[Mesa-dev] [PATCH 1/4] i965: tackle the occlusion query pipe control mess

[Daniel Vetter]

- Separate out the depth stall from the depth count write, workarounds
  say that a depth stall needs to be preceeded with a non-zero
  post-sync op.
- Implement the cs stall workaround like the kernel does.

I've hoped that this would fix a occlusion query issue on snb, but
alas, it doesn't seem to help.
---
 src/mesa/drivers/dri/i965/brw_queryobj.c |   53 ++++++++++++++++++++++--------
 1 file changed, 39 insertions(+), 14 deletions(-)

diff --git a/src/mesa/drivers/dri/i965/brw_queryobj.c b/src/mesa/drivers/dri/i965/brw_queryobj.c
index 240fe32..52db050 100644
--- a/src/mesa/drivers/dri/i965/brw_queryobj.c
+++ b/src/mesa/drivers/dri/i965/brw_queryobj.c
@@ -337,22 +337,34 @@ brw_emit_query_begin(struct brw_context *brw)
       return;
 
    if (intel->gen >= 6) {
-       BEGIN_BATCH(8);
+       BEGIN_BATCH(12);
 
-       /* workaround: CS stall required before depth stall. */
+       /* Workaround: A non-zero post-sync op (i.e. the DEPTH_COUNT write below
+	* needs a pipe control with CS_STALL set beforehand.
+	* Workaround: CS_STALL can't be set alone, we pick STALL_AT_SCOREBOARD
+	* like the kernel. */
        OUT_BATCH(_3DSTATE_PIPE_CONTROL);
-       OUT_BATCH(PIPE_CONTROL_CS_STALL);
-       OUT_BATCH(0); /* write address */
-       OUT_BATCH(0); /* write data */
+       OUT_BATCH(PIPE_CONTROL_CS_STALL |
+		 PIPE_CONTROL_STALL_AT_SCOREBOARD);
+       OUT_BATCH(0);
+       OUT_BATCH(0);
 
+       /* The actual DEPTH_COUNT write. */
        OUT_BATCH(_3DSTATE_PIPE_CONTROL);
-       OUT_BATCH(PIPE_CONTROL_DEPTH_STALL |
-	         PIPE_CONTROL_WRITE_DEPTH_COUNT);
+       OUT_BATCH(PIPE_CONTROL_WRITE_DEPTH_COUNT);
        OUT_RELOC(brw->query.bo,
 	         I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 		 PIPE_CONTROL_GLOBAL_GTT_WRITE |
 		 ((brw->query.index * 2) * sizeof(uint64_t)));
        OUT_BATCH(0);
+
+       /* We need to emit depth stall to get the right value for the depth
+	* count. As a workaround this needs a preceeding pipe control with a
+	* non-zero post-sync op, the depth count write above does that for us. */
+       OUT_BATCH(_3DSTATE_PIPE_CONTROL);
+       OUT_BATCH(PIPE_CONTROL_DEPTH_STALL);
+       OUT_BATCH(0);
+       OUT_BATCH(0);
        ADVANCE_BATCH();
        
    } else {
@@ -395,21 +407,34 @@ brw_emit_query_end(struct brw_context *brw)
       return;
 
    if (intel->gen >= 6) {
-       BEGIN_BATCH(8);
-       /* workaround: CS stall required before depth stall. */
+       BEGIN_BATCH(12);
+
+       /* Workaround: A non-zero post-sync op (i.e. the DEPTH_COUNT write below
+	* needs a pipe control with CS_STALL set beforehand.
+	* Workaround: CS_STALL can't be set alone, we pick STALL_AT_SCOREBOARD
+	* like the kernel. */
        OUT_BATCH(_3DSTATE_PIPE_CONTROL);
-       OUT_BATCH(PIPE_CONTROL_CS_STALL);
-       OUT_BATCH(0); /* write address */
-       OUT_BATCH(0); /* write data */
+       OUT_BATCH(PIPE_CONTROL_CS_STALL |
+		 PIPE_CONTROL_STALL_AT_SCOREBOARD);
+       OUT_BATCH(0);
+       OUT_BATCH(0);
 
+       /* The actual DEPTH_COUNT write. */
        OUT_BATCH(_3DSTATE_PIPE_CONTROL);
-       OUT_BATCH(PIPE_CONTROL_DEPTH_STALL |
-	         PIPE_CONTROL_WRITE_DEPTH_COUNT);
+       OUT_BATCH(PIPE_CONTROL_WRITE_DEPTH_COUNT);
        OUT_RELOC(brw->query.bo,
 	         I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 		 PIPE_CONTROL_GLOBAL_GTT_WRITE |
 		 ((brw->query.index * 2 + 1) * sizeof(uint64_t)));
        OUT_BATCH(0);
+
+       /* We need to emit depth stall to get the right value for the depth
+	* count. As a workaround this needs a preceeding pipe control with a
+	* non-zero post-sync op, the depth count write above does that for us. */
+       OUT_BATCH(_3DSTATE_PIPE_CONTROL);
+       OUT_BATCH(PIPE_CONTROL_DEPTH_STALL);
+       OUT_BATCH(0);
+       OUT_BATCH(0);
        ADVANCE_BATCH();
    
    } else {
-- 
1.7.10.4