sna: Force tiled modes for large pitches

If the surface is so big that the 2x2 texel sampling will cause a TLB
miss everytime, i.e. the row pitch exceeeds 4096, then we need to
encourage tiling to prevent attrocious performance.

For example, try downscaling a 2560x1600 background image on a gen3
device using I915_TILING_NONE...

Using slideshow-demo /usr/share/backgrounds/cosmos/whirlpool.jpg, on a
PineView netbook, fps goes from under 4 to over 40.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

3 files changed, 36 insertions, 8 deletions

diff --git a/src/sna/kgem.c b/src/sna/kgem.c
index 6fe6e936..38cf582c 100644
--- a/src/sna/kgem.c
+++ b/src/sna/kgem.c
@@ -1156,6 +1156,14 @@ int kgem_choose_tiling(struct kgem *kgem, int tiling, int width, int height, int
 	if (tiling < 0)
 		return tiling;
 
+	if (tiling && width * bpp > 8 * 4096) {
+		DBG(("%s: TLB miss between lines %dx%d (pitch=%d), forcing tiling %d\n",
+		     __FUNCTION__,
+		     width, height, width*bpp/8,
+		     tiling));
+		return -tiling;
+	}
+
 	if (tiling == I915_TILING_Y && height < 16) {
 		DBG(("%s: too short [%d] for TILING_Y\n",
 		     __FUNCTION__,height));
diff --git a/src/sna/kgem_debug.c b/src/sna/kgem_debug.c
index 20fe8a2e..745c7b90 100644
--- a/src/sna/kgem_debug.c
+++ b/src/sna/kgem_debug.c
@@ -80,6 +80,17 @@ static int kgem_debug_handle_is_fenced(struct kgem *kgem, uint32_t handle)
 	return 0;
 }
 
+static int kgem_debug_handle_tiling(struct kgem *kgem, uint32_t handle)
+{
+	struct kgem_bo *bo;
+
+	list_for_each_entry(bo, &kgem->next_request->buffers, request)
+		if (bo->handle == handle)
+			return bo->tiling;
+
+	return 0;
+}
+
 void
 kgem_debug_print(const uint32_t *data,
 		 uint32_t offset, unsigned int index,
@@ -254,11 +265,12 @@ decode_2d(struct kgem *kgem, uint32_t offset)
 			  data[3] & 0xffff, data[3] >> 16);
 		reloc = kgem_debug_get_reloc_entry(kgem, offset+4);
 		assert(reloc);
-		kgem_debug_print(data, offset, 4, "dst offset 0x%08x [handle=%d, delta=%d, read=%x, write=%x (fenced? %d)]\n",
+		kgem_debug_print(data, offset, 4, "dst offset 0x%08x [handle=%d, delta=%d, read=%x, write=%x (fenced? %d, tiling? %d)]\n",
 				 data[4],
 				 reloc->target_handle, reloc->delta,
 				 reloc->read_domains, reloc->write_domain,
-				 kgem_debug_handle_is_fenced(kgem, reloc->target_handle));
+				 kgem_debug_handle_is_fenced(kgem, reloc->target_handle),
+				 kgem_debug_handle_tiling(kgem, reloc->target_handle));
 		kgem_debug_print(data, offset, 5, "color\n");
 		return len;
 
@@ -299,22 +311,24 @@ decode_2d(struct kgem *kgem, uint32_t offset)
 				 data[3] & 0xffff, data[3] >> 16);
 		reloc = kgem_debug_get_reloc_entry(kgem, offset+4);
 		assert(reloc);
-		kgem_debug_print(data, offset, 4, "dst offset 0x%08x [handle=%d, delta=%d, read=%x, write=%x, (fenced? %d)]\n",
+		kgem_debug_print(data, offset, 4, "dst offset 0x%08x [handle=%d, delta=%d, read=%x, write=%x, (fenced? %d, tiling? %d)]\n",
 				 data[4],
 				 reloc->target_handle, reloc->delta,
 				 reloc->read_domains, reloc->write_domain,
-				 kgem_debug_handle_is_fenced(kgem, reloc->target_handle));
+				 kgem_debug_handle_is_fenced(kgem, reloc->target_handle),
+				 kgem_debug_handle_tiling(kgem, reloc->target_handle));
 		kgem_debug_print(data, offset, 5, "src (%d,%d)\n",
 				 data[5] & 0xffff, data[5] >> 16);
 		kgem_debug_print(data, offset, 6, "src pitch %d\n",
 				 (short)(data[6] & 0xffff));
 		reloc = kgem_debug_get_reloc_entry(kgem, offset+7);
 		assert(reloc);
-		kgem_debug_print(data, offset, 7, "src offset 0x%08x [handle=%d, delta=%d, read=%x, write=%x (fenced? %d)]\n",
+		kgem_debug_print(data, offset, 7, "src offset 0x%08x [handle=%d, delta=%d, read=%x, write=%x (fenced? %d, tiling? %d)]\n",
 				 data[7],
 				 reloc->target_handle, reloc->delta,
 				 reloc->read_domains, reloc->write_domain,
-				 kgem_debug_handle_is_fenced(kgem, reloc->target_handle));
+				 kgem_debug_handle_is_fenced(kgem, reloc->target_handle),
+				 kgem_debug_handle_tiling(kgem, reloc->target_handle));
 		return len;
 	}
 
diff --git a/src/sna/kgem_debug_gen3.c b/src/sna/kgem_debug_gen3.c
index 6709a8ec..17ddb6bd 100644
--- a/src/sna/kgem_debug_gen3.c
+++ b/src/sna/kgem_debug_gen3.c
@@ -1155,13 +1155,19 @@ gen3_decode_3d_1d(struct kgem *kgem, uint32_t offset)
 		for (map = 0; map <= 15; map++) {
 			if (data[1] & (1 << map)) {
 				int width, height, pitch, dword;
+				struct drm_i915_gem_relocation_entry *reloc;
+				struct kgem_bo *bo = NULL;
 				const char *tiling;
 
+				reloc = kgem_debug_get_reloc_entry(kgem, &data[i] - kgem->batch);
+				assert(reloc->target_handle);
+
 				dword = data[i];
-				kgem_debug_print(data, offset, i++, "map %d MS2 %s%s%s\n", map,
+				kgem_debug_print(data, offset, i++, "map %d MS2 %s%s%s, handle=%d\n", map,
 					  dword&(1<<31)?"untrusted surface, ":"",
 					  dword&(1<<1)?"vertical line stride enable, ":"",
-					  dword&(1<<0)?"vertical ofs enable, ":"");
+					  dword&(1<<0)?"vertical ofs enable, ":"",
+					  reloc->target_handle);
 
 				dword = data[i];
 				width = ((dword >> 10) & ((1 << 11) - 1))+1;