Initial support for HalfGrid and ComputeType #1787
Conversation
Initial support for HalfGrid and ComputeType, where the latter enables float computations for HalfGrid which increases accuracy and speed. Signed-off-by: ghurstunither <[email protected]>
Remove HalfGrid from RealGridTypes until a consensus is reached. Signed-off-by: ghurstunither <[email protected]>
Fixes trailingspaces errors Signed-off-by: ghurstunither <[email protected]>
| @@ -1746,10 +1747,11 @@ template<typename GridType> | |||
| typename GridType::Ptr | |||
| createLevelSet(Real voxelSize, Real halfWidth) | |||
| { | |||
| using ValueType = typename GridType::ValueType; | |||
| using ValueType = typename GridType::ValueType; | |||
| using ComputeType = typename GridType::ComputeType; | |||
|
|
|||
| // GridType::ValueType is required to be a floating-point scalar. | |||
There was a problem hiding this comment.
The comment should match what the code does. But I think in this case, the comment is right. ValueType reflects the payload stored in the leaf nodes and in the background value. The requirement for a level-set is for these two things to be floating point.
| template<> struct Tolerance<float> { static float value() { return 1e-8f; } }; | ||
| template<> struct Tolerance<double> { static double value() { return 1e-15; } }; | ||
| template<typename T> struct Tolerance { static T value() { return zeroVal<T>(); } }; | ||
| template<> struct Tolerance<math::half> { static math::half value() { return 0.00097656; } }; |
There was a problem hiding this comment.
Please add a comment on the reference of where this value comes from.
|
|
||
| // isNegative and operator-() are bitwise ops | ||
| // all half types in HalfDecl.h _could_ define abs() as half(FromBits, bits() & 0x7fff) | ||
| inline math::half Abs(math::half x) { return x.isNegative() ? -x : x; } |
There was a problem hiding this comment.
Happy to see the comment :)
| @@ -362,6 +370,10 @@ isApproxZero(const Type& x, const Type& tolerance) | |||
| } | |||
|
|
|||
|
|
|||
| /// Return @c true if @a x is less than zero. | |||
| inline bool | |||
| isNegative(math::half& x) { return x.isNegative(); } | |||
There was a problem hiding this comment.
We probably want to add a const keyword in the argument.
|
@ghurstunither. I'm wondering if you have ever compiled this PR with |
|
|
Should we close this PR and re-open using the feature branch as the source? |
Yes, I’ve gone ahead and closed this one. |
Initial support for
HalfGridandComputeTypein OpenVDB, where the latter enables intermediatefloatcomputations forHalfGridto increase accuracy and speed.ValueToComputeMapinTypes.hto setComputeTypeat the leaf level and then to propagate all the way up to the grid level.math::half,ComputeTypeisfloatand for any other typeT,ComputeTypeisT.HalfGrid(half vectors not in this PR).math::halffromTypes.hto its ownHalfDecl.hmath::halfinMath.h(to avoid casting in some places) so there were now circular dependancies.ComputeTypeinstead ofValueType.Note, I decided not to define
openvdb::is_floating_pointand friends. Instead I opted to usestd::is_floating_point<ComputeType>()::valueinstead.Still to come in this PR
*Uunit tests for half and to verify no unit tests have broken.