Skip to contentSkip to conversation

drm: vop2: Do not divide height twice for YUV

PATCH Applied Detlev Casanova 3 participants 3 messages Apr 14, 2024 – Apr 19, 2024

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm: vop2: Do not divide height twice for YUV

For the cbcr format, gt2 and gt4 are computed again after src_h has been divided by vsub.

As src_h as already been divided by 2 before, introduce cbcr_src_h and cbcr_src_w to keep a copy of those values to be used for cbcr gt2 and gt4 computation.

This fixes yuv planes being unaligned vertically when down scaling to 1080 pixels from 2160.

Fixes 604be85547ce ("drm/rockchip: Add VOP2 driver")
---
 drivers/gpu/drm/rockchip/rockchip_drm_vop2.c | 22 +++++++++++---------
 1 file changed, 12 insertions(+), 10 deletions(-)
drivers/gpu/drm/rockchip/rockchip_drm_vop2.c +12 -10
index fdd768bbd487c..62ebbdb16253d 100644
@@ -706,6 +706,8 @@ static void vop2_setup_scale(struct vop2 *vop2, const struct vop2_win *win,
 	const struct drm_format_info *info;
 	u16 hor_scl_mode, ver_scl_mode;
 	u16 hscl_filter_mode, vscl_filter_mode;
+	uint16_t cbcr_src_w = src_w;
+	uint16_t cbcr_src_h = src_h;
 	u8 gt2 = 0;
 	u8 gt4 = 0;
 	u32 val;
@@ -763,27 +765,27 @@ static void vop2_setup_scale(struct vop2 *vop2, const struct vop2_win *win,
 	vop2_win_write(win, VOP2_WIN_YRGB_VSCL_FILTER_MODE, vscl_filter_mode);
 
 	if (info->is_yuv) {
-		src_w /= info->hsub;
-		src_h /= info->vsub;
+		cbcr_src_w /= info->hsub;
+		cbcr_src_h /= info->vsub;
 
 		gt4 = 0;
 		gt2 = 0;
 
-		if (src_h >= (4 * dst_h)) {
+		if (cbcr_src_h >= (4 * dst_h)) {
 			gt4 = 1;
-			src_h >>= 2;
-		} else if (src_h >= (2 * dst_h)) {
+			cbcr_src_h >>= 2;
+		} else if (cbcr_src_h >= (2 * dst_h)) {
 			gt2 = 1;
-			src_h >>= 1;
+			cbcr_src_h >>= 1;
 		}
 
-		hor_scl_mode = scl_get_scl_mode(src_w, dst_w);
-		ver_scl_mode = scl_get_scl_mode(src_h, dst_h);
+		hor_scl_mode = scl_get_scl_mode(cbcr_src_w, dst_w);
+		ver_scl_mode = scl_get_scl_mode(cbcr_src_h, dst_h);
 
-		val = vop2_scale_factor(src_w, dst_w);
+		val = vop2_scale_factor(cbcr_src_w, dst_w);
 		vop2_win_write(win, VOP2_WIN_SCALE_CBCR_X, val);
 
-		val = vop2_scale_factor(src_h, dst_h);
+		val = vop2_scale_factor(cbcr_src_h, dst_h);
 		vop2_win_write(win, VOP2_WIN_SCALE_CBCR_Y, val);
 
 		vop2_win_write(win, VOP2_WIN_VSD_CBCR_GT4, gt4);
Signature · 1 line
2.43.2
Acked-by on v1 1/1
#1 lore raw

Hi,

Quoted 7 lines · Detlev Casanova

For the cbcr format, gt2 and gt4 are computed again after src_h has been divided by vsub.

As src_h as already been divided by 2 before, introduce cbcr_src_h and cbcr_src_w to keep a copy of those values to be used for cbcr gt2 and gt4 computation.

Quoted 5 lines

This fixes yuv planes being unaligned vertically when down scaling to 1080 pixels from 2160.

Fixes 604be85547ce ("drm/rockchip: Add VOP2 driver")
Quoted 45 lines
---
  drivers/gpu/drm/rockchip/rockchip_drm_vop2.c | 22 +++++++++++---------
  1 file changed, 12 insertions(+), 10 deletions(-)
drivers/gpu/drm/rockchip/rockchip_drm_vop2.c +10 -8
index fdd768bbd487c..62ebbdb16253d 100644
@@ -706,6 +706,8 @@ static void vop2_setup_scale(struct vop2 *vop2, const struct vop2_win *win,
  	const struct drm_format_info *info;
  	u16 hor_scl_mode, ver_scl_mode;
  	u16 hscl_filter_mode, vscl_filter_mode;
+	uint16_t cbcr_src_w = src_w;
+	uint16_t cbcr_src_h = src_h;
  	u8 gt2 = 0;
  	u8 gt4 = 0;
  	u32 val;
@@ -763,27 +765,27 @@ static void vop2_setup_scale(struct vop2 *vop2, const struct vop2_win *win,
  	vop2_win_write(win, VOP2_WIN_YRGB_VSCL_FILTER_MODE, vscl_filter_mode);
 
  	if (info->is_yuv) {
-		src_w /= info->hsub;
-		src_h /= info->vsub;
+		cbcr_src_w /= info->hsub;
+		cbcr_src_h /= info->vsub;
 
  		gt4 = 0;
  		gt2 = 0;
 
-		if (src_h >= (4 * dst_h)) {
+		if (cbcr_src_h >= (4 * dst_h)) {
  			gt4 = 1;
-			src_h >>= 2;
-		} else if (src_h >= (2 * dst_h)) {
+			cbcr_src_h >>= 2;
+		} else if (cbcr_src_h >= (2 * dst_h)) {
  			gt2 = 1;
-			src_h >>= 1;
+			cbcr_src_h >>= 1;
  		}
 
-		hor_scl_mode = scl_get_scl_mode(src_w, dst_w);
-		ver_scl_mode = scl_get_scl_mode(src_h, dst_h);
+		hor_scl_mode = scl_get_scl_mode(cbcr_src_w, dst_w);
+		ver_scl_mode = scl_get_scl_mode(cbcr_src_h, dst_h);

drivers/gpu/drm/rockchip/rockchip_drm_vop2.c static void vop2_setup_scale(struct vop2 *vop2, const struct vop2_win *win, · line 791

Quoted 10 lines
drivers/gpu/drm/rockchip/rockchip_drm_vop2.c +2 -2
 
-		val = vop2_scale_factor(src_w, dst_w);
+		val = vop2_scale_factor(cbcr_src_w, dst_w);
  		vop2_win_write(win, VOP2_WIN_SCALE_CBCR_X, val);
 
-		val = vop2_scale_factor(src_h, dst_h);
+		val = vop2_scale_factor(cbcr_src_h, dst_h);
  		vop2_win_write(win, VOP2_WIN_SCALE_CBCR_Y, val);
 
  		vop2_win_write(win, VOP2_WIN_VSD_CBCR_GT4, gt4);

_______________________________________________
Linux-rockchip mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-rockchip

maintainer Applied
#2 lore raw
In reply to Detlev Casanova
Quoted 8 lines · Detlev Casanova

For the cbcr format, gt2 and gt4 are computed again after src_h has been divided by vsub.

As src_h as already been divided by 2 before, introduce cbcr_src_h and cbcr_src_w to keep a copy of those values to be used for cbcr gt2 and gt4 computation.

[...]

Applied, thanks!

[1/1] drm: vop2: Do not divide height twice for YUV

      commit: e80c219f52861e756181d7f88b0d341116daac2b

But changed the subject to

    drm/rockchip: vop2: .....

Best regards,

Signature · 1 line
Heiko Stuebner <[email protected]>