GS: Add secondary vertex buffer for copy/modifying vertices.

Currently only used in HW renderer to fix vertices for provoking-first-vertex APIs.

pcsx2/GS/GSState.cpp

@@ -125,6 +125,8 @@ GSState::~GSState()
 {
 	if (m_vertex.buff)
 		_aligned_free(m_vertex.buff);
+	if (m_vertex.buff_copy)
+		_aligned_free(m_vertex.buff_copy);
 	if (m_index.buff)
 		_aligned_free(m_index.buff);
 	if (m_draw_vertex.buff)
@@ -2859,56 +2861,53 @@ void GSState::UpdateVertexKick()
 void GSState::GrowVertexBuffer()
 {
 	const u32 maxcount = std::max<u32>(m_vertex.maxcount * 3 / 2, 10000);
+	const u32 old_vertex_size = sizeof(GSVertex) * m_vertex.tail;
+	const u32 new_vertex_size = sizeof(GSVertex) * maxcount;
+	const u32 old_index_size = sizeof(u16) * m_index.tail;
+	const u32 new_index_size = sizeof(u16) * maxcount * 6; // Worst case index list is a list of points with vs expansion, 6 indices per point
 
-	GSVertex* vertex = static_cast<GSVertex*>(_aligned_malloc(sizeof(GSVertex) * maxcount, 32));
-	GSVertex* draw_vertex = static_cast<GSVertex*>(_aligned_malloc(sizeof(GSVertex) * maxcount, 32));
-	// Worst case index list is a list of points with vs expansion, 6 indices per point
-	u16* index = static_cast<u16*>(_aligned_malloc(sizeof(u16) * maxcount * 6, 32));
-	u16* draw_index = static_cast<u16*>(_aligned_malloc(sizeof(u16) * maxcount * 6, 32));
-
-	if (!vertex || !index)
+	// Structure describing buffers to reallocate
+	struct AllocDesc
 	{
-		const u32 vert_byte_count = sizeof(GSVertex) * maxcount;
-		const u32 idx_byte_count = sizeof(u16) * maxcount * 3;
+		void** pbuff;
+		u32 old_size;
+		u32 new_size;
+	};
+	std::vector<AllocDesc> alloc_desc = {
+		{reinterpret_cast<void**>(&m_vertex.buff),      old_vertex_size, new_vertex_size},
+		// discard contents of buff_copy by setting old_size = 0
+		{reinterpret_cast<void**>(&m_vertex.buff_copy), 0,               new_vertex_size},
+		{reinterpret_cast<void**>(&m_draw_vertex.buff), old_vertex_size, new_vertex_size},
+		{reinterpret_cast<void**>(&m_index.buff),       old_index_size,  new_index_size},
+		{reinterpret_cast<void**>(&m_draw_index.buff),  old_index_size,  new_index_size}
+	};
 
-		Console.Error("GS: failed to allocate %zu bytes for vertices and %zu for indices.",
-			vert_byte_count, idx_byte_count);
-		pxFailRel("Memory allocation failed");
+	// For logging
+	u32 total_size = 0;
+	for (const auto& desc : alloc_desc)
+		total_size += desc.new_size;
+
+	// Reallocate each of the needed buffers
+	for (const auto [pbuff, old_size, new_size] : alloc_desc)
+	{
+		void* new_buff = _aligned_malloc(new_size, 32);
+		if (!new_buff)
+		{
+			Console.Error("GS: failed to allocate %zu bytes for vertices and indices.", total_size);
+			pxFailRel("Memory allocation failed");
+		}
+		if (*pbuff)
+		{
+			if (old_size)
+			{
+				std::memcpy(new_buff, *pbuff, old_size);
+			}
+			_aligned_free(*pbuff);
+		}
+		*pbuff = new_buff;
 	}
 
-	if (m_vertex.buff)
-	{
-		std::memcpy(vertex, m_vertex.buff, sizeof(GSVertex) * m_vertex.tail);
-
-		_aligned_free(m_vertex.buff);
-	}
-
-	if (m_index.buff)
-	{
-		std::memcpy(index, m_index.buff, sizeof(u16) * m_index.tail);
-
-		_aligned_free(m_index.buff);
-	}
-
-	if (m_draw_vertex.buff)
-	{
-		std::memcpy(draw_vertex, m_draw_vertex.buff, sizeof(GSVertex) * m_vertex.tail);
-
-		_aligned_free(m_draw_vertex.buff);
-	}
-
-	if (m_draw_index.buff)
-	{
-		std::memcpy(draw_index, m_draw_index.buff, sizeof(u16) * m_index.tail);
-
-		_aligned_free(m_draw_index.buff);
-	}
-
-	m_draw_vertex.buff = draw_vertex;
-	m_draw_index.buff = draw_index;
-	m_vertex.buff = vertex;
 	m_vertex.maxcount = maxcount - 3; // -3 to have some space at the end of the buffer before DrawingKick can grow it
-	m_index.buff = index;
 }
 
 bool GSState::TrianglesAreQuads(bool shuffle_check)

pcsx2/GS/GSState.h

@@ -136,6 +136,7 @@ protected:
 	struct
 	{
 		GSVertex* buff;
+		GSVertex* buff_copy;            // same size buffer to copy/modify the original buffer
 		u32 head, tail, next, maxcount; // head: first vertex, tail: last vertex + 1, next: last indexed + 1
 		u32 xy_tail;
 		GSVector4i xy[4];

pcsx2/GS/Renderers/HW/GSRendererHW.cpp

@@ -4834,34 +4834,15 @@ void GSRendererHW::HandleProvokingVertexFirst()
 	if (first_eq_last)
 		return;
 
-	// De-index the vertices either in place or by reallocating the vertex buffer.
-	if (m_vertex.next <= m_index.tail && m_index.tail <= m_vertex.maxcount)
+	// De-index the vertices using the copy buffer
+	while (m_vertex.maxcount < m_index.tail)
+		GrowVertexBuffer();
+	for (int i = static_cast<int>(m_index.tail) - 1; i >= 0; i--)
 	{
-		// De-index in place
-		for (int i = static_cast<int>(m_index.tail) - 1; i >= 0; i--)
-		{
-			// FIXME: This might not hold with a large triangle fan with gaps (since gaps are not
-			// yet removed)!
-			pxAssert(m_index.buff[i] <= i); // At any point, there can never be more vertices than indices
-			m_vertex.buff[i] = m_vertex.buff[m_index.buff[i]];
-			m_index.buff[i] = static_cast<u16>(i);
-		}
-	}
-	else
-	{
-		// Reallocate the vertex buffer
-		m_vertex.maxcount = std::max(m_vertex.maxcount, m_index.tail);
-		GSVertex* vert_buff = static_cast<GSVertex*>(_aligned_malloc(sizeof(GSVertex) * m_vertex.maxcount, 32));
-
-		// De-index and copy the vertices
-		for (u32 i = 0; i < m_index.tail; i++)
-		{
-			vert_buff[i] = m_vertex.buff[m_index.buff[i]];
-			m_index.buff[i] = static_cast<u16>(i);
-		}
-		std::swap(vert_buff, m_vertex.buff);
-		_aligned_free(vert_buff);
+		m_vertex.buff_copy[i] = m_vertex.buff[m_index.buff[i]];
+		m_index.buff[i] = static_cast<u16>(i);
 	}
+	std::swap(m_vertex.buff, m_vertex.buff_copy);
 	m_vertex.head = m_vertex.next = m_vertex.tail = m_index.tail;
 
 	// Put correct color in the first vertex