On Monday, 8 May 2023 at 08:05:13 UTC, Richard (Rikki) Andrew
Cattermole wrote:
Yes, there is a pragma msg bug, but there is also a functionality
'bug'.
I collected some more info:
```
import std, core.simd, ldc.llvmasm;
T pshufb(T, U)(T a, in U b) { return __asm!ubyte16("pshufb $2,
$1", "=x,0,x", a, b); }
void main()
{
enum ubyte16
input = mixin(iota(100, 116).array),
good = mixin([1, 2, 3, 4, 5, 6, 7]),
bad1 = [1, 2, 3, 4, 5, 6, 7];
static immutable
bad2 = [1, 2, 3, 4, 5, 6, 7];
//pragma(msg, somewhat_good); <-crash
writeln(good);
writeln(pshufb(input, good));
writeln;
writeln(bad1);
writeln(pshufb(input, bad1));
writeln;
writeln(bad2);
writeln(pshufb(input, bad2));
}
[1, 2, 3, 4, 5, 6, 7, 0, 0, 0, 0, 0, 0, 0, 0, 0]
[101, 102, 103, 104, 105, 106, 107, 100, 100, 100, 100, 100, 100,
100, 100, 100]
[1, 2, 3, 4, 5, 6, 7, 0, 0, 0, 0, 0, 0, 0, 0, 0]
[100, 100, 100, 100, 100, 100, 100, 100, 100, 100, 100, 100, 100,
100, 100, 100]
[1, 2, 3, 4, 5, 6, 7]
[107, 100, 100, 100, 100, 100, 100, 100, 100, 100, 100, 100, 100,
100, 100, 100]
```
In the declaration of pshufb I let anything pass through to the
__asm statement.
Only the enum ubyte16 = mixin(...); variant calculates the
correct results, and luckily that can accept iota... calculated
inputs as well.
But at this point I have to memorize this and be aware when I'm
having weird results.
After all, this is the best SSE assembler I ever used so far *big
thumbs up*. I like that I don't even have to allocate registers
manually.
