pan/mid/fuse_io_cvt: Disable fusion on highp

As discussed on gitlab, we should not convert before interpolation
at highp.

Signed-off-by: Lorenzo Rossi <lorenzo.rossi@collabora.com>
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
This commit is contained in:
Lorenzo Rossi 2026-05-05 16:05:22 +02:00
parent ee721bb668
commit c92853198e

View file

@ -41,6 +41,10 @@ nir_fuse_io_16(nir_shader *shader)
if (intr->def.bit_size != 32)
continue;
/* Do not change interpolation precision in highp */
if (!nir_intrinsic_io_semantics(intr).medium_precision)
continue;
/* We swizzle at a 32-bit level so need a multiple of 2. We could
* do a bit better and handle even components though */
if (nir_intrinsic_component(intr))