The LLVM 21 compiler stack was branched today as release preparations get underway for shipping this next half-year compiler release as stable in late August or early September. In turn that now opens up LLVM 22 for development.
LLVM 21 is introducing the AMD GFX1250 target for what is expected to be an APU with an RDNA4 refresh (RDNA 4.5?) while the additions for the GFX1250 target remain ongoing. There are also other AMDGPU LLVM back-end enhancements for helping their ROCm stack and other libc-for-GPU efforts. There is also now -mcpu=gb10 support for NVIDIA’s GB10 Superchip, various RISC-V improvements, many RISC-V back-end enhancements, LLDB debugger improvements, and more.
On the Clang 21 C/C++ compiler front-end side there are new LLVM optimizations around pointer arithmetic on null pointers, new C++2c feature additions, various other C++ language features added, new C warnings, additional C2y feature work, and many compiler diagnostic enhancements. Clang 21 also addresses Intel’s changes to AVX10 with dropping the AVX10-256 support to now make 512-bit support unconditional for the maximum vector register size.
More details on the LLVM 21 branching today via the LLVM.org Discourse. The LLVM 21 release candidate should be out in the coming days while LLVM 21.1 stable compiler stack should be ready in roughly six weeks. LLVM/Clang 22 stable (LLVM 22.1 / Clang 22.1) in turn will be out around next March.
