i965: Increase G4X default VS URB allocation to actually allow 32 threads.

This improves the performance of my GLSL demo by 30%.  It also fixes the
VS deadlock that ut2004 had, for reasons I can't explain. Bug #21330.
This commit is contained in:
Eric Anholt 2009-06-30 14:26:06 -07:00
parent f359165a82
commit c3499f6c66

View file

@ -143,7 +143,19 @@ static void recalculate_urb_fence( struct brw_context *brw )
brw->urb.nr_clip_entries = limits[CLP].preferred_nr_entries;
brw->urb.nr_sf_entries = limits[SF].preferred_nr_entries;
brw->urb.nr_cs_entries = limits[CS].preferred_nr_entries;
brw->urb.constrained = 0;
if (BRW_IS_G4X(brw)) {
brw->urb.nr_vs_entries = 64;
if (check_urb_layout(brw)) {
goto done;
} else {
brw->urb.constrained = 1;
brw->urb.nr_vs_entries = limits[VS].preferred_nr_entries;
}
}
if (!check_urb_layout(brw)) {
brw->urb.nr_vs_entries = limits[VS].min_nr_entries;
brw->urb.nr_gs_entries = limits[GS].min_nr_entries;
@ -169,9 +181,8 @@ static void recalculate_urb_fence( struct brw_context *brw )
if (INTEL_DEBUG & (DEBUG_URB|DEBUG_FALLBACKS))
_mesa_printf("URB CONSTRAINED\n");
}
else
brw->urb.constrained = 0;
done:
if (INTEL_DEBUG & DEBUG_URB)
_mesa_printf("URB fence: %d ..VS.. %d ..GS.. %d ..CLP.. %d ..SF.. %d ..CS.. %d\n",
brw->urb.vs_start,