dolphin/Source/Core/VideoCommon/VertexLoaderManager.cpp

297 lines
8.0 KiB
C++
Raw Normal View History

// Copyright 2008 Dolphin Emulator Project
2015-05-17 23:08:10 +00:00
// Licensed under GPLv2+
// Refer to the license.txt file included.
#include <algorithm>
#include <memory>
#include <mutex>
#include <unordered_map>
#include <utility>
#include <vector>
#include "Common/CommonFuncs.h"
#include "Core/HW/Memmap.h"
#include "VideoCommon/BPMemory.h"
#include "VideoCommon/IndexGenerator.h"
#include "VideoCommon/Statistics.h"
#include "VideoCommon/VertexLoaderBase.h"
#include "VideoCommon/VertexLoaderManager.h"
#include "VideoCommon/VertexManagerBase.h"
#include "VideoCommon/VertexShaderManager.h"
#include "VideoCommon/VideoCommon.h"
namespace VertexLoaderManager
{
float position_cache[3][4];
u32 position_matrix_index[3];
typedef std::unordered_map<PortableVertexDeclaration, std::unique_ptr<NativeVertexFormat>> NativeVertexFormatMap;
static NativeVertexFormatMap s_native_vertex_map;
static NativeVertexFormat* s_current_vtx_fmt;
typedef std::unordered_map<VertexLoaderUID, std::unique_ptr<VertexLoaderBase>> VertexLoaderMap;
static std::mutex s_vertex_loader_map_lock;
static VertexLoaderMap s_vertex_loader_map;
// TODO - change into array of pointers. Keep a map of all seen so far.
u8 *cached_arraybases[12];
void Init()
{
MarkAllDirty();
for (auto& map_entry : g_main_cp_state.vertex_loaders)
map_entry = nullptr;
for (auto& map_entry : g_preprocess_cp_state.vertex_loaders)
map_entry = nullptr;
SETSTAT(stats.numVertexLoaders, 0);
}
void Shutdown()
{
std::lock_guard<std::mutex> lk(s_vertex_loader_map_lock);
s_vertex_loader_map.clear();
s_native_vertex_map.clear();
}
void UpdateVertexArrayPointers()
{
// Anything to update?
if (!g_main_cp_state.bases_dirty)
return;
// Some games such as Burnout 2 can put invalid addresses into
// the array base registers. (see issue 8591)
// But the vertex arrays with invalid addresses aren't actually enabled.
// Note: Only array bases 0 through 11 are used by the Vertex loaders.
// 12 through 15 are used for loading data into xfmem.
for (int i = 0; i < 12; i++)
{
// Only update the array base if the vertex description states we are going to use it.
if (g_main_cp_state.vtx_desc.GetVertexArrayStatus(i) >= 0x2)
cached_arraybases[i] = Memory::GetPointer(g_main_cp_state.array_bases[i]);
}
g_main_cp_state.bases_dirty = false;
}
namespace
{
struct entry
{
std::string text;
u64 num_verts;
bool operator < (const entry &other) const
{
return num_verts > other.num_verts;
}
};
}
void AppendListToString(std::string *dest)
{
std::lock_guard<std::mutex> lk(s_vertex_loader_map_lock);
std::vector<entry> entries;
size_t total_size = 0;
for (const auto& map_entry : s_vertex_loader_map)
{
entry e;
map_entry.second->AppendToString(&e.text);
e.num_verts = map_entry.second->m_numLoadedVertices;
entries.push_back(e);
total_size += e.text.size() + 1;
}
sort(entries.begin(), entries.end());
dest->reserve(dest->size() + total_size);
for (const entry& entry : entries)
{
*dest += entry.text;
*dest += '\n';
}
}
void MarkAllDirty()
{
g_main_cp_state.attr_dirty = BitSet32::AllTrue(8);
g_preprocess_cp_state.attr_dirty = BitSet32::AllTrue(8);
}
static VertexLoaderBase* RefreshLoader(int vtx_attr_group, bool preprocess = false)
{
CPState* state = preprocess ? &g_preprocess_cp_state : &g_main_cp_state;
VertexLoaderBase* loader;
if (state->attr_dirty[vtx_attr_group])
{
// We are not allowed to create a native vertex format on preprocessing as this is on the wrong thread
bool check_for_native_format = !preprocess;
VertexLoaderUID uid(state->vtx_desc, state->vtx_attr[vtx_attr_group]);
std::lock_guard<std::mutex> lk(s_vertex_loader_map_lock);
VertexLoaderMap::iterator iter = s_vertex_loader_map.find(uid);
if (iter != s_vertex_loader_map.end())
{
loader = iter->second.get();
check_for_native_format &= !loader->m_native_vertex_format;
}
else
{
loader = VertexLoaderBase::CreateVertexLoader(state->vtx_desc, state->vtx_attr[vtx_attr_group]);
s_vertex_loader_map[uid] = std::unique_ptr<VertexLoaderBase>(loader);
INCSTAT(stats.numVertexLoaders);
}
if (check_for_native_format)
{
// search for a cached native vertex format
const PortableVertexDeclaration& format = loader->m_native_vtx_decl;
std::unique_ptr<NativeVertexFormat>& native = s_native_vertex_map[format];
if (!native)
{
native.reset(g_vertex_manager->CreateNativeVertexFormat());
native->Initialize(format);
native->m_components = loader->m_native_components;
}
loader->m_native_vertex_format = native.get();
}
state->vertex_loaders[vtx_attr_group] = loader;
state->attr_dirty[vtx_attr_group] = false;
} else {
loader = state->vertex_loaders[vtx_attr_group];
}
// Lookup pointers for any vertex arrays.
if (!preprocess)
UpdateVertexArrayPointers();
return loader;
}
int RunVertices(int vtx_attr_group, int primitive, int count, DataReader src, bool skip_drawing, bool is_preprocess)
{
if (!count)
2014-12-09 07:35:04 +00:00
return 0;
VertexLoaderBase* loader = RefreshLoader(vtx_attr_group, is_preprocess);
int size = count * loader->m_VertexSize;
2014-12-09 07:35:04 +00:00
if ((int)src.size() < size)
return -1;
Refactor opcode decoding a bit to kill FifoCommandRunnable. Separated out from my gpu-determinism branch by request. It's not a big commit; I just like to write long commit messages. The main reason to kill it is hopefully a slight performance improvement from avoiding the double switch (especially in single core mode); however, this also improves cycle calculation, as described below. - FifoCommandRunnable is removed; in its stead, Decode returns the number of cycles (which only matters for "sync" GPU mode), or 0 if there was not enough data, and is also responsible for unknown opcode alerts. Decode and DecodeSemiNop are almost identical, so the latter is replaced with a skipped_frame parameter to Decode. Doesn't mean we can't improve skipped_frame mode to do less work; if, at such a point, branching on it has too much overhead (it certainly won't now), it can always be changed to a template parameter. - FifoCommandRunnable used a fixed, large cycle count for display lists, regardless of the contents. Presumably the actual hardware's processing time is mostly the processing time of whatever commands are in the list, and with this change InterpretDisplayList can just return the list's cycle count to be added to the total. (Since the calculation for this is part of Decode, it didn't seem easy to split this change up.) To facilitate this, Decode also gains an explicit 'end' parameter in lieu of FifoCommandRunnable's call to GetVideoBufferEndPtr, which can point to there or to the end of a display list (or elsewhere in gpu-determinism, but that's another story). Also, as a small optimization, InterpretDisplayList now calls OpcodeDecoder_Run rather than having its own Decode loop, to allow Decode to be inlined (haven't checked whether this actually happens though). skipped_frame mode still does not traverse display lists and uses the old fake value of 45 cycles. degasus has suggested that this hack is not essential for performance and can be removed, but I want to separate any potential performance impact of that from this commit.
2014-09-01 05:11:32 +00:00
if (skip_drawing || is_preprocess)
2014-12-09 07:35:04 +00:00
return size;
// If the native vertex format changed, force a flush.
if (loader->m_native_vertex_format != s_current_vtx_fmt)
VertexManager::Flush();
s_current_vtx_fmt = loader->m_native_vertex_format;
// 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);
2014-12-09 07:35:04 +00:00
DataReader dst = VertexManager::PrepareForAdditionalData(primitive, count,
loader->m_native_vtx_decl.stride, cullall);
count = loader->RunVertices(src, dst, count);
IndexGenerator::AddIndices(primitive, count);
VertexManager::FlushData(count, loader->m_native_vtx_decl.stride);
2014-12-09 07:35:04 +00:00
ADDSTAT(stats.thisFrame.numPrims, count);
INCSTAT(stats.thisFrame.numPrimitiveJoins);
2014-12-09 07:35:04 +00:00
return size;
}
NativeVertexFormat* GetCurrentVertexFormat()
{
return s_current_vtx_fmt;
}
} // namespace
void LoadCPReg(u32 sub_cmd, u32 value, bool is_preprocess)
{
bool update_global_state = !is_preprocess;
CPState* state = is_preprocess ? &g_preprocess_cp_state : &g_main_cp_state;
switch (sub_cmd & 0xF0)
{
case 0x30:
if (update_global_state)
VertexShaderManager::SetTexMatrixChangedA(value);
break;
case 0x40:
if (update_global_state)
VertexShaderManager::SetTexMatrixChangedB(value);
break;
case 0x50:
state->vtx_desc.Hex &= ~0x1FFFF; // keep the Upper bits
state->vtx_desc.Hex |= value;
state->attr_dirty = BitSet32::AllTrue(8);
state->bases_dirty = true;
break;
case 0x60:
state->vtx_desc.Hex &= 0x1FFFF; // keep the lower 17Bits
state->vtx_desc.Hex |= (u64)value << 17;
state->attr_dirty = BitSet32::AllTrue(8);
state->bases_dirty = true;
break;
case 0x70:
_assert_((sub_cmd & 0x0F) < 8);
state->vtx_attr[sub_cmd & 7].g0.Hex = value;
state->attr_dirty[sub_cmd & 7] = true;
break;
case 0x80:
_assert_((sub_cmd & 0x0F) < 8);
state->vtx_attr[sub_cmd & 7].g1.Hex = value;
state->attr_dirty[sub_cmd & 7] = true;
break;
case 0x90:
_assert_((sub_cmd & 0x0F) < 8);
state->vtx_attr[sub_cmd & 7].g2.Hex = value;
state->attr_dirty[sub_cmd & 7] = true;
break;
// Pointers to vertex arrays in GC RAM
case 0xA0:
state->array_bases[sub_cmd & 0xF] = value;
state->bases_dirty = true;
break;
case 0xB0:
state->array_strides[sub_cmd & 0xF] = value & 0xFF;
break;
}
}
void FillCPMemoryArray(u32 *memory)
{
memory[0x30] = g_main_cp_state.matrix_index_a.Hex;
memory[0x40] = g_main_cp_state.matrix_index_b.Hex;
memory[0x50] = (u32)g_main_cp_state.vtx_desc.Hex;
memory[0x60] = (u32)(g_main_cp_state.vtx_desc.Hex >> 17);
for (int i = 0; i < 8; ++i)
{
memory[0x70 + i] = g_main_cp_state.vtx_attr[i].g0.Hex;
memory[0x80 + i] = g_main_cp_state.vtx_attr[i].g1.Hex;
memory[0x90 + i] = g_main_cp_state.vtx_attr[i].g2.Hex;
}
for (int i = 0; i < 16; ++i)
{
memory[0xA0 + i] = g_main_cp_state.array_bases[i];
memory[0xB0 + i] = g_main_cp_state.array_strides[i];
}
}