Update bench pictures to time image decode & upload costs

tools/bench_pictures_main.cpp

Index: tools/bench_pictures_main.cpp
===================================================================
--- tools/bench_pictures_main.cpp	(revision 12709)
+++ tools/bench_pictures_main.cpp	(working copy)
@@ -41,6 +41,8 @@
 DEFINE_int32(repeat, 1, "Set the number of times to repeat each test.");
 DEFINE_bool(timeIndividualTiles, false, "Report times for drawing individual tiles, rather than "
             "times for drawing the whole page. Requires tiled rendering.");
+DEFINE_bool(purgeDecodedTex, false, "Purge decoded textures after each iteration "
+            "(only makes sense with --deferImageDecoding).");

[bsalomon, 2013/12/18 14:51:36]

I don't totally follow why this only makes sense with deferred decoding.

[robertphillips, 2013/12/18 15:59:13]

Done. You're right, if we allow it anytime we can compare the cost of just
uploading to VRAM to the cost of decoding and uploading to VRAM.

Case in point:

simple viewport 1500x1000
      lazydecode+purge    just purge    
gpu    271.83                45.24
8888   184.47                 2.73

 DEFINE_string(timers, "c", "[wcgWC]*: Display wall, cpu, gpu, truncated wall or truncated cpu time"
               " for each picture.");
 DEFINE_bool(trackDeferredCaching, false, "Only meaningful with --deferImageDecoding and "
@@ -338,6 +340,16 @@
         benchmark->setTimeIndividualTiles(true);
     }
 
+    if (FLAGS_purgeDecodedTex) {
+        if (!FLAGS_deferImageDecoding) {
+            gLogger.logError("--purgeDecodedTex doesn't make sense "
+                             "without --deferImageDecoding.\n");
+            exit(-1);
+        }
+
+        benchmark->setPurgeDecodedTex(true);
+    }
+
     if (FLAGS_readPath.count() < 1) {
         gLogger.logError(".skp files or directories are required.\n");
         exit(-1);