[x265] [PATCH 1 of 3] testbench: support float ret value

[Steve Borho]

On Fri, Apr 25, 2014 at 9:36 PM, chen <chenm003 at 163.com> wrote:
> At 2014-04-25 22:41:12,"Steve Borho" <steve at borho.org> wrote:
>
>>On Thu, Apr 24, 2014 at 10:02 PM, Min Chen <chenm003 at 163.com> wrote:
>>> # HG changeset patch
>>> # User Min Chen <chenm003 at 163.com>
>>> # Date 1398394872 -28800
>>> # Node ID 9505724b11bc944eaaefd35d396f92f4c2c3f88f
>>> # Parent  c630b0b393eec3a2e2bedf137fba0d56764672ef
>>> testbench: support float ret value
>>>
>>> diff -r c630b0b393ee -r 9505724b11bc source/test/checkasm-a.asm
>>> --- a/source/test/checkasm-a.asm        Thu Apr 24 15:59:05 2014 -0500
>>> +++ b/source/test/checkasm-a.asm        Fri Apr 25 11:01:12 2014 +0800
>>> @@ -87,6 +87,7 @@
>>>
>>> ;-----------------------------------------------------------------------------
>>>  ; intptr_t x265_checkasm_call( intptr_t (*func)(), int *ok, ... )
>>>
>>> ;-----------------------------------------------------------------------------
>>> +cglobal checkasm_call_float
>>>  INIT_XMM
>>>  cglobal checkasm_call, 2,15,16,max_args*8+8
>>>      mov  r6, r0
>>> @@ -170,6 +171,7 @@
>>>
>>> ;-----------------------------------------------------------------------------
>>>  ; intptr_t x264_checkasm_call( intptr_t (*func)(), int *ok, ... )
>>>
>>> ;-----------------------------------------------------------------------------
>>> +cglobal checkasm_call_float
>>>  cglobal checkasm_call, 1,7
>>>      mov  r3, n3
>>>      mov  r4, n4
>>> diff -r c630b0b393ee -r 9505724b11bc source/test/pixelharness.cpp
>>> --- a/source/test/pixelharness.cpp      Thu Apr 24 15:59:05 2014 -0500
>>> +++ b/source/test/pixelharness.cpp      Fri Apr 25 11:01:12 2014 +0800
>>> @@ -915,7 +915,7 @@
>>>
>>>          int width = (rand() % 4) + 1; // range[1-4]
>>>          float cres = ref(sum0, sum1, width);
>>> -        float vres = (float)checked(opt, sum0, sum1, width);
>>> +        float vres = (float)checked_float(opt, sum0, sum1, width);
>>>          if (fabs(vres - cres) > 0.00001)
>>>              return false;
>>>
>>> diff -r c630b0b393ee -r 9505724b11bc source/test/testharness.h
>>> --- a/source/test/testharness.h Thu Apr 24 15:59:05 2014 -0500
>>> +++ b/source/test/testharness.h Fri Apr 25 11:01:12 2014 +0800
>>> @@ -121,6 +121,7 @@
>>>  /* detect when callee-saved regs aren't saved
>>>   * needs an explicit asm check because it only sometimes crashes in
>>> normal use. */
>>>  intptr_t x265_checkasm_call(intptr_t (*func)(), int *ok, ...);
>>> +float x265_checkasm_call_float(intptr_t (*func)(), int *ok, ...);
>>
>>I suppose it might be a bit pedantic, but shouldn't the argument
>>funcdef return a float instead of an intptr_t?
>>
> the ABI problem, intptr_t store in rax, and float in XMM0, so I made two
> version

I was speaking of the function cast of the first argument passed to
x265_checkasm_call_float()

I changed "intptr_t (*func)()" to "float (*func)()" and it worked
fine, so I pushed it that way, more informative.

-- 
Steve Borho