Decrease SkClipStack memory allocations & deallocations

Going with 10 records per block but that could change with SkPicture-based information.

bench_compare.py -o new1.log -n new5.log
              aaclip_rect_AA 8888     1.95       2.08      -0.13    -6.7%
               dash_0_noclip 8888     1.65       1.73      -0.08    -4.8%
               dash_1_noclip 8888     9.53       9.78      -0.25    -2.6%
               dash_4_noclip 8888     3.61       3.62      -0.01    -0.3%
        aaclip_build_path_AA 8888     7.24       7.23      +0.01    +0.1%
              aaclip_path_BW 8888     2.26       2.25      +0.01    +0.4%
            nested_aaclip_BW 8888     2.08       2.07      +0.01    +0.5%
            aaclip_setregion 8888     7.11       7.00      +0.11    +1.5%
              aaclip_path_AA 8888     3.92       3.77      +0.15    +3.8%
              aaclip_rect_BW 8888     0.69       0.65      +0.04    +5.8%
        aaclip_build_path_BW 8888     3.79       3.53      +0.26    +6.9%
            nested_aaclip_AA 8888     3.31       3.08      +0.23    +6.9%
        aaclip_build_rect_BW 8888     0.78       0.70      +0.08   +10.3%
              dash_1_clipped 8888     0.26       0.21      +0.05   +19.2%
        aaclip_build_rect_AA 8888     1.29       1.03      +0.26   +20.2%

bench_compare.py -o new1.log -n new10.log
            nested_aaclip_BW 8888     2.08       2.17      -0.09    -4.3%
               dash_1_noclip 8888     9.53       9.54      -0.01    -0.1%
        aaclip_build_path_AA 8888     7.24       7.08      +0.16    +2.2%
               dash_4_noclip 8888     3.61       3.53      +0.08    +2.2%
            aaclip_setregion 8888     7.11       6.93      +0.18    +2.5%
              aaclip_rect_AA 8888     1.95       1.90      +0.05    +2.6%
               dash_0_noclip 8888     1.65       1.59      +0.06    +3.6%
              aaclip_path_AA 8888     3.92       3.73      +0.19    +4.8%
        aaclip_build_path_BW 8888     3.79       3.52      +0.27    +7.1%
              aaclip_rect_BW 8888     0.69       0.64      +0.05    +7.2%
              aaclip_path_BW 888      2.26       2.09      +0.17    +7.5%
            nested_aaclip_AA 8888     3.31       3.05      +0.26    +7.9%
        aaclip_build_rect_BW 8888     0.78       0.70      +0.08   +10.3%
              dash_1_clipped 8888     0.26       0.21      +0.05   +19.2%
        aaclip_build_rect_AA 8888     1.29       1.02      +0.27   +20.9%

bench_compare.py -o new1.log -n new20.log
              aaclip_rect_BW 8888     0.69       0.76      -0.07   -10.1%
              aaclip_rect_AA 8888     1.95       2.07      -0.12    -6.2%
               dash_0_noclip 8888     1.65       1.69      -0.04    -2.4%
               dash_1_noclip 8888     9.53       9.74      -0.21    -2.2%
        aaclip_build_rect_BW 8888     0.78       0.79      -0.01    -1.3%
            nested_aaclip_BW 8888     2.08       2.09      -0.01    -0.5%
            aaclip_setregion 8888     7.11       7.13      -0.02    -0.3%
               dash_4_noclip 8888     3.61       3.62      -0.01    -0.3%
        aaclip_build_path_AA 8888     7.24       7.21      +0.03    +0.4%
              aaclip_path_AA 8888     3.92       3.86      +0.06    +1.5%
              aaclip_path_BW 8888     2.26       2.18      +0.08    +3.5%
        aaclip_build_path_BW 8888     3.79       3.61      +0.18    +4.7%
            nested_aaclip_AA 8888     3.31       3.02      +0.29    +8.8%
        aaclip_build_rect_AA 8888     1.29       1.10      +0.19   +14.7%
              dash_1_clipped 8888     0.26       0.21      +0.05   +19.2%

[reed1, 2012-08-16 20:15:04]

lgtm w/ request for comment(s)

http://codereview.appspot.com/6443138/diff/1/src/core/SkClipStack.cpp
File src/core/SkClipStack.cpp (right):

http://codereview.appspot.com/6443138/diff/1/src/core/SkClipStack.cpp#newcode441
src/core/SkClipStack.cpp:441: 
// This is a balance between allocating too much memory, and wanting to
// avoid too many allocations for a given lifetime of a canvas (i.e. what is the
// deepest save/restore stack we expect to see).

1. We should have a mild TODO somewhere to measure this exact value when we have
our 100s of sample pictures
2. Can we please please please make it a power-of-two ? :)

[robertphillips, 2012-08-17 14:25:50]

committed as r5151

http://codereview.appspot.com/6443138/diff/1/src/core/SkClipStack.cpp
File src/core/SkClipStack.cpp (right):

http://codereview.appspot.com/6443138/diff/1/src/core/SkClipStack.cpp#newcode441
src/core/SkClipStack.cpp:441: 
On 2012/08/16 20:15:04, reed1 wrote:
> // This is a balance between allocating too much memory, and wanting to
> // avoid too many allocations for a given lifetime of a canvas (i.e. what is
the
> // deepest save/restore stack we expect to see).
> 
> 1. We should have a mild TODO somewhere to measure this exact value when we
have
> our 100s of sample pictures
> 2. Can we please please please make it a power-of-two ? :)

Done.