void GLGSRender::load_program(u32 vertex_base, u32 vertex_count)
{
auto &fragment_program = current_fragment_program;
auto &vertex_program = current_vertex_program;
for (auto &vtx : vertex_program.rsx_vertex_inputs)
{
auto &array_info = rsx::method_registers.vertex_arrays_info[vtx.location];
if (array_info.type() == rsx::vertex_base_type::s1 ||
array_info.type() == rsx::vertex_base_type::cmp)
{
//Some vendors do not support GL_x_SNORM buffer textures
verify(HERE), vtx.flags == 0;
vtx.flags |= GL_VP_FORCE_ATTRIB_SCALING | GL_VP_ATTRIB_S16_INT;
}
}
vertex_program.skip_vertex_input_check = true; //not needed for us since decoding is done server side
void* pipeline_properties = nullptr;
m_program = &m_prog_buffer.getGraphicPipelineState(vertex_program, fragment_program, pipeline_properties);
m_program->use();
if (m_prog_buffer.check_cache_missed())
m_shaders_cache->store(pipeline_properties, vertex_program, fragment_program);
u8 *buf;
u32 vertex_state_offset;
u32 vertex_constants_offset;
u32 fragment_constants_offset;
const u32 fragment_constants_size = (const u32)m_prog_buffer.get_fragment_constants_buffer_size(fragment_program);
const u32 fragment_buffer_size = fragment_constants_size + (17 * 4 * sizeof(float));
if (manually_flush_ring_buffers)
{
m_vertex_state_buffer->reserve_storage_on_heap(512);
m_fragment_constants_buffer->reserve_storage_on_heap(align(fragment_buffer_size, 256));
if (m_transform_constants_dirty) m_transform_constants_buffer->reserve_storage_on_heap(8192);
}
// Vertex state
auto mapping = m_vertex_state_buffer->alloc_from_heap(512, m_uniform_buffer_offset_align);
buf = static_cast<u8*>(mapping.first);
vertex_state_offset = mapping.second;
fill_scale_offset_data(buf, false);
fill_user_clip_data(buf + 64);
*(reinterpret_cast<u32*>(buf + 128)) = rsx::method_registers.transform_branch_bits();
*(reinterpret_cast<u32*>(buf + 132)) = vertex_base;
fill_vertex_layout_state(m_vertex_layout, vertex_count, reinterpret_cast<s32*>(buf + 144));
if (m_transform_constants_dirty)
{
// Vertex constants
mapping = m_transform_constants_buffer->alloc_from_heap(8192, m_uniform_buffer_offset_align);
buf = static_cast<u8*>(mapping.first);
vertex_constants_offset = mapping.second;
fill_vertex_program_constants_data(buf);
}
// Fragment constants
mapping = m_fragment_constants_buffer->alloc_from_heap(fragment_buffer_size, m_uniform_buffer_offset_align);
buf = static_cast<u8*>(mapping.first);
fragment_constants_offset = mapping.second;
if (fragment_constants_size)
m_prog_buffer.fill_fragment_constants_buffer({ reinterpret_cast<float*>(buf), gsl::narrow<int>(fragment_constants_size) }, fragment_program);
// Fragment state
fill_fragment_state_buffer(buf+fragment_constants_size, fragment_program);
m_vertex_state_buffer->bind_range(0, vertex_state_offset, 512);
m_fragment_constants_buffer->bind_range(2, fragment_constants_offset, fragment_buffer_size);
if (m_transform_constants_dirty) m_transform_constants_buffer->bind_range(1, vertex_constants_offset, 8192);
if (manually_flush_ring_buffers)
{
m_vertex_state_buffer->unmap();
m_fragment_constants_buffer->unmap();
if (m_transform_constants_dirty) m_transform_constants_buffer->unmap();
}
m_transform_constants_dirty = false;
}
bool GLGSRender::load_program()
{
auto rtt_lookup_func = [this](u32 texaddr, bool is_depth) -> std::tuple<bool, u16>
{
gl::render_target *surface = nullptr;
if (!is_depth)
surface = m_rtts.get_texture_from_render_target_if_applicable(texaddr);
else
surface = m_rtts.get_texture_from_depth_stencil_if_applicable(texaddr);
if (!surface) return std::make_tuple(false, 0);
return std::make_tuple(true, surface->get_native_pitch());
};
RSXVertexProgram vertex_program = get_current_vertex_program();
RSXFragmentProgram fragment_program = get_current_fragment_program(rtt_lookup_func);
std::array<float, 16> rtt_scaling;
u32 unnormalized_rtts = 0;
for (auto &vtx : vertex_program.rsx_vertex_inputs)
{
auto &array_info = rsx::method_registers.vertex_arrays_info[vtx.location];
if (array_info.type() == rsx::vertex_base_type::s1 ||
array_info.type() == rsx::vertex_base_type::cmp)
{
//Some vendors do not support GL_x_SNORM buffer textures
verify(HERE), vtx.flags == 0;
vtx.flags |= GL_VP_FORCE_ATTRIB_SCALING | GL_VP_ATTRIB_S16_INT;
}
}
auto old_program = m_program;
m_program = &m_prog_buffer.getGraphicPipelineState(vertex_program, fragment_program, nullptr);
m_program->use();
if (old_program == m_program && !m_transform_constants_dirty)
{
//This path is taken alot so the savings are tangible
struct scale_offset_layout
{
u16 clip_w, clip_h;
float scale_x, offset_x, scale_y, offset_y, scale_z, offset_z;
float fog0, fog1;
u32 alpha_tested;
float alpha_ref;
}
tmp = {};
tmp.clip_w = rsx::method_registers.surface_clip_width();
tmp.clip_h = rsx::method_registers.surface_clip_height();
tmp.scale_x = rsx::method_registers.viewport_scale_x();
tmp.offset_x = rsx::method_registers.viewport_offset_x();
tmp.scale_y = rsx::method_registers.viewport_scale_y();
tmp.offset_y = rsx::method_registers.viewport_offset_y();
tmp.scale_z = rsx::method_registers.viewport_scale_z();
tmp.offset_z = rsx::method_registers.viewport_offset_z();
tmp.fog0 = rsx::method_registers.fog_params_0();
tmp.fog1 = rsx::method_registers.fog_params_1();
tmp.alpha_tested = rsx::method_registers.alpha_test_enabled();
tmp.alpha_ref = rsx::method_registers.alpha_ref();
size_t old_hash = m_transform_buffer_hash;
m_transform_buffer_hash = 0;
u8 *data = reinterpret_cast<u8*>(&tmp);
for (int i = 0; i < sizeof(tmp); ++i)
m_transform_buffer_hash ^= std::hash<char>()(data[i]);
if (old_hash == m_transform_buffer_hash)
return true;
}
m_transform_constants_dirty = false;
u32 fragment_constants_size = m_prog_buffer.get_fragment_constants_buffer_size(fragment_program);
u32 fragment_buffer_size = fragment_constants_size + (17 * 4 * sizeof(float));
u32 max_buffer_sz = 512 + 8192 + align(fragment_constants_size, m_uniform_buffer_offset_align);
if (manually_flush_ring_buffers)
m_uniform_ring_buffer->reserve_storage_on_heap(align(max_buffer_sz, 512));
u8 *buf;
u32 scale_offset_offset;
u32 vertex_constants_offset;
u32 fragment_constants_offset;
// Scale offset
auto mapping = m_uniform_ring_buffer->alloc_from_heap(512, m_uniform_buffer_offset_align);
buf = static_cast<u8*>(mapping.first);
scale_offset_offset = mapping.second;
fill_scale_offset_data(buf, false);
// Vertex constants
mapping = m_uniform_ring_buffer->alloc_from_heap(8192, m_uniform_buffer_offset_align);
buf = static_cast<u8*>(mapping.first);
vertex_constants_offset = mapping.second;
fill_vertex_program_constants_data(buf);
// Fragment constants
mapping = m_uniform_ring_buffer->alloc_from_heap(fragment_buffer_size, m_uniform_buffer_offset_align);
buf = static_cast<u8*>(mapping.first);
fragment_constants_offset = mapping.second;
if (fragment_constants_size)
m_prog_buffer.fill_fragment_constants_buffer({ reinterpret_cast<float*>(buf), gsl::narrow<int>(fragment_constants_size) }, fragment_program);
// Fragment state
fill_fragment_state_buffer(buf+fragment_constants_size, fragment_program);
m_uniform_ring_buffer->bind_range(0, scale_offset_offset, 512);
m_uniform_ring_buffer->bind_range(1, vertex_constants_offset, 8192);
m_uniform_ring_buffer->bind_range(2, fragment_constants_offset, fragment_buffer_size);
if (manually_flush_ring_buffers)
m_uniform_ring_buffer->unmap();
return true;
}
