[LLVMdev] Opinions Wanted: New asm Comments

[Chris Lattner]

On Jul 12, 2011, at 7:45 AM, David A. Greene wrote:

> Chris Lattner <clattner at apple.com> writes:
> 
>>> Is a new flag appropriate for this or should I just put it under
>>> -asm-verbose with everything else?  If we want a new flag, does anyone
>>> have a spelling they'd recommend?  I'm not particularly fond of
>>> -asm-pattern but it was the best concise name my uncreative brain could
>>> conjur up.
>> 
>> How do you plan to implement this?  This is a compiler-hacker-only
>> feature, can it be protected fully in !NDEBUG code?
> 
> The codegen portion certainly can be protected with !NDEBUG.
> 
> The change has two parts: a TableGen enhancement to output a table of
> pattern strings and some code to pair up MachineInstrs with the
> appropriate index into the pattern string table and an AsmPrinter
> portion to actually print the comments.
> 
> The latter can be easily controlled via NDEBUG.  The former is more
> difficult since at the time TableGen is run we don't know how the rest
> of the compiler will be built.
> 
> How would you prefer this work?  Even if the output were controlled by
> NDEBUG, I feel the added pattern comments make the asm file too
> cluttered even for day-to-day compiler developers.  This is really a
> feature to debug instruction selection problems.  That's why I put it
> under the control of a separate option.

It's not really clear how this should work.  I'm primarily concerned that it will cause substantiate table/code bloat that doesn't make sense for a shipping compiler.  I agree that not all compiler hackers will want to see it, I'd suggest adding a cl::opt that is only even available when built with assertions on.

-Chris