Re: [Committed] riscv: relax splitter restrictions for creating pseudos

[Jeff Law <jeffreyalaw@gmail.com>]

On 4/25/23 14:20, Vineet Gupta wrote:
>> Just a note, there are some regressions in this table.  For example xz 
>> input #2.  So to be fair when making comparisons it's probably worth 
>> noting the regressions as well as improvements.
> 
> Fair point, but in my defense I dropped anything under 0.1 % (including 
> improvements).
No worries.


>> Anyway, my results are in line with yours. Given the instruction 
>> counts and known IPC, I think we're taking about a 2.5% hit on 
>> deepsjeng and about a 1% hit on leela and x264#2 comparing our 
>> internal gcc-12 vs gcc-13 trees.
> 
> I think you are referring to the bigger deepsjeng regression due to 
> commit 2e886eef7f2b5a which introduced the splitter movconst_internal. 
> This patch is about a small improvement to all splitters which reduces 
> that perf loss slightly (but does regress some tests too as pointed above).
Correct.  I'm referring to the larger gcc-12 to gcc-13 comparison, not 
your patch in my comment about regressions.  Sorry for not being clear 
about that.


Jeff