1 files changed, 20 insertions, 2 deletions
diff --git a/man/savage.man b/man/savage.man
index 6454681..822a233 100644
--- a/man/savage.man
+++ b/man/savage.man
@@ -169,8 +169,26 @@ applies to Savage4 and newer chips.  Default: \*qoff\*q (use COB).
 .BI "Option \*qBCIforXv\*q \*q" boolean \*q  
 Use the BCI to copy and reformat Xv pixel data.  Using the BCI for Xv causes 
 graphics artifacts on some chips.  This option only applies to Savage4 and 
-prosavage/twister chips.  Default: on for prosavage and twister (use BCI for Xv); 
-off for savage4 (do not use the BCI for Xv).
+prosavage/twister chips. On some combinations of chipsets and video players,
+BCI formatting might actually be slower than software formatting (\*qAGPforXv\*q 
+might help in this case). BCI formatting can only be used on video data with
+a width that is a multiple of 16 pixels (which is the vast majority of videos). 
+Other widths are handled through software formatting. Default: on for prosavage 
+and twister (use BCI for Xv); off for savage4 (do not use the BCI for Xv).
+.TP
+.BI "Option \*qAGPforXv\*q \*q" boolean \*q
+Instructs the BCI Xv pixel formatter to use AGP memory as a scratch buffer.
+Ordinarily the BCI formatter uses a an area in framebuffer memory to hold 
+YV12 planar data to be converted for display. This requires a somewhat expensive
+upload of YV12 data to framebuffer memory. The \*qAGPforXv\*q causes the BCI
+formatter to place the YV12 data in AGP memory instead, which can be uploaded
+faster than the framebuffer. Use of this option cuts upload overhead by 25%
+according to benchmarks. This option also smooths out most of the shearing
+present when using BCI for pixel conversion. Currently this option is 
+.B experimental
+and is disabled by default. Video width restrictions that apply to \*qBCIforXv\*q 
+also apply here. Only valid when \*qDRI\*q and \*qBCIforXv\*q are both active, 
+and only on AGP chipsets. Default: \*qoff\*q.
 .TP 
 .BI "Option \*qAGPMode\*q \*q" integer \*q
 Set AGP data transfer rate.