[Piglit] [PATCH] arb_copy_image: add new format swizzle test

[Ilia Mirkin]

On Tue, Sep 1, 2015 at 1:14 PM, Fredrik Höglund <fredrik at kde.org> wrote:
> On Tuesday 01 September 2015, Ilia Mirkin wrote:
>> On Tue, Sep 1, 2015 at 12:22 PM, Fredrik Höglund <fredrik at kde.org> wrote:
>> > On Tuesday 01 September 2015, Brian Paul wrote:
>> >> On 08/31/2015 05:38 PM, Ilia Mirkin wrote:
>> >> > On Mon, Aug 31, 2015 at 7:29 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>> >> >> Reviewed-by: Ilia Mirkin <imirkin at alum.mit.edu>
>> >> >>
>> >> >> Thanks a lot for doing this! I assume that it fails with your gallium
>> >> >> ARB_copy_image implementation as I predicted?
>> >>
>> >> Yes.  In Mesa, I added some code that checks for swizzles and uses
>> >> pipe_context::blit() instead.  Seems to fix things, but I need to spend
>> >> more time on it.
>> >
>> > This test is not sufficient though, because it's when you have a copy
>> > from B8G8R8A8 to R16G16, or a similar situation, that things get really
>> > interesting.
>>
>> Yeah, so the thing is that you can't do this with unsized internal
>> formats, since they're not part of the view classes, but you could
>> (carefully) arrange for a *sized* internal format, like GL_RGBA8 to
>> come out as a BGRA pipe format, and then try to copy it to RG16. And
>> enjoy the fireworks :)
>
> st/mesa uses the same rules when it selects a pipe format for
> GL_RGBA8 as it does for GL_RGBA.  See find_exact_format()
> in st_format.c

I know, but this test only uses unsized internal formats :)

>
>> But the test is a great start, still has my r-b.
>
> Sure, my point is just that we are going to need additional tests for
> compatible formats with different component orders.
>
> Fredrik
>