CCF_100/yuzu-mirror
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

MacroHLE: Reduce massive calculations on sizing estimation.

Browse Source
This commit is contained in:
Fernando Sahmkow
2022-11-17 16:36:53 +01:00
parent aad0cbf024
commit 18637766ef
9 changed files with 238 additions and 95 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
src/video_core/dma_pusher.cpp
View File

@@ -75,7 +75,8 @@ bool DmaPusher::Step() {
// Push buffer non-empty, read a word
command_headers.resize_destructive(command_list_header.size);
if (Settings::IsGPULevelExtreme()) {
constexpr u32 MacroRegistersStart = 0xE00;
if (dma_state.method < MacroRegistersStart) {
memory_manager.ReadBlock(dma_state.dma_get, command_headers.data(),
command_list_header.size * sizeof(u32));
} else {

15
src/video_core/engines/maxwell_3d.cpp
View File

@@ -157,6 +157,21 @@ void Maxwell3D::RefreshParameters() {
}
}
bool Maxwell3D::AnyParametersDirty() {
size_t current_index = 0;
for (auto& segment : macro_segments) {
if (segment.first == 0) {
current_index += segment.second;
continue;
}
if (memory_manager.IsMemoryDirty(segment.first, sizeof(u32) * segment.second)) {
return true;
}
current_index += segment.second;
}
return false;
}
u32 Maxwell3D::GetMaxCurrentVertices() {
u32 num_vertices = 0;
for (size_t index = 0; index < Regs::NumVertexArrays; ++index) {

2
src/video_core/engines/maxwell_3d.h
View File

@@ -3092,6 +3092,8 @@ public:
void RefreshParameters();
bool AnyParametersDirty();
u32 GetMaxCurrentVertices();
size_t EstimateIndexBufferSize();

91
src/video_core/memory_manager.cpp
View File

@@ -25,7 +25,8 @@ MemoryManager::MemoryManager(Core::System& system_, u64 address_space_bits_, u64
address_space_bits{address_space_bits_}, page_bits{page_bits_}, big_page_bits{big_page_bits_},
entries{}, big_entries{}, page_table{address_space_bits, address_space_bits + page_bits - 38,
page_bits != big_page_bits ? page_bits : 0},
unique_identifier{unique_identifier_generator.fetch_add(1, std::memory_order_acq_rel)} {
kind_map{PTEKind::INVALID}, unique_identifier{unique_identifier_generator.fetch_add(
1, std::memory_order_acq_rel)} {
address_space_size = 1ULL << address_space_bits;
page_size = 1ULL << page_bits;
page_mask = page_size - 1ULL;
@@ -41,11 +42,7 @@ MemoryManager::MemoryManager(Core::System& system_, u64 address_space_bits_, u64
big_entries.resize(big_page_table_size / 32, 0);
big_page_table_cpu.resize(big_page_table_size);
big_page_continous.resize(big_page_table_size / continous_bits, 0);
std::array<PTEKind, 32> kind_valus;
kind_valus.fill(PTEKind::INVALID);
big_kinds.resize(big_page_table_size / 32, kind_valus);
entries.resize(page_table_size / 32, 0);
kinds.resize(page_table_size / 32, kind_valus);
}
MemoryManager::~MemoryManager() = default;
@@ -83,38 +80,7 @@ void MemoryManager::SetEntry(size_t position, MemoryManager::EntryType entry) {
}
PTEKind MemoryManager::GetPageKind(GPUVAddr gpu_addr) const {
auto entry = GetEntry<true>(gpu_addr);
if (entry == EntryType::Mapped || entry == EntryType::Reserved) [[likely]] {
return GetKind<true>(gpu_addr);
} else {
return GetKind<false>(gpu_addr);
}
}
template <bool is_big_page>
PTEKind MemoryManager::GetKind(size_t position) const {
if constexpr (is_big_page) {
position = position >> big_page_bits;
const size_t sub_index = position % 32;
return big_kinds[position / 32][sub_index];
} else {
position = position >> page_bits;
const size_t sub_index = position % 32;
return kinds[position / 32][sub_index];
}
}
template <bool is_big_page>
void MemoryManager::SetKind(size_t position, PTEKind kind) {
if constexpr (is_big_page) {
position = position >> big_page_bits;
const size_t sub_index = position % 32;
big_kinds[position / 32][sub_index] = kind;
} else {
position = position >> page_bits;
const size_t sub_index = position % 32;
kinds[position / 32][sub_index] = kind;
}
return kind_map.GetValueAt(gpu_addr);
}
inline bool MemoryManager::IsBigPageContinous(size_t big_page_index) const {
@@ -141,7 +107,6 @@ GPUVAddr MemoryManager::PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cp
const GPUVAddr current_gpu_addr = gpu_addr + offset;
[[maybe_unused]] const auto current_entry_type = GetEntry<false>(current_gpu_addr);
SetEntry<false>(current_gpu_addr, entry_type);
SetKind<false>(current_gpu_addr, kind);
if (current_entry_type != entry_type) {
rasterizer->ModifyGPUMemory(unique_identifier, gpu_addr, page_size);
}
@@ -153,6 +118,7 @@ GPUVAddr MemoryManager::PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cp
}
remaining_size -= page_size;
}
kind_map.Map(gpu_addr, gpu_addr + size, kind);
return gpu_addr;
}
@@ -164,7 +130,6 @@ GPUVAddr MemoryManager::BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr
const GPUVAddr current_gpu_addr = gpu_addr + offset;
[[maybe_unused]] const auto current_entry_type = GetEntry<true>(current_gpu_addr);
SetEntry<true>(current_gpu_addr, entry_type);
SetKind<true>(current_gpu_addr, kind);
if (current_entry_type != entry_type) {
rasterizer->ModifyGPUMemory(unique_identifier, gpu_addr, big_page_size);
}
@@ -193,6 +158,7 @@ GPUVAddr MemoryManager::BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr
}
remaining_size -= big_page_size;
}
kind_map.Map(gpu_addr, gpu_addr + size, kind);
return gpu_addr;
}
@@ -578,52 +544,7 @@ size_t MemoryManager::MaxContinousRange(GPUVAddr gpu_addr, size_t size) const {
}
size_t MemoryManager::GetMemoryLayoutSize(GPUVAddr gpu_addr, size_t max_size) const {
PTEKind base_kind = GetPageKind(gpu_addr);
if (base_kind == PTEKind::INVALID) {
return 0;
}
size_t range_so_far = 0;
bool result{false};
auto fail = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset,
std::size_t copy_amount) {
result = true;
return true;
};
auto short_check = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
PTEKind base_kind_other = GetKind<false>((page_index << page_bits) + offset);
if (base_kind != base_kind_other) {
result = true;
return true;
}
range_so_far += copy_amount;
if (range_so_far >= max_size) {
result = true;
return true;
}
return false;
};
auto big_check = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
PTEKind base_kind_other = GetKind<true>((page_index << big_page_bits) + offset);
if (base_kind != base_kind_other) {
result = true;
return true;
}
range_so_far += copy_amount;
if (range_so_far >= max_size) {
result = true;
return true;
}
return false;
};
auto check_short_pages = [&](std::size_t page_index, std::size_t offset,
std::size_t copy_amount) {
GPUVAddr base = (page_index << big_page_bits) + offset;
MemoryOperation<false>(base, copy_amount, short_check, fail, fail);
return result;
};
MemoryOperation<true>(gpu_addr, address_space_size - gpu_addr, big_check, fail,
check_short_pages);
return range_so_far;
return kind_map.GetContinousSizeFrom(gpu_addr);
}
void MemoryManager::InvalidateRegion(GPUVAddr gpu_addr, size_t size) const {

11
src/video_core/memory_manager.h
View File

@@ -10,6 +10,7 @@
#include "common/common_types.h"
#include "common/multi_level_page_table.h"
#include "common/range_map.h"
#include "common/virtual_buffer.h"
#include "video_core/pte_kind.h"
@@ -186,16 +187,8 @@ private:
template <bool is_big_page>
inline void SetEntry(size_t position, EntryType entry);
std::vector<std::array<PTEKind, 32>> kinds;
std::vector<std::array<PTEKind, 32>> big_kinds;
template <bool is_big_page>
inline PTEKind GetKind(size_t position) const;
template <bool is_big_page>
inline void SetKind(size_t position, PTEKind kind);
Common::MultiLevelPageTable<u32> page_table;
Common::RangeMap<GPUVAddr, PTEKind> kind_map;
Common::VirtualBuffer<u32> big_page_table_cpu;
std::vector<u64> big_page_continous;