@charliermarsh: I hadn't really thought of this before but it seems like we can use thin LTO for the instrumented binary we build durin…
Summary
Charlie Marsh 发现可以在 PGO 过程中构建的插桩二进制文件上使用 thin LTO，从而显著缩短构建时间，且最终二进制文件的性能似乎不受影响。
View Cached Full Text
Cached at: 10/10/26, 02:54 PM
I hadn’t really thought of this before but it seems like we can use thin LTO for the instrumented binary we build during PGO, which significantly reduces build time at seemingly no performance cost in the final binary. https://t.co/qNCZjSUqk5
Similar Articles
@charliermarsh: We now use full LTO for CPython 3.12 and later in python-build-standalone, which should modestly improve runtime perfor…
python-build-standalone now uses full LTO for CPython 3.12 and later, which should modestly improve runtime performance. CPython defaults to ThinLTO when building with clang on these versions.
@charliermarsh: Is Tibo talking about my small-string optimization work?
Charlie Marsh asks if Tibo's praise of GPT-5.6 Sol's performance and efficiency improvements is referring to his small-string optimization work.
Hooking into the Go Toolchain
A deep-dive tutorial on Go's -toolexec flag, showing how to intercept every step of the go build pipeline to time, rewrite, and instrument compilation, drawing on garble and otelc for real-world examples.
Leaving performance on the table
A technical blog post demonstrating how Profile-Guided Optimization (PGO) with LLVM can significantly improve binary performance beyond standard -O3 and LTO, using SQLite as a benchmark.
@BohuTANG: After quantification, it turns out that pi's current 4 tools (read / bash / edit / write) plus glob and grep can go up another level. These two are high-frequency tools, but relying on the LLM to piece together a usable bash to replace them is actually quite difficult — it often...
BohuTANG found through quantitative analysis that adding dedicated grep and glob functions to the pi tool reduces the detours of LLM-generated bash and improves efficiency.