5 files changed, 24 insertions, 21 deletions
diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_draw.h b/src/gallium/drivers/freedreno/a5xx/fd5_draw.h
index 3edfc391fdf..c0d50b29cfd 100644
--- a/src/gallium/drivers/freedreno/a5xx/fd5_draw.h
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_draw.h
@@ -32,6 +32,7 @@
 #include "freedreno_draw.h"
 
 #include "fd5_context.h"
+#include "fd5_screen.h"
 
 /* some bits in common w/ a4xx: */
 #include "a4xx/fd4_draw.h"
diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_emit.c b/src/gallium/drivers/freedreno/a5xx/fd5_emit.c
index 944c62e29c3..9d17bda476c 100644
--- a/src/gallium/drivers/freedreno/a5xx/fd5_emit.c
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_emit.c
@@ -42,6 +42,7 @@
 #include "fd5_program.h"
 #include "fd5_rasterizer.h"
 #include "fd5_texture.h"
+#include "fd5_screen.h"
 #include "fd5_format.h"
 #include "fd5_zsa.h"
 
@@ -1101,7 +1102,15 @@ t7              opcode: CP_WAIT_FOR_IDLE (26) (1 dwords)
 static void
 fd5_emit_ib(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
 {
+	/* for debug after a lock up, write a unique counter value
+	 * to scratch6 for each IB, to make it easier to match up
+	 * register dumps to cmdstream.  The combination of IB and
+	 * DRAW (scratch7) is enough to "triangulate" the particular
+	 * draw that caused lockup.
+	 */
+	emit_marker5(ring, 6);
 	__OUT_IB5(ring, target);
+	emit_marker5(ring, 6);
 }
 
 static void
diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_emit.h b/src/gallium/drivers/freedreno/a5xx/fd5_emit.h
index 809ae475cee..bed52d4e87f 100644
--- a/src/gallium/drivers/freedreno/a5xx/fd5_emit.h
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_emit.h
@@ -33,6 +33,7 @@
 #include "fd5_context.h"
 #include "fd5_format.h"
 #include "fd5_program.h"
+#include "fd5_screen.h"
 #include "ir3_shader.h"
 
 struct fd_ringbuffer;
diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_screen.h b/src/gallium/drivers/freedreno/a5xx/fd5_screen.h
index ba0c7f15ff9..0a65b3b0737 100644
--- a/src/gallium/drivers/freedreno/a5xx/fd5_screen.h
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_screen.h
@@ -29,6 +29,19 @@
 
 #include "pipe/p_screen.h"
 
+#include "freedreno_util.h"
+
+#include "a5xx.xml.h"
+
 void fd5_screen_init(struct pipe_screen *pscreen);
 
+static inline void
+emit_marker5(struct fd_ringbuffer *ring, int scratch_idx)
+{
+	extern unsigned marker_cnt;
+	unsigned reg = REG_A5XX_CP_SCRATCH_REG(scratch_idx);
+	OUT_PKT4(ring, reg, 1);
+	OUT_RING(ring, ++marker_cnt);
+}
+
 #endif /* FD5_SCREEN_H_ */
diff --git a/src/gallium/drivers/freedreno/freedreno_util.h b/src/gallium/drivers/freedreno/freedreno_util.h
index bfeec4c17dd..b3511d8f15e 100644
--- a/src/gallium/drivers/freedreno/freedreno_util.h
+++ b/src/gallium/drivers/freedreno/freedreno_util.h
@@ -183,7 +183,6 @@ fd_half_precision(struct pipe_framebuffer_state *pfb)
 #define LOG_DWORDS 0
 
 static inline void emit_marker(struct fd_ringbuffer *ring, int scratch_idx);
-static inline void emit_marker5(struct fd_ringbuffer *ring, int scratch_idx);
 
 static inline void
 OUT_RING(struct fd_ringbuffer *ring, uint32_t data)
@@ -372,14 +371,6 @@ __OUT_IB5(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
 {
 	unsigned count = fd_ringbuffer_cmd_count(target);
 
-	/* for debug after a lock up, write a unique counter value
-	 * to scratch6 for each IB, to make it easier to match up
-	 * register dumps to cmdstream.  The combination of IB and
-	 * DRAW (scratch7) is enough to "triangulate" the particular
-	 * draw that caused lockup.
-	 */
-	emit_marker5(ring, 6);
-
 	for (unsigned i = 0; i < count; i++) {
 		uint32_t dwords;
 		OUT_PKT7(ring, CP_INDIRECT_BUFFER, 3);
@@ -387,8 +378,6 @@ __OUT_IB5(struct fd_ringbuffer *ring, struct fd_ringbuffer *target)
 		assert(dwords > 0);
 		OUT_RING(ring, dwords);
 	}
-
-	emit_marker5(ring, 6);
 }
 
 /* CP_SCRATCH_REG4 is used to hold base address for query results: */
@@ -409,16 +398,6 @@ emit_marker(struct fd_ringbuffer *ring, int scratch_idx)
 	OUT_RING(ring, ++marker_cnt);
 }
 
-static inline void
-emit_marker5(struct fd_ringbuffer *ring, int scratch_idx)
-{
-	extern unsigned marker_cnt;
-//XXX	unsigned reg = REG_A5XX_CP_SCRATCH_REG(scratch_idx);
-	unsigned reg = 0x00000b78 + scratch_idx;
-	OUT_PKT4(ring, reg, 1);
-	OUT_RING(ring, ++marker_cnt);
-}
-
 /* helper to get numeric value from environment variable..  mostly
  * just leaving this here because it is helpful to brute-force figure
  * out unknown formats, etc, which blob driver does not support: