Incorrect march names when building toolchain with `--with-multilib-generator` option

[por3bski]

Building toolchain with following custom multilib configuration results in unexpected march names:

./configure --prefix=/opt/riscv --with-multilib-generator="rv32imac-ilp32--;rv64imac-lp64--"

riscv64-unknown-elf-gcc --print-multi-lib
.;
rv32imac_zaamo_zalrsc/ilp32;@march=rv32imac_zaamo_zalrsc@mabi=ilp32
rv64imac_zaamo_zalrsc/lp64;@march=rv64imac_zaamo_zalrsc@mabi=lp64

Each march has _zaamo_zalrsc suffix.

When compiling with default multilib configuration, names are as I would expect:

./configure --prefix=/opt/riscv --enable-multilib

riscv64-unknown-elf-gcc --print-multi-lib
.;
rv32i/ilp32;@march=rv32i@mabi=ilp32
rv32im/ilp32;@march=rv32im@mabi=ilp32
rv32iac/ilp32;@march=rv32iac@mabi=ilp32
rv32imac/ilp32;@march=rv32imac@mabi=ilp32
rv32imafc/ilp32f;@march=rv32imafc@mabi=ilp32f
rv64imac/lp64;@march=rv64imac@mabi=lp64
rv64imafdc/lp64d;@march=rv64imafdc@mabi=lp64d 

Are these differences correct behavior?

[pz9115]

Hi @por3bski, I think the --enable-multilib was missing adding imply relationship for a extension, I will update this with a new pr, thanks for you report it.

[pz9115]

Hi @por3bski, I checked the process again, the --with-multilib-generator actually dealed by gcc part, and the --enable-multilib is handled by configure file, gcc added additional imply info to a extension. In fact, they are equivalent, since gcc will aslo replace a with a_zaamo_zalrsc.

The current --enable-multilib option is more secure, since old version gcc can not recognize the zaamo&zalrsc extension, so just ignore this difference.

[TommyMurphyTM1234]

The current --enable-multilib option is more secure, since old version gcc can not recognize the zaamo&zalrsc extension, so just ignore this difference.

Just to clarify, the above means that the following is true?

Are these differences correct behavior?

[pz9115]

The current option is more secure, since old version gcc can not recognize the extension, so just ignore this difference.--enable-multilib``zaamo&zalrsc

Just to clarify, the above means that the following is true?

Are these differences correct behavior?

Yes, I think, actually now a equals to a+zaamo+zalrsc.

[TommyMurphyTM1234]

Thanks @pz9115 - is there any link to information about this change?

[pz9115]

Thanks @pz9115 - is there any link to information about this change?

Here are some refer links——

Spec: https://github.com/riscv/riscv-zaamo-zalrsc

ISA manual PR: riscv/riscv-isa-manual#1304

[TommyMurphyTM1234]

Thanks a lot @pz9115. 👍