| Message ID | 20180918093421.12930-1-p.zabel@pengutronix.de (mailing list archive) |
|---|---|
| Headers | show |
| Series | i.MX media mem2mem scaler | expand |
Hi Philipp, I finished some fairly comprehensive testing on this series, scaling to and from the following common resolutions: 640x480, 720x480, 720x576, 800x480, 800x600, 1024x768, 1152x720, 1280x720, 1280x768, 1280x800, 1280x1024, 1920x1080, 1920x1200, 2560x1440, 2560x1600, 2592x1944 and within each scaling conversion, I tested converting to and from all the pixel formats supported by the mem2mem device: 422p, nv12, rgb4, yu12, bgr3, nv16, rgbp, yuyv, bgr4, rgb3, uyvy, yv12 What I found is that most of these conversions are working, producing no mis-aligned tile buffer addresses, except for one exception: converting to or from a packed 24-bit pixel format (rgb3 or bgr3). For example, try 640x480.rgb3 -> 1920x1080.yu12. I found the cause of that, I added the following patch to my fork git@github.com:slongerbeam/mediatree.git, branch imx-mem2mem.3: "gpu: ipu-v3: image-convert: Fix tile_left_align()" Feel free to squash it with "gpu: ipu-v3: image-convert: select optimal seam positions" I've also added a few other patches to my branch: "gpu: ipu-v3: Add chroma plane offset overrides to ipu_cpmem_set_image()" which fixes a false warning in ipu_cpmem_set_yuv_planar_full(). Also added: "gpu: ipu-v3: image-convert: Prevent race between run and unprepare" "gpu: ipu-v3: image-convert: Only wait for abort completion if active run" "gpu: ipu-v3: image-convert: Allow reentrancy into abort" "gpu: ipu-v3: image-convert: Remove need_abort flag" "gpu: ipu-v3: image-convert: Catch unaligned tile offsets" With the fix to tile_left_align(), all conversions for the above scaling resolutions are not producing mis-aligned tile buffers anymore, for all pixel formats. But one last thing. Conversions to and from YV12 are producing images with wrong colors, it looks like the .uv_swapped boolean needs to be checked additionally somewhere. Any ideas? Steve On 09/18/2018 02:34 AM, Philipp Zabel wrote: > Hi, > > this is the third version of the i.MX mem2mem scaler series. > > The driver patch has been moved to the beginning, as Steve suggested. > I've added his warning patch to catch alignment bugs to the series, > seam position selection has been fixed for more corner cases, the > alignment restriction relaxation patches have been merged into one > patch, and support for tiling with three rows or columns has been added > to avoid unnecessary overhead. > > Changes since v2: > - Rely on ipu_image_convert_adjust() in mem2mem_try_fmt() for format > adjustments. This makes the mem2mem driver mostly a V4L2 mem2mem API > wrapper around the IPU image converter, and independent of the > internal image converter implementation. > - Remove the source and destination buffers on error in device_run(). > Otherwise the conversion is re-attempted apparently over and over > again (with WARN() backtraces). > - Allow subscribing to control changes. > - Fix seam position selection for more corner cases: > - Switch width/height properly and align tile top left positions to 8x8 > IRT block size when rotating. > - Align input width to input burst length in case the scaling step > flips horizontally. > - Fix bottom edge calculation. > > Changes since v1: > - Fix inverted allow_overshoot logic > - Correctly switch horizontal / vertical tile alignment when > determining seam positions with the 90° rotator active. > - Fix SPDX-License-Identifier and remove superfluous license > text. > - Fix uninitialized walign in try_fmt > > Previous cover letter: > > we have image conversion code for scaling and colorspace conversion in > the IPUv3 base driver for a while. Since the IC hardware can only write > up to 1024x1024 pixel buffers, it scales to larger output buffers by > splitting the input and output frame into similarly sized tiles. > > This causes the issue that the bilinear interpolation resets at the tile > boundary: instead of smoothly interpolating across the seam, there is a > jump in the input sample position that is very apparent for high > upscaling factors. This can be avoided by slightly changing the scaling > coefficients to let the left/top tiles overshoot their input sampling > into the first pixel / line of their right / bottom neighbors. The error > can be further reduced by letting tiles be differently sized and by > selecting seam positions that minimize the input sampling position error > at tile boundaries. > This is complicated by different DMA start address, burst size, and > rotator block size alignment requirements, depending on the input and > output pixel formats, and the fact that flipping happens in different > places depending on the rotation. > > This series implements optimal seam position selection and seam hiding > with per-tile resizing coefficients and adds a scaling mem2mem device > to the imx-media driver. > > regards > Philipp > > Philipp Zabel (15): > media: imx: add mem2mem device > gpu: ipu-v3: ipu-ic: allow to manually set resize coefficients > gpu: ipu-v3: image-convert: prepare for per-tile configuration > gpu: ipu-v3: image-convert: calculate per-tile resize coefficients > gpu: ipu-v3: image-convert: reconfigure IC per tile > gpu: ipu-v3: image-convert: store tile top/left position > gpu: ipu-v3: image-convert: calculate tile dimensions and offsets > outside fill_image > gpu: ipu-v3: image-convert: move tile alignment helpers > gpu: ipu-v3: image-convert: select optimal seam positions > gpu: ipu-v3: image-convert: fix debug output for varying tile sizes > gpu: ipu-v3: image-convert: relax alignment restrictions > gpu: ipu-v3: image-convert: fix bytesperline adjustment > gpu: ipu-v3: image-convert: add some ASCII art to the exposition > gpu: ipu-v3: image-convert: disable double buffering if necessary > gpu: ipu-v3: image-convert: allow three rows or columns > > Steve Longerbeam (1): > gpu: ipu-cpmem: add WARN_ON_ONCE() for unaligned dma buffers > > drivers/gpu/ipu-v3/ipu-cpmem.c | 6 + > drivers/gpu/ipu-v3/ipu-ic.c | 52 +- > drivers/gpu/ipu-v3/ipu-image-convert.c | 919 +++++++++++++++--- > drivers/staging/media/imx/Kconfig | 1 + > drivers/staging/media/imx/Makefile | 1 + > drivers/staging/media/imx/imx-media-dev.c | 11 + > drivers/staging/media/imx/imx-media-mem2mem.c | 873 +++++++++++++++++ > drivers/staging/media/imx/imx-media.h | 10 + > include/video/imx-ipu-v3.h | 6 + > 9 files changed, 1727 insertions(+), 152 deletions(-) > create mode 100644 drivers/staging/media/imx/imx-media-mem2mem.c >
Hi Philipp, On 10/12/18 5:29 PM, Steve Longerbeam wrote: > <snip> > > But one last thing. Conversions to and from YV12 are producing images > with wrong colors, it looks like the .uv_swapped boolean needs to be > checked > additionally somewhere. Any ideas? Sorry, this was my fault. I fixed this in "gpu: ipu-v3: Add chroma plane offset overrides to ipu_cpmem_set_image()" in my fork git@github.com:slongerbeam/mediatree.git, branch imx-mem2mem.3. Steve
Hi Steve, On Wed, 2018-10-17 at 16:46 -0700, Steve Longerbeam wrote: > Hi Philipp, > > On 10/12/18 5:29 PM, Steve Longerbeam wrote: > > <snip> > > > > But one last thing. Conversions to and from YV12 are producing images > > with wrong colors, it looks like the .uv_swapped boolean needs to be > > checked > > additionally somewhere. Any ideas? > > > Sorry, this was my fault. I fixed this in > > "gpu: ipu-v3: Add chroma plane offset overrides to ipu_cpmem_set_image()" > > in my fork git@github.com:slongerbeam/mediatree.git, branch imx-mem2mem.3. > > Steve Thanks a lot for testing, fixes, and integration. Basically I've just resubmitted that branch as v4. After this round I'll pick up all non-controversial ipu-v3 / image- convert patches. regards Philipp
Hi, this is the third version of the i.MX mem2mem scaler series. The driver patch has been moved to the beginning, as Steve suggested. I've added his warning patch to catch alignment bugs to the series, seam position selection has been fixed for more corner cases, the alignment restriction relaxation patches have been merged into one patch, and support for tiling with three rows or columns has been added to avoid unnecessary overhead. Changes since v2: - Rely on ipu_image_convert_adjust() in mem2mem_try_fmt() for format adjustments. This makes the mem2mem driver mostly a V4L2 mem2mem API wrapper around the IPU image converter, and independent of the internal image converter implementation. - Remove the source and destination buffers on error in device_run(). Otherwise the conversion is re-attempted apparently over and over again (with WARN() backtraces). - Allow subscribing to control changes. - Fix seam position selection for more corner cases: - Switch width/height properly and align tile top left positions to 8x8 IRT block size when rotating. - Align input width to input burst length in case the scaling step flips horizontally. - Fix bottom edge calculation. Changes since v1: - Fix inverted allow_overshoot logic - Correctly switch horizontal / vertical tile alignment when determining seam positions with the 90° rotator active. - Fix SPDX-License-Identifier and remove superfluous license text. - Fix uninitialized walign in try_fmt Previous cover letter: we have image conversion code for scaling and colorspace conversion in the IPUv3 base driver for a while. Since the IC hardware can only write up to 1024x1024 pixel buffers, it scales to larger output buffers by splitting the input and output frame into similarly sized tiles. This causes the issue that the bilinear interpolation resets at the tile boundary: instead of smoothly interpolating across the seam, there is a jump in the input sample position that is very apparent for high upscaling factors. This can be avoided by slightly changing the scaling coefficients to let the left/top tiles overshoot their input sampling into the first pixel / line of their right / bottom neighbors. The error can be further reduced by letting tiles be differently sized and by selecting seam positions that minimize the input sampling position error at tile boundaries. This is complicated by different DMA start address, burst size, and rotator block size alignment requirements, depending on the input and output pixel formats, and the fact that flipping happens in different places depending on the rotation. This series implements optimal seam position selection and seam hiding with per-tile resizing coefficients and adds a scaling mem2mem device to the imx-media driver. regards Philipp Philipp Zabel (15): media: imx: add mem2mem device gpu: ipu-v3: ipu-ic: allow to manually set resize coefficients gpu: ipu-v3: image-convert: prepare for per-tile configuration gpu: ipu-v3: image-convert: calculate per-tile resize coefficients gpu: ipu-v3: image-convert: reconfigure IC per tile gpu: ipu-v3: image-convert: store tile top/left position gpu: ipu-v3: image-convert: calculate tile dimensions and offsets outside fill_image gpu: ipu-v3: image-convert: move tile alignment helpers gpu: ipu-v3: image-convert: select optimal seam positions gpu: ipu-v3: image-convert: fix debug output for varying tile sizes gpu: ipu-v3: image-convert: relax alignment restrictions gpu: ipu-v3: image-convert: fix bytesperline adjustment gpu: ipu-v3: image-convert: add some ASCII art to the exposition gpu: ipu-v3: image-convert: disable double buffering if necessary gpu: ipu-v3: image-convert: allow three rows or columns Steve Longerbeam (1): gpu: ipu-cpmem: add WARN_ON_ONCE() for unaligned dma buffers drivers/gpu/ipu-v3/ipu-cpmem.c | 6 + drivers/gpu/ipu-v3/ipu-ic.c | 52 +- drivers/gpu/ipu-v3/ipu-image-convert.c | 919 +++++++++++++++--- drivers/staging/media/imx/Kconfig | 1 + drivers/staging/media/imx/Makefile | 1 + drivers/staging/media/imx/imx-media-dev.c | 11 + drivers/staging/media/imx/imx-media-mem2mem.c | 873 +++++++++++++++++ drivers/staging/media/imx/imx-media.h | 10 + include/video/imx-ipu-v3.h | 6 + 9 files changed, 1727 insertions(+), 152 deletions(-) create mode 100644 drivers/staging/media/imx/imx-media-mem2mem.c