying to provoke deadlock - bonus: fix blocking executor implementation (did not work with threads <= task limit) | ||||
* | minor naming adjustments | Håvard Pettersen | 2020-06-11 | 1 | -7/+7 |
| | |||||
* | common code for operation inlining | Håvard Pettersen | 2020-06-11 | 4 | -1/+231 |
| | | | | | | | | | | - add common code to make selecting the appropriate template function easier (vespa/vespalib/util/typify.h) - enable detection of lambda functions matching all low-level operations. (lookup_op1, lookup_op2) - add typifiers to decide which low-level operations should be inlined (TypifyOp1, TypifyOp2) - integrate into dense_simple_join as a pilot customer | ||||
* | Add static_assert for sanity of template arguments. | Henning Baldersheim | 2020-06-09 | 1 | -0/+4 |
| | |||||
* | Revert "Revert "When we pull in a cacheline, we should use it too."" | Henning Baldersheim | 2020-06-08 | 10 | -7/+211 |
| | |||||
* | Revert "When we pull in a cacheline, we should use it too." | Harald Musum | 2020-06-08 | 10 | -211/+7 |
| | |||||
* | - Stick to void * and byte offsets. | Henning Baldersheim | 2020-06-07 | 10 | -31/+37 |
| | | | | - Correct spelling error. | ||||
* | - Test both normal and inverted bit vectors. | Henning Baldersheim | 2020-06-07 | 1 | -34/+70 |
| | | | | | - Use 64 byte alignment of buffer. - Improve error messages. | ||||
* | When we pull in a cacheline, we should use it too. | Henning Baldersheim | 2020-06-04 | 9 | -3/+165 |
| | | | | There is possibly wasting 7/8 of it and very likely suffer a cache miss. | ||||
* | rename regex option MultiLine -> DotMatchesNewline | Arne Juul | 2020-05-28 | 2 | -4/+4 |
| | |||||
* | add multiline regex option | Arne Juul | 2020-05-28 | 2 | -1/+5 |
| | |||||
* | Merge pull request #13368 from ↵ | Tor Brede Vekterli | 2020-05-26 | 1 | -0/+8 |
|\ | | | | | | | | | vespa-engine/toregge/detect-clang-version-of-address-sanitizer Detect use of clang version of address sanitizer. | ||||
| * | Detect use of clang version of address sanitizer. | Tor Egge | 2020-05-25 | 1 | -0/+8 |
| | | |||||
* | | Keep const qualifier for allocator() return value. | Tor Egge | 2020-05-25 | 1 | -1/+1 |
|/ | |||||
* | Disable rlimit checking tests when using address sanitizer. | Tor Egge | 2020-05-24 | 1 | -0/+2 |
| | |||||
* | Use correct argument for length of vector. | Henning Baldersheim | 2020-05-12 | 1 | -1/+1 |
| | |||||
* | Revert "Revert "Use a smart allocator for allocating memory for large 'long' ↵ | Henning Baldersheim | 2020-05-11 | 10 | -22/+92 |
| | | | | lived"" | ||||
* | Revert "Use a smart allocator for allocating memory for large 'long' lived" | Harald Musum | 2020-05-11 | 10 | -92/+22 |
| | |||||
* | Implemtent equality operator. | Henning Baldersheim | 2020-05-11 | 1 | -2/+10 |
| | |||||
* | Use a smart allocator for allocating memory for large 'long' lived | Henning Baldersheim | 2020-05-11 | 10 | -22/+84 |
| | | | | | vectors. Large vectors will be allocated directly with mmap. This cancels the main reason for using vespalib::Array. | ||||
* | Try to hide implementation. | Henning Baldersheim | 2020-05-10 | 1 | -3/+1 |
| | |||||
* | Rename namespace search::btree to vespalib::btree. | Tor Egge | 2020-05-04 | 109 | -139/+139 |
| | | | | Rename namespace search::datastore to vespalib::datastore. | ||||
* | Prepare for renaming namespace search::btree to vespalib::btree and | Tor Egge | 2020-04-28 | 1 | -3/+1 |
| | | | | namespace search::datastore to vespalib::datastore. | ||||
* | Remove bad namespace end comments. | Tor Egge | 2020-04-28 | 2 | -3/+2 |
| | |||||
* | Merge pull request #13066 from ↵ | Geir Storli | 2020-04-27 | 4 | -94/+0 |
|\ | | | | | | | | | vespa-engine/toregge/move-bufferwriter-back-to-searchlib Move BufferWriter back to searchlib. | ||||
| * | Move BufferWriter back to searchlib. | Tor Egge | 2020-04-25 | 4 | -94/+0 |
| | | |||||
* | | Merge pull request #13063 from vespa-engine/toregge/extend-btree-frozen-view | Geir Storli | 2020-04-27 | 3 | -0/+28 |
|\ \ | |/ |/| | Extend BTree frozen view with empty() and getAggregated() methods. | ||||
| * | Extend BTree frozen view with empty() and getAggregated() methods. | Tor Egge | 2020-04-25 | 3 | -0/+28 |
| | | |||||
* | | Reduce code duplication. Rand48 class is in vespalib. | Tor Egge | 2020-04-25 | 4 | -7/+9 |
|/ | |||||
* | Merge pull request #13034 from ↵ | Tor Brede Vekterli | 2020-04-24 | 8 | -24/+109 |
|\ | | | | | | | | | vespa-engine/vekterli/support-aggregating-over-btree-keys Add support for aggregating over B-tree keys | ||||
| * | Rename variable to be semantically accurate | Tor Brede Vekterli | 2020-04-23 | 1 | -4/+4 |
| | | |||||
| * | Add support for aggregating over B-tree keys | Tor Brede Vekterli | 2020-04-23 | 8 | -24/+109 |
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | This complements the existing support for aggregating over values. Let aggregate calculator specify whether it expects to be invoked with keys or values. In the current implementation there is some code duplication that could have been removed by using e.g. tag dispatch, but this is a pragmatic choice intended to guarantee these changes do not introduce any performance regressions for existing code using value aggregation. Can be refactored later once we have a functional baseline with this for both keys and values. | ||||
* | | If compiler is clang then disable the clock computation hack. | Tor Egge | 2020-04-23 | 1 | -0/+3 |
| | | |||||
* | | Fix buffer overflow in vespalib datastore buffer type test. | Tor Egge | 2020-04-22 | 1 | -3/+3 |
| | | |||||
* | | Merge pull request #12980 from vespa-engine/fix-foreach-variable-replacement | Arne H Juul | 2020-04-22 | 5 | -0/+99 |
|\ \ | | | | | | | Fix foreach variable replacement | ||||
| * | | cosmetic fixes in unit test | Arne Juul | 2020-04-20 | 1 | -1/+7 |
| | | | |||||
| * | | add utility for replacing a variable in a string | Arne Juul | 2020-04-20 | 5 | -0/+93 |
| | | | |||||
* | | | A failed attempt at unlinking a directory causes another exception | Tor Egge | 2020-04-21 | 1 | -0/+4 |
| | | | | | | | | | | | | | | | on Darwin due to unlink system call returning a different error code. | ||||
* | | | Merge pull request #13002 from ↵ | Henning Baldersheim | 2020-04-21 | 1 | -0/+3 |
|\ \ \ | | | | | | | | | | | | | | | | | vespa-engine/toregge/use-extended-locale-support-on-non-linux-platform Use extended locale support on non-linux platforms. | ||||
| * | | | Use extended locale support on non-linux platforms. | Tor Egge | 2020-04-21 | 1 | -0/+3 |
| | |/ | |/| | |||||
* / | | Directio support in vespa is limited to linux. | Tor Egge | 2020-04-21 | 1 | -0/+4 |
|/ / | |||||
* | | Merge pull request #12997 from ↵ | Tor Brede Vekterli | 2020-04-21 | 8 | -50/+47 |
|\ \ | | | | | | | | | | | | | vespa-engine/vekterli/constexpr-ifs-for-most-aggregation-call-sites Make most hasAggregated checks constexpr | ||||
| * | | Make most hasAggregated checks constexpr | Tor Brede Vekterli | 2020-04-21 | 8 | -50/+47 |
| |/ | | | | | | | | | Was previously _effectively_ constexpr so not likely to improve code generation in practice, but it's good to be explicit. | ||||
* | | GC unused code. | Henning Baldersheim | 2020-04-21 | 2 | -114/+0 |
| | | |||||
* | | Elaborate a bit more in the comment. | Henning Baldersheim | 2020-04-21 | 1 | -2/+7 |
| | | |||||
* | | Revert "Revert "There are so many combinations that the libstdc++ library ↵ | Henning Baldersheim | 2020-04-21 | 1 | -0/+26 |
| | | | | | | | | can be buil…"" | ||||
* | | Revert "There are so many combinations that the libstdc++ library can be ↵ | Henning Baldersheim | 2020-04-21 | 1 | -26/+0 |
| | | | | | | | | buil…" | ||||
* | | There are so many combinations that the libstdc++ library can be built that ↵ | Henning Baldersheim | 2020-04-20 | 1 | -0/+26 |
| | | | | | | | | | | | | | | the performance you get from std::chrono::system_clock and stxad::chrono::steady_clock has a dramatic performance difference. On RHEL7/Centos with an witout this patch is 18ns vs 550ns. | ||||
* | | Merge pull request #12985 from ↵ | Henning Baldersheim | 2020-04-20 | 1 | -1/+1 |
|\ \ | | | | | | | | | | | | | vespa-engine/toregge/handle-sh-builtin-echo-on-darwin Handle version of echo not understanding -n option. | ||||
| * | | Don't use shell builtin echo. | Tor Egge | 2020-04-20 | 1 | -5/+1 |
| | | |