SYSTEM Cited by 2 sources

Clang

Clang is the LLVM-project C / C++ / Objective-C compiler frontend. Canonical project documentation at clang.llvm.org.

This wiki page is a stub covering clang's relevance to FDO tooling. For the LLVM post-link optimiser, see systems/llvm-bolt; for general C++ build-system patterns, see concepts/monorepo and the various code-analysis systems (systems/include-what-you-use, etc.).

PGO flags

Clang implements profile-guided optimization via a two-phase build:

• -fprofile-generate[=<dir>] — phase 1. Instruments the binary. Profile data written on program exit.
• -fprofile-use=<path> / -fprofile-instr-use=<path> — phase 2. Consumes the profile for code-layout, inlining, and branch-hint decisions.
• -fprofile-sample-use=<path> — AutoFDO / CSSPGO variant. Consumes a sampling profile (from Linux perf or Strobelight) instead of the instrumented .profdata.

See instrumented-vs-sampling-profile for the trade-off.

Role on this wiki

• The compile-time FDO implementer — PGO / CSSPGO / AutoFDO all live behind clang flags.
• Pairs with systems/llvm-bolt as compile-time + post-link siblings in the FDO family.
• Redpanda's 26.1 PGO rollout uses clang (Source: sources/2026-04-02-redpanda-supercharging-streaming-with-profile-guided-optimization) — the Tier-3 canonical wiki example of clang PGO delivering ~15% CPU reduction + 47% p999 latency reduction on a streaming-broker binary.

Seen in

• sources/2026-04-02-redpanda-supercharging-streaming-with-profile-guided-optimization — Redpanda Streaming 26.1 PGO rollout; clang is the canonical build-time FDO implementer named in the post.
• sources/2025-03-07-meta-strobelight-a-profiling-service-built-on-open-source-technology — Meta's fleet-scale CSSPGO deployment through clang.

Related

• systems/llvm-bolt — the post-link sibling.
• systems/meta-bolt-binary-optimizer — Meta's BOLT perspective.
• systems/linux-perf — the sampling-profile source.
• systems/redpanda — Tier-3 adopter.
• profile-guided-optimization — the optimisation clang's PGO flags implement.
• feedback-directed-optimization — the umbrella.
• instrumented-vs-sampling-profile — the two profile-collection shapes clang supports.
• hot-cold-code-splitting / concepts/cache-locality — the transformations clang PGO applies.
• pgo-for-frontend-bound-application — the canonical apply pattern.