- memset(key, 0, sizeof(*key));
-
- /* CACHE_NEW_GS_PROG */
- key->prog_active = brw->gs.prog_active;
- if (key->prog_active) {
- key->total_grf = brw->gs.prog_data->total_grf;
- key->urb_entry_read_length = brw->gs.prog_data->urb_read_length;
- } else {
- key->total_grf = 1;
- key->urb_entry_read_length = 1;
+ struct brw_gs_unit_state *gs;
+
+ gs = brw_state_batch(brw, AUB_TRACE_GS_STATE,
+ sizeof(*gs), 32, &brw->ff_gs.state_offset);
+
+ memset(gs, 0, sizeof(*gs));
+
+ /* BRW_NEW_PROGRAM_CACHE | BRW_NEW_GS_PROG_DATA */
+ if (brw->ff_gs.prog_active) {
+ gs->thread0.grf_reg_count = (ALIGN(brw->ff_gs.prog_data->total_grf, 16) /
+ 16 - 1);
+
+ gs->thread0.kernel_start_pointer =
+ brw_program_reloc(brw,
+ brw->ff_gs.state_offset +
+ offsetof(struct brw_gs_unit_state, thread0),
+ brw->ff_gs.prog_offset +
+ (gs->thread0.grf_reg_count << 1)) >> 6;
+
+ gs->thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
+ gs->thread1.single_program_flow = 1;
+
+ gs->thread3.dispatch_grf_start_reg = 1;
+ gs->thread3.const_urb_entry_read_offset = 0;
+ gs->thread3.const_urb_entry_read_length = 0;
+ gs->thread3.urb_entry_read_offset = 0;
+ gs->thread3.urb_entry_read_length =
+ brw->ff_gs.prog_data->urb_read_length;
+
+ /* BRW_NEW_URB_FENCE */
+ gs->thread4.nr_urb_entries = brw->urb.nr_gs_entries;
+ gs->thread4.urb_entry_allocation_size = brw->urb.vsize - 1;
+
+ if (brw->urb.nr_gs_entries >= 8)
+ gs->thread4.max_threads = 1;
+ else
+ gs->thread4.max_threads = 0;