LLVM 11 Feature Development Is Over With Many Changes

LLVM 11 feature development has ended with the code having been branched in Git this morning and the first release candidate expected shortly.

LLVM 11.0 was branched today in its mono repository including sub-projects like Clang. This branching is going as planned with aiming to ship LLVM 11.0 officially around 26 August.

Among the many changes found in this half-year update to the LLVM compiler stack includes:

- Clang now uses C17/GNU17 by default if no C standard version is specified. Previously it was C11.

- Clang 11 now supports -fstack-clash-protection for helping fight off possible stack clash attacks on x86/x86_64, s390x, and POWER 64-bit architectures.

- Clang can now parse the asm inline GNU C extension albeit is currently unhandled.

- Continued C++20 support.

- Clang -O behavior better matches GCC.

- Early bits around Intel AMX support.

- Support for newer ARMv8 instructions.

- AMD Navi 2 (Sienna Cichlid) support within the AMDGPU shader compiler as GFX1030.

- Radeon GCN offloading for OpenMP.

- Intel load hardening mitigation and SESES speculative execution pass.

- The AVR target for LLVM has been promoted from experimental status to an official back-end.

- Various WebAssembly target improvements.

- F18/Flang were merged into the LLVM 11 source tree albeit still under development.

- Intel SERIALIZE instruction support for Golden Cove.

- Support for the Marvell ThunderX3.

- Arm Cortex-A77 support.

- Fujitsu A64FX support.

- NVIDIA Carmel bits were finally upstreamed.

Benchmarks of LLVM Clang 11 will begin soon on Phoronix.