i965: Emit depth stalls and flushes before changing depth state on Gen6+.
Fixes OpenArena on Gen7. Technically, adding only the first depth stall
fixes it, but the documentation says to do all three, and the Windows
driver seems to do it.
Not observed to fix anything on Gen6 yet.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=38863
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
diff --git a/src/mesa/drivers/dri/i965/brw_misc_state.c b/src/mesa/drivers/dri/i965/brw_misc_state.c
index ce4fc84..7b83ff5 100644
--- a/src/mesa/drivers/dri/i965/brw_misc_state.c
+++ b/src/mesa/drivers/dri/i965/brw_misc_state.c
@@ -226,8 +226,10 @@
/* 3DSTATE_DEPTH_BUFFER, 3DSTATE_STENCIL_BUFFER are both
* non-pipelined state that will need the PIPE_CONTROL workaround.
*/
- if (intel->gen == 6)
+ if (intel->gen == 6) {
intel_emit_post_sync_nonzero_flush(intel);
+ intel_emit_depth_stall_flushes(intel);
+ }
/*
* If either depth or stencil buffer has packed depth/stencil format,
diff --git a/src/mesa/drivers/dri/i965/gen7_misc_state.c b/src/mesa/drivers/dri/i965/gen7_misc_state.c
index 7544f96..9eb75e2 100644
--- a/src/mesa/drivers/dri/i965/gen7_misc_state.c
+++ b/src/mesa/drivers/dri/i965/gen7_misc_state.c
@@ -80,6 +80,8 @@
struct intel_renderbuffer *drb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
struct intel_renderbuffer *srb = intel_get_renderbuffer(fb, BUFFER_STENCIL);
+ intel_emit_depth_stall_flushes(intel);
+
/* Gen7 doesn't support packed depth/stencil */
assert(srb == NULL || srb != drb);
diff --git a/src/mesa/drivers/dri/intel/intel_batchbuffer.c b/src/mesa/drivers/dri/intel/intel_batchbuffer.c
index bb14cc2..37c13c9 100644
--- a/src/mesa/drivers/dri/intel/intel_batchbuffer.c
+++ b/src/mesa/drivers/dri/intel/intel_batchbuffer.c
@@ -297,6 +297,45 @@
}
/**
+ * Restriction [DevSNB, DevIVB]:
+ *
+ * Prior to changing Depth/Stencil Buffer state (i.e. any combination of
+ * 3DSTATE_DEPTH_BUFFER, 3DSTATE_CLEAR_PARAMS, 3DSTATE_STENCIL_BUFFER,
+ * 3DSTATE_HIER_DEPTH_BUFFER) SW must first issue a pipelined depth stall
+ * (PIPE_CONTROL with Depth Stall bit set), followed by a pipelined depth
+ * cache flush (PIPE_CONTROL with Depth Flush Bit set), followed by
+ * another pipelined depth stall (PIPE_CONTROL with Depth Stall bit set),
+ * unless SW can otherwise guarantee that the pipeline from WM onwards is
+ * already flushed (e.g., via a preceding MI_FLUSH).
+ */
+void
+intel_emit_depth_stall_flushes(struct intel_context *intel)
+{
+ assert(intel->gen >= 6 && intel->gen <= 7);
+
+ BEGIN_BATCH(4);
+ OUT_BATCH(_3DSTATE_PIPE_CONTROL);
+ OUT_BATCH(PIPE_CONTROL_DEPTH_STALL);
+ OUT_BATCH(0); /* address */
+ OUT_BATCH(0); /* write data */
+ ADVANCE_BATCH()
+
+ BEGIN_BATCH(4);
+ OUT_BATCH(_3DSTATE_PIPE_CONTROL);
+ OUT_BATCH(PIPE_CONTROL_DEPTH_CACHE_FLUSH);
+ OUT_BATCH(0); /* address */
+ OUT_BATCH(0); /* write data */
+ ADVANCE_BATCH();
+
+ BEGIN_BATCH(4);
+ OUT_BATCH(_3DSTATE_PIPE_CONTROL);
+ OUT_BATCH(PIPE_CONTROL_DEPTH_STALL);
+ OUT_BATCH(0); /* address */
+ OUT_BATCH(0); /* write data */
+ ADVANCE_BATCH();
+}
+
+/**
* Emits a PIPE_CONTROL with a non-zero post-sync operation, for
* implementing two workarounds on gen6. From section 1.4.7.1
* "PIPE_CONTROL" of the Sandy Bridge PRM volume 2 part 1: