even though ASTC could encode other values of A, still clamp as if it was BC6H for consistency
now all compressed HDR formats should clamp the same way that Oodle Texture currently does
#preflight 6296556c452ffe576a816993
#rb fabian.giesen
[CL 20440882 by charles bloom in ue5-main branch]
Moved Math forward declarations from CoreFwd.h to Math/MathFwd.h
#rb steve.robb
#preflight 62837df43c1cdc59e2b07dd5
[CL 20240602 by Andrew Davidson in ue5-main branch]
* Brushes or rotated volumes were including more actors than they should have
#rb jeanfrancois.dube
#jira none
#preflight 62824c41734d06577025f099
[CL 20222727 by Sebastien Lussier in ue5-main branch]
AIGraph::UpdateClassData: Fix incorrect recursive child call: call needs to be made on the child not on the current node.
From PR #8466 submitted by LindyHopperGT.
#rb Josh.Adams
#rnx
#preflight 627bb7842d678960301025a0
[CL 20150316 by Matt Peters in ue5-main branch]
#jira UE-151249
#rb Zak.Middleton
#ushell-cherrypick of 20130528 by dmytro.vovk
#preflight 627ae02e68422389764e57c5
[CL 20131539 by dmytro vovk in ue5-main branch]
In the Shirring Test (in development editor), the Bending Constraints change drops the single iteration BendingApply from 2.1ms --> 350us. For reference, Bending Springs (with ISPC) are about 150us.
For XPBDSpringConstraints, there is no way to enable XPBD from the UI, so this does not affect existing behavior, but I use XPBD for testing high res cloth, so the ISPC XPBD speedups are important there.
This also includes code for using PhysicsParallelFor (following other parallel constraint examples). This is significantly slower than ISPC on PC (similar to other cloth constraints). I don't know if this would be better on other platforms, but the default is to use ISPC (following same convention as other cloth constraints).
#preflight 6271affa2f4fb6234ec50446
#rb kriss.gossart
[CL 20034474 by Alex McAdams in ue5-main branch]
Given that the second argument is always 360, we can remove some tests for bad ranges of input and use simple truncation/multiplication to compute the result, while retaining good accuracy within reasonable FRotator input value ranges.
#jira UE-129730
#preflight 62699851f97c319beba69337
#rb Dmytro.Vovk
[CL 19947858 by Zak Middleton in ue5-main branch]
- FMath::Fmod() changed to use standard fmod() function, because old version could go out of range for combinations of very large and very small values. VectorMod() changed where possible to do the same.
- Fixed ambiguous naming conventions for many functions (inconsistency between InvSqrt, Reciprocal, etc and their estimate functions). Now the default for such functions is to be accurate, and the estimate functions are explicitly named with "Estimate".
- VectorReciprocalSqrt() implementation no longer refines an estimate, as this can diverge between Intel and AMD; this was an issue addressed earlier in UnrealPlatformMathSSE.h (CL 16466371) and we now match that implementation. VectorReciprocal() also no longer refines an estimate for the same reason. Attempted to use SVML InvSqrt however it suffered from measurable variations on AMD/Intel platforms.
- Added missing vectorized implementation for VectorPow() for doubles in SSE.
- Added missing VectorSqrt() implementations for doubles..
- Did a small pass at fixing up some missing functions in UnrealMathFPU so it can pass engine startup tests and HeadlessChaos tests when enabled, though there are a few more that need implementation updates for Niagara (to come soon).
#jira UE-129730
#rb Andrew.Davidson, Dmytro.Vovk
#preflight 62632206e70367fd1e0d29db
[CL 19906229 by Zak Middleton in ue5-main branch]
#fyi Zak.Middleton
Original CL Desc
-----------------------------------------------------------------
#ue5 - LWC: Improve precision of some math functions (FMod / VectorMod) on SSE by using SVML where possible. FMath::Fmod() changed to use standard fmod() function. Fixed ambiguous naming conventions for many functions (inconsistency between InvSqrt, Reciprocal, etc and their estimate functions). Now the default for such functions is to be accurate, and the estimate functions are explicitly named with "Estimate".
Added missing vectorized implementation for VectorPow() for doubles. Fixed some missing VectorSqrt() overrides.
Did a small pass at fixing up some missing functions in UnrealMathFPU so it can pass start tests and HeadlessChaos tests when enabled, though there are a few more that need updates for Niagara (to come soon).
#jira UE-129730
#rb Andrew.Davidson, Dmytro.Vovk
#preflight 6260739e080c6600634ca48b, 625f0f5d5f498a37c9263f29
[CL 19870070 by Zak Middleton in ue5-main branch]
Added missing vectorized implementation for VectorPow() for doubles. Fixed some missing VectorSqrt() overrides.
Did a small pass at fixing up some missing functions in UnrealMathFPU so it can pass start tests and HeadlessChaos tests when enabled, though there are a few more that need updates for Niagara (to come soon).
#jira UE-129730
#rb Andrew.Davidson, Dmytro.Vovk
#preflight 6260739e080c6600634ca48b, 625f0f5d5f498a37c9263f29
[CL 19868793 by Zak Middleton in ue5-main branch]
