[PATCH vkd3d 2/5] tests: Test the shape of the sum of two numeric values.

[Zebediah Figura (she/her)]

On 9/1/21 2:04 AM, Giovanni Mascellani wrote:
> Hi,
> 
> thanks for the review.
> 
> Il 31/08/21 19:32, Zebediah Figura (she/her) ha scritto:
>> On 8/31/21 9:40 AM, Giovanni Mascellani wrote:
>>> diff --git a/tests/hlsl-shape.shader_test b/tests/hlsl-shape.shader_test
>>> new file mode 100644
>>> index 00000000..0d488da4
>>> --- /dev/null
>>> +++ b/tests/hlsl-shape.shader_test
>>
>> I'm not a huge fan of this name; it's not immediately clear to me what
>> it means :-(
>>
>> Maybe "hlsl-expression-dimensions"?
> 
> I though "shape" was a rather standard term. For sure it is standard in
> scientific computing (a trend probably set by MatLab[1]), but maybe only
> there. No problem in switching to your suggestion.
> 
>    [1] https://www.mathworks.com/help/matlab/matrices-and-arrays.html
> 

Alright, I wasn't familiar with it. If it's a standard term then no 
objections from me.

>> This doesn't compile with native for ps_2_0 either (see my comment on
>> 5/5). Part of this is because it uses SV_Position, which is kind of my
>> fault too as I've been using that. We should probably find a way to
>> change that to texcoords instead.
> 
> Ok, I'll try to set it up.
> 
>> With that hacked out of the way, it breaks because of the following error:
>>
>> Z:\home\hazel\test.hlsl(83,38): error X5608: Compiled shader code uses
>> too many arithmetic instruction slots (83). Max. allowed by the target
>> (ps_2_0) is 64. Consider increasing optimization level to reduce
>> instruction count.
>>
>> (Increasing optimization level does nothing, of course.)
> 
> I guess the only way there is to split the test in many smaller shaders.
> Not a big problem.
> 
> Giovanni.
>