[MPlayer-dev-eng] [PATCH] vidix nvidia_vic patch for 256M memory, bps fix etc

[Sascha Sommer]

Hi

On Wednesday 08 February 2006 06:36, Pascal Yu wrote:
> Attached is the patch for nvidia_vid vidix driver fix for 256M VRAM. 256M
> card will not show 0 MB any more.
> Also add fix for bps, (byte per stride, not bpp byte per pixel).
> Along with this added device ID for Geforce FX5500 in pci.db file.
>
> The diagonally lines issue I mentioned earlier today for some media files,
> seems this CVS version I got today already included the same fix I did for
> my 1.0 pre7 try2 code - dstride.y=64 instead of 16 since it's a 128 bit
> memory I guess. Also the missing expand filter features for roundup and
> aspect from pre7try2 found in CVS souce, so I didn't add my patch, they are
> almost identical, except I used double, it's float in CVS.
> Do we have a patch ready to get rid of the menu OSD left on blank bands
> added by expand filter?
>

I don't know of any patch for this.


> --- nvidia_vid.c.org	2005-11-09 11:10:45.000000000 -0500
> +++ nvidia_vid.c	2006-02-08 00:05:06.000000000 -0500
> @@ -31,7 +31,7 @@
>
>
> #define MAX_FRAMES 3
> -#define NV04_BES_SIZE 1024*2000*4
> +#define NV04_BES_SIZE 2048*2000*4
>

Why did you change this?


>
> static vidix_capability_t nvidia_cap = {
> @@ -144,6 +144,7 @@
>    {DEVICE_NVIDIA_NV31_GEFORCE_FX2,NV_ARCH_30},
>    {DEVICE_NVIDIA_NV34_GEFORCE_FX,NV_ARCH_30},
>    {DEVICE_NVIDIA_NV34_GEFORCE_FX2,NV_ARCH_30},
> +  {DEVICE_NVIDIA_NV34_GEFORCE_FX3,NV_ARCH_30},
>    {DEVICE_NVIDIA_NV34M_GEFORCE_FX,NV_ARCH_30},
>    {DEVICE_NVIDIA_NV34GL_QUADRO_FX,NV_ARCH_30},
>    {DEVICE_NVIDIA_NV35_GEFORCE_FX,NV_ARCH_30},
> @@ -328,7 +329,11 @@
> }
>

Ok.

> static unsigned long rivatv_fbsize_nv10 (struct rivatv_chip *chip){
> -	return ((VID_RD32 (chip->PFB, 0x20C) >> 20) & 0x000000FF) * 1024 * 1024;
> +	return VID_RD32 (chip->PFB, 0x20C) & 0x0FF00000;
> +}
> +
> +static unsigned long rivatv_fbsize_nv30 (struct rivatv_chip *chip){
> +	return VID_RD32 (chip->PFB, 0x20C) & 0x1FF00000;
> }
>

Hm looks like X does something like
VID_RD32(chip->PFB,0x20C) &0xFFF00000 so maybe this would work for all cards 
>= nv10


> //lock funcs
> @@ -479,15 +484,13 @@
> }
>
> static void nv_getscreenproperties(struct rivatv_info *info){
> -  uint32_t bpp=0;
> +  uint32_t bpp=0,x;
>    info->chip.lock(&info->chip, 0);
>    /*get screen depth*/
>    VID_WR08(info->chip.PCIO, 0x03D4,0x28);
>    bpp = VID_RD08(info->chip.PCIO,0x03D5)&0x3;
> -  if(bpp==3)bpp=4;
>    if((bpp == 2) && (VID_RD32(info->chip.PVIDEO,0x600) & 0x00001000) ==
> 0x0)info->depth=15;
> -  else info->depth = bpp*8;
> -  info->bps=bpp;
> +  else info->depth = 0x04 << bpp;

Are you sure this is correct?
Before bpp was either 2 or 4 so depth was 16 or 32
With your patch it can be 24, too and as far as I remember at least the 
riva128 did not like this.

Regards

Sascha