brw: only consider cross lane access on non scalar VGRFs

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 1bff4f93ca ("brw: Basic infrastructure to store convergent values as scalars")
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37939>

src/intel/compiler/brw/brw_schedule_instructions.cpp

@@ -1128,7 +1128,11 @@ has_cross_lane_access(const brw_inst *inst)
 
    for (unsigned s = 0; s < inst->sources; s++) {
       if (inst->src[s].file == VGRF) {
-         if (inst->src[s].stride == 0)
+         /* The instruction reads a particular lane (only relevant with non
+          * scalar values, otherwise this is just the way we read uniform
+          * values produced in reduced SIMD size).
+          */
+         if (!inst->src[s].is_scalar && inst->src[s].stride == 0)
             return true;
       }
    }