Add -mabi=ilp32e #110
Add -mabi=ilp32e #110
Conversation
|
Thanks, Kito. I probably won't have time to look at this for a bit, but I appreciate the help :) |
There was a problem hiding this comment.
ABI_SPEC in riscv.h also needs to be updated, to get the library paths right for ilp32e.
There is also a MUSL_ABI_SUFFIX in linux.h, but I think rv32e and linux are contradictory, though maybe we should emit an error somewhere if someone tries to use rv32e with linux. This part is optional.
gcc/config.gcc
Outdated
| @@ -4056,11 +4057,12 @@ case "${target}" in | |||
| # pick a default based on the ISA, preferring soft-float | |||
| # unless the D extension is present. | |||
| case "${with_abi}" in | |||
| ilp32 | ilp32f | ilp32d | lp64 | lp64f | lp64d) | |||
| ilp32 | ilp32f | ilp32d | lp64 | lp64f | lp64d | lp64q) | |||
There was a problem hiding this comment.
ilp32e needs to be in this list of valid abis, otherwise you are overriding the user's choice, and won't give an error if the user makes a mistake with non-matching arch/abi.
There was a problem hiding this comment.
Fixed, that's my mistake when rebase with Q-extension...
gcc/config/riscv/riscv.c
Outdated
| if (riscv_abi == ABI_ILP32E) | ||
| { | ||
| for (int r = 16; r <= 31; r++) | ||
| call_used_regs[r] = 1; |
There was a problem hiding this comment.
An extra space before the =, minor nit.
kito-cheng
force-pushed
the
mabi-rv32e
branch
from
a45f189
to
13be76f
Compare
|
@palmer-dabbelt @jim-wilson thanks for your review :) |
|
I am unfamiliar with github workflow in general, and riscv github workflow in particular. I was assuming that the person who submitted the patch would merge them after approval. This is how FSF binutils/gcc work. If you are waiting for me to merge them, let me know. |
riscv-non-isa/riscv-toolchain-conventions#3