Hi Jeff
Thanks for the review!
> Note this is more general than just FP related addresses. The variable
> name in that routine isn't great in that regard as it would seem to
> imply this code is only used on frame pointer based addresses.
The address being legitimized is a virtual-stack-vars based address rather
than a frame-pointer one, so the "fp" name is misleading. I've renamed it
to "stack_base" in v1.
> Note that riscv_add_offset handles SMALL_OPERAND just fine. So it seems
> like we could call it unconditionally [...]
> reg1 = force_reg (Pmode, riscv_add_offset (NULL, fp, offset));
> Can you try that?
Done. Dropping the SMALL_OPERAND conditional and calling
riscv_add_offset unconditionally produces identical code for the small
offset case and keeps the large offset case in a single pseudo, so v1
now just does:
reg1 = force_reg (Pmode, riscv_add_offset (NULL, stack_base, offset));
This makes the patch considerably simpler.
> Finally, how was this tested?
Tested on riscv64-unknown-linux-gnu:
- Full GCC regression run with no new failures; the new
frame-indexed-basec tests pass.
- SPEC CPU 2017: on 505.mcf_r the change gives the expected ~3%
reduction in dynamic instruction count.
I'll send v1 shortly.
Best regards,
Wang Jue
