JIT: Unroll GT_BLK via two regs instead of one

[EgorBo]

From #93214 (comment)

Currently, we unroll GT_BLK (for copy or init) via a single SIMD reg even for large blocks, e.g.:

[StructLayout(LayoutKind.Sequential, Size = 64)]
private struct Block64 { }

static void CopyBlock64(ref byte dest, ref byte src)
{
    Unsafe.As<byte, Block64>(ref dest) = Unsafe.As<byte, Block64>(ref src);
}

       vmovdqu  ymm0, ymmword ptr [rdx]
       vmovdqu  ymmword ptr [rcx], ymm0
       vmovdqu  ymm0, ymmword ptr [rdx+0x20]
       vmovdqu  ymmword ptr [rcx+0x20], ymm0

It'd be nice if GT_BLK unrolling could rent 2 regs for unrolling instead of 1, SIMD regs usually don't have big RA pressure anyway. so it could be then:

       vmovdqu  ymm0, ymmword ptr [rdx]
       vmovdqu  ymm1, ymmword ptr [rdx+0x20]
       vmovdqu  ymmword ptr [rcx], ymm0
       vmovdqu  ymmword ptr [rcx+0x20], ymm1

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Currently, we unroll GT_BLK (for copy or init) via a single SIMD reg even for large blocks, e.g.:

[StructLayout(LayoutKind.Sequential, Size = 64)]
private struct Block64 { }

static void CopyBlock64(ref byte dest, ref byte src)
{
    Unsafe.As<byte, Block64>(ref dest) = Unsafe.As<byte, Block64>(ref src);
}

       vmovdqu  ymm0, ymmword ptr [rdx]
       vmovdqu  ymmword ptr [rcx], ymm0
       vmovdqu  ymm0, ymmword ptr [rdx+0x20]
       vmovdqu  ymmword ptr [rcx+0x20], ymm0

It'd be nice if GT_BLK unrolling could rent 2 regs for unrolling instead of 1, SIMD regs usually don't have big RA pressure anyway. so it could be then:

       vmovdqu  ymm0, ymmword ptr [rdx]
       vmovdqu  ymm1, ymmword ptr [rdx+0x20]
       vmovdqu  ymmword ptr [rcx], ymm0
       vmovdqu  ymmword ptr [rcx+0x20], ymm1

Author:	EgorBo
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[EgorBo]

I wasn't able to find any cases on both arm and x86 where it was profitable