[Mesa-dev] [PATCH 12/16] st/nine: Fix computation of const_used_size

[Ilia Mirkin]

On Sun, Apr 26, 2015 at 5:14 AM, Axel Davy <axel.davy at ens.fr> wrote:
> On 24/04/2015 23:40, Ilia Mirkin wrote :
>>
>> On Fri, Apr 24, 2015 at 4:09 PM, Axel Davy <axel.davy at ens.fr> wrote:
>>>
>>> From: Xavier Bouchoux <xavierb at gmail.com>
>>>
>>> Was sometimes too large for PS.
>>>
>>> Reviewed-by: Axel Davy <axel.davy at ens.fr>
>>> Signed-off-by: Xavier Bouchoux <xavierb at gmail.com>
>>> ---
>>>   src/gallium/state_trackers/nine/nine_shader.c | 8 +++++---
>>>   1 file changed, 5 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/src/gallium/state_trackers/nine/nine_shader.c
>>> b/src/gallium/state_trackers/nine/nine_shader.c
>>> index 1f1f7c8..7beb1ab 100644
>>> --- a/src/gallium/state_trackers/nine/nine_shader.c
>>> +++ b/src/gallium/state_trackers/nine/nine_shader.c
>>> @@ -3082,6 +3082,7 @@ nine_translate_shader(struct NineDevice9 *device,
>>> struct nine_shader_info *info)
>>>       HRESULT hr = D3D_OK;
>>>       const unsigned processor = tgsi_processor_from_type(info->type);
>>>       unsigned s, slot_max;
>>> +    unsigned max_const_f;
>>>
>>>       user_assert(processor != ~0, D3DERR_INVALIDCALL);
>>>
>>> @@ -3221,11 +3222,12 @@ nine_translate_shader(struct NineDevice9 *device,
>>> struct nine_shader_info *info)
>>>       if (tx->indirect_const_access) /* vs only */
>>>           info->const_float_slots = device->max_vs_const_f;
>>>
>>> +    max_const_f = IS_VS ? device->max_vs_const_f :
>>> device->max_ps_const_f;
>>>       slot_max = info->const_bool_slots > 0 ?
>>> -                   device->max_vs_const_f + NINE_MAX_CONST_I
>>> -                   + info->const_bool_slots :
>>> +                   max_const_f + NINE_MAX_CONST_I
>>> +                   + (info->const_bool_slots+3)/4 :
>>
>> aka DIV_ROUND_UP(info->const_bool_slots, 4) ?
>>
>> With that fixed up,
>>
>> Reviewed-by: Ilia Mirkin <imirkin at alum.mit.edu>
>>
> DIV_ROUND_UP seems to be defined in src/mesa/main/macros.h, thus we don't
> have access to it
> in gallium nine. What is the problem with (x+3)/4 ?

The problem is that you have 50 implementations of the same thing, and
you have to ask yourself, "what is this doing". We already have

src/gallium/drivers/svga/include/svga3d_surfacedefs.h:#define
DIV_ROUND_UP(x, y)  (((x) + (y) - 1) / (y))
src/mesa/main/macros.h:#define DIV_ROUND_UP( A, B )  ( (A) % (B) == 0
? (A)/(B) : (A)/(B)+1 )

I think the one in macros.h also works for negative numbers? Not
sure... seems odd. I'd recommend unifying into src/util/macros.h and
removing both of the above.

  -ilia