search

kiwifig / libyuv

Automatically exported from code.google.com/p/libyuv
BSD 3-Clause "New" or "Revised" License
0 stars 0 forks source link

ARGBToI411 test fails on ARM platform #491

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago 
What steps will reproduce the problem?
1. build libyuvtest from master branch
2. run libyuvtest with following command on Android device:
LIBYUV_WIDTH=130 LIBYUV_HEIGHT=80 LIBYUV_REPEAT=1 LIBYUV_DISABLE_NEON=0 
./libyuv_unittest
3.

What is the expected output? What do you see instead?
Expected: Pass all tests.
Instead:
[  FAILED  ] 4 tests, listed below:
[  FAILED  ] libyuvTest.ARGBToI411_Any
[  FAILED  ] libyuvTest.ARGBToI411_Unaligned
[  FAILED  ] libyuvTest.ARGBToI411_Invert
[  FAILED  ] libyuvTest.ARGBToI411_Opt

What version of the product are you using? On what operating system?

Please provide any additional information below.

Original issue reported on code.google.com by zhongwei...@linaro.org on 15 Sep 2015 at 7:55

GoogleCodeExporter commented 9 years ago 
Passes on Intel

LIBYUV_WIDTH=130 LIBYUV_HEIGHT=80 out/Release/libyuv_unittest 
--gtest_filter=*ARGBToI411*
Note: Google Test filter = *ARGBToI411*
[==========] Running 4 tests from 1 test case.
[----------] Global test environment set-up.
[----------] 4 tests from libyuvTest
[ RUN      ] libyuvTest.ARGBToI411_Any
[       OK ] libyuvTest.ARGBToI411_Any (2 ms)
[ RUN      ] libyuvTest.ARGBToI411_Unaligned
[       OK ] libyuvTest.ARGBToI411_Unaligned (1 ms)
[ RUN      ] libyuvTest.ARGBToI411_Invert
[       OK ] libyuvTest.ARGBToI411_Invert (1 ms)
[ RUN      ] libyuvTest.ARGBToI411_Opt
[       OK ] libyuvTest.ARGBToI411_Opt (1 ms)
[----------] 4 tests from libyuvTest (5 ms total)

[----------] Global test environment tear-down
[==========] 4 tests from 1 test case ran. (5 ms total)
[  PASSED  ] 4 tests.

Original comment by fbarch...@google.com on 17 Sep 2015 at 8:04

GoogleCodeExporter commented 9 years ago

Original comment by fbarch...@google.com on 17 Sep 2015 at 8:09

GoogleCodeExporter commented 9 years ago

Original comment by fbarch...@chromium.org on 12 Oct 2015 at 11:48

GoogleCodeExporter commented 9 years ago 
With default parameters, this function passes
[ RUN      ] LibYUVConvertTest.ARGBToI411_Any
[       OK ] LibYUVConvertTest.ARGBToI411_Any (5 ms)
[ RUN      ] LibYUVConvertTest.ARGBToI411_Unaligned
[       OK ] LibYUVConvertTest.ARGBToI411_Unaligned (4 ms)
[ RUN      ] LibYUVConvertTest.ARGBToI411_Invert
[       OK ] LibYUVConvertTest.ARGBToI411_Invert (4 ms)
[ RUN      ] LibYUVConvertTest.ARGBToI411_Opt
[       OK ] LibYUVConvertTest.ARGBToI411_Opt (4 ms)
[----------] 4 tests from LibYUVConvertTest (18 ms total)

Original comment by fbarch...@chromium.org on 14 Oct 2015 at 10:33

GoogleCodeExporter commented 9 years ago 
LIBYUV_WIDTH=130 LIBYUV_HEIGHT=80 LIBYUV_REPEAT=1 LIBYUV_DISABLE_ASM=1 
out/Release/libyuv_unittest

This is the Intel code

     0000000000556f00 <ARGBToUV411Row_C>:
       push   %r13
       cmp    $0x3,%ecx
       push   %r12
       push   %rbp
       push   %rbx
     ↓ jle    2c3
       lea    -0x4(%rcx),%r12d
       mov    %rsi,%rbx
       mov    %rdi,%rax
       shr    $0x2,%r12d
       add    $0x1,%r12
       lea    (%rdx,%r12,1),%rbp
       nop

 28:   movzbl (%rax),%r9d
       movzbl 0x4(%rax),%r8d
       add    $0x1,%rdx
       movzbl 0x1(%rax),%r10d
       movzbl 0x2(%rax),%r11d
       add    $0x1,%rbx
       add    $0x10,%rax
       add    %r9d,%r8d
       movzbl -0x8(%rax),%r9d
       add    %r9d,%r8d
       movzbl -0x4(%rax),%r9d
       add    %r9d,%r8d
       movzbl -0xb(%rax),%r9d
       sar    $0x2,%r8d
       mov    %r8d,%r13d
       lea    (%r8,%r8,8),%r8d
       add    %r10d,%r9d
       movzbl -0x7(%rax),%r10d
       add    %r8d,%r8d
       add    %r10d,%r9d
       movzbl -0x3(%rax),%r10d
       add    %r10d,%r9d
       movzbl -0xa(%rax),%r10d
       sar    $0x2,%r9d
       add    %r11d,%r10d
       movzbl -0x6(%rax),%r11d
       add    %r11d,%r10d
       movzbl -0x2(%rax),%r11d
       add    %r11d,%r10d
       mov    $0x70,%r11d
       imul   %r11d,%r13d
       sar    $0x2,%r10d
       imul   $0xffffffb6,%r9d,%r11d
       add    %r11d,%r13d
       imul   $0xffffffda,%r10d,%r11d
       lea    0x8080(%r13,%r11,1),%r11d
       mov    $0xffffffa2,%r13d
       imul   %r13d,%r9d
       sar    $0x8,%r11d
       mov    %r11b,-0x1(%rbx)
       mov    $0x70,%r11d
       imul   %r11d,%r10d
       add    %r10d,%r9d
       sub    %r8d,%r9d
       add    $0x8080,%r9d
       sar    $0x8,%r9d
       mov    %r9b,-0x1(%rdx)
       cmp    %rbp,%rdx
     ↑ jne    28

       mov    %r12,%rax
       add    %r12,%rsi
       shl    $0x4,%rax
       add    %rax,%rdi
104:   and    $0x3,%ecx
       cmp    $0x3,%ecx
     ↓ je     198
       cmp    $0x2,%ecx
     ↓ je     238
       cmp    $0x1,%ecx
     ↓ je     128
       pop    %rbx
       pop    %rbp
       pop    %r12
       pop    %r13
     ← retq
       nop

Original comment by fbarch...@chromium.org on 16 Oct 2015 at 6:16

GoogleCodeExporter commented 9 years ago 
I've reproduced the error... had to hard code the unittest/unittest.cc 
dimensions to 130x80.  130x72 also fails... its the width not being a multiple 
of 4.

411 subsampling is really not used in practice, so it may be better to remove 
the functionality than spend time fixing it.

Original comment by fbarch...@google.com on 20 Oct 2015 at 1:37

GoogleCodeExporter commented 9 years ago 
#if defined(HAS_ARGBTOYROW_NEON)
  if (TestCpuFlag(kCpuHasNEON)) {
    ARGBToYRow = ARGBToYRow_Any_NEON;
    if (IS_ALIGNED(width, 8)) {
      ARGBToYRow = ARGBToYRow_NEON;
    }
  }
#endif
#if defined(HAS_ARGBTOUV411ROW_NEON)
  if (TestCpuFlag(kCpuHasNEON)) {
    ARGBToUV411Row = ARGBToUV411Row_Any_NEON;
    if (IS_ALIGNED(width, 32)) {
      ARGBToUV411Row = ARGBToUV411Row_NEON;
    }
  }
#endif

Original comment by fbarch...@google.com on 20 Oct 2015 at 1:45

GoogleCodeExporter commented 9 years ago 
There are 2 issues with ARGBTo411.
1. any handles odd 422 by duplicating the last ARGB.  It doesnt support 
remainder of 2 or 3 bytes.
// Any 1 to 2.  Outputs UV planes.
#define ANY12(NAMEANY, ANY_SIMD, UVSHIFT, BPP, DUVSHIFT, MASK)                 \
    void NAMEANY(const uint8* src_ptr, uint8* dst_u, uint8* dst_v, int width) {\
      SIMD_ALIGNED(uint8 temp[128 * 3]);                                       \
      memset(temp, 0, 128);  /* for msan */                                    \
      int r = width & MASK;                                                    \
      int n = width & ~MASK;                                                   \
      if (n > 0) {                                                             \
        ANY_SIMD(src_ptr, dst_u, dst_v, n);                                    \
      }                                                                        \
      memcpy(temp, src_ptr  + (n >> UVSHIFT) * BPP, SS(r, UVSHIFT) * BPP);     \
      if ((width & 1) && BPP == 4) {  /* repeat last 4 bytes for subsampler */ \
        memcpy(temp + SS(r, UVSHIFT) * BPP,                                    \
               temp + SS(r, UVSHIFT) * BPP - BPP, 4);                          \
      }                                                                        \
      ANY_SIMD(temp, temp + 128, temp + 256, MASK + 1);                        \
      memcpy(dst_u + (n >> DUVSHIFT), temp + 128, SS(r, DUVSHIFT));            \
      memcpy(dst_v + (n >> DUVSHIFT), temp + 256, SS(r, DUVSHIFT));            \
    }
ANY12(ARGBToUV444Row_Any_NEON, ARGBToUV444Row_NEON, 0, 4, 0, 7)
ANY12(ARGBToUV422Row_Any_NEON, ARGBToUV422Row_NEON, 0, 4, 1, 15)
ANY12(ARGBToUV411Row_Any_NEON, ARGBToUV411Row_NEON, 0, 4, 2, 31)

so the 411 averages in zeros when using neon for remainder of 2 or 3.

2. The C version handles remainder differently than Neon

void ARGBToUV411Row_C(const uint8* src_argb,
                      uint8* dst_u, uint8* dst_v, int width) {
  int x;
  for (x = 0; x < width - 3; x += 4) {
    uint8 ab = (src_argb[0] + src_argb[4] + src_argb[8] + src_argb[12]) >> 2;
    uint8 ag = (src_argb[1] + src_argb[5] + src_argb[9] + src_argb[13]) >> 2;
    uint8 ar = (src_argb[2] + src_argb[6] + src_argb[10] + src_argb[14]) >> 2;
    dst_u[0] = RGBToU(ar, ag, ab);
    dst_v[0] = RGBToV(ar, ag, ab);
    src_argb += 16;
    dst_u += 1;
    dst_v += 1;
  }
  if ((width & 3) == 3) {
    uint8 ab = (src_argb[0] + src_argb[4] + src_argb[8]) / 3;
    uint8 ag = (src_argb[1] + src_argb[5] + src_argb[9]) / 3;
    uint8 ar = (src_argb[2] + src_argb[6] + src_argb[10]) / 3;
    dst_u[0] = RGBToU(ar, ag, ab);
    dst_v[0] = RGBToV(ar, ag, ab);
  } else if ((width & 3) == 2) {
    uint8 ab = (src_argb[0] + src_argb[4]) >> 1;
    uint8 ag = (src_argb[1] + src_argb[5]) >> 1;
    uint8 ar = (src_argb[2] + src_argb[6]) >> 1;
    dst_u[0] = RGBToU(ar, ag, ab);
    dst_v[0] = RGBToV(ar, ag, ab);
  } else if ((width & 3) == 1) {
    uint8 ab = src_argb[0];
    uint8 ag = src_argb[1];
    uint8 ar = src_argb[2];
    dst_u[0] = RGBToU(ar, ag, ab);
    dst_v[0] = RGBToV(ar, ag, ab);
  }
}

The Any function used to use C, so it would have been self consistent, but now 
it uses SIMD the difference causes failure.

2 solutions:
1. C and ANY should match behavior on odd width
2. remove 411 support.  422 is easy enough and popular but 411 is difficult and 
unpopular.

Original comment by fbarch...@chromium.org on 21 Oct 2015 at 6:49

GoogleCodeExporter commented 9 years ago 
util/android/test_runner.py gtest -t 1800 -s libyuv_unittest --verbose 
--release --gtest_filter=*ARGBToI4*_Opt -a "--libyuv_width=642 
--libyuv_height=360 --libyuv_repeat=1000 --libyuv_flags=0"

I   13.367s run_tests_on_device(HT4A2JT03762)  [==========] Running 5 tests 
from 1 test case.
I   13.367s run_tests_on_device(HT4A2JT03762)  [----------] Global test 
environment set-up.
I   13.367s run_tests_on_device(HT4A2JT03762)  [----------] 5 tests from 
LibYUVConvertTest
I   13.367s run_tests_on_device(HT4A2JT03762)  [ RUN      ] 
LibYUVConvertTest.ARGBToI420_Opt
I   13.367s run_tests_on_device(HT4A2JT03762)  [       OK ] 
LibYUVConvertTest.ARGBToI420_Opt (792 ms)
I   13.367s run_tests_on_device(HT4A2JT03762)  [ RUN      ] 
LibYUVConvertTest.ARGBToI411_Opt
I   13.367s run_tests_on_device(HT4A2JT03762)  [       OK ] 
LibYUVConvertTest.ARGBToI411_Opt (896 ms)
I   13.367s run_tests_on_device(HT4A2JT03762)  [ RUN      ] 
LibYUVConvertTest.ARGBToI422_Opt
I   13.367s run_tests_on_device(HT4A2JT03762)  [       OK ] 
LibYUVConvertTest.ARGBToI422_Opt (661 ms)
I   13.367s run_tests_on_device(HT4A2JT03762)  [ RUN      ] 
LibYUVConvertTest.ARGBToI444_Opt
I   13.367s run_tests_on_device(HT4A2JT03762)  [       OK ] 
LibYUVConvertTest.ARGBToI444_Opt (743 ms)
I   13.368s run_tests_on_device(HT4A2JT03762)  [ RUN      ] 
LibYUVConvertTest.ARGBToI400_Opt
I   13.368s run_tests_on_device(HT4A2JT03762)  [       OK ] 
LibYUVConvertTest.ARGBToI400_Opt (345 ms)
I   13.368s run_tests_on_device(HT4A2JT03762)  [----------] 5 tests from 
LibYUVConvertTest (3440 ms total)
I   13.368s run_tests_on_device(HT4A2JT03762)  
I   13.368s run_tests_on_device(HT4A2JT03762)  [----------] Global test 
environment tear-down
I   13.368s run_tests_on_device(HT4A2JT03762)  [==========] 5 tests from 1 test 
case ran. (3441 ms total)
I   13.368s run_tests_on_device(HT4A2JT03762)  [  PASSED  ] 5 tests.

Original comment by fbarch...@google.com on 21 Oct 2015 at 7:53