radeonsi: sample shading state fixes

- really update sample shading state when it's changed
- reduce log state bits in the shader key to 2 because we don't support
  16x EQAA
- exit early from si_update_ps_iter_samples if ps_iter_sample has the same
  value since the last call
- set missing wqm for the PS prolog (this might fix tests)

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33024>

src/gallium/drivers/radeonsi/si_blit.c

@@ -468,7 +468,7 @@ static void si_blit_decompress_color(struct si_context *sctx, struct si_texture
       /* DCC_DECOMPRESS and ELIMINATE_FAST_CLEAR require MSAA_NUM_SAMPLES=0. */
       if (sctx->gfx_level >= GFX11) {
          sctx->gfx11_force_msaa_num_samples_zero = true;
-         si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
+         si_update_ps_iter_samples(sctx);
       }
 
       assert(vi_dcc_enabled(tex, first_level));
@@ -555,7 +555,7 @@ static void si_blit_decompress_color(struct si_context *sctx, struct si_texture
    /* Restore gfx11_force_msaa_num_samples_zero. */
    if (sctx->gfx11_force_msaa_num_samples_zero) {
       sctx->gfx11_force_msaa_num_samples_zero = false;
-      si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
+      si_update_ps_iter_samples(sctx);
    }
 
 expand_fmask:

src/gallium/drivers/radeonsi/si_pipe.h

@@ -1140,7 +1140,8 @@ struct si_context {
    struct si_resource *last_const_upload_buffer;
 
    /* MSAA config state. */
-   int ps_iter_samples;
+   uint8_t ps_iter_samples;
+   uint8_t last_ps_iter_samples;
    bool ps_uses_fbfetch;
    bool smoothing_enabled;
 

src/gallium/drivers/radeonsi/si_shader.c

@@ -3307,7 +3307,8 @@ static void si_get_ps_prolog_key(struct si_shader *shader, union si_shader_part_
        key->ps_prolog.states.force_linear_sample_interp ||
        key->ps_prolog.states.force_persp_center_interp ||
        key->ps_prolog.states.force_linear_center_interp ||
-       key->ps_prolog.states.bc_optimize_for_persp || key->ps_prolog.states.bc_optimize_for_linear);
+       key->ps_prolog.states.bc_optimize_for_persp || key->ps_prolog.states.bc_optimize_for_linear ||
+       key->ps_prolog.states.samplemask_log_ps_iter);
    key->ps_prolog.num_fragcoord_components = shader->info.num_fragcoord_components;
 
    if (shader->key.ps.part.prolog.poly_stipple)

src/gallium/drivers/radeonsi/si_shader.h

@@ -671,7 +671,7 @@ struct si_ps_prolog_bits {
    unsigned force_linear_center_interp : 1;
    unsigned bc_optimize_for_persp : 1;
    unsigned bc_optimize_for_linear : 1;
-   unsigned samplemask_log_ps_iter : 3;
+   unsigned samplemask_log_ps_iter : 2;
 };
 
 /* Common PS bits between the shader key and the epilog key. */

src/gallium/drivers/radeonsi/si_state.c

@@ -2786,6 +2786,7 @@ static void si_set_framebuffer_state(struct pipe_context *ctx,
    si_ps_key_update_framebuffer(sctx);
    si_ps_key_update_framebuffer_blend_dsa_rasterizer(sctx);
    si_ps_key_update_framebuffer_rasterizer_sample_shading(sctx);
+   si_ps_key_update_sample_shading(sctx);
    si_vs_ps_key_update_rast_prim_smooth_stipple(sctx);
    si_update_ps_inputs_read_or_disabled(sctx);
    si_update_vrs_flat_shading(sctx);
@@ -3607,6 +3608,11 @@ static void si_emit_msaa_config(struct si_context *sctx, unsigned index)
 
 void si_update_ps_iter_samples(struct si_context *sctx)
 {
+   if (sctx->ps_iter_samples == sctx->last_ps_iter_samples)
+      return;
+
+   sctx->last_ps_iter_samples = sctx->ps_iter_samples;
+   si_ps_key_update_sample_shading(sctx);
    if (sctx->framebuffer.nr_samples > 1)
       si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
    if (sctx->screen->dpbb_allowed)
@@ -3625,7 +3631,6 @@ static void si_set_min_samples(struct pipe_context *ctx, unsigned min_samples)
 
    sctx->ps_iter_samples = min_samples;
 
-   si_ps_key_update_sample_shading(sctx);
    si_ps_key_update_framebuffer_rasterizer_sample_shading(sctx);
    sctx->do_update_shaders = true;
 

src/gallium/drivers/radeonsi/si_state_shaders.cpp

@@ -2811,15 +2811,17 @@ void si_ps_key_update_dsa(struct si_context *sctx)
 void si_ps_key_update_sample_shading(struct si_context *sctx)
 {
    struct si_shader_selector *sel = sctx->shader.ps.cso;
-   union si_shader_key *key = &sctx->shader.ps.key;
-
    if (!sel)
       return;
 
-   if (sctx->ps_iter_samples > 1 && sel->info.reads_samplemask)
-      key->ps.part.prolog.samplemask_log_ps_iter = util_logbase2(sctx->ps_iter_samples);
-   else
+   union si_shader_key *key = &sctx->shader.ps.key;
+   unsigned ps_iter_samples = si_get_ps_iter_samples(sctx);
+
+   if (ps_iter_samples > 1 && sel->info.reads_samplemask) {
+      key->ps.part.prolog.samplemask_log_ps_iter = util_logbase2(ps_iter_samples);
+   } else {
       key->ps.part.prolog.samplemask_log_ps_iter = 0;
+   }
 }
 
 void si_ps_key_update_framebuffer_rasterizer_sample_shading(struct si_context *sctx)