Comme=
nt # 17
on bug 10460=
2
from Timothy Arceri
After talking this over with Marek here is a summary of the pr=
oblem.
LLVM's VGPR indexing code on gfx9+ is broken for immediate arrays. Usually =
this
is not a problem as GLSL IR in mesa will lower these to Uniforms via
lower_const_arrays_to_uniforms(). However this does not work for the shader=
s in
Civ6 because these arrays are not actually defined as constant arrays for
example the original shader looks like this:
vec4 x0[3];
vec4 x1[6];
vec4 x2[6];
vec4 x3[6];
x0[0].xy =3D vec2(0.031250, 0.500000);
x0[1].xy =3D vec2(0.968750, 0.031250);
x0[2].xy =3D vec2(0.968750, 0.968750);
x1[0].xy =3D vec2(1.000000, 1.000000);
x1[1].xy =3D vec2(0.000000, 1.000000);
x1[2].xy =3D vec2(-1.000000, 0.000000);
x1[3].xy =3D vec2(-1.000000, -1.000000);
x1[4].xy =3D vec2(0.000000, -1.000000);
x1[5].xy =3D vec2(1.000000, 0.000000);
x2[0].xy =3D vec2(1.000000, -1.000000);
x2[1].xy =3D vec2(2.000000, 1.000000);
x2[2].xy =3D vec2(1.000000, 2.000000);
x2[3].xy =3D vec2(-1.000000, 1.000000);
x2[4].xy =3D vec2(-2.000000, -1.000000);
x2[5].xy =3D vec2(-1.000000, -2.000000);
Without SSA I don't see any way for GLSL IR to easily recognise this as a
constant array. Unfortunately by the time LLVM is done it is recognised and=
is
exposed to the buggy indexing support.