Re: [PATCH v7 4/8] drm/msm: Add MSM-specific DSC helper methods

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 





On 5/10/2023 11:15 PM, Marijn Suijten wrote:
On 2023-05-10 14:03:14, Jessica Zhang wrote:


On 5/9/2023 11:33 PM, Marijn Suijten wrote:
On 2023-05-09 15:06:50, Jessica Zhang wrote:
Introduce MSM-specific DSC helper methods, as some calculations are
common between DP and DSC.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx>
Signed-off-by: Jessica Zhang <quic_jesszhan@xxxxxxxxxxx>
---
   drivers/gpu/drm/msm/Makefile         |  1 +
   drivers/gpu/drm/msm/msm_dsc_helper.c | 26 ++++++++++++++
   drivers/gpu/drm/msm/msm_dsc_helper.h | 69 ++++++++++++++++++++++++++++++++++++
   3 files changed, 96 insertions(+)

diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile
index 7274c41228ed..b814fc80e2d5 100644
--- a/drivers/gpu/drm/msm/Makefile
+++ b/drivers/gpu/drm/msm/Makefile
@@ -94,6 +94,7 @@ msm-y += \
   	msm_atomic_tracepoints.o \
   	msm_debugfs.o \
   	msm_drv.o \
+	msm_dsc_helper.o \
   	msm_fb.o \
   	msm_fence.o \
   	msm_gem.o \
diff --git a/drivers/gpu/drm/msm/msm_dsc_helper.c b/drivers/gpu/drm/msm/msm_dsc_helper.c
new file mode 100644
index 000000000000..29feb3e3b5a4
--- /dev/null
+++ b/drivers/gpu/drm/msm/msm_dsc_helper.c
@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2023 Qualcomm Innovation Center, Inc. All rights reserved
+ */
+
+#include <linux/kernel.h>
+#include <drm/drm_fixed.h>
+
+#include "msm_dsc_helper.h"
+
+s64 msm_dsc_get_bytes_per_soft_slice(struct drm_dsc_config *dsc)
+{
+	return drm_fixp_from_fraction(dsc->slice_width * msm_dsc_get_bpp_int(dsc), 8);

How about using dsc->slice_chunk_size?

Hi Marijn,

Thanks for pointing this out. However, I would prefer to keep this fixed
point version of the slice_chunk_size math as the downstream DP math
also uses fixed point [1].

That looks like a totally different computation.  For this specific
value it appears drm_fixp_from_fraction does rounding whereas
DIV_ROUND_UP always rounds up when used for slice_chunk_size, but these
values are generally multiples of 8.  slice_chunk_size will also take
fractional bpp into account (even if unsupported by DPU otherwise).

If we are able to confirm that integer math also works for DP, we will
make the change to use slice_chunk_size within the DP DSC series.

Don't think the DP series alone should point that out, as it heavily
depends on the relation between slice size and bpp parameters, and
whether those end up with a fractional part or not (are you able to test
and confirm all those combinations?).  Regardless it seems more natural
to use slice_chunk_size which is used in various other ways and sent in
the PPS: it wouldn't make sense to adhere to a different slice byte
count elsewhere.


They are not totally different.

I guess what Jessica is trying to say here is we will have to do more validation with slice/bpp combinations for DP to conclude whether there is a difference in math. We can go with two approaches:

1) Either go with slice_chunk_size for now and re-validate with the monitors we have and drop this particular helper 2) Keep this particular helper for now and upon more validation if we see it's same across more combinations, drop this later.

I just have a slight preference for (2), but looks like your preference is for (1) but this is just more validation work for us which is fine I guess this time since we will revalidate DP again anyway.

I also want to note that this math has stayed the same throughout all 7
revisions. In the interest of making review more efficient, I think it
would be helpful to point out important details like this early on in
the process. That way we can address major concerns early on and keep
the number of revisions per series low.

I've already expressed to Abhinav and you that the revisions for all
these series were coming in way too hot, leaving no time for review and
discussions before the next revision hits the list.  If you want to keep
the number of revisions low, v8 shouln't have already been sent.


You had the concern for DSC 1.2 DPU series from kuogee. So to respect that we held that back the entire end of last week and early this week (~6 days) and posted today.

Now, you have the same concern with this series from jessica. Sorry, as much as i would like to get your feedback on every series, I cannot hold back every QC display developer to wait for your reviews on patch rev 1 before posting patch rev 2. They all work mostly independently. So if one reviewer say dmitry has reviewed one revision and we agreed on the comments, in the absence of another comment from another reviewer, the developer is going to post the next revision for more reviews as its more efficient to manage their time as they are in the same context. This is nothing unusual from normal upstream development.

If by the time you start reviewing its going to be revision 7 or 8, then start it as a fresh review which it is for you, folks are obviously going to question why you didnt review revisions 1-7 so far. So I cannot take sides on this.

I wish I could help more to accommodate your schedules but its hard to control who pushes when with distributed development and tell each of them to hold back.

I desire to review *and test* all these patches (which I believe is in
everyone's best interest) and have not initially been able to do so as
all these efforts come out of my free time, which is sometimes limited
of has been used to finalize the INTF TE series (which is a dependency
of these series).

It should be pretty obvious to see that this patch has not had a reply
from me on any of the previous revisions, and there are more patches
that I have not been able to comment on yet.

The alternative to that is of course not seeing enough value in my
review and testing (or ""late"" comments...) and merging it regardless
at some point, but that's not up to me to decide.

- Marijn

Thats true even for mine or other's reviews :) that if i am not able to get to some patches on time and other reviewers are happy with it they do get landed in the tree. Has happened many times and I have not complained. I just fix the issues on top of them (which again happens a lot) and move on.

So there doesnt seem to be any rule as such today that wait for this particular developer to review and validate.



[1]
https://github.com/ianmacd/gts6lwifi/blob/master/drivers/gpu/drm/msm/dp/dp_panel.c#L335


+}
+
+u32 msm_dsc_get_bytes_per_intf(struct drm_dsc_config *dsc, int intf_width)
+{
+	u32 bytes_per_soft_slice;
+	s64 bytes_per_soft_slice_fp;
+	int slice_per_intf = msm_dsc_get_slice_per_intf(dsc, intf_width);
+
+	bytes_per_soft_slice_fp = msm_dsc_get_bytes_per_soft_slice(dsc);
+	bytes_per_soft_slice = drm_fixp2int_ceil(bytes_per_soft_slice_fp);
+
+	return bytes_per_soft_slice * slice_per_intf;
+}
diff --git a/drivers/gpu/drm/msm/msm_dsc_helper.h b/drivers/gpu/drm/msm/msm_dsc_helper.h
new file mode 100644
index 000000000000..38f3651d0b79
--- /dev/null
+++ b/drivers/gpu/drm/msm/msm_dsc_helper.h
@@ -0,0 +1,69 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2023 Qualcomm Innovation Center, Inc. All rights reserved
+ */
+
+#ifndef MSM_DSC_HELPER_H_
+#define MSM_DSC_HELPER_H_
+
+#include <linux/bug.h>
+#include <linux/math.h>
+#include <drm/display/drm_dsc_helper.h>
+
+/*
+ * Helper methods for MSM specific DSC calculations that are common between timing engine,
+ * DSI, and DP.
+ */
+
+/**
+ * msm_dsc_get_bpp_int - get bits per pixel integer value

For all function docs, don't forget the trailing parenthesis after the
function name: msm_dsc_get_bpp_int()

https://www.kernel.org/doc/html/next/doc-guide/kernel-doc.html#function-documentation

Acked.


+ * @dsc: Pointer to drm dsc config struct
+ * Returns: BPP integer value
+ */
+static inline int msm_dsc_get_bpp_int(struct drm_dsc_config *dsc)
+{
+	WARN_ON_ONCE(dsc->bits_per_pixel & 0xf);
+	return dsc->bits_per_pixel >> 4;
+}
+
+/**
+ * msm_dsc_get_slice_per_intf - get number of slices per interface
+ * @dsc: Pointer to drm dsc config struct
+ * @intf_width: interface width
+ * Returns: Integer representing the slice per interface
+ */
+static inline int msm_dsc_get_slice_per_intf(struct drm_dsc_config *dsc, int intf_width)
+{
+	return DIV_ROUND_UP(intf_width, dsc->slice_width);

Looks good.

+}
+
+/**
+ * msm_dsc_get_bytes_per_line - Calculate bytes per line
+ * @dsc: Pointer to drm dsc config struct
+ * Returns: Integer value representing pclk per interface
+ *
+ * Note: This value will then be passed along to DSI and DP for some more
+ * calculations. This is because DSI and DP divide the pclk_per_intf value
+ * by different values depending on if widebus is enabled.
+ */
+static inline int msm_dsc_get_bytes_per_line(struct drm_dsc_config *dsc)
+{
+	return DIV_ROUND_UP(dsc->slice_width * dsc->slice_count * msm_dsc_get_bpp_int(dsc), 8);

dsc->slice_chunk_size * dsc->slice_count?

Acked.


+}
+
+/**
+ * msm_dsc_get_bytes_per_soft_slice - get size of each soft slice for dsc

Explain to the reader what a "soft" slice is?

A soft slice is a slice defined in software as opposed to "hard slices"
that are defined by hardware.

Since the slice-related variables in drm_dsc_config hold information
related to soft slices and there is no separate variable for hard
slices, I'll change this name to *_get_bytes_per_slice instead.

Thanks,

Jessica Zhang


- Marijn

+ * @dsc: Pointer to drm dsc config struct
+ * Returns: s31.32 fixed point value representing bytes per soft slice
+ */
+s64 msm_dsc_get_bytes_per_soft_slice(struct drm_dsc_config *dsc);
+
+/**
+ * msm_dsc_get_bytes_per_intf - get total bytes per interface
+ * @dsc: Pointer to drm dsc config struct
+ * @intf_width: interface width
+ * Returns: u32 value representing bytes per interface
+ */
+u32 msm_dsc_get_bytes_per_intf(struct drm_dsc_config *dsc, int intf_width);
+
+#endif /* MSM_DSC_HELPER_H_ */

--
2.40.1




[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux