From: Ville Syrjälä ville.syrjala@linux.intel.com
Having both modes can be beneficial for video playback cases. If you can match the video framerate exactly, and the audio and video clocks come from the same source, you should be able to avoid dropped/repeated frames without expensive operations such as resampling the audio to match video output rate.
Rather than add both variants based on the CEA extension short video descriptors in do_cea_modes(), add only one variant there. Once all the EDID has been fully probed, do a loop over the entire probed mode list, during which we add the other variants for all modes that match CEA modes. This allows us to match modes that didn't come via the CEA short video descriptors. For example one Samsung TV here doesn't have the 640x480-60 mode as a SVD, but instead it's specified via a detailed timing descriptor.
Signed-off-by: Ville Syrjälä ville.syrjala@linux.intel.com --- A few people requested this. Originally I was a bit opposed to it, but when I thought about it a bit more I figured if the audio and video clocks come from the same source (or happen to be close enough w/o significant drift), this could provide a better A/V sync w/o resampling tricks.
drivers/gpu/drm/drm_edid.c | 102 ++++++++++++++++++++++++++++++++++++++------- 1 file changed, 88 insertions(+), 14 deletions(-)
diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c index 9e62bbe..e8d17ee 100644 --- a/drivers/gpu/drm/drm_edid.c +++ b/drivers/gpu/drm/drm_edid.c @@ -2321,6 +2321,31 @@ u8 *drm_find_cea_extension(struct edid *edid) } EXPORT_SYMBOL(drm_find_cea_extension);
+/* + * Calculate the alternate clock for the CEA mode + * (60Hz vs. 59.94Hz etc.) + */ +static unsigned int +cea_mode_alternate_clock(const struct drm_display_mode *cea_mode) +{ + unsigned int clock = cea_mode->clock; + + if (cea_mode->vrefresh % 6 != 0) + return clock; + + /* + * edid_cea_modes contains the 59.94Hz + * variant for 240 and 480 line modes, + * and the 60Hz variant otherwise. + */ + if (cea_mode->vdisplay == 240 || cea_mode->vdisplay == 480) + clock = clock * 1001 / 1000; + else + clock = DIV_ROUND_UP(clock * 1000, 1001); + + return clock; +} + /** * drm_match_cea_mode - look for a CEA mode matching given mode * @to_match: display mode @@ -2339,21 +2364,9 @@ u8 drm_match_cea_mode(const struct drm_display_mode *to_match) const struct drm_display_mode *cea_mode = &edid_cea_modes[mode]; unsigned int clock1, clock2;
- clock1 = clock2 = cea_mode->clock; - /* Check both 60Hz and 59.94Hz */ - if (cea_mode->vrefresh % 6 == 0) { - /* - * edid_cea_modes contains the 59.94Hz - * variant for 240 and 480 line modes, - * and the 60Hz variant otherwise. - */ - if (cea_mode->vdisplay == 240 || - cea_mode->vdisplay == 480) - clock1 = clock1 * 1001 / 1000; - else - clock2 = DIV_ROUND_UP(clock2 * 1000, 1001); - } + clock1 = cea_mode->clock; + clock2 = cea_mode_alternate_clock(cea_mode);
if ((KHZ2PICOS(to_match->clock) == KHZ2PICOS(clock1) || KHZ2PICOS(to_match->clock) == KHZ2PICOS(clock2)) && @@ -2364,6 +2377,66 @@ u8 drm_match_cea_mode(const struct drm_display_mode *to_match) } EXPORT_SYMBOL(drm_match_cea_mode);
+static int +add_alternate_cea_modes(struct drm_connector *connector, struct edid *edid) +{ + struct drm_device *dev = connector->dev; + struct drm_display_mode *mode, *tmp; + LIST_HEAD(list); + int modes = 0; + + /* Don't add CEA modes if the CEA extension block is missing */ + if (!drm_find_cea_extension(edid)) + return 0; + + /* + * Go through all probed modes and create a new mode + * with the alternate clock for certain CEA modes. + */ + list_for_each_entry(mode, &connector->probed_modes, head) { + const struct drm_display_mode *cea_mode; + struct drm_display_mode *newmode; + u8 cea_mode_idx = drm_match_cea_mode(mode) - 1; + unsigned int clock1, clock2; + + if (cea_mode_idx >= ARRAY_SIZE(edid_cea_modes)) + continue; + + cea_mode = &edid_cea_modes[cea_mode_idx]; + + clock1 = cea_mode->clock; + clock2 = cea_mode_alternate_clock(cea_mode); + + if (clock1 == clock2) + continue; + + if (mode->clock != clock1 && mode->clock != clock2) + continue; + + newmode = drm_mode_duplicate(dev, cea_mode); + if (!newmode) + continue; + + /* + * The current mode could be either variant. Make + * sure to pick the "other" clock for the new mode. + */ + if (mode->clock != clock1) + newmode->clock = clock1; + else + newmode->clock = clock2; + + list_add_tail(&newmode->head, &list); + } + + list_for_each_entry_safe(mode, tmp, &list, head) { + list_del(&mode->head); + drm_mode_probed_add(connector, mode); + modes++; + } + + return modes; +}
static int do_cea_modes (struct drm_connector *connector, u8 *db, u8 len) @@ -2946,6 +3019,7 @@ int drm_add_edid_modes(struct drm_connector *connector, struct edid *edid) if (edid->features & DRM_EDID_FEATURE_DEFAULT_GTF) num_modes += add_inferred_modes(connector, edid); num_modes += add_cea_modes(connector, edid); + num_modes += add_alternate_cea_modes(connector, edid);
if (quirks & (EDID_QUIRK_PREFER_LARGE_60 | EDID_QUIRK_PREFER_LARGE_75)) edid_fixup_preferred(connector, quirks);
On Fri, May 31, 2013 at 03:23:41PM +0300, ville.syrjala@linux.intel.com wrote:
From: Ville Syrjälä ville.syrjala@linux.intel.com
Having both modes can be beneficial for video playback cases. If you can match the video framerate exactly, and the audio and video clocks come from the same source, you should be able to avoid dropped/repeated frames without expensive operations such as resampling the audio to match video output rate.
Rather than add both variants based on the CEA extension short video descriptors in do_cea_modes(), add only one variant there. Once all the EDID has been fully probed, do a loop over the entire probed mode list, during which we add the other variants for all modes that match CEA modes. This allows us to match modes that didn't come via the CEA short video descriptors. For example one Samsung TV here doesn't have the 640x480-60 mode as a SVD, but instead it's specified via a detailed timing descriptor.
Signed-off-by: Ville Syrjälä ville.syrjala@linux.intel.com
A few people requested this. Originally I was a bit opposed to it, but when I thought about it a bit more I figured if the audio and video clocks come from the same source (or happen to be close enough w/o significant drift), this could provide a better A/V sync w/o resampling tricks.
Yeah, I think this should be useful for a bunch of people. I've recently chatted with a few xbmc folks on #irc and one thing they've requested is mode fine-tuning. For DP we should have plenty of precision, but for HDMI we'd need to (slightly) frob the vtotal ever so often to compensate. With some runtime-tuning a la npt we should have perfect a/v sync without any audio resampling.
Anyway, on the patch:
Reviewed-by: Daniel Vetter daniel.vetter@ffwll.ch
drivers/gpu/drm/drm_edid.c | 102 ++++++++++++++++++++++++++++++++++++++------- 1 file changed, 88 insertions(+), 14 deletions(-)
diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c index 9e62bbe..e8d17ee 100644 --- a/drivers/gpu/drm/drm_edid.c +++ b/drivers/gpu/drm/drm_edid.c @@ -2321,6 +2321,31 @@ u8 *drm_find_cea_extension(struct edid *edid) } EXPORT_SYMBOL(drm_find_cea_extension);
+/*
- Calculate the alternate clock for the CEA mode
- (60Hz vs. 59.94Hz etc.)
- */
+static unsigned int +cea_mode_alternate_clock(const struct drm_display_mode *cea_mode) +{
- unsigned int clock = cea_mode->clock;
- if (cea_mode->vrefresh % 6 != 0)
return clock;- /*
* edid_cea_modes contains the 59.94Hz* variant for 240 and 480 line modes,* and the 60Hz variant otherwise.*/- if (cea_mode->vdisplay == 240 || cea_mode->vdisplay == 480)
clock = clock * 1001 / 1000;- else
clock = DIV_ROUND_UP(clock * 1000, 1001);- return clock;
+}
/**
- drm_match_cea_mode - look for a CEA mode matching given mode
- @to_match: display mode
@@ -2339,21 +2364,9 @@ u8 drm_match_cea_mode(const struct drm_display_mode *to_match) const struct drm_display_mode *cea_mode = &edid_cea_modes[mode]; unsigned int clock1, clock2;
clock1 = clock2 = cea_mode->clock;- /* Check both 60Hz and 59.94Hz */
if (cea_mode->vrefresh % 6 == 0) {/** edid_cea_modes contains the 59.94Hz* variant for 240 and 480 line modes,* and the 60Hz variant otherwise.*/if (cea_mode->vdisplay == 240 ||cea_mode->vdisplay == 480)clock1 = clock1 * 1001 / 1000;elseclock2 = DIV_ROUND_UP(clock2 * 1000, 1001);}
clock1 = cea_mode->clock;clock2 = cea_mode_alternate_clock(cea_mode);if ((KHZ2PICOS(to_match->clock) == KHZ2PICOS(clock1) || KHZ2PICOS(to_match->clock) == KHZ2PICOS(clock2)) &&
@@ -2364,6 +2377,66 @@ u8 drm_match_cea_mode(const struct drm_display_mode *to_match) } EXPORT_SYMBOL(drm_match_cea_mode);
+static int +add_alternate_cea_modes(struct drm_connector *connector, struct edid *edid) +{
- struct drm_device *dev = connector->dev;
- struct drm_display_mode *mode, *tmp;
- LIST_HEAD(list);
- int modes = 0;
- /* Don't add CEA modes if the CEA extension block is missing */
- if (!drm_find_cea_extension(edid))
return 0;- /*
* Go through all probed modes and create a new mode* with the alternate clock for certain CEA modes.*/- list_for_each_entry(mode, &connector->probed_modes, head) {
const struct drm_display_mode *cea_mode;struct drm_display_mode *newmode;u8 cea_mode_idx = drm_match_cea_mode(mode) - 1;unsigned int clock1, clock2;if (cea_mode_idx >= ARRAY_SIZE(edid_cea_modes))continue;cea_mode = &edid_cea_modes[cea_mode_idx];clock1 = cea_mode->clock;clock2 = cea_mode_alternate_clock(cea_mode);if (clock1 == clock2)continue;if (mode->clock != clock1 && mode->clock != clock2)continue;newmode = drm_mode_duplicate(dev, cea_mode);if (!newmode)continue;/** The current mode could be either variant. Make* sure to pick the "other" clock for the new mode.*/if (mode->clock != clock1)newmode->clock = clock1;elsenewmode->clock = clock2;list_add_tail(&newmode->head, &list);- }
- list_for_each_entry_safe(mode, tmp, &list, head) {
list_del(&mode->head);drm_mode_probed_add(connector, mode);modes++;- }
- return modes;
+}
static int do_cea_modes (struct drm_connector *connector, u8 *db, u8 len) @@ -2946,6 +3019,7 @@ int drm_add_edid_modes(struct drm_connector *connector, struct edid *edid) if (edid->features & DRM_EDID_FEATURE_DEFAULT_GTF) num_modes += add_inferred_modes(connector, edid); num_modes += add_cea_modes(connector, edid);
num_modes += add_alternate_cea_modes(connector, edid);
if (quirks & (EDID_QUIRK_PREFER_LARGE_60 | EDID_QUIRK_PREFER_LARGE_75)) edid_fixup_preferred(connector, quirks);
-- 1.8.1.5
dri-devel mailing list dri-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/dri-devel
Daniel Vetter wrote:
On Fri, May 31, 2013 at 03:23:41PM +0300, ville.syrjala@linux.intel.com wrote:
From: Ville Syrjälä ville.syrjala@linux.intel.com
Having both modes can be beneficial for video playback cases. If you can match the video framerate exactly, and the audio and video clocks come from the same source, you should be able to avoid dropped/repeated frames without expensive operations such as resampling the audio to match video output rate.
Rather than add both variants based on the CEA extension short video descriptors in do_cea_modes(), add only one variant there. Once all the EDID has been fully probed, do a loop over the entire probed mode list, during which we add the other variants for all modes that match CEA modes. This allows us to match modes that didn't come via the CEA short video descriptors. For example one Samsung TV here doesn't have the 640x480-60 mode as a SVD, but instead it's specified via a detailed timing descriptor.
Signed-off-by: Ville Syrjälä ville.syrjala@linux.intel.com
A few people requested this. Originally I was a bit opposed to it, but when I thought about it a bit more I figured if the audio and video clocks come from the same source (or happen to be close enough w/o significant drift), this could provide a better A/V sync w/o resampling tricks.
Yeah, I think this should be useful for a bunch of people. I've recently chatted with a few xbmc folks on #irc and one thing they've requested is mode fine-tuning. For DP we should have plenty of precision, but for HDMI we'd need to (slightly) frob the vtotal ever so often to compensate. With some runtime-tuning a la npt we should have perfect a/v sync without any audio resampling.
Apologies, for jumping in here, but assuming I haven't missed anything you've done already: do you have any plans for supporting CEA interlaced modes?
That's assuming they actually need any more support - I only know that they are slightly wrong for me testing with radeon + my TV - are they tested and known to work / not work on intel hw?
The symptom I get is not instantly obvious - the second half of the bottom line gets rendered and repeated as the top line.
For me the reason that proper support could in theory be useful, is that were there in future a way to sync up properly, my TV will do a nice "free" deinterlace. Even on a fast quad core PC doing one as good on cpu is not possible for HD 50i.
On Fri, May 31, 2013 at 8:56 PM, Andy Furniss adf.lists@gmail.com wrote:
Daniel Vetter wrote:
On Fri, May 31, 2013 at 03:23:41PM +0300, ville.syrjala@linux.intel.com wrote:
From: Ville Syrjälä ville.syrjala@linux.intel.com
Having both modes can be beneficial for video playback cases. If you can match the video framerate exactly, and the audio and video clocks come from the same source, you should be able to avoid dropped/repeated frames without expensive operations such as resampling the audio to match video output rate.
Rather than add both variants based on the CEA extension short video descriptors in do_cea_modes(), add only one variant there. Once all the EDID has been fully probed, do a loop over the entire probed mode list, during which we add the other variants for all modes that match CEA modes. This allows us to match modes that didn't come via the CEA short video descriptors. For example one Samsung TV here doesn't have the 640x480-60 mode as a SVD, but instead it's specified via a detailed timing descriptor.
Signed-off-by: Ville Syrjälä ville.syrjala@linux.intel.com
A few people requested this. Originally I was a bit opposed to it, but when I thought about it a bit more I figured if the audio and video clocks come from the same source (or happen to be close enough w/o significant drift), this could provide a better A/V sync w/o resampling tricks.
Yeah, I think this should be useful for a bunch of people. I've recently chatted with a few xbmc folks on #irc and one thing they've requested is mode fine-tuning. For DP we should have plenty of precision, but for HDMI we'd need to (slightly) frob the vtotal ever so often to compensate. With some runtime-tuning a la npt we should have perfect a/v sync without any audio resampling.
Apologies, for jumping in here, but assuming I haven't missed anything you've done already: do you have any plans for supporting CEA interlaced modes?
That's assuming they actually need any more support - I only know that they are slightly wrong for me testing with radeon + my TV - are they tested and known to work / not work on intel hw?
The symptom I get is not instantly obvious - the second half of the bottom line gets rendered and repeated as the top line.
For me the reason that proper support could in theory be useful, is that were there in future a way to sync up properly, my TV will do a nice "free" deinterlace. Even on a fast quad core PC doing one as good on cpu is not possible for HD 50i.
Interlaced modes are supported on all intel platforms (except i8xx, hw can't do it there) and we have full support for CEA modes. -Daniel -- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
Daniel Vetter wrote:
Interlaced modes are supported on all intel platforms (except i8xx, hw can't do it there) and we have full support for CEA modes. -Daniel
OK, thanks, I guess it's an AMD driver issue rather than a more generic drm thing then.
On Sat, Jun 1, 2013 at 5:13 AM, Andy Furniss adf.lists@gmail.com wrote:
Daniel Vetter wrote:
Interlaced modes are supported on all intel platforms (except i8xx, hw can't do it there) and we have full support for CEA modes. -Daniel
OK, thanks, I guess it's an AMD driver issue rather than a more generic drm thing then.
You might try removing this chunk of code in radeon_atom_mode_fixup() in atombios_encoders.c
/* hw bug */ if ((mode->flags & DRM_MODE_FLAG_INTERLACE) && (mode->crtc_vsync_start < (mode->crtc_vdisplay + 2))) adjusted_mode->crtc_vsync_start = adjusted_mode->crtc_vdisplay + 2;
Alex
Alex Deucher wrote:
On Sat, Jun 1, 2013 at 5:13 AM, Andy Furniss adf.lists@gmail.com wrote:
Daniel Vetter wrote:
Interlaced modes are supported on all intel platforms (except i8xx, hw can't do it there) and we have full support for CEA modes. -Daniel
OK, thanks, I guess it's an AMD driver issue rather than a more generic drm thing then.
You might try removing this chunk of code in radeon_atom_mode_fixup() in atombios_encoders.c
/* hw bug */ if ((mode->flags & DRM_MODE_FLAG_INTERLACE) && (mode->crtc_vsync_start < (mode->crtc_vdisplay + 2))) adjusted_mode->crtc_vsync_start =adjusted_mode->crtc_vdisplay + 2;
Thanks, but it doesn't make any difference.
To slightly correct what I said earlier when describing what I see -
at 1920x1080i the right quarter of the bottom line is rendering repeated x4 as the top line.
ville.syrjala@linux.intel.com wrote:
From: Ville Syrjälä ville.syrjala@linux.intel.com
Having both modes can be beneficial for video playback cases. If you can match the video framerate exactly, and the audio and video clocks come from the same source, you should be able to avoid dropped/repeated frames without expensive operations such as resampling the audio to match video output rate.
Rather than add both variants based on the CEA extension short video descriptors in do_cea_modes(), add only one variant there. Once all the EDID has been fully probed, do a loop over the entire probed mode list, during which we add the other variants for all modes that match CEA modes. This allows us to match modes that didn't come via the CEA short video descriptors. For example one Samsung TV here doesn't have the 640x480-60 mode as a SVD, but instead it's specified via a detailed timing descriptor.
Signed-off-by: Ville Syrjälä ville.syrjala@linux.intel.com
A few people requested this. Originally I was a bit opposed to it, but when I thought about it a bit more I figured if the audio and video clocks come from the same source (or happen to be close enough w/o significant drift), this could provide a better A/V sync w/o resampling tricks.
I see this has gone in now, one thing I notice is that xorg/apps/xrandr only prints Hz to 1dp so you can't see which mode is which for the 24p and 30i cases.
Maybe someone reading has commit access for xorg?
On Tue, Jun 18, 2013 at 11:10:30AM +0100, Andy Furniss wrote:
ville.syrjala@linux.intel.com wrote:
From: Ville Syrjälä ville.syrjala@linux.intel.com
Having both modes can be beneficial for video playback cases. If you can match the video framerate exactly, and the audio and video clocks come from the same source, you should be able to avoid dropped/repeated frames without expensive operations such as resampling the audio to match video output rate.
Rather than add both variants based on the CEA extension short video descriptors in do_cea_modes(), add only one variant there. Once all the EDID has been fully probed, do a loop over the entire probed mode list, during which we add the other variants for all modes that match CEA modes. This allows us to match modes that didn't come via the CEA short video descriptors. For example one Samsung TV here doesn't have the 640x480-60 mode as a SVD, but instead it's specified via a detailed timing descriptor.
Signed-off-by: Ville Syrjälä ville.syrjala@linux.intel.com
A few people requested this. Originally I was a bit opposed to it, but when I thought about it a bit more I figured if the audio and video clocks come from the same source (or happen to be close enough w/o significant drift), this could provide a better A/V sync w/o resampling tricks.
I see this has gone in now, one thing I notice is that xorg/apps/xrandr only prints Hz to 1dp so you can't see which mode is which for the 24p and 30i cases.
Maybe someone reading has commit access for xorg?
Not sure if you noticed but I posted some relevant xrandr patches to xorg-devel. Unfortunately I got no response, and I've been too lazy to figure out who I need to pester.
Ville Syrjälä wrote:
On Tue, Jun 18, 2013 at 11:10:30AM +0100, Andy Furniss wrote:
ville.syrjala@linux.intel.com wrote:
From: Ville Syrjälä ville.syrjala@linux.intel.com
Having both modes can be beneficial for video playback cases. If you can match the video framerate exactly, and the audio and video clocks come from the same source, you should be able to avoid dropped/repeated frames without expensive operations such as resampling the audio to match video output rate.
Rather than add both variants based on the CEA extension short video descriptors in do_cea_modes(), add only one variant there. Once all the EDID has been fully probed, do a loop over the entire probed mode list, during which we add the other variants for all modes that match CEA modes. This allows us to match modes that didn't come via the CEA short video descriptors. For example one Samsung TV here doesn't have the 640x480-60 mode as a SVD, but instead it's specified via a detailed timing descriptor.
Signed-off-by: Ville Syrjälä ville.syrjala@linux.intel.com
A few people requested this. Originally I was a bit opposed to it, but when I thought about it a bit more I figured if the audio and video clocks come from the same source (or happen to be close enough w/o significant drift), this could provide a better A/V sync w/o resampling tricks.
I see this has gone in now, one thing I notice is that xorg/apps/xrandr only prints Hz to 1dp so you can't see which mode is which for the 24p and 30i cases.
Maybe someone reading has commit access for xorg?
Not sure if you noticed but I posted some relevant xrandr patches to xorg-devel. Unfortunately I got no response, and I've been too lazy to figure out who I need to pester.
Ahh, sorry I didn't see those, I just has a look at the current code to check it was still the same.
dri-devel@lists.freedesktop.org
-
Alex Deucher -
Andy Furniss -
Daniel Vetter -
Ville Syrjälä -
ville.syrjala＠linux.intel.com