Can we improve ROS<byte>.ToArray from RVA static? #68103
Description
When doing new byte[] { const, const, const, ... }, the JIT generates code that calls the new array helper and then emits an unrolled loop to copy the data into the array. When doing ((ReadOnlySpan<byte>)new byte[] { const, const, const, ... }).ToArray(), the copy is handled via ToArray's call to Buffer.MemMove. Is it feasible to do better?
using System;
public class C
{
public byte[] Implicit() => new byte[] { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16 };
public byte[] Explicit() => ((ReadOnlySpan<byte>)new byte[] { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16 }).ToArray();
}; Core CLR 6.0.222.6406 on amd64
C.Implicit()
L0000: sub rsp, 0x28
L0004: vzeroupper
L0007: mov rcx, 0x7ffa6ca70a90
L0011: mov edx, 0x10
L0016: call 0x00007ffacc4fb1b0
L001b: mov rdx, 0x221bb0a0894
L0025: vmovupd xmm0, [rdx]
L0029: vmovupd [rax+0x10], xmm0
L002e: add rsp, 0x28
L0032: ret
C.Explicit()
L0000: push rsi
L0001: sub rsp, 0x20
L0005: mov edx, 0x10
L000a: mov rcx, 0x7ffa6ca70a90
L0014: call 0x00007ffacc4fb1b0
L0019: mov rsi, rax
L001c: lea rcx, [rsi+0x10]
L0020: mov rdx, 0x221bb0a0894
L002a: mov r8d, 0x10
L0030: call 0x00007ffa6c97e808
L0035: mov rax, rsi
L0038: add rsp, 0x20
L003c: pop rsi
L003d: ret
category:cq
theme:optimization
skill-level:beginner
cost:small
impact:small