Cuda safety fixes #198
Merged
Cuda safety fixes #198
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Many places (basically everywhere), we had the following idiom:
template<class T>
class Foo {
IMATH_HOSTDEVICE float bar(); // declaration
};
...
template<class T> float Foo<T>::bar() { return 0.0; } // implementation
But this is wrong! When actually compiled in Cuda mode (maybe depending
on the compiler?), you can get errors about how you can't overload
a `__host__ __device__` declaration with a `__host__`-only implementation.
Which kinda makes sense. You have to match the two. So I have a WHOLE LOT
of places where I had to add IMATH_HOSTDEVICE.
Also, in ImathMath.h, we used this idiom:
IMATH_HOSTDEVICE IMATH_DEPRECATED("reason") float foo() { ... }
Now, this seems to work with `cudacc`, but when you use
`clang++ --language=cuda` to compile Cuda to PTX (it can do that!), clang
really doesn't like it when the `__host__ __device__` comes before the
`[[ deprecated("msg") ]]`, it has an error about how the deprecated
attribute can't go there. So we have to transpose these so that the
IMATH_DEPRECATED is alwys the first thing in the declaration (which is
the way we almost always write it anyway).
Signed-off-by: Larry Gritz <lg@larrygritz.com>
|
This PR is against RB-3.1 because I noticed the problem when trying to use 3.1. When accepted, it should also be cherry-picked into master and RB-3.0. (2.x is unnecessary because it predates the HOSTDEVICE annotations.) I would very much appreciate a 3.1.3 release that incorporates these fixes (when convenient), since the Imath headers are quite broken under Cuda without it. |
meshula
approved these changes
Aug 13, 2021
|
Some search & replace aid, but mostly manual, and double checking that my OSL Cuda builds against it stopped spitting out error messages. I think I got all the right spots. |
cary-ilm
reviewed
Aug 17, 2021
cary-ilm
reviewed
Aug 17, 2021
Signed-off-by: Larry Gritz <lg@larrygritz.com>
|
Other places that appear to be missing IMATH_HOSTDEVICE: ImathShear.h line 681 (operator*) I grepped for "inline" and looked for missing IMATH_HOSTDEVICE. |
Signed-off-by: Larry Gritz <lg@larrygritz.com>
|
Thanks for the extra pair of eyes, Cary! |
lgritz
added a commit
to lgritz/Imath
that referenced
this pull request
Aug 25, 2021
* Cuda safety fixes
Many places (basically everywhere), we had the following idiom:
template<class T>
class Foo {
IMATH_HOSTDEVICE float bar(); // declaration
};
...
template<class T> float Foo<T>::bar() { return 0.0; } // implementation
But this is wrong! When actually compiled in Cuda mode (maybe depending
on the compiler?), you can get errors about how you can't overload
a `__host__ __device__` declaration with a `__host__`-only implementation.
Which kinda makes sense. You have to match the two. So I have a WHOLE LOT
of places where I had to add IMATH_HOSTDEVICE.
Also, in ImathMath.h, we used this idiom:
IMATH_HOSTDEVICE IMATH_DEPRECATED("reason") float foo() { ... }
Now, this seems to work with `cudacc`, but when you use
`clang++ --language=cuda` to compile Cuda to PTX (it can do that!), clang
really doesn't like it when the `__host__ __device__` comes before the
`[[ deprecated("msg") ]]`, it has an error about how the deprecated
attribute can't go there. So we have to transpose these so that the
IMATH_DEPRECATED is alwys the first thing in the declaration (which is
the way we almost always write it anyway).
Signed-off-by: Larry Gritz <lg@larrygritz.com>
* Add missing HOSTDEVICE
Signed-off-by: Larry Gritz <lg@larrygritz.com>
* Add missing HOSTDEVICE
Signed-off-by: Larry Gritz <lg@larrygritz.com>
Merged
lgritz
added a commit
that referenced
this pull request
Aug 26, 2021
* Cuda safety fixes from #198 Many places (basically everywhere), we had the following idiom: template<class T> class Foo { IMATH_HOSTDEVICE float bar(); // declaration }; ... template<class T> float Foo<T>::bar() { return 0.0; } // implementation But this is wrong! When actually compiled in Cuda mode (maybe depending on the compiler?), you can get errors about how you can't overload a `__host__ __device__` declaration with a `__host__`-only implementation. Which kinda makes sense. You have to match the two. So I have a WHOLE LOT of places where I had to add IMATH_HOSTDEVICE. Also, in ImathMath.h, we used this idiom: IMATH_HOSTDEVICE IMATH_DEPRECATED("reason") float foo() { ... } Now, this seems to work with `cudacc`, but when you use `clang++ --language=cuda` to compile Cuda to PTX (it can do that!), clang really doesn't like it when the `__host__ __device__` comes before the `[[ deprecated("msg") ]]`, it has an error about how the deprecated attribute can't go there. So we have to transpose these so that the IMATH_DEPRECATED is alwys the first thing in the declaration (which is the way we almost always write it anyway). * Some minor cuda corrections to address warnings Signed-off-by: Larry Gritz <lg@larrygritz.com>
cary-ilm
pushed a commit
to cary-ilm/Imath
that referenced
this pull request
Jan 17, 2022
* Cuda safety fixes from AcademySoftwareFoundation#198 Many places (basically everywhere), we had the following idiom: template<class T> class Foo { IMATH_HOSTDEVICE float bar(); // declaration }; ... template<class T> float Foo<T>::bar() { return 0.0; } // implementation But this is wrong! When actually compiled in Cuda mode (maybe depending on the compiler?), you can get errors about how you can't overload a `__host__ __device__` declaration with a `__host__`-only implementation. Which kinda makes sense. You have to match the two. So I have a WHOLE LOT of places where I had to add IMATH_HOSTDEVICE. Also, in ImathMath.h, we used this idiom: IMATH_HOSTDEVICE IMATH_DEPRECATED("reason") float foo() { ... } Now, this seems to work with `cudacc`, but when you use `clang++ --language=cuda` to compile Cuda to PTX (it can do that!), clang really doesn't like it when the `__host__ __device__` comes before the `[[ deprecated("msg") ]]`, it has an error about how the deprecated attribute can't go there. So we have to transpose these so that the IMATH_DEPRECATED is alwys the first thing in the declaration (which is the way we almost always write it anyway). * Some minor cuda corrections to address warnings Signed-off-by: Larry Gritz <lg@larrygritz.com> Signed-off-by: Cary Phillips <cary@ilm.com>
cary-ilm
pushed a commit
that referenced
this pull request
Jan 20, 2022
* Cuda safety fixes from #198 Many places (basically everywhere), we had the following idiom: template<class T> class Foo { IMATH_HOSTDEVICE float bar(); // declaration }; ... template<class T> float Foo<T>::bar() { return 0.0; } // implementation But this is wrong! When actually compiled in Cuda mode (maybe depending on the compiler?), you can get errors about how you can't overload a `__host__ __device__` declaration with a `__host__`-only implementation. Which kinda makes sense. You have to match the two. So I have a WHOLE LOT of places where I had to add IMATH_HOSTDEVICE. Also, in ImathMath.h, we used this idiom: IMATH_HOSTDEVICE IMATH_DEPRECATED("reason") float foo() { ... } Now, this seems to work with `cudacc`, but when you use `clang++ --language=cuda` to compile Cuda to PTX (it can do that!), clang really doesn't like it when the `__host__ __device__` comes before the `[[ deprecated("msg") ]]`, it has an error about how the deprecated attribute can't go there. So we have to transpose these so that the IMATH_DEPRECATED is alwys the first thing in the declaration (which is the way we almost always write it anyway). * Some minor cuda corrections to address warnings Signed-off-by: Larry Gritz <lg@larrygritz.com> Signed-off-by: Cary Phillips <cary@ilm.com>
cary-ilm
added a commit
to cary-ilm/Imath
that referenced
this pull request
May 19, 2023
PR AcademySoftwareFoundation#198 added IMATH_HOSTDEVICE to the RB-3.1 branch, and PR AcademySoftwareFoundation#202 cherry-picked the change into main, but that cherry-pick somehow lost the IMATH_HOSTDEVICE on Matrix33<T>::invert(bool). Signed-off-by: Cary Phillips <cary@ilm.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Many places (basically everywhere), we had the following idiom:
But this is wrong! When actually compiled in Cuda mode (maybe depending
on the compiler?), you can get errors about how you can't overload
a
__host__ __device__declaration with a__host__-only implementation.Which kinda makes sense. You have to match the two. So I have a WHOLE LOT
of places where I had to add IMATH_HOSTDEVICE.
Also, in ImathMath.h, we used this idiom:
Now, this seems to work with
cudacc, but when you useclang++ --language=cudato compile Cuda to PTX (it can do that!), clangreally doesn't like it when the
__host__ __device__comes before the[[ deprecated("msg") ]], it has an error about how the deprecatedattribute can't go there. So we have to transpose these so that the
IMATH_DEPRECATED is alwys the first thing in the declaration (which is
the way we almost always write it anyway).
Signed-off-by: Larry Gritz lg@larrygritz.com