Remove unsafe bool casts
#111024
Remove unsafe bool casts
#111024
Conversation
dotnet-issue-labeler
bot
added
the
needs-area-label
dotnet-policy-service
bot
added
the
community-contribution
| @@ -1075,17 +1075,15 @@ public static explicit operator float(Half value) | |||
| // Extract exponent bits of value (BiasedExponent is not for here as it performs unnecessary shift) | |||
| uint offsetExponent = bitValueInProcess & HalfExponentMask; | |||
| // ~0u when value is subnormal, 0 otherwise | |||
| uint subnormalMask = (uint)-Unsafe.BitCast<bool, byte>(offsetExponent == 0u); | |||
| // ~0u when value is either Infinity or NaN, 0 otherwise | |||
There was a problem hiding this comment.
this comment is incorrect
|
@tannergooding Is this beneficial? 2 bytes higher in code size, but 0.75 lower PerfScore. ; Assembly listing for method System.Half:op_Explicit(float):System.Half (FullOpts)
+ xor ecx, ecx
+ mov edx, -1
vucomiss xmm0, xmm0
- setp cl
- movzx rcx, cl
- dec ecx
+ cmovnp ecx, edxRelated: #61761 (comment) |
| // TODO: Replace with log2ToPow10[BitOperations.Log2(value)] once https://github.com/dotnet/runtime/issues/79257 is fixed | ||
| uint index = Unsafe.Add(ref MemoryMarshal.GetReference(log2ToPow10), BitOperations.Log2(value)); | ||
| nint elementOffset = Unsafe.Add(ref MemoryMarshal.GetReference(log2ToPow10), BitOperations.Log2(value)); |
There was a problem hiding this comment.
why nint isn't a sign rather than zero extension. Also, this looks like a hack that JIT should do, is it really needed? I know we do it in other places, but ideally it should be handled by JIT (and we move into that direction)
There was a problem hiding this comment.
It's a zero-extension from byte. We could use nuint instead, with some extra casts, which in any case would have no impact on code generation.
There was a problem hiding this comment.
have you checked that it comes with codegen improvements? if not (or it's not clear whether it brings any benefits) let's revert it. We don't want to pollute our codebase with hacks which often becomes unnecessary eventually.
There was a problem hiding this comment.
Using early-extension does provide a code generation improvement, e.g.
; Assembly listing for method System.Int128:TryFormat(System.Span`1[ushort],byref,System.ReadOnlySpan`1[ushort],System.IFormatProvider):ubyte:this (FullOpts)
mov rdx, 0xD1FFAB1E ; static handle
movzx rdi, byte ptr [rsi+rdx]
- mov esi, edi
- mov rdx, 0xD1FFAB1E ; static handle
- cmp rax, qword ptr [rdx+8*rsi]
+ mov rsi, 0xD1FFAB1E ; static handle
+ cmp rax, qword ptr [rsi+8*rdi]But it is not dependant on using nint vs nuint.
These were the only conversions I found searching for:
|
| @@ -24,8 +24,9 @@ public static int CountDigits(ulong value) | |||
| ]; | |||
| Debug.Assert(log2ToPow10.Length == 64); | |||
|
|
|||
| // We use 'nint' because it gives us a free early zero-extension to 64 bits when running on a 64-bit platform. | |||
There was a problem hiding this comment.
| // We use 'nint' because it gives us a free early zero-extension to 64 bits when running on a 64-bit platform. | |
| // We use a native-sized integer | |
| // because it gives us a free early zero-extension to 64 bits when running on a 64-bit platform. |
Contributes to #94941.