GS/OGL: Align texture uploads to 64 bytes

Fixes potential crash in some games with odd-sized targets
and preloading (e.g. Densha De Go).

pcsx2/GS/Renderers/OpenGL/GSTextureOGL.cpp

@@ -21,10 +21,17 @@
 #include "GS/GSPerfMon.h"
 #include "GS/GSPng.h"
 #include "GS/GSGL.h"
+#include "common/Align.h"
 #include "common/AlignedMalloc.h"
 #include "common/StringUtil.h"
 
-static constexpr u32 TEXTURE_UPLOAD_ALIGNMENT = 256;
+// Looking across a range of GPUs, the optimal copy alignment for Vulkan drivers seems
+// to be between 1 (AMD/NV) and 64 (Intel). So, we'll go with 64 here.
+static constexpr u32 TEXTURE_UPLOAD_ALIGNMENT = 64;
+
+// The pitch alignment must be less or equal to the upload alignment.
+// We need 32 here for AVX2, so 64 is also fine.
+static constexpr u32 TEXTURE_UPLOAD_PITCH_ALIGNMENT = 64;
 
 GSTextureOGL::GSTextureOGL(Type type, int width, int height, int levels, Format format)
 {
@@ -214,8 +221,8 @@ bool GSTextureOGL::Update(const GSVector4i& r, const void* data, int pitch, int
 
 	m_clean = false;
 
-	u32 row_byte = r.width() << m_int_shift;
+	const u32 preferred_pitch = Common::AlignUpPow2(r.width() << m_int_shift, TEXTURE_UPLOAD_PITCH_ALIGNMENT);
-	u32 map_size = r.height() * row_byte;
+	const u32 map_size = r.height() * preferred_pitch;
 
 #if 0
 	if (r.height() == 1) {
@@ -250,7 +257,7 @@ bool GSTextureOGL::Update(const GSVector4i& r, const void* data, int pitch, int
 		GL::StreamBuffer* const sb = GSDeviceOGL::GetTextureUploadBuffer();
 
 		const auto map = sb->Map(TEXTURE_UPLOAD_ALIGNMENT, map_size);
-		StringUtil::StrideMemCpy(map.pointer, row_byte, data, pitch, row_byte, r.height());
+		StringUtil::StrideMemCpy(map.pointer, preferred_pitch, data, pitch, r.width() << m_int_shift, r.height());
 		sb->Unmap(map_size);
 		sb->Bind();
 
@@ -275,13 +282,13 @@ bool GSTextureOGL::Map(GSMap& m, const GSVector4i* _r, int layer)
 	ASSERT(r.width() != 0);
 	ASSERT(r.height() != 0);
 
-	u32 row_byte = r.width() << m_int_shift;
+	const u32 pitch = Common::AlignUpPow2(r.width() << m_int_shift, TEXTURE_UPLOAD_PITCH_ALIGNMENT);
-	m.pitch = row_byte;
+	m.pitch = pitch;
 
 	if (m_type == Type::Texture || m_type == Type::RenderTarget)
 	{
-		const u32 map_size = r.height() * row_byte;
+		const u32 upload_size = CalcUploadSize(r.height(), pitch);
-		if (GLLoader::buggy_pbo || map_size > GSDeviceOGL::GetTextureUploadBuffer()->GetChunkSize())
+		if (GLLoader::buggy_pbo || upload_size > GSDeviceOGL::GetTextureUploadBuffer()->GetChunkSize())
 			return false;
 
 		GL_PUSH_("Upload Texture %d", m_texture_id); // POP is in Unmap
@@ -289,7 +296,7 @@ bool GSTextureOGL::Map(GSMap& m, const GSVector4i* _r, int layer)
 
 		m_clean = false;
 
-		const auto map = GSDeviceOGL::GetTextureUploadBuffer()->Map(TEXTURE_UPLOAD_ALIGNMENT, map_size);
+		const auto map = GSDeviceOGL::GetTextureUploadBuffer()->Map(TEXTURE_UPLOAD_ALIGNMENT, upload_size);
 		m.bits = static_cast<u8*>(map.pointer);
 
 		// Save the area for the unmap
@@ -310,14 +317,20 @@ void GSTextureOGL::Unmap()
 {
 	if (m_type == Type::Texture || m_type == Type::RenderTarget)
 	{
-		const u32 map_size = (m_r_w << m_int_shift) * m_r_h;
+		const u32 pitch = Common::AlignUpPow2(m_r_w << m_int_shift, TEXTURE_UPLOAD_PITCH_ALIGNMENT);
+		const u32 upload_size = pitch * m_r_h;
 		GL::StreamBuffer* sb = GSDeviceOGL::GetTextureUploadBuffer();
-		sb->Unmap(map_size);
+		sb->Unmap(upload_size);
 		sb->Bind();
 
+		const u32 row_length = CalcUploadRowLengthFromPitch(pitch);
+		glPixelStorei(GL_UNPACK_ROW_LENGTH, row_length);
+
 		glTextureSubImage2D(m_texture_id, m_layer, m_r_x, m_r_y, m_r_w, m_r_h, m_int_format, m_int_type,
 			reinterpret_cast<void*>(static_cast<uintptr_t>(m_map_offset)));
 
+		glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
+
 		sb->Unbind();
 
 		m_needs_mipmaps_generated = true;