Skip to content
Snippets Groups Projects

x86: Add refmvs.save_tmvs SSSE3 asm

Merged Victorien Le Couviour--Tuffet requested to merge psilokos/dav1d:refmvs_simd into master

64-bit:

(10 checkasm runs)
Speed ups:
c..ssse3: 2.921x (o=0.0249)
c..avx2: 3.134x (o=0.0289)
Speed diffs:
c..ssse3: 34.24% (o=0.29)
c..avx2: 31.91% (o=0.29)
save_tmvs_c:      25681.4 ( 1.00x)
save_tmvs_ssse3:   8711.6 ( 2.95x)
save_tmvs_avx2:    8075.1 ( 3.18x)
chimera: 453.64 => 455.93 (~+0.5%)

32-bit:

(10 checkasm runs)
Speed ups:
c..ssse3: 2.353x (o=0.0253)
Speed diffs:
c..ssse3: 42.51% (o=0.46)
save_tmvs_c:      23775.4 ( 1.00x)
save_tmvs_ssse3:   9799.6 ( 2.43x)
chimera: 408.40 => 412.16 (~+0.9%)

These chimera decodes numbers aren't very stable, it's probably better to just look at the checkasm ones.

Edited by Victorien Le Couviour--Tuffet

Merge request reports

Loading
Loading

Activity

Filter activity
  • Approvals
  • Assignees & reviewers
  • Comments (from bots)
  • Comments (from users)
  • Commits & branches
  • Edits
  • Labels
  • Lock status
  • Mentions
  • Merge request status
  • Tracking
  • Henrik Gramner
  • Henrik Gramner
  • Henrik Gramner
  • Henrik Gramner
  • Henrik Gramner
  • added 1 commit

    • 4bf74338 - x86: Add refmvs.save_tmvs SSSE3 asm

    Compare with previous version

  • resolved all threads

  • resolved all threads

  • resolved all threads

  • Victorien Le Couviour--Tuffet changed the description

    changed the description

  • added 2 commits

    • da7dba4e - x86: Optimize refmvs.save_tmvs AVX2 asm
    • 69aa94c4 - x86: Add refmvs.save_tmvs SSSE3 asm

    Compare with previous version

  • resolved all threads

  • Loading
  • Loading
  • Loading
  • Loading
  • Loading
  • Loading
  • Loading
  • Loading
  • Loading
  • Loading
  • Please register or sign in to reply
    Loading