MantisBT - ITK |
| View Issue Details |
|
| ID | Project | Category | View Status | Date Submitted | Last Update |
| 0007207 | ITK | | public | 2008-06-19 09:49 | 2010-11-02 12:54 |
|
| Reporter | Hauke Heibel | |
| Assigned To | Bradley Lowekamp | |
| Priority | normal | Severity | major | Reproducibility | always |
| Status | assigned | Resolution | open | |
| Platform | | OS | | OS Version | |
| Product Version | | |
| Target Version | | Fixed in Version | | |
| Resolution Date | |
| Sprint | |
| Sprint Status | |
|
| Summary | 0007207: vnl_sse<double>::matrix_x_vector(...) yields wrong result when VNL_CONFIG_ENABLE_SSE2:BOOL=ON |
| Description | I build ITK with VNL_CONFIG_ENABLE_SSE2:BOOL=ON (the CMakeCache.txt is attached in the .zip file).
When running the project RUN_TESTS, I recognized some of the VNL related tests failing and I started digging into test no. 35 vnl_test_cholesky. In line 43, b=A*x0, I recognized that the result of the multiplication yielded a wrong result. The result has been verified with a single 3x3 time 3x1 matrix vector multiplication, using fixed values, that have been precomputed with Matlab.
A test showing this result is attached as well. |
| Steps To Reproduce | |
| Additional Information | Build environment:
- Windows XP Professional (english)
- Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 14.00.50727.762
- Microsoft (R) Incremental Linker Version 8.00.50727.762
System:
- Intel Mobile Core 2 Duo T7400 (Merom)
- Instructions: MMX, SSE, SSE2, SSE3, SSSE3, EM64T
- L1 Cache: 32 KBytes, L2 Cache: 4096 KBytes
- 2048 MBytes DDR2 RAM
Important notes:
I did not yet with VNL_CONFIG_ENABLE_SSE2:BOOL=OFF, it's just an educated guess.
I did not check for vnl_sse<float>::matrix_x_vector(...) |
| Tags | No tags attached. |
| Relationships | | related to | 0007214 | closed | Luis Ibanez | ITK CVS does not build with VNL_CONFIG_ENABLE_SSE2 ON when building Mac Universal Binaries |
|
| Attached Files |  vnl_sse_bug.zip (25,236) 2008-06-19 09:49
https://public.kitware.com/Bug/file/1539/vnl_sse_bug.zip |
|
| Issue History |
| Date Modified | Username | Field | Change |
| 2008-06-19 09:49 | Hauke Heibel | New Issue | |
| 2008-06-19 09:49 | Hauke Heibel | File Added: vnl_sse_bug.zip | |
| 2008-06-20 10:16 | Sean McBride | Note Added: 0012438 | |
| 2008-06-20 10:20 | Sean McBride | Relationship added | related to 0007214 |
| 2010-11-02 12:47 | Hans Johnson | Status | new => assigned |
| 2010-11-02 12:47 | Hans Johnson | Assigned To | => Bradley Lowekamp |
| 2010-11-02 12:49 | Hans Johnson | Note Added: 0022786 | |
| 2010-11-02 12:54 | Sean McBride | Note Added: 0022790 | |
|
Notes |
|
|
(0012438)
|
|
Sean McBride
|
|
2008-06-20 10:16
|
|
I reproed most of your test failures:
35/1258 Testing vnl_test_cholesky ***Failed
44/1258 Testing vnl_test_ldl_cholesky ***Failed
49/1258 Testing vnl_test_qr ***Failed
61/1258 Testing vnl_test_svd ***Failed
91/1258 Testing vnl_test_quaternion Passed
vnl_test_quaternion is passing however. I am using Mac OS X 10.5.3, gcc 4.0, and running in 64 bit. |
|
|
|
(0022786)
|
|
Hans Johnson
|
|
2010-11-02 12:49
|
|
Brad,
This seems related to other work you are interested in. This bug may already be done by the related bug 0007214 , could you please confirm?
Thanks,
Hans |
|
|
|
(0022790)
|
|
Sean McBride
|
|
2010-11-02 12:54
|
|
|
Hans, the fix for 0007214 will not have fixed this bug. |
|