[llvm-commits] [llvm] r132882 - in /llvm/trunk: lib/CodeGen/AsmPrinter/AsmPrinter.cpp lib/Target/X86/X86InstrInfo.cpp test/CodeGen/Thumb2/machine-licm.ll

[Jakob Stoklund Olesen]

On Jun 13, 2011, at 10:16 AM, Dan Gohman wrote:

> On Jun 12, 2011, at 12:57 AM, Jakob Stoklund Olesen wrote:
> 
>> 
>> On Jun 12, 2011, at 12:38 AM, Rafael Ávila de Espíndola wrote:
>> 
>>>> The double-CFG-edge hack was added as a compile time optimization.
> 
> It was also a simplification. Before that, several places in codegen that modified
> successor lists needed extra code to avoid duplicates.

Do we need to add that back in?

It seems to me that we can accept double CFG edges as long as it only happens on degenerate IR input.

I am more concerned with adding double edges on 'normal' code. They look a lot like critical edges, disabling optimizations.

> Additional speedup comes from the assembler spending less time parsing
> and managing extra symbols.  But this is all less important now that assembly
> printing is no longer a hot path.

Rafael, are you emitting more labels now, or does the extra AnalyzeBranch call take care of that?

/jakob