Repository navigation
JIT: Redundant bounds check for when loopUpperBound==arrayLength #114744
Description
Loop cloning generates a fast path without bounds check but this can be statically optimized without needing to keep the slow path: https://godbolt.org/z/3WeaEfWh3
static void Test(int count)
{
int[] arrA = new int[count];
float[] arrB = new float[count];
for (int i = 0; i < count; i++)
{
arrA[i] = 0; // bounds check
arrB[i] = 0.0f; // bounds check
}
}Program:Test(int) (FullOpts):
push rbp
push r15
push rbx
lea rbp, [rsp+0x10]
mov ebx, edi
movsxd rsi, ebx
mov rdi, 0x79E0784C8C88 ; int[]
call CORINFO_HELP_NEWARR_1_VC
mov r15, rax
movsxd rsi, ebx
mov rdi, 0x79E078711E90 ; float[]
call CORINFO_HELP_NEWARR_1_VC
xor ecx, ecx
test ebx, ebx
jle SHORT G_M29168_IG06
cmp dword ptr [r15+0x08], ebx
jl SHORT G_M29168_IG07
cmp dword ptr [rax+0x08], ebx
jl SHORT G_M29168_IG07
mov ecx, 16
align [0 bytes for IG04]
G_M29168_IG04: ;; offset=0x0048
xor edx, edx
mov dword ptr [r15+rcx], edx
mov dword ptr [rax+rcx], edx
add rcx, 4
dec ebx
jne SHORT G_M29168_IG04
G_M29168_IG06: ;; offset=0x0059
pop rbx
pop r15
pop rbp
ret
G_M29168_IG07: ;; offset=0x005E
cmp ecx, dword ptr [r15+0x08]
jae SHORT G_M29168_IG09
mov edx, ecx
xor edi, edi
mov dword ptr [r15+4*rdx+0x10], edi
cmp ecx, dword ptr [rax+0x08]
jae SHORT G_M29168_IG09
mov edx, ecx
mov dword ptr [rax+4*rdx+0x10], edi
inc ecx
cmp ecx, ebx
jl SHORT G_M29168_IG07
jmp SHORT G_M29168_IG06
G_M29168_IG09: ;; offset=0x0080
call CORINFO_HELP_RNGCHKFAIL
int3
Activity
dotnet-policy-service commentedon Apr 16, 2025
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.
JulieLeeMSFT commentedon Apr 17, 2025
CC @BruceForstall, @dotnet/jit-contrib.
EgorBo commentedon Apr 20, 2025
The LC's checks should be foldable at VN level, we basically have: