mirror of
https://github.com/citra-emu/citra.git
synced 2024-11-25 17:30:16 +00:00
Made vertex batch vector reserve appropiate space in the Rasterizer
This commit is contained in:
parent
26028969f8
commit
60059377de
@ -44,12 +44,6 @@ static const u32 expand_bits_to_bytes[] = {
|
||||
0xff000000, 0xff0000ff, 0xff00ff00, 0xff00ffff, 0xffff0000, 0xffff00ff, 0xffffff00, 0xffffffff,
|
||||
};
|
||||
|
||||
constexpr u32 VERTEX_CACHE_MAX_SIZE = (256 * 256); // 16-bit indices max
|
||||
constexpr u32 VERTEX_CACHE_MIN_SIZE = 256; // 8-bit indices max
|
||||
|
||||
static bool vertex_cache_ids[VERTEX_CACHE_MAX_SIZE] = {false}; // 64 Kb index cache
|
||||
static std::array<Shader::OutputVertex, VERTEX_CACHE_MAX_SIZE> vertex_cache; // 8MB Cache
|
||||
|
||||
MICROPROFILE_DEFINE(GPU_Drawing, "GPU", "Drawing", MP_RGB(50, 50, 240));
|
||||
|
||||
static void WritePicaReg(u32 id, u32 value, u32 mask) {
|
||||
@ -221,6 +215,30 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
|
||||
const u16* index_address_16 = reinterpret_cast<const u16*>(index_address_8);
|
||||
bool index_u16 = index_info.format != 0;
|
||||
|
||||
constexpr size_t VERTEX_CACHE_MAX_SIZE = (256 * 256); // 16-bit indices max
|
||||
|
||||
static std::array<bool, VERTEX_CACHE_MAX_SIZE> vertex_cache_ids = {}; // 64 Kb index cache
|
||||
static std::array<Shader::OutputVertex, VERTEX_CACHE_MAX_SIZE> vertex_cache; // 8MB Cache
|
||||
if (is_indexed) {
|
||||
u32 min_index;
|
||||
u32 max_index;
|
||||
const size_t last_index = regs.num_vertices;
|
||||
// TODO: is there any easier way to get the min/max indices?
|
||||
// NOTE: also regs.num_vertices is actualy the number of indices
|
||||
// in the element being rendered.
|
||||
if (index_u16) {
|
||||
auto result = std::minmax_element(index_address_16, index_address_16 + last_index);
|
||||
min_index = *result.first;
|
||||
max_index = *result.second;
|
||||
} else {
|
||||
auto result = std::minmax_element(index_address_8, index_address_8 + last_index);
|
||||
min_index = *result.first;
|
||||
max_index = *result.second;
|
||||
}
|
||||
std::fill(vertex_cache_ids.begin() + min_index,
|
||||
vertex_cache_ids.begin() + max_index + 1, false);
|
||||
}
|
||||
|
||||
PrimitiveAssembler<Shader::OutputVertex>& primitive_assembler = g_state.primitive_assembler;
|
||||
|
||||
if (g_debug_context && g_debug_context->recorder) {
|
||||
@ -240,11 +258,6 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
|
||||
DebugUtils::MemoryAccessTracker memory_accesses;
|
||||
Shader::OutputVertex output_vertex;
|
||||
|
||||
if (is_indexed) {
|
||||
const u32 num_indices = index_u16 ? VERTEX_CACHE_MAX_SIZE : VERTEX_CACHE_MIN_SIZE;
|
||||
std::memset(vertex_cache_ids, false, sizeof(bool) * num_indices);
|
||||
}
|
||||
|
||||
Shader::UnitState shader_unit;
|
||||
g_state.vs.Setup();
|
||||
|
||||
@ -276,7 +289,7 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
|
||||
}
|
||||
}
|
||||
|
||||
// if the vertex wasn'tcached, then we pass it to the shader, process
|
||||
// if the vertex wasn't cached, then we pass it to the shader, process
|
||||
// it and cache it.
|
||||
if (!vertex_cache_hit) {
|
||||
// Initialize data for the current vertex
|
||||
|
@ -53,7 +53,10 @@ RasterizerOpenGL::RasterizerOpenGL() : shader_dirty(true) {
|
||||
state.draw.uniform_buffer = uniform_buffer.handle;
|
||||
state.Apply();
|
||||
|
||||
vertex_batch.reserve(256 * 256); // 8MB for worst case scenario
|
||||
// reserve 8MB for worst case scenario.
|
||||
// this will avoid vector resizing as we send vertices.
|
||||
constexpr size_t VERTEX_ARRAY_MAX_SIZE = 256 * 256;
|
||||
vertex_batch.reserve(VERTEX_ARRAY_MAX_SIZE);
|
||||
|
||||
// Bind the UBO to binding point 0
|
||||
glBindBufferBase(GL_UNIFORM_BUFFER, 0, uniform_buffer.handle);
|
||||
|
Loading…
Reference in New Issue
Block a user