search

dinhminhquoi / webm

Automatically exported from code.google.com/p/webm
0 stars 0 forks source link

[NEON] fdct8x8_test failure #716

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago 
v1.3.0-1442-g0eca2cd

FwdTrans8x8HT / vp9_iht8x8_64_add_neon

Tests are available:
https://gerrit.chromium.org/gerrit/#/c/68970/

---
[ RUN      ] DISABLED_NEON/FwdTrans8x8HT.RoundTripErrorCheck/0
test/fdct8x8_test.cc:167: Failure
Expected: (1) >= (max_error), actual: 1 vs 45369
Error: 8x8 FDCT/IDCT or FHT/IHT has an individual roundtrip error > 1
...
[ RUN      ] DISABLED_NEON/FwdTrans8x8HT.ExtremalCheck/0
test/fdct8x8_test.cc:207: Failure
Expected: (1) >= (max_error), actual: 1 vs 65025
Error: Extremal 8x8 FDCT/IDCT or FHT/IHT hasan individual roundtrip error > 1
test/fdct8x8_test.cc:211: Failure
Expected: (count_test_block/5) >= (total_error), actual: 20000 vs 1083478
Error: Extremal 8x8 FDCT/IDCT or FHT/IHT has average roundtrip error > 1/5 per 
block
test/fdct8x8_test.cc:207: Failure
Expected: (1) >= (max_error), actual: 1 vs 65025
...

Original issue reported on code.google.com by jz...@google.com on 26 Feb 2014 at 8:14

GoogleCodeExporter commented 9 years ago 
vp9_iht8x8_64_add_neon is using the q4-q7 registers but not saving/restoring 
them.

Original comment by fgalli...@google.com on 26 Feb 2014 at 4:48

GoogleCodeExporter commented 9 years ago 
Can a test for this be added to REGISTER_STATE_CHECK()?

Original comment by jz...@google.com on 26 Feb 2014 at 6:05

GoogleCodeExporter commented 9 years ago 
Consider taking this patch if perf is similar
https://gerrit.chromium.org/gerrit/#/c/68922/

Original comment by johannkoenig@google.com on 27 Feb 2014 at 3:24

GoogleCodeExporter commented 9 years ago 
ping

Original comment by jz...@google.com on 16 Jul 2014 at 6:48

GoogleCodeExporter commented 9 years ago

Original comment by hku...@google.com on 17 Jul 2014 at 6:34

GoogleCodeExporter commented 9 years ago 
still failing:
fdct4x4_test.cc:    DISABLED_NEON, Trans4x4HT,
fdct8x8_test.cc:    DISABLED_NEON, FwdTrans8x8HT,

Original comment by johannkoenig@google.com on 6 Nov 2014 at 1:20

GoogleCodeExporter commented 9 years ago 
Issue 898 has been merged into this issue.

Original comment by johannkoenig@google.com on 12 Dec 2014 at 5:23

GoogleCodeExporter commented 9 years ago 
https://gerrit.chromium.org/gerrit/68922

e test exercised tx_type 0 but the code is never called that way in the encoder:
void vp9_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest, 
int stride, int eob) {
  if (tx_type == DCT_DCT) {
    vp9_idct8x8_add(input, dest, stride, eob);
  } else {
    vp9_iht8x8_64_add(input, dest, stride, tx_type);
  }
}

The updated patch contains all the tx_types and calls the C code if the DCT_DCT 
is specified. It would probably be better to fix the test and the reference 
code to indicate this is not an expected code path.

Original comment by johannkoenig@google.com on 15 Dec 2014 at 10:55