dolphin/Source/Core/VideoCommon/VertexLoaderManager.cpp

282 lines
6.3 KiB
C++
Raw Normal View History

// Copyright 2013 Dolphin Emulator Project
// Licensed under GPLv2
// Refer to the license.txt file included.
#include <algorithm>
#include <memory>
#include <unordered_map>
#include <utility>
#include <vector>
#include "Core/HW/Memmap.h"
#include "VideoCommon/BPMemory.h"
#include "VideoCommon/IndexGenerator.h"
#include "VideoCommon/Statistics.h"
#include "VideoCommon/VertexLoader.h"
#include "VideoCommon/VertexLoaderManager.h"
#include "VideoCommon/VertexManagerBase.h"
#include "VideoCommon/VertexShaderManager.h"
#include "VideoCommon/VideoCommon.h"
static int s_attr_dirty; // bitfield
static NativeVertexFormat* s_current_vtx_fmt;
typedef std::pair<VertexLoader*, NativeVertexFormat*> VertexLoaderCacheItem;
static VertexLoaderCacheItem s_VertexLoaders[8];
namespace std
{
template <>
struct hash<VertexLoaderUID>
{
size_t operator()(const VertexLoaderUID& uid) const
{
return uid.GetHash();
}
};
}
typedef std::unordered_map<VertexLoaderUID, VertexLoaderCacheItem> VertexLoaderMap;
typedef std::map<PortableVertexDeclaration, std::unique_ptr<NativeVertexFormat>> NativeVertexLoaderMap;
namespace VertexLoaderManager
{
static VertexLoaderMap s_VertexLoaderMap;
static NativeVertexLoaderMap s_native_vertex_map;
// TODO - change into array of pointers. Keep a map of all seen so far.
void Init()
{
MarkAllDirty();
for (auto& map_entry : s_VertexLoaders)
{
map_entry.first = nullptr;
map_entry.second = nullptr;
}
RecomputeCachedArraybases();
}
void Shutdown()
{
for (auto& map_entry : s_VertexLoaderMap)
{
delete map_entry.second.first;
}
s_VertexLoaderMap.clear();
s_native_vertex_map.clear();
}
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::vector<entry> entries;
size_t total_size = 0;
for (const auto& map_entry : s_VertexLoaderMap)
{
entry e;
map_entry.second.first->AppendToString(&e.text);
e.num_verts = map_entry.second.first->GetNumLoadedVerts();
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->append(entry.text);
}
}
void MarkAllDirty()
{
s_attr_dirty = 0xff;
}
static NativeVertexFormat* GetNativeVertexFormat(const PortableVertexDeclaration& format,
u32 components)
{
auto& native = s_native_vertex_map[format];
if (!native)
{
auto raw_pointer = g_vertex_manager->CreateNativeVertexFormat();
native = std::unique_ptr<NativeVertexFormat>(raw_pointer);
native->Initialize(format);
native->m_components = components;
}
return native.get();
}
static VertexLoaderCacheItem RefreshLoader(int vtx_attr_group)
{
if ((s_attr_dirty >> vtx_attr_group) & 1)
{
VertexLoaderUID uid(g_VtxDesc, g_VtxAttr[vtx_attr_group]);
VertexLoaderMap::iterator iter = s_VertexLoaderMap.find(uid);
if (iter != s_VertexLoaderMap.end())
{
s_VertexLoaders[vtx_attr_group] = iter->second;
}
else
{
VertexLoader* loader = new VertexLoader(g_VtxDesc, g_VtxAttr[vtx_attr_group]);
NativeVertexFormat* vtx_fmt = GetNativeVertexFormat(
loader->GetNativeVertexDeclaration(),
loader->GetNativeComponents());
s_VertexLoaderMap[uid] = std::make_pair(loader, vtx_fmt);
s_VertexLoaders[vtx_attr_group] = std::make_pair(loader, vtx_fmt);
INCSTAT(stats.numVertexLoaders);
}
}
s_attr_dirty &= ~(1 << vtx_attr_group);
return s_VertexLoaders[vtx_attr_group];
}
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.
2014-08-31 23:11:32 -06:00
bool RunVertices(int vtx_attr_group, int primitive, int count, size_t buf_size)
{
if (!count)
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.
2014-08-31 23:11:32 -06:00
return true;
auto loader = RefreshLoader(vtx_attr_group);
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.
2014-08-31 23:11:32 -06:00
size_t size = count * loader.first->GetVertexSize();
if (buf_size < size)
return false;
if (bpmem.genMode.cullmode == GenMode::CULL_ALL && primitive < 5)
{
// if cull mode is CULL_ALL, ignore triangles and quads
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.
2014-08-31 23:11:32 -06:00
DataSkip((u32)size);
return true;
}
// If the native vertex format changed, force a flush.
if (loader.second != s_current_vtx_fmt)
VertexManager::Flush();
s_current_vtx_fmt = loader.second;
VertexManager::PrepareForAdditionalData(primitive, count,
loader.first->GetNativeVertexDeclaration().stride);
loader.first->RunVertices(g_VtxAttr[vtx_attr_group], primitive, count);
IndexGenerator::AddIndices(primitive, count);
ADDSTAT(stats.thisFrame.numPrims, count);
INCSTAT(stats.thisFrame.numPrimitiveJoins);
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.
2014-08-31 23:11:32 -06:00
return true;
}
int GetVertexSize(int vtx_attr_group)
{
return RefreshLoader(vtx_attr_group).first->GetVertexSize();
}
NativeVertexFormat* GetCurrentVertexFormat()
{
return s_current_vtx_fmt;
}
} // namespace
void LoadCPReg(u32 sub_cmd, u32 value)
{
switch (sub_cmd & 0xF0)
{
case 0x30:
VertexShaderManager::SetTexMatrixChangedA(value);
break;
case 0x40:
VertexShaderManager::SetTexMatrixChangedB(value);
break;
case 0x50:
g_VtxDesc.Hex &= ~0x1FFFF; // keep the Upper bits
g_VtxDesc.Hex |= value;
s_attr_dirty = 0xFF;
break;
case 0x60:
g_VtxDesc.Hex &= 0x1FFFF; // keep the lower 17Bits
g_VtxDesc.Hex |= (u64)value << 17;
s_attr_dirty = 0xFF;
break;
case 0x70:
_assert_((sub_cmd & 0x0F) < 8);
g_VtxAttr[sub_cmd & 7].g0.Hex = value;
s_attr_dirty |= 1 << (sub_cmd & 7);
break;
case 0x80:
_assert_((sub_cmd & 0x0F) < 8);
g_VtxAttr[sub_cmd & 7].g1.Hex = value;
s_attr_dirty |= 1 << (sub_cmd & 7);
break;
case 0x90:
_assert_((sub_cmd & 0x0F) < 8);
g_VtxAttr[sub_cmd & 7].g2.Hex = value;
s_attr_dirty |= 1 << (sub_cmd & 7);
break;
// Pointers to vertex arrays in GC RAM
case 0xA0:
arraybases[sub_cmd & 0xF] = value;
cached_arraybases[sub_cmd & 0xF] = Memory::GetPointer(value);
break;
case 0xB0:
arraystrides[sub_cmd & 0xF] = value & 0xFF;
break;
}
}
void FillCPMemoryArray(u32 *memory)
{
memory[0x30] = MatrixIndexA.Hex;
memory[0x40] = MatrixIndexB.Hex;
memory[0x50] = (u32)g_VtxDesc.Hex;
memory[0x60] = (u32)(g_VtxDesc.Hex >> 17);
for (int i = 0; i < 8; ++i)
{
memory[0x70 + i] = g_VtxAttr[i].g0.Hex;
memory[0x80 + i] = g_VtxAttr[i].g1.Hex;
memory[0x90 + i] = g_VtxAttr[i].g2.Hex;
}
for (int i = 0; i < 16; ++i)
{
memory[0xA0 + i] = arraybases[i];
memory[0xB0 + i] = arraystrides[i];
}
}
void RecomputeCachedArraybases()
{
for (int i = 0; i < 16; i++)
{
cached_arraybases[i] = Memory::GetPointer(arraybases[i]);
}
}