Stock function with recursion not removed if not used #358
Comments
dvander
added a commit
that referenced
this issue
Sep 10, 2019
…embly. The two-pass system in the compiler is able to roughly guess which functions are not used, but it's a poor guess. Stocks called by other stocks, including recursive functions, all get compiled in. We can fix this by traversing the call graph to determine the actual live set. Note, the assembler now skips functions with uDEFINED but not uPUBLIC. uDEFINED is cleared by delete_symbols, so it looks like a pointless check. This patch shaves about 10% off some .smx files sampled from the SourceMod tree. Bug: issue #358 Test: manual test
dvander
added a commit
that referenced
this issue
Sep 10, 2019
For some reason, natives had a weird relationship with the assembler. The code generator assigned an address based on a global counter, and then the assembler was responsible for building the table in the same order. (Similar nonsense exists with the staging counter, but we'll save that for another day.) Instead, emit sysreq instructions by name as we do for the call instruction. This allows us to build the table during code generation, simplifying another piece of the assembler pipeline. Bug: issue #358 Test: manual test
dvander
added a commit
that referenced
this issue
Sep 10, 2019
…embly. The two-pass system in the compiler is able to roughly guess which functions are not used, but it's a poor guess. Stocks called by other stocks, including recursive functions, all get compiled in. We can fix this by traversing the call graph to determine the actual live set. Note, the assembler now skips functions with uDEFINED but not uPUBLIC. uDEFINED is cleared by delete_symbols, so it looks like a pointless check. This patch shaves about 10% off some .smx files sampled from the SourceMod tree. Bug: issue #358 Test: manual test
dvander
added a commit
that referenced
this issue
Sep 10, 2019
…embly. The two-pass system in the compiler is able to roughly guess which functions are not used, but it's a poor guess. Stocks called by other stocks, including recursive functions, all get compiled in. We can fix this by traversing the call graph to determine the actual live set. Note, the assembler now skips functions with uDEFINED but not uPUBLIC. uDEFINED is cleared by delete_symbols, so it looks like a pointless check. This patch shaves about 10% off some .smx files sampled from the SourceMod tree. Bug: issue #358 Test: manual test
dvander
added a commit
that referenced
this issue
Sep 10, 2019
…embly. The two-pass system in the compiler is able to roughly guess which functions are not used, but it's a poor guess. Stocks called by other stocks, including recursive functions, all get compiled in. We can fix this by traversing the call graph to determine the actual live set. Note, the assembler now skips functions with uDEFINED but not uPUBLIC. uDEFINED is cleared by delete_symbols, so it looks like a pointless check. This patch shaves about 10% off some .smx files sampled from the SourceMod tree. Bug: issue #358 Test: manual test
|
Thanks for the report, this was a good catch. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Bug: So, i wrote some lib functions that required using of recursion call. Functions not used but stay to compile (not small functions, make +8k bytes to plugin code).
Tested with 1.9.0.6270, 1.10.0.6435 and latest sourcepawn master branch build.
The text was updated successfully, but these errors were encountered: