use persistent code caches for improved performance on short-running large apps like chrome

[derekbruening]

From [email protected] on January 28, 2012 14:54:17

short-running apps that execute a lot of code with little code re-use preset a worst-case scenario for dynamic code caching tools like Dr. Memory.

this issue covers employing a persistent cache in Dr. Memory, initially targeting chrome ui_tests.

first I plan to implement partial support from DR in https://code.google.com/p/dynamorio/issues/detail?id=40

Original issue: http://code.google.com/p/drmemory/issues/detail?id=769

[derekbruening]

From [email protected] on April 18, 2012 14:10:43

persistence of light mode is feature-complete and works fine on small apps. I split full mode off into a separate issue #867 b/c I may postpone that work. bugs in light mode persistence will be filed as separate issues now.

------------------------------------------------------------------------ r820 | derek.bruening | 2012-04-17 17:31:12 -0400 (Tue, 17 Apr 2012) | 2 lines

issue #769: persistent cache support, part 3

• add a test to the suite for "-light -persist_code"

------------------------------------------------------------------------ r818 | derek.bruening | 2012-04-17 17:31:02 -0400 (Tue, 17 Apr 2012) | 2 lines

issue #769: persistent cache support, part 2

• add hashtable persistence, using DRi#708

------------------------------------------------------------------------ r817 | derek.bruening | 2012-04-17 17:30:52 -0400 (Tue, 17 Apr 2012) | 12 lines

issue #769: persistent cache support, part 1

• options -persist_code and -persist_dir
• default dir is logs/codecache/
• persist for all but full mode
• set a preferred base and only resurrect
  when base matches to support absolute
  refs into lib w/o patching
• version and options checks for pcache
  consistency

still TODO:

• persist hashtables

Status: Verified