Re: [SciPy-User] Maximum file size for .npz format?
Hey Francesc, On 2010-03-15, at 4:16 PM, Francesc Alted <faltet@pytables.org> wrote:
Good! So, I should start by looking at how it achieves that then :)
Actually I did build a 4-way universal binary of HDF5. Long story short, because of some quirks of the build process for HDF5 it isn't possible to add some -arch flags to CFLAGS -- you actually have to build it separately for each architecture and then do something like make install DESTDIR=path/for/arch and then stitch them together manually with otool. I'm also not sure of the correct way, if there is one, to handle "hdf5.settings" -- it's unclear to me whether programs ever look at this at build time/ runtime, or if it's just there for the user's convenience. An installer for the binaries I made is at http://www.cs.toronto.edu/~dwf/mirror/hdf5-1.8.4-quad.pkg I can write up some instructions if that would be helpful. David
A Tuesday 16 March 2010 16:05:05 David Warde-Farley escrigué:
Hey Francesc,
On 2010-03-15, at 4:16 PM, Francesc Alted <faltet@pytables.org> wrote:
Good! So, I should start by looking at how it achieves that then :)
Actually I did build a 4-way universal binary of HDF5. Long story short, because of some quirks of the build process for HDF5 it isn't possible to add some -arch flags to CFLAGS -- you actually have to build it separately for each architecture and then do something like
make install DESTDIR=path/for/arch
and then stitch them together manually with otool. I'm also not sure of the correct way, if there is one, to handle "hdf5.settings" -- it's unclear to me whether programs ever look at this at build time/ runtime, or if it's just there for the user's convenience.
Yes, I think "hdf5.settings" it is for convenience purposes only (a fast way to look for configuration of the compiler flags for the library). So you don't need to worry too much about this.
An installer for the binaries I made is at
Hey, that's great.
I can write up some instructions if that would be helpful.
Please do. I'm definitely interested! Thanks, -- Francesc Alted
David Warde-Farley wrote:
Actually I did build a 4-way universal binary of HDF5.
An installer for the binaries I made is at
http://www.cs.toronto.edu/~dwf/mirror/hdf5-1.8.4-quad.pkg
I can write up some instructions if that would be helpful.
Wonderful! Any chance you've tackled netcdf4 as well? Thanks, -Chris -- Christopher Barker, Ph.D. Oceanographer Emergency Response Division NOAA/NOS/OR&R (206) 526-6959 voice 7600 Sand Point Way NE (206) 526-6329 fax Seattle, WA 98115 (206) 526-6317 main reception Chris.Barker@noaa.gov
On Tue, Mar 16, 2010 at 09:49:12AM -0700, Christopher Barker wrote:
David Warde-Farley wrote:
Actually I did build a 4-way universal binary of HDF5.
An installer for the binaries I made is at
http://www.cs.toronto.edu/~dwf/mirror/hdf5-1.8.4-quad.pkg
I can write up some instructions if that would be helpful.
Wonderful! Any chance you've tackled netcdf4 as well?
Sorry, I haven't, but it should be roughly the same procedure. I'll write up some instructions. One thing with HDF5 was that I believe I had to compile i386 and x86_64 binaries on an Intel Mac and ppc/ppc64 on a PowerPC Mac, there was some issue with cross-compiling that I never quite got sorted out. It may compile a binary which it then wants to actually *run* for another step of the build process. Rosetta should take care of this on Intel but the machine I was sitting in front of was a G5, and the Intel build on that machine failed miserably, so I logged in remotely to an Intel Mac. I will try compiling all 4 on an Intel machine and see if that works. David
On Tue, Mar 16, 2010 at 10:05 AM, David Warde-Farley <dwf@cs.toronto.edu> wrote:
Hey Francesc,
On 2010-03-15, at 4:16 PM, Francesc Alted <faltet@pytables.org> wrote:
Good! So, I should start by looking at how it achieves that then :)
Actually I did build a 4-way universal binary of HDF5. Long story short, because of some quirks of the build process for HDF5 it isn't possible to add some -arch flags to CFLAGS -- you actually have to build it separately for each architecture and then do something like
make install DESTDIR=path/for/arch
and then stitch them together manually with otool.
Actually, that's a much better way of doing things generally. I wish this was the standard way of building universal binaries, it is usually more robust than using multiple -arch arguments, especially when you export a public API with C headers. I use this technique myself for all my packages pure C libraries (audiolab and samplerate in particular), cheers, David
participants (4)
-
Christopher Barker -
David Cournapeau -
David Warde-Farley -
Francesc Alted