Re: [racket-dev] Release for v5.2 has begun
I downloaded the latest pre-release version (5.1.900.1) to update the science collection to use the new plot collection. The good news is that that went very smoothly. The bad news is that some of my FFT routines seem to be getting incorrect numeric results - at least the plots are very bad. But, since all of the other plots seem fine, I don't see why these would be any different. So, I suspect that something has changed that affects the numeric calculations. It is just the radix-2 FFTs that are having the problem. They do some low-level bit fiddling to do the in-place butterfly addressing for the FFTs - using things like unsafe-fxlshift (which just looks unfriendly) - so, I suspect something there. Are the any recent changes that would affect these kinds of fixed-point operations? I've run the code under a previous version of Racket on a 64-bit Linux (Scientific Linux 6.0) computer and a 32-bit Windows XP computer and get correct results. So, I don't think it is simply a 32/64 bit problem - unless it is limited to 64-bit Windows. Sorry that was kind of rambling and non-specific, but I was wondering where to start looking at the problem or what would help someone else look into it. Doug On Fri, Oct 7, 2011 at 10:19 PM, Ryan Culpepper r...@cs.utah.edu wrote: The release process for v5.2 has begun: the `release' branch was created for any work that is left and is now bumped to v5.1.90. You can go on using the `master' branch as usual, it is now bumped to v5.2.0.1 (to avoid having two different trees with the same version). If you have any bug-fixes and changes that need to go in the release then make sure to specify that in the commit message or mail me the commit SHA1s. You can `git checkout release' to try it out directly if needed -- but do not try to push commits on it (the server will forbid it). Please make sure that code that you're responsible for is as stable as possible, and let me know if there is any new work that should not be included in this release. NOW IS THE TIME TO FIX BUGS THAT YOU KNOW ABOUT The time between the `release' branch creation and the actual release is for fixing new errors that prevent proper functioning of major components and that show up during the preparation for a release. You can also finalize piece of work that is not yet complete, but please avoid merging new features. Note that nightly builds will go on as usual (as v5.2.0.1), and pre-release builds will be available shortly at http://pre.racket-lang.org/release/ Please tell me if you think that this release is significant enough that it should be announced on the users list for wider testing. -- Ryan Culpepper _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev
Re: [racket-dev] Release for v5.2 has begun
I reloaded the older version of Racket on my Windows 7 computer and the radix-2 FFTs run fine there. I also tested both versions on my Macbook Pro (32 bit) and got the same behavior - correct results (plots) on 5.1.2 and bad radix-2 results (plots) on 5.1.900.1. code for the mixed-radix and radix-2. So, it isn't limited to Windows 7 or 64-bit. Doug On Sun, Oct 9, 2011 at 1:44 PM, Doug Williams m.douglas.willi...@gmail.comwrote: I downloaded the latest pre-release version (5.1.900.1) to update the science collection to use the new plot collection. The good news is that that went very smoothly. The bad news is that some of my FFT routines seem to be getting incorrect numeric results - at least the plots are very bad. But, since all of the other plots seem fine, I don't see why these would be any different. So, I suspect that something has changed that affects the numeric calculations. It is just the radix-2 FFTs that are having the problem. They do some low-level bit fiddling to do the in-place butterfly addressing for the FFTs - using things like unsafe-fxlshift (which just looks unfriendly) - so, I suspect something there. Are the any recent changes that would affect these kinds of fixed-point operations? I've run the code under a previous version of Racket on a 64-bit Linux (Scientific Linux 6.0) computer and a 32-bit Windows XP computer and get correct results. So, I don't think it is simply a 32/64 bit problem - unless it is limited to 64-bit Windows. Sorry that was kind of rambling and non-specific, but I was wondering where to start looking at the problem or what would help someone else look into it. Doug On Fri, Oct 7, 2011 at 10:19 PM, Ryan Culpepper r...@cs.utah.edu wrote: The release process for v5.2 has begun: the `release' branch was created for any work that is left and is now bumped to v5.1.90. You can go on using the `master' branch as usual, it is now bumped to v5.2.0.1 (to avoid having two different trees with the same version). If you have any bug-fixes and changes that need to go in the release then make sure to specify that in the commit message or mail me the commit SHA1s. You can `git checkout release' to try it out directly if needed -- but do not try to push commits on it (the server will forbid it). Please make sure that code that you're responsible for is as stable as possible, and let me know if there is any new work that should not be included in this release. NOW IS THE TIME TO FIX BUGS THAT YOU KNOW ABOUT The time between the `release' branch creation and the actual release is for fixing new errors that prevent proper functioning of major components and that show up during the preparation for a release. You can also finalize piece of work that is not yet complete, but please avoid merging new features. Note that nightly builds will go on as usual (as v5.2.0.1), and pre-release builds will be available shortly at http://pre.racket-lang.org/release/ Please tell me if you think that this release is significant enough that it should be announced on the users list for wider testing. -- Ryan Culpepper _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev
Re: [racket-dev] Release for v5.2 has begun
Fixed. Some unsafe primitives were marked internally as functional, but this annotation was used with two different interpretations: sometimes as non-mutating, sometimes as always produces the same result. The `unsafe-vector-ref' primitive was marked as functional with the former intent, but the compiler could use the latter interpretation to move it past an `unsafe-vector-set!'. The solution, of course, is to have two distinct annotations. At Sun, 9 Oct 2011 15:02:13 -0600, Doug Williams wrote: I reloaded the older version of Racket on my Windows 7 computer and the radix-2 FFTs run fine there. I also tested both versions on my Macbook Pro (32 bit) and got the same behavior - correct results (plots) on 5.1.2 and bad radix-2 results (plots) on 5.1.900.1. code for the mixed-radix and radix-2. So, it isn't limited to Windows 7 or 64-bit. Doug On Sun, Oct 9, 2011 at 1:44 PM, Doug Williams m.douglas.willi...@gmail.comwrote: I downloaded the latest pre-release version (5.1.900.1) to update the science collection to use the new plot collection. The good news is that that went very smoothly. The bad news is that some of my FFT routines seem to be getting incorrect numeric results - at least the plots are very bad. But, since all of the other plots seem fine, I don't see why these would be any different. So, I suspect that something has changed that affects the numeric calculations. It is just the radix-2 FFTs that are having the problem. They do some low-level bit fiddling to do the in-place butterfly addressing for the FFTs - using things like unsafe-fxlshift (which just looks unfriendly) - so, I suspect something there. Are the any recent changes that would affect these kinds of fixed-point operations? I've run the code under a previous version of Racket on a 64-bit Linux (Scientific Linux 6.0) computer and a 32-bit Windows XP computer and get correct results. So, I don't think it is simply a 32/64 bit problem - unless it is limited to 64-bit Windows. Sorry that was kind of rambling and non-specific, but I was wondering where to start looking at the problem or what would help someone else look into it. Doug On Fri, Oct 7, 2011 at 10:19 PM, Ryan Culpepper r...@cs.utah.edu wrote: The release process for v5.2 has begun: the `release' branch was created for any work that is left and is now bumped to v5.1.90. You can go on using the `master' branch as usual, it is now bumped to v5.2.0.1 (to avoid having two different trees with the same version). If you have any bug-fixes and changes that need to go in the release then make sure to specify that in the commit message or mail me the commit SHA1s. You can `git checkout release' to try it out directly if needed -- but do not try to push commits on it (the server will forbid it). Please make sure that code that you're responsible for is as stable as possible, and let me know if there is any new work that should not be included in this release. NOW IS THE TIME TO FIX BUGS THAT YOU KNOW ABOUT The time between the `release' branch creation and the actual release is for fixing new errors that prevent proper functioning of major components and that show up during the preparation for a release. You can also finalize piece of work that is not yet complete, but please avoid merging new features. Note that nightly builds will go on as usual (as v5.2.0.1), and pre-release builds will be available shortly at http://pre.racket-lang.org/release/ Please tell me if you think that this release is significant enough that it should be announced on the users list for wider testing. -- Ryan Culpepper _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev
Re: [racket-dev] Release for v5.2 has begun
Let me know if you want me to break anything else. My mom always hated it when I was a kid, but it seems to be useful on pre-releases. On Sun, Oct 9, 2011 at 7:46 PM, Matthew Flatt mfl...@cs.utah.edu wrote: Fixed. Some unsafe primitives were marked internally as functional, but this annotation was used with two different interpretations: sometimes as non-mutating, sometimes as always produces the same result. The `unsafe-vector-ref' primitive was marked as functional with the former intent, but the compiler could use the latter interpretation to move it past an `unsafe-vector-set!'. The solution, of course, is to have two distinct annotations. At Sun, 9 Oct 2011 15:02:13 -0600, Doug Williams wrote: I reloaded the older version of Racket on my Windows 7 computer and the radix-2 FFTs run fine there. I also tested both versions on my Macbook Pro (32 bit) and got the same behavior - correct results (plots) on 5.1.2 and bad radix-2 results (plots) on 5.1.900.1. code for the mixed-radix and radix-2. So, it isn't limited to Windows 7 or 64-bit. Doug On Sun, Oct 9, 2011 at 1:44 PM, Doug Williams m.douglas.willi...@gmail.comwrote: I downloaded the latest pre-release version (5.1.900.1) to update the science collection to use the new plot collection. The good news is that that went very smoothly. The bad news is that some of my FFT routines seem to be getting incorrect numeric results - at least the plots are very bad. But, since all of the other plots seem fine, I don't see why these would be any different. So, I suspect that something has changed that affects the numeric calculations. It is just the radix-2 FFTs that are having the problem. They do some low-level bit fiddling to do the in-place butterfly addressing for the FFTs - using things like unsafe-fxlshift (which just looks unfriendly) - so, I suspect something there. Are the any recent changes that would affect these kinds of fixed-point operations? I've run the code under a previous version of Racket on a 64-bit Linux (Scientific Linux 6.0) computer and a 32-bit Windows XP computer and get correct results. So, I don't think it is simply a 32/64 bit problem - unless it is limited to 64-bit Windows. Sorry that was kind of rambling and non-specific, but I was wondering where to start looking at the problem or what would help someone else look into it. Doug On Fri, Oct 7, 2011 at 10:19 PM, Ryan Culpepper r...@cs.utah.edu wrote: The release process for v5.2 has begun: the `release' branch was created for any work that is left and is now bumped to v5.1.90. You can go on using the `master' branch as usual, it is now bumped to v5.2.0.1 (to avoid having two different trees with the same version). If you have any bug-fixes and changes that need to go in the release then make sure to specify that in the commit message or mail me the commit SHA1s. You can `git checkout release' to try it out directly if needed -- but do not try to push commits on it (the server will forbid it). Please make sure that code that you're responsible for is as stable as possible, and let me know if there is any new work that should not be included in this release. NOW IS THE TIME TO FIX BUGS THAT YOU KNOW ABOUT The time between the `release' branch creation and the actual release is for fixing new errors that prevent proper functioning of major components and that show up during the preparation for a release. You can also finalize piece of work that is not yet complete, but please avoid merging new features. Note that nightly builds will go on as usual (as v5.2.0.1), and pre-release builds will be available shortly at http://pre.racket-lang.org/release/ Please tell me if you think that this release is significant enough that it should be announced on the users list for wider testing. -- Ryan Culpepper _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev
