This patch series contains some of the work done by Vladislav in the beginning of March, that seems to have been forgotten.
For reference, that series, with review comments, can be found here: https://lists.freedesktop.org/archives/mesa-dev/2017-January/139892.html I've adressed some of the review comments, most notably redoing the string buffer implementation quite a bit, and adding tests. The major change is that it is now decoupled from the parser and instead lives in /src/util/ as a utility for easier reuse. I've also closely followed the structure of our hash table and set so that it should be quite familiar for everyone. My implementation is also null terminated, while Vladislav's implementation was not. My reasoning behind this was that it's less fragile if someone where to access the contents of the buffer directly. I've also expanded with some different functions that allow us to do more stuff. Obviously this could be expanded upon further if need be. I've included some of Vladislav's less involved patches that I think we should get in as soon as possible. I've added Ian's r-b on the ones that he reviewed (I hope that's OK). The one patch that is missing from this series is the hand-written custom parser. While this was the thing that gave the biggest speed-up, unfortunately it's also a bit harder to review, so I've left that as a future exercise. I've run it through my complete shader-db and there's no changes or breakages, so that's encouraging. It amounts to about a 3% reduction in runtime and executed instructions on the shader-db run. It should be noted that the i965 backend is the primary bottleneck when running shader-db on my computer, so the numbers don't look all that impressive due to this. I'll see if I can come up with some numbers using the gallium noop driver. I have not done a very thorough job on testing that the output of the preprocessor/parser is the same after this series. However, there's no changes here that I believe should cause any issues, so I feel quite confident that this series should not cause any trouble. The original series did have some issues discovered by running a fuzzer over it, however I expect that was caused by the introduced hand-written fast-path scanner and associated macro substitutaion that are not a part of this series. CC: Vladislav Egorov <vegorov...@gmail.com> CC: Ian Romanick <i...@freedesktop.com> Thomas Helland (5): util: Add a string buffer implementation util: Add tests for the string buffer glsl: Change the parser to use the string buffer glcpp: Use string_buffer for line continuation removal glcpp: Avoid unnecessary call to strlen Vladislav Egorov (4): glcpp: Avoid unnecessary strcmp() glcpp: Skip unnecessary line continuations removal ralloc: Use strnlen() inside of strncat() glcpp: Use Bloom filter before identifier search configure.ac | 1 + src/compiler/glsl/glcpp/glcpp-lex.l | 3 +- src/compiler/glsl/glcpp/glcpp-parse.y | 123 +++++++++------- src/compiler/glsl/glcpp/glcpp.h | 18 ++- src/compiler/glsl/glcpp/pp.c | 73 ++++++---- src/util/Makefile.am | 3 +- src/util/Makefile.sources | 2 + src/util/ralloc.c | 7 +- src/util/string_buffer.c | 180 ++++++++++++++++++++++++ src/util/string_buffer.h | 75 ++++++++++ src/util/tests/string_buffer/Makefile.am | 34 +++++ src/util/tests/string_buffer/append_and_print.c | 82 +++++++++++ 12 files changed, 505 insertions(+), 96 deletions(-) create mode 100644 src/util/string_buffer.c create mode 100644 src/util/string_buffer.h create mode 100644 src/util/tests/string_buffer/Makefile.am create mode 100644 src/util/tests/string_buffer/append_and_print.c -- 2.13.0 _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
