Add BFloat16 #98643
Add BFloat16 #98643
Conversation
|
Note regarding the This serves as a reminder for when your PR is modifying a ref *.cs file and adding/modifying public APIs, please make sure the API implementation in the src *.cs file is documented with triple slash comments, so the PR reviewers can sign off that change. |
|
Tagging subscribers to this area: @dotnet/area-system-numerics Issue DetailsCloses #96295. Delegating all functional members to float, since the upcast is trivial. The only logic is rounding from double.
|
huoyaoyuan
added
the
community-contribution
| result = (actualValue >= new BFloat16(0x7F00)) ? MaxValue : | ||
| (actualValue <= new BFloat16(0xFF00)) ? MinValue : (Int128)actualValue; |
There was a problem hiding this comment.
nit: for Half we're using == Half.PositiveInfinity and == Half.NegativeInfinity for clarity.
We should do the equivalent here for BFloat16, either using an existing named constant or documenting what the boundary values used are.
There was a problem hiding this comment.
The cases are actually different here. For Half the max possible value is 65504, which is in range of larger integer.
The value range of BFloat16 are about the same of float which is larger than the integers. The value is actually a manual truncate of (float)Int128.MaxValue.
There was a problem hiding this comment.
I meant conceptually the same. That is, we should use a well named member where it exists -or- document what value the bits represent.
0x7F00 says nothing to the typical reader. Adding a comment that explains it means (float)Int128.MaxValue and corresponds roughly to 340282366920938463463374607431768211456 (adjusting for precision and the actual number stored) then means a lot
| { | ||
| BFloat16 actualValue = (BFloat16)(object)value; | ||
| result = (actualValue >= new BFloat16(0x4700)) ? MaxValue : | ||
| (actualValue <= new BFloat16(0xB700)) ? MinValue : (short)actualValue; |
There was a problem hiding this comment.
We should have and use an internal UInt16BitsToBFloat16 for parity with the other floating-point support.
| @@ -1015,7 +1015,7 @@ public static explicit operator double(Half value) | |||
| exp -= 1; | |||
| } | |||
|
|
|||
| return CreateDouble(sign, (ushort)(exp + 0x3F0), (ulong)sig << 42); | |||
| return Math.CreateDouble(sign, (ushort)(exp + 0x3F0), (ulong)sig << 42); | |||
There was a problem hiding this comment.
nit: Math is basically "frozen" and not a place we want to find new methods, even internal ones
These would be "better" as internal double.Create and float.Create methods instead
Closes #96295.
Delegating all functional members to float, since the upcast is trivial. The only logic is rounding from double.
Tests are borrowed from Half.