From: Ian Armstrong Date: Sat, 3 Feb 2007 09:37:25 +0000 (-0300) Subject: V4L/DVB (5184): Add cx23415 decoder register documentation X-Git-Tag: v2.6.21-rc2~66^2~50 X-Git-Url: http://pilppa.com/gitweb/?a=commitdiff_plain;h=32ec5332f987435d42371c1c47e310c9cc211cf7;p=linux-2.6-omap-h63xx.git V4L/DVB (5184): Add cx23415 decoder register documentation Many thanks to Ian Armstrong for figuring out what all these registers do. Signed-off-by: Ian Armstrong Signed-off-by: Hans Verkuil Signed-off-by: Mauro Carvalho Chehab --- diff --git a/Documentation/video4linux/cx2341x/fw-decoder-regs.txt b/Documentation/video4linux/cx2341x/fw-decoder-regs.txt new file mode 100644 index 00000000000..db2366c634e --- /dev/null +++ b/Documentation/video4linux/cx2341x/fw-decoder-regs.txt @@ -0,0 +1,815 @@ +PVR350 Video decoder registers 0x02002800 -> 0x02002B00 +======================================================= + +This list has been worked out through trial and error. There will be mistakes +and omissions. Some registers have no obvious effect so it's hard to say what +they do, while others interact with each other, or require a certain load +sequence. Horizontal filter setup is one example, with six registers working +in unison and requiring a certain load sequence to correctly configure. The +indexed colour palette is much easier to set at just two registers, but again +it requires a certain load sequence. + +Some registers are fussy about what they are set to. Load in a bad value & the +decoder will fail. A firmware reload will often recover, but sometimes a reset +is required. For registers containing size information, setting them to 0 is +generally a bad idea. For other control registers i.e. 2878, you'll only find +out what values are bad when it hangs. + +-------------------------------------------------------------------------------- +2800 + bit 0 + Decoder enable + 0 = disable + 1 = enable +-------------------------------------------------------------------------------- +2804 + bits 0:31 + Decoder horizontal Y alias register 1 +--------------- +2808 + bits 0:31 + Decoder horizontal Y alias register 2 +--------------- +280C + bits 0:31 + Decoder horizontal Y alias register 3 +--------------- +2810 + bits 0:31 + Decoder horizontal Y alias register 4 +--------------- +2814 + bits 0:31 + Decoder horizontal Y alias register 5 +--------------- +2818 + bits 0:31 + Decoder horizontal Y alias trigger + + These six registers control the horizontal aliasing filter for the Y plane. + The first five registers must all be loaded before accessing the trigger + (2818), as this register actually clocks the data through for the first + five. + + To correctly program set the filter, this whole procedure must be done 16 + times. The actual register contents are copied from a lookup-table in the + firmware which contains 4 different filter settings. + +-------------------------------------------------------------------------------- +281C + bits 0:31 + Decoder horizontal UV alias register 1 +--------------- +2820 + bits 0:31 + Decoder horizontal UV alias register 2 +--------------- +2824 + bits 0:31 + Decoder horizontal UV alias register 3 +--------------- +2828 + bits 0:31 + Decoder horizontal UV alias register 4 +--------------- +282C + bits 0:31 + Decoder horizontal UV alias register 5 +--------------- +2830 + bits 0:31 + Decoder horizontal UV alias trigger + + These six registers control the horizontal aliasing for the UV plane. + Operation is the same as the Y filter, with 2830 being the trigger + register. + +-------------------------------------------------------------------------------- +2834 + bits 0:15 + Decoder Y source width in pixels + + bits 16:31 + Decoder Y destination width in pixels +--------------- +2838 + bits 0:15 + Decoder UV source width in pixels + + bits 16:31 + Decoder UV destination width in pixels + + NOTE: For both registers, the resulting image must be fully visible on + screen. If the image exceeds the right edge both the source and destination + size must be adjusted to reflect the visible portion. For the source width, + you must take into account the scaling when calculating the new value. +-------------------------------------------------------------------------------- + +283C + bits 0:31 + Decoder Y horizontal scaling + Normally = Reg 2854 >> 2 +--------------- +2840 + bits 0:31 + Decoder ?? unknown - horizontal scaling + Usually 0x00080514 +--------------- +2844 + bits 0:31 + Decoder UV horizontal scaling + Normally = Reg 2854 >> 2 +--------------- +2848 + bits 0:31 + Decoder ?? unknown - horizontal scaling + Usually 0x00100514 +--------------- +284C + bits 0:31 + Decoder ?? unknown - Y plane + Usually 0x00200020 +--------------- +2850 + bits 0:31 + Decoder ?? unknown - UV plane + Usually 0x00200020 +--------------- +2854 + bits 0:31 + Decoder 'master' value for horizontal scaling +--------------- +2858 + bits 0:31 + Decoder ?? unknown + Usually 0 +--------------- +285C + bits 0:31 + Decoder ?? unknown + Normally = Reg 2854 >> 1 +--------------- +2860 + bits 0:31 + Decoder ?? unknown + Usually 0 +--------------- +2864 + bits 0:31 + Decoder ?? unknown + Normally = Reg 2854 >> 1 +--------------- +2868 + bits 0:31 + Decoder ?? unknown + Usually 0 + + Most of these registers either control horizontal scaling, or appear linked + to it in some way. Register 2854 contains the 'master' value & the other + registers can be calculated from that one. You must also remember to + correctly set the divider in Reg 2874. + + To enlarge: + Reg 2854 = (source_width * 0x00200000) / destination_width + Reg 2874 = No divide + + To reduce from full size down to half size: + Reg 2854 = (source_width/2 * 0x00200000) / destination width + Reg 2874 = Divide by 2 + + To reduce from half size down to quarter size: + Reg 2854 = (source_width/4 * 0x00200000) / destination width + Reg 2874 = Divide by 4 + + The result is always rounded up. + +-------------------------------------------------------------------------------- +286C + bits 0:15 + Decoder horizontal Y buffer offset + + bits 15:31 + Decoder horizontal UV buffer offset + + Offset into the video image buffer. If the offset is gradually incremented, + the on screen image will move left & wrap around higher up on the right. + +-------------------------------------------------------------------------------- +2870 + bits 0:15 + Decoder horizontal Y output offset + + bits 16:31 + Decoder horizontal UV output offset + + Offsets the actual video output. Controls output alignment of the Y & UV + planes. The higher the value, the greater the shift to the left. Use + reg 2890 to move the image right. + +-------------------------------------------------------------------------------- +2874 + bits 0:1 + Decoder horizontal Y output size divider + 00 = No divide + 01 = Divide by 2 + 10 = Divide by 3 + + bits 4:5 + Decoder horizontal UV output size divider + 00 = No divide + 01 = Divide by 2 + 10 = Divide by 3 + + bit 8 + Decoder ?? unknown + 0 = Normal + 1 = Affects video output levels + + bit 16 + Decoder ?? unknown + 0 = Normal + 1 = Disable horizontal filter + +-------------------------------------------------------------------------------- +2878 + bit 0 + ?? unknown + + bit 1 + osd on/off + 0 = osd off + 1 = osd on + + bit 2 + Decoder + osd video timing + 0 = NTSC + 1 = PAL + + bits 3:4 + ?? unknown + + bit 5 + Decoder + osd + Swaps upper & lower fields + +-------------------------------------------------------------------------------- +287C + bits 0:10 + Decoder & osd ?? unknown + Moves entire screen horizontally. Starts at 0x005 with the screen + shifted heavily to the right. Incrementing in steps of 0x004 will + gradually shift the screen to the left. + + bits 11:31 + ?? unknown + + Normally contents are 0x00101111 (NTSC) or 0x1010111d (PAL) + +-------------------------------------------------------------------------------- +2880 -------- ?? unknown +2884 -------- ?? unknown +-------------------------------------------------------------------------------- +2888 + bit 0 + Decoder + osd ?? unknown + 0 = Normal + 1 = Misaligned fields (Correctable through 289C & 28A4) + + bit 4 + ?? unknown + + bit 8 + ?? unknown + + Warning: Bad values will require a firmware reload to recover. + Known to be bad are 0x000,0x011,0x100,0x111 +-------------------------------------------------------------------------------- +288C + bits 0:15 + osd ?? unknown + Appears to affect the osd position stability. The higher the value the + more unstable it becomes. Decoder output remains stable. + + bits 16:31 + osd ?? unknown + Same as bits 0:15 + +-------------------------------------------------------------------------------- +2890 + bits 0:11 + Decoder output horizontal offset. + + Horizontal offset moves the video image right. A small left shift is + possible, but it's better to use reg 2870 for that due to its greater + range. + + NOTE: Video corruption will occur if video window is shifted off the right + edge. To avoid this read the notes for 2834 & 2838. +-------------------------------------------------------------------------------- +2894 + bits 0:23 + Decoder output video surround colour. + + Contains the colour (in yuv) used to fill the screen when the video is + running in a window. +-------------------------------------------------------------------------------- +2898 + bits 0:23 + Decoder video window colour + Contains the colour (in yuv) used to fill the video window when the + video is turned off. + + bit 24 + Decoder video output + 0 = Video on + 1 = Video off + + bit 28 + Decoder plane order + 0 = Y,UV + 1 = UV,Y + + bit 29 + Decoder second plane byte order + 0 = Normal (UV) + 1 = Swapped (VU) + + In normal usage, the first plane is Y & the second plane is UV. Though the + order of the planes can be swapped, only the byte order of the second plane + can be swapped. This isn't much use for the Y plane, but can be useful for + the UV plane. + +-------------------------------------------------------------------------------- +289C + bits 0:15 + Decoder vertical field offset 1 + + bits 16:31 + Decoder vertical field offset 2 + + Controls field output vertical alignment. The higher the number, the lower + the image on screen. Known starting values are 0x011E0017 (NTSC) & + 0x01500017 (PAL) +-------------------------------------------------------------------------------- +28A0 + bits 0:15 + Decoder & osd width in pixels + + bits 16:31 + Decoder & osd height in pixels + + All output from the decoder & osd are disabled beyond this area. Decoder + output will simply go black outside of this region. If the osd tries to + exceed this area it will become corrupt. +-------------------------------------------------------------------------------- +28A4 + bits 0:11 + osd left shift. + + Has a range of 0x770->0x7FF. With the exception of 0, any value outside of + this range corrupts the osd. +-------------------------------------------------------------------------------- +28A8 + bits 0:15 + osd vertical field offset 1 + + bits 16:31 + osd vertical field offset 2 + + Controls field output vertical alignment. The higher the number, the lower + the image on screen. Known starting values are 0x011E0017 (NTSC) & + 0x01500017 (PAL) +-------------------------------------------------------------------------------- +28AC -------- ?? unknown + | + V +28BC -------- ?? unknown +-------------------------------------------------------------------------------- +28C0 + bit 0 + Current output field + 0 = first field + 1 = second field + + bits 16:31 + Current scanline + The scanline counts from the top line of the first field + through to the last line of the second field. +-------------------------------------------------------------------------------- +28C4 -------- ?? unknown + | + V +28F8 -------- ?? unknown +-------------------------------------------------------------------------------- +28FC + bit 0 + ?? unknown + 0 = Normal + 1 = Breaks decoder & osd output +-------------------------------------------------------------------------------- +2900 + bits 0:31 + Decoder vertical Y alias register 1 +--------------- +2904 + bits 0:31 + Decoder vertical Y alias register 2 +--------------- +2908 + bits 0:31 + Decoder vertical Y alias trigger + + These three registers control the vertical aliasing filter for the Y plane. + Operation is similar to the horizontal Y filter (2804). The only real + difference is that there are only two registers to set before accessing + the trigger register (2908). As for the horizontal filter, the values are + taken from a lookup table in the firmware, and the procedure must be + repeated 16 times to fully program the filter. +-------------------------------------------------------------------------------- +290C + bits 0:31 + Decoder vertical UV alias register 1 +--------------- +2910 + bits 0:31 + Decoder vertical UV alias register 2 +--------------- +2914 + bits 0:31 + Decoder vertical UV alias trigger + + These three registers control the vertical aliasing filter for the UV + plane. Operation is the same as the Y filter, with 2914 being the trigger. +-------------------------------------------------------------------------------- +2918 + bits 0:15 + Decoder Y source height in pixels + + bits 16:31 + Decoder Y destination height in pixels +--------------- +291C + bits 0:15 + Decoder UV source height in pixels divided by 2 + + bits 16:31 + Decoder UV destination height in pixels + + NOTE: For both registers, the resulting image must be fully visible on + screen. If the image exceeds the bottom edge both the source and + destination size must be adjusted to reflect the visible portion. For the + source height, you must take into account the scaling when calculating the + new value. +-------------------------------------------------------------------------------- +2920 + bits 0:31 + Decoder Y vertical scaling + Normally = Reg 2930 >> 2 +--------------- +2924 + bits 0:31 + Decoder Y vertical scaling + Normally = Reg 2920 + 0x514 +--------------- +2928 + bits 0:31 + Decoder UV vertical scaling + When enlarging = Reg 2930 >> 2 + When reducing = Reg 2930 >> 3 +--------------- +292C + bits 0:31 + Decoder UV vertical scaling + Normally = Reg 2928 + 0x514 +--------------- +2930 + bits 0:31 + Decoder 'master' value for vertical scaling +--------------- +2934 + bits 0:31 + Decoder ?? unknown - Y vertical scaling +--------------- +2938 + bits 0:31 + Decoder Y vertical scaling + Normally = Reg 2930 +--------------- +293C + bits 0:31 + Decoder ?? unknown - Y vertical scaling +--------------- +2940 + bits 0:31 + Decoder UV vertical scaling + When enlarging = Reg 2930 >> 1 + When reducing = Reg 2930 +--------------- +2944 + bits 0:31 + Decoder ?? unknown - UV vertical scaling +--------------- +2948 + bits 0:31 + Decoder UV vertical scaling + Normally = Reg 2940 +--------------- +294C + bits 0:31 + Decoder ?? unknown - UV vertical scaling + + Most of these registers either control vertical scaling, or appear linked + to it in some way. Register 2930 contains the 'master' value & all other + registers can be calculated from that one. You must also remember to + correctly set the divider in Reg 296C + + To enlarge: + Reg 2930 = (source_height * 0x00200000) / destination_height + Reg 296C = No divide + + To reduce from full size down to half size: + Reg 2930 = (source_height/2 * 0x00200000) / destination height + Reg 296C = Divide by 2 + + To reduce from half down to quarter. + Reg 2930 = (source_height/4 * 0x00200000) / destination height + Reg 296C = Divide by 4 + +-------------------------------------------------------------------------------- +2950 + bits 0:15 + Decoder Y line index into display buffer, first field + + bits 16:31 + Decoder Y vertical line skip, first field +-------------------------------------------------------------------------------- +2954 + bits 0:15 + Decoder Y line index into display buffer, second field + + bits 16:31 + Decoder Y vertical line skip, second field +-------------------------------------------------------------------------------- +2958 + bits 0:15 + Decoder UV line index into display buffer, first field + + bits 16:31 + Decoder UV vertical line skip, first field +-------------------------------------------------------------------------------- +295C + bits 0:15 + Decoder UV line index into display buffer, second field + + bits 16:31 + Decoder UV vertical line skip, second field +-------------------------------------------------------------------------------- +2960 + bits 0:15 + Decoder destination height minus 1 + + bits 16:31 + Decoder destination height divided by 2 +-------------------------------------------------------------------------------- +2964 + bits 0:15 + Decoder Y vertical offset, second field + + bits 16:31 + Decoder Y vertical offset, first field + + These two registers shift the Y plane up. The higher the number, the + greater the shift. +-------------------------------------------------------------------------------- +2968 + bits 0:15 + Decoder UV vertical offset, second field + + bits 16:31 + Decoder UV vertical offset, first field + + These two registers shift the UV plane up. The higher the number, the + greater the shift. +-------------------------------------------------------------------------------- +296C + bits 0:1 + Decoder vertical Y output size divider + 00 = No divide + 01 = Divide by 2 + 10 = Divide by 4 + + bits 8:9 + Decoder vertical UV output size divider + 00 = No divide + 01 = Divide by 2 + 10 = Divide by 4 +-------------------------------------------------------------------------------- +2970 + bit 0 + Decoder ?? unknown + 0 = Normal + 1 = Affect video output levels + + bit 16 + Decoder ?? unknown + 0 = Normal + 1 = Disable vertical filter + +-------------------------------------------------------------------------------- +2974 -------- ?? unknown + | + V +29EF -------- ?? unknown +-------------------------------------------------------------------------------- +2A00 + bits 0:2 + osd colour mode + 001 = 16 bit (565) + 010 = 15 bit (555) + 011 = 12 bit (444) + 100 = 32 bit (8888) + 101 = 8 bit indexed + + bits 4:5 + osd display bpp + 01 = 8 bit + 10 = 16 bit + 11 = 32 bit + + bit 8 + osd global alpha + 0 = Off + 1 = On + + bit 9 + osd local alpha + 0 = Off + 1 = On + + bit 10 + osd colour key + 0 = Off + 1 = On + + bit 11 + osd ?? unknown + Must be 1 + + bit 13 + osd colour space + 0 = ARGB + 1 = AYVU + + bits 16:31 + osd ?? unknown + Must be 0x001B (some kind of buffer pointer ?) + + When the bits-per-pixel is set to 8, the colour mode is ignored and + assumed to be 8 bit indexed. For 16 & 32 bits-per-pixel the colour depth + is honoured, and when using a colour depth that requires fewer bytes than + allocated the extra bytes are used as padding. So for a 32 bpp with 8 bit + index colour, there are 3 padding bytes per pixel. It's also possible to + select 16bpp with a 32 bit colour mode. This results in the pixel width + being doubled, but the color key will not work as expected in this mode. + + Colour key is as it suggests. You designate a colour which will become + completely transparent. When using 565, 555 or 444 colour modes, the + colour key is always 16 bits wide. The colour to key on is set in Reg 2A18. + + Local alpha is a per-pixel 256 step transparency, with 0 being transparent + and 255 being solid. This is only available in 32 bit & 8 bit indexed + colour modes. + + Global alpha is a 256 step transparency that applies to the entire osd, + with 0 being transparent & 255 being solid. + + It's possible to combine colour key, local alpha & global alpha. +-------------------------------------------------------------------------------- +2A04 + bits 0:15 + osd x coord for left edge + + bits 16:31 + osd y coord for top edge +--------------- +2A08 + bits 0:15 + osd x coord for right edge + + bits 16:31 + osd y coord for bottom edge + + For both registers, (0,0) = top left corner of the display area. These + registers do not control the osd size, only where it's positioned & how + much is visible. The visible osd area cannot exceed the right edge of the + display, otherwise the osd will become corrupt. See reg 2A10 for + setting osd width. +-------------------------------------------------------------------------------- +2A0C + bits 0:31 + osd buffer index + + An index into the osd buffer. Slowly incrementing this moves the osd left, + wrapping around onto the right edge +-------------------------------------------------------------------------------- +2A10 + bits 0:11 + osd buffer 32 bit word width + + Contains the width of the osd measured in 32 bit words. This means that all + colour modes are restricted to a byte width which is divisible by 4. +-------------------------------------------------------------------------------- +2A14 + bits 0:15 + osd height in pixels + + bits 16:32 + osd line index into buffer + osd will start displaying from this line. +-------------------------------------------------------------------------------- +2A18 + bits 0:31 + osd colour key + + Contains the colour value which will be transparent. +-------------------------------------------------------------------------------- +2A1C + bits 0:7 + osd global alpha + + Contains the global alpha value (equiv ivtvfbctl --alpha XX) +-------------------------------------------------------------------------------- +2A20 -------- ?? unknown + | + V +2A2C -------- ?? unknown +-------------------------------------------------------------------------------- +2A30 + bits 0:7 + osd colour to change in indexed palette +--------------- +2A34 + bits 0:31 + osd colour for indexed palette + + To set the new palette, first load the index of the colour to change into + 2A30, then load the new colour into 2A34. The full palette is 256 colours, + so the index range is 0x00-0xFF +-------------------------------------------------------------------------------- +2A38 -------- ?? unknown +2A3C -------- ?? unknown +-------------------------------------------------------------------------------- +2A40 + bits 0:31 + osd ?? unknown + + Affects overall brightness, wrapping around to black +-------------------------------------------------------------------------------- +2A44 + bits 0:31 + osd ?? unknown + + Green tint +-------------------------------------------------------------------------------- +2A48 + bits 0:31 + osd ?? unknown + + Red tint +-------------------------------------------------------------------------------- +2A4C + bits 0:31 + osd ?? unknown + + Affects overall brightness, wrapping around to black +-------------------------------------------------------------------------------- +2A50 + bits 0:31 + osd ?? unknown + + Colour shift +-------------------------------------------------------------------------------- +2A54 + bits 0:31 + osd ?? unknown + + Colour shift +-------------------------------------------------------------------------------- +2A58 -------- ?? unknown + | + V +2AFC -------- ?? unknown +-------------------------------------------------------------------------------- +2B00 + bit 0 + osd filter control + 0 = filter off + 1 = filter on + + bits 1:4 + osd ?? unknown + +-------------------------------------------------------------------------------- + +v0.3 - 2 February 2007 - Ian Armstrong (ian@iarmst.demon.co.uk) +