Commits

Branch selector

 master

User selector

All users

All time

Commit History

Commits on Jun 11, 2016

• Vec4i operator *: use shift/and/or instead of 3 shuffles

  Show description for 7752228
  pcordes

  committed
  7752228

  Copy full SHA for 7752228

Commits on Jun 10, 2016

• Vec2/4q operator >> use immediate blend, SSE2 custom blend strategy, and SSE4.1 improvements

  Show description for ae03cf5
  pcordes

  committed
  ae03cf5

  Copy full SHA for ae03cf5

Commits on Jun 5, 2016

• i256: horizontal_add_x(Vec32c): port xor + psadbw code from 128b to 256b

  Show description for 11aa770
  pcordes

  committed
  11aa770

  Copy full SHA for 11aa770

• i256: bugfix: return type of horizontal_add (Vec32c) should be signed

  Show description for 59486df
  pcordes

  committed
  59486df

  Copy full SHA for 59486df

• i256: Replace all _mm256_extractf128_si256 with the hi128i wrapper macro

  Show description for ca8a65f
  pcordes

  committed
  ca8a65f

  Copy full SHA for ca8a65f

• some TODO comments

  pcordes

  committed
  3c3757f

  Copy full SHA for 3c3757f

Commits on Jun 2, 2016

• i128 hsums: tune instruction selection for add_x sign/zero extension

  Show description for 1ca9d14
  pcordes

  committed
  1ca9d14

  Copy full SHA for 1ca9d14

• i128 hsums: shift instead of shuffle for HILO32 and 16 with AVX, and defeat clang shuffle pessimization.

  Show description for 5dc5b82
  pcordes

  committed
  5dc5b82

  Copy full SHA for 5dc5b82

• horizontal_add_x (Vec16c): range-shift to unsigned for psadbw

  Show description for 630ca80
  pcordes

  committed
  630ca80

  Copy full SHA for 630ca80

• horizontal_add_x (Vec8s): tune for code size / insns.

  Show description for 6d3a068
  pcordes

  committed
  6d3a068

  Copy full SHA for 6d3a068

• horizontal_add_x (Vec4ui): try 4 different ways (odd/even vs. extend/unpack)

  Show description for 7162ad7
  pcordes

  committed
  7162ad7

  Copy full SHA for 7162ad7

• i128 hsum: tidy comments

  pcordes

  committed
  0397c63

  Copy full SHA for 0397c63

• i128 hsums: fix syntax error with -m32

  Show description for 7a2a570
  pcordes

  committed
  7a2a570

  Copy full SHA for 7a2a570

• horizontal_add_x (Vec4i): save a movdqa with pmovsx. Also opens up ILP

  Show description for 56dae2f
  pcordes

  committed
  56dae2f

  Copy full SHA for 56dae2f

• i128 hsums: Remove SSSE3 phadd ifdefs, and macro the shuffles to save a byte with AVX

  Show description for ad44f29
  pcordes

  committed
  ad44f29

  Copy full SHA for ad44f29

Commits on Jun 1, 2016

• i128 hsum: separator comments

  pcordes

  committed
  abe7f24

  Copy full SHA for abe7f24

• i128: define most horizontal_sum_x functions in terms of the next wider vector

  Show description for 98d1347
  pcordes

  committed
  98d1347

  Copy full SHA for 98d1347

• i128: fiddle with hsum comments

  pcordes

  committed
  e4583a0

  Copy full SHA for e4583a0

• i128: move Vec8 horizontal add functions down so things are in descending order of element size

  pcordes

  committed
  c220cc9

  Copy full SHA for c220cc9

• i128: extract_lowi64: factor out the 32bit workaround for getting the low [u]int64_t

  Show description for d1dfb51
  pcordes

  committed
  d1dfb51

  Copy full SHA for d1dfb51

• i128: Bugfix: Vec8us compares and some selects used signed instead of boolean vectors

  Show description for 77c189e
  pcordes

  committed
  77c189e

  Copy full SHA for 77c189e

• f256: add a comment about assuming boolean vectors really are

  pcordes

  committed
  cae0769

  Copy full SHA for cae0769

• i256: fiddle with a couple hsum things: hadd_x(Vec32c) uses Vec8s

  pcordes

  committed
  57b1a9c

  Copy full SHA for 57b1a9c

• i256: fix hi128i/lo128i macros

  pcordes

  committed
  a8d09e3

  Copy full SHA for a8d09e3

• i128: 8sb 4ib and 2qb horizontal and/or use the 16cb def, so we can tune it

  Show description for 63e4807
  pcordes

  committed
  63e4807

  Copy full SHA for 63e4807

• 128i: group the horizontal_add functions together at the bottom

  Show description for 37e9d49
  pcordes

  committed
  37e9d49

  Copy full SHA for 37e9d49

Commits on May 31, 2016

• Work-in-progress: i256: move some definitions lower down so they can reuse existing code

  Show description for b2e0960
  pcordes

  committed
  b2e0960

  Copy full SHA for b2e0960

• Changelog: update for recent patches

  pcordes

  committed
  2b7bcaf

  Copy full SHA for 2b7bcaf

• f256: more efficient horizontal add

  Show description for bafcbf9
  pcordes

  committed
  bafcbf9

  Copy full SHA for bafcbf9

• 128bit integer abs using non-destructive _mm_abs to give the compiler more options

  Show description for b71cf88
  pcordes

  committed
  b71cf88

  Copy full SHA for b71cf88

Commits on May 30, 2016

• Vec128b horizontal bitwise and/or: use pcmpeq

  Show description for 9bf1f64
  pcordes

  committed
  9bf1f64

  Copy full SHA for 9bf1f64

Commits on May 3, 2016

• Update to v1.22

  darealshinji

  committed
  2caaa9b

  Copy full SHA for 2caaa9b

Commits on Apr 29, 2016

• Update to v1.21

  darealshinji

  committed
  5663c03

  Copy full SHA for 5663c03

Commits on Dec 18, 2015

• Apply changes from 2015-Dec-15 version

  darealshinji

  committed
  8c4fd67

  Copy full SHA for 8c4fd67

Commits on Dec 5, 2015

• Update to v1.20

  darealshinji

  committed
  7b5e9a2

  Copy full SHA for 7b5e9a2

Pagination

Previous

Next