Mesa/i965_aperture_call_for_Q3.patch

diff -urNap mesa_orig//src/mesa/drivers/dri/i965/brw_context.c mesa_haihao//src/mesa/drivers/dri/i965/brw_context.c
--- mesa_orig//src/mesa/drivers/dri/i965/brw_context.c	2009-01-12 13:29:40.000000000 +0800
+++ mesa_haihao//src/mesa/drivers/dri/i965/brw_context.c	2009-01-13 16:43:37.000000000 +0800
@@ -33,6 +33,7 @@
 #include "brw_context.h"
 #include "brw_defines.h"
 #include "brw_draw.h"
+#include "brw_state.h"
 #include "brw_vs.h"
 #include "imports.h"
 #include "intel_tex.h"
diff -urNap mesa_orig//src/mesa/drivers/dri/i965/brw_context.h mesa_haihao//src/mesa/drivers/dri/i965/brw_context.h
--- mesa_orig//src/mesa/drivers/dri/i965/brw_context.h	2009-01-12 13:29:36.000000000 +0800
+++ mesa_haihao//src/mesa/drivers/dri/i965/brw_context.h	2009-01-13 16:43:37.000000000 +0800
@@ -431,6 +431,19 @@ struct brw_context
       GLuint nr_draw_regions;
       struct intel_region *draw_regions[MAX_DRAW_BUFFERS];
       struct intel_region *depth_region;
+
+      /**
+       * List of buffers accumulated in brw_validate_state to receive
+       * dri_bo_check_aperture treatment before exec, so we can know if we
+       * should flush the batch and try again before emitting primitives.
+       *
+       * This can be a fixed number as we only have a limited number of
+       * objects referenced from the batchbuffer in a primitive emit,
+       * consisting of the vertex buffers, pipelined state pointers,
+       * the CURBE, the depth buffer, and a query BO.
+       */
+      dri_bo *validated_bos[VERT_ATTRIB_MAX + 16];
+      int validated_bo_count;
    } state;
 
    struct brw_state_pointers attribs;
@@ -659,14 +672,6 @@ GLboolean brwCreateContext( const __GLco
 
 
 /*======================================================================
- * brw_state.c
- */
-void brw_validate_state( struct brw_context *brw );
-void brw_init_state( struct brw_context *brw );
-void brw_destroy_state( struct brw_context *brw );
-
-
-/*======================================================================
  * brw_state_dump.c
  */
 void brw_debug_batch(struct intel_context *intel);
diff -urNap mesa_orig//src/mesa/drivers/dri/i965/brw_curbe.c mesa_haihao//src/mesa/drivers/dri/i965/brw_curbe.c
--- mesa_orig//src/mesa/drivers/dri/i965/brw_curbe.c	2009-01-12 13:29:40.000000000 +0800
+++ mesa_haihao//src/mesa/drivers/dri/i965/brw_curbe.c	2009-01-13 16:47:46.000000000 +0800
@@ -307,6 +307,7 @@ static void prepare_constant_buffer(stru
       dri_bo_subdata(brw->curbe.curbe_bo, brw->curbe.curbe_offset, bufsz, buf);
    }
 
+   brw_add_validated_bo(brw, brw->curbe.curbe_bo);
 
    /* Because this provokes an action (ie copy the constants into the
     * URB), it shouldn't be shortcircuited if identical to the
@@ -328,13 +329,6 @@ static void emit_constant_buffer(struct
 {
    struct intel_context *intel = &brw->intel;
    GLuint sz = brw->curbe.total_size;
-   dri_bo *aper_array[] = {
-      brw->intel.batch->buf,
-      brw->curbe.curbe_bo,
-   };
-
-   if (dri_bufmgr_check_aperture_space(aper_array, ARRAY_SIZE(aper_array)))
-      intel_batchbuffer_flush(intel->batch);
 
    BEGIN_BATCH(2, IGNORE_CLIPRECTS);
    if (sz == 0) {
diff -urNap mesa_orig//src/mesa/drivers/dri/i965/brw_draw.c mesa_haihao//src/mesa/drivers/dri/i965/brw_draw.c
--- mesa_orig//src/mesa/drivers/dri/i965/brw_draw.c	2009-01-12 13:29:37.000000000 +0800
+++ mesa_haihao//src/mesa/drivers/dri/i965/brw_draw.c	2009-01-13 16:43:37.000000000 +0800
@@ -105,6 +105,7 @@ static GLuint brw_set_prim(struct brw_co
       }
 
       brw_validate_state(brw);
+      brw_upload_state(brw);
    }
 
    return hw_prim[prim];
@@ -256,6 +257,7 @@ static GLboolean brw_try_draw_prims( GLc
    struct intel_context *intel = intel_context(ctx);
    struct brw_context *brw = brw_context(ctx);
    GLboolean retval = GL_FALSE;
+   GLboolean warn = GL_FALSE;
    GLuint i;
 
    if (ctx->NewState)
@@ -304,8 +306,6 @@ static GLboolean brw_try_draw_prims( GLc
        */
       brw_set_prim(brw, prim[0].mode);
 
-      /* XXX:  Need to separate validate and upload of state.  
-       */
       brw_validate_state( brw );
 
       /* Various fallback checks:
@@ -316,6 +316,31 @@ static GLboolean brw_try_draw_prims( GLc
       if (check_fallbacks( brw, prim, nr_prims ))
 	 goto out;
 
+      /* Check that we can fit our state in with our existing batchbuffer, or
+       * flush otherwise.
+       */
+      if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
+					  brw->state.validated_bo_count)) {
+	 static GLboolean warned;
+	 intel_batchbuffer_flush(intel->batch);
+
+	 /* Validate the state after we flushed the batch (which would have
+	  * changed the set of dirty state).  If we still fail to
+	  * check_aperture, warn of what's happening, but attempt to continue
+	  * on since it may succeed anyway, and the user would probably rather
+	  * see a failure and a warning than a fallback.
+	  */
+	 brw_validate_state(brw);
+	 if (!warned &&
+	     dri_bufmgr_check_aperture_space(brw->state.validated_bos,
+					     brw->state.validated_bo_count)) {
+	    warn = GL_TRUE;
+	    warned = GL_TRUE;
+	 }
+      }
+
+      brw_upload_state(brw);
+
       for (i = 0; i < nr_prims; i++) {
 	 brw_emit_prim(brw, &prim[i]);
       }
@@ -326,6 +351,10 @@ static GLboolean brw_try_draw_prims( GLc
  out:
    UNLOCK_HARDWARE(intel);
 
+   if (warn)
+      fprintf(stderr, "i965: Single primitive emit potentially exceeded "
+	      "available aperture space\n");
+
    if (!retval)
       DBG("%s failed\n", __FUNCTION__);
 
diff -urNap mesa_orig//src/mesa/drivers/dri/i965/brw_draw_upload.c mesa_haihao//src/mesa/drivers/dri/i965/brw_draw_upload.c
--- mesa_orig//src/mesa/drivers/dri/i965/brw_draw_upload.c	2009-01-12 13:29:36.000000000 +0800
+++ mesa_haihao//src/mesa/drivers/dri/i965/brw_draw_upload.c	2009-01-13 16:43:37.000000000 +0800
@@ -250,10 +250,10 @@ static void get_space( struct brw_contex
       wrap_buffers(brw, size);
    }
 
+   assert(*bo_return == NULL);
    dri_bo_reference(brw->vb.upload.bo);
    *bo_return = brw->vb.upload.bo;
    *offset_return = brw->vb.upload.offset;
-
    brw->vb.upload.offset += size;
 }
 
@@ -353,12 +353,21 @@ static void brw_prepare_vertices(struct
 	    intel_buffer_object(input->glarray->BufferObj);
 
 	 /* Named buffer object: Just reference its contents directly. */
+         dri_bo_unreference(input->bo);
 	 input->bo = intel_bufferobj_buffer(intel, intel_buffer,
 					    INTEL_READ);
 	 dri_bo_reference(input->bo);
 	 input->offset = (unsigned long)input->glarray->Ptr;
 	 input->stride = input->glarray->StrideB;
       } else {
+	 if (input->bo != NULL) {
+	    /* Already-uploaded vertex data is present from a previous
+	     * prepare_vertices, but we had to re-validate state due to
+	     * check_aperture failing and a new batch being produced.
+	     */
+	    continue;
+	 }
+
 	 /* Queue the buffer object up to be uploaded in the next pass,
 	  * when we've decided if we're doing interleaved or not.
 	  */
@@ -415,6 +424,11 @@ static void brw_prepare_vertices(struct
           copy_array_to_vbo_array(brw, upload[i], upload[i]->element_size);
       }
    }
+
+   for (i = 0; i < nr_enabled; i++) {
+       struct brw_vertex_element *input = enabled[i];
+       brw_add_validated_bo(brw, input->bo);
+   }
 }
 
 static void brw_emit_vertices(struct brw_context *brw)
@@ -509,7 +523,7 @@ static void brw_prepare_indices(struct b
    struct intel_context *intel = &brw->intel;
    const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
    GLuint ib_size;
-   dri_bo *bo;
+   dri_bo *bo = NULL;
    struct gl_buffer_object *bufferobj;
    GLuint offset;
 
@@ -558,6 +572,8 @@ static void brw_prepare_indices(struct b
    dri_bo_unreference(brw->ib.bo);
    brw->ib.bo = bo;
    brw->ib.offset = offset;
+
+   brw_add_validated_bo(brw, brw->ib.bo);
 }
 
 static void brw_emit_indices(struct brw_context *brw)
diff -urNap mesa_orig//src/mesa/drivers/dri/i965/brw_misc_state.c mesa_haihao//src/mesa/drivers/dri/i965/brw_misc_state.c
--- mesa_orig//src/mesa/drivers/dri/i965/brw_misc_state.c	2009-01-12 13:29:39.000000000 +0800
+++ mesa_haihao//src/mesa/drivers/dri/i965/brw_misc_state.c	2009-01-13 16:50:17.000000000 +0800
@@ -71,6 +71,11 @@ const struct brw_tracked_state brw_blend
    .emit = upload_blend_constant_color
 };
 
+static void prepare_binding_table_pointers(struct brw_context *brw)
+{
+   brw_add_validated_bo(brw, brw->wm.bind_bo);
+}
+
 /**
  * Upload the binding table pointers, which point each stage's array of surface
  * state pointers.
@@ -81,13 +86,6 @@ const struct brw_tracked_state brw_blend
 static void upload_binding_table_pointers(struct brw_context *brw)
 {
    struct intel_context *intel = &brw->intel;
-   dri_bo *aper_array[] = {
-      intel->batch->buf,
-      brw->wm.bind_bo,
-   };
-
-   if (dri_bufmgr_check_aperture_space(aper_array, ARRAY_SIZE(aper_array)))
-      intel_batchbuffer_flush(intel->batch);
 
    BEGIN_BATCH(6, IGNORE_CLIPRECTS);
    OUT_BATCH(CMD_BINDING_TABLE_PTRS << 16 | (6 - 2));
@@ -107,6 +105,7 @@ const struct brw_tracked_state brw_bindi
       .brw = BRW_NEW_BATCH,
       .cache = CACHE_NEW_SURF_BIND,
    },
+   .prepare = prepare_binding_table_pointers,
    .emit = upload_binding_table_pointers,
 };
 
@@ -140,21 +139,18 @@ static void upload_pipelined_state_point
    brw->state.dirty.brw |= BRW_NEW_PSP;
 }
 
+
+static void prepare_psp_urb_cbs(struct brw_context *brw)
+{
+   brw_add_validated_bo(brw, brw->vs.state_bo);
+   brw_add_validated_bo(brw, brw->gs.state_bo);
+   brw_add_validated_bo(brw, brw->clip.state_bo);
+   brw_add_validated_bo(brw, brw->wm.state_bo);
+   brw_add_validated_bo(brw, brw->cc.state_bo);
+}
+  
 static void upload_psp_urb_cbs(struct brw_context *brw )
 {
-   struct intel_context *intel = &brw->intel;
-   dri_bo *aper_array[] = {
-      intel->batch->buf,
-      brw->vs.state_bo,
-      brw->gs.state_bo,
-      brw->clip.state_bo,
-      brw->wm.state_bo,
-      brw->cc.state_bo,
-   };
-
-   if (dri_bufmgr_check_aperture_space(aper_array, ARRAY_SIZE(aper_array)))
-      intel_batchbuffer_flush(intel->batch);
-
    upload_pipelined_state_pointers(brw);
    brw_upload_urb_fence(brw);
    brw_upload_constant_buffer_state(brw);
@@ -172,9 +168,18 @@ const struct brw_tracked_state brw_psp_u
 		CACHE_NEW_WM_UNIT | 
 		CACHE_NEW_CC_UNIT)
    },
+   .prepare = prepare_psp_urb_cbs,
    .emit = upload_psp_urb_cbs,
 };
 
+static void prepare_depthbuffer(struct brw_context *brw)
+{
+   struct intel_region *region = brw->state.depth_region;
+
+   if (region != NULL)
+      brw_add_validated_bo(brw, region->buffer);
+}
+
 static void emit_depthbuffer(struct brw_context *brw)
 {
    struct intel_context *intel = &brw->intel;
@@ -196,10 +201,6 @@ static void emit_depthbuffer(struct brw_
       ADVANCE_BATCH();
    } else {
       unsigned int format;
-      dri_bo *aper_array[] = {
-	 intel->batch->buf,
-	 region->buffer
-      };
 
       switch (region->cpp) {
       case 2:
@@ -216,9 +217,6 @@ static void emit_depthbuffer(struct brw_
 	 return;
       }
 
-      if (dri_bufmgr_check_aperture_space(aper_array, ARRAY_SIZE(aper_array)))
-	 intel_batchbuffer_flush(intel->batch);
-
       BEGIN_BATCH(len, IGNORE_CLIPRECTS);
       OUT_BATCH(CMD_DEPTH_BUFFER << 16 | (len - 2));
       OUT_BATCH(((region->pitch * region->cpp) - 1) |
@@ -247,6 +245,7 @@ const struct brw_tracked_state brw_depth
       .brw = BRW_NEW_DEPTH_BUFFER | BRW_NEW_BATCH,
       .cache = 0,
    },
+   .prepare = prepare_depthbuffer,
    .emit = emit_depthbuffer,
 };
 
diff -urNap mesa_orig//src/mesa/drivers/dri/i965/brw_state.h mesa_haihao//src/mesa/drivers/dri/i965/brw_state.h
--- mesa_orig//src/mesa/drivers/dri/i965/brw_state.h	2009-01-12 13:29:39.000000000 +0800
+++ mesa_haihao//src/mesa/drivers/dri/i965/brw_state.h	2009-01-13 16:43:37.000000000 +0800
@@ -35,6 +35,16 @@
 
 #include "brw_context.h"
 
+static inline void
+brw_add_validated_bo(struct brw_context *brw, dri_bo *bo)
+{
+   assert(brw->state.validated_bo_count < ARRAY_SIZE(brw->state.validated_bos));
+
+   if (bo != NULL) {
+      dri_bo_reference(bo);
+      brw->state.validated_bos[brw->state.validated_bo_count++] = bo;
+   }
+};
 
 const struct brw_tracked_state brw_blend_constant_color;
 const struct brw_tracked_state brw_cc_unit;
@@ -84,6 +94,14 @@ const struct brw_tracked_state brw_indic
 const struct brw_tracked_state brw_vertices;
 
 /***********************************************************************
+ * brw_state.c
+ */
+void brw_validate_state(struct brw_context *brw);
+void brw_upload_state(struct brw_context *brw);
+void brw_init_state(struct brw_context *brw);
+void brw_destroy_state(struct brw_context *brw);
+
+/***********************************************************************
  * brw_state_cache.c
  */
 dri_bo *brw_cache_data(struct brw_cache *cache,
diff -urNap mesa_orig//src/mesa/drivers/dri/i965/brw_state_upload.c mesa_haihao//src/mesa/drivers/dri/i965/brw_state_upload.c
--- mesa_orig//src/mesa/drivers/dri/i965/brw_state_upload.c	2009-01-12 13:29:39.000000000 +0800
+++ mesa_haihao//src/mesa/drivers/dri/i965/brw_state_upload.c	2009-01-13 16:43:37.000000000 +0800
@@ -169,6 +169,18 @@ static void xor_states( struct brw_state
    result->cache = a->cache ^ b->cache;
 }
 
+static void
+brw_clear_validated_bos(struct brw_context *brw)
+{
+   int i;
+
+   /* Clear the last round of validated bos */
+   for (i = 0; i < brw->state.validated_bo_count; i++) {
+      dri_bo_unreference(brw->state.validated_bos[i]);
+      brw->state.validated_bos[i] = NULL;
+   }
+   brw->state.validated_bo_count = 0;
+}
 
 /***********************************************************************
  * Emit all state:
@@ -177,12 +189,15 @@ void brw_validate_state( struct brw_cont
 {
    struct intel_context *intel = &brw->intel;
    struct brw_state_flags *state = &brw->state.dirty;
-   GLuint i, count, pass = 0;
-   dri_bo *last_batch_bo = NULL;
+   GLuint i;
+
+   brw_clear_validated_bos(brw);
 
    state->mesa |= brw->intel.NewGLState;
    brw->intel.NewGLState = 0;
 
+   brw_add_validated_bo(brw, intel->batch->buf);
+
    if (brw->wrap)
       state->brw |= BRW_NEW_CONTEXT;
 
@@ -211,8 +226,6 @@ void brw_validate_state( struct brw_cont
 
    brw->intel.Fallback = 0;
 
-   count = 0;
-
    /* do prepare stage for all atoms */
    for (i = 0; i < Elements(atoms); i++) {
       const struct brw_tracked_state *atom = brw->state.atoms[i];
@@ -226,19 +239,15 @@ void brw_validate_state( struct brw_cont
         }
       }
    }
+}
 
-   if (brw->intel.Fallback)
-      return;
 
-   /* We're about to try to set up a coherent state in the batchbuffer for
-    * the emission of primitives.  If we exceed the aperture size in any of the
-    * emit() calls, we need to go back to square 1 and try setting up again.
-    */
-got_flushed:
-   dri_bo_unreference(last_batch_bo);
-   last_batch_bo = intel->batch->buf;
-   dri_bo_reference(last_batch_bo);
-   assert(pass++ <= 2);
+void brw_upload_state(struct brw_context *brw)
+{
+   struct brw_state_flags *state = &brw->state.dirty;
+   int i;
+
+   brw_clear_validated_bos(brw);
 
    if (INTEL_DEBUG) {
       /* Debug version which enforces various sanity checks on the
@@ -263,8 +272,6 @@ got_flushed:
 	 if (check_state(state, &atom->dirty)) {
 	    if (atom->emit) {
 	       atom->emit( brw );
-	       if (intel->batch->buf != last_batch_bo)
-		  goto got_flushed;
 	    }
 	 }
 
@@ -289,15 +296,11 @@ got_flushed:
 	 if (check_state(state, &atom->dirty)) {
 	    if (atom->emit) {
 	       atom->emit( brw );
-	       if (intel->batch->buf != last_batch_bo)
-		  goto got_flushed;
 	    }
 	 }
       }
    }
 
-   dri_bo_unreference(last_batch_bo);
-
    if (!brw->intel.Fallback)
       memset(state, 0, sizeof(*state));
 }