DescriptionThis is a somewhat speculative attempt to address this regression: https://code.google.com/p/chromium/issues/detail?id=180345.
On Win7 (old on left; new on right) we get (removing any differences <= .5ms or <= 3%):
rects_1_stroke_4 GPU c 0.78 1.40 -0.62 -79.5%
bitmap_8888_A_scale_rotate_filter GPU c 1.40 2.34 -0.94 -67.1%
path_stroke_small_triangle GPU c 26.83 28.55 -1.72 -6.4%
readpix GPU c 79.25 84.08 -4.83 -6.1%
giantdashline_vert_2 GPU c 13.42 14.20 -0.78 -5.8%
giantdashline_hori_0 GPU c 12.95 13.57 -0.62 -4.8%
path_fill_small_sawtooth GPU c 16.54 17.32 -0.78 -4.7%
path_stroke_small_rect GPU c 25.27 26.21 -0.94 -3.7%
polygon GPU c 51.48 53.20 -1.72 -3.3%
path_fill_small_long_line GPU c 20.12 19.50 +0.62 +3.1%
bitmap_8888_update GPU c 41.50 40.09 +1.41 +3.4%
aaclip_rect_AA GPU c 18.56 17.78 +0.78 +4.2%
bitmap_8888_update_scale_filter GPU c 43.37 41.50 +1.87 +4.3%
giantdashline_hori_2 GPU c 42.74 40.56 +2.18 +5.1%
matrixconvolution GPU c 11.08 10.30 +0.78 +7.0%
blurrect_boxfilter_1.50 GPU c 8.58 7.80 +0.78 +9.1%
maskshader GPU c 3.28 2.65 +0.63 +19.2%
lightingpointlitspecular_large GPU c 1.09 0.47 +0.62 +56.9%
aarects_rotated GPU c 113.41 22.31 +91.10 +80.3%
Of these times I believe only the 'aarects_rotated' difference is real.
For SampleApp::ManyRects (with rotations and AA turned on and clipping turned off) we get:
without Opt: ~2ms
with Opt: ~1ms
Patch Set 1 #Patch Set 2 : Updated to ToT #
Messages
Total messages: 3 (0 generated)
|