search

tozhovez / cudpp

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

Tridiagonal fails for fp64 systems of 128 elements. #99

Closed GoogleCodeExporter closed 8 years ago

GoogleCodeExporter commented 8 years ago 
What steps will reproduce the problem?
On a system with a GeForce GTX 480 (and presumably similar GPUs, on CUDA 4.0, 
Windows 7 x64:

Run 
    cudpp_testrig -tridiagonal -n=1

What is the expected output? What do you see instead?
All should pass, but I get:

Using device 0:
GeForce GTX 480; global mem: 1543176192B; compute v2.0; clock: 1401000 kHz
Runing a fp32 CR-PCR tridiagonal solver solving 512 systems of 128 equations
GPU execution time: 0.320065 ms
CPU execution time: 1.303725 ms
test PASSED

Runing a fp64 CR-PCR tridiagonal solver solving 512 systems of 128 equations
GPU execution time: 0.493480 ms
CPU execution time: 2.379409 ms
test failed, error is larger than 0.001
test FAILED

1 tests failed

Original issue reported on code.google.com by harr...@gmail.com on 1 Aug 2011 at 3:04

GoogleCodeExporter commented 8 years ago 
Fixed

Original comment by harr...@gmail.com on 3 Aug 2011 at 6:23