Should the macro __riscv_c and __riscv_compressed defined for Zca?

[KaiYG]

Should the macro __riscv_c (and __riscv_compressed) defined when only Zca is enabled? Also, seems GCC and Clang behave differently in this case, e.g., for -march=rv32i_zca, GCC: undefined __riscv_c, while defined __riscv_compressed LLVM: undefined both.

[cmuellner]

Related questions:

• https://github.com/riscv-non-isa/riscv-asm-manual/issues/91
• https://github.com/riscv/riscv-isa-manual/pull/1121

[kito-cheng]

__riscv_compressed was deprecated, but I guess we should consider redefine that as defined if any extension provided compressed instruction is available? We have C and Zca now, but not guarantee we won't have more independent Zc* extensions .

cc @a4lg

[a4lg]

While deprecated (according to @kito-cheng), considering the context of __riscv_compressed, I think existence of the compressed instructions is better to be reported by Zca, as well as C (though I'm not sure whether I can use any stronger words like "should").

[KaiYG]

As the Zc spec v1.0.4-2 added the rule that C implies Zca, I think existence of the compressed instructions can be reported by __riscv_zca now.

[cmuellner]

Thanks for the update and closing the ticket!