lkml.org 
[lkml]   [2019]   [Mar]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH v6 3/7] gpu: ipu-v3: ipu-ic: Fully describe colorspace conversions
From
Date


On 3/8/19 3:46 AM, Philipp Zabel wrote:
> On Thu, 2019-03-07 at 15:33 -0800, Steve Longerbeam wrote:
>> Only providing the input and output RGB/YUV space to the IC task init
>> functions is not sufficient. To fully characterize a colorspace
>> conversion, the colorspace (chromaticities), Y'CbCr encoding standard,
>> and quantization also need to be specified.
>>
>> Define a 'struct ipu_ic_colorspace' that includes all the above, and pass
>> the input and output ipu_ic_colorspace to the IC task init functions.
>>
>> This allows to actually enforce the fact that the IC:
>>
>> - can only encode to/from YUV full range (follow-up patch will remove
>> this restriction).
>> - can only encode to/from RGB full range.
>> - can only encode using BT.601 standard (follow-up patch will add
>> Rec.709 encoding support).
>> - cannot convert colorspaces from input to output, the
>> input and output colorspace chromaticities must be the same.
>>
>> The determination of the CSC coefficients based on the input/output
>> colorspace parameters are moved to a new function calc_csc_coeffs(),
>> called by init_csc().
>>
>> Signed-off-by: Steve Longerbeam <slongerbeam@gmail.com>
>> ---
>> drivers/gpu/ipu-v3/ipu-ic.c | 136 +++++++++++++-------
>> drivers/gpu/ipu-v3/ipu-image-convert.c | 27 ++--
>> drivers/staging/media/imx/imx-ic-prpencvf.c | 22 +++-
>> include/video/imx-ipu-v3.h | 37 +++++-
>> 4 files changed, 154 insertions(+), 68 deletions(-)
>>
>> diff --git a/drivers/gpu/ipu-v3/ipu-ic.c b/drivers/gpu/ipu-v3/ipu-ic.c
>> index b63a2826b629..c4048c921801 100644
>> --- a/drivers/gpu/ipu-v3/ipu-ic.c
>> +++ b/drivers/gpu/ipu-v3/ipu-ic.c
>> @@ -146,8 +146,10 @@ struct ipu_ic {
>> const struct ic_task_regoffs *reg;
>> const struct ic_task_bitfields *bit;
>>
>> - enum ipu_color_space in_cs, g_in_cs;
>> - enum ipu_color_space out_cs;
>> + struct ipu_ic_colorspace in_cs;
>> + struct ipu_ic_colorspace g_in_cs;
>> + struct ipu_ic_colorspace out_cs;
>> +
>> bool graphics;
>> bool rotation;
>> bool in_use;
>> @@ -235,42 +237,83 @@ static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = {
>> .scale = 2,
>> };
>>
>> +static int calc_csc_coeffs(struct ipu_ic_priv *priv,
>> + struct ic_encode_coeff *coeff_out,
>> + const struct ipu_ic_colorspace *in,
>> + const struct ipu_ic_colorspace *out)
>> +{
>> + bool inverse_encode;
>> +
>> + if (in->colorspace != out->colorspace) {
>> + dev_err(priv->ipu->dev, "Cannot convert colorspaces\n");
>> + return -ENOTSUPP;
>> + }
> I don't think this is useful enough to warrant having the colorspace
> field in ipu_ic_colorspace. Let the caller make sure of this, same as
> for xfer_func.

Ok, for xfer_func it is implicit that the gamma function must be the
same for input and output, so I agree it might as well be implicit for
chromaticities too.


>
>> + if (out->enc != V4L2_YCBCR_ENC_601) {
>> + dev_err(priv->ipu->dev, "Only BT.601 encoding supported\n");
>> + return -ENOTSUPP;
>> + }
> This is only important if out->cs is IPUV3_COLORSPACE_YUV, right? If the
> output is RGB this field shouldn't matter.

It matters for encoding YUV to RGB, or the inverse RGB to YUV. The
encoding standard doesn't matter only if no encoding/inverse encoding is
requested (YUV to YUV or RGB to RGB).

>
>> +
>> + if ((in->cs == IPUV3_COLORSPACE_YUV &&
>> + in->quant != V4L2_QUANTIZATION_FULL_RANGE) ||
>> + (out->cs == IPUV3_COLORSPACE_YUV &&
>> + out->quant != V4L2_QUANTIZATION_FULL_RANGE)) {
>> + dev_err(priv->ipu->dev, "Limited range YUV not supported\n");
>> + return -ENOTSUPP;
>> + }
>> +
>> + if ((in->cs == IPUV3_COLORSPACE_RGB &&
>> + in->quant != V4L2_QUANTIZATION_FULL_RANGE) ||
>> + (out->cs == IPUV3_COLORSPACE_RGB &&
>> + out->quant != V4L2_QUANTIZATION_FULL_RANGE)) {
>> + dev_err(priv->ipu->dev, "Limited range RGB not supported\n");
>> + return -ENOTSUPP;
>> + }
>> +
>> + if (in->cs == out->cs) {
>> + *coeff_out = ic_encode_identity;
>> +
>> + return 0;
>> + }
>> +
>> + inverse_encode = (in->cs == IPUV3_COLORSPACE_YUV);
> What does inverse_encode mean in this context?

It means YUV to RGB. At this point in the function it is determined that
encoding or inverse encoding is requested.

>
>> +
>> + *coeff_out = inverse_encode ?
>> + ic_encode_ycbcr2rgb_601 : ic_encode_rgb2ycbcr_601;
>> +
>> + return 0;
>> +}
>> +
>> static int init_csc(struct ipu_ic *ic,
>> - enum ipu_color_space inf,
>> - enum ipu_color_space outf,
>> + const struct ipu_ic_colorspace *in,
>> + const struct ipu_ic_colorspace *out,
>> int csc_index)
>> {
>> struct ipu_ic_priv *priv = ic->priv;
>> - const struct ic_encode_coeff *coeff;
>> + struct ic_encode_coeff coeff;
> I understand this is a preparation for patch 5, but on its own this
> introduces an unnecessary copy.

True, I'll try to remove the copy in this patch.

>
>> u32 __iomem *base;
>> const u16 (*c)[3];
>> const u16 *a;
>> u32 param;
>> + int ret;
>> +
>> + ret = calc_csc_coeffs(priv, &coeff, in, out);
>> + if (ret)
>> + return ret;
>>
>> base = (u32 __iomem *)
>> (priv->tpmem_base + ic->reg->tpmem_csc[csc_index]);
>>
>> - if (inf == IPUV3_COLORSPACE_YUV && outf == IPUV3_COLORSPACE_RGB)
>> - coeff = &ic_encode_ycbcr2rgb_601;
>> - else if (inf == IPUV3_COLORSPACE_RGB && outf == IPUV3_COLORSPACE_YUV)
>> - coeff = &ic_encode_rgb2ycbcr_601;
>> - else if (inf == IPUV3_COLORSPACE_RGB && outf == IPUV3_COLORSPACE_RGB)
>> - coeff = &ic_encode_identity;
>> - else {
>> - dev_err(priv->ipu->dev, "Unsupported color space conversion\n");
>> - return -EINVAL;
>> - }
>> -
>> /* Cast to unsigned */
>> - c = (const u16 (*)[3])coeff->coeff;
>> - a = (const u16 *)coeff->offset;
>> + c = (const u16 (*)[3])coeff.coeff;
>> + a = (const u16 *)coeff.offset;
>>
>> param = ((a[0] & 0x1f) << 27) | ((c[0][0] & 0x1ff) << 18) |
>> ((c[1][1] & 0x1ff) << 9) | (c[2][2] & 0x1ff);
>> writel(param, base++);
>>
>> - param = ((a[0] & 0x1fe0) >> 5) | (coeff->scale << 8) |
>> - (coeff->sat << 10);
>> + param = ((a[0] & 0x1fe0) >> 5) | (coeff.scale << 8) |
>> + (coeff.sat << 10);
>> writel(param, base++);
>>
>> param = ((a[1] & 0x1f) << 27) | ((c[0][1] & 0x1ff) << 18) |
>> @@ -357,14 +400,14 @@ void ipu_ic_task_enable(struct ipu_ic *ic)
>> if (ic->rotation)
>> ic_conf |= ic->bit->ic_conf_rot_en;
>>
>> - if (ic->in_cs != ic->out_cs)
>> + if (ic->in_cs.cs != ic->out_cs.cs)
>> ic_conf |= ic->bit->ic_conf_csc1_en;
>>
>> if (ic->graphics) {
>> ic_conf |= ic->bit->ic_conf_cmb_en;
>> ic_conf |= ic->bit->ic_conf_csc1_en;
>>
>> - if (ic->g_in_cs != ic->out_cs)
>> + if (ic->g_in_cs.cs != ic->out_cs.cs)
>> ic_conf |= ic->bit->ic_conf_csc2_en;
>> }
>>
>> @@ -399,7 +442,7 @@ void ipu_ic_task_disable(struct ipu_ic *ic)
>> EXPORT_SYMBOL_GPL(ipu_ic_task_disable);
>>
>> int ipu_ic_task_graphics_init(struct ipu_ic *ic,
>> - enum ipu_color_space in_g_cs,
>> + const struct ipu_ic_colorspace *g_in_cs,
> What made you decide not to expose the task parameter structure?
>
> I was hoping we could eventually move the V4L2 colorimetry settings to
> conversion matrix translation into imx-media.

Sure, I'm fine with that. I'll move the task parameter struct to
imx-ipu-v3.h.

>
> Btw, do you have any plans for using IC composition?
> ipu_ic_task_graphics_init() is currently unused...

No plans for IC composition, I've only been keeping the function
up-to-date. I've never even tested this, it might not even work. Should
it be removed?

>
>> bool galpha_en, u32 galpha,
>> bool colorkey_en, u32 colorkey)
>> {
>> @@ -416,20 +459,25 @@ int ipu_ic_task_graphics_init(struct ipu_ic *ic,
>> ic_conf = ipu_ic_read(ic, IC_CONF);
>>
>> if (!(ic_conf & ic->bit->ic_conf_csc1_en)) {
>> + struct ipu_ic_colorspace rgb_cs;
>> +
>> + ipu_ic_fill_colorspace(&rgb_cs,
>> + V4L2_COLORSPACE_SRGB,
>> + V4L2_YCBCR_ENC_601,
>> + V4L2_QUANTIZATION_FULL_RANGE,
>> + IPUV3_COLORSPACE_RGB);
>> +
>> /* need transparent CSC1 conversion */
>> - ret = init_csc(ic, IPUV3_COLORSPACE_RGB,
>> - IPUV3_COLORSPACE_RGB, 0);
>> + ret = init_csc(ic, &rgb_cs, &rgb_cs, 0);
>> if (ret)
>> goto unlock;
>> }
>>
>> - ic->g_in_cs = in_g_cs;
>> + ic->g_in_cs = *g_in_cs;
>>
>> - if (ic->g_in_cs != ic->out_cs) {
>> - ret = init_csc(ic, ic->g_in_cs, ic->out_cs, 1);
>> - if (ret)
>> - goto unlock;
>> - }
>> + ret = init_csc(ic, &ic->g_in_cs, &ic->out_cs, 1);
>> + if (ret)
>> + goto unlock;
> I had to look twice, but this is fine. If ic->g_in_cs == ic->out_cs,
> ipu_ic_task_enable() won't enable CSC2 in IC_CONF, and these TPMEM
> values will be ignored.
>
>>
>> if (galpha_en) {
>> ic_conf |= IC_CONF_IC_GLB_LOC_A;
>> @@ -456,10 +504,10 @@ int ipu_ic_task_graphics_init(struct ipu_ic *ic,
>> EXPORT_SYMBOL_GPL(ipu_ic_task_graphics_init);
>>
>> int ipu_ic_task_init_rsc(struct ipu_ic *ic,
>> + const struct ipu_ic_colorspace *in_cs,
>> + const struct ipu_ic_colorspace *out_cs,
>> int in_width, int in_height,
>> int out_width, int out_height,
>> - enum ipu_color_space in_cs,
>> - enum ipu_color_space out_cs,
>> u32 rsc)
>> {
>> struct ipu_ic_priv *priv = ic->priv;
>> @@ -491,28 +539,24 @@ int ipu_ic_task_init_rsc(struct ipu_ic *ic,
>> ipu_ic_write(ic, rsc, ic->reg->rsc);
>>
>> /* Setup color space conversion */
>> - ic->in_cs = in_cs;
>> - ic->out_cs = out_cs;
>> + ic->in_cs = *in_cs;
>> + ic->out_cs = *out_cs;
>>
>> - if (ic->in_cs != ic->out_cs) {
>> - ret = init_csc(ic, ic->in_cs, ic->out_cs, 0);
>> - if (ret)
>> - goto unlock;
>> - }
>> + ret = init_csc(ic, &ic->in_cs, &ic->out_cs, 0);
> Same as above for CSC1.
>
>> -unlock:
>> spin_unlock_irqrestore(&priv->lock, flags);
>> return ret;
>> }
>>
>> int ipu_ic_task_init(struct ipu_ic *ic,
>> + const struct ipu_ic_colorspace *in_cs,
>> + const struct ipu_ic_colorspace *out_cs,
>> int in_width, int in_height,
>> - int out_width, int out_height,
>> - enum ipu_color_space in_cs,
>> - enum ipu_color_space out_cs)
>> + int out_width, int out_height)
>> {
>> - return ipu_ic_task_init_rsc(ic, in_width, in_height, out_width,
>> - out_height, in_cs, out_cs, 0);
>> + return ipu_ic_task_init_rsc(ic, in_cs, out_cs,
>> + in_width, in_height,
>> + out_width, out_height, 0);
>> }
>> EXPORT_SYMBOL_GPL(ipu_ic_task_init);
>>
>> diff --git a/drivers/gpu/ipu-v3/ipu-image-convert.c b/drivers/gpu/ipu-v3/ipu-image-convert.c
>> index 13103ab86050..ccbc8f4d95d7 100644
>> --- a/drivers/gpu/ipu-v3/ipu-image-convert.c
>> +++ b/drivers/gpu/ipu-v3/ipu-image-convert.c
>> @@ -1317,7 +1317,7 @@ static int convert_start(struct ipu_image_convert_run *run, unsigned int tile)
>> struct ipu_image_convert_priv *priv = chan->priv;
>> struct ipu_image_convert_image *s_image = &ctx->in;
>> struct ipu_image_convert_image *d_image = &ctx->out;
>> - enum ipu_color_space src_cs, dest_cs;
>> + struct ipu_ic_colorspace src_cs, dest_cs;
>> unsigned int dst_tile = ctx->out_tile_map[tile];
>> unsigned int dest_width, dest_height;
>> unsigned int col, row;
>> @@ -1327,8 +1327,16 @@ static int convert_start(struct ipu_image_convert_run *run, unsigned int tile)
>> dev_dbg(priv->ipu->dev, "%s: task %u: starting ctx %p run %p tile %u -> %u\n",
>> __func__, chan->ic_task, ctx, run, tile, dst_tile);
>>
>> - src_cs = ipu_pixelformat_to_colorspace(s_image->fmt->fourcc);
>> - dest_cs = ipu_pixelformat_to_colorspace(d_image->fmt->fourcc);
>> + ipu_ic_fill_colorspace(&src_cs,
>> + s_image->base.pix.colorspace,
>> + s_image->base.pix.ycbcr_enc,
>> + s_image->base.pix.quantization,
>> + ipu_pixelformat_to_colorspace(s_image->fmt->fourcc));
>> + ipu_ic_fill_colorspace(&dest_cs,
>> + d_image->base.pix.colorspace,
>> + d_image->base.pix.ycbcr_enc,
>> + d_image->base.pix.quantization,
>> + ipu_pixelformat_to_colorspace(d_image->fmt->fourcc));
> If ipu_ic_task_init(_rsc) could be passed the task parameter structure,
> it could be calculated once in ipu_image_convert_prepare and stored in
> ipu_image_convert_ctx for repeated use.

Yes, I'll add this for v7.

Steve

\
 
 \ /
  Last update: 2019-03-09 02:17    [W:0.460 / U:0.228 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site