panfrost: change tie-breaking rule for 16x MSAA

When using 16x MSAA, we have two sample-positions on the negative
boundary of the unit-square covering the pixel. This causes problems
when using the default tie-breaking rule, where we miss some
sample-positions when rasterizing primitives covering the entire
viewport.

This works fine on Bifrost and later, but this setting is ignored on
those GPUs, and they assume the default (e.g MINUS_180_OUT_0_IN).
Because we'd prefer for rasterization to match between Midgard and
Bifrost when we can, we only apply this when we have 16x MSAA.

As an added bonus, this behavior matches what the DDK does.

Fixes these tests when 16x MSAA is enabled:
- dEQP-GLES31.functional.texture.multisample.samples_16.use_texture_*
- dEQP-GLES3.functional.multisample.fbo_max_samples.proportionality_alpha_to_coverage

Reviewed-by: Eric R. Smith <eric.smith@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35184>

src/panfrost/lib/pan_desc.c

@@ -874,7 +874,10 @@ GENX(pan_emit_fbd)(const struct pan_fb_info *fb, unsigned layer_idx,
       cfg.bound_max_y = fb->height - 1;
 
       cfg.effective_tile_size = fb->tile_size;
-      cfg.tie_break_rule = MALI_TIE_BREAK_RULE_MINUS_180_IN_0_OUT;
+      /* Ensure we cover the samples on the edge for 16x MSAA */
+      cfg.tie_break_rule = fb->nr_samples == 16 ?
+         MALI_TIE_BREAK_RULE_MINUS_180_OUT_0_IN :
+         MALI_TIE_BREAK_RULE_MINUS_180_IN_0_OUT;
       cfg.render_target_count = MAX2(fb->rt_count, 1);
 
       /* Default to 24 bit depth if there's no surface. */