[JIT] X64 - Recognize patterns to emit `bts`, `btc`, `btr` instructions

[TIHan]

@tannergooding and I were having a discussion and these examples were brought up.

public class C {
    static uint SetBit(int position, uint bits)
    {
        return bits | (1u << position);
    }

    static uint ToggleBit(int position, uint bits)
    {
        return bits ^ (1u << position);
    }

    static uint ResetBit(int position, uint bits)
    {
        return bits & ~(1u << position);
    }
}

These examples currently emit:

C.SetBit(Int32, UInt32)
    L0000: mov eax, 1
    L0005: shlx eax, eax, ecx
    L000a: or eax, edx
    L000c: ret

C.ToggleBit(Int32, UInt32)
    L0000: mov eax, 1
    L0005: shlx eax, eax, ecx
    L000a: xor eax, edx
    L000c: ret

C.ResetBit(Int32, UInt32)
    L0000: mov eax, 1
    L0005: shlx eax, eax, ecx
    L000a: andn eax, eax, edx
    L000f: ret

They could emit:

SetBit(int, unsigned int):                            # @SetBit(int, unsigned int)
        mov     eax, ecx
        bts     eax, edx
        ret
ToggleBit(int, unsigned int):                         # @ToggleBit(int, unsigned int)
        mov     eax, ecx
        btc     eax, edx
        ret
ResetBit(int, unsigned int):                          # @ResetBit(int, unsigned int)
        mov     eax, ecx
        btr     eax, edx
        ret

In XARCH LIR, we have a GT_BT already, we can just add three more GT_BTS, GT_BTC, GT_BTR and recognize the patterns above to transform to those ops in lowering.

Related: #27382

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

Issue Details

---

@tannergooding and I were having a discussion and these examples were brought up.

public class C {
    static uint SetBit(int position, uint bits)
    {
        return bits | (1u << position);
    }

    static uint ToggleBit(int position, uint bits)
    {
        return bits ^ (1u << position);
    }

    static uint ResetBit(int position, uint bits)
    {
        return bits & ~(1u << position);
    }
}

These examples currently emit:

C.SetBit(Int32, UInt32)
    L0000: mov eax, 1
    L0005: shlx eax, eax, ecx
    L000a: or eax, edx
    L000c: ret

C.ToggleBit(Int32, UInt32)
    L0000: mov eax, 1
    L0005: shlx eax, eax, ecx
    L000a: xor eax, edx
    L000c: ret

C.ResetBit(Int32, UInt32)
    L0000: mov eax, 1
    L0005: shlx eax, eax, ecx
    L000a: andn eax, eax, edx
    L000f: ret

They could emit:

SetBit(int, unsigned int):                            # @SetBit(int, unsigned int)
        mov     eax, esi
        bts     eax, edi
        ret
ToggleBit(int, unsigned int):                         # @ToggleBit(int, unsigned int)
        mov     eax, esi
        btc     eax, edi
        ret
ResetBit(int, unsigned int):                          # @ResetBit(int, unsigned int)
        mov     eax, esi
        btr     eax, edi
        ret

In XARCH LIR, we have a GT_BT already, we can just add three more GT_BTS, GT_BTC, GT_BTR and recognize the patterns above to transform to those ops in lowering.

Related: #6815

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

[xtqqczze]

static uint SetBit(int position, uint bits) {}
static uint ToggleBit(int position, uint bits) {}
static uint ResetBit(int position, uint bits) {}

We could also implement these as public APIs in BitOperations?

[tannergooding]

Exposing the apis via public helpers is a separate consideration and one that would need to be in its own API proposal

[xtqqczze]

Merged in #83400:

runtime/src/libraries/System.Private.CoreLib/src/System/Numerics/BitOperations.cs

Lines 947 to 959 in c96c15c

	/// <summary>
	/// Flip the bit at a specific position in a given value.
	/// Similar in behavior to the x86 instruction BTC (Bit Test and Complement).
	/// </summary>
	/// <param name="value">The value.</param>
	/// <param name="index">The zero-based index of the bit to flip.
	/// Any value outside the range [0..31] is treated as congruent mod 32.</param>
	/// <returns>The new value.</returns>
	[MethodImpl(MethodImplOptions.AggressiveInlining)]
	internal static uint FlipBit(uint value, int index)
	{
	return value ^ (1u << index);
	}