i965/eu: Take into account the target cache argument in brw_set_dp_read_message.

[mesa.git] / src / mesa / drivers / dri / i965 / brw_fs_saturate_propagation.cpp

diff --git a/src/mesa/drivers/dri/i965/brw_fs_saturate_propagation.cpp b/src/mesa/drivers/dri/i965/brw_fs_saturate_propagation.cpp
index e406c2899e89c954b257364ad1da979f70de42d9..f59fdbddfa69c08c827f19f3407605d94ee16b4d 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs_saturate_propagation.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_saturate_propagation.cpp
@@ -52,27 +52,60 @@ opt_saturate_propagation_local(fs_visitor *v, bblock_t *block)
       ip--;
 
       if (inst->opcode != BRW_OPCODE_MOV ||
-          inst->dst.file != GRF ||
-          inst->src[0].file != GRF ||
-          inst->src[0].abs ||
-          inst->src[0].negate ||
-          !inst->saturate)
+          !inst->saturate ||
+          inst->dst.file != VGRF ||
+          inst->dst.type != inst->src[0].type ||
+          inst->src[0].file != VGRF ||
+          inst->src[0].abs)
          continue;
 
       int src_var = v->live_intervals->var_from_reg(inst->src[0]);
       int src_end_ip = v->live_intervals->end[src_var];
 
       bool interfered = false;
-      foreach_inst_in_block_reverse_starting_from(fs_inst, scan_inst, inst, block) {
+      foreach_inst_in_block_reverse_starting_from(fs_inst, scan_inst, inst) {
          if (scan_inst->overwrites_reg(inst->src[0])) {
-            if (scan_inst->is_partial_write())
+            if (scan_inst->is_partial_write() ||
+                (scan_inst->dst.type != inst->dst.type &&
+                 !scan_inst->can_change_types()))
                break;
 
             if (scan_inst->saturate) {
                inst->saturate = false;
                progress = true;
-            } else if (src_end_ip <= ip || inst->dst.equals(inst->src[0])) {
+            } else if (src_end_ip == ip || inst->dst.equals(inst->src[0])) {
                if (scan_inst->can_do_saturate()) {
+                  if (scan_inst->dst.type != inst->dst.type) {
+                     scan_inst->dst.type = inst->dst.type;
+                     for (int i = 0; i < scan_inst->sources; i++) {
+                        scan_inst->src[i].type = inst->dst.type;
+                     }
+                  }
+
+                  if (inst->src[0].negate) {
+                     if (scan_inst->opcode == BRW_OPCODE_MUL) {
+                        scan_inst->src[0].negate = !scan_inst->src[0].negate;
+                        inst->src[0].negate = false;
+                     } else if (scan_inst->opcode == BRW_OPCODE_MAD) {
+                        scan_inst->src[0].negate = !scan_inst->src[0].negate;
+                        scan_inst->src[1].negate = !scan_inst->src[1].negate;
+                        inst->src[0].negate = false;
+                     } else if (scan_inst->opcode == BRW_OPCODE_ADD) {
+                        if (scan_inst->src[1].file == IMM) {
+                           if (!brw_negate_immediate(scan_inst->src[1].type,
+                                                     &scan_inst->src[1].as_brw_reg())) {
+                              break;
+                           }
+                        } else {
+                           scan_inst->src[1].negate = !scan_inst->src[1].negate;
+                        }
+                        scan_inst->src[0].negate = !scan_inst->src[0].negate;
+                        inst->src[0].negate = false;
+                     } else {
+                        break;
+                     }
+                  }
+
                   scan_inst->saturate = true;
                   inst->saturate = false;
                   progress = true;
@@ -81,13 +114,15 @@ opt_saturate_propagation_local(fs_visitor *v, bblock_t *block)
             break;
          }
          for (int i = 0; i < scan_inst->sources; i++) {
-            if (scan_inst->src[i].file == GRF &&
-                scan_inst->src[i].reg == inst->src[0].reg &&
+            if (scan_inst->src[i].file == VGRF &&
+                scan_inst->src[i].nr == inst->src[0].nr &&
                 scan_inst->src[i].reg_offset == inst->src[0].reg_offset) {
                if (scan_inst->opcode != BRW_OPCODE_MOV ||
                    !scan_inst->saturate ||
                    scan_inst->src[0].abs ||
-                   scan_inst->src[0].negate) {
+                   scan_inst->src[0].negate ||
+                   scan_inst->src[0].abs != inst->src[0].abs ||
+                   scan_inst->src[0].negate != inst->src[0].negate) {
                   interfered = true;
                   break;
                }