BeagleBoard/GSoC/2010 Projects/FFTW - Revision history

Pmeerw: /* Build Instructions */

2011-07-05T15:25:55Z

‎Build Instructions

2011-07-05T14:30:23Z

‎Performance Counter

2011-04-13T05:29:22Z

Added to category BeagleBoard and GSOC

2010-08-23T15:25:51Z

‎Test Instructions

2010-08-23T15:24:51Z

‎Test Instructions

2010-08-23T15:23:59Z

‎Test Instructions

2010-08-23T15:23:08Z

‎Build Instructions: seperated into section Test Instructions

2010-08-20T12:41:42Z

‎Build Instructions

2010-08-20T12:30:31Z

‎Build Instructions

2010-08-20T12:29:56Z

‎Build Instructions

@@ Line 112: / Line 112: @@
 sh bootstrap.sh
 </pre>
-and ignore any messages that you see. The additional configure options are '--enable-armv7a-cycle-counter', '--enable-neon', '--enable-neon-intrinsics', '--enable-ffmpeg', and '--enable-ffmpeg-test'. I would strongly suggest that you enable the cycle counter, otherwise your FFTW build will always use ESTIMATE mode, which will likely result in degraded performance.
+and ignore any messages that you see. The additional configure options are '--enable-armv7-cycle-counter', '--enable-neon', '--enable-neon-intrinsics', '--enable-ffmpeg', and '--enable-ffmpeg-test'. I would strongly suggest that you enable the cycle counter, otherwise your FFTW build will always use ESTIMATE mode, which will likely result in degraded performance.
 '''IMPORTANT'''
-If you choose to use '--enable-armv7a-cycle-counter', you will need a recent kernel with the Mans Rullgard's USER_PMON patch applied. You can obtain his patch from [http://git.mansr.com/?p=linux-omap;a=commitdiff;h=5170038 here].
+If you choose to use '--enable-armv7-cycle-counter', you will need a recent kernel with the Mans Rullgard's USER_PMON patch applied. You can obtain his patch from [http://git.mansr.com/?p=linux-omap;a=commitdiff;h=5170038 here].
 '''Patch FFTW'''

@@ Line 59: / Line 59: @@
 ==== Performance Counter ====
-The cycle counter (sometimes called performance counter) was originally given to me in a demo app by my mentor, Mans, and I basically just had to coerce it into the FFTW codebase and ensure that it was getting turned off and on correctly during regular FFTW usage (which required a bit of debugging). You can enable the cycle counter with '--enable-armv7a-cycle-counter'. The cycle counter is crucial to having an optimally performing FFTW library. If it is not enabled, then FFTW purely uses estimation methods for optimization.
+The cycle counter (sometimes called performance counter) was originally given to me in a demo app by my mentor, Mans, and I basically just had to coerce it into the FFTW codebase and ensure that it was getting turned off and on correctly during regular FFTW usage (which required a bit of debugging). You can enable the cycle counter with '--enable-armv7-cycle-counter'. The cycle counter is crucial to having an optimally performing FFTW library. If it is not enabled, then FFTW purely uses estimation methods for optimization.
 ==== Speed, Speed and more Speed ====

@@ Line 211: / Line 211: @@
 </pre>
-If you do indeed have a recent version of ffmpeg, then download my octave demo to your beagleboard, extract it in the root directory, and then run the demo script.
+If you do indeed have a recent version of ffmpeg, then download [http://code.google.com/p/fftw-neon/downloads/list my octave demo] to your beagleboard, extract it in the root directory, and then run the demo script.
 <pre>

@@ Line 205: / Line 205: @@
 '''GNU Octave'''
-Alternatively you could simply use Octave's 'fft' command. I've set up a fairly straight-forward demonstration that will compare the average execution times for power-of-two cases. For this demonstration, there are a few prerequisites; you need to have GNU Octave, fftw3, and a recent version of ffmpeg (from the git repository) installed. The first two should be fairly simple to test, but to test out the ffmpeg version (specifically libavcodec), se the command below and ensure the following returns a valid symbol.
+Alternatively you could simply use Octave's 'fft' command. I've set up a fairly straight-forward demonstration that will compare the average execution times for power-of-two cases. For this demonstration, there are a few prerequisites; you need to have GNU Octave, fftw3, and a recent version of ffmpeg (from the git repository) installed. The first two should be fairly simple to test, but to test out the ffmpeg version (specifically libavcodec), use the command below and ensure the following returns a valid symbol.
 <pre>

@@ Line 177: / Line 177: @@
 </pre>
-==== Test Instructions ====
+=== Test Instructions ===
 '''Use libfftw3f.so Directly'''

@@ Line 177: / Line 177: @@
 </pre>
-'''A Sample Test Program'''
+==== Test Instructions ====
-You can write your own apps with a sample directly from the fftw documentation.
+'''Use libfftw3f.so Directly'''
 <pre>
@@ Line 199: / Line 203: @@
 </pre>
-Alternatively you could simply use GNU Octave's 'fft' command.
+'''GNU Octave'''
 <pre>
-@@@
+tar xpjf fftw-neon-demo-octave*.tar.bz2 -C /
 </pre>

@@ Line 179: / Line 179: @@
 '''A Sample Test Program'''
-You can write your own apps with a sample directly from the fftw documentation. Alternatively you could simply use GNU Octave's 'fft' command.
+You can write your own apps with a sample directly from the fftw documentation.
 <pre>
@@ Line 197: / Line 197: @@
       fftwf_free(in); fftwf_free(out);
+}
 </pre>

@@ Line 119: / Line 119: @@
 '''Patch FFTW'''
 Change to the base directory of your FFTW source and run the following.
 <pre>

@@ Line 79: / Line 79: @@
 '''Prerequisites'''
-# a cross-compiler may be downloaded from [www.codesourcery.com CodeSourcery]
+# a cross-compiler may be downloaded from [http://www.codesourcery.com CodeSourcery]
-# a NEON-enabled ARM device for testing (such as the BeagleBoard)
+# a NEON-enabled ARM device for testing such as the [http://beagleboard.org BeagleBoard]
 # a temporary directory (substituted for /tmp/my-install)