[libcamera-devel] [PATCH] ipa: ipu3: Replace ipa::ipu3::algorithms::Ipu3AwbCell
Jean-Michel Hautbois
jeanmichel.hautbois at ideasonboard.com
Thu Sep 2 18:08:00 CEST 2021
Hi Kieran,
On 02/09/2021 17:46, Kieran Bingham wrote:
> On 02/09/2021 14:29, Jean-Michel Hautbois wrote:
>> This patch was proposed in:
>> https://lore.kernel.org/linux-media/20210831185140.77400-1-jeanmichel.hautbois@ideasonboard.com/
>
> That's not accurate, and it dives off to an external reference without
> explaining why. /This/ patch is not proposed for linux-media.
>
>
> Try to introduce your patches, and think about your commit message in
> distinct sections such as:
>
> - Describe the issue
> - Describe any context
> - Describe solution
>
>
> - the issue
> """
> The intel-ipu3.h public interface from the kernel does not define how to
> parse the statistics for a cell. This had to be identified by a process
> of reverse engineering, and later identifying the structures from
> https://chromium.googlesource.com/chromiumos/platform/arc-camera/+/refs/heads/master/hal/intel/include/ia_imaging/awb_public.h
> leading to our custom definition of struct Ipu3AwbCell.
>
> """
>
>
> - context
> """
> To improve the kernel interface, a proposal has been made to the
> linux-kernel [0] to incorporate the memory layout for each cell into the
> intel-ipu3 header directly.
>
> [0]
> https://lore.kernel.org/linux-media/20210831185140.77400-1-jeanmichel.hautbois@ideasonboard.com/
> """
>
>
> - solution / What this patch does
> """
> Update our local copy of the intel-ipu3.h to match the proposal and
> change the AGC and AWB algorithms to reference that structure directly,
> allowing us to remove the deprecated custom Ipu3AwbCell definition.
> """
>
> The only issue is that the proposal to linux-media has not yet been
> reviewed or approved/integrated, so until then this patch would cause us
> to diverge from the kernel version of the header and would require
> special attention.
>
> That may not be an issue in this case, as we can track the progress
> upstream - but normally we would expect to keep our linux headers
> matching the kernel unmodified.
>
Thanks a lot for this really better description, I sent a v2 with it.
I know we would normally keep the headers matching the kernel, but it
could take some time before we can do that, can we afford it ?
>
>
>
>> It introduces the AWB metadata layout as a structure in the kernel
>> header, to make it clear for userspace applications. As intel-ipu3.h now
>> has a defined AWB layout, use that one instead of the
>> ipa::ipu3::algorithms::Ipu3AwbCell.
>>
>> Signed-off-by: Jean-Michel Hautbois <jeanmichel.hautbois at ideasonboard.com>
>> ---
>> include/linux/intel-ipu3.h | 38 +++++++++++++++++++++++++--------
>> src/ipa/ipu3/algorithms/agc.cpp | 7 +++---
>> src/ipa/ipu3/algorithms/awb.cpp | 10 ++++-----
>> src/ipa/ipu3/algorithms/awb.h | 10 ---------
>> 4 files changed, 38 insertions(+), 27 deletions(-)
>>
>> diff --git a/include/linux/intel-ipu3.h b/include/linux/intel-ipu3.h
>> index ee0e6d0e..0633e04e 100644
>> --- a/include/linux/intel-ipu3.h
>> +++ b/include/linux/intel-ipu3.h
>> @@ -59,20 +59,40 @@ struct ipu3_uapi_grid_config {
>> __u16 y_end;
>> } __attribute__((packed));
>>
>> +/**
>> + * struct ipu3_uapi_awb_set_item - Memory layout for each cell in AWB
>> + *
>> + * @Gr_avg: Green average for red lines in the cell.
>> + * @R_avg: Red average in the cell.
>> + * @B_avg: Blue average in the cell.
>> + * @Gb_avg: Green average for blue lines in the cell.
>> + * @sat_ratio: Saturation ratio in the cell.
>> + * @padding0: Unused byte for padding.
>> + * @padding1: Unused byte for padding.
>> + * @padding2: Unused byte for padding.
>> + */
>> +struct ipu3_uapi_awb_set_item {
>> + unsigned char Gr_avg;
>> + unsigned char R_avg;
>> + unsigned char B_avg;
>> + unsigned char Gb_avg;
>> + unsigned char sat_ratio;
>> + unsigned char padding0;
>> + unsigned char padding1;
>> + unsigned char padding2;
>> +} __attribute__((packed));
>> +
>> /*
>> * The grid based data is divided into "slices" called set, each slice of setX
>> * refers to ipu3_uapi_grid_config width * height_per_slice.
>> */
>> #define IPU3_UAPI_AWB_MAX_SETS 60
>> -/* Based on grid size 80 * 60 and cell size 16 x 16 */
>> -#define IPU3_UAPI_AWB_SET_SIZE 1280
>> -#define IPU3_UAPI_AWB_MD_ITEM_SIZE 8
>> -#define IPU3_UAPI_AWB_SPARE_FOR_BUBBLES \
>> - (IPU3_UAPI_MAX_BUBBLE_SIZE * IPU3_UAPI_MAX_STRIPES * \
>> - IPU3_UAPI_AWB_MD_ITEM_SIZE)
>> +#define AWB_PUBLIC_NUM_OF_ITEMS_IN_SET 160
>> +/* Based on max grid height + Spare for bubbles */
>> +#define AWB_PUBLIC_NUM_OF_SETS_IN_BUFFER IPU3_UAPI_AWB_MAX_SETS + \
>> + (IPU3_UAPI_MAX_BUBBLE_SIZE * IPU3_UAPI_MAX_STRIPES)
>> #define IPU3_UAPI_AWB_MAX_BUFFER_SIZE \
>> - (IPU3_UAPI_AWB_MAX_SETS * \
>> - (IPU3_UAPI_AWB_SET_SIZE + IPU3_UAPI_AWB_SPARE_FOR_BUBBLES))
>> + AWB_PUBLIC_NUM_OF_SETS_IN_BUFFER * AWB_PUBLIC_NUM_OF_ITEMS_IN_SET
>>
>>
>> /**
>> @@ -82,7 +102,7 @@ struct ipu3_uapi_grid_config {
>> * the average values for each color channel.
>> */
>> struct ipu3_uapi_awb_raw_buffer {
>> - __u8 meta_data[IPU3_UAPI_AWB_MAX_BUFFER_SIZE]
>> + struct ipu3_uapi_awb_set_item meta_data[IPU3_UAPI_AWB_MAX_BUFFER_SIZE]
>> __attribute__((aligned(32)));
>> } __attribute__((packed));
>>
>> diff --git a/src/ipa/ipu3/algorithms/agc.cpp b/src/ipa/ipu3/algorithms/agc.cpp
>> index 5ff50f4a..36f648be 100644
>> --- a/src/ipa/ipu3/algorithms/agc.cpp
>> +++ b/src/ipa/ipu3/algorithms/agc.cpp
>> @@ -96,9 +96,10 @@ void Agc::processBrightness(const ipu3_uapi_stats_3a *stats,
>> * We observed a bit shift which makes the value 160 to be 32 in the stats grid.
>> * Use the one passed at init time.
>> */
>> - if (stats->awb_raw_buffer.meta_data[i + 4 + j * grid.width] == 0) {
>> - uint8_t Gr = stats->awb_raw_buffer.meta_data[i + 0 + j * grid.width];
>> - uint8_t Gb = stats->awb_raw_buffer.meta_data[i + 3 + j * grid.width];
>> + const ipu3_uapi_awb_set_item *currentCell = &stats->awb_raw_buffer.meta_data[i + 4 + j * grid.width];
>> + if (currentCell->sat_ratio == 0) {
>> + uint8_t Gr = currentCell->Gr_avg;
>> + uint8_t Gb = currentCell->Gb_avg;
>> hist[(Gr + Gb) / 2]++;
>> count++;
>> }
>> diff --git a/src/ipa/ipu3/algorithms/awb.cpp b/src/ipa/ipu3/algorithms/awb.cpp
>> index ea567f96..db0a22d3 100644
>> --- a/src/ipa/ipu3/algorithms/awb.cpp
>> +++ b/src/ipa/ipu3/algorithms/awb.cpp
>> @@ -207,14 +207,14 @@ void Awb::generateAwbStats(const ipu3_uapi_stats_3a *stats,
>> cellPosition *= 8;
>>
>> /* Cast the initial IPU3 structure to simplify the reading */
>> - Ipu3AwbCell *currentCell = reinterpret_cast<Ipu3AwbCell *>(const_cast<uint8_t *>(&stats->awb_raw_buffer.meta_data[cellPosition]));
>> - if (currentCell->satRatio == 0) {
>> + const ipu3_uapi_awb_set_item *currentCell = &stats->awb_raw_buffer.meta_data[cellPosition];
>> + if (currentCell->sat_ratio == 0) {
>> /* The cell is not saturated, use the current cell */
>> awbStats_[awbRegionPosition].counted++;
>> - uint32_t greenValue = currentCell->greenRedAvg + currentCell->greenBlueAvg;
>> + uint32_t greenValue = currentCell->Gr_avg + currentCell->Gb_avg;
>> awbStats_[awbRegionPosition].sum.green += greenValue / 2;
>> - awbStats_[awbRegionPosition].sum.red += currentCell->redAvg;
>> - awbStats_[awbRegionPosition].sum.blue += currentCell->blueAvg;
>> + awbStats_[awbRegionPosition].sum.red += currentCell->R_avg;
>> + awbStats_[awbRegionPosition].sum.blue += currentCell->B_avg;
>> }
>> }
>> }
>> diff --git a/src/ipa/ipu3/algorithms/awb.h b/src/ipa/ipu3/algorithms/awb.h
>> index 23f3e872..a9320b4c 100644
>> --- a/src/ipa/ipu3/algorithms/awb.h
>> +++ b/src/ipa/ipu3/algorithms/awb.h
>> @@ -23,16 +23,6 @@ namespace ipa::ipu3::algorithms {
>> static constexpr uint32_t kAwbStatsSizeX = 16;
>> static constexpr uint32_t kAwbStatsSizeY = 12;
>>
>> -/* \todo Move the cell layout into intel-ipu3.h kernel header */
>> -struct Ipu3AwbCell {
>> - unsigned char greenRedAvg;
>> - unsigned char redAvg;
>> - unsigned char blueAvg;
>> - unsigned char greenBlueAvg;
>> - unsigned char satRatio;
>> - unsigned char padding[3];
>> -} __attribute__((packed));
>> -
>> struct Accumulator {
>> unsigned int counted;
>> unsigned int total;
>>
More information about the libcamera-devel
mailing list