Hardware accelerated implementation of Quaternion multiply #96624
Conversation
This file contains hidden or 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
I saw on the Discord that Quaternion multiply wasnt yet vectorised, this PR adds that. The non accelerated path is the exact same as before, and the accelerated path appears to return the same results.
ghost
added
community-contribution
|
Tagging subscribers to this area: @dotnet/area-system-numerics Issue DetailsI saw on the Discord that Quaternion multiply wasnt yet vectorised, this PR adds that. The non accelerated path is the exact same as before, and the accelerated path returns the same results.
|
AustinWise
reviewed
Jan 8, 2024
src/libraries/System.Private.CoreLib/src/System/Numerics/Quaternion.cs
Outdated
Show resolved
Hide resolved
…rnion.cs Copy paste error! Co-authored-by: Austin Wise <AustinWise@gmail.com>
This was referenced Jan 9, 2024
Comment on lines
191
to
205
| if (Vector.IsHardwareAccelerated) | ||
| { | ||
| // Concatenate rotation is actually q2 * q1 instead of q1 * q2. | ||
| // So that's why value2 goes q1 and value1 goes q2. | ||
| float q1w = value2.W; | ||
| float q2w = value1.W; | ||
|
|
||
| Vector3 q1V = new Vector3(value2.X, value2.Y, value2.Z); | ||
| Vector3 q2V = new Vector3(value1.X, value1.Y, value1.Z); | ||
|
|
||
| return new Quaternion( | ||
| q1w * q2V + q2w * q1V + Vector3.Cross(q1V, q2V), | ||
| q1w * q2w - Vector3.Dot(q1V, q2V) | ||
| ); | ||
| } |
There was a problem hiding this comment.
I would expect something like the following to perform better:
if (Vector128.IsHardwareAccelerated)
{
var left = value1.AsVector128();
var right = value2.AsVector128();
var result = right * left.GetElement(3);
result += (Vector128.Shuffle(right, Vector128.Create(3, 2, 1, 0)) * left.GetElement(0)) * Vector128.Create(+1.0f, -1.0f, +1.0f, -1.0f);
result += (Vector128.Shuffle(right, Vector128.Create(2, 3, 0, 1)) * left.GetElement(1)) * Vector128.Create(+1.0f, +1.0f, -1.0f, -1.0f);
result += (Vector128.Shuffle(right, Vector128.Create(1, 0, 3, 2)) * left.GetElement(2)) * Vector128.Create(-1.0f, +1.0f, +1.0f, -1.0f);
return result.AsQuaternion();
}There was a problem hiding this comment.
Thanks, that is indeed quite a lot faster, and the results still looked correct.
| Method | Mean | Error | StdDev |
|----------------------- |----------:|----------:|----------:|
| MulOperator | 4.5998 ns | 0.0843 ns | 0.0788 ns |
| VectorMul | 2.9167 ns | 0.0145 ns | 0.0128 ns |
| IntrinsicMulGetElement | 0.8482 ns | 0.0113 ns | 0.0106 ns |
I've modified it to use the GetElementUnsafe method, i saw other Numerics types use this too.
Tanner provided an even quicker version. I've changed it to use GetElementUnsafe as i saw other API's doing the same, and we can reasonably argue these fields to exist.
src/libraries/System.Private.CoreLib/src/System/Numerics/Quaternion.cs
Outdated
Show resolved
Hide resolved
…rnion.cs Co-authored-by: Tanner Gooding <tagoo@outlook.com>
src/libraries/System.Private.CoreLib/src/System/Numerics/Quaternion.cs
Outdated
Show resolved
Hide resolved
tannergooding
approved these changes
Jan 17, 2024
This was referenced Jan 17, 2024
|
Failures are unrelated issues that are already tracked. |
tmds
pushed a commit
to tmds/runtime
that referenced
this pull request
Jan 23, 2024
) * Vector based implementation of Quaternion multiply I saw on the Discord that Quaternion multiply wasnt yet vectorised, this PR adds that. The non accelerated path is the exact same as before, and the accelerated path appears to return the same results. * Convert tabs to spaces * Update src/libraries/System.Private.CoreLib/src/System/Numerics/Quaternion.cs Copy paste error! Co-authored-by: Austin Wise <AustinWise@gmail.com> * Even faster version Tanner provided an even quicker version. I've changed it to use GetElementUnsafe as i saw other API's doing the same, and we can reasonably argue these fields to exist. * Update src/libraries/System.Private.CoreLib/src/System/Numerics/Quaternion.cs Co-authored-by: Tanner Gooding <tagoo@outlook.com> * Fix incorrect non-hardware accelerated version --------- Co-authored-by: Austin Wise <AustinWise@gmail.com> Co-authored-by: Tanner Gooding <tagoo@outlook.com>
|
Looks like we have some small number of regressions for this (rest is not interesting): dotnet/perf-autofiling-issues#27922 |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
5 participants
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.
I saw on the Discord that Quaternion multiply wasnt yet vectorised, this PR adds that. The non accelerated path is the exact same as before, and the accelerated path returns the same results.