[Mesa-dev] [PATCH 3/4] i965: adjust gen6+ timestamp pipe_control writes

[Daniel Vetter]

Similar treatment to the depth count pipe_control writes
- Add the CS_STALL workaround, timestamp writes are non-zero post-sync
  ops, too.
- Also ensure that we write the full 64bits by using the 5 dword long
  variant of pipe_control.

v2: Implement |(5-2) suggestion from Kenneth Graunke.
---
 src/mesa/drivers/dri/i965/brw_queryobj.c |   32 ++++++++++++++++++++++++++----
 1 file changed, 28 insertions(+), 4 deletions(-)

diff --git a/src/mesa/drivers/dri/i965/brw_queryobj.c b/src/mesa/drivers/dri/i965/brw_queryobj.c
index 8049580..a6b3e22 100644
--- a/src/mesa/drivers/dri/i965/brw_queryobj.c
+++ b/src/mesa/drivers/dri/i965/brw_queryobj.c
@@ -131,14 +131,26 @@ brw_begin_query(struct gl_context *ctx, struct gl_query_object *q)
 				     4096, 4096);
 
       if (intel->gen >= 6) {
-	  BEGIN_BATCH(4);
-	  OUT_BATCH(_3DSTATE_PIPE_CONTROL);
+	  BEGIN_BATCH(9);
+
+          /* 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 |
+                    PIPE_CONTROL_STALL_AT_SCOREBOARD);
+          OUT_BATCH(0);
+          OUT_BATCH(0);
+
+	  OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
 	  OUT_BATCH(PIPE_CONTROL_WRITE_TIMESTAMP);
 	  OUT_RELOC(query->bo,
 		  I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 		  PIPE_CONTROL_GLOBAL_GTT_WRITE |
 		  0);
 	  OUT_BATCH(0);
+	  OUT_BATCH(0);
 	  ADVANCE_BATCH();
       
       } else {
@@ -201,14 +213,26 @@ brw_end_query(struct gl_context *ctx, struct gl_query_object *q)
    switch (query->Base.Target) {
    case GL_TIME_ELAPSED_EXT:
       if (intel->gen >= 6) {
-	  BEGIN_BATCH(4);
-	  OUT_BATCH(_3DSTATE_PIPE_CONTROL);
+	  BEGIN_BATCH(9);
+
+          /* 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 |
+                    PIPE_CONTROL_STALL_AT_SCOREBOARD);
+          OUT_BATCH(0);
+          OUT_BATCH(0);
+
+	  OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
 	  OUT_BATCH(PIPE_CONTROL_WRITE_TIMESTAMP);
 	  OUT_RELOC(query->bo,
 		  I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 		  PIPE_CONTROL_GLOBAL_GTT_WRITE |
 		  8);
 	  OUT_BATCH(0);
+	  OUT_BATCH(0);
 	  ADVANCE_BATCH();
       
       } else {
-- 
1.7.10.4