ppsspp/GPU/D3D11/DrawEngineD3D11.cpp

970 lines
32 KiB
C++
Raw Normal View History

// Copyright (c) 2012- PPSSPP Project.
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 2.0 or later versions.
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License 2.0 for more details.
// A copy of the GPL 2.0 should have been included with the program.
// If not, see http://www.gnu.org/licenses/
// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
#include "base/logging.h"
#include "base/timeutil.h"
#include "Common/MemoryUtil.h"
#include "Core/MemMap.h"
#include "Core/Host.h"
#include "Core/System.h"
#include "Core/Reporting.h"
#include "Core/Config.h"
#include "Core/CoreTiming.h"
#include "GPU/Math3D.h"
#include "GPU/GPUState.h"
#include "GPU/ge_constants.h"
#include "GPU/Common/TextureDecoder.h"
#include "GPU/Common/SplineCommon.h"
#include "GPU/Common/TransformCommon.h"
#include "GPU/Common/VertexDecoderCommon.h"
#include "GPU/Common/SoftwareTransformCommon.h"
#include "GPU/D3D11/TextureCacheD3D11.h"
#include "GPU/D3D11/DrawEngineD3D11.h"
#include "GPU/D3D11/ShaderManagerD3D11.h"
#include "GPU/D3D11/GPU_D3D11.h"
const D3D11_PRIMITIVE_TOPOLOGY glprim[8] = {
D3D11_PRIMITIVE_TOPOLOGY_POINTLIST,
D3D11_PRIMITIVE_TOPOLOGY_LINELIST,
D3D11_PRIMITIVE_TOPOLOGY_LINESTRIP,
D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST,
D3D11_PRIMITIVE_TOPOLOGY_TRIANGLESTRIP,
D3D11_PRIMITIVE_TOPOLOGY_UNDEFINED, // Fans not supported
D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST, // Need expansion - though we could do it with geom shaders in most cases
};
enum {
TRANSFORMED_VERTEX_BUFFER_SIZE = VERTEX_BUFFER_MAX * sizeof(TransformedVertex)
};
#define VERTEXCACHE_DECIMATION_INTERVAL 17
enum { VAI_KILL_AGE = 120, VAI_UNRELIABLE_KILL_AGE = 240, VAI_UNRELIABLE_KILL_MAX = 4 };
enum {
VERTEX_PUSH_SIZE = 1024 * 1024 * 16,
INDEX_PUSH_SIZE = 1024 * 1024 * 4,
};
static const D3D11_INPUT_ELEMENT_DESC TransformedVertexElements[] = {
{ "POSITION", 0, DXGI_FORMAT_R32G32B32A32_FLOAT, 0, 0, D3D11_INPUT_PER_VERTEX_DATA, 0 },
{ "TEXCOORD", 0, DXGI_FORMAT_R32G32B32_FLOAT, 0, 16, D3D11_INPUT_PER_VERTEX_DATA, 0 },
{ "COLOR", 0, DXGI_FORMAT_R8G8B8A8_UNORM, 0, 28, D3D11_INPUT_PER_VERTEX_DATA, 0 },
{ "COLOR", 1, DXGI_FORMAT_R8G8B8A8_UNORM, 0, 32, D3D11_INPUT_PER_VERTEX_DATA, 0 },
};
2017-02-10 09:26:18 +00:00
DrawEngineD3D11::DrawEngineD3D11(Draw::DrawContext *draw, ID3D11Device *device, ID3D11DeviceContext *context)
: draw_(draw),
device_(device),
context_(context),
decodedVerts_(0),
prevPrim_(GE_PRIM_INVALID),
lastVType_(-1),
shaderManager_(0),
textureCache_(0),
framebufferManager_(0),
numDrawCalls(0),
vertexCountInDrawCalls(0),
decodeCounter_(0),
dcid_(0),
fboTexNeedBind_(false),
fboTexBound_(false) {
decOptions_.expandAllWeightsToFloat = true;
decOptions_.expand8BitNormalsToFloat = true;
decimationCounter_ = VERTEXCACHE_DECIMATION_INTERVAL;
// Allocate nicely aligned memory. Maybe graphics drivers will
// appreciate it.
// All this is a LOT of memory, need to see if we can cut down somehow.
decoded = (u8 *)AllocateMemoryPages(DECODED_VERTEX_BUFFER_SIZE, MEM_PROT_READ | MEM_PROT_WRITE);
decIndex = (u16 *)AllocateMemoryPages(DECODED_INDEX_BUFFER_SIZE, MEM_PROT_READ | MEM_PROT_WRITE);
splineBuffer = (u8 *)AllocateMemoryPages(SPLINE_BUFFER_SIZE, MEM_PROT_READ | MEM_PROT_WRITE);
transformed = (TransformedVertex *)AllocateMemoryPages(TRANSFORMED_VERTEX_BUFFER_SIZE, MEM_PROT_READ | MEM_PROT_WRITE);
transformedExpanded = (TransformedVertex *)AllocateMemoryPages(3 * TRANSFORMED_VERTEX_BUFFER_SIZE, MEM_PROT_READ | MEM_PROT_WRITE);
indexGen.Setup(decIndex);
InitDeviceObjects();
tessDataTransfer = new TessellationDataTransferD3D11();
// Vertex pushing buffers.
pushVerts_ = new PushBufferD3D11(device, VERTEX_PUSH_SIZE, D3D11_BIND_VERTEX_BUFFER);
pushInds_ = new PushBufferD3D11(device, INDEX_PUSH_SIZE, D3D11_BIND_INDEX_BUFFER);
}
DrawEngineD3D11::~DrawEngineD3D11() {
delete pushVerts_;
delete pushInds_;
2017-02-10 09:26:18 +00:00
for (auto decl = inputLayoutMap_.begin(); decl != inputLayoutMap_.end(); ++decl) {
if (decl->second) {
decl->second->Release();
}
}
for (auto &depth : depthStencilCache_) {
depth.second->Release();
}
for (auto &blend : blendCache_) {
blend.second->Release();
}
DestroyDeviceObjects();
FreeMemoryPages(decoded, DECODED_VERTEX_BUFFER_SIZE);
FreeMemoryPages(decIndex, DECODED_INDEX_BUFFER_SIZE);
FreeMemoryPages(splineBuffer, SPLINE_BUFFER_SIZE);
FreeMemoryPages(transformed, TRANSFORMED_VERTEX_BUFFER_SIZE);
FreeMemoryPages(transformedExpanded, 3 * TRANSFORMED_VERTEX_BUFFER_SIZE);
delete tessDataTransfer;
}
void DrawEngineD3D11::InitDeviceObjects() {
}
void DrawEngineD3D11::DestroyDeviceObjects() {
ClearTrackedVertexArrays();
}
struct DeclTypeInfo {
DXGI_FORMAT type;
const char * name;
};
static const DeclTypeInfo VComp[] = {
{ DXGI_FORMAT_UNKNOWN, "NULL" }, // DEC_NONE,
{ DXGI_FORMAT_R32_FLOAT, "D3DDECLTYPE_FLOAT1 " }, // DEC_FLOAT_1,
{ DXGI_FORMAT_R32G32_FLOAT, "D3DDECLTYPE_FLOAT2 " }, // DEC_FLOAT_2,
{ DXGI_FORMAT_R32G32B32_FLOAT, "D3DDECLTYPE_FLOAT3 " }, // DEC_FLOAT_3,
{ DXGI_FORMAT_R32G32B32A32_FLOAT, "D3DDECLTYPE_FLOAT4 " }, // DEC_FLOAT_4,
{ DXGI_FORMAT_R8G8B8A8_SNORM, "UNUSED" }, // DEC_S8_3,
{ DXGI_FORMAT_R16G16B16A16_SNORM, "D3DDECLTYPE_SHORT4N " }, // DEC_S16_3,
{ DXGI_FORMAT_R8G8B8A8_UNORM, "D3DDECLTYPE_UBYTE4N " }, // DEC_U8_1,
{ DXGI_FORMAT_R8G8B8A8_UNORM, "D3DDECLTYPE_UBYTE4N " }, // DEC_U8_2,
{ DXGI_FORMAT_R8G8B8A8_UNORM, "D3DDECLTYPE_UBYTE4N " }, // DEC_U8_3,
{ DXGI_FORMAT_R8G8B8A8_UNORM, "D3DDECLTYPE_UBYTE4N " }, // DEC_U8_4,
{ DXGI_FORMAT_UNKNOWN, "UNUSED_DEC_U16_1" }, // DEC_U16_1,
{ DXGI_FORMAT_UNKNOWN, "UNUSED_DEC_U16_2" }, // DEC_U16_2,
{ DXGI_FORMAT_R16G16B16A16_UNORM ,"D3DDECLTYPE_USHORT4N "}, // DEC_U16_3,
{ DXGI_FORMAT_R16G16B16A16_UNORM ,"D3DDECLTYPE_USHORT4N "}, // DEC_U16_4,
{ DXGI_FORMAT_UNKNOWN, "UNUSED_DEC_U8A_2"}, // DEC_U8A_2,
{ DXGI_FORMAT_UNKNOWN, "UNUSED_DEC_U16A_2" }, // DEC_U16A_2,
};
static void VertexAttribSetup(D3D11_INPUT_ELEMENT_DESC * VertexElement, u8 fmt, u8 offset, const char *semantic, u8 semantic_index = 0) {
memset(VertexElement, 0, sizeof(D3D11_INPUT_ELEMENT_DESC));
VertexElement->AlignedByteOffset = offset;
VertexElement->Format = VComp[fmt].type;
VertexElement->SemanticName = semantic;
VertexElement->SemanticIndex = semantic_index;
}
ID3D11InputLayout *DrawEngineD3D11::SetupDecFmtForDraw(D3D11VertexShader *vshader, const DecVtxFormat &decFmt, u32 pspFmt) {
2017-02-10 09:26:18 +00:00
// TODO: Instead of one for each vshader, we can reduce it to one for each type of shader
// that reads TEXCOORD or not, etc. Not sure if worth it.
InputLayoutKey key{ pspFmt, vshader };
auto vertexDeclCached = inputLayoutMap_.find(key);
if (vertexDeclCached == inputLayoutMap_.end()) {
D3D11_INPUT_ELEMENT_DESC VertexElements[8];
D3D11_INPUT_ELEMENT_DESC *VertexElement = &VertexElements[0];
// Vertices Elements orders
// WEIGHT
if (decFmt.w0fmt != 0) {
VertexAttribSetup(VertexElement, decFmt.w0fmt, decFmt.w0off, "TEXCOORD", 1);
VertexElement++;
}
if (decFmt.w1fmt != 0) {
VertexAttribSetup(VertexElement, decFmt.w1fmt, decFmt.w1off, "TEXCOORD", 2);
VertexElement++;
}
// TC
if (decFmt.uvfmt != 0) {
VertexAttribSetup(VertexElement, decFmt.uvfmt, decFmt.uvoff, "TEXCOORD", 0);
VertexElement++;
}
// COLOR
if (decFmt.c0fmt != 0) {
VertexAttribSetup(VertexElement, decFmt.c0fmt, decFmt.c0off, "COLOR", 0);
VertexElement++;
}
// Never used ?
if (decFmt.c1fmt != 0) {
VertexAttribSetup(VertexElement, decFmt.c1fmt, decFmt.c1off, "COLOR", 1);
VertexElement++;
}
// NORMAL
if (decFmt.nrmfmt != 0) {
VertexAttribSetup(VertexElement, decFmt.nrmfmt, decFmt.nrmoff, "NORMAL", 0);
VertexElement++;
}
// POSITION
// Always
VertexAttribSetup(VertexElement, decFmt.posfmt, decFmt.posoff, "POSITION", 0);
VertexElement++;
// Create declaration
ID3D11InputLayout *inputLayout = nullptr;
HRESULT hr = device_->CreateInputLayout(VertexElements, VertexElement - VertexElements, vshader->bytecode().data(), vshader->bytecode().size(), &inputLayout);
if (FAILED(hr)) {
ERROR_LOG(G3D, "Failed to create input layout!");
inputLayout = nullptr;
}
// Add it to map
2017-02-10 09:26:18 +00:00
inputLayoutMap_[key] = inputLayout;
return inputLayout;
} else {
// Set it from map
return vertexDeclCached->second;
}
}
void DrawEngineD3D11::SetupVertexDecoder(u32 vertType) {
SetupVertexDecoderInternal(vertType);
}
inline void DrawEngineD3D11::SetupVertexDecoderInternal(u32 vertType) {
// As the decoder depends on the UVGenMode when we use UV prescale, we simply mash it
// into the top of the verttype where there are unused bits.
const u32 vertTypeID = (vertType & 0xFFFFFF) | (gstate.getUVGenMode() << 24);
// If vtype has changed, setup the vertex decoder.
if (vertTypeID != lastVType_) {
dec_ = GetVertexDecoder(vertTypeID);
lastVType_ = vertTypeID;
}
}
void DrawEngineD3D11::SubmitPrim(void *verts, void *inds, GEPrimitiveType prim, int vertexCount, u32 vertType, int *bytesRead) {
if (!indexGen.PrimCompatible(prevPrim_, prim) || numDrawCalls >= MAX_DEFERRED_DRAW_CALLS || vertexCountInDrawCalls + vertexCount > VERTEX_BUFFER_MAX)
Flush();
// TODO: Is this the right thing to do?
if (prim == GE_PRIM_KEEP_PREVIOUS) {
prim = prevPrim_ != GE_PRIM_INVALID ? prevPrim_ : GE_PRIM_POINTS;
} else {
prevPrim_ = prim;
}
SetupVertexDecoderInternal(vertType);
*bytesRead = vertexCount * dec_->VertexSize();
if ((vertexCount < 2 && prim > 0) || (vertexCount < 3 && prim > 2 && prim != GE_PRIM_RECTANGLES))
return;
DeferredDrawCall &dc = drawCalls[numDrawCalls];
dc.verts = verts;
dc.inds = inds;
dc.vertType = vertType;
dc.indexType = (vertType & GE_VTYPE_IDX_MASK) >> GE_VTYPE_IDX_SHIFT;
dc.prim = prim;
dc.vertexCount = vertexCount;
u32 dhash = dcid_;
dhash ^= (u32)(uintptr_t)verts;
dhash = __rotl(dhash, 13);
dhash ^= (u32)(uintptr_t)inds;
dhash = __rotl(dhash, 13);
dhash ^= (u32)vertType;
dhash = __rotl(dhash, 13);
dhash ^= (u32)vertexCount;
dhash = __rotl(dhash, 13);
dhash ^= (u32)prim;
dcid_ = dhash;
if (inds) {
GetIndexBounds(inds, vertexCount, vertType, &dc.indexLowerBound, &dc.indexUpperBound);
} else {
dc.indexLowerBound = 0;
dc.indexUpperBound = vertexCount - 1;
}
uvScale[numDrawCalls] = gstate_c.uv;
numDrawCalls++;
vertexCountInDrawCalls += vertexCount;
if (g_Config.bSoftwareSkinning && (vertType & GE_VTYPE_WEIGHT_MASK)) {
DecodeVertsStep();
decodeCounter_++;
}
if (prim == GE_PRIM_RECTANGLES && (gstate.getTextureAddress(0) & 0x3FFFFFFF) == (gstate.getFrameBufAddress() & 0x3FFFFFFF)) {
// Rendertarget == texture?
if (!g_Config.bDisableSlowFramebufEffects) {
gstate_c.Dirty(DIRTY_TEXTURE_PARAMS);
Flush();
}
}
}
void DrawEngineD3D11::DecodeVerts() {
const UVScale origUV = gstate_c.uv;
for (; decodeCounter_ < numDrawCalls; decodeCounter_++) {
gstate_c.uv = uvScale[decodeCounter_];
DecodeVertsStep();
}
gstate_c.uv = origUV;
// Sanity check
if (indexGen.Prim() < 0) {
ERROR_LOG_REPORT(G3D, "DecodeVerts: Failed to deduce prim: %i", indexGen.Prim());
// Force to points (0)
indexGen.AddPrim(GE_PRIM_POINTS, 0);
}
}
void DrawEngineD3D11::DecodeVertsStep() {
const int i = decodeCounter_;
const DeferredDrawCall &dc = drawCalls[i];
indexGen.SetIndex(decodedVerts_);
int indexLowerBound = dc.indexLowerBound, indexUpperBound = dc.indexUpperBound;
u32 indexType = dc.indexType;
void *inds = dc.inds;
if (indexType == GE_VTYPE_IDX_NONE >> GE_VTYPE_IDX_SHIFT) {
// Decode the verts and apply morphing. Simple.
dec_->DecodeVerts(decoded + decodedVerts_ * (int)dec_->GetDecVtxFmt().stride,
dc.verts, indexLowerBound, indexUpperBound);
decodedVerts_ += indexUpperBound - indexLowerBound + 1;
indexGen.AddPrim(dc.prim, dc.vertexCount);
} else {
// It's fairly common that games issue long sequences of PRIM calls, with differing
// inds pointer but the same base vertex pointer. We'd like to reuse vertices between
// these as much as possible, so we make sure here to combine as many as possible
// into one nice big drawcall, sharing data.
// 1. Look ahead to find the max index, only looking as "matching" drawcalls.
// Expand the lower and upper bounds as we go.
int lastMatch = i;
const int total = numDrawCalls;
for (int j = i + 1; j < total; ++j) {
if (drawCalls[j].verts != dc.verts)
break;
if (memcmp(&uvScale[j], &uvScale[i], sizeof(uvScale[0])) != 0)
break;
indexLowerBound = std::min(indexLowerBound, (int)drawCalls[j].indexLowerBound);
indexUpperBound = std::max(indexUpperBound, (int)drawCalls[j].indexUpperBound);
lastMatch = j;
}
// 2. Loop through the drawcalls, translating indices as we go.
switch (indexType) {
case GE_VTYPE_IDX_8BIT >> GE_VTYPE_IDX_SHIFT:
for (int j = i; j <= lastMatch; j++) {
indexGen.TranslatePrim(drawCalls[j].prim, drawCalls[j].vertexCount, (const u8 *)drawCalls[j].inds, indexLowerBound);
}
break;
case GE_VTYPE_IDX_16BIT >> GE_VTYPE_IDX_SHIFT:
for (int j = i; j <= lastMatch; j++) {
indexGen.TranslatePrim(drawCalls[j].prim, drawCalls[j].vertexCount, (const u16_le *)drawCalls[j].inds, indexLowerBound);
}
break;
case GE_VTYPE_IDX_32BIT >> GE_VTYPE_IDX_SHIFT:
for (int j = i; j <= lastMatch; j++) {
indexGen.TranslatePrim(drawCalls[j].prim, drawCalls[j].vertexCount, (const u32_le *)drawCalls[j].inds, indexLowerBound);
}
break;
}
const int vertexCount = indexUpperBound - indexLowerBound + 1;
// This check is a workaround for Pangya Fantasy Golf, which sends bogus index data when switching items in "My Room" sometimes.
if (decodedVerts_ + vertexCount > VERTEX_BUFFER_MAX) {
return;
}
// 3. Decode that range of vertex data.
dec_->DecodeVerts(decoded + decodedVerts_ * (int)dec_->GetDecVtxFmt().stride,
dc.verts, indexLowerBound, indexUpperBound);
decodedVerts_ += vertexCount;
// 4. Advance indexgen vertex counter.
indexGen.Advance(vertexCount);
decodeCounter_ = lastMatch;
}
}
inline u32 ComputeMiniHashRange(const void *ptr, size_t sz) {
// Switch to u32 units.
const u32 *p = (const u32 *)ptr;
sz >>= 2;
if (sz > 100) {
size_t step = sz / 4;
u32 hash = 0;
for (size_t i = 0; i < sz; i += step) {
hash += DoReliableHash32(p + i, 100, 0x3A44B9C4);
}
return hash;
} else {
return p[0] + p[sz - 1];
}
}
u32 DrawEngineD3D11::ComputeMiniHash() {
u32 fullhash = 0;
const int vertexSize = dec_->GetDecVtxFmt().stride;
const int indexSize = IndexSize(dec_->VertexType());
int step;
if (numDrawCalls < 3) {
step = 1;
} else if (numDrawCalls < 8) {
step = 4;
} else {
step = numDrawCalls / 8;
}
for (int i = 0; i < numDrawCalls; i += step) {
const DeferredDrawCall &dc = drawCalls[i];
if (!dc.inds) {
fullhash += ComputeMiniHashRange(dc.verts, vertexSize * dc.vertexCount);
} else {
int indexLowerBound = dc.indexLowerBound, indexUpperBound = dc.indexUpperBound;
fullhash += ComputeMiniHashRange((const u8 *)dc.verts + vertexSize * indexLowerBound, vertexSize * (indexUpperBound - indexLowerBound));
fullhash += ComputeMiniHashRange(dc.inds, indexSize * dc.vertexCount);
}
}
return fullhash;
}
void DrawEngineD3D11::MarkUnreliable(VertexArrayInfoD3D11 *vai) {
vai->status = VertexArrayInfoD3D11::VAI_UNRELIABLE;
if (vai->vbo) {
vai->vbo->Release();
vai->vbo = nullptr;
}
if (vai->ebo) {
vai->ebo->Release();
vai->ebo = nullptr;
}
}
ReliableHashType DrawEngineD3D11::ComputeHash() {
ReliableHashType fullhash = 0;
const int vertexSize = dec_->GetDecVtxFmt().stride;
const int indexSize = IndexSize(dec_->VertexType());
// TODO: Add some caps both for numDrawCalls and num verts to check?
// It is really very expensive to check all the vertex data so often.
for (int i = 0; i < numDrawCalls; i++) {
const DeferredDrawCall &dc = drawCalls[i];
if (!dc.inds) {
fullhash += DoReliableHash((const char *)dc.verts, vertexSize * dc.vertexCount, 0x1DE8CAC4);
} else {
int indexLowerBound = dc.indexLowerBound, indexUpperBound = dc.indexUpperBound;
int j = i + 1;
int lastMatch = i;
while (j < numDrawCalls) {
if (drawCalls[j].verts != dc.verts)
break;
indexLowerBound = std::min(indexLowerBound, (int)dc.indexLowerBound);
indexUpperBound = std::max(indexUpperBound, (int)dc.indexUpperBound);
lastMatch = j;
j++;
}
// This could get seriously expensive with sparse indices. Need to combine hashing ranges the same way
// we do when drawing.
fullhash += DoReliableHash((const char *)dc.verts + vertexSize * indexLowerBound,
vertexSize * (indexUpperBound - indexLowerBound), 0x029F3EE1);
// Hm, we will miss some indices when combining above, but meh, it should be fine.
fullhash += DoReliableHash((const char *)dc.inds, indexSize * dc.vertexCount, 0x955FD1CA);
i = lastMatch;
}
}
if (uvScale) {
fullhash += DoReliableHash(&uvScale[0], sizeof(uvScale[0]) * numDrawCalls, 0x0123e658);
}
return fullhash;
}
void DrawEngineD3D11::ClearTrackedVertexArrays() {
for (auto vai = vai_.begin(); vai != vai_.end(); vai++) {
delete vai->second;
}
vai_.clear();
}
2017-02-10 09:26:18 +00:00
void DrawEngineD3D11::BeginFrame() {
pushVerts_->Reset();
pushInds_->Reset();
if (--decimationCounter_ <= 0) {
decimationCounter_ = VERTEXCACHE_DECIMATION_INTERVAL;
} else {
return;
}
const int threshold = gpuStats.numFlips - VAI_KILL_AGE;
const int unreliableThreshold = gpuStats.numFlips - VAI_UNRELIABLE_KILL_AGE;
int unreliableLeft = VAI_UNRELIABLE_KILL_MAX;
for (auto iter = vai_.begin(); iter != vai_.end(); ) {
bool kill;
if (iter->second->status == VertexArrayInfoD3D11::VAI_UNRELIABLE) {
// We limit killing unreliable so we don't rehash too often.
kill = iter->second->lastFrame < unreliableThreshold && --unreliableLeft >= 0;
} else {
kill = iter->second->lastFrame < threshold;
}
if (kill) {
delete iter->second;
vai_.erase(iter++);
} else {
++iter;
}
}
// Enable if you want to see vertex decoders in the log output. Need a better way.
#if 0
char buffer[16384];
for (std::map<u32, VertexDecoder*>::iterator dec = decoderMap_.begin(); dec != decoderMap_.end(); ++dec) {
char *ptr = buffer;
ptr += dec->second->ToString(ptr);
// *ptr++ = '\n';
NOTICE_LOG(G3D, buffer);
}
#endif
}
VertexArrayInfoD3D11::~VertexArrayInfoD3D11() {
if (vbo) {
vbo->Release();
}
if (ebo) {
ebo->Release();
}
}
static uint32_t SwapRB(uint32_t c) {
return (c & 0xFF00FF00) | ((c >> 16) & 0xFF) | ((c << 16) & 0xFF0000);
}
// The inline wrapper in the header checks for numDrawCalls == 0
void DrawEngineD3D11::DoFlush() {
gpuStats.numFlushes++;
gpuStats.numTrackedVertexArrays = (int)vai_.size();
// This is not done on every drawcall, we should collect vertex data
// until critical state changes. That's when we draw (flush).
GEPrimitiveType prim = prevPrim_;
ApplyDrawState(prim);
bool useHWTransform = CanUseHardwareTransform(prim);
if (useHWTransform) {
ID3D11Buffer *vb_ = nullptr;
ID3D11Buffer *ib_ = nullptr;
int vertexCount = 0;
int maxIndex = 0;
bool useElements = true;
// Cannot cache vertex data with morph enabled.
bool useCache = g_Config.bVertexCache && !(lastVType_ & GE_VTYPE_MORPHCOUNT_MASK);
// Also avoid caching when software skinning.
if (g_Config.bSoftwareSkinning && (lastVType_ & GE_VTYPE_WEIGHT_MASK))
useCache = false;
if (useCache) {
u32 id = dcid_ ^ gstate.getUVGenMode(); // This can have an effect on which UV decoder we need to use! And hence what the decoded data will look like. See #9263
auto iter = vai_.find(id);
VertexArrayInfoD3D11 *vai;
if (iter != vai_.end()) {
// We've seen this before. Could have been a cached draw.
vai = iter->second;
} else {
vai = new VertexArrayInfoD3D11();
vai_[id] = vai;
}
switch (vai->status) {
case VertexArrayInfoD3D11::VAI_NEW:
{
// Haven't seen this one before.
ReliableHashType dataHash = ComputeHash();
vai->hash = dataHash;
vai->minihash = ComputeMiniHash();
vai->status = VertexArrayInfoD3D11::VAI_HASHING;
vai->drawsUntilNextFullHash = 0;
DecodeVerts(); // writes to indexGen
vai->numVerts = indexGen.VertexCount();
vai->prim = indexGen.Prim();
vai->maxIndex = indexGen.MaxIndex();
vai->flags = gstate_c.vertexFullAlpha ? VAI11_FLAG_VERTEXFULLALPHA : 0;
goto rotateVBO;
}
// Hashing - still gaining confidence about the buffer.
// But if we get this far it's likely to be worth creating a vertex buffer.
case VertexArrayInfoD3D11::VAI_HASHING:
{
vai->numDraws++;
if (vai->lastFrame != gpuStats.numFlips) {
vai->numFrames++;
}
if (vai->drawsUntilNextFullHash == 0) {
// Let's try to skip a full hash if mini would fail.
const u32 newMiniHash = ComputeMiniHash();
ReliableHashType newHash = vai->hash;
if (newMiniHash == vai->minihash) {
newHash = ComputeHash();
}
if (newMiniHash != vai->minihash || newHash != vai->hash) {
MarkUnreliable(vai);
DecodeVerts();
goto rotateVBO;
}
if (vai->numVerts > 64) {
// exponential backoff up to 16 draws, then every 24
vai->drawsUntilNextFullHash = std::min(24, vai->numFrames);
} else {
// Lower numbers seem much more likely to change.
vai->drawsUntilNextFullHash = 0;
}
// TODO: tweak
//if (vai->numFrames > 1000) {
// vai->status = VertexArrayInfo::VAI_RELIABLE;
//}
} else {
vai->drawsUntilNextFullHash--;
u32 newMiniHash = ComputeMiniHash();
if (newMiniHash != vai->minihash) {
MarkUnreliable(vai);
DecodeVerts();
goto rotateVBO;
}
}
if (vai->vbo == 0) {
DecodeVerts();
vai->numVerts = indexGen.VertexCount();
vai->prim = indexGen.Prim();
vai->maxIndex = indexGen.MaxIndex();
vai->flags = gstate_c.vertexFullAlpha ? VAI11_FLAG_VERTEXFULLALPHA : 0;
useElements = !indexGen.SeenOnlyPurePrims() || prim == GE_PRIM_TRIANGLE_FAN;
if (!useElements && indexGen.PureCount()) {
vai->numVerts = indexGen.PureCount();
}
_dbg_assert_msg_(G3D, gstate_c.vertBounds.minV >= gstate_c.vertBounds.maxV, "Should not have checked UVs when caching.");
// TODO: Combine these two into one buffer?
u32 size = dec_->GetDecVtxFmt().stride * indexGen.MaxIndex();
2017-02-09 23:01:56 +00:00
D3D11_BUFFER_DESC desc{ size, D3D11_USAGE_IMMUTABLE, D3D11_BIND_VERTEX_BUFFER, 0 };
D3D11_SUBRESOURCE_DATA data{ decoded };
device_->CreateBuffer(&desc, &data, &vai->vbo);
if (useElements) {
u32 size = sizeof(short) * indexGen.VertexCount();
2017-02-09 23:01:56 +00:00
D3D11_BUFFER_DESC desc{ size, D3D11_USAGE_IMMUTABLE, D3D11_BIND_INDEX_BUFFER, 0 };
2017-02-10 09:26:18 +00:00
D3D11_SUBRESOURCE_DATA data{ decIndex };
device_->CreateBuffer(&desc, &data, &vai->ebo);
} else {
vai->ebo = 0;
}
} else {
gpuStats.numCachedDrawCalls++;
useElements = vai->ebo ? true : false;
gpuStats.numCachedVertsDrawn += vai->numVerts;
gstate_c.vertexFullAlpha = vai->flags & VAI11_FLAG_VERTEXFULLALPHA;
}
vb_ = vai->vbo;
ib_ = vai->ebo;
vertexCount = vai->numVerts;
maxIndex = vai->maxIndex;
prim = static_cast<GEPrimitiveType>(vai->prim);
break;
}
// Reliable - we don't even bother hashing anymore. Right now we don't go here until after a very long time.
case VertexArrayInfoD3D11::VAI_RELIABLE:
{
vai->numDraws++;
if (vai->lastFrame != gpuStats.numFlips) {
vai->numFrames++;
}
gpuStats.numCachedDrawCalls++;
gpuStats.numCachedVertsDrawn += vai->numVerts;
vb_ = vai->vbo;
ib_ = vai->ebo;
vertexCount = vai->numVerts;
maxIndex = vai->maxIndex;
prim = static_cast<GEPrimitiveType>(vai->prim);
gstate_c.vertexFullAlpha = vai->flags & VAI11_FLAG_VERTEXFULLALPHA;
break;
}
case VertexArrayInfoD3D11::VAI_UNRELIABLE:
{
vai->numDraws++;
if (vai->lastFrame != gpuStats.numFlips) {
vai->numFrames++;
}
DecodeVerts();
goto rotateVBO;
}
}
vai->lastFrame = gpuStats.numFlips;
} else {
DecodeVerts();
rotateVBO:
gpuStats.numUncachedVertsDrawn += indexGen.VertexCount();
useElements = !indexGen.SeenOnlyPurePrims() || prim == GE_PRIM_TRIANGLE_FAN;
vertexCount = indexGen.VertexCount();
maxIndex = indexGen.MaxIndex();
if (!useElements && indexGen.PureCount()) {
vertexCount = indexGen.PureCount();
}
prim = indexGen.Prim();
}
VERBOSE_LOG(G3D, "Flush prim %i! %i verts in one go", prim, vertexCount);
bool hasColor = (lastVType_ & GE_VTYPE_COL_MASK) != GE_VTYPE_COL_NONE;
if (gstate.isModeThrough()) {
gstate_c.vertexFullAlpha = gstate_c.vertexFullAlpha && (hasColor || gstate.getMaterialAmbientA() == 255);
} else {
gstate_c.vertexFullAlpha = gstate_c.vertexFullAlpha && ((hasColor && (gstate.materialupdate & 1)) || gstate.getMaterialAmbientA() == 255) && (!gstate.isLightingEnabled() || gstate.getAmbientA() == 255);
}
ApplyDrawStateLate(false, 0);
D3D11VertexShader *vshader;
D3D11FragmentShader *fshader;
shaderManager_->GetShaders(prim, lastVType_, &vshader, &fshader, useHWTransform);
2017-02-10 09:26:18 +00:00
ID3D11InputLayout *inputLayout = SetupDecFmtForDraw(vshader, dec_->GetDecVtxFmt(), dec_->VertexType());
2017-02-09 23:01:56 +00:00
context_->PSSetShader(fshader->GetShader(), nullptr, 0);
context_->VSSetShader(vshader->GetShader(), nullptr, 0);
2017-02-12 11:02:13 +00:00
shaderManager_->UpdateUniforms();
shaderManager_->BindUniforms();
2017-02-10 09:26:18 +00:00
context_->IASetInputLayout(inputLayout);
UINT stride = dec_->GetDecVtxFmt().stride;
context_->IASetPrimitiveTopology(glprim[prim]);
if (!vb_) {
// Push!
2017-02-10 09:26:18 +00:00
UINT vOffset;
int vSize = (maxIndex + 1) * dec_->GetDecVtxFmt().stride;
uint8_t *vptr = pushVerts_->BeginPush(context_, &vOffset, vSize);
memcpy(vptr, decoded, vSize);
pushVerts_->EndPush(context_);
ID3D11Buffer *buf = pushVerts_->Buf();
context_->IASetVertexBuffers(0, 1, &buf, &stride, &vOffset);
if (useElements) {
2017-02-10 09:26:18 +00:00
UINT iOffset;
int iSize = 2 * indexGen.VertexCount();
2017-02-10 09:26:18 +00:00
uint8_t *iptr = pushInds_->BeginPush(context_, &iOffset, iSize);
memcpy(iptr, decIndex, iSize);
pushInds_->EndPush(context_);
context_->IASetIndexBuffer(pushInds_->Buf(), DXGI_FORMAT_R16_UINT, iOffset);
context_->DrawIndexed(vertexCount, 0, 0);
} else {
2017-02-10 09:26:18 +00:00
context_->Draw(vertexCount, 0);
}
} else {
UINT offset = 0;
context_->IASetVertexBuffers(0, 1, &vb_, &stride, &offset);
if (useElements) {
context_->IASetIndexBuffer(ib_, DXGI_FORMAT_R16_UINT, 0);
context_->DrawIndexed(vertexCount, 0, 0);
} else {
context_->Draw(vertexCount, 0);
}
}
} else {
DecodeVerts();
bool hasColor = (lastVType_ & GE_VTYPE_COL_MASK) != GE_VTYPE_COL_NONE;
if (gstate.isModeThrough()) {
gstate_c.vertexFullAlpha = gstate_c.vertexFullAlpha && (hasColor || gstate.getMaterialAmbientA() == 255);
} else {
gstate_c.vertexFullAlpha = gstate_c.vertexFullAlpha && ((hasColor && (gstate.materialupdate & 1)) || gstate.getMaterialAmbientA() == 255) && (!gstate.isLightingEnabled() || gstate.getAmbientA() == 255);
}
gpuStats.numUncachedVertsDrawn += indexGen.VertexCount();
prim = indexGen.Prim();
// Undo the strip optimization, not supported by the SW code yet.
if (prim == GE_PRIM_TRIANGLE_STRIP)
prim = GE_PRIM_TRIANGLES;
VERBOSE_LOG(G3D, "Flush prim %i SW! %i verts in one go", prim, indexGen.VertexCount());
int numTrans = 0;
bool drawIndexed = false;
u16 *inds = decIndex;
TransformedVertex *drawBuffer = NULL;
SoftwareTransformResult result;
memset(&result, 0, sizeof(result));
SoftwareTransformParams params;
memset(&params, 0, sizeof(params));
params.decoded = decoded;
params.transformed = transformed;
params.transformedExpanded = transformedExpanded;
params.fbman = framebufferManager_;
params.texCache = textureCache_;
params.allowSeparateAlphaClear = true;
int maxIndex = indexGen.MaxIndex();
SoftwareTransform(
prim, indexGen.VertexCount(),
dec_->VertexType(), inds, GE_VTYPE_IDX_16BIT, dec_->GetDecVtxFmt(),
maxIndex, drawBuffer, numTrans, drawIndexed, &params, &result);
ApplyDrawStateLate(result.setStencil, result.stencilValue);
2017-02-10 09:26:18 +00:00
if (result.action == SW_DRAW_PRIMITIVES) {
const int vertexSize = sizeof(transformed[0]);
D3D11VertexShader *vshader;
D3D11FragmentShader *fshader;
shaderManager_->GetShaders(prim, lastVType_, &vshader, &fshader, false);
context_->PSSetShader(fshader->GetShader(), nullptr, 0);
context_->VSSetShader(vshader->GetShader(), nullptr, 0);
shaderManager_->UpdateUniforms();
shaderManager_->BindUniforms();
2017-02-10 09:26:18 +00:00
// We really do need a vertex layout for each vertex shader (or at least check its ID bits for what inputs it uses)!
// Some vertex shaders ignore one of the inputs, and then the layout created from it will lack it, which will be a problem for others.
InputLayoutKey key{ 0xFFFFFFFF, vshader }; // Let's use 0xFFFFFFFF to signify TransformedVertex
auto iter = inputLayoutMap_.find(key);
ID3D11InputLayout *layout;
if (iter == inputLayoutMap_.end()) {
device_->CreateInputLayout(TransformedVertexElements, ARRAY_SIZE(TransformedVertexElements), vshader->bytecode().data(), vshader->bytecode().size(), &layout);
inputLayoutMap_[key] = layout;
} else {
layout = iter->second;
}
2017-02-10 09:26:18 +00:00
context_->IASetInputLayout(layout);
UINT stride = sizeof(TransformedVertex);
UINT vOffset = 0;
int vSize = maxIndex * stride;
2017-02-10 09:26:18 +00:00
uint8_t *vptr = pushVerts_->BeginPush(context_, &vOffset, vSize);
memcpy(vptr, drawBuffer, vSize);
pushVerts_->EndPush(context_);
ID3D11Buffer *buf = pushVerts_->Buf();
context_->IASetVertexBuffers(0, 1, &buf, &stride, &vOffset);
if (drawIndexed) {
2017-02-10 09:26:18 +00:00
UINT iOffset;
int iSize = sizeof(uint16_t) * numTrans;
2017-02-10 09:26:18 +00:00
uint8_t *iptr = pushInds_->BeginPush(context_, &iOffset, iSize);
memcpy(iptr, inds, iSize);
pushInds_->EndPush(context_);
context_->IASetIndexBuffer(pushInds_->Buf(), DXGI_FORMAT_R16_UINT, iOffset);
context_->DrawIndexed(numTrans, 0, 0);
} else {
context_->Draw(numTrans, 0);
}
2017-02-10 09:26:18 +00:00
} else if (result.action == SW_CLEAR) {
u32 clearColor = result.color;
float clearDepth = result.depth;
2017-02-10 09:32:32 +00:00
uint32_t clearFlag = 0;
2017-02-10 09:32:32 +00:00
if (gstate.isClearModeColorMask()) clearFlag |= Draw::COLOR;
if (gstate.isClearModeAlphaMask()) clearFlag |= Draw::STENCIL;
if (gstate.isClearModeDepthMask()) clearFlag |= Draw::DEPTH;
if (clearColor) {
framebufferManager_->SetColorUpdated(gstate_c.skipDrawReason);
}
2017-02-10 09:32:32 +00:00
if (clearFlag & Draw::DEPTH) {
framebufferManager_->SetDepthUpdated();
}
int colorMask = 0;
if (clearColor)
colorMask |= 7;
2017-02-10 09:32:32 +00:00
if (clearFlag & Draw::STENCIL) {
colorMask |= 8;
}
context_->OMSetBlendState(stockD3D11.blendStateDisabledWithColorMask[colorMask], nullptr, 0xFFFFFFFF);
2017-02-10 09:32:32 +00:00
uint8_t clearStencil = clearColor >> 24;
draw_->Clear(clearFlag, clearColor, clearDepth, clearStencil);
int scissorX2 = gstate.getScissorX2() + 1;
int scissorY2 = gstate.getScissorY2() + 1;
framebufferManager_->SetSafeSize(scissorX2, scissorY2);
if (g_Config.bBlockTransferGPU && (gstate_c.featureFlags & GPU_USE_CLEAR_RAM_HACK) && gstate.isClearModeColorMask() && (gstate.isClearModeAlphaMask() || gstate.FrameBufFormat() == GE_FORMAT_565)) {
2017-02-10 09:26:18 +00:00
int scissorX1 = gstate.getScissorX1();
int scissorY1 = gstate.getScissorY1();
ApplyClearToMemory(scissorX1, scissorY1, scissorX2, scissorY2, clearColor);
}
2017-02-10 09:26:18 +00:00
}
}
gpuStats.numDrawCalls += numDrawCalls;
gpuStats.numVertsSubmitted += vertexCountInDrawCalls;
indexGen.Reset();
decodedVerts_ = 0;
numDrawCalls = 0;
vertexCountInDrawCalls = 0;
decodeCounter_ = 0;
dcid_ = 0;
prevPrim_ = GE_PRIM_INVALID;
gstate_c.vertexFullAlpha = true;
framebufferManager_->SetColorUpdated(gstate_c.skipDrawReason);
// Now seems as good a time as any to reset the min/max coords, which we may examine later.
gstate_c.vertBounds.minU = 512;
gstate_c.vertBounds.minV = 512;
gstate_c.vertBounds.maxU = 0;
gstate_c.vertBounds.maxV = 0;
host->GPUNotifyDraw();
}
void DrawEngineD3D11::Resized() {
decJitCache_->Clear();
lastVType_ = -1;
dec_ = NULL;
for (auto iter = decoderMap_.begin(); iter != decoderMap_.end(); iter++) {
delete iter->second;
}
decoderMap_.clear();
}
bool DrawEngineD3D11::IsCodePtrVertexDecoder(const u8 *ptr) const {
return decJitCache_->IsInSpace(ptr);
}
void DrawEngineD3D11::TessellationDataTransferD3D11::SendDataToShader(const float * pos, const float * tex, const float * col, int size, bool hasColor, bool hasTexCoords)
{
}