-
Notifications
You must be signed in to change notification settings - Fork 4.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
JIT: Intrinsify ClearWithoutReferences and Fill #98700
Conversation
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsThis PR enables more memset-like unrollings for these APIs. Also, it fixes some potential regressions from #98623 which removes Example of what it does compared to Main/.NET 7.0: static void Test(Span<int> span)
{
span.Slice(0, 10).Fill(0);
} Current codegen:; Assembly listing for method Prog:Test(System.Span`1[int]) (FullOpts)
G_M6718_IG01:
sub rsp, 40
G_M6718_IG02:
cmp dword ptr [rcx+0x08], 10
jb SHORT G_M6718_IG04
mov rcx, bword ptr [rcx]
mov edx, 10
xor r8d, r8d
call [System.SpanHelpers:Fill[int](byref,ulong,int)]
nop
G_M6718_IG03:
add rsp, 40
ret
G_M6718_IG04:
call [System.ThrowHelper:ThrowArgumentOutOfRangeException()]
int3
; Total bytes of code 40 New codegen; Assembly listing for method Prog:Test(System.Span`1[int]) (FullOpts)
G_M6718_IG01:
sub rsp, 40
G_M6718_IG02:
cmp dword ptr [rcx+0x08], 10
jb SHORT G_M6718_IG04
mov rax, bword ptr [rcx]
vxorps ymm0, ymm0, ymm0
vmovdqu ymmword ptr [rax], ymm0
vmovdqu xmmword ptr [rax+0x18], xmm0
G_M6718_IG03:
add rsp, 40
ret
G_M6718_IG04:
call [System.ThrowHelper:ThrowArgumentOutOfRangeException()]
int3
; Total bytes of code 38
|
Co-authored-by: SingleAccretion <62474226+SingleAccretion@users.noreply.github.com>
@jakobbotsch @dotnet/jit-contrib PTAL. This is needed to unblock #98623 which removes unrollable Unsafe.InitBlock from two places. This impl is similar to LowerMemmove that @jakobbotsch reviewed in the past |
@jakobbotsch I've addressed your feedback (thanks, it made it simpler!) PTAL |
Grr.. clicked "Update branch" instead of "Merge" 😠 |
Possible improvements (Only the clear related benchmarks): |
This PR enables more memset-like unrollings for these APIs. Also, it fixes some potential regressions from #98623 which removes
Unsafe.InitBlockUnaligned
fromSpanHelpers.ClearWithoutReferences
andSpan.Fill
.Example of what it does compared to Main/.NET 8.0:
Current codegen:
New codegen