[llvm-bugs] [Bug 25685] New: [ppc] bad scheduling for special register access instructions

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=25685

            Bug ID: 25685
           Summary: [ppc] bad scheduling for special register access
                    instructions
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: PowerPC
          Assignee: unassignedbugs at nondot.org
          Reporter: carrot at google.com
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

This is a very common instruction sequence. 

One example is Perl_sv_setsv_flags from perlbench. On power8, llvm generated
Perl_sv_setsv_flags consumes 4.26% of run time, gcc generated function consumes
3.5% of run time.

perf annotate shows following instructions and cycles of llvm generated code:

      │      00000000100c67d0 <Perl_sv_setsv_flags>:                            
       │        addis  r2,r12,26                                                
       │        addi   r2,r2,31576                                              
  4.95 │        mfcr   r12                                                      
  4.18 │        mflr   r0                                                       
  0.07 │        std    r31,-8(r1)                                               
       │        std    r0,16(r1)                                                
  0.02 │        stw    r12,8(r1) 
                ...                         

The two instructions mfcr/mflr consume many time, these are slow instructions,
but the results are used immediately, causes stalling.

For comparison following is gcc generated code:

       │      00000000100d02f0 <Perl_sv_setsv_flags>:                           
       │        addis  r2,r12,24                                                
       │        addi   r2,r2,21112                                              
  1.63 │        mflr   r0                                                       
       │        cmpld  cr7,r4,r3                                                
  1.57 │        std    r30,-16(r1)                                              
  0.01 │        std    r31,-8(r1)                                               
  1.66 │        mfocrf r12,8                                                    
       │        std    r26,-48(r1)                                              
       │        std    r27,-40(r1)                                              
  0.02 │        mr     r31,r3                                                   
       │        mr     r30,r4                                                   
       │        std    r28,-32(r1)                                              
  2.64 │        std    r29,-24(r1)                                              
       │        stw    r12,8(r1)                                                
       │        std    r0,16(r1)   
                ...                        

It has much better scheduling of mflr/mfocrf and corresponding usage
instructions.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20151201/481091cd/attachment-0001.html>