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 Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel