[PATCH] D153159: [RISCV] Add errors for mixing Zcmp with C/Zcd and D.

[Xinlong Wu via Phabricator via llvm-commits]

VincentWu added inline comments.


================
Comment at: llvm/lib/Support/RISCVISAInfo.cpp:900
 
-  if (HasZcmt && HasD && HasC)
+  if ((HasZcmt || Exts.count("zcmp")) && Exts.count("d") &&
+      (HasC || Exts.count("zcd")))
----------------
I found there might be a problem about predicate of `Zcd`

Zcd contains: `c.fld, c.fldsp, c.fsd, c.fsdsp`, witch is a subset of D ext.
it directly depends on Zca instead of C ext.

Thus, `c.fld, c.fldsp, c.fsd, c.fsdsp` are valid when `HasZcd || (HasD && HasC)`

Zcmp is directly incompatible with Zcd, so there should be two error msg

`'Zcmp' extension is incompatible with 'Zcd'` 
and `'Zcmp' extension is incompatible with both 'C' and 'D' are enabled`


================
Comment at: llvm/lib/Support/RISCVISAInfo.cpp:900
 
-  if (HasZcmt && HasD && HasC)
+  if ((HasZcmt || Exts.count("zcmp")) && Exts.count("d") &&
+      (HasC || Exts.count("zcd")))
----------------
VincentWu wrote:
> I found there might be a problem about predicate of `Zcd`
> 
> Zcd contains: `c.fld, c.fldsp, c.fsd, c.fsdsp`, witch is a subset of D ext.
> it directly depends on Zca instead of C ext.
> 
> Thus, `c.fld, c.fldsp, c.fsd, c.fsdsp` are valid when `HasZcd || (HasD && HasC)`
> 
> Zcmp is directly incompatible with Zcd, so there should be two error msg
> 
> `'Zcmp' extension is incompatible with 'Zcd'` 
> and `'Zcmp' extension is incompatible with both 'C' and 'D' are enabled`
same problem also happens on MC layer.

I am trying to fix it by 
```
def HasCompressedDoubleFloat
    : Predicate<"Subtarget->hasStdExtZcd() || (Subtarget->hasStdExtC() && Subtarget->hasStdExtC())">,
                AssemblerPredicate<(any_of FeatureStdExtZcd, (all_of FeatureStdExtD, FeatureStdExtC)),
                                   "'Zcd' (Compressed Double-Precision Floating-Point Instructions) or "
                                   "both 'C' (Compressed Instructions) and "
                                   "'D' (Double-Precision Floating-Point)">;
```

but it seems `AssemblerPredicate` can't process nested conditions 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D153159/new/

https://reviews.llvm.org/D153159