fix(simd.h): Make all-architecture matrix44::inverse() #4076
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
We had an awkward situation of the simd matrix44::inverse() only being defined for Intel SSE, and in all other cases falling back on Imath::M44f::inverse. The problem with that is that it made simd.h depend either directly on ImathMatrix.h, or assume that it would be included prior to inclusion of simd.h. This was not good. This patch rewrites inverse() to eliminate all of the direct SSE intrinsics in terms of other simd wrapper functions, which means that they will be defined one way or another on all architectures. Along the way I added a new vfloat4 shuffle that combines two vectors (corresponding to _mm_shuffle_ps in SSE land), and a vfloat4 constructor from 2 float*', where the first two elements come from the first ptr and the second two elements come from the second ptr. Ultimately, the final implementation I settled on for the new inverse didn't use these (an earlier version did), but I want to keep them around anyway in case they are useful in the future. I did add a test for them. Signed-off-by: Larry Gritz <[email protected]>
|
Does anybody think this is a bad idea? |
ThiagoIze
approved these changes
Jan 1, 2024
|
simd_test benchmarks these, so I know there is no slowdown (though I haven't tested on ARM). It's not especially as much faster than the Imath one as I would have hoped, so to the extent that it's ever performance critical, I'm sure it could use some improvement. But for now, I just wanted to get some implementation so we could remove the Imath header inclusion from simd.h. |
ThiagoIze
approved these changes
Jan 1, 2024
lgritz
added a commit
to lgritz/OpenImageIO
that referenced
this pull request
Jan 2, 2024
…reFoundation#4076) We had an awkward situation of the simd matrix44::inverse() only being defined for Intel SSE, and in all other cases falling back on Imath::M44f::inverse. The problem with that is that it made simd.h depend either directly on ImathMatrix.h, or assume that it would be included prior to inclusion of simd.h. This was not good. This patch rewrites inverse() to eliminate all of the direct SSE intrinsics in terms of other simd wrapper functions, which means that they will be defined one way or another on all architectures. Along the way I added a new vfloat4 shuffle that combines two vectors (corresponding to `_mm_shuffle_ps` in SSE land), and a vfloat4 constructor from 2 float*', where the first two elements come from the first ptr and the second two elements come from the second ptr. Ultimately, the final implementation I settled on for the new inverse didn't use these (an earlier version did), but I want to keep them around anyway in case they are useful in the future. I did add a test for them. Signed-off-by: Larry Gritz <[email protected]>
1div0
pushed a commit
to 1div0/OpenImageIO
that referenced
this pull request
Feb 24, 2024
…reFoundation#4076) We had an awkward situation of the simd matrix44::inverse() only being defined for Intel SSE, and in all other cases falling back on Imath::M44f::inverse. The problem with that is that it made simd.h depend either directly on ImathMatrix.h, or assume that it would be included prior to inclusion of simd.h. This was not good. This patch rewrites inverse() to eliminate all of the direct SSE intrinsics in terms of other simd wrapper functions, which means that they will be defined one way or another on all architectures. Along the way I added a new vfloat4 shuffle that combines two vectors (corresponding to `_mm_shuffle_ps` in SSE land), and a vfloat4 constructor from 2 float*', where the first two elements come from the first ptr and the second two elements come from the second ptr. Ultimately, the final implementation I settled on for the new inverse didn't use these (an earlier version did), but I want to keep them around anyway in case they are useful in the future. I did add a test for them. Signed-off-by: Larry Gritz <[email protected]> Signed-off-by: Peter Kovář <[email protected]>
|
@lgritz do you know why lgritz@4cb956b did not remove this: // Without SSE, we need to fall back on Imath for matrix44 invert
#if !OIIO_SIMD_SSE
# include <OpenImageIO/Imath.h>
#endifwhile the 2.6 branch does have it removed? It's a shame 2.5 still depends on Imath.h because of that, especially since the comment says this include is for matrix44 invert which is what this ticket was for. |
|
I don't remember now. We could try removing it to see if anything breaks? |
5 tasks
lgritz
pushed a commit
that referenced
this pull request
Aug 11, 2024
lgritz@4cb956b lost this removal when resolving a merge conflict. This properly removes it. #4076 implemented matrix44::inverse() so that we could stop including Imath.h. It removed the Imath.h include in the master/dev-2.6 branch but the dev-2.5 branch kept the include, likely because of an improperly resolved merge conflict. This PR removes the include like in master. No tests are needed. As long as it compiles, this should be good. I verified this builds on arm64. Signed-off-by: Thiago Ize <[email protected]> Co-authored-by: Thiago Ize <[email protected]>
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.
We had an awkward situation of the simd matrix44::inverse() only being defined for Intel SSE, and in all other cases falling back on Imath::M44f::inverse. The problem with that is that it made simd.h depend either directly on ImathMatrix.h, or assume that it would be included prior to inclusion of simd.h. This was not good.
This patch rewrites inverse() to eliminate all of the direct SSE intrinsics in terms of other simd wrapper functions, which means that they will be defined one way or another on all architectures.
Along the way I added a new vfloat4 shuffle that combines two vectors (corresponding to
_mm_shuffle_psin SSE land), and a vfloat4 constructor from 2 float*', where the first two elements come from the first ptr and the second two elements come from the second ptr. Ultimately, the final implementation I settled on for the new inverse didn't use these (an earlier version did), but I want to keep them around anyway in case they are useful in the future. I did add a test for them.