i965: Track the windowizer's dispatch for kill pixel, promoted, and OQ

Looks like the problem was we weren't passing the depth to the render
target as expected, so the chip would wedge.  Fixes GPU hang in
occlusion-query-discard.

Bug #30097

3 files changed, 23 insertions, 6 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c
index d70be7bda2..17ff3e2b2b 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.c
+++ b/src/mesa/drivers/dri/i965/brw_wm.c
@@ -215,6 +215,7 @@ static void do_wm_prog( struct brw_context *brw,
 static void brw_wm_populate_key( struct brw_context *brw,
 				 struct brw_wm_prog_key *key )
 {
+   struct intel_context *intel = &brw->intel;
    GLcontext *ctx = &brw->intel.ctx;
    /* BRW_NEW_FRAGMENT_PROGRAM */
    const struct brw_fragment_program *fp = 
@@ -279,7 +280,8 @@ static void brw_wm_populate_key( struct brw_context *brw,
       }
    }
 	 
-   brw_wm_lookup_iz(line_aa,
+   brw_wm_lookup_iz(intel,
+		    line_aa,
 		    lookup,
 		    uses_depth,
 		    key);
diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h
index c28acad24b..fed1372969 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.h
+++ b/src/mesa/drivers/dri/i965/brw_wm.h
@@ -299,7 +299,8 @@ void brw_wm_print_insn( struct brw_wm_compile *c,
 void brw_wm_print_program( struct brw_wm_compile *c,
 			   const char *stage );
 
-void brw_wm_lookup_iz( GLuint line_aa,
+void brw_wm_lookup_iz( struct intel_context *intel,
+		       GLuint line_aa,
 		       GLuint lookup,
 		       GLboolean ps_uses_depth,
 		       struct brw_wm_prog_key *key );
diff --git a/src/mesa/drivers/dri/i965/brw_wm_iz.c b/src/mesa/drivers/dri/i965/brw_wm_iz.c
index 8505ef1951..62e556698b 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_iz.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_iz.c
@@ -120,24 +120,38 @@ const struct {
  * \param line_aa  AA_NEVER, AA_ALWAYS or AA_SOMETIMES
  * \param lookup  bitmask of IZ_* flags
  */
-void brw_wm_lookup_iz( GLuint line_aa,
+void brw_wm_lookup_iz( struct intel_context *intel,
+		       GLuint line_aa,
 		       GLuint lookup,
 		       GLboolean ps_uses_depth,
 		       struct brw_wm_prog_key *key )
 {
    GLuint reg = 2;
+   GLboolean kill_stats_promoted_workaround = GL_FALSE;
 
    assert (lookup < IZ_BIT_MAX);
-      
+
+   /* Crazy workaround in the windowizer, which we need to track in
+    * our register allocation and render target writes.  See the "If
+    * statistics are enabled..." paragraph of 11.5.3.2: Early Depth
+    * Test Cases [Pre-DevGT] of the 3D Pipeline - Windower B-Spec.
+    */
+   if (intel->stats_wm &&
+       (lookup & IZ_PS_KILL_ALPHATEST_BIT) &&
+       wm_iz_table[lookup].mode == P) {
+      kill_stats_promoted_workaround = GL_TRUE;
+   }
+
    if (lookup & IZ_PS_COMPUTES_DEPTH_BIT)
       key->computes_depth = 1;
 
-   if (wm_iz_table[lookup].sd_present || ps_uses_depth) {
+   if (wm_iz_table[lookup].sd_present || ps_uses_depth ||
+       kill_stats_promoted_workaround) {
       key->source_depth_reg = reg;
       reg += 2;
    }
 
-   if (wm_iz_table[lookup].sd_to_rt)
+   if (wm_iz_table[lookup].sd_to_rt || kill_stats_promoted_workaround)
       key->source_depth_to_render_target = 1;
 
    if (wm_iz_table[lookup].ds_present || line_aa != AA_NEVER) {