[Mesa-dev] [PATCH] i915g: fix incorrect gl_FragCoord value

[Roland Scheidegger]

Am 22.09.2016 um 17:40 schrieb Emil Velikov:
> Hi Nicholas,
> 
> On 26 August 2016 at 00:31, Nicholas Bishop <nbishop at neverware.com> wrote:
>> From: Nicholas Bishop <nicholasbishop at gmail.com>
>>
>> On Intel Pineview M hardware, the i915 gallium driver doesn't output
>> the correct gl_FragCoord. It seems to always have an X coord of 0.0
>> and a Y coord of the window's height in pixels, e.g. 600.0f or such.
>>
>> I believe this is a regression caused in part by this commit:
>> afa035031ff9e0c07a2297d864e46c76f7bfff58
>>
>> The old behavior used the output at index zero, while the new behavior
>> uses actual zeroes. In the case of gl_FragCoord the output at index
>> zero happened to be the correct one, so the behavior appeared correct
>> although the code already had a bug.
>>
>> Fixed by checking for I915_SEMANTIC_POS when setting up texCoords. If
>> the generic_mapping is I915_SEMANTIC_POS, look for the
>> TGSI_SEMANTIC_POSITION instead of a TGSI_SEMANTIC_GENERIC output.
>>
>> https://urldefense.proofpoint.com/v2/url?u=https-3A__bugs.freedesktop.org_show-5Fbug.cgi-3Fid-3D97477&d=CwIFaQ&c=Sqcl0Ez6M0X8aeM67LKIiDJAXVeAw-YihVMNtXt-uEs&r=Vjtt0vs_iqoI31UfJxBl7yv9I2FeiaeAYgMTLKRBc_I&m=Mt7V3FvGsInekgNmMd2jOoCYRJDl4If-uvNtbOhGkRQ&s=JlfHHRDgBQvDfacPpT-qFUmY4drEprImNSdf1L6ch3c&e= 
>> ---
>>  src/gallium/drivers/i915/i915_state_derived.c | 7 ++++++-
>>  1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/src/gallium/drivers/i915/i915_state_derived.c b/src/gallium/drivers/i915/i915_state_derived.c
>> index 177b854..dbfbc84 100644
>> --- a/src/gallium/drivers/i915/i915_state_derived.c
>> +++ b/src/gallium/drivers/i915/i915_state_derived.c
>> @@ -145,7 +145,12 @@ static void calculate_vertex_layout(struct i915_context *i915)
>>        uint hwtc;
>>        if (texCoords[i]) {
>>           hwtc = TEXCOORDFMT_4D;
>> -         src = draw_find_shader_output(i915->draw, TGSI_SEMANTIC_GENERIC, fs->generic_mapping[i]);
>> +         if (fs->generic_mapping[i] == I915_SEMANTIC_POS) {
>> +            src = draw_find_shader_output(i915->draw, TGSI_SEMANTIC_POSITION, 0);
>> +         }
>> +         else {
>> +            src = draw_find_shader_output(i915->draw, TGSI_SEMANTIC_GENERIC, fs->generic_mapping[i]);
>> +         }
> 
> Personally I'm unfamiliar with the i915g driver, so I've Cc'd Roland
> (the author of the commit 'breaking' this) and Stéphane (the author of
> this driver).
> 
> Gents, any input on the above ?

Oh, I wasn't aware I broke this...
This looks to me like before my patch there i915g was indeed just
relying on draw code redirecting undefined inputs to 0, which near
certainly was unspecified behavior.
I don't really know i915g, but this looks like it should be correct.

Reviewed-by: Roland Scheidegger <sroland at vmware.com>

> 
> That said, please drop the unneeded brackets and add the following
> line in the commit message for v2.
> Cc: mesa-stable at lists.freedesktop.org
> 
> Thanks,
> Emil
>