Skip to contentSkip to conversation

tests/gem_ringfill: Exercise all rings

PATCH In review Chris Wilson 2 participants 4 messages Feb 20, 2012 – Feb 22, 2012 1 message not in the archive Library code

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

PATCH 1/1 tests/gem_ringfill: Exercise all rings

On SandyBridge, the BLT commands were split from the RENDER commands as well as the BSD split inherited from Ironlake. So we need to make sure we do exercise each ring, and in order to do so we also need to make sure each batch takes longer to execute than it takes for us to submit it.

---
 lib/intel_gpu_tools.h |    5 ++
 lib/rendercopy.h      |    5 ++
 tests/gem_ringfill.c  |  174 ++++++++++++++++++++++++++++++++++---------------
 3 files changed, 131 insertions(+), 53 deletions(-)
lib/intel_gpu_tools.h +5 -0
index f46abfa..eb21a16 100644
@@ -25,6 +25,9 @@
  *
  */
 
+#ifndef INTEL_GPU_TOOLS_H
+#define INTEL_GPU_TOOLS_H
+
 #include <stdint.h>
 #include <sys/types.h>
 #include <pciaccess.h>
@@ -94,3 +97,5 @@ extern enum pch_type pch;
 void intel_check_pch(void);
 
 #define HAS_CPT (pch == PCH_CPT)
+
+#endif /* INTEL_GPU_TOOLS_H */
lib/rendercopy.h +5 -0
index 7547ac4..4fcc817 100644
@@ -58,6 +58,11 @@ static inline unsigned buf_height(struct scratch_buf *buf)
 	return buf->size/buf->stride;
 }
 
+typedef void (*render_copyfunc_t)(struct intel_batchbuffer *batch,
+				  struct scratch_buf *src, unsigned src_x, unsigned src_y,
+				  unsigned width, unsigned height,
+				  struct scratch_buf *dst, unsigned dst_x, unsigned dst_y);
+
 void gen6_render_copyfunc(struct intel_batchbuffer *batch,
 			  struct scratch_buf *src, unsigned src_x, unsigned src_y,
 			  unsigned width, unsigned height,
tests/gem_ringfill.c +121 -53

210 lines not on this page · raw message

Signature · 1 line
1.7.9
#2 lore raw
Quoted 5 lines · Chris Wilson

On SandyBridge, the BLT commands were split from the RENDER commands as well as the BSD split inherited from Ironlake. So we need to make sure we do exercise each ring, and in order to do so we also need to make sure each batch takes longer to execute than it takes for us to submit it.

Whilst this found the bug, it does presume the BLT and RENDER copy are equally fast. So on a SNB without the autoreported head, it fails to fill the BLT ring as the untiled RENDER copy is far too slow. -Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#3 lore raw

PATCH 1/1 tests/gem_ringfill: Exercise all rings

On SandyBridge, the BLT commands were split from the RENDER commands as well as the BSD split inherited from Ironlake. So we need to make sure we do exercise each ring, and in order to do so we also need to make sure each batch takes longer to execute than it takes for us to submit it.

v2: Exercise each ring sequentially.

---
 lib/intel_gpu_tools.h |    5 ++
 lib/rendercopy.h      |    5 ++
 tests/gem_ringfill.c  |  185 ++++++++++++++++++++++++++++++++++---------------
 3 files changed, 140 insertions(+), 55 deletions(-)
lib/intel_gpu_tools.h +5 -0
index f46abfa..eb21a16 100644
@@ -25,6 +25,9 @@
  *
  */
 
+#ifndef INTEL_GPU_TOOLS_H
+#define INTEL_GPU_TOOLS_H
+
 #include <stdint.h>
 #include <sys/types.h>
 #include <pciaccess.h>
@@ -94,3 +97,5 @@ extern enum pch_type pch;
 void intel_check_pch(void);
 
 #define HAS_CPT (pch == PCH_CPT)
+
+#endif /* INTEL_GPU_TOOLS_H */
lib/rendercopy.h +5 -0
index 7547ac4..4fcc817 100644
@@ -58,6 +58,11 @@ static inline unsigned buf_height(struct scratch_buf *buf)
 	return buf->size/buf->stride;
 }
 
+typedef void (*render_copyfunc_t)(struct intel_batchbuffer *batch,
+				  struct scratch_buf *src, unsigned src_x, unsigned src_y,
+				  unsigned width, unsigned height,
+				  struct scratch_buf *dst, unsigned dst_x, unsigned dst_y);
+
 void gen6_render_copyfunc(struct intel_batchbuffer *batch,
 			  struct scratch_buf *src, unsigned src_x, unsigned src_y,
 			  unsigned width, unsigned height,
tests/gem_ringfill.c +130 -55

228 lines not on this page · raw message

Signature · 1 line
1.7.9
#4 lore raw
Quoted 9 lines · Chris Wilson

On SandyBridge, the BLT commands were split from the RENDER commands as well as the BSD split inherited from Ironlake. So we need to make sure we do exercise each ring, and in order to do so we also need to make sure each batch takes longer to execute than it takes for us to submit it.

v2: Exercise each ring sequentially.

Slurped in, thanks for improving the test.
-Daniel

Signature · 3 lines
Daniel Vetter
Mail: [email protected]
Mobile: +41 (0)79 365 57 48