[cfe-dev] "Optimized implementations"?

[Stefan Kanthak via cfe-dev]

"Craig Topper" <craig.topper at gmail.com> wrote;

> Clang never generates calls to ___paritysi2, ___paritydi2, ___cmpdi2, or
> ___ucmpdi2 on X86 so its not clear the performance of this matters at all.

So you can safely remove them for X86 and all the other targets where such
unoptimized code is never called!
But fix these routines for targets where they are called.

The statement does NOT make any exceptions, and it does not say
| ships unoptimized routines the compiler never calls
but
| optimized target-independent implementations

Stefan

BTW: do builtins like __builtin_*parity* exist?
     If yes: do they generate the same bad code?

> On Sun, Sep 6, 2020 at 12:31 PM Stefan Kanthak via cfe-dev <
> cfe-dev at lists.llvm.org> wrote:
> 
>> <https://compiler-rt.llvm.org/index.html> boasts:
>>
>> | The builtins library provides optimized implementations of this
>> | and other low-level routines, either in target-independent C form,
>> | or as a heavily-optimized assembly.
>>
>> Really?
>>
>> Left: inperformant code shipped in    # Right: slightly improved code,
>>       clang_rt.builtins-*             #        which the optimiser REALLY
>>                                       #        should have generated
>>
>> ___cmpdi2:
>>         mov     ecx, [esp+16]         #       mov     ecx, [esp+16]
>>         xor     eax, eax              #       xor     eax, eax
>>         cmp     [esp+8], ecx          #       cmp     ecx, [esp+8]
>>         jl      @f                    #       jg      @f
>>         mov     eax, 2                #       mov     eax, 2
>>         jg      @f                    #       jl      @f
>>         mov     ecx, [esp+4]          #
>>         mov     edx, [esp+12]         #       mov     ecx, [esp+12]
>>         mov     eax, 0                #       xor     eax, eax
>>         cmp     ecx, edx              #       cmp     ecx, [esp+4]
>>         jb      @f                    #       ja      @f
>>         cmp     edx, ecx              #
>>         mov     eax, 1                #
>>         adc     eax, 0                #       adc     eax, 1
>> @@:                                   # @@:
>>         ret                           #       ret
>>
>>                                       # 3 instructions less, 10 bytes saved
>>
>> ___ucmpdi2:
>>         mov     ecx, [esp+16]         #       mov     ecx, [esp+16]
>>         xor     eax, eax              #       xor     eax, eax
>>         cmp     [esp+8], ecx          #       cmp     ecx, [esp+8]
>>         jb      @f                    #       ja      @f
>>         mov     eax, 2                #       mov     eax, 2
>>         ja      @f                    #       jb      @f
>>         mov     ecx, [esp+4]          #
>>         mov     edx, [esp+12]         #       mov     ecx, [esp+12]
>>         mov     eax, 0                #       xor     eax, eax
>>         cmp     ecx, edx              #       cmp     ecx, [esp+4]
>>         jb      @f                    #       ja      @f
>>         cmp     edx, ecx              #
>>         mov     eax, 1                #
>>         adc     eax, 0                #       adc     eax, 1
>> @@:                                   # @@:
>>         ret                           #       ret
>>
>>                                       # 3 instructions less, 10 bytes saved
>>
>>
>> Now properly written code, of course branch-free, faster and shorter:
>>
>> # Copyright (C) 2004-2020, Stefan Kanthak <stefan.kanthak at nexgo.de>
>>
>> ___cmpdi2:
>>         mov     ecx, [esp+4]
>>         mov     edx, [esp+12]
>>         cmp     ecx, edx
>>         mov     eax, [esp+8]
>>         sbb     eax, [esp+16]
>>         setl    ah
>>         cmp     edx, ecx
>>         mov     edx, [esp+16]
>>         sbb     edx, [esp+8]
>>         setl    al
>>         sub     al, ah
>>         movsx   eax, al
>>         inc     eax
>>         ret
>>
>> ___ucmpdi2:
>>         mov     ecx, [esp+4]
>>         mov     edx, [esp+12]
>>         cmp     ecx, edx
>>         mov     eax, [esp+8]
>>         sbb     eax, [esp+16]
>>         sbb     eax, eax
>>         cmp     edx, ecx
>>         mov     edx, [esp+16]
>>         sbb     edx, [esp+8]
>>         adc     eax, 1
>>         ret
>>
>>
>> AGAIN:
>> Remove every occurance of the word "optimized" on the above web page.
>>
>> 'nuff said
>> Stefan
>> _______________________________________________
>> cfe-dev mailing list
>> cfe-dev at lists.llvm.org
>> https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
>>