Optimize Compilation #15
Conversation
Xeratec
force-pushed
the
pr/optimize_compilation
branch
from
601b31d to
e75301f
Compare
Xeratec
force-pushed
the
pr/optimize_compilation
branch
from
e75301f to
37d16e0
Compare
| @@ -32,6 +32,7 @@ project(chimera-sdk LANGUAGES C ASM) | |||
| set(ABI ilp32) | |||
| set(ISA_CLUSTER_SNITCH rv32im) | |||
| set(ISA_HOST rv32imc) | |||
| set(DISASSEMBLE_LIBRARIES CACHE BOOL OFF) | |||
There was a problem hiding this comment.
I'd argue it almost always makes sense to just provide the objdump for all binaries. If you don't need it you won't feel the overhead, if you need it, you don't want to go digging in the make flow.
There was a problem hiding this comment.
Disassembling of the libraries is enabled per default. This flag only triggers to also disassemble the static libraries.
|
|
||
| target_link_options(${TEST_NAME} | ||
| PUBLIC | ||
| -Wl,--print-memory-usage |
There was a problem hiding this comment.
--print-memory-usage is GCC only, irc.
There was a problem hiding this comment.
Since release 17.0.1 it is also would also be supported by the LLD linker. However, with the current flow, we use the LD linker for the LLVM and GCC build flow. Hence, I do not see an issue, and it is tested with the current setup, using toolchain_llvm.cmake.
References:
Xeratec
force-pushed
the
pr/optimize_compilation
branch
from
37d16e0 to
7f199aa
Compare
Xeratec
force-pushed
the
pr/optimize_compilation
branch
from
7f199aa to
96b4a66
Compare
Changelog
This PR tries to reduce the binary size and improve the debugging functionality of the generated code. One aspect is to separate create function and data sections to allow garbage collection during linking, which significantly reduces the binary size.
Bugs
The separate section works on the object and library level, but somehow the linker merges everything into the
.textor.datasections for the final executable and does not keep the individual subsections.Added
Changed
Fixed
ToDo