summaryrefslogtreecommitdiff
path: root/compiler/optimizing/graph_visualizer.cc
diff options
context:
space:
mode:
authorVladimir Marko <vmarko@google.com>2018-05-04 15:04:45 +0100
committerVladimir Marko <vmarko@google.com>2018-05-08 17:16:51 +0100
commita90dd515c4451ad3d6e7ecdf0f0527fcc1e32d91 (patch)
tree1d551b2ce84dd87e7622d011354cea24b856047a /compiler/optimizing/graph_visualizer.cc
parentb865d9af3bb5f4777dbe9cbd60cdb2a8fe5f0557 (diff)
Reimplement HInstruction::Is##type().
Use a table-based implementation for abstract kinds. This implementation can be fully inlined and even eliminated if the range of possible kinds is known. There is a small (<0.5%) improvement of dex2oat compile time for a big app (sum of "Compile Dex File Quick" dumped for --dump-timings with the timing of compiler passes hacked away) on aosp_taimen-userdebug, little cores, -j4. The sizes of libart-compiler.so have changed: - lib/libart-compiler.so: 2178608 -> 2169136 (-9472) .rodata: 0xd36c -> 0xd55c (+496) .text: 0x175914 -> 0x175f44 (+1584) .data.rel.ro: 0xa858 -> 0x84e8 (-9072) - lib64/libart-compiler.so: 3091432 -> 3064248 (-27184) .rodata: 0x15aa4 -> 0x15c94 (+496) .text: 0x203304 -> 0x202294 (-4208) .data.rel.ro: 0x151d8 -> 0x10a68 (-18288) Note that .data.rel.ro is always dirty memory, .rodata and .text is always clean. Test: Rely on TreeHugger Change-Id: I95e6040ecd23cad83f024970c3bf785d32169deb
Diffstat (limited to 'compiler/optimizing/graph_visualizer.cc')
0 files changed, 0 insertions, 0 deletions