Hi Ryan,
On 5/31/2019 9:32 AM, Ryan Chen wrote:
On 5/24/19 6:17 PM, Jae Hyun Yoo wrote:
AST2500 silicon revision A1 and A2 have a silicon bug which causes
extremly long capturing time on specific resolutions (1680 width).
To fix the bug, this commit adjusts the capturing window register
setting to 1728 if detected width is 1680. The compression window
register setting will be kept as the original width so output result
will be the same.
This is a bit curious, why 1728 in particular? And what is the
behavior of the VE when the capture window is larger than the actual
source resolution?
For an example, if resolution is 1680x1050, capturing operation takes very long time because VE has the silicon bug. So this patch adjusts the 'Capture Window' register slightly larger than >1680 to avoid the issue. As a result, source buffer will copy 1728x1050 frames from the original screen buffer but the image is still has valid information.
As the next step in compression phase, it will set the 'Compression Window' register as '1680x1050' so it will compress using the original image resolution which is a cropped image from the >'1728x1050' source buffer.
You can compare results using these shell commands in Ubuntu GUI desktop.
$ xrandr --newmode "1680x1050_60.00" 146.25 1680 1784 1960 1240 1050
1053 1059 1089 -hsync +vsync
$ xrandr --addmode VGA-1 1680x1050_60.00 $ xrandr --output VGA-1 --mode 1680x1050_60.00
I'm also curious about why that is 1728. Actually, this workaround was provided from the chip vendor, Aspeed, and they use this in their SDK code too. Let's check it to Ryan.
Hi Ryan,
Can you please explain why that is 1728 in particular.
Thanks,
Jae
That have two factor, one is data too huge cause the memory bandwidth is too busy.
The other is vga resolution width is not align to 32 or 64 pixels (32bpp or 16bpp).
Those will cause engine read latency time too long, maybe engine will hange.
Thanks a lot for the detail. I'll add it as an additional comment in
code.
Thanks,
Jae
Ryan
Thanks,
Eddie
Signed-off-by: Jae Hyun Yoo <jae.hyun.yoo@xxxxxxxxxxxxxxx>
---
v1 -> v2:
New.
drivers/media/platform/aspeed-video.c | 26
+++++++++++++++++++-------
1 file changed, 19 insertions(+), 7 deletions(-)
diff --git a/drivers/media/platform/aspeed-video.c
b/drivers/media/platform/aspeed-video.c
index b05b073b63bc..f93989f532d6 100644
--- a/drivers/media/platform/aspeed-video.c
+++ b/drivers/media/platform/aspeed-video.c
@@ -824,8 +824,27 @@ static void aspeed_video_set_resolution(struct
aspeed_video *video)
struct v4l2_bt_timings *act = &video->active_timings;
unsigned int size = act->width * act->height;
+ /* Set capture/compression frame sizes */
aspeed_video_calc_compressed_size(video, size);
+ if (video->active_timings.width == 1680) {
+ /*
+ * This is a workaround to fix a silicon bug on A1 and A2
+ * revisions. Since it doesn't break capturing operation on
+A0
+ * revision, use it for all revisions without checking the
+ * revision ID.
+ */
+ aspeed_video_write(video, VE_CAP_WINDOW,
+ 1728 << 16 | act->height);
+ size += (1728 - 1680) * video->active_timings.height;
+ } else {
+ aspeed_video_write(video, VE_CAP_WINDOW,
+ act->width << 16 | act->height);
+ }
+ aspeed_video_write(video, VE_COMP_WINDOW,
+ act->width << 16 | act->height);
+ aspeed_video_write(video, VE_SRC_SCANLINE_OFFSET, act->width *
+4);
+
/* Don't use direct mode below 1024 x 768 (irqs don't fire) */
if (size < DIRECT_FETCH_THRESHOLD) {
aspeed_video_write(video, VE_TGS_0, @@ -842,13 +861,6 @@
static void aspeed_video_set_resolution(struct
aspeed_video *video)
aspeed_video_update(video, VE_CTRL, 0,
VE_CTRL_DIRECT_FETCH);
}
- /* Set capture/compression frame sizes */
- aspeed_video_write(video, VE_CAP_WINDOW,
- act->width << 16 | act->height);
- aspeed_video_write(video, VE_COMP_WINDOW,
- act->width << 16 | act->height);
- aspeed_video_write(video, VE_SRC_SCANLINE_OFFSET, act->width *
4);
-
size *= 4;
if (size == video->srcs[0].size / 2) {
