This was solution i wanted to implement (but not time to do :(), i really
think that os x driver use this to upload texture. I don't think this
will change
anything for x86 setup. Thus if no one is against, may we apply this change ?
Moreover could this change also affect way X do bitblit with xrender
acceleration ?
Because we got this issue too with X bitblit (there have been discusion on this
on xorg list).
Jerome Glisse
On Wed, 9 Mar 2005 11:24:02 +1100, Paul Mackerras <[EMAIL PROTECTED]> wrote:
> I started looking into the issue of how we handle various texture
> formats on R300 on big-endian machines. It became evident that
> textures were getting byte-swapped on their way to the framebuffer.
> Setting RADEON_HOST_DATA_SWAP_32BIT in RADEON_RBBM_GUICNTL doesn't
> seem to have any effect on R300.
>
> We can cope with the byte-swap for textures with 4 bytes/texel, but
> not for textures with 2 or 1 byte/texel. So instead of using a
> HOSTDATA_BLT in radeon_cp_dispatch_texture, I changed it to use a
> BITBLT_MULTI. I still copy the texture into gart memory, but instead
> of using an indirect buffer I just put the blit command into the ring
> buffer. This avoids the byte swap that the CP does and gets the
> texture to the framebuffer without being byte-swapped. It should be
> just as fast this way as with the HOSTDATA_BLT.
>
> The patch below implements this. With this patch we also need a patch
> to the r300 client driver code, which I will post shortly.
>
> Paul.
>
> diff -urN cvs/r300_driver/drm/shared-core/radeon_state.c
> r300_driver/drm/shared-core/radeon_state.c
> --- cvs/r300_driver/drm/shared-core/radeon_state.c 2005-03-05
> 09:26:06.000000000 +1100
> +++ r300_driver/drm/shared-core/radeon_state.c 2005-03-09 10:32:45.000000000
> +1100
> @@ -1471,7 +1471,7 @@
>
> }
>
> -#define RADEON_MAX_TEXTURE_SIZE (RADEON_BUFFER_SIZE - 8 * sizeof(u32))
> +#define RADEON_MAX_TEXTURE_SIZE RADEON_BUFFER_SIZE
>
> static int radeon_cp_dispatch_texture(DRMFILE filp,
> drm_device_t * dev,
> @@ -1488,6 +1488,7 @@
> u32 height;
> int i;
> u32 texpitch, microtile;
> + u32 offset;
> RING_LOCALS;
>
> DRM_GET_PRIV_WITH_RETURN(filp_priv, filp);
> @@ -1508,16 +1509,6 @@
> RADEON_WAIT_UNTIL_IDLE();
> ADVANCE_RING();
>
> -#ifdef __BIG_ENDIAN
> - /* The Mesa texture functions provide the data in little endian as the
> - * chip wants it, but we need to compensate for the fact that the CP
> - * ring gets byte-swapped
> - */
> - BEGIN_RING(2);
> - OUT_RING_REG(RADEON_RBBM_GUICNTL, RADEON_HOST_DATA_SWAP_32BIT);
> - ADVANCE_RING();
> -#endif
> -
> /* The compiler won't optimize away a division by a variable,
> * even if the only legal values are powers of two. Thus, we'll
> * use a shift instead.
> @@ -1601,23 +1592,6 @@
> buffer =
> (u32 *) ((char *)dev->agp_buffer_map->handle +
> buf->offset);
> dwords = size / 4;
> - buffer[0] = CP_PACKET3(RADEON_CNTL_HOSTDATA_BLT, dwords + 6);
> - buffer[1] = (RADEON_GMC_DST_PITCH_OFFSET_CNTL |
> - RADEON_GMC_BRUSH_NONE |
> - (format << 8) |
> - RADEON_GMC_SRC_DATATYPE_COLOR |
> - RADEON_ROP3_S |
> - RADEON_DP_SRC_SOURCE_HOST_DATA |
> - RADEON_GMC_CLR_CMP_CNTL_DIS |
> - RADEON_GMC_WR_MSK_DIS);
> -
> - buffer[2] = (texpitch << 22) | (tex->offset >> 10);
> - buffer[3] = 0xffffffff;
> - buffer[4] = 0xffffffff;
> - buffer[5] = (image->y << 16) | image->x;
> - buffer[6] = (height << 16) | image->width;
> - buffer[7] = dwords;
> - buffer += 8;
>
> if (microtile) {
> /* texture micro tiling in use, minimum texture width
> is thus 16 bytes.
> @@ -1726,8 +1700,26 @@
> }
>
> buf->filp = filp;
> - buf->used = (dwords + 8) * sizeof(u32);
> - radeon_cp_dispatch_indirect(dev, buf, 0, buf->used);
> + buf->used = size;
> + offset = dev_priv->gart_buffers_offset + buf->offset;
> + BEGIN_RING(7);
> + OUT_RING(CP_PACKET3(RADEON_CNTL_BITBLT_MULTI, 5));
> + OUT_RING(RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
> + RADEON_GMC_DST_PITCH_OFFSET_CNTL |
> + RADEON_GMC_BRUSH_NONE |
> + (format << 8) |
> + RADEON_GMC_SRC_DATATYPE_COLOR |
> + RADEON_ROP3_S |
> + RADEON_DP_SRC_SOURCE_MEMORY |
> + RADEON_GMC_CLR_CMP_CNTL_DIS |
> + RADEON_GMC_WR_MSK_DIS );
> + OUT_RING((texpitch << 22) | (offset >> 10));
> + OUT_RING((texpitch << 22) | (tex->offset >> 10));
> + OUT_RING(0);
> + OUT_RING((image->x << 16) | image->y);
> + OUT_RING((image->width << 16) | height);
> + ADVANCE_RING();
> +
> radeon_cp_discard_buffer(dev, buf);
>
> /* Update the input parameters for next time */
>
> -------------------------------------------------------
> SF email is sponsored by - The IT Product Guide
> Read honest & candid reviews on hundreds of IT Products from real users.
> Discover which products truly live up to the hype. Start reading now.
> http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click
> --
> _______________________________________________
> Dri-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/dri-devel
>
-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click
--
_______________________________________________
Dri-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dri-devel
