Code

Double-Precision towering#155

Code

Merged

mratsim merged 22 commits intomastermratsim/constantine:masterfrom tower-2xmratsim/constantine:tower-2xCopy head branch name to clipboardFeb 9, 2021

+1,572-699

Commits

Commits on Feb 9, 2021

• consistent naming for dbl-width

  mratsim

  committed
  f41aeb9

• Isolate double-width Fp2 mul

  mratsim

  committed
  6f56899

• Implement double-width complex multiplication

  mratsim

  committed
  180c839

• Lay out Fp4 double-width mul

  mratsim

  committed
  acc4afe

• Off by p in square Fp4 as well :/

  mratsim

  committed
  e07a947

• less copies and stack space in addition chains

  mratsim

  committed
  b5e213b

• Address #154 partly

  mratsim

  committed
  3929d19

• Fix #154, faster Fp4 square: less non-residue, no Mul, only square (bit more ops total)

  mratsim

  committed
  f8f051a

• Fix typo

  mratsim

  committed
  bb4a379

• better assembly scheduling for add/sub

  mratsim

  committed
  6b09da0

• Double-width -> Double-precision

  mratsim

  committed
  c0a5000

• Unred -> Unr

  mratsim

  committed
  6702941

• double-precision modular addition

  mratsim

  committed
  80434b2

• Replace canUseNoCarryMontyMul and canUseNoCarryMontySquare by getSpareBits

  mratsim

  committed
  00d6e9a

• Complete the double-precision implementation

  mratsim

  committed
  a265872

• Use double-precision path for Fp4 squaring and mul

  mratsim

  committed
  f5eb4e2

• remove mixin annotations

  mratsim

  committed
  0fe1f2b

• Lazy reduction in Fp4 prod

  mratsim

  committed
  234147b

• Fix assembly for sum2xMod

  mratsim

  committed
  0df6bb0

• Assembly for double-precision negation

  mratsim

  committed
  4d0c92f

• reduce white spaces in pairing benchmarks

  mratsim

  committed
  427143c

• ADX implies BMI2

  mratsim

  committed
  052f4bf