inline VertexLoaderBase *GetOrAddLoader(const TVtxDesc &VtxDesc, const VAT &VtxAttr)
{
VertexLoaderUID uid(VtxDesc, VtxAttr);
VertexLoaderMap::iterator iter = s_vertex_loader_map.find(uid);
if (iter == s_vertex_loader_map.end())
{
s_vertex_loader_map[uid] = VertexLoaderBase::CreateVertexLoader(VtxDesc, VtxAttr);
VertexLoaderBase* loader = s_vertex_loader_map[uid].get();
loader->m_native_vertex_format = GetNativeVertexFormat(loader->m_native_vtx_decl);
VertexLoaderBase * fallback = loader->GetFallback();
if (fallback)
{
fallback->m_native_vertex_format = GetNativeVertexFormat(fallback->m_native_vtx_decl);
}
INCSTAT(stats.numVertexLoaders);
return loader;
}
return iter->second.get();
}
int RunVertices(int vtx_attr_group, int primitive, int count, DataReader src, bool skip_drawing,
bool is_preprocess)
{
if (!count)
return 0;
VertexLoaderBase* loader = RefreshLoader(vtx_attr_group, is_preprocess);
int size = count * loader->m_VertexSize;
if ((int)src.size() < size)
return -1;
if (skip_drawing || is_preprocess)
return size;
// If the native vertex format changed, force a flush.
if (loader->m_native_vertex_format != s_current_vtx_fmt ||
loader->m_native_components != g_current_components)
{
VertexManagerBase::Flush();
}
s_current_vtx_fmt = loader->m_native_vertex_format;
g_current_components = loader->m_native_components;
// if cull mode is CULL_ALL, tell VertexManager to skip triangles and quads.
// They still need to go through vertex loading, because we need to calculate a zfreeze refrence
// slope.
bool cullall = (bpmem.genMode.cullmode == GenMode::CULL_ALL && primitive < 5);
DataReader dst = VertexManagerBase::PrepareForAdditionalData(
primitive, count, loader->m_native_vtx_decl.stride, cullall);
count = loader->RunVertices(src, dst, count);
IndexGenerator::AddIndices(primitive, count);
VertexManagerBase::FlushData(count, loader->m_native_vtx_decl.stride);
ADDSTAT(stats.thisFrame.numPrims, count);
INCSTAT(stats.thisFrame.numPrimitiveJoins);
return size;
}
TEST_F(VertexLoaderTest, PositionDirectU16XYSpeed)
{
m_vtx_desc.Position = 1; // Direct
m_vtx_attr.g0.PosElements = 0; // XY
m_vtx_attr.g0.PosFormat = 2; // U16
VertexLoaderBase* loader = VertexLoaderBase::CreateVertexLoader(m_vtx_desc, m_vtx_attr);
ASSERT_EQ(3 * sizeof (float), (u32)loader->m_native_vtx_decl.stride);
ASSERT_EQ(2 * sizeof (u16), (u32)loader->m_VertexSize);
for (int i = 0; i < 1000; ++i)
{
ResetPointers();
int count = loader->RunVertices(7, 100000, src, dst);
src.Skip(100000 * loader->m_VertexSize);
dst.Skip(count * loader->m_native_vtx_decl.stride);
}
delete loader;
}
TEST_F(VertexLoaderTest, PositionDirectU16XY)
{
m_vtx_desc.Position = 1; // Direct
m_vtx_attr.g0.PosElements = 0; // XY
m_vtx_attr.g0.PosFormat = 2; // U16
VertexLoaderBase* loader = VertexLoaderBase::CreateVertexLoader(m_vtx_desc, m_vtx_attr);
ASSERT_EQ(3 * sizeof (float), (u32)loader->m_native_vtx_decl.stride);
ASSERT_EQ(2 * sizeof (u16), (u32)loader->m_VertexSize);
// Write some vertices.
Input<u16>(0); Input<u16>(0);
Input<u16>(1); Input<u16>(2);
Input<u16>(256); Input<u16>(257);
Input<u16>(65535); Input<u16>(65534);
Input<u16>(12345); Input<u16>(54321);
// Convert 5 points. "7" -> primitive are points.
int count = loader->RunVertices(7, 5, src, dst);
src.Skip(5 * loader->m_VertexSize);
dst.Skip(count * loader->m_native_vtx_decl.stride);
delete loader;
ExpectOut(0.0f); ExpectOut(0.0f); ExpectOut(0.0f);
ExpectOut(1.0f); ExpectOut(2.0f); ExpectOut(0.0f);
ExpectOut(256.0f); ExpectOut(257.0f); ExpectOut(0.0f);
ExpectOut(65535.0f); ExpectOut(65534.0f); ExpectOut(0.0f);
ExpectOut(12345.0f); ExpectOut(54321.0f); ExpectOut(0.0f);
// Test that scale works on U16 inputs.
Input<u16>(42); Input<u16>(24);
m_vtx_attr.g0.PosFrac = 1;
m_vtx_attr.g0.ByteDequant = 1;
loader = VertexLoaderBase::CreateVertexLoader(m_vtx_desc, m_vtx_attr);
count = loader->RunVertices(7, 1, src, dst);
src.Skip(1 * loader->m_VertexSize);
dst.Skip(count * loader->m_native_vtx_decl.stride);
ExpectOut(21.0f); ExpectOut(12.0f); ExpectOut(0.0f);
delete loader;
}
TEST_F(VertexLoaderTest, PositionDirectFloatXYZ)
{
m_vtx_desc.Position = 1; // Direct
m_vtx_attr.g0.PosElements = 1; // XYZ
m_vtx_attr.g0.PosFormat = 4; // Float
VertexLoaderBase* loader = VertexLoaderBase::CreateVertexLoader(m_vtx_desc, m_vtx_attr);
ASSERT_EQ(3 * sizeof (float), (u32)loader->m_native_vtx_decl.stride);
ASSERT_EQ(3 * sizeof (float), (u32)loader->m_VertexSize);
// Write some vertices.
Input(0.0f); Input(0.0f); Input(0.0f);
Input(1.0f); Input(0.0f); Input(0.0f);
Input(0.0f); Input(1.0f); Input(0.0f);
Input(0.0f); Input(0.0f); Input(1.0f);
// Convert 4 points. "7" -> primitive are points.
int count = loader->RunVertices(7, 4, src, dst);
src.Skip(4 * loader->m_VertexSize);
dst.Skip(count * loader->m_native_vtx_decl.stride);
delete loader;
ExpectOut(0.0f); ExpectOut(0.0f); ExpectOut(0.0f);
ExpectOut(1.0f); ExpectOut(0.0f); ExpectOut(0.0f);
ExpectOut(0.0f); ExpectOut(1.0f); ExpectOut(0.0f);
ExpectOut(0.0f); ExpectOut(0.0f); ExpectOut(1.0f);
// Test that scale does nothing for floating point inputs.
Input(1.0f); Input(2.0f); Input(4.0f);
m_vtx_attr.g0.PosFrac = 1;
loader = VertexLoaderBase::CreateVertexLoader(m_vtx_desc, m_vtx_attr);
count = loader->RunVertices(7, 1, src, dst);
src.Skip(1 * loader->m_VertexSize);
dst.Skip(count * loader->m_native_vtx_decl.stride);
ExpectOut(1.0f); ExpectOut(2.0f); ExpectOut(4.0f);
delete loader;
}
VertexLoaderBase* VertexLoaderBase::CreateVertexLoader(const TVtxDesc& vtx_desc, const VAT& vtx_attr)
{
VertexLoaderBase* loader;
//#define COMPARE_VERTEXLOADERS
#if defined(COMPARE_VERTEXLOADERS) && defined(_M_X86_64)
// first try: Any new VertexLoader vs the old one
loader = new VertexLoaderTester(
new VertexLoader(vtx_desc, vtx_attr), // the software one
new VertexLoaderX64(vtx_desc, vtx_attr), // the new one to compare
vtx_desc, vtx_attr);
if (loader->IsInitialized())
return loader;
delete loader;
#elif defined(_M_X86_64)
loader = new VertexLoaderX64(vtx_desc, vtx_attr);
if (loader->IsInitialized())
return loader;
delete loader;
#elif defined(_M_ARM_64)
loader = new VertexLoaderARM64(vtx_desc, vtx_attr);
if (loader->IsInitialized())
return loader;
delete loader;
#endif
// last try: The old VertexLoader
loader = new VertexLoader(vtx_desc, vtx_attr);
if (loader->IsInitialized())
return loader;
delete loader;
PanicAlert("No Vertex Loader found.");
return nullptr;
}
TEST_F(VertexLoaderTest, LargeFloatVertexSpeed)
{
// Enables most attributes in floating point direct mode to test speed.
m_vtx_desc.PosMatIdx = 1;
m_vtx_desc.Tex0MatIdx = 1;
m_vtx_desc.Tex1MatIdx = 1;
m_vtx_desc.Tex2MatIdx = 1;
m_vtx_desc.Tex3MatIdx = 1;
m_vtx_desc.Tex4MatIdx = 1;
m_vtx_desc.Tex5MatIdx = 1;
m_vtx_desc.Tex6MatIdx = 1;
m_vtx_desc.Tex7MatIdx = 1;
m_vtx_desc.Position = 1;
m_vtx_desc.Normal = 1;
m_vtx_desc.Color0 = 1;
m_vtx_desc.Color1 = 1;
m_vtx_desc.Tex0Coord = 1;
m_vtx_desc.Tex1Coord = 1;
m_vtx_desc.Tex2Coord = 1;
m_vtx_desc.Tex3Coord = 1;
m_vtx_desc.Tex4Coord = 1;
m_vtx_desc.Tex5Coord = 1;
m_vtx_desc.Tex6Coord = 1;
m_vtx_desc.Tex7Coord = 1;
m_vtx_attr.g0.PosElements = 1; // XYZ
m_vtx_attr.g0.PosFormat = 4; // Float
m_vtx_attr.g0.NormalElements = 1; // NBT
m_vtx_attr.g0.NormalFormat = 4; // Float
m_vtx_attr.g0.Color0Elements = 1; // Has Alpha
m_vtx_attr.g0.Color0Comp = 5; // RGBA8888
m_vtx_attr.g0.Color1Elements = 1; // Has Alpha
m_vtx_attr.g0.Color1Comp = 5; // RGBA8888
m_vtx_attr.g0.Tex0CoordElements = 1; // ST
m_vtx_attr.g0.Tex0CoordFormat = 4; // Float
m_vtx_attr.g1.Tex1CoordElements = 1; // ST
m_vtx_attr.g1.Tex1CoordFormat = 4; // Float
m_vtx_attr.g1.Tex2CoordElements = 1; // ST
m_vtx_attr.g1.Tex2CoordFormat = 4; // Float
m_vtx_attr.g1.Tex3CoordElements = 1; // ST
m_vtx_attr.g1.Tex3CoordFormat = 4; // Float
m_vtx_attr.g1.Tex4CoordElements = 1; // ST
m_vtx_attr.g1.Tex4CoordFormat = 4; // Float
m_vtx_attr.g2.Tex5CoordElements = 1; // ST
m_vtx_attr.g2.Tex5CoordFormat = 4; // Float
m_vtx_attr.g2.Tex6CoordElements = 1; // ST
m_vtx_attr.g2.Tex6CoordFormat = 4; // Float
m_vtx_attr.g2.Tex7CoordElements = 1; // ST
m_vtx_attr.g2.Tex7CoordFormat = 4; // Float
VertexLoaderBase* loader = VertexLoaderBase::CreateVertexLoader(m_vtx_desc, m_vtx_attr);
// This test is only done 100x in a row since it's ~20x slower using the
// current vertex loader implementation.
for (int i = 0; i < 100; ++i)
{
ResetPointers();
int count = loader->RunVertices(7, 100000, src, dst);
src.Skip(100000 * loader->m_VertexSize);
dst.Skip(count * loader->m_native_vtx_decl.stride);
}
delete loader;
}
