The-Homebrew-Cloud/Lime3DS
2
0
Fork 0
mirror of https://github.com/Lime3DS/Lime3DS.git synced 2024-11-14 05:55:13 +01:00
Code Issues Packages Projects Releases Wiki Activity

Videocore: Replace std::stack in shader interpreter with static_vector

Browse Source 
Shaves off 1/3rd of the vertex shader time in Fire Emblem
This commit is contained in:
Yuri Kunde Schlesner 2015-07-26 07:27:36 -03:00
parent 392c7feba0
commit 877d2a0e48
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
src/video_core/vertex_shader.cpp
View File

@ -2,8 +2,7 @@
// Licensed under GPLv2 or any later version // Licensed under GPLv2 or any later version
// Refer to the license.txt file included. // Refer to the license.txt file included.
#include <stack> #include <boost/container/static_vector.hpp>
#include <boost/range/algorithm.hpp> #include <boost/range/algorithm.hpp>
#include <common/file_util.h> #include <common/file_util.h>
@ -53,7 +52,7 @@ struct VertexShaderState {
}; };
// TODO: Is there a maximal size for this? // TODO: Is there a maximal size for this?
std::stack<CallStackElement> call_stack; boost::container::static_vector<CallStackElement, 16> call_stack;
struct { struct {
u32 max_offset; // maximum program counter ever reached u32 max_offset; // maximum program counter ever reached
@ -71,13 +70,13 @@ static void ProcessShaderCode(VertexShaderState& state) {
while (true) { while (true) {
if (!state.call_stack.empty()) { if (!state.call_stack.empty()) {
auto& top = state.call_stack.top(); auto& top = state.call_stack.back();
if (state.program_counter - program_code.data() == top.final_address) { if (state.program_counter - program_code.data() == top.final_address) {
state.address_registers[2] += top.loop_increment; state.address_registers[2] += top.loop_increment;
if (top.repeat_counter-- == 0) { if (top.repeat_counter-- == 0) {
state.program_counter = &program_code[top.return_address]; state.program_counter = &program_code[top.return_address];
state.call_stack.pop(); state.call_stack.pop_back();
} else { } else {
state.program_counter = &program_code[top.loop_address]; state.program_counter = &program_code[top.loop_address];
} }
@ -94,7 +93,8 @@ static void ProcessShaderCode(VertexShaderState& state) {
static auto call = [&program_code](VertexShaderState& state, u32 offset, u32 num_instructions, static auto call = [&program_code](VertexShaderState& state, u32 offset, u32 num_instructions,
u32 return_offset, u8 repeat_count, u8 loop_increment) { u32 return_offset, u8 repeat_count, u8 loop_increment) {
state.program_counter = &program_code[offset] - 1; // -1 to make sure when incrementing the PC we end up at the correct offset state.program_counter = &program_code[offset] - 1; // -1 to make sure when incrementing the PC we end up at the correct offset
state.call_stack.push({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset }); ASSERT(state.call_stack.size() < state.call_stack.capacity());
state.call_stack.push_back({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset });
}; };
u32 binary_offset = state.program_counter - program_code.data(); u32 binary_offset = state.program_counter - program_code.data();