1. 08 Nov, 2016 1 commit
  2. 07 Nov, 2016 1 commit
  3. 01 Nov, 2016 1 commit
  4. 27 Oct, 2016 1 commit
  5. 26 Oct, 2016 3 commits
  6. 25 Oct, 2016 3 commits
  7. 24 Oct, 2016 1 commit
    • Frank Barchard's avatar
      Add MSA optimized I422ToARGBRow_MSA and I422ToRGBARow_MSA functions · f5d5bd88
      Frank Barchard authored
      R=fbarchard@google.com
      BUG=libyuv:634
      
      Performance Gains :- (vs C vectorized)
      
      I422ToARGBRow_MSA     : ~1.6x
      I422ToRGBARow_MSA     : ~1.6x
      
      I422ToARGBRow_Any_MSA : ~1.58x
      I422ToRGBARow_Any_MSA : ~1.6x
      
      Performance Gains :- (vs C non-vectorized)
      
      I422ToARGBRow_MSA     : ~7x
      I422ToRGBARow_MSA     : ~7x
      
      I422ToARGBRow_Any_MSA : ~6.9x
      I422ToRGBARow_Any_MSA : ~6.8x
      
      Regarding performance measurement, We have created standalone tests which pass in row's data from a 1920x1080 filled buffer to both the C and MSA functions. And such N iterations are executed to get more accurate timings of C vs MSA.
      
      Review URL: https://codereview.chromium.org/2430313005 .
      f5d5bd88
  8. 21 Oct, 2016 1 commit
    • Frank Barchard's avatar
      scale by 1 for neon implemented · 451af5e9
      Frank Barchard authored
      void HalfFloat1Row_NEON(const uint16* src, uint16* dst, float, int width) {
        asm volatile (
        "1:                                          \n"
          MEMACCESS(0)
          "ld1        {v1.16b}, [%0], #16            \n"  // load 8 shorts
          "subs       %w2, %w2, #8                   \n"  // 8 pixels per loop
          "uxtl       v2.4s, v1.4h                   \n"  // 8 int's
          "uxtl2      v1.4s, v1.8h                   \n"
          "scvtf      v2.4s, v2.4s                   \n"  // 8 floats
          "scvtf      v1.4s, v1.4s                   \n"
          "fcvtn      v4.4h, v2.4s                   \n"  // 8 floatsgit
          "fcvtn2     v4.8h, v1.4s                   \n"
         MEMACCESS(1)
          "st1        {v4.16b}, [%1], #16            \n"  // store 8 shorts
          "b.gt       1b                             \n"
        : "+r"(src),    // %0
          "+r"(dst),    // %1
          "+r"(width)   // %2
        :
        : "cc", "memory", "v1", "v2", "v4"
        );
      }
      
      void HalfFloatRow_NEON(const uint16* src, uint16* dst, float scale, int width) {
        asm volatile (
        "1:                                          \n"
          MEMACCESS(0)
          "ld1        {v1.16b}, [%0], #16            \n"  // load 8 shorts
          "subs       %w2, %w2, #8                   \n"  // 8 pixels per loop
          "uxtl       v2.4s, v1.4h                   \n"  // 8 int's
          "uxtl2      v1.4s, v1.8h                   \n"
          "scvtf      v2.4s, v2.4s                   \n"  // 8 floats
          "scvtf      v1.4s, v1.4s                   \n"
          "fmul       v2.4s, v2.4s, %3.s[0]          \n"  // adjust exponent
          "fmul       v1.4s, v1.4s, %3.s[0]          \n"
          "uqshrn     v4.4h, v2.4s, #13              \n"  // isolate halffloat
          "uqshrn2    v4.8h, v1.4s, #13              \n"
         MEMACCESS(1)
          "st1        {v4.16b}, [%1], #16            \n"  // store 8 shorts
          "b.gt       1b                             \n"
        : "+r"(src),    // %0
          "+r"(dst),    // %1
          "+r"(width)   // %2
        : "w"(scale * 1.9259299444e-34f)    // %3
        : "cc", "memory", "v1", "v2", "v4"
        );
      }
      
      TEST=LibYUVPlanarTest.TestHalfFloatPlane_One
      BUG=libyuv:560
      R=hubbe@chromium.org
      
      Review URL: https://codereview.chromium.org/2430313008 .
      451af5e9
  9. 20 Oct, 2016 2 commits
  10. 19 Oct, 2016 1 commit
  11. 18 Oct, 2016 3 commits
  12. 17 Oct, 2016 3 commits
  13. 15 Oct, 2016 1 commit
  14. 14 Oct, 2016 2 commits
  15. 13 Oct, 2016 2 commits
  16. 12 Oct, 2016 2 commits
  17. 11 Oct, 2016 2 commits
  18. 08 Oct, 2016 2 commits
  19. 07 Oct, 2016 1 commit
  20. 06 Oct, 2016 1 commit
    • Frank Barchard's avatar
      YUY2ToI422_Any_Neon clean up to not require 16 pixels · 3b88a19a
      Frank Barchard authored
      YUY2ToI422_Any_Neon previously required 16 pixels and duplicated
      the last pixel.  The replication was not necessary after a previous
      change to treat YUY2 to 4 byte macro pixels.
      
      TBR=harryjin@google.com
      BUG=libyuv:648
      TESTED=util/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release --gtest_filter=*YUY2ToI422* -a "--libyuv_width=17 --libyuv_height=7 --libyuv_repeat=999 --libyuv_flags=1"
      
      Review URL: https://codereview.chromium.org/2399143002 .
      3b88a19a
  21. 05 Oct, 2016 1 commit
  22. 04 Oct, 2016 2 commits
  23. 03 Oct, 2016 1 commit
  24. 30 Sep, 2016 1 commit
  25. 29 Sep, 2016 1 commit