mirror of
https://github.com/apache/impala.git
synced 2025-12-19 09:58:28 -05:00
7230c57f67
Adds the 'codegen_opt_level' query option to select LLVM optimization level for generated code. Retains the prior behavior - O2 - as default. If optimization level is changed for an entry already in cache, the cache entry will be used unless the new optimization level is higher than the cached level. Adds additional counters for NumOptimizedFunctions and NumOptimizedInstructions, which allow observing some impacts from codegen optimization. These additional counters, and tracking opt level for cached entries, increases the size of each cached entry. Adds unit tests for all optimizition levels checking - that small functions are inlined at higher levels (as a way to verify that optimization level has an effect) - codegen cache entries are updated when optimizing the same fragment at a higher level, and not updated the rest of the time Change-Id: I371f8758b6552263e91a1fbfd9a6e1c28e1fa2bd Reviewed-on: http://gerrit.cloudera.org:8080/20399 Reviewed-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com> Tested-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>