[965] Bring back no-backing-store hack for classic-mode performance.

This is not the complete version of the previous hack -- most importantly,
the NO_FENCE_SUBDATA is missing.  However, the previous implementation of that
lacked some required synchronization for the state cache.

This improves openarena performance by almost 50%, to about a quarter of what
it was before bufmgr conversion.

4 files changed, 79 insertions, 27 deletions

diff --git a/src/mesa/drivers/dri/common/dri_bufmgr.h b/src/mesa/drivers/dri/common/dri_bufmgr.h
index f1863c7056..1290fa956b 100644
--- a/src/mesa/drivers/dri/common/dri_bufmgr.h
+++ b/src/mesa/drivers/dri/common/dri_bufmgr.h
@@ -202,6 +202,10 @@ dri_bufmgr *dri_bufmgr_fake_init(unsigned long low_offset, void *low_virtual,
 				 int (*fence_wait)(void *private,
 						   unsigned int cookie),
 				 void *driver_priv);
+void dri_bo_fake_disable_backing_store(dri_bo *bo,
+				       void (*invalidate_cb)(dri_bo *bo,
+							     void *ptr),
+				       void *ptr);
 void dri_bufmgr_destroy(dri_bufmgr *bufmgr);
 dri_bo *dri_ttm_bo_create_from_handle(dri_bufmgr *bufmgr, const char *name,
 				      unsigned int handle);
diff --git a/src/mesa/drivers/dri/common/dri_bufmgr_fake.c b/src/mesa/drivers/dri/common/dri_bufmgr_fake.c
index c31bd7c6b0..9081f23e4a 100644
--- a/src/mesa/drivers/dri/common/dri_bufmgr_fake.c
+++ b/src/mesa/drivers/dri/common/dri_bufmgr_fake.c
@@ -181,7 +181,7 @@ typedef struct _dri_bo_fake {
 
    struct block *block;
    void *backing_store;
-   void (*invalidate_cb)(dri_bufmgr *bufmgr, void * );
+   void (*invalidate_cb)(dri_bo *bo, void *ptr);
    void *invalidate_ptr;
 } dri_bo_fake;
 
@@ -318,9 +318,9 @@ static void
 free_backing_store(dri_bo *bo)
 {
    dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
-   assert(!(bo_fake->flags & (BM_PINNED|BM_NO_BACKING_STORE)));
 
    if (bo_fake->backing_store) {
+      assert(!(bo_fake->flags & (BM_PINNED|BM_NO_BACKING_STORE)));
       ALIGN_FREE(bo_fake->backing_store);
       bo_fake->backing_store = NULL;
    }
@@ -329,11 +329,10 @@ free_backing_store(dri_bo *bo)
 static void
 set_dirty(dri_bo *bo)
 {
-   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
    dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
 
-   if (bo_fake->flags & BM_NO_BACKING_STORE)
-      bo_fake->invalidate_cb(&bufmgr_fake->bufmgr, bo_fake->invalidate_ptr);
+   if (bo_fake->flags & BM_NO_BACKING_STORE && bo_fake->invalidate_cb != NULL)
+      bo_fake->invalidate_cb(bo, bo_fake->invalidate_ptr);
 
    assert(!(bo_fake->flags & BM_PINNED));
 
@@ -678,6 +677,43 @@ dri_fake_bo_unreference(dri_bo *bo)
 }
 
 /**
+ * Set the buffer as not requiring backing store, and instead get the callback
+ * invoked whenever it would be set dirty.
+ *
+ * \param dont_fence_subdata Disables waiting for idle before mapping the
+ *	  buffer.
+ */
+void dri_bo_fake_disable_backing_store(dri_bo *bo,
+				       void (*invalidate_cb)(dri_bo *bo,
+							     void *ptr),
+				       void *ptr)
+{
+   dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bo->bufmgr;
+   dri_bo_fake *bo_fake = (dri_bo_fake *)bo;
+
+   _glthread_LOCK_MUTEX(bufmgr_fake->mutex);
+
+   if (bo_fake->backing_store)
+      free_backing_store(bo);
+
+   bo_fake->flags |= BM_NO_BACKING_STORE;
+
+   DBG("disable_backing_store set buf %d dirty\n", bo_fake->id);
+   bo_fake->dirty = 1;
+   bo_fake->invalidate_cb = invalidate_cb;
+   bo_fake->invalidate_ptr = ptr;
+
+   /* Note that it is invalid right from the start.  Also note
+    * invalidate_cb is called with the bufmgr locked, so cannot
+    * itself make bufmgr calls.
+    */
+   if (invalidate_cb != NULL)
+      invalidate_cb(bo, ptr);
+
+   _glthread_UNLOCK_MUTEX(bufmgr_fake->mutex);
+}
+
+/**
  * Map a buffer into bo->virtual, allocating either card memory space (If
  * BM_NO_BACKING_STORE or BM_PINNED) or backing store, as necessary.
  */
diff --git a/src/mesa/drivers/dri/i965/brw_draw.c b/src/mesa/drivers/dri/i965/brw_draw.c
index 223a1c1c9d..87e2202029 100644
--- a/src/mesa/drivers/dri/i965/brw_draw.c
+++ b/src/mesa/drivers/dri/i965/brw_draw.c
@@ -447,13 +447,6 @@ void brw_draw_prims( GLcontext *ctx,
    }
 }
 
-#if 0
-static void brw_invalidate_vbo_cb( struct intel_context *intel, void *ptr )
-{
-   /* nothing to do, we don't rely on the contents being preserved */
-}
-#endif
-
 void brw_draw_init( struct brw_context *brw )
 {
    GLcontext *ctx = &brw->intel.ctx;
@@ -469,23 +462,24 @@ void brw_draw_init( struct brw_context *brw )
    for (i = 0; i < BRW_NR_UPLOAD_BUFS; i++) {
       brw->vb.upload.vbo[i] = ctx->Driver.NewBufferObject(ctx, 1, GL_ARRAY_BUFFER_ARB);
 
-#if 0
-      /* NOTE:  These are set to no-backing-store.
+      ctx->Driver.BufferData(ctx,
+			     GL_ARRAY_BUFFER_ARB,
+			     BRW_UPLOAD_INIT_SIZE,
+			     NULL,
+			     GL_DYNAMIC_DRAW_ARB,
+			     brw->vb.upload.vbo[i]);
+
+      /* Set the internal VBOs to no-backing-store.  We only use them as a
+       * temporary within a brw_try_draw_prims while the lock is held.
        */
-      bmBufferSetInvalidateCB(&brw->intel,
-			      intel_bufferobj_buffer(intel_buffer_object(brw->vb.upload.vbo[i])),
-			      brw_invalidate_vbo_cb,
-			      &brw->intel,
-			      GL_TRUE);
-#endif
-   }
+      if (!brw->intel.intelScreen->ttm) {
+	 struct intel_buffer_object *intel_bo =
+	    intel_buffer_object(brw->vb.upload.vbo[i]);
 
-   ctx->Driver.BufferData( ctx, 
-			   GL_ARRAY_BUFFER_ARB, 
-			   BRW_UPLOAD_INIT_SIZE,
-			   NULL,
-			   GL_DYNAMIC_DRAW_ARB,
-			   brw->vb.upload.vbo[0] );
+	 dri_bo_fake_disable_backing_store(intel_bufferobj_buffer(intel_bo),
+					   NULL, NULL);
+      }
+   }
 }
 
 void brw_draw_destroy( struct brw_context *brw )
diff --git a/src/mesa/drivers/dri/i965/brw_state_pool.c b/src/mesa/drivers/dri/i965/brw_state_pool.c
index 63daa9da34..0826d1ded3 100644
--- a/src/mesa/drivers/dri/i965/brw_state_pool.c
+++ b/src/mesa/drivers/dri/i965/brw_state_pool.c
@@ -70,6 +70,14 @@ void brw_invalidate_pool( struct intel_context *intel,
    brw_clear_all_caches(pool->brw);
 }
 
+static void
+brw_invalidate_pool_cb(dri_bo *bo, void *ptr)
+{
+   struct brw_mem_pool *pool = ptr;
+   struct brw_context *brw = pool->brw;
+
+   brw_invalidate_pool(&brw->intel, pool);
+}
 
 static void brw_init_pool( struct brw_context *brw,
 			   GLuint pool_id,
@@ -84,6 +92,16 @@ static void brw_init_pool( struct brw_context *brw,
    pool->buffer = dri_bo_alloc(brw->intel.intelScreen->bufmgr,
 			       (pool_id == BRW_GS_POOL) ? "GS pool" : "SS pool",
 			       size, 4096, DRM_BO_FLAG_MEM_TT);
+
+   /* Disable the backing store for the state cache.  It's not worth the
+    * cost of keeping a backing store copy, since we can just regenerate
+    * the contents at approximately the same cost as the memcpy, and only
+    * if the contents are lost.
+    */
+   if (!brw->intel.intelScreen->ttm) {
+      dri_bo_fake_disable_backing_store(pool->buffer, brw_invalidate_pool_cb,
+					pool);
+   }
 }
 
 static void brw_destroy_pool( struct brw_context *brw,