3a8ff230ab
Add SkOpts variant for avx2 to get FMA Decrease tolerance now that we're testing the function Before 15/15 MB 1 1.13ms 1.17ms 1.18ms 1.26ms 4% cubicmap_0_1_1_1 15/15 MB 1 1.08ms 1.13ms 1.12ms 1.17ms 3% cubicmap_0_1_1_0 15/15 MB 1 862µs 904µs 900µs 937µs 3% cubicmap_0_1_0_1 15/15 MB 1 861µs 878µs 882µs 934µs 3% cubicmap_0_1_0_0 15/15 MB 1 1.44ms 1.47ms 1.49ms 1.55ms 3% cubicmap_1_0_1_1 15/15 MB 1 1.44ms 1.48ms 1.48ms 1.55ms 3% cubicmap_1_0_1_0 15/15 MB 1 1.42ms 1.42ms 1.46ms 1.53ms 3% cubicmap_1_0_0_1 15/15 MB 1 1.42ms 1.42ms 1.44ms 1.51ms 2% cubicmap_1_0_0_0 After moving the check to the function, not delta_t 15/15 MB 1 900µs 900µs 915µs 971µs 3% cubicmap_0_1_1_1 15/15 MB 1 899µs 900µs 914µs 988µs 3% cubicmap_0_1_1_0 15/15 MB 1 865µs 896µs 890µs 946µs 3% cubicmap_0_1_0_1 15/15 MB 1 866µs 910µs 914µs 959µs 3% cubicmap_0_1_0_0 15/15 MB 1 1.29ms 1.29ms 1.33ms 1.44ms 4% cubicmap_1_0_1_1 15/15 MB 1 1.28ms 1.29ms 1.34ms 1.54ms 6% cubicmap_1_0_1_0 15/15 MB 1 1.26ms 1.26ms 1.27ms 1.34ms 3% cubicmap_1_0_0_1 15/15 MB 1 1.26ms 1.26ms 1.27ms 1.3ms 2% cubicmap_1_0_0_0 After SkOpts (on an avx2 machine) 15/15 MB 1 613µs 613µs 616µs 646µs 2% cubicmap_0_1_1_1 15/15 MB 1 613µs 613µs 624µs 654µs 3% cubicmap_0_1_1_0 15/15 MB 1 862µs 865µs 867µs 887µs 1% cubicmap_0_1_0_1 15/15 MB 1 865µs 901µs 896µs 949µs 3% cubicmap_0_1_0_0 15/15 MB 1 849µs 850µs 868µs 929µs 4% cubicmap_1_0_1_1 15/15 MB 1 849µs 850µs 873µs 940µs 4% cubicmap_1_0_1_0 15/15 MB 1 831µs 831µs 856µs 950µs 5% cubicmap_1_0_0_1 15/15 MB 1 831µs 831µs 848µs 911µs 3% cubicmap_1_0_0_0 (not checked in) if we also enable the pragma in cubic_solver 15/15 MB 1 593µs 594µs 597µs 623µs 2% cubicmap_0_1_1_1 15/15 MB 1 593µs 595µs 605µs 629µs 2% cubicmap_0_1_1_0 15/15 MB 1 864µs 867µs 869µs 890µs 1% cubicmap_0_1_0_1 15/15 MB 1 864µs 866µs 886µs 950µs 4% cubicmap_0_1_0_0 15/15 MB 1 809µs 831µs 841µs 891µs 4% cubicmap_1_0_1_1 15/15 MB 1 809µs 810µs 855µs 1.11ms 11% cubicmap_1_0_1_0 15/15 MB 1 794µs 861µs 856µs 914µs 4% cubicmap_1_0_0_1 15/15 MB 1 794µs 821µs 818µs 853µs 3% cubicmap_1_0_0_0 Change-Id: I260391be956d31a5cf3d0367d1285e56af7568f8 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/226499 Reviewed-by: Mike Reed <reed@google.com> Reviewed-by: Mike Klein <mtklein@google.com> Commit-Queue: Mike Reed <reed@google.com> Auto-Submit: Mike Reed <reed@google.com> |
||
|---|---|---|
| .. | ||
| android | ||
| atlastext | ||
| c | ||
| codec | ||
| config | ||
| core | ||
| docs | ||
| effects | ||
| encode | ||
| gpu | ||
| pathops | ||
| ports | ||
| private | ||
| svg | ||
| third_party | ||
| utils | ||