OSDN Git Service

i965/fs: Move perf_debug about register spilling to a more obvious spot.
authorKenneth Graunke <kenneth@whitecape.org>
Wed, 14 May 2014 03:41:27 +0000 (20:41 -0700)
committerKenneth Graunke <kenneth@whitecape.org>
Mon, 19 May 2014 06:35:18 +0000 (23:35 -0700)
The if (!allocated_without_spills) block is an obvious spot for this
performance warning message.

In the Vec4 backend, scratch is also used for indirect access of
temporary arrays.  The FS backend doesn't implement that yet, but
if it did, this message would be inaccurate, since scratch access
wouldn't necessarily mean spilling.  Moving it preemptively fixes that.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
src/mesa/drivers/dri/i965/brw_fs.cpp
src/mesa/drivers/dri/i965/brw_wm.c

index 6b1b866..ed5bb07 100644 (file)
@@ -3064,6 +3064,10 @@ fs_visitor::run()
          if (dispatch_width == 16) {
             fail("Failure to register allocate.  Reduce number of "
                  "live scalar values to avoid this.");
+         } else {
+            perf_debug("Fragment shader triggered register spilling.  "
+                       "Try reducing the number of live scalar values to "
+                       "improve performance.\n");
          }
 
          /* Since we're out of heuristics, just go spill registers until we
index 8530f4f..cda8689 100644 (file)
@@ -186,10 +186,6 @@ bool do_wm_prog(struct brw_context *brw,
 
    /* Scratch space is used for register spilling */
    if (c->last_scratch) {
-      perf_debug("Fragment shader triggered register spilling.  "
-                 "Try reducing the number of live scalar values to "
-                 "improve performance.\n");
-
       c->prog_data.total_scratch = brw_get_scratch_size(c->last_scratch);
 
       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,