You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
The bitcode format writes each module to a temporary file, then reads and parses it #273
In the bc format of --compile-ir-format, the default, each source is compiled to a temporary .bc file. The analyzer then reads that file into a string, copies the string into a buffer and parses the buffer into its module (InputPipeline.cpp L907-L918 and L1155-L1198 at db7fb25). coretrace-compiler held that module in memory before it wrote the file.
CoreTrace/coretrace-compiler#265, a draft, adds compileToModule: it gives the module of a compilation, created in the caller's llvm::LLVMContext, and writes no output file.
Measured on macOS arm64 (M3, 8 cores) with LLVM 20.1.2, --format=json --timing, --jobs=1, no IR cache, medians of 3 to 5 runs. The analyzer at db7fb25 with coretrace-compiler built from its main branch is compared with the same analyzer taking each module from compileToModule:
In the bc format, take each module from compileToModule. With the compile IR cache, store the bitcode of that module.
With --timing, a compiled source then has no bitcode parse to report. The regression test check_compile_ir_format_switch recognizes the bc format by that log line, so it needs the line of the new path.
Activity
docs(input): say why the cache's bitcode keeps the order of uses
test(unit): check that a module read from the cache keeps the order o…
build(deps): update coretrace-compiler to v0.13.0
docs(changelog): say that coretrace-compiler is updated to v0.13.0