Mesa (master): i965: Pack the tracked state atoms into separate arrays for prepare/emit.

[Eric Anholt]

Module: Mesa
Branch: master
Commit: ea004a3aed89ec02bb5f2356479cd038d41f73c1
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=ea004a3aed89ec02bb5f2356479cd038d41f73c1

Author: Chris Wilson <chris at chris-wilson.co.uk>
Date:   Mon Mar  7 16:43:45 2011 -0800

i965: Pack the tracked state atoms into separate arrays for prepare/emit.

Improves performance of a hacked-up scissor-many (to reuse a small set
of scissors instead of blowing out the cache, and then to run 100x
more iterations so it actually took some time) by 3.6% +/- 1.2% (n=10)

---

 src/mesa/drivers/dri/i965/brw_context.h      |    3 +
 src/mesa/drivers/dri/i965/brw_state_upload.c |   82 +++++++++++++-------------
 2 files changed, 43 insertions(+), 42 deletions(-)

diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index 26a6388..7b0551a 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -701,6 +701,9 @@ struct brw_context
    /* Used to give every program string a unique id
     */
    GLuint program_id;
+
+   int num_prepare_atoms, num_emit_atoms;
+   struct brw_tracked_state prepare_atoms[64], emit_atoms[64];
 };
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c
index 2d6fb37..6f521be 100644
--- a/src/mesa/drivers/dri/i965/brw_state_upload.c
+++ b/src/mesa/drivers/dri/i965/brw_state_upload.c
@@ -104,7 +104,7 @@ static const struct brw_tracked_state *gen4_atoms[] =
    &brw_constant_buffer
 };
 
-const struct brw_tracked_state *gen6_atoms[] =
+static const struct brw_tracked_state *gen6_atoms[] =
 {
    &brw_check_fallback,
 
@@ -169,7 +169,32 @@ const struct brw_tracked_state *gen6_atoms[] =
 
 void brw_init_state( struct brw_context *brw )
 {
+   const struct brw_tracked_state **atoms;
+   int num_atoms;
+
    brw_init_caches(brw);
+
+   if (brw->intel.gen >= 6) {
+      atoms = gen6_atoms;
+      num_atoms = ARRAY_SIZE(gen6_atoms);
+   } else {
+      atoms = gen4_atoms;
+      num_atoms = ARRAY_SIZE(gen4_atoms);
+   }
+
+   while (num_atoms--) {
+      assert((*atoms)->dirty.mesa |
+	     (*atoms)->dirty.brw |
+	     (*atoms)->dirty.cache);
+
+      if ((*atoms)->prepare)
+	 brw->prepare_atoms[brw->num_prepare_atoms++] = **atoms;
+      if ((*atoms)->emit)
+	 brw->emit_atoms[brw->num_emit_atoms++] = **atoms;
+      atoms++;
+   }
+   assert(brw->num_emit_atoms <= ARRAY_SIZE(brw->emit_atoms));
+   assert(brw->num_prepare_atoms <= ARRAY_SIZE(brw->prepare_atoms));
 }
 
 
@@ -186,7 +211,7 @@ static GLuint check_state( const struct brw_state_flags *a,
 {
    return ((a->mesa & b->mesa) |
 	   (a->brw & b->brw) |
-	   (a->cache & b->cache));
+	   (a->cache & b->cache)) != 0;
 }
 
 static void accumulate_state( struct brw_state_flags *a,
@@ -342,9 +367,9 @@ void brw_validate_state( struct brw_context *brw )
    struct gl_context *ctx = &brw->intel.ctx;
    struct intel_context *intel = &brw->intel;
    struct brw_state_flags *state = &brw->state.dirty;
+   const struct brw_tracked_state *atoms = brw->prepare_atoms;
+   int num_atoms = brw->num_prepare_atoms;
    GLuint i;
-   const struct brw_tracked_state **atoms;
-   int num_atoms;
 
    brw_clear_validated_bos(brw);
 
@@ -353,14 +378,6 @@ void brw_validate_state( struct brw_context *brw )
 
    brw_add_validated_bo(brw, intel->batch.bo);
 
-   if (intel->gen >= 6) {
-      atoms = gen6_atoms;
-      num_atoms = ARRAY_SIZE(gen6_atoms);
-   } else {
-      atoms = gen4_atoms;
-      num_atoms = ARRAY_SIZE(gen4_atoms);
-   }
-
    if (brw->emit_state_always) {
       state->mesa |= ~0;
       state->brw |= ~0;
@@ -384,15 +401,13 @@ void brw_validate_state( struct brw_context *brw )
 
    /* do prepare stage for all atoms */
    for (i = 0; i < num_atoms; i++) {
-      const struct brw_tracked_state *atom = atoms[i];
-
-      if (brw->intel.Fallback)
-         break;
+      const struct brw_tracked_state *atom = &atoms[i];
 
       if (check_state(state, &atom->dirty)) {
-         if (atom->prepare) {
-            atom->prepare(brw);
-        }
+	 atom->prepare(brw);
+
+	 if (brw->intel.Fallback)
+	    break;
       }
    }
 
@@ -415,20 +430,11 @@ void brw_validate_state( struct brw_context *brw )
 
 void brw_upload_state(struct brw_context *brw)
 {
-   struct intel_context *intel = &brw->intel;
    struct brw_state_flags *state = &brw->state.dirty;
+   const struct brw_tracked_state *atoms = brw->emit_atoms;
+   int num_atoms = brw->num_emit_atoms;
    int i;
    static int dirty_count = 0;
-   const struct brw_tracked_state **atoms;
-   int num_atoms;
-
-   if (intel->gen >= 6) {
-      atoms = gen6_atoms;
-      num_atoms = ARRAY_SIZE(gen6_atoms);
-   } else {
-      atoms = gen4_atoms;
-      num_atoms = ARRAY_SIZE(gen4_atoms);
-   }
 
    brw_clear_validated_bos(brw);
 
@@ -442,20 +448,14 @@ void brw_upload_state(struct brw_context *brw)
       prev = *state;
 
       for (i = 0; i < num_atoms; i++) {
-	 const struct brw_tracked_state *atom = atoms[i];
+	 const struct brw_tracked_state *atom = &atoms[i];
 	 struct brw_state_flags generated;
 
-	 assert(atom->dirty.mesa ||
-		atom->dirty.brw ||
-		atom->dirty.cache);
-
 	 if (brw->intel.Fallback)
 	    break;
 
 	 if (check_state(state, &atom->dirty)) {
-	    if (atom->emit) {
-	       atom->emit( brw );
-	    }
+	    atom->emit(brw);
 	 }
 
 	 accumulate_state(&examined, &atom->dirty);
@@ -471,15 +471,13 @@ void brw_upload_state(struct brw_context *brw)
    }
    else {
       for (i = 0; i < num_atoms; i++) {
-	 const struct brw_tracked_state *atom = atoms[i];
+	 const struct brw_tracked_state *atom = &atoms[i];
 
 	 if (brw->intel.Fallback)
 	    break;
 
 	 if (check_state(state, &atom->dirty)) {
-	    if (atom->emit) {
-	       atom->emit( brw );
-	    }
+	    atom->emit(brw);
 	 }
       }
    }