Method VertexKick

pcsx2/GS/GSState.cpp:5860–6117

template <u32 prim, bool auto_flush>
__forceinline void GSState::VertexKick(u32 skip)
{
	constexpr u32 n = NumIndicesForPrim(prim);
	constexpr int primclass = GSUtil::GetPrimClass(prim);
	static_assert(n > 0);

	pxAssert(m_vertex->tail < m_vertex->maxcount + 3);

	if constexpr (prim == GS_INVALID)
	{
		m_vertex->tail = m_vertex->head;
		return;
	}

	if (CheckOverlapVerts(n))
		Flush(CONTEXTCHANGE);
	
	if (auto_flush && skip == 0 && m_index->tail > 0 && ((m_vertex->tail + 1) - m_vertex->head) >= n)
	{
		HandleAutoFlush<prim>();
	}

	u32 head = m_vertex->head;
	u32 tail = m_vertex->tail;
	u32 next = m_vertex->next;
	u32 xy_tail = m_vertex->xy_tail;

	if (GSIsHardwareRenderer() && GSLocalMemory::m_psm[m_context->ZBUF.PSM].bpp == 32)
	{
		if (GSConfig.UserHacks_Limit24BitDepth == GSLimit24BitDepth::PrioritizeUpper)
			m_v.XYZ.Z = ((m_v.XYZ.Z >> 8) & ~0xFF) | (m_v.XYZ.Z & 0xFF);
		else if (GSConfig.UserHacks_Limit24BitDepth == GSLimit24BitDepth::PrioritizeLower)
			m_v.XYZ.Z &= 0x00FFFFFF;
	}

	// callers should write XYZUVF to m_v.m[1] in one piece to have this load store-forwarded, either by the cpu or the compiler when this function is inlined

	const GSVector4i new_v0(m_v.m[0]);
	const GSVector4i new_v1(m_v.m[1]);

	GSVector4i* RESTRICT tailptr = (GSVector4i*)&m_vertex->buff[tail];

	tailptr[0] = new_v0;
	tailptr[1] = new_v1;

	// We maintain the X/Y coordinates for the last 4 vertices, as well as the head for triangle fans, so we can compute
	// the min/max, and cull degenerate triangles, which saves draws in some cases. Why 4? Mod 4 is cheaper than Mod 3.
	const GSVector4i xy = new_v1.xxxx().u16to32().sub32(m_xyof);
	m_vertex->xy[xy_tail & 3] = xy;

	// Backup head for triangle fans so we can read it later, otherwise it'll get lost after the 4th vertex.
	if (prim == GS_TRIANGLEFAN && tail == head)
		m_vertex->xyhead = xy;

	m_vertex->tail = ++tail;
	m_vertex->xy_tail = ++xy_tail;

	const u32 m = tail - head;