• Locked Locked
  • Replies 2 replies
  • Subscribers 119 subscribers
  • Views 2387 views
  • Users 0 members are here
2025 survey

We are running a survey to help us improve the experience for all of our members. If you see the survey appear, please take the time to tell us about your experience if you can.

Options
Related
How was your experience today?
This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

Bug in ARM assembler optimezed OpenMAX implementation?

bobosv
Note: This was originally posted on 28th October 2009 at http://forums.arm.com

I believed i've found a bug in the assembler optimized version of the
OpenMAX implementation, can anyone here confirm this bug?


First, i begin with the software version i'm using!

* Arm Assembler optimized openmax (OX001-BU-00010-r2p0-00rel0)
Filename: ARM_OpenMax_OX001-BU-0010-r2p0-00rel0.tgz
md5sum: 2eb4fb315e5935bcfa5cf9ed28e63954

* ARM C implementation of openmax (OX000-BU-00010-r1p0-00bet0)
Filename: arm-download-98665.tgz
md5sum c093bd73e6e48c13952855a63ec5baab


* Compiler Toolchains
RVDS: RVDS v 4.0

$ arm-eabi-gcc --version
arm-eabi-gcc (GCC) 4.4.0
Copyright © 2009 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.



* Report
The function omxSP_FIR_Direct_S16 provides different results when
comparing the C implementation and the Assembler optimized version
when TAP length is over 9 TAPS!
Also, TAP length of 4, 8, 12, 16 results in segmention fault on my
target (Android telephone HTC Magic)

** Specifics

*** TAP length = 13
OMX_S16 FIR_TESTVECTOR[100] = {
  0,0,0,0,0,0,0,0,0,0,
  0,0,0,0,0,0,0,0,0,0,
  10000,0,0,0,0,0,0,0,0,
  0,0,0,0,0,0,0,0,0,0,
  0,0,0,0,0,0,0,0,0,0
};

const OMX_S16 Coeff[13]={16384, 8192,4096,2048,1024,512,256,128,64,32,16,16,0};
int Coefflen=13;

delayline is zeroed
delaylinIndex 0

result = omxSP_FIR_Direct_S16(FIR_TESTVECTOR, Destination, len, Coeff,
Coefflen,  delayline, delaylineIndex); 


Provides the result:
**** CORRECT C implementation  (performed on PC)
...
Destination[18]==0
Destination[19]==0
Destination[20]==5000
Destination[21]==2500
Destination[22]==1250
Destination[23]==625
Destination[24]==313
Destination[25]==156
Destination[26]==78
Destination[27]==39
Destination[28]==20
Destination[29]==10
Destination[30]==5
Destination[31]==5
Destination[32]==0
...

**** FAILED!! Assembler implementation (performed on target, ARM1176)
Destination[18]==0
Destination[19]==0
Destination[20]==-2
Destination[21]==-6641
Destination[22]==-9926
Destination[23]==-2435
Destination[24]==-7426
Destination[25]==-2436
Destination[26]==-4177
Destination[27]==-2418
Destination[28]==-7426
Destination[29]==-2319
Destination[30]==0
Destination[31]==-2435
Destination[32]==0



*** TAP length = 9
OMX_S16 FIR_TESTVECTOR[100] = {
  0,0,0,0,0,0,0,0,0,0,
  0,0,0,0,0,0,0,0,0,0,
  10000,0,0,0,0,0,0,0,0,
  0,0,0,0,0,0,0,0,0,0,
  0,0,0,0,0,0,0,0,0,0
};

const OMX_S16 Coeff[13]={16384, 8192,4096,2048,1024,512,256,128,64,32,16,16,0};
int Coefflen=9;

delayline is zeroed
delaylinIndex 0

result = omxSP_FIR_Direct_S16(FIR_TESTVECTOR, Destination, len, Coeff,
Coefflen,  delayline, delaylineIndex); 


Provides the result:
**** CORRECT C implementation  (performed on PC)
...
Destination[18]==0
Destination[19]==0
Destination[20]==5000
Destination[21]==2500
Destination[22]==1250
Destination[23]==625
Destination[24]==313
Destination[25]==156
Destination[26]==78
Destination[27]==39
Destination[28]==20
Destination[29]==0
Destination[30]==0
Destination[31]==0
Destination[32]==0
...

**** CORRECT Assembler implementation (performed on target, ARM1176)
Destination[18]==0
Destination[19]==0
Destination[20]==5000
Destination[21]==2500
Destination[22]==1250
Destination[23]==625
Destination[24]==313
Destination[25]==156
Destination[26]==78
Destination[27]==39
Destination[28]==20
Destination[29]==0
Destination[30]==0
Destination[31]==0
Destination[32]==0

Do anyone here have possibility to confirm this bug??


/Bo Svangård,
System Designer, EmbeddedArt AB, Swden
Telephon +46-736 631 628
Parents
  • Note: This was originally posted on 29th October 2009 at http://forums.arm.com

    Hello,
    I am using OX002-BU-0010-r2p0-00rel0 for SP function and I have not tried to do comparison with C version of it.
    Have you used omxSP_FFTFwd_CToC_SC32_Sfs and omxSP_FFTInv_CToC_SC32_Sfs  SP functions ?

    no

    I want to use float values as input of this functions ?
    ok

    Do you have any idea about that?

    no

    Thanks,
    rach



    I've solved my problem, more on that later, The _assembler_ function omxSP_FIR_Direct_S16 do contain errors when using a taplength greater than 11 taps.
    My solution to the problem is to user the _C_ function omxSP_FIR_Direct_S16 from ARM, which calls a function named omxSP_FIROne_Direct_S16
    Then when i link my program i link it with the _assembler_ version of omxSP_FIROne_Direct_S16, This provided my application with correct results of the FIR filter!!


    confirmation of my error considering the ARM function is still helpful, or some help where to find support of the assembler optimized functions and where to send patches, so the implementation may become better!!
Reply
  • Note: This was originally posted on 29th October 2009 at http://forums.arm.com

    Hello,
    I am using OX002-BU-0010-r2p0-00rel0 for SP function and I have not tried to do comparison with C version of it.
    Have you used omxSP_FFTFwd_CToC_SC32_Sfs and omxSP_FFTInv_CToC_SC32_Sfs  SP functions ?

    no

    I want to use float values as input of this functions ?
    ok

    Do you have any idea about that?

    no

    Thanks,
    rach



    I've solved my problem, more on that later, The _assembler_ function omxSP_FIR_Direct_S16 do contain errors when using a taplength greater than 11 taps.
    My solution to the problem is to user the _C_ function omxSP_FIR_Direct_S16 from ARM, which calls a function named omxSP_FIROne_Direct_S16
    Then when i link my program i link it with the _assembler_ version of omxSP_FIROne_Direct_S16, This provided my application with correct results of the FIR filter!!


    confirmation of my error considering the ARM function is still helpful, or some help where to find support of the assembler optimized functions and where to send patches, so the implementation may become better!!
Children
No data