mbox series

[v2,00/16] RISC-V: assorted fixes and (hopefully) improvements

Message ID c32643f7-c8b2-4a0e-9eeb-4ba00369b147@suse.com
Headers
Series RISC-V: assorted fixes and (hopefully) improvements |

Message

Jan Beulich May 15, 2026, 1:27 p.m. UTC 
  There may not be many dependencies among the patches, but the issues
were all noticed more or less together (i.e. while addressing one,
the next one popped up). I've not made an attempt at sorting the V vs
Z?inx issue, as per Andrew's request.

Quite a few further points are made in remarks in individual patches.
Input there is very welcome.

Jiawei kindly reviewed v1, and I committed a few patches from there
which I assumed were pretty sure to be uncontroversial. For the others
I'd prefer RISC-V maintainer approval, yet I'm not going to wait
indefinitely.

There aren't many changes in v2, the most prominent perhaps being the
new 1st patch. See individual changes for what changed, if anything.

01: rename operand descriptor O4
02: RISC-V: fold redundant code in riscv_ip()
03: RISC-V: add dedicated vector arithmetic .insn forms
04: RISC-V: match_*() improvements
05: RISC-V: EEW64 checking
06: bfd/RISC-V: Zvfbfwma implies Zvfbfmin
07: RISC-V: drop FCVT.Q.L{,U} forms with rounding mode operand
08: RISC-V: make FP rounding mode an optional argument
09: RISC-V: check operands for Zdinx in RV32
10: RISC-V: riscv_set_arch() can fail
11: RISC-V: riscv_parse_subset_t's isa_spec is only parser input
12: RISC-V: error handling for subset parsing
13: RISC-V: avoid buffer underrun in subset parsing
14: RISC-V: drop dead code from subset parsing
15: bfd/RISC-V: move _bfd_riscv_elf_merge_private_bfd_data()
16: RISC-V/gas: .attribute vs .insn

I'd also like to point out that [1] continues to be pending. I'm
far from insisting that that patch be taken, but something needs
doing about the issue.

Jan

[1] https://sourceware.org/pipermail/binutils/2023-March/126601.html

Comments

Jan Beulich June 5, 2026, 9:27 a.m. UTC | #1 
On 15.05.2026 15:27, Jan Beulich wrote:
> There may not be many dependencies among the patches, but the issues
> were all noticed more or less together (i.e. while addressing one,
> the next one popped up). I've not made an attempt at sorting the V vs
> Z?inx issue, as per Andrew's request.
> 
> Quite a few further points are made in remarks in individual patches.
> Input there is very welcome.
> 
> Jiawei kindly reviewed v1, and I committed a few patches from there
> which I assumed were pretty sure to be uncontroversial. For the others
> I'd prefer RISC-V maintainer approval, yet I'm not going to wait
> indefinitely.
> 
> There aren't many changes in v2, the most prominent perhaps being the
> new 1st patch. See individual changes for what changed, if anything.
> 
> 01: rename operand descriptor O4
> 02: RISC-V: fold redundant code in riscv_ip()
> 03: RISC-V: add dedicated vector arithmetic .insn forms
> 04: RISC-V: match_*() improvements
> 05: RISC-V: EEW64 checking
> 06: bfd/RISC-V: Zvfbfwma implies Zvfbfmin
> 07: RISC-V: drop FCVT.Q.L{,U} forms with rounding mode operand
> 08: RISC-V: make FP rounding mode an optional argument
> 09: RISC-V: check operands for Zdinx in RV32
> 10: RISC-V: riscv_set_arch() can fail
> 11: RISC-V: riscv_parse_subset_t's isa_spec is only parser input
> 12: RISC-V: error handling for subset parsing
> 13: RISC-V: avoid buffer underrun in subset parsing
> 14: RISC-V: drop dead code from subset parsing
> 15: bfd/RISC-V: move _bfd_riscv_elf_merge_private_bfd_data()
> 16: RISC-V/gas: .attribute vs .insn

And some follow-on-s to observations noted in some of the above:

17: warn about non-boolean unaligned-access attribute
18: warn about non-power-of-2 stack-align attribute
19: check operands for Zqinx

Jan