Re: [PATCH 2/5] wined3d: Handle the SM4 SV_POSITION pixel shader input semantic in the GLSL shader backend.

[Stefan Dösinger]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Am 2015-01-19 um 13:44 schrieb Henri Verbeet:
> This is (mostly) equivalent to the vPos register in SM3.
> ...
> -
>        if (!strcmp(semantic_name, "SV_POSITION"))
> -
>            shader_addline(buffer, "ps_in[%u]%s = vpos%s;\n",
> -
>                    shader->u.ps.input_reg_map[i], reg_mask, reg_mask);

The vpos variable has the d3d9 0.5 pixel offset applied, which to my knowledge is not correct for d3d10. Is this intentional? (I guess yes, either for simplicity to make patch 3 work and / or because we're still adding the offset on the vertex side even in d3d10.)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
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=0cAR
-----END PGP SIGNATURE-----

[PATCH 2/5] wined3d: Handle the SM4 SV_POSITION pixel shader input semantic in the GLSL shader backend.

[Henri Verbeet]

On 19 January 2015 at 14:31, Stefan Dösinger  wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Am 2015-01-19 um 13:44 schrieb Henri Verbeet:
> > This is (mostly) equivalent to the vPos register in SM3.
> > ...
> > -
> >        if (!strcmp(semantic_name, "SV_POSITION"))
> > -
> >            shader_addline(buffer, "ps_in[%u]%s = vpos%s;\n",
> > -
> >                    shader->u.ps.input_reg_map[i], reg_mask, reg_mask);
>
> The vpos variable has the d3d9 0.5 pixel offset applied, which to my knowledge is not correct for d3d10. Is this intentional? (I guess yes, either for simplicity to make patch 3 work and / or because we're still adding the offset on the vertex side even in d3d10.)

Sort of. I'm deferring a bunch of details like pixel centers until
there's enough infrastructure in place to just port the equivalent
d3d9 tests to d3d10.