[Beignet] [BUG] built-in function pow conformance test fail

Sun, Yi yi.sun at intel.com
Wed Aug 28 01:12:40 PDT 2013


I failed to fix this issue. Since many math functions have involved __ocl_function_pow, I think this issue should affect a lot.
Anyone would like give me a favor on debugging this issue?

Thanks
  --Sun, Yi

From: Sun, Yi
Sent: Tuesday, August 20, 2013 3:07 PM
To: 'beignet at lists.freedesktop.org'
Cc: Jin, Gordon
Subject: [Beignet] [BUG] built-in function pow conformance test fail

A test case is implemented as attached.
Following are some failure example.

0/0: x:-20.500000, y:-20.500000 -> gpu:0.000000  cpu:-nan
1/0: x:-20.500000, y:-1.000000 -> gpu:-20.500000  cpu:-0.048780
2/0: x:-20.500000, y:-0.900000 -> gpu:-20.500000  cpu:-nan
3/0: x:-20.500000, y:-0.010000 -> gpu:-20.500000  cpu:-nan
4/0: x:-20.500000, y:0.000000 -> gpu:-20.500000  cpu:1.000000
5/0: x:-20.500000, y:0.010000 -> gpu:-20.500000  cpu:-nan
6/0: x:-20.500000, y:0.900000 -> gpu:-20.500000  cpu:-nan
8/0: x:-20.500000, y:20.500000 -> gpu:-20.500000  cpu:-nan
9/0: x:-1.000000, y:-20.500000 -> gpu:-20.500000  cpu:-nan
11/0: x:-1.000000, y:-0.900000 -> gpu:-1.000000  cpu:-nan
12/0: x:-1.000000, y:-0.010000 -> gpu:-1.000000  cpu:-nan
13/0: x:-1.000000, y:0.000000 -> gpu:-1.000000  cpu:1.000000
14/0: x:-1.000000, y:0.010000 -> gpu:-1.000000  cpu:-nan
15/0: x:-1.000000, y:0.900000 -> gpu:-1.000000  cpu:-nan
17/0: x:-1.000000, y:20.500000 -> gpu:-1.000000  cpu:-nan
18/0: x:-0.900000, y:-20.500000 -> gpu:-1.000000  cpu:-nan
19/0: x:-0.900000, y:-1.000000 -> gpu:-0.900000  cpu:-1.111111
20/0: x:-0.900000, y:-0.900000 -> gpu:-0.900000  cpu:-nan
21/0: x:-0.900000, y:-0.010000 -> gpu:-0.900000  cpu:-nan
22/0: x:-0.900000, y:0.000000 -> gpu:-0.900000  cpu:1.000000
23/0: x:-0.900000, y:0.010000 -> gpu:-0.900000  cpu:-nan
24/0: x:-0.900000, y:0.900000 -> gpu:-0.900000  cpu:-nan

Thanks
  --Sun, Yi

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/beignet/attachments/20130828/3ecbd588/attachment.html>


More information about the Beignet mailing list