Add Zvfbfa, Zvfofp8min, Zvfofp4min support

[drom]

This PR implements Zvfbfa, Zvfofp8min, Zvfofp4min extensions according to spec proposals:

• https://github.com/aswaterman/riscv-misc/blob/main/isa/zvfbfa.adoc
• https://github.com/aswaterman/riscv-misc/blob/main/isa/zvfofp8min.adoc
• https://github.com/aswaterman/riscv-misc/blob/main/isa/zvfofp4min.adoc

[nadime15]

That’s off topic, but it would be nice if additions to the SoftFloat library could be pushed upstream to the official Berkeley SoftFloat repo, so we have a common place to share improvements, report bugs, and keep things consistent.

By the way, there are already functions (most of the bf16 stuff) that exist only in Spike and aren’t part of the official repo.

[aswaterman]

@nadime15 indeed, that is a topic for another day. (It isn't sustainable for RVIA to keep modifying softfloat at ucb-bar; we'll need to fork it to riscv-software-src. I will take that responsibility, but it'll be a few months before I work on that.)

[aswaterman]

Thanks @drom, this looks good overall. I'll merge after my minor comments are addressed.

[aswaterman]

It's getting pretty close. I think we can merge it once my remaining open comments are addressed.

[aswaterman]

@mslijepc I think we might accidentally be relying on Zvfh being present when Zvfbfa is present. My concern stems from this:

#define VI_VFP_COMMON \
  VI_VFP_BASE; \
  require((P.VU.vsew == e16 && p->extension_enabled(EXT_ZVFH)) || \
          (P.VU.vsew == e32 && p->get_isa().get_zvf()) || \
          (P.VU.vsew == e64 && p->get_isa().get_zvd())); \

Don't we need to modify that to something like P.VU.vsew == e16 && p->extension_enabled(P.VU.altfmt ? EXT_ZVFBFA : EXT_ZVFH)?

[mslijepc]

@aswaterman I agree. I was thinking to use

 #define require_zvfbfa \
-  require(P.VU.altfmt == 0 || p->extension_enabled(EXT_ZVFBFA)); \
+  require_extension(P.VU.altfmt ? EXT_ZVFBFA : EXT_ZVFH);

and maybe

 #define require_zvfbfa_zvfhmin \
-  require(P.VU.altfmt == 0 || p->extension_enabled(EXT_ZVFBFA)); \
+  require_extension(P.VU.altfmt ? EXT_ZVFBFA : EXT_ZVFHMIN);

there and also in other cases you mentioned. What do you think? Feel free to propose better naming for macros

[mslijepc]

@aswaterman @youngar I addressed all the comments. Could you please double check if it's all good in PR (except conflicts we have)?

[aswaterman]

@aswaterman I agree. I was thinking to use

 #define require_zvfbfa \
-  require(P.VU.altfmt == 0 || p->extension_enabled(EXT_ZVFBFA)); \
+  require_extension(P.VU.altfmt ? EXT_ZVFBFA : EXT_ZVFH);

and maybe

 #define require_zvfbfa_zvfhmin \
-  require(P.VU.altfmt == 0 || p->extension_enabled(EXT_ZVFBFA)); \
+  require_extension(P.VU.altfmt ? EXT_ZVFBFA : EXT_ZVFHMIN);

there and also in other cases you mentioned. What do you think? Feel free to propose better naming for macros

This looks right, but I don't see it in the PR.

[aswaterman]

@aswaterman I agree. I was thinking to use

 #define require_zvfbfa \
-  require(P.VU.altfmt == 0 || p->extension_enabled(EXT_ZVFBFA)); \
+  require_extension(P.VU.altfmt ? EXT_ZVFBFA : EXT_ZVFH);

and maybe

 #define require_zvfbfa_zvfhmin \
-  require(P.VU.altfmt == 0 || p->extension_enabled(EXT_ZVFBFA)); \
+  require_extension(P.VU.altfmt ? EXT_ZVFBFA : EXT_ZVFHMIN);

there and also in other cases you mentioned. What do you think? Feel free to propose better naming for macros

This looks right, but I don't see it in the PR.

To clarify, these do make sense, but only if they are used in the right places, e.g. inside the CHECK32 macros, not in the global scope of an instruction.

[aswaterman]

This looks good to go. Thanks for the hard work, @drom @youngar @mslijepc!