X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fi965%2Fgen6_urb.c;h=c7311fd0b03137bf92b731f3515a5cbdbc5fd8a5;hb=5c0436dbf87fef76ba67456f215d9285c38f1816;hp=d47bf9ea46681a383a03dbb515ede0abf6c69f40;hpb=5fff3752c88255ea3f4eb26cddb2c996694b33b1;p=mesa.git

diff --git a/src/mesa/drivers/dri/i965/gen6_urb.c b/src/mesa/drivers/dri/i965/gen6_urb.c
index d47bf9ea466..c7311fd0b03 100644
--- a/src/mesa/drivers/dri/i965/gen6_urb.c
+++ b/src/mesa/drivers/dri/i965/gen6_urb.c
@@ -49,23 +49,32 @@
 static void
 gen6_upload_urb( struct brw_context *brw )
 {
-   struct intel_context *intel = &brw->intel;
    int nr_vs_entries, nr_gs_entries;
    int total_urb_size = brw->urb.size * 1024; /* in bytes */
 
-   /* CACHE_NEW_VS_PROG */
+   bool gs_present = brw->ff_gs.prog_active || brw->geometry_program;
+
+   /* BRW_NEW_VS_PROG_DATA */
    unsigned vs_size = MAX2(brw->vs.prog_data->base.urb_entry_size, 1);
 
-   /* We use the same VUE layout for VS outputs and GS outputs (as it's what
-    * the SF and Clipper expect), so we can simply make the GS URB entry size
-    * the same as for the VS.  This may technically be too large in cases
-    * where we have few vertex attributes and a lot of varyings, since the VS
-    * size is determined by the larger of the two.  For now, it's safe.
+   /* Whe using GS to do transform feedback only we use the same VUE layout for
+    * VS outputs and GS outputs (as it's what the SF and Clipper expect), so we
+    * can simply make the GS URB entry size the same as for the VS.  This may
+    * technically be too large in cases where we have few vertex attributes and
+    * a lot of varyings, since the VS size is determined by the larger of the
+    * two. For now, it's safe.
+    *
+    * For user-provided GS the assumption above does not hold since the GS
+    * outputs can be different from the VS outputs.
     */
    unsigned gs_size = vs_size;
+   if (brw->geometry_program) {
+      gs_size = brw->gs.prog_data->base.urb_entry_size;
+      assert(gs_size >= 1);
+   }
 
    /* Calculate how many entries fit in each stage's section of the URB */
-   if (brw->gs.prog_active) {
+   if (gs_present) {
       nr_vs_entries = (total_urb_size/2) / (vs_size * 128);
       nr_gs_entries = (total_urb_size/2) / (gs_size * 128);
    } else {
@@ -84,11 +93,11 @@ gen6_upload_urb( struct brw_context *brw )
    brw->urb.nr_vs_entries = ROUND_DOWN_TO(nr_vs_entries, 4);
    brw->urb.nr_gs_entries = ROUND_DOWN_TO(nr_gs_entries, 4);
 
-   assert(brw->urb.nr_vs_entries >= 24);
+   assert(brw->urb.nr_vs_entries >= brw->urb.min_vs_entries);
    assert(brw->urb.nr_vs_entries % 4 == 0);
    assert(brw->urb.nr_gs_entries % 4 == 0);
-   assert(vs_size < 5);
-   assert(gs_size < 5);
+   assert(vs_size <= 5);
+   assert(gs_size <= 5);
 
    BEGIN_BATCH(3);
    OUT_BATCH(_3DSTATE_URB << 16 | (3 - 2));
@@ -110,16 +119,19 @@ gen6_upload_urb( struct brw_context *brw )
     * doesn't exist on Gen6).  So for now we just do a full pipeline flush as
     * a workaround.
     */
-   if (brw->urb.gen6_gs_previously_active && !brw->gs.prog_active)
-      intel_batchbuffer_emit_mi_flush(intel);
-   brw->urb.gen6_gs_previously_active = brw->gs.prog_active;
+   if (brw->urb.gs_present && !gs_present)
+      brw_emit_mi_flush(brw);
+   brw->urb.gs_present = gs_present;
 }
 
 const struct brw_tracked_state gen6_urb = {
    .dirty = {
       .mesa = 0,
-      .brw = BRW_NEW_CONTEXT,
-      .cache = (CACHE_NEW_VS_PROG | CACHE_NEW_GS_PROG),
+      .brw = BRW_NEW_CONTEXT |
+             BRW_NEW_FF_GS_PROG_DATA |
+             BRW_NEW_GEOMETRY_PROGRAM |
+             BRW_NEW_GS_PROG_DATA |
+             BRW_NEW_VS_PROG_DATA,
    },
    .emit = gen6_upload_urb,
 };