Make sure bmBufferOffset is called for all active buffers every time

we render.  Currenly requires that some state be re-examined after
every LOCK_HARDWARE().

9 files changed, 20 insertions, 26 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index 48d2cce448..1137bfd2c7 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -135,6 +135,7 @@ struct brw_context;
 #define BRW_NEW_PSP                     0x800
 #define BRW_NEW_METAOPS                 0x1000
 #define BRW_NEW_FENCE                   0x2000
+#define BRW_NEW_LOCK                    0x4000
 
 
 
@@ -429,8 +430,8 @@ struct brw_context
    GLuint primitive;
 
    GLboolean emit_state_always;
-   
    GLboolean wrap;
+   GLboolean tmp_fallback;
 
    struct {
       struct brw_state_flags dirty;
@@ -443,9 +444,6 @@ struct brw_context
    } state;
 
    struct brw_state_pointers attribs;
-
-   GLboolean tmp_fallback;
-
    struct brw_mem_pool pool[BRW_MAX_POOL];
    struct brw_cache cache[BRW_MAX_CACHE];
    struct brw_cached_batch_item *cached_batch_items;
diff --git a/src/mesa/drivers/dri/i965/brw_misc_state.c b/src/mesa/drivers/dri/i965/brw_misc_state.c
index 49a8a6e4a8..29296c17e9 100644
--- a/src/mesa/drivers/dri/i965/brw_misc_state.c
+++ b/src/mesa/drivers/dri/i965/brw_misc_state.c
@@ -276,6 +276,7 @@ static void upload_depthbuffer(struct brw_context *brw)
    bd.dword1.bits.tiled_surface = intel->depth_region->tiled;
    bd.dword1.bits.surface_type = BRW_SURFACE_2D;
 
+   /* BRW_NEW_LOCK */
    bd.dword2_base_addr = bmBufferOffset(intel, region->buffer);    
 
    bd.dword3.bits.mipmap_layout = BRW_SURFACE_MIPMAPLAYOUT_BELOW;
@@ -292,7 +293,7 @@ static void upload_depthbuffer(struct brw_context *brw)
 const struct brw_tracked_state brw_depthbuffer = {
    .dirty = {
       .mesa = 0,
-      .brw = BRW_NEW_CONTEXT | BRW_NEW_FENCE,
+      .brw = BRW_NEW_CONTEXT | BRW_NEW_LOCK,
       .cache = 0
    },
    .update = upload_depthbuffer
@@ -505,9 +506,11 @@ static void upload_state_base_address( struct brw_context *brw )
    sba.header.opcode = CMD_STATE_BASE_ADDRESS;
    sba.header.length = 0x4;
 
+   /* BRW_NEW_LOCK */
    sba.bits0.general_state_address = bmBufferOffset(intel, brw->pool[BRW_GS_POOL].buffer) >> 5;
    sba.bits0.modify_enable = 1;
 
+   /* BRW_NEW_LOCK */
    sba.bits1.surface_state_address = bmBufferOffset(intel, brw->pool[BRW_SS_POOL].buffer) >> 5;
    sba.bits1.modify_enable = 1;
 
@@ -522,7 +525,7 @@ static void upload_state_base_address( struct brw_context *brw )
 const struct brw_tracked_state brw_state_base_address = {
    .dirty = {
       .mesa = 0,
-      .brw = BRW_NEW_CONTEXT | BRW_NEW_FENCE,
+      .brw = BRW_NEW_CONTEXT | BRW_NEW_LOCK,
       .cache = 0
    },
    .update = upload_state_base_address
diff --git a/src/mesa/drivers/dri/i965/brw_vtbl.c b/src/mesa/drivers/dri/i965/brw_vtbl.c
index 4a15a7763d..4896882034 100644
--- a/src/mesa/drivers/dri/i965/brw_vtbl.c
+++ b/src/mesa/drivers/dri/i965/brw_vtbl.c
@@ -123,6 +123,8 @@ static void brw_note_unlock( struct intel_context *intel )
 
    brw_pool_check_wrap(brw, &brw->pool[BRW_GS_POOL]);
    brw_pool_check_wrap(brw, &brw->pool[BRW_SS_POOL]);
+
+   brw_context(&intel->ctx)->state.dirty.brw |= BRW_NEW_LOCK;
 }
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_wm_fp.c b/src/mesa/drivers/dri/i965/brw_wm_fp.c
index 73df3e5ff5..203eeead0f 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_fp.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_fp.c
@@ -776,7 +776,7 @@ void brw_wm_pass_fp( struct brw_wm_compile *c )
 
    if (INTEL_DEBUG & DEBUG_WM) {
       _mesa_printf("\n\n\npre-fp:\n");
-/*       _mesa_print_program(&fp->program); */
+      _mesa_print_program(&fp->program.Base); 
       _mesa_printf("\n");
    }
 
diff --git a/src/mesa/drivers/dri/i965/brw_wm_state.c b/src/mesa/drivers/dri/i965/brw_wm_state.c
index 8e708d742d..accf1243ec 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_state.c
@@ -98,6 +98,10 @@ static void upload_wm_unit(struct brw_context *brw )
 		   
       assert(per_thread <= 12 * 1024);
       wm.thread2.per_thread_scratch_space = (per_thread / 1024) - 1;
+
+      /* XXX: could make this dynamic as this is so rarely active:
+       */
+      /* BRW_NEW_LOCK */
       wm.thread2.scratch_space_base_pointer = 
 	 bmBufferOffset(intel, brw->wm.scratch_buffer) >> 10;
    }
@@ -179,7 +183,7 @@ const struct brw_tracked_state brw_wm_unit = {
 
       .brw = (BRW_NEW_FRAGMENT_PROGRAM | 
 	      BRW_NEW_CURBE_OFFSETS |
-	      BRW_NEW_FENCE),
+	      BRW_NEW_LOCK),
 
       .cache = (CACHE_NEW_SURFACE | 
 		CACHE_NEW_WM_PROG | 
diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
index 51a56d6745..6ccf56e41a 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
@@ -137,6 +137,7 @@ void brw_update_texture_surface( GLcontext *ctx,
     */
 /*    surf->ss0.data_return_format = BRW_SURFACERETURNFORMAT_S1; */
 
+   /* BRW_NEW_LOCK */
    surf->ss1.base_addr = bmBufferOffset(intel,
 					intelObj->mt->region->buffer);
 
@@ -198,6 +199,7 @@ static void upload_wm_surfaces(struct brw_context *brw )
       surf.ss0.writedisable_blue =  !brw->attribs.Color->ColorMask[2];
       surf.ss0.writedisable_alpha = !brw->attribs.Color->ColorMask[3];
 
+      /* BRW_NEW_LOCK */
       surf.ss1.base_addr = bmBufferOffset(&brw->intel, region->buffer);
 
 
@@ -240,7 +242,7 @@ const struct brw_tracked_state brw_wm_surfaces = {
    .dirty = {
       .mesa = _NEW_COLOR | _NEW_TEXTURE | _NEW_BUFFERS,
       .brw = (BRW_NEW_CONTEXT | 
-	      BRW_NEW_FENCE),	/* required for bmBufferOffset */
+	      BRW_NEW_LOCK),	/* required for bmBufferOffset */
       .cache = 0
    },
    .update = upload_wm_surfaces
diff --git a/src/mesa/drivers/dri/i965/intel_context.c b/src/mesa/drivers/dri/i965/intel_context.c
index 37e51f376d..e21c7d89b9 100644
--- a/src/mesa/drivers/dri/i965/intel_context.c
+++ b/src/mesa/drivers/dri/i965/intel_context.c
@@ -201,6 +201,7 @@ static const struct dri_debug_control debug_control[] =
     { "sing",  DEBUG_SINGLE_THREAD },
     { "thre",  DEBUG_SINGLE_THREAD },
     { "wm",    DEBUG_WM },
+    { "vs",    DEBUG_VS },
     { NULL,    0 }
 };
 
@@ -558,7 +559,6 @@ static void intelContendedLock( struct intel_context *intel, GLuint flags )
    /* Lost context?
     */
    if (sarea->ctxOwner != me) {
-      intel->perf_boxes |= I830_BOX_LOST_CONTEXT;
       sarea->ctxOwner = me;
 
       /* Should also fence the frontbuffer even if ctxOwner doesn't
@@ -572,12 +572,6 @@ static void intelContendedLock( struct intel_context *intel, GLuint flags )
       intel->vtbl.lost_hardware( intel );
    }
 
-   /* Because the X server issues drawing commands without properly
-    * fencing them, we need to be paraniod about waiting for hardware
-    * rendering to finish after a contended lock.
-    */
-   intel->flushBeforeFallback = GL_TRUE;
-
    /* Drawable changed?
     */
    if (dPriv && intel->lastStamp != dPriv->lastStamp) {
diff --git a/src/mesa/drivers/dri/i965/intel_context.h b/src/mesa/drivers/dri/i965/intel_context.h
index 8fa9619a52..0328cb900a 100644
--- a/src/mesa/drivers/dri/i965/intel_context.h
+++ b/src/mesa/drivers/dri/i965/intel_context.h
@@ -174,8 +174,6 @@ struct intel_context
    
    GLboolean aub_wrap;
 
-   GLboolean flushBeforeFallback;
-
    struct intel_batchbuffer *batch;
 
    struct {
@@ -200,7 +198,6 @@ struct intel_context
    GLboolean hw_stencil;
    GLboolean hw_stipple;
    GLboolean depth_buffer_is_float;
-   GLboolean perf_boxes;
    GLboolean no_hw;
    GLboolean no_rast;
    GLboolean thrashing;
@@ -385,6 +382,7 @@ extern int INTEL_DEBUG;
 #define DEBUG_SINGLE_THREAD   0x8000
 #define DEBUG_WM        0x10000
 #define DEBUG_URB       0x20000
+#define DEBUG_VS        0x40000
 
 
 #define PCI_CHIP_845_G			0x2562
diff --git a/src/mesa/drivers/dri/i965/intel_span.c b/src/mesa/drivers/dri/i965/intel_span.c
index da6c119bb2..c68def5a9f 100644
--- a/src/mesa/drivers/dri/i965/intel_span.c
+++ b/src/mesa/drivers/dri/i965/intel_span.c
@@ -209,13 +209,6 @@ void intelSpanRenderStart( GLcontext *ctx )
 
    LOCK_HARDWARE(intel);
 
-#if 0
-   if (intel->flushBeforeFallback) {
-      intelFinish(&intel->ctx);
-      intel->flushBeforeFallback = GL_FALSE;
-   }
-#endif
-
    /* Just map the framebuffer and all textures.  Bufmgr code will
     * take care of waiting on the necessary fences:
     */