[Mesa-dev] [PATCH 11/14] i965/blorp: Ignore color setup in fast clears and resolves

[Topi Pohjolainen]

Before:

START B0
    mov(4)      g114<1>F    g2.3<8,2,4>F  { align1 WE_all };
    sendc(16)   null<1>UW   g114<8,8,1>F
    render RT write SIMD16/RepData LastRT Surface = 0 mlen 1 rlen 0
        { align1 1H EOT };
END B0

After:

START B0
    sendc(16)   null<1>UW   g114<8,8,1>F
    render RT write SIMD16/RepData LastRT Surface = 0 mlen 1 rlen 0
        { align1 1H EOT };
END B0

Signed-off-by: Topi Pohjolainen <topi.pohjolainen at intel.com>
---
 src/mesa/drivers/dri/i965/brw_blorp_clear.cpp | 43 ++++++++++++++++++++-------
 1 file changed, 33 insertions(+), 10 deletions(-)

diff --git a/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp b/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
index 5aa420f..3848e3e 100644
--- a/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
+++ b/src/mesa/drivers/dri/i965/brw_blorp_clear.cpp
@@ -44,17 +44,35 @@
 struct brw_blorp_const_color_prog_key
 {
    bool use_simd16_replicated_data;
-   bool pad[3];
+
+   /* Fast color clears and resolves ignore the actual color values in
+    * the rep-clear message payload. Therefore the actual setup of the
+    * color values in the program can be omitted also.
+    *
+    * From the BSpec: 3D Pipeline Stages - Pixel - Pixel Backend [IVB+] -
+    *                 Render Target Resolve [IVB+] Issues Info
+    *
+    * The pixel shader kernel requires no attributes, but must deliver a
+    * render target write message covering all pixels and all render targets
+    * desired to be resolved The color data in these messages is ignored
+    * (the replicated color message is required).
+    */
+   bool ignores_color_value;
+   bool pad[2];
 };
 
 static void
 brw_blorp_params_get_clear_kernel(struct brw_context *brw,
                                   struct brw_blorp_params *params,
-                                  bool use_replicated_data)
+                                  bool use_replicated_data,
+                                  bool ignores_color_value)
 {
+   assert(!ignores_color_value || use_replicated_data);
+
    struct brw_blorp_const_color_prog_key blorp_key;
    memset(&blorp_key, 0, sizeof(blorp_key));
    blorp_key.use_simd16_replicated_data = use_replicated_data;
+   blorp_key.ignores_color_value = ignores_color_value;
 
    if (brw_search_cache(&brw->cache, BRW_CACHE_BLORP_PROG,
                         &blorp_key, sizeof(blorp_key),
@@ -67,16 +85,18 @@ brw_blorp_params_get_clear_kernel(struct brw_context *brw,
    nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
    b.shader->info.name = ralloc_strdup(b.shader, "BLORP-clear");
 
-   nir_variable *u_color = nir_variable_create(b.shader, nir_var_uniform,
-                                               glsl_vec4_type(), "u_color");
-   u_color->data.location = 0;
-
    nir_variable *frag_color = nir_variable_create(b.shader, nir_var_shader_out,
                                                   glsl_vec4_type(),
                                                   "gl_FragColor");
    frag_color->data.location = FRAG_RESULT_COLOR;
 
-   nir_copy_var(&b, frag_color, u_color);
+   if (!ignores_color_value) {
+      nir_variable *u_color = nir_variable_create(b.shader, nir_var_uniform,
+                                                  glsl_vec4_type(), "u_color");
+      u_color->data.location = 0;
+
+      nir_copy_var(&b, frag_color, u_color);
+   }
 
    struct brw_wm_prog_key wm_key;
    brw_blorp_init_wm_prog_key(&wm_key);
@@ -84,7 +104,8 @@ brw_blorp_params_get_clear_kernel(struct brw_context *brw,
    struct brw_blorp_prog_data prog_data;
    unsigned program_size;
    const int flags =
-      use_replicated_data ? BRW_FS_COMPILER_OPTION_REP_SEND : 0;
+      (use_replicated_data ? BRW_FS_COMPILER_OPTION_REP_SEND : 0) |
+      (ignores_color_value ? BRW_FS_COMPILER_OPTION_IGNORE_COLOR : 0);
    const unsigned *program =
       brw_blorp_compile_nir_shader(brw, b.shader, &wm_key, flags,
                                    &prog_data, &program_size);
@@ -187,7 +208,9 @@ do_single_blorp_clear(struct brw_context *brw, struct gl_framebuffer *fb,
                                &params.x1, &params.y1);
    }
 
-   brw_blorp_params_get_clear_kernel(brw, &params, use_simd16_replicated_data);
+   brw_blorp_params_get_clear_kernel(brw, &params,
+                                     use_simd16_replicated_data,
+                                     params.fast_clear_op);
 
    const bool is_fast_clear =
       params.fast_clear_op == GEN7_PS_RENDER_TARGET_FAST_CLEAR_ENABLE;
@@ -325,7 +348,7 @@ brw_blorp_resolve_color(struct brw_context *brw, struct intel_mipmap_tree *mt)
     * color" message.
     */
 
-   brw_blorp_params_get_clear_kernel(brw, &params, true);
+   brw_blorp_params_get_clear_kernel(brw, &params, true, true);
 
    brw_blorp_exec(brw, &params);
    mt->fast_clear_state = INTEL_FAST_CLEAR_STATE_RESOLVED;
-- 
2.5.5