| Message ID | 20200819205336.fce24lioz34vbcd2@smtp.gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/vkms: add alpha-premultiplied color blending | expand |
On Wed, 19 Aug 2020 17:53:36 -0300 Melissa Wen <melissa.srw@gmail.com> wrote: > The current VKMS blend function ignores alpha channel and just overwrites > vaddr_src with vaddr_dst. This XRGB approach triggers a warning when > running the kms_cursor_crc/cursor-alpha-transparent test case. In IGT > tests, cairo_format_argb32 uses premultiplied alpha (according to > documentation), so this patch considers premultiplied alpha colors to > compose vaddr_src with vaddr_dst. > > This change removes the following cursor-alpha-transparent warning: > Suspicious CRC: All values are 0. > > Cc: Daniel Vetter <daniel@ffwll.ch> > Cc: Rodrigo Siqueira <rodrigosiqueiramelo@gmail.com> > Cc: Haneen Mohammed <hamohammed.sa@gmail.com> > > Signed-off-by: Melissa Wen <melissa.srw@gmail.com> > --- > drivers/gpu/drm/vkms/vkms_composer.c | 43 +++++++++++++++++++++------- > 1 file changed, 33 insertions(+), 10 deletions(-) > > diff --git a/drivers/gpu/drm/vkms/vkms_composer.c b/drivers/gpu/drm/vkms/vkms_composer.c > index 4f3b07a32b60..6aac962d3e2e 100644 > --- a/drivers/gpu/drm/vkms/vkms_composer.c > +++ b/drivers/gpu/drm/vkms/vkms_composer.c > @@ -32,8 +32,6 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) > src_offset = composer->offset > + (i * composer->pitch) > + (j * composer->cpp); > - /* XRGB format ignores Alpha channel */ > - bitmap_clear(vaddr_out + src_offset, 24, 8); > crc = crc32_le(crc, vaddr_out + src_offset, > sizeof(u32)); > } > @@ -42,6 +40,32 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) > return crc; > } > > +u8 blend_channel(u8 c_src, u8 c_dst, u8 a_src) > +{ > + u32 pre_blend; > + u8 new_color; > + > + /* Premultiplied alpha blending - IGT + cairo context */ > + pre_blend = (c_src * 255 + c_dst * (255 - a_src)); > + > + /* Faster div by 255 */ > + new_color = ((pre_blend + ((pre_blend + 257) >> 8)) >> 8); > + > + return new_color; > +} > + > +void alpha_blending(u8 *argb_src, u8 *argb_dst) > +{ > + u8 a_src; > + > + a_src = argb_src[3]; Hi, DRM pixel formats are often defined as "bits in a 32-bit word", but here you are accessing it as an array of bytes. To me that looks suspicious wrt. big-endian architectures. Unfortunately I have again forgot how DRM pixel formats should be interpreted on a big-endian machine, if I ever even understood it, so I can't say if this is right or not. Thanks, pq > + argb_dst[0] = blend_channel(argb_src[0], argb_dst[0], a_src); > + argb_dst[1] = blend_channel(argb_src[1], argb_dst[1], a_src); > + argb_dst[2] = blend_channel(argb_src[2], argb_dst[2], a_src); > + /* Opaque primary */ > + argb_dst[3] = 0xFF; > +} > + > /** > * blend - blend value at vaddr_src with value at vaddr_dst > * @vaddr_dst: destination address > @@ -50,12 +74,9 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) > * @src_composer: source framebuffer's metadata > * > * Blend value at vaddr_src with value at vaddr_dst. > - * Currently, this function write value of vaddr_src on value > - * at vaddr_dst using buffer's metadata to locate the new values > - * from vaddr_src and their destination at vaddr_dst. > - * > - * TODO: Use the alpha value to blend vaddr_src with vaddr_dst > - * instead of overwriting it. > + * Currently, this function considers premultiplied alpha for blending, as used > + * by Cairo. It uses buffer's metadata to locate the new composite values at > + * vaddr_dst. > */ > static void blend(void *vaddr_dst, void *vaddr_src, > struct vkms_composer *dest_composer, > @@ -63,6 +84,7 @@ static void blend(void *vaddr_dst, void *vaddr_src, > { > int i, j, j_dst, i_dst; > int offset_src, offset_dst; > + u8 *p_dst, *p_src; > > int x_src = src_composer->src.x1 >> 16; > int y_src = src_composer->src.y1 >> 16; > @@ -84,8 +106,9 @@ static void blend(void *vaddr_dst, void *vaddr_src, > + (i * src_composer->pitch) > + (j * src_composer->cpp); > > - memcpy(vaddr_dst + offset_dst, > - vaddr_src + offset_src, sizeof(u32)); > + p_src = (u8 *)(vaddr_src + offset_src); > + p_dst = (u8 *)(vaddr_dst + offset_dst); > + alpha_blending(p_src, p_dst); > } > i_dst++; > }
Hi Melissa, First of all, thanks a lot for your patch! Follows my inline comments. On 08/19, Melissa Wen wrote: > The current VKMS blend function ignores alpha channel and just overwrites > vaddr_src with vaddr_dst. This XRGB approach triggers a warning when > running the kms_cursor_crc/cursor-alpha-transparent test case. In IGT > tests, cairo_format_argb32 uses premultiplied alpha (according to > documentation), so this patch considers premultiplied alpha colors to > compose vaddr_src with vaddr_dst. > > This change removes the following cursor-alpha-transparent warning: > Suspicious CRC: All values are 0. > > Cc: Daniel Vetter <daniel@ffwll.ch> > Cc: Rodrigo Siqueira <rodrigosiqueiramelo@gmail.com> > Cc: Haneen Mohammed <hamohammed.sa@gmail.com> > > Signed-off-by: Melissa Wen <melissa.srw@gmail.com> > --- > drivers/gpu/drm/vkms/vkms_composer.c | 43 +++++++++++++++++++++------- > 1 file changed, 33 insertions(+), 10 deletions(-) > > diff --git a/drivers/gpu/drm/vkms/vkms_composer.c b/drivers/gpu/drm/vkms/vkms_composer.c > index 4f3b07a32b60..6aac962d3e2e 100644 > --- a/drivers/gpu/drm/vkms/vkms_composer.c > +++ b/drivers/gpu/drm/vkms/vkms_composer.c > @@ -32,8 +32,6 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) > src_offset = composer->offset > + (i * composer->pitch) > + (j * composer->cpp); > - /* XRGB format ignores Alpha channel */ > - bitmap_clear(vaddr_out + src_offset, 24, 8); > crc = crc32_le(crc, vaddr_out + src_offset, > sizeof(u32)); > } > @@ -42,6 +40,32 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) > return crc; > } > > +u8 blend_channel(u8 c_src, u8 c_dst, u8 a_src) Use static here. Also, replace c_src to src, c_dst to dst, and a_src to alpha. > +{ > + u32 pre_blend; > + u8 new_color; > + > + /* Premultiplied alpha blending - IGT + cairo context */ You can drop the part that says "IGT + cairo context", this explanation better suit the commit message. > + pre_blend = (c_src * 255 + c_dst * (255 - a_src)); > + > + /* Faster div by 255 */ > + new_color = ((pre_blend + ((pre_blend + 257) >> 8)) >> 8); > + > + return new_color; > +} > + > +void alpha_blending(u8 *argb_src, u8 *argb_dst) Use static. Looks like that argb_src is a read-only variable, in this sense add const. > +{ > + u8 a_src; > + > + a_src = argb_src[3]; change a_src to alpha. > + argb_dst[0] = blend_channel(argb_src[0], argb_dst[0], a_src); > + argb_dst[1] = blend_channel(argb_src[1], argb_dst[1], a_src); > + argb_dst[2] = blend_channel(argb_src[2], argb_dst[2], a_src); > + /* Opaque primary */ > + argb_dst[3] = 0xFF; > +} > + > /** > * blend - blend value at vaddr_src with value at vaddr_dst > * @vaddr_dst: destination address > @@ -50,12 +74,9 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) > * @src_composer: source framebuffer's metadata > * > * Blend value at vaddr_src with value at vaddr_dst. > - * Currently, this function write value of vaddr_src on value > - * at vaddr_dst using buffer's metadata to locate the new values > - * from vaddr_src and their destination at vaddr_dst. > - * > - * TODO: Use the alpha value to blend vaddr_src with vaddr_dst > - * instead of overwriting it. > + * Currently, this function considers premultiplied alpha for blending, as used > + * by Cairo. It uses buffer's metadata to locate the new composite values at > + * vaddr_dst. > */ > static void blend(void *vaddr_dst, void *vaddr_src, > struct vkms_composer *dest_composer, > @@ -63,6 +84,7 @@ static void blend(void *vaddr_dst, void *vaddr_src, > { > int i, j, j_dst, i_dst; > int offset_src, offset_dst; > + u8 *p_dst, *p_src; I suppose that p stands for "pixel", right? In this case how about use pixel? Best Regards Rodrigo Siqueira > > int x_src = src_composer->src.x1 >> 16; > int y_src = src_composer->src.y1 >> 16; > @@ -84,8 +106,9 @@ static void blend(void *vaddr_dst, void *vaddr_src, > + (i * src_composer->pitch) > + (j * src_composer->cpp); > > - memcpy(vaddr_dst + offset_dst, > - vaddr_src + offset_src, sizeof(u32)); > + p_src = (u8 *)(vaddr_src + offset_src); > + p_dst = (u8 *)(vaddr_dst + offset_dst); > + alpha_blending(p_src, p_dst); > } > i_dst++; > } > -- > 2.28.0 >
Hi Rodrigo, Thanks for the review! I just sent a v2 applying your suggestions. Best Regards, Melissa Wen Melissa Wen On Tue, Aug 25, 2020 at 12:04 AM Rodrigo Siqueira <rodrigosiqueiramelo@gmail.com> wrote: > > Hi Melissa, > > First of all, thanks a lot for your patch! > > Follows my inline comments. > > On 08/19, Melissa Wen wrote: > > The current VKMS blend function ignores alpha channel and just overwrites > > vaddr_src with vaddr_dst. This XRGB approach triggers a warning when > > running the kms_cursor_crc/cursor-alpha-transparent test case. In IGT > > tests, cairo_format_argb32 uses premultiplied alpha (according to > > documentation), so this patch considers premultiplied alpha colors to > > compose vaddr_src with vaddr_dst. > > > > This change removes the following cursor-alpha-transparent warning: > > Suspicious CRC: All values are 0. > > > > Cc: Daniel Vetter <daniel@ffwll.ch> > > Cc: Rodrigo Siqueira <rodrigosiqueiramelo@gmail.com> > > Cc: Haneen Mohammed <hamohammed.sa@gmail.com> > > > > Signed-off-by: Melissa Wen <melissa.srw@gmail.com> > > --- > > drivers/gpu/drm/vkms/vkms_composer.c | 43 +++++++++++++++++++++------- > > 1 file changed, 33 insertions(+), 10 deletions(-) > > > > diff --git a/drivers/gpu/drm/vkms/vkms_composer.c b/drivers/gpu/drm/vkms/vkms_composer.c > > index 4f3b07a32b60..6aac962d3e2e 100644 > > --- a/drivers/gpu/drm/vkms/vkms_composer.c > > +++ b/drivers/gpu/drm/vkms/vkms_composer.c > > @@ -32,8 +32,6 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) > > src_offset = composer->offset > > + (i * composer->pitch) > > + (j * composer->cpp); > > - /* XRGB format ignores Alpha channel */ > > - bitmap_clear(vaddr_out + src_offset, 24, 8); > > crc = crc32_le(crc, vaddr_out + src_offset, > > sizeof(u32)); > > } > > @@ -42,6 +40,32 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) > > return crc; > > } > > > > +u8 blend_channel(u8 c_src, u8 c_dst, u8 a_src) > > Use static here. > > Also, replace c_src to src, c_dst to dst, and a_src to alpha. > > > +{ > > + u32 pre_blend; > > + u8 new_color; > > + > > + /* Premultiplied alpha blending - IGT + cairo context */ > > You can drop the part that says "IGT + cairo context", this explanation > better suit the commit message. > > > + pre_blend = (c_src * 255 + c_dst * (255 - a_src)); > > + > > + /* Faster div by 255 */ > > + new_color = ((pre_blend + ((pre_blend + 257) >> 8)) >> 8); > > + > > + return new_color; > > +} > > + > > +void alpha_blending(u8 *argb_src, u8 *argb_dst) > > Use static. > > Looks like that argb_src is a read-only variable, in this sense add > const. > > > +{ > > + u8 a_src; > > + > > + a_src = argb_src[3]; > > change a_src to alpha. > > > + argb_dst[0] = blend_channel(argb_src[0], argb_dst[0], a_src); > > + argb_dst[1] = blend_channel(argb_src[1], argb_dst[1], a_src); > > + argb_dst[2] = blend_channel(argb_src[2], argb_dst[2], a_src); > > + /* Opaque primary */ > > + argb_dst[3] = 0xFF; > > +} > > + > > /** > > * blend - blend value at vaddr_src with value at vaddr_dst > > * @vaddr_dst: destination address > > @@ -50,12 +74,9 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) > > * @src_composer: source framebuffer's metadata > > * > > * Blend value at vaddr_src with value at vaddr_dst. > > - * Currently, this function write value of vaddr_src on value > > - * at vaddr_dst using buffer's metadata to locate the new values > > - * from vaddr_src and their destination at vaddr_dst. > > - * > > - * TODO: Use the alpha value to blend vaddr_src with vaddr_dst > > - * instead of overwriting it. > > + * Currently, this function considers premultiplied alpha for blending, as used > > + * by Cairo. It uses buffer's metadata to locate the new composite values at > > + * vaddr_dst. > > */ > > static void blend(void *vaddr_dst, void *vaddr_src, > > struct vkms_composer *dest_composer, > > @@ -63,6 +84,7 @@ static void blend(void *vaddr_dst, void *vaddr_src, > > { > > int i, j, j_dst, i_dst; > > int offset_src, offset_dst; > > + u8 *p_dst, *p_src; > > I suppose that p stands for "pixel", right? In this case how about use > pixel? > > Best Regards > Rodrigo Siqueira > > > > > int x_src = src_composer->src.x1 >> 16; > > int y_src = src_composer->src.y1 >> 16; > > @@ -84,8 +106,9 @@ static void blend(void *vaddr_dst, void *vaddr_src, > > + (i * src_composer->pitch) > > + (j * src_composer->cpp); > > > > - memcpy(vaddr_dst + offset_dst, > > - vaddr_src + offset_src, sizeof(u32)); > > + p_src = (u8 *)(vaddr_src + offset_src); > > + p_dst = (u8 *)(vaddr_dst + offset_dst); > > + alpha_blending(p_src, p_dst); > > } > > i_dst++; > > } > > -- > > 2.28.0 > > > > -- > Rodrigo Siqueira > https://siqueira.tech
diff --git a/drivers/gpu/drm/vkms/vkms_composer.c b/drivers/gpu/drm/vkms/vkms_composer.c index 4f3b07a32b60..6aac962d3e2e 100644 --- a/drivers/gpu/drm/vkms/vkms_composer.c +++ b/drivers/gpu/drm/vkms/vkms_composer.c @@ -32,8 +32,6 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) src_offset = composer->offset + (i * composer->pitch) + (j * composer->cpp); - /* XRGB format ignores Alpha channel */ - bitmap_clear(vaddr_out + src_offset, 24, 8); crc = crc32_le(crc, vaddr_out + src_offset, sizeof(u32)); } @@ -42,6 +40,32 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) return crc; } +u8 blend_channel(u8 c_src, u8 c_dst, u8 a_src) +{ + u32 pre_blend; + u8 new_color; + + /* Premultiplied alpha blending - IGT + cairo context */ + pre_blend = (c_src * 255 + c_dst * (255 - a_src)); + + /* Faster div by 255 */ + new_color = ((pre_blend + ((pre_blend + 257) >> 8)) >> 8); + + return new_color; +} + +void alpha_blending(u8 *argb_src, u8 *argb_dst) +{ + u8 a_src; + + a_src = argb_src[3]; + argb_dst[0] = blend_channel(argb_src[0], argb_dst[0], a_src); + argb_dst[1] = blend_channel(argb_src[1], argb_dst[1], a_src); + argb_dst[2] = blend_channel(argb_src[2], argb_dst[2], a_src); + /* Opaque primary */ + argb_dst[3] = 0xFF; +} + /** * blend - blend value at vaddr_src with value at vaddr_dst * @vaddr_dst: destination address @@ -50,12 +74,9 @@ static uint32_t compute_crc(void *vaddr_out, struct vkms_composer *composer) * @src_composer: source framebuffer's metadata * * Blend value at vaddr_src with value at vaddr_dst. - * Currently, this function write value of vaddr_src on value - * at vaddr_dst using buffer's metadata to locate the new values - * from vaddr_src and their destination at vaddr_dst. - * - * TODO: Use the alpha value to blend vaddr_src with vaddr_dst - * instead of overwriting it. + * Currently, this function considers premultiplied alpha for blending, as used + * by Cairo. It uses buffer's metadata to locate the new composite values at + * vaddr_dst. */ static void blend(void *vaddr_dst, void *vaddr_src, struct vkms_composer *dest_composer, @@ -63,6 +84,7 @@ static void blend(void *vaddr_dst, void *vaddr_src, { int i, j, j_dst, i_dst; int offset_src, offset_dst; + u8 *p_dst, *p_src; int x_src = src_composer->src.x1 >> 16; int y_src = src_composer->src.y1 >> 16; @@ -84,8 +106,9 @@ static void blend(void *vaddr_dst, void *vaddr_src, + (i * src_composer->pitch) + (j * src_composer->cpp); - memcpy(vaddr_dst + offset_dst, - vaddr_src + offset_src, sizeof(u32)); + p_src = (u8 *)(vaddr_src + offset_src); + p_dst = (u8 *)(vaddr_dst + offset_dst); + alpha_blending(p_src, p_dst); } i_dst++; }
The current VKMS blend function ignores alpha channel and just overwrites vaddr_src with vaddr_dst. This XRGB approach triggers a warning when running the kms_cursor_crc/cursor-alpha-transparent test case. In IGT tests, cairo_format_argb32 uses premultiplied alpha (according to documentation), so this patch considers premultiplied alpha colors to compose vaddr_src with vaddr_dst. This change removes the following cursor-alpha-transparent warning: Suspicious CRC: All values are 0. Cc: Daniel Vetter <daniel@ffwll.ch> Cc: Rodrigo Siqueira <rodrigosiqueiramelo@gmail.com> Cc: Haneen Mohammed <hamohammed.sa@gmail.com> Signed-off-by: Melissa Wen <melissa.srw@gmail.com> --- drivers/gpu/drm/vkms/vkms_composer.c | 43 +++++++++++++++++++++------- 1 file changed, 33 insertions(+), 10 deletions(-)