Suboptimal ASM emmited for Vector256<T>.Zero and Vector128<T>.Zero

[israellot]

When using Vector256.Zero, I would expect it to be kept in a fixed register and reused. Instead, what I see is a vxorps operation emitted every time.

AVX2 has 16 YMM registers.

Bellow is one example. I can get the desired behavior by forcing a zero vector variable instead of using Vector256.Zero;
Assigning Vector256<byte>.Zero to a variable alone does not do the trick. Only the extra xor operation ensures it stays in a fixed register.

var byteVector = Vector256.LoadUnsafe<byte>(ref spanRef);
           
var low = Avx2.UnpackLow(byteVector, Vector256<byte>.Zero);
var high = Avx2.UnpackHigh(byteVector, Vector256<byte>.Zero);

var added = Avx2.Add(low.AsInt16(), high.AsInt16());

added = Avx2.HorizontalAdd(added, Vector256<short>.Zero);
added = Avx2.HorizontalAdd(added, Vector256<short>.Zero);
added = Avx2.HorizontalAdd(added, Vector256<short>.Zero);

//ASM

mov      rax, bword ptr [rcx]
vmovdqu  ymm0, ymmword ptr[rax]
vxorps   ymm1, ymm1, ymm1
vpunpcklbw ymm1, ymm0, ymm1
vxorps   ymm2, ymm2, ymm2
vpunpckhbw ymm0, ymm0, ymm2
vpaddw   ymm0, ymm1, ymm0
vxorps   ymm1, ymm1, ymm1
vphaddw  ymm0, ymm0, ymm1
vxorps   ymm1, ymm1, ymm1
vphaddw  ymm0, ymm0, ymm1
vxorps   ymm1, ymm1, ymm1
vphaddw  ymm0, ymm0, ymm1

var byteVector = Vector256.LoadUnsafe<byte>(ref spanRef);

var zero = Vector256<byte>.Zero;
zero = Avx2.Xor(zero, zero); //forces fixed register
            
var low = Avx2.UnpackLow(byteVector, zero);
var high = Avx2.UnpackHigh(byteVector, zero);

var added = Avx2.Add(low.AsInt16(), high.AsInt16());
added = Avx2.HorizontalAdd(added, zero.AsInt16());
added = Avx2.HorizontalAdd(added, zero.AsInt16());
added = Avx2.HorizontalAdd(added, zero.AsInt16());


//ASM
mov      rax, bword ptr [rcx]
vxorps   ymm0, ymm0, ymm0
vmovdqu  ymm1, ymmword ptr[rax]
vpunpcklbw ymm2, ymm1, ymm0
vpunpckhbw ymm1, ymm1, ymm0
vpaddw   ymm1, ymm2, ymm1
vphaddw  ymm1, ymm1, ymm0
vphaddw  ymm1, ymm1, ymm0
vphaddw  ymm1, ymm1, ymm0

category:cq
theme:cse
skill-level:intermediate
cost:medium
impact:small

[ghost]

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

Issue Details

---

When using Vector256.Zero, I would expect it to be kept in a fixed register and reused. Instead, what I see is a vxorps operation emitted every time.

AVX2 has 16 YMM registers.

Bellow is one example. I can get the desired behavior by forcing a zero vector variable instead of using Vector256.Zero;
Assigning Vector256<byte>.Zero to a variable alone does not do the trick. Only the extra xor operation ensures it stays in a fixed register.

var byteVector = Vector256.LoadUnsafe<byte>(ref spanRef);
           
var low = Avx2.UnpackLow(byteVector, Vector256<byte>.Zero);
var high = Avx2.UnpackHigh(byteVector, Vector256<byte>.Zero);

var added = Avx2.Add(low.AsInt16(), high.AsInt16());

added = Avx2.HorizontalAdd(added, Vector256<short>.Zero);
added = Avx2.HorizontalAdd(added, Vector256<short>.Zero);
added = Avx2.HorizontalAdd(added, Vector256<short>.Zero);

//ASM

mov      rax, bword ptr [rcx]
vmovdqu  ymm0, ymmword ptr[rax]
vxorps   ymm1, ymm1, ymm1
vpunpcklbw ymm1, ymm0, ymm1
vxorps   ymm2, ymm2, ymm2
vpunpckhbw ymm0, ymm0, ymm2
vpaddw   ymm0, ymm1, ymm0
vxorps   ymm1, ymm1, ymm1
vphaddw  ymm0, ymm0, ymm1
vxorps   ymm1, ymm1, ymm1
vphaddw  ymm0, ymm0, ymm1
vxorps   ymm1, ymm1, ymm1
vphaddw  ymm0, ymm0, ymm1

var byteVector = Vector256.LoadUnsafe<byte>(ref spanRef);

var zero = Vector256<byte>.Zero;
zero = Avx2.Xor(zero, zero); //forces fixed register
            
var low = Avx2.UnpackLow(byteVector, zero);
var high = Avx2.UnpackHigh(byteVector, zero);

var added = Avx2.Add(low.AsInt16(), high.AsInt16());
added = Avx2.HorizontalAdd(added, zero.AsInt16());
added = Avx2.HorizontalAdd(added, zero.AsInt16());
added = Avx2.HorizontalAdd(added, zero.AsInt16());

mov      rax, bword ptr [rcx]
vxorps   ymm0, ymm0, ymm0
vmovdqu  ymm1, ymmword ptr[rax]
vpunpcklbw ymm2, ymm1, ymm0
vpunpckhbw ymm1, ymm1, ymm0
vpaddw   ymm1, ymm2, ymm1
vphaddw  ymm1, ymm1, ymm0
vphaddw  ymm1, ymm1, ymm0
vphaddw  ymm1, ymm1, ymm0

Author:	israellot
Assignees:	-
Labels:	tenet-performance, area-CodeGen-coreclr
Milestone:	-

[EgorBo]

This is a known issue that we might eventually address with constant materialization during LSRA.
We don't currently do CSE for Vector<>.Zero because 1) it's cheap to materialize 2) we plan to rely on it in optimizations in lower*.cpp

[israellot]

Thanks for the info @EgorBo

[TIHan]

Discussed with @tannergooding for a bit regarding this, as an immediate thing we could do, is a peephole optimization to eliminate unnecessary vxorps ymm1, ymm1, ymm1 instructions by looking back to see if the same instruction occurred, as long as we know that ymm1 was not written to in any other way.

[EgorBo]

Another manifistation of this problem:

void Foo(ref byte b)
{
    Unsafe.As<byte, Vector128<byte>>(ref Unsafe.Add(ref b, 0)) = default;
    Unsafe.As<byte, Vector128<byte>>(ref Unsafe.Add(ref b, 16)) = default;
    Unsafe.As<byte, Vector128<byte>>(ref Unsafe.Add(ref b, 32)) = default;
    Unsafe.As<byte, Vector128<byte>>(ref Unsafe.Add(ref b, 48)) = default;
}

Emits:

            movi    v16.4s, #0
            str     q16, [x1]
            movi    v16.4s, #0
            str     q16, [x1, #0x10]
            movi    v16.4s, #0
            str     q16, [x1, #0x20]
            movi    v16.4s, #0
            str     q16, [x1, #0x30]

Expected:

            movi    v16.4s, #0
            str     q16, q16, [x1]
            str     q16, q16, [x1, #0x20]

eliminate unnecessary vxorps ymm1, ymm1, ymm1 instructions by looking back to see if the same instruction occurred, as long as we know that ymm1 was not written to in any other way.

TLDR: that didn't work due to ABI issues, although, it won't be a real fix anyway because it wouldn't help with loop hoisting.