Remove boxing from "value is S" for nullables #95711
Conversation
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsCloses #95685 It turns out to be a simple fix bool Test1<T>(T value) // T is int?
{
return value is int;
}
void Test2<T>(T value) // T is int?
{
if (value is int)
Console.WriteLine();
}; Assembly listing for method Prog:Test1[System.Nullable`1[int]]
- sub rsp, 40
- mov qword ptr [rsp+0x38], rdx
- lea rdx, [rsp+0x38]
- mov rcx, 0xD1FFAB1E ; System.Nullable`1[int]
- call CORINFO_HELP_BOX_NULLABLE
- test rax, rax
- je SHORT G_M33107_IG04
- mov rcx, 0xD1FFAB1E ; System.Int32
- xor rdx, rdx
- cmp qword ptr [rax], rcx
- cmovne rax, rdx
-G_M33107_IG04:
- test rax, rax
- setne al
- movzx rax, al
- add rsp, 40
+ xor eax, eax
ret
-; Total bytes of code 67
+; Total bytes of code 3
; Assembly listing for method Prog:Test2[System.Nullable`1[int]]
- sub rsp, 40
- mov qword ptr [rsp+0x38], rdx
- cmp byte ptr [rsp+0x38], 0
- je SHORT G_M14037_IG04
- call [System.Console:WriteLine()]
-G_M14037_IG04:
- nop
- add rsp, 40
ret
-; Total bytes of code 28
+; Total bytes of code 1
|
362da42 to
f0c4fff
Compare
| case CEE_LDNULL: | ||
| { | ||
| // "ldnull + cgt_un" is used when BOX+ISINST is not fed into a branch, e.g.: | ||
| // | ||
| // if (obj is string) { <--- box + isinst + br_true | ||
| // | ||
| // bool b = obj is string; <--- box + isinst + ldnull + cgt_un | ||
| // | ||
| // bool b = obj is not string; <--- box + isinst + ldnull + cgt_un + ldc.i4.0 + ceq | ||
| // |
There was a problem hiding this comment.
I did similar transform in #47920 (comment) . It's interesting to see which implementation covers more cases.
I'll push it later today.
There was a problem hiding this comment.
I think I tried this transformation I while back and gave up due to small diffs, decided to push it because of customer reported issue
There was a problem hiding this comment.
main...huoyaoyuan:runtime:isinst-null-cmp
It doesn't help, and doesn't conflict either. Let's do separately.
There was a problem hiding this comment.
I think my change is simpler? (if you ignore whitespaces).
Also, not sure if box + isinst + ldnull + ceq ever shows up - Roslyn seems to not emitting it so I didn't bother
E.g. sharplab
|
PTAL @dotnet/jit-contrib, @jakobbotsch A small change (if you disable whitespaces) that closes #95685 Almost no hits in BCL (except a few tests in minopts) |
Closes #95685
I recommend reviewing with "hide whitespaces" mode on the review tab. I removed some codeAddr validations and replaced those with
impGetNonPrefixOpcodethat does the same.