h.mask |= bits << h.valid
		h.valid += valid
		return h
	}
	// Too many bits to fit in this word. Write the current word
	// out and move on to the next word.

	data := h.mask | bits<<h.valid       // mask for this word
	h.mask = bits >> (ptrBits - h.valid) // leftover for next word
	h.valid += valid - ptrBits           // have h.valid+valid bits, writing ptrBits of them

	// Flush mask to the memory bitmap.

		p.dstStackSize = alignUp(p.dstStackSize, uintptr(t.Align_))
		return
	}

	// In the C ABI, we're already on a word boundary.
	// Also, sub-word-sized fastcall register arguments
	// are stored to the least-significant bytes of the
	// argument word and all supported Windows
	// architectures are little endian, so srcStackOffset
	// is already pointing to the right place for smaller

	MOVD  8(R8), R11
	MOVD  8(R9), R16
	SUBE  R16, R11, R20
	MOVD  R20, 8(R10)

final:
	ADDZE R4
	XOR   $1, R4

done:
	MOVD  R4, c+72(FP)
	RET

// func addVW(z, x []Word, y Word) (c Word)
TEXT ·addVW(SB), NOSPLIT, $0
	MOVD z+0(FP), R10	// R10 = z[]
	MOVD x+24(FP), R8	// R8 = x[]
	MOVD y+48(FP), R4	// R4 = y = c
	MOVD z_len+8(FP), R11	// R11 = z_len

		// Value is indirect, but interface is direct. We need
		// to load the data at v.ptr into the interface data word.
		e.Data = *(*unsafe.Pointer)(v.ptr)
	default:
		// Value is direct, and so is the interface.
		e.Data = v.ptr
	}
	// Now, fill in the type portion. We're very careful here not
	// to have any operation between the e.word and e.typ assignments
	// that would let the garbage collector observe the partially-built

I think `Union[SomeType, None]` is more explicit about what it means.

It's just about the words and names. But those words can affect how you and your teammates think about the code.

//
// On ARM, 386, and 32-bit MIPS, it is the caller's responsibility to arrange
// for 64-bit alignment of 64-bit words accessed atomically via the primitive
// atomic functions (types [Int64] and [Uint64] are automatically aligned).
// The first word in an allocated struct, array, or slice; in a global
// variable; or in a local variable (because the subject of all atomic operations

}

const (
	PT_TRACE_ME             = 0  // Debug this process
	PT_READ_I               = 1  // Read a full word
	PT_READ_D               = 2  // Read a full word
	PT_READ_U               = 3  // Read control info
	PT_WRITE_I              = 4  //Write a full word
	PT_WRITE_D              = 5  //Write a full word
	PT_CONTINUE             = 7  //Continue the process
	PT_KILL                 = 8  //Terminate the process

	HasAVX5124VNNIW     bool // Advanced vector extension 512 Vector Neural Network Instructions Word variable precision
	HasAVX5124FMAPS     bool // Advanced vector extension 512 Fused Multiply Accumulation Packed Single precision
	HasAVX512VPOPCNTDQ  bool // Advanced vector extension 512 Double and quad word population count instructions
	HasAVX512VPCLMULQDQ bool // Advanced vector extension 512 Vector carry-less multiply operations

}

// mul sets y to y*H, where H is the GCM key, fixed during NewGCMWithNonceSize.
func (g *gcm) mul(y *gcmFieldElement) {
	var z gcmFieldElement

	for i := 0; i < 2; i++ {
		word := y.high
		if i == 1 {
			word = y.low
		}

		// Multiplication works by multiplying z by 16 and adding in
		// one of the precomputed multiples of H.
		for j := 0; j < 64; j += 4 {
			msw := z.high & 0xf
			z.high >>= 4

// which can express repetition compactly. In either form, the
// information is used by the runtime to initialize the heap bitmap,
// and for large types (like 128 or more words), they are roughly the
// same speed. GC programs are never much larger and often more
// compact. (If large arrays are involved, they can be arbitrarily
// more compact.)
//