Mismatch decoded output for C-only build and build with full optimizaiton.

[GoogleCodeExporter]

What is the expected behavior? What do you see instead?
vpxdec produces different result for C-only build

What version are you using? On what operating system?
TBD

Can you reproduce using the vpxdec or vpxenc tools? What command line are
you using?
Yes
Can you reproduce using the vpxdec or vpxenc tools? What command line are
you using?
Yes

Please provide any additional information below.
attached file and: 
debug build with c only gets the following md5: 
bcfa5757625391d51caefeefa1b20519
regular build has the following md5:
31ac8443b3069121e02bc58574a8dc05 

Please provide any additional information below.
attached file and: 
debug build with c only gets the following md5: 
bcfa5757625391d51caefeefa1b20519
regular build has the following md5:
31ac8443b3069121e02bc58574a8dc05 

Original issue reported on code.google.com by ya...@google.com on 2 Oct 2014 at 6:43

Attachments:

• mismatching_stream.vp9

[GoogleCodeExporter]

The mismatch happens on the second frame mi_row=20 mi_col=0, (167, 11) and 
(168, 11). D63 and 16x16 transform. It seems to be due to the mismatch between 
vp9_iht16x16_256_add_sse2 and vp9_iht16x16_256_add_c. 

Original comment by ya...@google.com on 3 Oct 2014 at 1:12

• Changed state: Assigned

[GoogleCodeExporter]

https://gerrit.chromium.org/gerrit/#/c/71741/

Original comment by ya...@google.com on 7 Oct 2014 at 4:41

[GoogleCodeExporter]

The added check should report out-of-range coefficients for invalid input 
streams when build with --enable-coefficient-range-checking --enable-debug. 

Original comment by ya...@google.com on 7 Oct 2014 at 7:52

[GoogleCodeExporter]

Original comment by ya...@google.com on 7 Oct 2014 at 7:52

• Changed state: Fixed