Support RISC-V p-ext-proposal v0.9.12 for RV64

[jason01180118]

See the previous PR #637

Hello, we’re the team from Programming Langauge Lab, NTHU (National Tsing-Hua University), Taiwan.

This PR implements all instructions in the latest P-Extension specification (link)

Branches are split for RV32 and RV64, since some instructions may conflict between architectures.

[arromanoff]

v0.9.12 is quite old. AFAIK P extension is not ratified and still under development (I've seen a very recent v0.11 update here). I am not really against this change but I am concerned about maintenance. Is P extension relatively stable now? I'm afraid I am lost on the subject of P extension standardization. Please correct me if I am wrong here

[chihminchao]

Based on the reference link, the PR should use latest 0.18 spec located in jhauser's personal folder. I guess the 0.9.12 is a plus of final draft version, 0.9.11

[ChiaHuiSu]

Hello all, the version number 0.9.12 is from Rich Fuler's talk(at 5:43) in RISC-V Summit 2025, North America, and it's the same as the latest 0.18 spec lacated jhauser's personal folder.

[arromanoff]

The versioning situation seems very confusing to me (0.12 / 0.9.12 / 0.92 / 0.18) but it actually looks like all of them are basically the same. I am still a little concerned about merging such often-changing extension. @aswaterman what are your thoughts on this? Is this the version of zpa you were talking about here?

[chihminchao]

@jason01180118 There are some instructions having the same name but different encoding for rv32 and rv64. How is it handled in these PRs ?

ex: SATI/SRARI/USATI

[jason01180118]

Since spike does not permit instructions with the same name but different encodings within a single target, we split the RV32 and RV64 implementations into separate branches.
For example, the encoding of SATI differs between RV64 and RV32:

RV64 (#2211 ): MATCH_SATI = 0xe400401b, MASK_SATI = 0xfc00707f

RV32 (#2210 ): MATCH_SATI = 0xe200401b, MASK_SATI = 0xfe00707f

[chihminchao]

It is just my thinking to use different name sto tell them if the encoding doesn't be changed.
ex
sati_rv32, sati_rv64

[chihminchao]

@jason01180118 There are some compilation warnings related to macro expansion

../riscv/p_ext_macros.h:335:25: warning: second operand of conditional expression has no effect [-Wunused-value]
  335 |       P_USAT(BIT, p_res); \
      |                         ^
../riscv/p_ext_macros.h:494:3: note: in expansion of macro ‘P_REDUCTION_ULOOP_END’
  494 |   P_REDUCTION_ULOOP_END(BIT, IS_SAT)
      |   ^~~~~~~~~~~~~~~~~~~~~
../riscv/insns/pabdsumau_b.h:1:1: note: in expansion of macro ‘P_REDUCTION_ULOOP’
    1 | P_REDUCTION_ULOOP(64, 8, true, false, {
      | ^~~~~~~~~~~~~~~~~
../riscv/p_ext_macros.h:335:19: warning: second operand of conditional expression has no effect [-Wunused-value]
  335 |       P_USAT(BIT, p_res); \
      |                   ^
../riscv/p_ext_macros.h:513:20: note: in definition of macro ‘P_USAT’
  513 |   ((BIT) == 64) ? (R) : \
      |                    ^
../riscv/p_ext_macros.h:494:3: note: in expan

It happens with gcc 11, 12 , and 13.