media: imx: Add support for BT.709 encoding
Subsystems (8)
Steve Longerbeam
maintainerv6 cover media: imx: Add support for BT.709 encoding
This patchset adds support for the BT.709 encoding and inverse encoding matrices to the ipu_ic task init functions. The imx-media driver can now support both BT.601 and BT.709 encoding.
History:
v6:
- tweak some of the coefficients slightly, they were not getting rounded correctly.
- move the introduction of calc_csc_coeffs() to an earlier patch for easier patch readability.
v5:
- the hard-coded encode coefficients now convert only between full-range quantization. A new function is added to transform the coefficients to limited-range at input or output.
- add a bug fix patch for saturation bit in TPMEM register.
- add a patch to fully describe input and output colorspace to the IC task init functions.
- add imx_media_try_colorimetry(), called at all sink/source pad try_fmt.
v4:
- fix a compile error in init_csc(), reported by Tim Harvey.
v3:
- fix some inconsistent From: and Signed-off-by:'s. No functional changes.
v2:
- rename ic_csc_rgb2rgb matrix to ic_csc_identity.
- only return "Unsupported YCbCr encoding" error if inf != outf, since if inf == outf, the identity matrix can be used. Reported by Tim Harvey.
- move ic_route check above default colorimetry checks, and fill default colorspace for ic_route, otherwise it's not possible to set BT.709 encoding for ic routes.
Steve Longerbeam (7):
gpu: ipu-v3: ipu-ic: Fix saturation bit offset in TPMEM
gpu: ipu-v3: ipu-ic: Fix BT.601 coefficients
gpu: ipu-v3: ipu-ic: Fully describe colorspace conversions
gpu: ipu-v3: ipu-ic: Add support for Rec.709 encoding
gpu: ipu-v3: ipu-ic: Add support for limited range encoding
media: imx: Try colorimetry at both sink and source pads
media: imx: Allow BT.709 encoding for IC routes
drivers/gpu/ipu-v3/ipu-ic.c | 487 +++++++++++++++++--- drivers/gpu/ipu-v3/ipu-image-convert.c | 27 +- drivers/staging/media/imx/imx-ic-prp.c | 6 +- drivers/staging/media/imx/imx-ic-prpencvf.c | 30 +- drivers/staging/media/imx/imx-media-csi.c | 19 +- drivers/staging/media/imx/imx-media-utils.c | 70 +-- drivers/staging/media/imx/imx-media-vdic.c | 5 +- drivers/staging/media/imx/imx-media.h | 5 +- drivers/staging/media/imx/imx7-media-csi.c | 8 +- include/video/imx-ipu-v3.h | 37 +- 10 files changed, 547 insertions(+), 147 deletions(-)
Signature · 1 line
2.17.1
Steve Longerbeam
maintainerPATCH v6 1/7 gpu: ipu-v3: ipu-ic: Fix saturation bit offset in TPMEM
The saturation bit was being set at bit 9 in the second 32-bit word of the TPMEM CSC. This isn't correct, the saturation bit is bit 42, which is bit 10 of the second word.
--- drivers/gpu/ipu-v3/ipu-ic.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/ipu-v3/ipu-ic.c +1 -1
@@ -257,7 +257,7 @@ static int init_csc(struct ipu_ic *ic,writel (param, base++); param = ((a[0 ] &0x1fe0 ) >>5 ) | (params->scale <<8 ) | - (params->sat <<9 ); + (params->sat <<10 );writel (param, base++); param = ((a[1] & 0x1f) << 27) | ((c[0][1] & 0x1ff) << 18) |
Signature · 1 line
2.17.1
Re: [PATCH v6 1/7] gpu: ipu-v3: ipu-ic: Fix saturation bit offset in TPMEM
Quoted 17 lines · Steve Longerbeam
The saturation bit was being set at bit 9 in the second 32-bit word of the TPMEM CSC. This isn't correct, the saturation bit is bit 42, which is bit 10 of the second word.
Fixes 1aa8ea0d2bd5d ("gpu: ipu-v3: Add Image Converter unit")Signed-off-by Steve Longerbeam <[email protected]>--- drivers/gpu/ipu-v3/ipu-ic.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/gpu/ipu-v3/ipu-ic.c +0
-0
drivers/gpu/ipu-v3/ipu-ic.c static int init_csc(struct ipu_ic *ic, · line 263
Quoted 8 lines
drivers/gpu/ipu-v3/ipu-ic.c +1
-1@@ -257,7 +257,7 @@ static int init_csc(struct ipu_ic *ic,writel (param, base++); param = ((a[0 ] &0x1fe0 ) >>5 ) | (params->scale <<8 ) | - (params->sat <<9 ); + (params->sat <<10 );writel (param, base++); param = ((a[1] & 0x1f) << 27) | ((c[0][1] & 0x1ff) << 18) |
regards
Philipp
Steve Longerbeam
maintainerPATCH v6 2/7 gpu: ipu-v3: ipu-ic: Fix BT.601 coefficients
In reply to Steve LongerbeamThe ycbcr2rgb and inverse rgb2ycbcr tables define the BT.601 Y'CbCr encoding coefficients.
The rgb2ycbcr table specifically describes the BT.601 encoding from full range RGB to full range YUV. Add table comments to make this more clear.
The ycbcr2rgb inverse table describes encoding YUV limited range to RGB full range. To be consistent with the rgb2ycbcr table, convert this to YUV full range to RGB full range, and adjust/expand on the comments.
The ic_csc_rgb2rgb table is just an identity matrix, so rename to ic_encode_identity.
--- drivers/gpu/ipu-v3/ipu-ic.c | 61 ++++++++++++++++++++++--------------- 1 file changed, 37 insertions(+), 24 deletions(-)
drivers/gpu/ipu-v3/ipu-ic.c +37 -24
@@ -175,7 +175,7 @@ static inline void ipu_ic_write(struct ipu_ic *ic, u32 value, unsigned offset)writel (value, ic->priv->base + offset); } -struct ic_csc_params { +struct ic_encode_coeff {s16 coeff[3 ][3 ];/* signed 9-bit integer coefficients */ s16 offset[3 ];/* signed 11+2-bit fixed point offset */ u8 scale:2 ;/* scale coefficients * 2^(scale-1) */
@@ -183,13 +183,15 @@ struct ic_csc_params { };/* -* Y = R * .299 + G * .587 + B * .114; -* U = R * -.169 + G * -.332 + B * .500 + 128.; -* V = R * .500 + G * -.419 + B * -.0813 + 128.; +* BT.601 encoding from RGB full range to YUV full range: +* +* Y = .2990 * R + .5870 * G + .1140 * B +* U = -.1687 * R - .3313 * G + .5000 * B + 128 +* V = .5000 * R - .4187 * G - .0813 * B + 128 */ -static const struct ic_csc_params ic_csc_rgb2ycbcr = { +static const struct ic_encode_coeff ic_encode_rgb2ycbcr_601 = { .coeff = { - {77 ,150 ,29 }, + {77 ,150 ,29 }, {469 ,427 ,128 }, {128 ,405 ,491 }, },
@@ -197,8 +199,11 @@ static const struct ic_csc_params ic_csc_rgb2ycbcr = { .scale = 1, }; -/* transparent RGB->RGB matrix for graphics combining */ -static const struct ic_csc_params ic_csc_rgb2rgb = { +/* +* identity matrix, used for transparent RGB->RGB graphics +* combining. +*/ +static const struct ic_encode_coeff ic_encode_identity = { .coeff = { {128 ,0 ,0 }, {0 ,128 ,0 },
@@ -208,17 +213,25 @@ static const struct ic_csc_params ic_csc_rgb2rgb = { };/* -* R = (1.164 * (Y - 16)) + (1.596 * (Cr - 128)); -* G = (1.164 * (Y - 16)) - (0.392 * (Cb - 128)) - (0.813 * (Cr - 128)); -* B = (1.164 * (Y - 16)) + (2.017 * (Cb - 128); +* Inverse BT.601 encoding from YUV full range to RGB full range: +* +* R = 1. * Y + 0 * (Cb - 128) + 1.4020 * (Cr - 128) +* G = 1. * Y - .3442 * (Cb - 128) - 0.7142 * (Cr - 128) +* B = 1. * Y + 1.7720 * (Cb - 128) + 0 * (Cr - 128) +* +* equivalently (factoring out the offsets): +* +* R = 1. * Y + 0 * Cb + 1.4020 * Cr - 179.456 +* G = 1. * Y - .3442 * Cb - 0.7142 * Cr + 135.475 +* B = 1. * Y + 1.7720 * Cb + 0 * Cr - 226.816 */ -static const struct ic_csc_params ic_csc_ycbcr2rgb = { +static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = { .coeff = { - {149 ,0 ,204 }, - {149 ,462 ,408 }, - {149 ,255 ,0 }, + {128 ,0 ,179 }, + {128 ,468 ,421 }, + {128 ,227 ,0 }, }, - .offset = {-446 ,266 ,-554 }, + .offset = {-359 ,271 ,-454 }, .scale =2 , };
@@ -228,7 +241,7 @@ static int init_csc(struct ipu_ic *ic, int csc_index) {struct ipu_ic_priv *priv = ic->priv; -const struct ic_csc_params *params; +const struct ic_encode_coeff *coeff;u32 __iomem *base;const u16 (*c)[3 ];const u16 *a;
@@ -238,26 +251,26 @@ static int init_csc(struct ipu_ic *ic, (priv->tpmem_base + ic->reg->tpmem_csc[csc_index]);if (inf ==IPUV3_COLORSPACE_YUV && outf ==IPUV3_COLORSPACE_RGB ) - params = &ic_csc_ycbcr2rgb; + coeff = &ic_encode_ycbcr2rgb_601;else if (inf ==IPUV3_COLORSPACE_RGB && outf ==IPUV3_COLORSPACE_YUV ) - params = &ic_csc_rgb2ycbcr; + coeff = &ic_encode_rgb2ycbcr_601;else if (inf ==IPUV3_COLORSPACE_RGB && outf ==IPUV3_COLORSPACE_RGB ) - params = &ic_csc_rgb2rgb; + coeff = &ic_encode_identity;else {dev_err (priv->ipu->dev,"Unsupported color space conversion\n" );return -EINVAL ; }/* Cast to unsigned */ - c = (const u16 (*)[3 ])params->coeff; - a = (const u16 *)params->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 ) | (params->scale <<8 ) | - (params->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) |
Signature · 1 line
2.17.1
Philipp Zabel
maintainerRe: [PATCH v6 2/7] gpu: ipu-v3: ipu-ic: Fix BT.601 coefficients
Hi Steve,
Quoted 28 lines · Steve Longerbeam
The ycbcr2rgb and inverse rgb2ycbcr tables define the BT.601 Y'CbCr encoding coefficients.
The rgb2ycbcr table specifically describes the BT.601 encoding from full range RGB to full range YUV. Add table comments to make this more clear.
The ycbcr2rgb inverse table describes encoding YUV limited range to RGB full range. To be consistent with the rgb2ycbcr table, convert this to YUV full range to RGB full range, and adjust/expand on the comments.
The
ic_csc_rgb2rgbtable is just an identity matrix, so rename toic_encode_identity.Fixes 1aa8ea0d2bd5d ("gpu: ipu-v3: Add Image Converter unit")Suggested-by Philipp Zabel <[email protected]>Signed-off-by Steve Longerbeam <[email protected]>--- drivers/gpu/ipu-v3/ipu-ic.c | 61 ++++++++++++++++++++++--------------- 1 file changed, 37 insertions(+), 24 deletions(-)drivers/gpu/ipu-v3/ipu-ic.c +0
-0
drivers/gpu/ipu-v3/ipu-ic.c static inline void ipu_ic_write(struct ipu_ic *ic, u32 value, unsigned offset) · line 178
Quoted 5 lines
drivers/gpu/ipu-v3/ipu-ic.c +1
-1@@ -175,7 +175,7 @@ static inline void ipu_ic_write(struct ipu_ic *ic, u32 value, unsigned offset)writel (value, ic->priv->base + offset); } -struct ic_csc_params { +struct ic_encode_coeff {
This less accurate. These are called IC (Task) Parameters in the reference manual, the 64-bit aligned words are called CSC words. Beside the coefficients, this structure also contains the coefficient scale, the offsets, and the saturation mode flag.
Quoted 13 lines
drivers/gpu/ipu-v3/ipu-ic.c +3
-3s16 coeff[3 ][3 ];/* signed 9-bit integer coefficients */ s16 offset[3 ];/* signed 11+2-bit fixed point offset */ u8 scale:2; /* scale coefficients * 2^(scale-1) */@@ -183,13 +183,15 @@ struct ic_csc_params { };/* -* Y = R * .299 + G * .587 + B * .114; -* U = R * -.169 + G * -.332 + B * .500 + 128.; -* V = R * .500 + G * -.419 + B * -.0813 + 128.; +* BT.601 encoding from RGB full range to YUV full range: +* +* Y = .2990 * R + .5870 * G + .1140 * B
drivers/gpu/ipu-v3/ipu-ic.c struct ic_csc_params { · line 196
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +4
-2+* U = -.1687 * R - .3313 * G + .5000 * B + 128 +* V = .5000 * R - .4187 * G - .0813 * B + 128 */ -static const struct ic_csc_params ic_csc_rgb2ycbcr = { +static const struct ic_encode_coeff ic_encode_rgb2ycbcr_601 = { .coeff = { - {77 ,150 ,29 }, + {77 ,150 ,29 }, {469 ,427 ,128 }, {128 ,405 ,491 },
We could subtract 512 from the negative values, to improve readability.
Quoted 16 lines
drivers/gpu/ipu-v3/ipu-ic.c +5
-2},@@ -197,8 +199,11 @@ static const struct ic_csc_params ic_csc_rgb2ycbcr = { .scale = 1, }; -/* transparent RGB->RGB matrix for graphics combining */ -static const struct ic_csc_params ic_csc_rgb2rgb = { +/* +* identity matrix, used for transparent RGB->RGB graphics +* combining. +*/ +static const struct ic_encode_coeff ic_encode_identity = { .coeff = { {128 ,0 ,0 }, {0 ,128 ,0 },
drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_csc_params ic_csc_rgb2rgb = { · line 219
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +4
-3@@ -208,17 +213,25 @@ static const struct ic_csc_params ic_csc_rgb2rgb = { };/* -* R = (1.164 * (Y - 16)) + (1.596 * (Cr - 128)); -* G = (1.164 * (Y - 16)) - (0.392 * (Cb - 128)) - (0.813 * (Cr - 128)); -* B = (1.164 * (Y - 16)) + (2.017 * (Cb - 128); +* Inverse BT.601 encoding from YUV full range to RGB full range: +* +* R = 1. * Y + 0 * (Cb - 128) + 1.4020 * (Cr - 128) +* G = 1. * Y - .3442 * (Cb - 128) - 0.7142 * (Cr - 128)
Should that be ^^^^^ .3441 and ^^^^^ .7141 ?
The coefficients and offsets after rounding should end up the same.
Also, let's consistently either add the leading zero, or leave it out.
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +8
-1+* B = 1. * Y + 1.7720 * (Cb - 128) + 0 * (Cr - 128) +* +* equivalently (factoring out the offsets): +* +* R = 1. * Y + 0 * Cb + 1.4020 * Cr - 179.456 +* G = 1. * Y - .3442 * Cb - 0.7142 * Cr + 135.475 +* B = 1. * Y + 1.7720 * Cb + 0 * Cr - 226.816 */ -static const struct ic_csc_params ic_csc_ycbcr2rgb = { +static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = {
drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_csc_params ic_csc_rgb2rgb = { · line 234
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +4
-4.coeff = { - { 149, 0, 204 }, - { 149, 462, 408 }, - { 149, 255, 0 }, + { 128, 0, 179 }, + { 128, 468, 421 }, + { 128, 227, 0 }, }, - .offset = { -446, 266, -554 }, + .offset = { -359, 271, -454 },
These seem to be correct. Again, I think this would be easier to read if the negative coefficients were written with a sign as well.
Quoted 24 lines
drivers/gpu/ipu-v3/ipu-ic.c +4
-4.scale = 2, };@@ -228,7 +241,7 @@ static int init_csc(struct ipu_ic *ic, int csc_index) {struct ipu_ic_priv *priv = ic->priv; -const struct ic_csc_params *params; +const struct ic_encode_coeff *coeff;u32 __iomem *base;const u16 (*c)[3 ];const u16 *a;@@ -238,26 +251,26 @@ static int init_csc(struct ipu_ic *ic, (priv->tpmem_base + ic->reg->tpmem_csc[csc_index]);if (inf ==IPUV3_COLORSPACE_YUV && outf ==IPUV3_COLORSPACE_RGB ) - params = &ic_csc_ycbcr2rgb; + coeff = &ic_encode_ycbcr2rgb_601;else if (inf ==IPUV3_COLORSPACE_RGB && outf ==IPUV3_COLORSPACE_YUV ) - params = &ic_csc_rgb2ycbcr; + coeff = &ic_encode_rgb2ycbcr_601;else if (inf ==IPUV3_COLORSPACE_RGB && outf ==IPUV3_COLORSPACE_RGB ) - params = &ic_csc_rgb2rgb; + coeff = &ic_encode_identity;
drivers/gpu/ipu-v3/ipu-ic.c static int init_csc(struct ipu_ic *ic, · line 266
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +2
-2else {dev_err (priv->ipu->dev,"Unsupported color space conversion\n" );return -EINVAL ; }/* Cast to unsigned */ - c = (const u16 (*)[3 ])params->coeff; - a = (const u16 *)params->offset; + c = (const u16 (*)[3 ])coeff->coeff; + a = (const u16 *)coeff->offset;
This looks weird to me. I'd be in favor of not renaming the type.
regards
Philipp
Steve Longerbeam
maintainerRe: [PATCH v6 2/7] gpu: ipu-v3: ipu-ic: Fix BT.601 coefficients
Quoted 34 lines · Philipp Zabel
Hi Steve,
Quoted 31 lines · Steve Longerbeam
The ycbcr2rgb and inverse rgb2ycbcr tables define the BT.601 Y'CbCr encoding coefficients.
The rgb2ycbcr table specifically describes the BT.601 encoding from full range RGB to full range YUV. Add table comments to make this more clear.
The ycbcr2rgb inverse table describes encoding YUV limited range to RGB full range. To be consistent with the rgb2ycbcr table, convert this to YUV full range to RGB full range, and adjust/expand on the comments.
The
ic_csc_rgb2rgbtable is just an identity matrix, so rename toic_encode_identity.Fixes 1aa8ea0d2bd5d ("gpu: ipu-v3: Add Image Converter unit")Suggested-by Philipp Zabel <[email protected]>Signed-off-by Steve Longerbeam <[email protected]>--- drivers/gpu/ipu-v3/ipu-ic.c | 61 ++++++++++++++++++++++--------------- 1 file changed, 37 insertions(+), 24 deletions(-)drivers/gpu/ipu-v3/ipu-ic.c +0
-0@@ -175,7 +175,7 @@ static inline void ipu_ic_write(struct ipu_ic *ic, u32 value, unsigned offset)writel (value, ic->priv->base + offset); }
Quoted 6 lines
drivers/gpu/ipu-v3/ipu-ic.c static inline void ipu_ic_write(struct ipu_ic *ic, u32 value, unsigned offset) · line 178
Quoted 2 lines
drivers/gpu/ipu-v3/ipu-ic.c +1
-1-struct ic_csc_params { +struct ic_encode_coeff {This less accurate. These are called IC (Task) Parameters in the reference manual, the 64-bit aligned words are called CSC words. Beside the coefficients, this structure also contains the coefficient scale, the offsets, and the saturation mode flag.
It seemed to me the renaming was more clear, but I agree the former name conforms better to the manual nomenclature. I will revert this renaming.
Quoted 17 lines
Quoted 17 lines
drivers/gpu/ipu-v3/ipu-ic.c +5
-3s16 coeff[3 ][3 ];/* signed 9-bit integer coefficients */ s16 offset[3 ];/* signed 11+2-bit fixed point offset */ u8 scale:2; /* scale coefficients * 2^(scale-1) */@@ -183,13 +183,15 @@ struct ic_csc_params { };/* -* Y = R * .299 + G * .587 + B * .114; -* U = R * -.169 + G * -.332 + B * .500 + 128.; -* V = R * .500 + G * -.419 + B * -.0813 + 128.; +* BT.601 encoding from RGB full range to YUV full range: +* +* Y = .2990 * R + .5870 * G + .1140 * B +* U = -.1687 * R - .3313 * G + .5000 * B + 128 +* V = .5000 * R - .4187 * G - .0813 * B + 128 */
Quoted 8 lines
drivers/gpu/ipu-v3/ipu-ic.c struct ic_csc_params { · line 196
Quoted 7 lines
drivers/gpu/ipu-v3/ipu-ic.c +2
-2-static const struct ic_csc_params ic_csc_rgb2ycbcr = { +static const struct ic_encode_coeff ic_encode_rgb2ycbcr_601 = { .coeff = { - {77 ,150 ,29 }, + {77 ,150 ,29 }, {469 ,427 ,128 }, {128 ,405 ,491 },We could subtract 512 from the negative values, to improve readability.
Agreed.
Quoted 19 lines
Quoted 19 lines
drivers/gpu/ipu-v3/ipu-ic.c +5
-2},@@ -197,8 +199,11 @@ static const struct ic_csc_params ic_csc_rgb2ycbcr = { .scale = 1, }; -/* transparent RGB->RGB matrix for graphics combining */ -static const struct ic_csc_params ic_csc_rgb2rgb = { +/* +* identity matrix, used for transparent RGB->RGB graphics +* combining. +*/ +static const struct ic_encode_coeff ic_encode_identity = { .coeff = { {128 ,0 ,0 }, {0 ,128 ,0 },@@ -208,17 +213,25 @@ static const struct ic_csc_params ic_csc_rgb2rgb = { };
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_csc_params ic_csc_rgb2rgb = { · line 219
Quoted 8 lines
drivers/gpu/ipu-v3/ipu-ic.c +4
-3/* -* R = (1.164 * (Y - 16)) + (1.596 * (Cr - 128)); -* G = (1.164 * (Y - 16)) - (0.392 * (Cb - 128)) - (0.813 * (Cr - 128)); -* B = (1.164 * (Y - 16)) + (2.017 * (Cb - 128); +* Inverse BT.601 encoding from YUV full range to RGB full range: +* +* R = 1. * Y + 0 * (Cb - 128) + 1.4020 * (Cr - 128) +* G = 1. * Y - .3442 * (Cb - 128) - 0.7142 * (Cr - 128) Should that be ^^^^^ .3441 and ^^^^^ .7141 ?
The coefficients and offsets after rounding should end up the same.
Ok.
Quoted 2 lines
Also, let's consistently either add the leading zero, or leave it out.
Yes.
Quoted 53 lines
Quoted 11 lines
drivers/gpu/ipu-v3/ipu-ic.c +8
-1+* B = 1. * Y + 1.7720 * (Cb - 128) + 0 * (Cr - 128) +* +* equivalently (factoring out the offsets): +* +* R = 1. * Y + 0 * Cb + 1.4020 * Cr - 179.456 +* G = 1. * Y - .3442 * Cb - 0.7142 * Cr + 135.475 +* B = 1. * Y + 1.7720 * Cb + 0 * Cr - 226.816 */ -static const struct ic_csc_params ic_csc_ycbcr2rgb = { +static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = {drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_csc_params ic_csc_rgb2rgb = { · line 234
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +4
-4.coeff = { - { 149, 0, 204 }, - { 149, 462, 408 }, - { 149, 255, 0 }, + { 128, 0, 179 }, + { 128, 468, 421 }, + { 128, 227, 0 }, }, - .offset = { -446, 266, -554 }, + .offset = { -359, 271, -454 },These seem to be correct. Again, I think this would be easier to read if the negative coefficients were written with a sign as well.
Quoted 29 lines
drivers/gpu/ipu-v3/ipu-ic.c +4
-4.scale = 2, };@@ -228,7 +241,7 @@ static int init_csc(struct ipu_ic *ic, int csc_index) {struct ipu_ic_priv *priv = ic->priv; -const struct ic_csc_params *params; +const struct ic_encode_coeff *coeff;u32 __iomem *base;const u16 (*c)[3 ];const u16 *a;@@ -238,26 +251,26 @@ static int init_csc(struct ipu_ic *ic, (priv->tpmem_base + ic->reg->tpmem_csc[csc_index]);if (inf ==IPUV3_COLORSPACE_YUV && outf ==IPUV3_COLORSPACE_RGB ) - params = &ic_csc_ycbcr2rgb; + coeff = &ic_encode_ycbcr2rgb_601;else if (inf ==IPUV3_COLORSPACE_RGB && outf ==IPUV3_COLORSPACE_YUV ) - params = &ic_csc_rgb2ycbcr; + coeff = &ic_encode_rgb2ycbcr_601;else if (inf ==IPUV3_COLORSPACE_RGB && outf ==IPUV3_COLORSPACE_RGB ) - params = &ic_csc_rgb2rgb; + coeff = &ic_encode_identity;else {dev_err (priv->ipu->dev,"Unsupported color space conversion\n" );return -EINVAL ; }
Quoted 6 lines
drivers/gpu/ipu-v3/ipu-ic.c static int init_csc(struct ipu_ic *ic, · line 266
Quoted 5 lines
drivers/gpu/ipu-v3/ipu-ic.c +2
-2/* Cast to unsigned */ - c = (const u16 (*)[3 ])params->coeff; - a = (const u16 *)params->offset; + c = (const u16 (*)[3 ])coeff->coeff; + a = (const u16 *)coeff->offset;This looks weird to me. I'd be in favor of not renaming the type.
Ok.
Steve
Steve Longerbeam
maintainerPATCH v6 3/7 gpu: ipu-v3: ipu-ic: Fully describe colorspace conversions
In reply to Steve LongerbeamOnly 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().
--- 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(-)
drivers/gpu/ipu-v3/ipu-ic.c +90 -46
@@ -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 ; + } + +if (out->enc !=V4L2_YCBCR_ENC_601 ) { +dev_err (priv->ipu->dev,"Only BT.601 encoding supported\n" ); +return -ENOTSUPP ; + } + +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 ); + + *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;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,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;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 ); -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);
drivers/gpu/ipu-v3/ipu-image-convert.c +17 -10
@@ -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_rot_mode_is_irt (ctx->rot_mode)) {/* swap width/height for resizer */
@@ -1352,13 +1360,12 @@ static int convert_start(struct ipu_image_convert_run *run, unsigned int tile) s_image->tile[tile].height, dest_width, dest_height, rsc);/* setup the IC resizer and CSC */ - ret =ipu_ic_task_init_rsc (chan->ic, - s_image->tile[tile].width, - s_image->tile[tile].height, - dest_width, - dest_height, - src_cs, dest_cs, - rsc); + ret =ipu_ic_task_init_rsc (chan->ic, &src_cs, &dest_cs, + s_image->tile[tile].width, + s_image->tile[tile].height, + dest_width, + dest_height, + rsc);if (ret) {dev_err (priv->ipu->dev,"ipu_ic_task_init failed, %d\n" , ret);return ret;
drivers/staging/media/imx/imx-ic-prpencvf.c +16 -6
@@ -458,6 +458,7 @@ static int prp_setup_rotation(struct prp_priv *priv)struct imx_media_video_dev *vdev = priv->vdev;struct imx_ic_priv *ic_priv = priv->ic_priv;const struct imx_media_pixfmt *outcc, *incc; +struct ipu_ic_colorspace in_cs, out_cs;struct v4l2_mbus_framefmt *infmt;struct v4l2_pix_format *outfmt;dma_addr_t phys[2 ];
@@ -468,6 +469,11 @@ static int prp_setup_rotation(struct prp_priv *priv) incc = priv->cc[PRPENCVF_SINK_PAD ]; outcc = vdev->cc; +ipu_ic_fill_colorspace (&in_cs, infmt->colorspace, infmt->ycbcr_enc, + infmt->quantization, incc->cs); +ipu_ic_fill_colorspace (&out_cs, outfmt->colorspace, outfmt->ycbcr_enc, + outfmt->quantization, outcc->cs); + ret =imx_media_alloc_dma_buf (priv->md, &priv->rot_buf[0 ], outfmt->sizeimage);if (ret) {
@@ -481,10 +487,9 @@ static int prp_setup_rotation(struct prp_priv *priv) goto free_rot0; } - ret =ipu_ic_task_init (priv->ic, + ret =ipu_ic_task_init (priv->ic, &in_cs, &out_cs, infmt->width, infmt->height, - outfmt->height, outfmt->width, - incc->cs, outcc->cs); + outfmt->height, outfmt->width);if (ret) {v4l2_err (&ic_priv->sd,"ipu_ic_task_init failed, %d\n" , ret); goto free_rot1;
@@ -574,6 +579,7 @@ static int prp_setup_norotation(struct prp_priv *priv)struct imx_media_video_dev *vdev = priv->vdev;struct imx_ic_priv *ic_priv = priv->ic_priv;const struct imx_media_pixfmt *outcc, *incc; +struct ipu_ic_colorspace in_cs, out_cs;struct v4l2_mbus_framefmt *infmt;struct v4l2_pix_format *outfmt;dma_addr_t phys[2 ];
@@ -584,10 +590,14 @@ static int prp_setup_norotation(struct prp_priv *priv) incc = priv->cc[PRPENCVF_SINK_PAD ]; outcc = vdev->cc; - ret =ipu_ic_task_init (priv->ic, +ipu_ic_fill_colorspace (&in_cs, infmt->colorspace, infmt->ycbcr_enc, + infmt->quantization, incc->cs); +ipu_ic_fill_colorspace (&out_cs, outfmt->colorspace, outfmt->ycbcr_enc, + outfmt->quantization, outcc->cs); + + ret =ipu_ic_task_init (priv->ic, &in_cs, &out_cs, infmt->width, infmt->height, - outfmt->width, outfmt->height, - incc->cs, outcc->cs); + outfmt->width, outfmt->height);if (ret) {v4l2_err (&ic_priv->sd,"ipu_ic_task_init failed, %d\n" , ret);return ret;
include/video/imx-ipu-v3.h +31 -6
@@ -386,20 +386,45 @@ enum ipu_ic_task { IC_NUM_TASKS, }; +/* +* The parameters that describe a colorspace according to the +* Image Converter: colorspace (chromaticities), Y'CbCr encoding, +* quantization, and "colorspace" (RGB or YUV). +*/ +struct ipu_ic_colorspace { +enum v4l2_colorspace colorspace; +enum v4l2_ycbcr_encoding enc; +enum v4l2_quantization quant; +enum ipu_color_space cs; +}; + +static inline void +ipu_ic_fill_colorspace (struct ipu_ic_colorspace *ic_cs, +enum v4l2_colorspace colorspace, +enum v4l2_ycbcr_encoding enc, +enum v4l2_quantization quant, +enum ipu_color_space cs) +{ + ic_cs->colorspace = colorspace; + ic_cs->enc = enc; + ic_cs->quant = quant; + ic_cs->cs = cs; +} +struct ipu_ic ;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);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);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,bool galpha_en,u32 galpha,bool colorkey_en,u32 colorkey);void ipu_ic_task_enable (struct ipu_ic *ic);
Signature · 1 line
2.17.1
Philipp Zabel
maintainerRe: [PATCH v6 3/7] gpu: ipu-v3: ipu-ic: Fully describe colorspace conversions
Quoted 52 lines · Steve Longerbeam
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 outputipu_ic_colorspaceto 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 byinit_csc().Signed-off-by Steve Longerbeam <[email protected]>--- 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(-)drivers/gpu/ipu-v3/ipu-ic.c +5
-2@@ -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,
drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = { · line 250
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +10
-0+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.
drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = { · line 255
Quoted 4 lines
drivers/gpu/ipu-v3/ipu-ic.c +4
-0+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.
Quoted 14 lines
drivers/gpu/ipu-v3/ipu-ic.c +14
-0+ +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" );
drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = { · line 279
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +10
-0+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?
Quoted 7 lines
drivers/gpu/ipu-v3/ipu-ic.c +7
-0+ + *coeff_out = inverse_encode ? + ic_encode_ycbcr2rgb_601 : ic_encode_rgb2ycbcr_601; + +return 0 ; +} +
drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = { · line 293
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +3
-3static 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.
Quoted 59 lines
drivers/gpu/ipu-v3/ipu-ic.c +11
-17u32 __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; }
drivers/gpu/ipu-v3/ipu-ic.c void ipu_ic_task_disable(struct ipu_ic *ic) · line 445
Quoted 5 lines
drivers/gpu/ipu-v3/ipu-ic.c +1
-1@@ -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.
Btw, do you have any plans for using IC composition? ipu_ic_task_graphics_init() is currently unused...
Quoted 24 lines
drivers/gpu/ipu-v3/ipu-ic.c +9
-3bool 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;
drivers/gpu/ipu-v3/ipu-ic.c int ipu_ic_task_graphics_init(struct ipu_ic *ic, · line 480
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +4
-5+ 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.
Quoted 21 lines
drivers/gpu/ipu-v3/ipu-ic.c +2
-3if (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;
drivers/gpu/ipu-v3/ipu-ic.c int ipu_ic_task_init_rsc(struct ipu_ic *ic, · line 545
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +3
-6- 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.
Quoted 41 lines
drivers/gpu/ipu-v3/ipu-ic.c +6
-6-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);drivers/gpu/ipu-v3/ipu-image-convert.c +1
-3@@ -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);
drivers/gpu/ipu-v3/ipu-image-convert.c static int convert_start(struct ipu_image_convert_run *run, unsigned int tile) · line 1339
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-image-convert.c +10
-0+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.
regards
Philipp
Steve Longerbeam
maintainerRe: [PATCH v6 3/7] gpu: ipu-v3: ipu-ic: Fully describe colorspace conversions
Quoted 52 lines · Philipp Zabel
Quoted 51 lines · Steve Longerbeam
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 outputipu_ic_colorspaceto 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 byinit_csc().Signed-off-by Steve Longerbeam <[email protected]>--- 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(-)drivers/gpu/ipu-v3/ipu-ic.c +4
-2@@ -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, };
Quoted 14 lines
drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = { · line 250
Quoted 11 lines
drivers/gpu/ipu-v3/ipu-ic.c +11
-0+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 forxfer_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.
Quoted 7 lines
drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = { · line 255
Quoted 5 lines
drivers/gpu/ipu-v3/ipu-ic.c +4
-0+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->csisIPUV3_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).
Quoted 18 lines
Quoted 18 lines
drivers/gpu/ipu-v3/ipu-ic.c +17
-0+ +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 ; + } +
Quoted 8 lines
drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = {
Quoted 7 lines
drivers/gpu/ipu-v3/ipu-ic.c +7
-0+if (in->cs = out->cs) { + *coeff_out = ic_encode_identity; + +return 0 ; + } + + inverse_encode = (in->cs =IPUV3_COLORSPACE_YUV );What does
inverse_encodemean in this context?
It means YUV to RGB. At this point in the function it is determined that encoding or inverse encoding is requested.
Quoted 12 lines
Quoted 12 lines
drivers/gpu/ipu-v3/ipu-ic.c +8
-2+ + *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,
Quoted 8 lines
drivers/gpu/ipu-v3/ipu-ic.c static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = { · line 293
Quoted 6 lines
drivers/gpu/ipu-v3/ipu-ic.c +2
-1+ 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.
Quoted 62 lines
Quoted 62 lines
drivers/gpu/ipu-v3/ipu-ic.c +11
-17u32 __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);
Quoted 7 lines
drivers/gpu/ipu-v3/ipu-ic.c void ipu_ic_task_disable(struct ipu_ic *ic) · line 445
Quoted 3 lines
drivers/gpu/ipu-v3/ipu-ic.c +1
-1int 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.
Quoted 3 lines
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?
Quoted 111 lines
Quoted 25 lines
drivers/gpu/ipu-v3/ipu-ic.c +9
-3bool 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;drivers/gpu/ipu-v3/ipu-ic.c int ipu_ic_task_graphics_init(struct ipu_ic *ic, · line 480
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +4
-5+ 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 inIC_CONF, and these TPMEM values will be ignored.Quoted 21 lines
drivers/gpu/ipu-v3/ipu-ic.c +2
-3if (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;drivers/gpu/ipu-v3/ipu-ic.c int ipu_ic_task_init_rsc(struct ipu_ic *ic, · line 545
Quoted 10 lines
drivers/gpu/ipu-v3/ipu-ic.c +3
-6- 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.
Quoted 39 lines
drivers/gpu/ipu-v3/ipu-ic.c +6
-6-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);drivers/gpu/ipu-v3/ipu-image-convert.c +1
-1@@ -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);
Quoted 15 lines
drivers/gpu/ipu-v3/ipu-image-convert.c static int convert_start(struct ipu_image_convert_run *run, unsigned int tile) · line 1339
Quoted 12 lines
drivers/gpu/ipu-v3/ipu-image-convert.c +10
-2- 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 inipu_image_convert_prepareand stored inipu_image_convert_ctxfor repeated use.
Yes, I'll add this for v7.
Steve
Steve Longerbeam
maintainerPATCH v6 4/7 gpu: ipu-v3: ipu-ic: Add support for Rec.709 encoding
In reply to Steve LongerbeamAdd support for Rec.709 encoding and inverse encoding.
---
Changes in v5:
- moved API changes to a previous patch.
- moved CSC coeff calc to new function calc_csc_coeffs(). Changes in v4:
- fix compile error.
Chnges in v3:
- none.
Changes in v2:
- only return "Unsupported YCbCr encoding" error if inf != outf, since if inf == outf, the identity matrix can be used. Reported by Tim Harvey.
--- drivers/gpu/ipu-v3/ipu-ic.c | 63 ++++++++++++++++++++++++++++++++----- 1 file changed, 56 insertions(+), 7 deletions(-)
drivers/gpu/ipu-v3/ipu-ic.c +56 -7
@@ -214,6 +214,23 @@ static const struct ic_encode_coeff ic_encode_identity = { .scale = 2, }; +/* +* REC.709 encoding from RGB full range to YUV full range: +* +* Y = .2126 * R + .7152 * G + .0722 * B +* U = -.1146 * R - .3854 * G + .5000 * B + 128 +* V = .5000 * R - .4542 * G - .0458 * B + 128 +*/ +static const struct ic_encode_coeff ic_encode_rgb2ycbcr_709 = { + .coeff = { + {54 ,183 ,19 }, + {483 ,413 ,128 }, + {128 ,396 ,500 }, + }, + .offset = {0 ,512 ,512 }, + .scale =1 , +}; +/* * Inverse BT.601 encoding from YUV full range to RGB full range: *
@@ -237,11 +254,35 @@ static const struct ic_encode_coeff ic_encode_ycbcr2rgb_601 = { .scale = 2, }; +/* +* Inverse REC.709 encoding from YUV full range to RGB full range: +* +* R = 1. * Y + 0 * (Cb - 128) + 1.5748 * (Cr - 128) +* G = 1. * Y - .1873 * (Cb - 128) - .4681 * (Cr - 128) +* B = 1. * Y + 1.8556 * (Cb - 128) + 0 * (Cr - 128) +* +* equivalently (factoring out the offsets): +* +* R = 1. * Y + 0 * Cb + 1.5748 * Cr - 201.574 +* G = 1. * Y - .1873 * Cb - .4681 * Cr + 83.891 +* B = 1. * Y + 1.8556 * Cb + 0 * Cr - 237.517 +*/ +static const struct ic_encode_coeff ic_encode_ycbcr2rgb_709 = { + .coeff = { + {128 ,0 ,202 }, + {128 ,488 ,452 }, + {128 ,238 ,0 }, + }, + .offset = {-403 ,168 ,-475 }, + .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) { +const struct ic_encode_coeff *encode_coeff;bool inverse_encode;if (in->colorspace != out->colorspace) {
@@ -249,11 +290,6 @@ static int calc_csc_coeffs(struct ipu_ic_priv *priv,return -ENOTSUPP ; } -if (out->enc !=V4L2_YCBCR_ENC_601 ) { -dev_err (priv->ipu->dev,"Only BT.601 encoding supported\n" ); -return -ENOTSUPP ; - } - if ((in->cs == IPUV3_COLORSPACE_YUV && in->quant != V4L2_QUANTIZATION_FULL_RANGE) || (out->cs == IPUV3_COLORSPACE_YUV &&
@@ -278,8 +314,21 @@ static int calc_csc_coeffs(struct ipu_ic_priv *priv, inverse_encode = (in->cs ==IPUV3_COLORSPACE_YUV ); - *coeff_out = inverse_encode ? - ic_encode_ycbcr2rgb_601 : ic_encode_rgb2ycbcr_601; +switch (out->enc) { +case V4L2_YCBCR_ENC_601 : + encode_coeff = inverse_encode ? + &ic_encode_ycbcr2rgb_601 : &ic_encode_rgb2ycbcr_601; +break ; +case V4L2_YCBCR_ENC_709 : + encode_coeff = inverse_encode ? + &ic_encode_ycbcr2rgb_709 : &ic_encode_rgb2ycbcr_709; +break ; +default : +dev_err (priv->ipu->dev,"Unsupported YCbCr encoding\n" ); +return -ENOTSUPP ; + } + + *coeff_out = *encode_coeff;return 0 ; }
Signature · 6 lines
2.17.1 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Steve Longerbeam
maintainerPATCH v6 5/7 gpu: ipu-v3: ipu-ic: Add support for limited range encoding
In reply to Steve LongerbeamAdd support for the following conversions:
- YUV full-range to YUV limited-range
- YUV limited-range to YUV full-range
- YUV limited-range to RGB full-range
- RGB full-range to YUV limited-range
The last two conversions require operating on the YUV full-range encoding and inverse encoding coefficients, with the YUV-to-YUV limited<->full coefficients. The formula to convert is
M_c = M_a * M_bO_c = M_a * O_b + O_a
For calculating the RGB full-range to YUV limited-range coefficients:
[M_a, O_a] = YUV full-range to YUV limited-range coefficients. [M_b, O_b] = RGB full-range to YUV full-range coefficients.
For calculating the YUV limited-range to RGB full-range coefficients:
[M_a, O_a] = YUV full-range to RGB full-range coefficients. [M_b, O_b] = YUV limited-range to YUV full-range coefficients.
The calculation of [M_c, O_c] is carried out by the function transform_coeffs().
In the future if RGB limited range encoding is required, the same function can be used. And cascaded to create all combinations of encoding for YUV limited/full range <-> RGB limited/full range, passing the output coefficients from one call as the input for the next.
For example, to create YUV full-range to RGB limited-range coefficients:
[M_a, O_a] = RGB full-range to RGB limited-range coefficients. [M_b, O_b] = YUV full-range to RGB full-range coefficients.
and that output sent as input to create YUV limited-range to RGB limited-range coefficients:
[M_a, O_a] = YUV full-range to RGB limited-range coefficients. [M_b, O_b] = YUV limited-range to YUV full-range coefficients.
--- drivers/gpu/ipu-v3/ipu-ic.c | 281 +++++++++++++++++++++++++++++++++--- 1 file changed, 263 insertions(+), 18 deletions(-)
drivers/gpu/ipu-v3/ipu-ic.c +263 -18
@@ -178,10 +178,10 @@ static inline void ipu_ic_write(struct ipu_ic *ic, u32 value, unsigned offset) }struct ic_encode_coeff { -s16 coeff[3 ][3 ];/* signed 9-bit integer coefficients */ -s16 offset[3 ];/* signed 11+2-bit fixed point offset */ -u8 scale:2 ;/* scale coefficients * 2^(scale-1) */ -bool sat:1 ;/* saturate to (16, 235(Y) / 240(U, V)) */ +int coeff[3 ][3 ];/* signed 9-bit integer coefficients */ +int offset[3 ];/* signed 13-bit integer offset */ +int scale;/* scale coefficients * 2^(scale-1) */ +bool sat;/* saturate to (16, 235(Y) / 240(U, V)) */ };/*
@@ -277,6 +277,231 @@ static const struct ic_encode_coeff ic_encode_ycbcr2rgb_709 = { .scale = 2, }; +/* +* YUV full range to YUV limited range: +* +* Y_lim = 0.8588 * Y_full + 16 +* Cb_lim = 0.8784 * (Cb_full - 128) + 128 +* Cr_lim = 0.8784 * (Cr_full - 128) + 128 +*/ +static const struct ic_encode_coeff ic_encode_ycbcr_full2lim = { + .coeff = { + {219 ,0 ,0 }, + {0 ,224 ,0 }, + {0 ,0 ,224 }, + }, + .offset = {64 ,62 ,62 }, + .scale =1 , +}; + +/* +* YUV limited range to YUV full range: +* +* Y_full = 1.1644 * (Y_lim - 16) +* Cb_full = 1.1384 * (Cb_lim - 128) + 128 +* Cr_full = 1.1384 * (Cr_lim - 128) + 128 +*/ +static const struct ic_encode_coeff ic_encode_ycbcr_lim2full = { + .coeff = { + {149 ,0 ,0 }, + {0 ,145 ,0 }, + {0 ,0 ,145 }, + }, + .offset = {-37 ,-35 ,-35 }, + .scale =2 , +}; + +/* +* RGB full range to RGB limited range: +* +* R_lim = 0.8588 * R_full + 16 +* G_lim = 0.8588 * G_full + 16 +* B_lim = 0.8588 * B_full + 16 +*/ +static const struct ic_encode_coeff +ic_encode_rgb_full2lim__maybe_unused = { + .coeff = { + {220 ,0 ,0 }, + {0 ,220 ,0 }, + {0 ,0 ,220 }, + }, + .offset = {64 ,64 ,64 }, + .scale =1 , +}; + +/* +* RGB limited range to RGB full range: +* +* R_full = 1.1644 * (R_lim - 16) +* G_full = 1.1644 * (G_lim - 16) +* B_full = 1.1644 * (B_lim - 16) +*/ +static const struct ic_encode_coeff +ic_encode_rgb_lim2full__maybe_unused = { + .coeff = { + {149 ,0 ,0 }, + {0 ,149 ,0 }, + {0 ,0 ,149 }, + }, + .offset = {-37 ,-37 ,-37 }, + .scale =2 , +}; + +/* +* Convert a coefficient and scale value in TPMEM register format +* to a signed int times 256 (fix the radix point). The TPMEM register +* coefficient format is a signed 9-bit value (sign bit at bit 8, +* mantissa = coeff * 2 ^ (8 - scale - 1)). +*/ +static int coeff_fix (int coeff,int scale) +{ +if (coeff >=256 ) + coeff -=512 ; +if (scale ==0 ) +return DIV_ROUND_CLOSEST (coeff,2 ); +return coeff << (scale -1 ); +} + +/* +* Convert a signed int coefficient times 256 to TPMEM register +* format, given a scale value = TPMEM scale - 1. +*/ +static int coeff_normalize (int coeff,int scale) +{ + coeff =DIV_ROUND_CLOSEST (coeff,1 << scale); +if (coeff <0 ) + coeff +=512 ; +return coeff; +} + +/* +* Convert an offset and scale value in TPMEM register format to a +* signed int times 256 (fix the radix point). The TPMEM register +* offset format is a signed 13-bit value (sign bit at bit 12, +* mantissa = offset * 2 ^ (2 - (scale - 1)). +*/ +static int offset_fix (int offset,int scale) +{ +return offset << (8 - (2 - (scale -1 ))); +} + +/* +* Convert a signed int offset times 256 to TPMEM register +* format, given a scale value = TPMEM scale - 1. +*/ +static int offset_normalize (int off,int scale) +{ +return DIV_ROUND_CLOSEST (off,1 << (8 - (2 - scale))); +} + +/* +* Find the scale value that fits the given coefficient within +* the 8-bit TPMEM mantissa. +*/ +static int get_coeff_scale (int coeff) +{ +int scale =0 ; + +while (abs (coeff) >=256 && scale <=2 ) { + coeff =DIV_ROUND_CLOSEST (coeff,2 ); + scale++; + } + +return scale; +} + +/* +* The above defined encoding coefficients all encode between +* full-range RGB and full-range YCbCr. +* +* This function calculates a matrix M_c and offset vector O_c, given +* input matrices M_a, M_b and offset vectors O_a, O_b, such that: +* +* M_c = M_a * M_b +* O_c = M_a * O_b + O_a +* +* This operation will transform the full-range coefficients to +* coefficients that encode to or from limited range YCbCr or RGB. +* +* For example, to transform ic_encode_rgb2ycbcr_601 to encode to +* limited-range YCbCr: +* +* [M_a, O_a] = ic_encode_ycbcr_full2lim +* [M_b, O_b] = ic_encode_rgb2ycbcr_601 +* +* To transform the inverse coefficients ic_encode_ycbcr2rgb_601 to +* encode from limited-range YCbCr: +* +* [M_a, O_a] = ic_encode_ycbcr2rgb_601 +* [M_b, O_b] = ic_encode_ycbcr_lim2full +* +* The function can also be used to create RGB limited range +* coefficients, and cascaded to create all combinations of +* encodings between YCbCr limited/full range <-> RGB limited/full +* range. +*/ +static void transform_coeffs (struct ic_encode_coeff *out, +const struct ic_encode_coeff *a, +const struct ic_encode_coeff *b) +{ +int c_a, c_b, c_out; +int o_a, o_b, o_out; +int outscale =0 ; +int i, j, k; + +for (i =0 ; i <3 ; i++) { + o_out =0 ; +for (j =0 ; j <3 ; j++) { +int scale; + +/* M_c[i,j] = M_a[i,k] * M_b[k,j] */ + c_out =0 ; +for (k =0 ; k <3 ; k++) { + c_a =coeff_fix (a->coeff[i][k], a->scale); + c_b =coeff_fix (b->coeff[k][j], b->scale); + c_out += c_a * c_b; + } + + c_out =DIV_ROUND_CLOSEST (c_out,1 <<8 ); + out->coeff[i][j] = c_out; + +/* +* get scale for this coefficient and update +* final output scale. +*/ + scale =get_coeff_scale (c_out); + outscale =max (outscale, scale); + +/* M_a[i,j] * O_b[j] */ + c_a =coeff_fix (a->coeff[i][j], a->scale); + o_b =offset_fix (b->offset[j], b->scale); + o_out +=DIV_ROUND_CLOSEST (c_a * o_b,1 <<8 ); + } + +/* O_c[i] = (M_a * O_b)[i] + O_a[i] */ + o_a =offset_fix (a->offset[i], a->scale); + o_out += o_a; + + out->offset[i] = o_out; + } + +/* +* normalize output coefficients and offsets to TPMEM +* register format. +*/ +for (i =0 ; i <3 ; i++) { +for (j =0 ; j <3 ; j++) { + c_out = out->coeff[i][j]; + out->coeff[i][j] =coeff_normalize (c_out, outscale); + } + + o_out = out->offset[i]; + out->offset[i] =offset_normalize (o_out, outscale); + } + + out->scale = outscale +1 ; +} + static int calc_csc_coeffs(struct ipu_ic_priv *priv, struct ic_encode_coeff *coeff_out, const struct ipu_ic_colorspace *in,
@@ -290,14 +515,6 @@ static int calc_csc_coeffs(struct ipu_ic_priv *priv,return -ENOTSUPP ; } -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 &&
@@ -307,7 +524,18 @@ static int calc_csc_coeffs(struct ipu_ic_priv *priv, }if (in->cs == out->cs) { - *coeff_out = ic_encode_identity; +if (in->quant == out->quant) { + *coeff_out = ic_encode_identity; + }else if (in->quant ==V4L2_QUANTIZATION_FULL_RANGE ) { +/* YUV full-range to YUV limited-range */ + *coeff_out = ic_encode_ycbcr_full2lim; + +/* set saturation bit for YUV limited-range output */ + coeff_out->sat = true; + }else { +/* YUV limited-range to YUV full-range */ + *coeff_out = ic_encode_ycbcr_lim2full; + }return 0 ; }
@@ -328,7 +556,24 @@ static int calc_csc_coeffs(struct ipu_ic_priv *priv,return -ENOTSUPP ; } - *coeff_out = *encode_coeff; +if (in->quant == out->quant) { +/* +* YUV full-range to RGB full-range, or +* RGB full-range to YUV full-range. +*/ + *coeff_out = *encode_coeff; + }else if (inverse_encode) { +/* YUV limited-range to RGB full-range */ +transform_coeffs (coeff_out, encode_coeff, + &ic_encode_ycbcr_lim2full); + }else { +/* RGB full-range to YUV limited-range */ +transform_coeffs (coeff_out, &ic_encode_ycbcr_full2lim, + encode_coeff); + +/* set saturation bit for YUV limited-range output */ + coeff_out->sat = true; + }return 0 ; }
@@ -340,9 +585,9 @@ static int init_csc(struct ipu_ic *ic, {struct ipu_ic_priv *priv = ic->priv;struct ic_encode_coeff coeff; +const unsigned int (*c)[3 ]; +const unsigned int *a;u32 __iomem *base; -const u16 (*c)[3 ]; -const u16 *a;u32 param;int ret;
@@ -354,8 +599,8 @@ static int init_csc(struct ipu_ic *ic, (priv->tpmem_base + ic->reg->tpmem_csc[csc_index]);/* Cast to unsigned */ - c = (const u16 (*)[3 ])coeff.coeff; - a = (const u16 *)coeff.offset; + c = (const unsigned int (*)[3 ])coeff.coeff; + a = (const unsigned int *)coeff.offset; param = ((a[0 ] &0x1f ) <<27 ) | ((c[0 ][0 ] &0x1ff ) <<18 ) | ((c[1 ][1 ] &0x1ff ) <<9 ) | (c[2 ][2 ] &0x1ff );
Signature · 1 line
2.17.1
Philipp Zabel
maintainerRe: [PATCH v6 5/7] gpu: ipu-v3: ipu-ic: Add support for limited range encoding
Quoted 38 lines · Steve Longerbeam
Add support for the following conversions:
- YUV full-range to YUV limited-range
- YUV limited-range to YUV full-range
- YUV limited-range to RGB full-range
- RGB full-range to YUV limited-rangeThe last two conversions require operating on the YUV full-range encoding and inverse encoding coefficients, with the YUV-to-YUV limited<->full coefficients. The formula to convert is
M_c=M_a*M_bO_c=M_a*O_b+O_aFor calculating the RGB full-range to YUV limited-range coefficients:
[
M_a,O_a] = YUV full-range to YUV limited-range coefficients. [M_b,O_b] = RGB full-range to YUV full-range coefficients.For calculating the YUV limited-range to RGB full-range coefficients:
[
M_a,O_a] = YUV full-range to RGB full-range coefficients. [M_b,O_b] = YUV limited-range to YUV full-range coefficients.The calculation of [
M_c,O_c] is carried out by the functiontransform_coeffs().In the future if RGB limited range encoding is required, the same function can be used. And cascaded to create all combinations of encoding for YUV limited/full range <-> RGB limited/full range, passing the output coefficients from one call as the input for the next.
For example, to create YUV full-range to RGB limited-range coefficients:
[
M_a,O_a] = RGB full-range to RGB limited-range coefficients. [M_b,O_b] = YUV full-range to RGB full-range coefficients.
Quoted 7 lines
and that output sent as input to create YUV limited-range to RGB limited-range coefficients:
[
M_a,O_a] = YUV full-range to RGB limited-range coefficients. [M_b,O_b] = YUV limited-range to YUV full-range coefficients.Signed-off-by Steve Longerbeam <[email protected]>
I'm not a big fan of this. Wouldn't it be much easier to compute all necessary task parameter sets offline with high precision, and store the precomputed sets in the compact representation?
regards
Philipp
Quoted 339 lines
--- drivers/gpu/ipu-v3/ipu-ic.c | 281 +++++++++++++++++++++++++++++++++--- 1 file changed, 263 insertions(+), 18 deletions(-)drivers/gpu/ipu-v3/ipu-ic.c +263
-18@@ -178,10 +178,10 @@ static inline void ipu_ic_write(struct ipu_ic *ic, u32 value, unsigned offset) }struct ic_encode_coeff { -s16 coeff[3 ][3 ];/* signed 9-bit integer coefficients */ -s16 offset[3 ];/* signed 11+2-bit fixed point offset */ -u8 scale:2 ;/* scale coefficients * 2^(scale-1) */ -bool sat:1 ;/* saturate to (16, 235(Y) / 240(U, V)) */ +int coeff[3 ][3 ];/* signed 9-bit integer coefficients */ +int offset[3 ];/* signed 13-bit integer offset */ +int scale;/* scale coefficients * 2^(scale-1) */ +bool sat;/* saturate to (16, 235(Y) / 240(U, V)) */ };/* @@ -277,6 +277,231 @@ static const struct ic_encode_coeff ic_encode_ycbcr2rgb_709 = { .scale = 2, }; +/* +* YUV full range to YUV limited range: +* +* Y_lim = 0.8588 * Y_full + 16 +* Cb_lim = 0.8784 * (Cb_full - 128) + 128 +* Cr_lim = 0.8784 * (Cr_full - 128) + 128 +*/ +static const struct ic_encode_coeff ic_encode_ycbcr_full2lim = { + .coeff = { + {219 ,0 ,0 }, + {0 ,224 ,0 }, + {0 ,0 ,224 }, + }, + .offset = {64 ,62 ,62 }, + .scale =1 , +}; + +/* +* YUV limited range to YUV full range: +* +* Y_full = 1.1644 * (Y_lim - 16) +* Cb_full = 1.1384 * (Cb_lim - 128) + 128 +* Cr_full = 1.1384 * (Cr_lim - 128) + 128 +*/ +static const struct ic_encode_coeff ic_encode_ycbcr_lim2full = { + .coeff = { + {149 ,0 ,0 }, + {0 ,145 ,0 }, + {0 ,0 ,145 }, + }, + .offset = {-37 ,-35 ,-35 }, + .scale =2 , +}; + +/* +* RGB full range to RGB limited range: +* +* R_lim = 0.8588 * R_full + 16 +* G_lim = 0.8588 * G_full + 16 +* B_lim = 0.8588 * B_full + 16 +*/ +static const struct ic_encode_coeff +ic_encode_rgb_full2lim__maybe_unused = { + .coeff = { + {220 ,0 ,0 }, + {0 ,220 ,0 }, + {0 ,0 ,220 }, + }, + .offset = {64 ,64 ,64 }, + .scale =1 , +}; + +/* +* RGB limited range to RGB full range: +* +* R_full = 1.1644 * (R_lim - 16) +* G_full = 1.1644 * (G_lim - 16) +* B_full = 1.1644 * (B_lim - 16) +*/ +static const struct ic_encode_coeff +ic_encode_rgb_lim2full__maybe_unused = { + .coeff = { + {149 ,0 ,0 }, + {0 ,149 ,0 }, + {0 ,0 ,149 }, + }, + .offset = {-37 ,-37 ,-37 }, + .scale =2 , +}; + +/* +* Convert a coefficient and scale value in TPMEM register format +* to a signed int times 256 (fix the radix point). The TPMEM register +* coefficient format is a signed 9-bit value (sign bit at bit 8, +* mantissa = coeff * 2 ^ (8 - scale - 1)). +*/ +static int coeff_fix (int coeff,int scale) +{ +if (coeff >=256 ) + coeff -=512 ; +if (scale ==0 ) +return DIV_ROUND_CLOSEST (coeff,2 ); +return coeff << (scale -1 ); +} + +/* +* Convert a signed int coefficient times 256 to TPMEM register +* format, given a scale value = TPMEM scale - 1. +*/ +static int coeff_normalize (int coeff,int scale) +{ + coeff =DIV_ROUND_CLOSEST (coeff,1 << scale); +if (coeff <0 ) + coeff +=512 ; +return coeff; +} + +/* +* Convert an offset and scale value in TPMEM register format to a +* signed int times 256 (fix the radix point). The TPMEM register +* offset format is a signed 13-bit value (sign bit at bit 12, +* mantissa = offset * 2 ^ (2 - (scale - 1)). +*/ +static int offset_fix (int offset,int scale) +{ +return offset << (8 - (2 - (scale -1 ))); +} + +/* +* Convert a signed int offset times 256 to TPMEM register +* format, given a scale value = TPMEM scale - 1. +*/ +static int offset_normalize (int off,int scale) +{ +return DIV_ROUND_CLOSEST (off,1 << (8 - (2 - scale))); +} + +/* +* Find the scale value that fits the given coefficient within +* the 8-bit TPMEM mantissa. +*/ +static int get_coeff_scale (int coeff) +{ +int scale =0 ; + +while (abs (coeff) >=256 && scale <=2 ) { + coeff =DIV_ROUND_CLOSEST (coeff,2 ); + scale++; + } + +return scale; +} + +/* +* The above defined encoding coefficients all encode between +* full-range RGB and full-range YCbCr. +* +* This function calculates a matrix M_c and offset vector O_c, given +* input matrices M_a, M_b and offset vectors O_a, O_b, such that: +* +* M_c = M_a * M_b +* O_c = M_a * O_b + O_a +* +* This operation will transform the full-range coefficients to +* coefficients that encode to or from limited range YCbCr or RGB. +* +* For example, to transform ic_encode_rgb2ycbcr_601 to encode to +* limited-range YCbCr: +* +* [M_a, O_a] = ic_encode_ycbcr_full2lim +* [M_b, O_b] = ic_encode_rgb2ycbcr_601 +* +* To transform the inverse coefficients ic_encode_ycbcr2rgb_601 to +* encode from limited-range YCbCr: +* +* [M_a, O_a] = ic_encode_ycbcr2rgb_601 +* [M_b, O_b] = ic_encode_ycbcr_lim2full +* +* The function can also be used to create RGB limited range +* coefficients, and cascaded to create all combinations of +* encodings between YCbCr limited/full range <-> RGB limited/full +* range. +*/ +static void transform_coeffs (struct ic_encode_coeff *out, +const struct ic_encode_coeff *a, +const struct ic_encode_coeff *b) +{ +int c_a, c_b, c_out; +int o_a, o_b, o_out; +int outscale =0 ; +int i, j, k; + +for (i =0 ; i <3 ; i++) { + o_out =0 ; +for (j =0 ; j <3 ; j++) { +int scale; + +/* M_c[i,j] = M_a[i,k] * M_b[k,j] */ + c_out =0 ; +for (k =0 ; k <3 ; k++) { + c_a =coeff_fix (a->coeff[i][k], a->scale); + c_b =coeff_fix (b->coeff[k][j], b->scale); + c_out += c_a * c_b; + } + + c_out =DIV_ROUND_CLOSEST (c_out,1 <<8 ); + out->coeff[i][j] = c_out; + +/* +* get scale for this coefficient and update +* final output scale. +*/ + scale =get_coeff_scale (c_out); + outscale =max (outscale, scale); + +/* M_a[i,j] * O_b[j] */ + c_a =coeff_fix (a->coeff[i][j], a->scale); + o_b =offset_fix (b->offset[j], b->scale); + o_out +=DIV_ROUND_CLOSEST (c_a * o_b,1 <<8 ); + } + +/* O_c[i] = (M_a * O_b)[i] + O_a[i] */ + o_a =offset_fix (a->offset[i], a->scale); + o_out += o_a; + + out->offset[i] = o_out; + } + +/* +* normalize output coefficients and offsets to TPMEM +* register format. +*/ +for (i =0 ; i <3 ; i++) { +for (j =0 ; j <3 ; j++) { + c_out = out->coeff[i][j]; + out->coeff[i][j] =coeff_normalize (c_out, outscale); + } + + o_out = out->offset[i]; + out->offset[i] =offset_normalize (o_out, outscale); + } + + out->scale = outscale +1 ; +} + static int calc_csc_coeffs(struct ipu_ic_priv *priv, struct ic_encode_coeff *coeff_out, const struct ipu_ic_colorspace *in,@@ -290,14 +515,6 @@ static int calc_csc_coeffs(struct ipu_ic_priv *priv,return -ENOTSUPP ; } -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 &&@@ -307,7 +524,18 @@ static int calc_csc_coeffs(struct ipu_ic_priv *priv, }if (in->cs == out->cs) { - *coeff_out = ic_encode_identity; +if (in->quant == out->quant) { + *coeff_out = ic_encode_identity; + }else if (in->quant ==V4L2_QUANTIZATION_FULL_RANGE ) { +/* YUV full-range to YUV limited-range */ + *coeff_out = ic_encode_ycbcr_full2lim; + +/* set saturation bit for YUV limited-range output */ + coeff_out->sat = true; + }else { +/* YUV limited-range to YUV full-range */ + *coeff_out = ic_encode_ycbcr_lim2full; + }return 0 ; }@@ -328,7 +556,24 @@ static int calc_csc_coeffs(struct ipu_ic_priv *priv,return -ENOTSUPP ; } - *coeff_out = *encode_coeff; +if (in->quant == out->quant) { +/* +* YUV full-range to RGB full-range, or +* RGB full-range to YUV full-range. +*/ + *coeff_out = *encode_coeff; + }else if (inverse_encode) { +/* YUV limited-range to RGB full-range */ +transform_coeffs (coeff_out, encode_coeff, + &ic_encode_ycbcr_lim2full); + }else { +/* RGB full-range to YUV limited-range */ +transform_coeffs (coeff_out, &ic_encode_ycbcr_full2lim, + encode_coeff); + +/* set saturation bit for YUV limited-range output */ + coeff_out->sat = true; + }return 0 ; }@@ -340,9 +585,9 @@ static int init_csc(struct ipu_ic *ic, {struct ipu_ic_priv *priv = ic->priv;struct ic_encode_coeff coeff; +const unsigned int (*c)[3 ]; +const unsigned int *a;u32 __iomem *base; -const u16 (*c)[3 ]; -const u16 *a;u32 param;int ret;@@ -354,8 +599,8 @@ static int init_csc(struct ipu_ic *ic, (priv->tpmem_base + ic->reg->tpmem_csc[csc_index]);/* Cast to unsigned */ - c = (const u16 (*)[3 ])coeff.coeff; - a = (const u16 *)coeff.offset; + c = (const unsigned int (*)[3 ])coeff.coeff; + a = (const unsigned int *)coeff.offset; param = ((a[0 ] &0x1f ) <<27 ) | ((c[0 ][0 ] &0x1ff ) <<18 ) | ((c[1 ][1 ] &0x1ff ) <<9 ) | (c[2 ][2 ] &0x1ff );
Steve Longerbeam
maintainerRe: [PATCH v6 5/7] gpu: ipu-v3: ipu-ic: Add support for limited range encoding
Quoted 42 lines · Philipp Zabel
Quoted 41 lines · Steve Longerbeam
Add support for the following conversions:
- YUV full-range to YUV limited-range
- YUV limited-range to YUV full-range
- YUV limited-range to RGB full-range
- RGB full-range to YUV limited-rangeThe last two conversions require operating on the YUV full-range encoding and inverse encoding coefficients, with the YUV-to-YUV limited<->full coefficients. The formula to convert is
M_c=M_a*M_bO_c=M_a*O_b+O_aFor calculating the RGB full-range to YUV limited-range coefficients:
[
M_a,O_a] = YUV full-range to YUV limited-range coefficients. [M_b,O_b] = RGB full-range to YUV full-range coefficients.For calculating the YUV limited-range to RGB full-range coefficients:
[
M_a,O_a] = YUV full-range to RGB full-range coefficients. [M_b,O_b] = YUV limited-range to YUV full-range coefficients.The calculation of [
M_c,O_c] is carried out by the functiontransform_coeffs().In the future if RGB limited range encoding is required, the same function can be used. And cascaded to create all combinations of encoding for YUV limited/full range <-> RGB limited/full range, passing the output coefficients from one call as the input for the next.
For example, to create YUV full-range to RGB limited-range coefficients:
[
M_a,O_a] = RGB full-range to RGB limited-range coefficients. [M_b,O_b] = YUV full-range to RGB full-range coefficients.and that output sent as input to create YUV limited-range to RGB limited-range coefficients:
Quoted 7 lines
Quoted 4 lines
[
M_a,O_a] = YUV full-range to RGB limited-range coefficients. [M_b,O_b] = YUV limited-range to YUV full-range coefficients.Signed-off-by Steve Longerbeam <[email protected]>I'm not a big fan of this. Wouldn't it be much easier to compute all necessary task parameter sets offline with high precision, and store the precomputed sets in the compact representation?
I am thinking of when support might be added for the other encoding standards. With this transform function, only two new task parameter structs need to be added, one for yuv-full-to-rgb-full, and one for rgb-full-to-yuv-full. Without transform_coeffs(), four structs would have to be added (adding encoding to and from yuv-limited). And if rgb-limited support is added, it would mean a total of eight new structs for a new encoding standard. But with transform_coeffs(), still only the two structs above are needed, and the function would compute the others automatically in runtime.
Steve
Quoted 333 lines
Quoted 333 lines
--- drivers/gpu/ipu-v3/ipu-ic.c | 281 +++++++++++++++++++++++++++++++++--- 1 file changed, 263 insertions(+), 18 deletions(-)drivers/gpu/ipu-v3/ipu-ic.c +261
-16@@ -178,10 +178,10 @@ static inline void ipu_ic_write(struct ipu_ic *ic, u32 value, unsigned offset) }struct ic_encode_coeff { -s16 coeff[3 ][3 ];/* signed 9-bit integer coefficients */ -s16 offset[3 ];/* signed 11+2-bit fixed point offset */ -u8 scale:2 ;/* scale coefficients * 2^(scale-1) */ -bool sat:1 ;/* saturate to (16, 235(Y) / 240(U, V)) */ +int coeff[3 ][3 ];/* signed 9-bit integer coefficients */ +int offset[3 ];/* signed 13-bit integer offset */ +int scale;/* scale coefficients * 2^(scale-1) */ +bool sat;/* saturate to (16, 235(Y) / 240(U, V)) */ };/* @@ -277,6 +277,231 @@ static const struct ic_encode_coeff ic_encode_ycbcr2rgb_709 = { .scale = 2, }; +/* +* YUV full range to YUV limited range: +* +* Y_lim = 0.8588 * Y_full + 16 +* Cb_lim = 0.8784 * (Cb_full - 128) + 128 +* Cr_lim = 0.8784 * (Cr_full - 128) + 128 +*/ +static const struct ic_encode_coeff ic_encode_ycbcr_full2lim = { + .coeff = { + {219 ,0 ,0 }, + {0 ,224 ,0 }, + {0 ,0 ,224 }, + }, + .offset = {64 ,62 ,62 }, + .scale =1 , +}; + +/* +* YUV limited range to YUV full range: +* +* Y_full = 1.1644 * (Y_lim - 16) +* Cb_full = 1.1384 * (Cb_lim - 128) + 128 +* Cr_full = 1.1384 * (Cr_lim - 128) + 128 +*/ +static const struct ic_encode_coeff ic_encode_ycbcr_lim2full = { + .coeff = { + {149 ,0 ,0 }, + {0 ,145 ,0 }, + {0 ,0 ,145 }, + }, + .offset = {-37 ,-35 ,-35 }, + .scale =2 , +}; + +/* +* RGB full range to RGB limited range: +* +* R_lim = 0.8588 * R_full + 16 +* G_lim = 0.8588 * G_full + 16 +* B_lim = 0.8588 * B_full + 16 +*/ +static const struct ic_encode_coeff +ic_encode_rgb_full2lim__maybe_unused = { + .coeff = { + {220 ,0 ,0 }, + {0 ,220 ,0 }, + {0 ,0 ,220 }, + }, + .offset = {64 ,64 ,64 }, + .scale =1 , +}; + +/* +* RGB limited range to RGB full range: +* +* R_full = 1.1644 * (R_lim - 16) +* G_full = 1.1644 * (G_lim - 16) +* B_full = 1.1644 * (B_lim - 16) +*/ +static const struct ic_encode_coeff +ic_encode_rgb_lim2full__maybe_unused = { + .coeff = { + {149 ,0 ,0 }, + {0 ,149 ,0 }, + {0 ,0 ,149 }, + }, + .offset = {-37 ,-37 ,-37 }, + .scale =2 , +}; + +/* +* Convert a coefficient and scale value in TPMEM register format +* to a signed int times 256 (fix the radix point). The TPMEM register +* coefficient format is a signed 9-bit value (sign bit at bit 8, +* mantissa = coeff * 2 ^ (8 - scale - 1)). +*/ +static int coeff_fix (int coeff,int scale) +{ +if (coeff >=256 ) + coeff -=512 ; +if (scale ==0 ) +return DIV_ROUND_CLOSEST (coeff,2 ); +return coeff << (scale -1 ); +} + +/* +* Convert a signed int coefficient times 256 to TPMEM register +* format, given a scale value = TPMEM scale - 1. +*/ +static int coeff_normalize (int coeff,int scale) +{ + coeff =DIV_ROUND_CLOSEST (coeff,1 << scale); +if (coeff <0 ) + coeff +=512 ; +return coeff; +} + +/* +* Convert an offset and scale value in TPMEM register format to a +* signed int times 256 (fix the radix point). The TPMEM register +* offset format is a signed 13-bit value (sign bit at bit 12, +* mantissa = offset * 2 ^ (2 - (scale - 1)). +*/ +static int offset_fix (int offset,int scale) +{ +return offset << (8 - (2 - (scale -1 ))); +} + +/* +* Convert a signed int offset times 256 to TPMEM register +* format, given a scale value = TPMEM scale - 1. +*/ +static int offset_normalize (int off,int scale) +{ +return DIV_ROUND_CLOSEST (off,1 << (8 - (2 - scale))); +} + +/* +* Find the scale value that fits the given coefficient within +* the 8-bit TPMEM mantissa. +*/ +static int get_coeff_scale (int coeff) +{ +int scale =0 ; + +while (abs (coeff) >=256 && scale <=2 ) { + coeff =DIV_ROUND_CLOSEST (coeff,2 ); + scale++; + } + +return scale; +} + +/* +* The above defined encoding coefficients all encode between +* full-range RGB and full-range YCbCr. +* +* This function calculates a matrix M_c and offset vector O_c, given +* input matrices M_a, M_b and offset vectors O_a, O_b, such that: +* +* M_c = M_a * M_b +* O_c = M_a * O_b + O_a +* +* This operation will transform the full-range coefficients to +* coefficients that encode to or from limited range YCbCr or RGB. +* +* For example, to transform ic_encode_rgb2ycbcr_601 to encode to +* limited-range YCbCr: +* +* [M_a, O_a] = ic_encode_ycbcr_full2lim +* [M_b, O_b] = ic_encode_rgb2ycbcr_601 +* +* To transform the inverse coefficients ic_encode_ycbcr2rgb_601 to +* encode from limited-range YCbCr: +* +* [M_a, O_a] = ic_encode_ycbcr2rgb_601 +* [M_b, O_b] = ic_encode_ycbcr_lim2full +* +* The function can also be used to create RGB limited range +* coefficients, and cascaded to create all combinations of +* encodings between YCbCr limited/full range <-> RGB limited/full +* range. +*/ +static void transform_coeffs (struct ic_encode_coeff *out, +const struct ic_encode_coeff *a, +const struct ic_encode_coeff *b) +{ +int c_a, c_b, c_out; +int o_a, o_b, o_out; +int outscale =0 ; +int i, j, k; + +for (i =0 ; i <3 ; i++) { + o_out =0 ; +for (j =0 ; j <3 ; j++) { +int scale; + +/* M_c[i,j] = M_a[i,k] * M_b[k,j] */ + c_out =0 ; +for (k =0 ; k <3 ; k++) { + c_a =coeff_fix (a->coeff[i][k], a->scale); + c_b =coeff_fix (b->coeff[k][j], b->scale); + c_out += c_a * c_b; + } + + c_out =DIV_ROUND_CLOSEST (c_out,1 <<8 ); + out->coeff[i][j] = c_out; + +/* +* get scale for this coefficient and update +* final output scale. +*/ + scale =get_coeff_scale (c_out); + outscale =max (outscale, scale); + +/* M_a[i,j] * O_b[j] */ + c_a =coeff_fix (a->coeff[i][j], a->scale); + o_b =offset_fix (b->offset[j], b->scale); + o_out +=DIV_ROUND_CLOSEST (c_a * o_b,1 <<8 ); + } + +/* O_c[i] = (M_a * O_b)[i] + O_a[i] */ + o_a =offset_fix (a->offset[i], a->scale); + o_out += o_a; + + out->offset[i] = o_out; + } + +/* +* normalize output coefficients and offsets to TPMEM +* register format. +*/ +for (i =0 ; i <3 ; i++) { +for (j =0 ; j <3 ; j++) { + c_out = out->coeff[i][j]; + out->coeff[i][j] =coeff_normalize (c_out, outscale); + } + + o_out = out->offset[i]; + out->offset[i] =offset_normalize (o_out, outscale); + } + + out->scale = outscale +1 ; +} + static int calc_csc_coeffs(struct ipu_ic_priv *priv, struct ic_encode_coeff *coeff_out, const struct ipu_ic_colorspace *in,@@ -290,14 +515,6 @@ static int calc_csc_coeffs(struct ipu_ic_priv *priv,return -ENOTSUPP ; } -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 &&@@ -307,7 +524,18 @@ static int calc_csc_coeffs(struct ipu_ic_priv *priv, }if (in->cs == out->cs) { - *coeff_out = ic_encode_identity; +if (in->quant == out->quant) { + *coeff_out = ic_encode_identity; + }else if (in->quant ==V4L2_QUANTIZATION_FULL_RANGE ) { +/* YUV full-range to YUV limited-range */ + *coeff_out = ic_encode_ycbcr_full2lim; + +/* set saturation bit for YUV limited-range output */ + coeff_out->sat = true; + }else { +/* YUV limited-range to YUV full-range */ + *coeff_out = ic_encode_ycbcr_lim2full; + }return 0 ; }@@ -328,7 +556,24 @@ static int calc_csc_coeffs(struct ipu_ic_priv *priv,return -ENOTSUPP ; } - *coeff_out = *encode_coeff; +if (in->quant == out->quant) { +/* +* YUV full-range to RGB full-range, or +* RGB full-range to YUV full-range. +*/ + *coeff_out = *encode_coeff; + }else if (inverse_encode) { +/* YUV limited-range to RGB full-range */ +transform_coeffs (coeff_out, encode_coeff, + &ic_encode_ycbcr_lim2full); + }else { +/* RGB full-range to YUV limited-range */ +transform_coeffs (coeff_out, &ic_encode_ycbcr_full2lim, + encode_coeff); + +/* set saturation bit for YUV limited-range output */ + coeff_out->sat = true; + }return 0 ; }@@ -340,9 +585,9 @@ static int init_csc(struct ipu_ic *ic, {struct ipu_ic_priv *priv = ic->priv;struct ic_encode_coeff coeff; +const unsigned int (*c)[3 ]; +const unsigned int *a;u32 __iomem *base; -const u16 (*c)[3 ]; -const u16 *a;u32 param;int ret;@@ -354,8 +599,8 @@ static int init_csc(struct ipu_ic *ic, (priv->tpmem_base + ic->reg->tpmem_csc[csc_index]);
Quoted 8 lines
drivers/gpu/ipu-v3/ipu-ic.c static int init_csc(struct ipu_ic *ic, · line 606
Quoted 8 lines
drivers/gpu/ipu-v3/ipu-ic.c +2
-2/* Cast to unsigned */ - c = (const u16 (*)[3 ])coeff.coeff; - a = (const u16 *)coeff.offset; + c = (const unsigned int (*)[3 ])coeff.coeff; + a = (const unsigned int *)coeff.offset; param = ((a[0 ] &0x1f ) <<27 ) | ((c[0 ][0 ] &0x1ff ) <<18 ) | ((c[1 ][1 ] &0x1ff ) <<9 ) | (c[2 ][2 ] &0x1ff );
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Steve Longerbeam
maintainerPATCH v6 6/7 media: imx: Try colorimetry at both sink and source pads
In reply to Steve LongerbeamThe colorimetry parameters need to be tested at both sink and source pads. Specifically, for prpencvf, the IC only supports RGB full-range quantization at input and output.
Fix this by cleaning up imx_media_fill_default_mbus_fields(), renaming to imx_media_try_colorimetry(), and call it at both sink and source pad try_fmt's. The unrelated check for uninitialized field value is moved out to appropriate places in each subdev try_fmt.
--- drivers/staging/media/imx/imx-ic-prp.c | 6 +- drivers/staging/media/imx/imx-ic-prpencvf.c | 8 +-- drivers/staging/media/imx/imx-media-csi.c | 19 +++--- drivers/staging/media/imx/imx-media-utils.c | 68 +++++++++++---------- drivers/staging/media/imx/imx-media-vdic.c | 5 +- drivers/staging/media/imx/imx-media.h | 5 +- drivers/staging/media/imx/imx7-media-csi.c | 8 +-- 7 files changed, 62 insertions(+), 57 deletions(-)
drivers/staging/media/imx/imx-ic-prp.c +4 -2
@@ -197,8 +197,8 @@ static int prp_set_fmt(struct v4l2_subdev *sd, sdformat->format.code = cc->codes[0 ]; } -imx_media_fill_default_mbus_fields (&sdformat->format, infmt, - true); +if (sdformat->format.field ==V4L2_FIELD_ANY ) + sdformat->format.field =V4L2_FIELD_NONE ;break ;case PRP_SRC_PAD_PRPENC :case PRP_SRC_PAD_PRPVF :
@@ -207,6 +207,8 @@ static int prp_set_fmt(struct v4l2_subdev *sd,break ; } +imx_media_try_colorimetry (&sdformat->format, true); + fmt =__prp_get_fmt (priv, cfg, sdformat->pad, sdformat->which); *fmt = sdformat->format; out:
drivers/staging/media/imx/imx-ic-prpencvf.c +4 -4
@@ -899,8 +899,6 @@ static void prp_try_fmt(struct prp_priv *priv,/* propagate colorimetry from sink */ sdformat->format.colorspace = infmt->colorspace; sdformat->format.xfer_func = infmt->xfer_func; - sdformat->format.quantization = infmt->quantization; - sdformat->format.ycbcr_enc = infmt->ycbcr_enc; }else { v4l_bound_align_image(&sdformat->format.width, MIN_W_SINK, MAX_W_SINK, W_ALIGN_SINK,
@@ -908,9 +906,11 @@ static void prp_try_fmt(struct prp_priv *priv,MIN_H_SINK ,MAX_H_SINK ,H_ALIGN_SINK , S_ALIGN); -imx_media_fill_default_mbus_fields (&sdformat->format, infmt, - true); +if (sdformat->format.field ==V4L2_FIELD_ANY ) + sdformat->format.field =V4L2_FIELD_NONE ; } + +imx_media_try_colorimetry (&sdformat->format, true); } static int prp_set_fmt(struct v4l2_subdev *sd,
drivers/staging/media/imx/imx-media-csi.c +11 -8
@@ -1369,9 +1369,15 @@ static void csi_try_field(struct csi_priv *priv,struct v4l2_mbus_framefmt *infmt =__csi_get_fmt (priv, cfg,CSI_SINK_PAD , sdformat->which); -/* no restrictions on sink pad field type */ -if (sdformat->pad ==CSI_SINK_PAD ) +/* +* no restrictions on sink pad field type except must +* be initialized. +*/ +if (sdformat->pad ==CSI_SINK_PAD ) { +if (sdformat->format.field ==V4L2_FIELD_ANY ) + sdformat->format.field =V4L2_FIELD_NONE ;return ; + }switch (infmt->field) {case V4L2_FIELD_SEQ_TB :
@@ -1449,8 +1455,6 @@ static void csi_try_fmt(struct csi_priv *priv,/* propagate colorimetry from sink */ sdformat->format.colorspace = infmt->colorspace; sdformat->format.xfer_func = infmt->xfer_func; - sdformat->format.quantization = infmt->quantization; - sdformat->format.ycbcr_enc = infmt->ycbcr_enc;break ;case CSI_SINK_PAD :
@@ -1470,10 +1474,6 @@ static void csi_try_fmt(struct csi_priv *priv,csi_try_field (priv, cfg, sdformat); -imx_media_fill_default_mbus_fields ( - &sdformat->format, infmt, - priv->active_output_pad ==CSI_SRC_PAD_DIRECT ); -/* Reset crop and compose rectangles */ crop->left =0 ; crop->top =0 ;
@@ -1489,6 +1489,9 @@ static void csi_try_fmt(struct csi_priv *priv,break ; } + +imx_media_try_colorimetry (&sdformat->format, + priv->active_output_pad ==CSI_SRC_PAD_DIRECT ); } static int csi_set_fmt(struct v4l2_subdev *sd,
drivers/staging/media/imx/imx-media-utils.c +36 -32
100 lines not on this page · raw message
drivers/staging/media/imx/imx-media-vdic.c +2 -3
@@ -615,14 +615,13 @@ static void vdic_try_fmt(struct vdic_priv *priv, &sdformat->format.height, MIN_H, MAX_H_VDIC, H_ALIGN, S_ALIGN); -imx_media_fill_default_mbus_fields (&sdformat->format, infmt, - true); -/* input must be interlaced! Choose SEQ_TB if not */ if (!V4L2_FIELD_HAS_BOTH (sdformat->format.field)) sdformat->format.field =V4L2_FIELD_SEQ_TB ;break ; } + +imx_media_try_colorimetry (&sdformat->format, true); } static int vdic_set_fmt(struct v4l2_subdev *sd,
drivers/staging/media/imx/imx-media.h +2 -3
@@ -176,9 +176,8 @@ int imx_media_init_mbus_fmt(struct v4l2_mbus_framefmt *mbus, const struct imx_media_pixfmt **cc);int imx_media_init_cfg (struct v4l2_subdev *sd,struct v4l2_subdev_pad_config *cfg); -void imx_media_fill_default_mbus_fields (struct v4l2_mbus_framefmt *tryfmt, -struct v4l2_mbus_framefmt *fmt, -bool ic_route); +void imx_media_try_colorimetry (struct v4l2_mbus_framefmt *tryfmt, +bool ic_route); int imx_media_mbus_fmt_to_pix_fmt(struct v4l2_pix_format *pix, struct v4l2_rect *compose, const struct v4l2_mbus_framefmt *mbus,
drivers/staging/media/imx/imx7-media-csi.c +3 -5
@@ -1003,8 +1003,6 @@ static int imx7_csi_try_fmt(struct imx7_csi *csi, sdformat->format.colorspace = in_fmt->colorspace; sdformat->format.xfer_func = in_fmt->xfer_func; - sdformat->format.quantization = in_fmt->quantization; - sdformat->format.ycbcr_enc = in_fmt->ycbcr_enc;break ;case IMX7_CSI_PAD_SINK : *cc = imx_media_find_mbus_format(sdformat->format.code,
@@ -1015,14 +1013,14 @@ static int imx7_csi_try_fmt(struct imx7_csi *csi, false); sdformat->format.code = (*cc)->codes[0 ]; } - -imx_media_fill_default_mbus_fields (&sdformat->format, in_fmt, - false);break ;default :return -EINVAL ;break ; } + +imx_media_try_colorimetry (&sdformat->format, false); +return 0 ; }
Signature · 1 line
2.17.1
Steve Longerbeam
maintainerPATCH v6 7/7 media: imx: Allow BT.709 encoding for IC routes
In reply to Steve LongerbeamThe IC now supports BT.709 Y'CbCr encoding, in addition to existing BT.601 encoding, so allow both, for pipelines that route through the IC.
---
Changes in v5:
- rebased this patch on top of repurposing the function to imx_media_try_colorimetry().
Changes in v2:
- move ic_route check above default colorimetry checks, and fill default colorimetry for ic_route, otherwise it's not possible to set BT.709 encoding for ic routes.
--- drivers/staging/media/imx/imx-media-utils.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/staging/media/imx/imx-media-utils.c +4 -2
@@ -524,7 +524,7 @@ EXPORT_SYMBOL_GPL(imx_media_init_cfg);* If this format is destined to be routed through the Image Converter, * quantization and Y`CbCr encoding must be fixed. The IC supports only * full-range quantization for RGB at its input and output, and only -* BT.601 Y`CbCr encoding. +* BT.601 or Rec.709 Y`CbCr encoding. */ void imx_media_try_colorimetry (struct v4l2_mbus_framefmt *tryfmt,bool ic_route)
@@ -563,7 +563,9 @@ void imx_media_try_colorimetry(struct v4l2_mbus_framefmt *tryfmt, tryfmt->quantization == V4L2_QUANTIZATION_DEFAULT) tryfmt->quantization = V4L2_QUANTIZATION_FULL_RANGE; - tryfmt->ycbcr_enc =V4L2_YCBCR_ENC_601 ; +if (tryfmt->ycbcr_enc !=V4L2_YCBCR_ENC_601 && + tryfmt->ycbcr_enc !=V4L2_YCBCR_ENC_709 ) + tryfmt->ycbcr_enc =V4L2_YCBCR_ENC_601 ; }else {if (tryfmt->ycbcr_enc ==V4L2_YCBCR_ENC_DEFAULT ) { tryfmt->ycbcr_enc =
Signature · 1 line
2.17.1