The funny thing is that I just encountered a somewhat similar bug in the i965 vec4 backend. :) The ability to mix sizes like that in the IR was a mistake. We should have added swizzles to code like this so that the types would match. Alas.
This patch is Reviewed-by: Ian Romanick <ian.d.roman...@intel.com> On 06/27/2018 04:23 PM, Timothy Arceri wrote: > The spec allows adding scalars a vector or matrix. In this case > the opt was losing any swizzle or size information. > > This fixes a bug with Doom (2016) shaders. > --- > src/compiler/glsl/opt_algebraic.cpp | 6 ++++++ > 1 file changed, 6 insertions(+) > > diff --git a/src/compiler/glsl/opt_algebraic.cpp > b/src/compiler/glsl/opt_algebraic.cpp > index 1a8ee361652..ff4be269578 100644 > --- a/src/compiler/glsl/opt_algebraic.cpp > +++ b/src/compiler/glsl/opt_algebraic.cpp > @@ -709,6 +709,12 @@ ir_algebraic_visitor::handle_expression(ir_expression > *ir) > if (!is_vec_zero(zero)) > continue; > > + /* We are allowed to add scalars with a vector or matrix. In that > + * case lets just exit early. > + */ > + if (add->operands[0]->type != add->operands[1]->type) > + continue; > + > /* Depending of the zero position we want to optimize > * (0 cmp x+y) into (-x cmp y) or (x+y cmp 0) into (x cmp -y) > */ > _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
