This is a beta for the upcoming 1.0.0 release. Major changes since the last beta: * Added support for LLVM 12 and 13 * Support for (previously deprecated) LLVM 3.5, 3.6 and 3.7 has been removed * Added support for fast-math flags * Added support for setting calling convention on functions * Experimental support for atomicrmw instruction * Experimental support for AMD GPU code generation * Build process for the released binaries has been improved and updated * Nix derivation has been substantially upgraded * Fixes for performance regressions in CUDA code generation since LLVM 3.8 * Various long-standing documentation issues have been addressed