shader: Always emit shader with alpha test support in core profile

This prevents shader recompilation when the guest wants to enable/disable
alpha test. Can make a performance difference when lots of fragment
shaders suddenly have to be recompiled only because the application
called `glEnable(GL_ALPHA_TEST)`.

Signed-off-by: Italo Nicola <italonicola@collabora.com>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>

src/vrend_renderer.c

@@ -418,6 +418,7 @@ struct vrend_linked_shader_program {
    GLint fs_stipple_loc;
 
    GLint fs_alpha_ref_val_loc;
+   GLint fs_alpha_func_loc;
 
    GLuint clip_locs[8];
 
@@ -1741,10 +1742,15 @@ static struct vrend_linked_shader_program *add_shader_program(struct vrend_sub_c
       sprog->fs_stipple_loc = glGetUniformLocation(prog_id, "pstipple_sampler");
    else
       sprog->fs_stipple_loc = -1;
-   if (vrend_shader_needs_alpha_func(&fs->key))
+
-      sprog->fs_alpha_ref_val_loc = glGetUniformLocation(prog_id, "alpha_ref_val");
+   if (vrend_state.use_core_profile) {
-   else
+       sprog->fs_alpha_ref_val_loc = glGetUniformLocation(prog_id, "alpha_ref_val");
-      sprog->fs_alpha_ref_val_loc = -1;
+       sprog->fs_alpha_func_loc = glGetUniformLocation(prog_id, "alpha_func");
+   } else {
+       sprog->fs_alpha_ref_val_loc = -1;
+       sprog->fs_alpha_func_loc = -1;
+   }
+
    sprog->vs_ws_adjust_loc = glGetUniformLocation(prog_id, "winsys_adjust_y");
 
    vrend_use_program(sub_ctx, prog_id);
@@ -3534,7 +3540,6 @@ static inline void vrend_fill_shader_key(struct vrend_sub_context *sub_ctx,
 
    if (vrend_state.use_core_profile) {
       int i;
-      bool add_alpha_test = true;
 
       // Only use integer info when drawing to avoid stale info.
       if (vrend_state.use_integer && sub_ctx->drawing &&
@@ -3549,17 +3554,12 @@ static inline void vrend_fill_shader_key(struct vrend_sub_context *sub_ctx,
             if (vrend_format_is_emulated_alpha(sub_ctx->surf[i]->format))
                key->fs.cbufs_are_a8_bitmask |= (1 << i);
             if (util_format_is_pure_integer(sub_ctx->surf[i]->format)) {
-            add_alpha_test = false;
             UPDATE_INT_SIGN_MASK(sub_ctx->surf[i]->format, i,
                                  key->fs.cbufs_signed_int_bitmask,
                                  key->fs.cbufs_unsigned_int_bitmask);
             }
             key->fs.surface_component_bits[i] = util_format_get_component_bits(sub_ctx->surf[i]->format, UTIL_FORMAT_COLORSPACE_RGB, 0);
          }
-         if (add_alpha_test) {
-            key->add_alpha_test = sub_ctx->dsa_state.alpha.enabled;
-            key->alpha_test = sub_ctx->dsa_state.alpha.func;
-         }
       }
 
       key->pstipple_tex = sub_ctx->rs_state.poly_stipple_enable;
@@ -4625,7 +4625,16 @@ static void vrend_draw_bind_objects(struct vrend_sub_context *sub_ctx, bool new_
       glUniform1i(sub_ctx->prog->fs_stipple_loc, next_sampler_id);
    }
 
-   if (vrend_state.use_core_profile && sub_ctx->prog->fs_alpha_ref_val_loc != -1) {
+   if (sub_ctx->prog->fs_alpha_ref_val_loc != -1) {
+      assert(sub_ctx->prog->fs_alpha_func_loc != -1);
+
+      /* If it's an integer format surface, alpha test shouldn't do anything. */
+      if (sub_ctx->dsa_state.alpha.enabled && sub_ctx->surf[0] &&
+          !util_format_is_pure_integer(sub_ctx->surf[0]->format))
+          glUniform1i(sub_ctx->prog->fs_alpha_func_loc, sub_ctx->dsa_state.alpha.func);
+      else
+          glUniform1i(sub_ctx->prog->fs_alpha_func_loc, PIPE_FUNC_ALWAYS);
+
       glUniform1f(sub_ctx->prog->fs_alpha_ref_val_loc, sub_ctx->dsa_state.alpha.ref_value);
    }
 }

src/vrend_shader.c

@@ -259,6 +259,7 @@ struct dump_ctx {
    int num_in_clip_dist;
    int num_out_clip_dist;
    int fs_uses_clipdist_input;
+   bool fs_has_color_outputs;
    int glsl_ver_required;
    int color_in_mask;
    int color_out_mask;
@@ -1403,6 +1404,7 @@ iter_declaration(struct tgsi_iterate_context *iter,
                                             ctx->key->fs.cbufs_unsigned_int_bitmask,
                                             ctx->outputs[i].sid);
             name_prefix = ctx->key->fs.logicop_enabled ? "fsout_tmp" : "fsout";
+            ctx->fs_has_color_outputs = true;
          } else {
             if (ctx->glsl_ver_required < 140) {
                ctx->outputs[i].glsl_no_index = true;
@@ -1839,23 +1841,10 @@ static void emit_a8_swizzle(struct vrend_glsl_strbufs *glsl_strbufs)
    emit_buf(glsl_strbufs, "fsout_c0.x = fsout_c0.w;\n");
 }
 
-static const char *atests[PIPE_FUNC_ALWAYS + 1] = {
-   "false",
-   "<",
-   "==",
-   "<=",
-   ">",
-   "!=",
-   ">=",
-   "true"
-};
-
 static void emit_alpha_test(const struct dump_ctx *ctx,
                             struct vrend_glsl_strbufs *glsl_strbufs)
 {
-   char comp_buf[128];
+   if (!ctx->fs_has_color_outputs)
-
-   if (!ctx->num_outputs)
       return;
 
    if (!ctx->write_all_cbufs) {
@@ -1863,26 +1852,19 @@ static void emit_alpha_test(const struct dump_ctx *ctx,
       if (ctx->outputs[0].sid != 0)
          return;
    }
-   switch (ctx->key->alpha_test) {
-   case PIPE_FUNC_NEVER:
-   case PIPE_FUNC_ALWAYS:
-      snprintf(comp_buf, 128, "%s", atests[ctx->key->alpha_test]);
-      break;
-   case PIPE_FUNC_LESS:
-   case PIPE_FUNC_EQUAL:
-   case PIPE_FUNC_LEQUAL:
-   case PIPE_FUNC_GREATER:
-   case PIPE_FUNC_NOTEQUAL:
-   case PIPE_FUNC_GEQUAL:
-      snprintf(comp_buf, 128, "%s %s alpha_ref_val", "fsout_c0.w", atests[ctx->key->alpha_test]);
-      break;
-   default:
-      vrend_printf( "invalid alpha-test: %x\n", ctx->key->alpha_test);
-      set_buf_error(glsl_strbufs);
-      return;
-   }
 
-   emit_buff(glsl_strbufs, "if (!(%s)) {\n\tdiscard;\n}\n", comp_buf);
+   emit_buff(glsl_strbufs, "if (alpha_func != 7 /* PIPE_FUNC_ALWAYS */) {\n"
+                           "  switch (alpha_func) {\n"
+                           "  case 0: discard; break;\n"
+                           "  case 1: if (!(fsout_c0.w < alpha_ref_val)) discard; break;\n"
+                           "  case 2: if (!(fsout_c0.w == alpha_ref_val)) discard; break;\n"
+                           "  case 3: if (!(fsout_c0.w <= alpha_ref_val)) discard; break;\n"
+                           "  case 4: if (!(fsout_c0.w > alpha_ref_val)) discard; break;\n"
+                           "  case 5: if (!(fsout_c0.w != alpha_ref_val)) discard; break;\n"
+                           "  case 6: if (!(fsout_c0.w >= alpha_ref_val)) discard; break;\n"
+                           "  default: break;\n"
+                           "  }\n"
+                           "}\n");
 }
 
 static void emit_pstipple_pass(struct vrend_glsl_strbufs *glsl_strbufs)
@@ -2340,9 +2322,8 @@ static void handle_fragment_proc_exit(const struct dump_ctx *ctx,
     if (ctx->key->fs.cbufs_are_a8_bitmask)
        emit_a8_swizzle(glsl_strbufs);
 
-    if (ctx->key->add_alpha_test)
+    if (ctx->cfg->use_core_profile)
-       emit_alpha_test(ctx, glsl_strbufs);
+        emit_alpha_test(ctx, glsl_strbufs);
-
 
     if (ctx->key->fs.logicop_enabled)
        emit_fragment_logicop(ctx, glsl_strbufs);
@@ -6653,8 +6634,9 @@ static void emit_ios_fs(const struct dump_ctx *ctx,
       }
    }
 
-   if (vrend_shader_needs_alpha_func(ctx->key)) {
+   if (ctx->cfg->use_core_profile) {
-      emit_hdr(glsl_strbufs, "uniform float alpha_ref_val;\n");
+       emit_hdr(glsl_strbufs, "uniform float alpha_ref_val;\n");
+       emit_hdr(glsl_strbufs, "uniform int alpha_func;\n");
    }
 
    if (ctx->key->color_two_side) {
@@ -7733,18 +7715,3 @@ fail:
    return false;
 }
 
-bool vrend_shader_needs_alpha_func(const struct vrend_shader_key *key) {
-   if (!key->add_alpha_test)
-      return false;
-   switch (key->alpha_test) {
-   default:
-      return false;
-   case PIPE_FUNC_LESS:
-   case PIPE_FUNC_EQUAL:
-   case PIPE_FUNC_LEQUAL:
-   case PIPE_FUNC_GREATER:
-   case PIPE_FUNC_NOTEQUAL:
-   case PIPE_FUNC_GEQUAL:
-      return true;
-   }
-}

src/vrend_shader.h

@@ -178,14 +178,12 @@ struct vrend_shader_key {
    uint32_t sampler_views_lower_swizzle_mask;
    uint16_t tex_swizzle[PIPE_MAX_SHADER_SAMPLER_VIEWS];
 
-   uint8_t alpha_test;
    uint8_t clip_plane_enable;
    uint8_t num_in_cull : 4;
    uint8_t num_in_clip : 4;
    uint8_t num_out_cull : 4;
    uint8_t num_out_clip : 4;
    uint8_t pstipple_tex : 1;
-   uint8_t add_alpha_test : 1;
    uint8_t color_two_side : 1;
    uint8_t gs_present : 1;
    uint8_t tcs_present : 1;