613ad38b10
Long term we'll need to spin up our own codegen rather than using xbyak, since it uses exceptions and does not support ARM. I really don't like how I've hacked in the caching of the JITted program, but that'll change anyway when switching away from xbyak. Performance looks good, better than (lowp) SkRasterPipeline but not quite as good as hand-tuned code: Opts: 0.2 ns/px RP: 0.7 F32: 0.8 I32: 0.6 I32_SWAR: 0.5 ninja -C out dm nanobench && out/dm -m SkVM && gobjdump -D -b binary -mi386:x86-64 /tmp/code.bin --insn-width=10 Change-Id: I34ce46e1d3fe5aa75f42709049a92ac79bf48bbd Reviewed-on: https://skia-review.googlesource.com/c/skia/+/219341 Reviewed-by: Herb Derby <herb@google.com> Commit-Queue: Mike Klein <mtklein@google.com> |
||
|---|---|---|
| .. | ||
| BUILD.gn | ||