Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

Merge pull request #6869 from unknownbrackets/d3d9

Browse source 
d3d: Centralize more framebuffer management
This commit is contained in:
Henrik Rydgård 2014-09-10 09:09:37 +02:00
commit cf9a6143ff
7 changed files with 553 additions and 590 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

276
GPU/Common/FramebufferCommon.cpp
View file

@ -17,7 +17,12 @@
#include <algorithm>
#include "Common/Common.h"
#include "Core/Config.h"
#include "Core/CoreParameter.h"
#include "Core/Reporting.h"
#include "Core/System.h"
#include "GPU/Common/FramebufferCommon.h"
#include "GPU/GPUInterface.h"
#include "GPU/GPUState.h"
FramebufferManagerCommon::FramebufferManagerCommon() :
@ -29,18 +34,57 @@ FramebufferManagerCommon::FramebufferManagerCommon() :
prevPrevDisplayFramebuf_(0),
frameLastFramebufUsed_(0),
currentRenderVfb_(0),
framebufRangeEnd_(0) {
framebufRangeEnd_(0),
hackForce04154000Download_(false) {
}
FramebufferManagerCommon::~FramebufferManagerCommon() {
}
void FramebufferManagerCommon::BeginFrame() {
DecimateFBOs();
currentRenderVfb_ = 0;
useBufferedRendering_ = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;
updateVRAM_ = !(g_Config.iRenderingMode == FB_NON_BUFFERED_MODE || g_Config.iRenderingMode == FB_BUFFERED_MODE);
}
void FramebufferManagerCommon::SetDisplayFramebuffer(u32 framebuf, u32 stride, GEBufferFormat format) {
displayFramebufPtr_ = framebuf;
displayStride_ = stride;
displayFormat_ = format;
}
VirtualFramebuffer *FramebufferManagerCommon::GetVFBAt(u32 addr) {
VirtualFramebuffer *match = NULL;
for (size_t i = 0; i < vfbs_.size(); ++i) {
VirtualFramebuffer *v = vfbs_[i];
if (MaskedEqual(v->fb_address, addr)) {
// Could check w too but whatever
if (match == NULL || match->last_frame_render < v->last_frame_render) {
match = v;
}
}
}
if (match != NULL) {
return match;
}
DEBUG_LOG(SCEGE, "Finding no FBO matching address %08x", addr);
return 0;
}
bool FramebufferManagerCommon::MaskedEqual(u32 addr1, u32 addr2) {
return (addr1 & 0x03FFFFFF) == (addr2 & 0x03FFFFFF);
}
u32 FramebufferManagerCommon::FramebufferByteSize(const VirtualFramebuffer *vfb) const {
return vfb->fb_stride * vfb->height * (vfb->format == GE_FORMAT_8888 ? 4 : 2);
}
bool FramebufferManagerCommon::ShouldDownloadFramebuffer(const VirtualFramebuffer *vfb) const {
return updateVRAM_ || (hackForce04154000Download_ && vfb->fb_address == 0x00154000);
}
// Heuristics to figure out the size of FBO to create.
void FramebufferManagerCommon::EstimateDrawingSize(int &drawing_width, int &drawing_height) {
static const int MAX_FRAMEBUF_HEIGHT = 512;
@ -115,3 +159,233 @@ void FramebufferManagerCommon::EstimateDrawingSize(int &drawing_width, int &draw
DEBUG_LOG(G3D, "Est: %08x V: %ix%i, R: %ix%i, S: %ix%i, STR: %i, THR:%i, Z:%08x = %ix%i", gstate.getFrameBufAddress(), viewport_width,viewport_height, region_width, region_height, scissor_width, scissor_height, fb_stride, gstate.isModeThrough(), gstate.isDepthWriteEnabled() ? gstate.getDepthBufAddress() : 0, drawing_width, drawing_height);
}
void FramebufferManagerCommon::DoSetRenderFrameBuffer() {
/*
if (useBufferedRendering_ && currentRenderVfb_) {
// Hack is enabled, and there was a previous framebuffer.
// Before we switch, let's do a series of trickery to copy one bit of stencil to
// destination alpha. Or actually, this is just a bunch of hackery attempts on Wipeout.
// Ignore for now.
glstate.depthTest.disable();
glstate.colorMask.set(GL_FALSE, GL_FALSE, GL_FALSE, GL_TRUE);
glstate.stencilTest.enable();
glstate.stencilOp.set(GL_KEEP, GL_KEEP, GL_KEEP); // don't modify stencil§
glstate.stencilFunc.set(GL_GEQUAL, 0xFE, 0xFF);
DrawPlainColor(0x00000000);
//glstate.stencilFunc.set(GL_LESS, 0x80, 0xFF);
//DrawPlainColor(0xFF000000);
glstate.stencilTest.disable();
glstate.colorMask.set(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
glstate.depthTest.disable();
glstate.colorMask.set(GL_FALSE, GL_FALSE, GL_FALSE, GL_TRUE);
DrawPlainColor(0x00000000);
shaderManager_->DirtyLastShader(); // dirty lastShader_
}
*/
gstate_c.framebufChanged = false;
// Get parameters
const u32 fb_address = gstate.getFrameBufRawAddress();
const int fb_stride = gstate.FrameBufStride();
const u32 z_address = gstate.getDepthBufRawAddress();
const int z_stride = gstate.DepthBufStride();
GEBufferFormat fmt = gstate.FrameBufFormat();
// As there are no clear "framebuffer width" and "framebuffer height" registers,
// we need to infer the size of the current framebuffer somehow.
int drawing_width, drawing_height;
EstimateDrawingSize(drawing_width, drawing_height);
gstate_c.cutRTOffsetX = 0;
bool vfbFormatChanged = false;
// Find a matching framebuffer
VirtualFramebuffer *vfb = 0;
for (size_t i = 0; i < vfbs_.size(); ++i) {
VirtualFramebuffer *v = vfbs_[i];
if (v->fb_address == fb_address) {
vfb = v;
// Update fb stride in case it changed
if (vfb->fb_stride != fb_stride || vfb->format != fmt) {
vfbFormatChanged = true;
vfb->fb_stride = fb_stride;
vfb->format = fmt;
}
// In throughmode, a higher height could be used. Let's avoid shrinking the buffer.
if (gstate.isModeThrough() && (int)vfb->width < fb_stride) {
vfb->width = std::max((int)vfb->width, drawing_width);
vfb->height = std::max((int)vfb->height, drawing_height);
} else {
vfb->width = drawing_width;
vfb->height = drawing_height;
}
break;
} else if (v->fb_address < fb_address && v->fb_address + v->fb_stride * 4 > fb_address) {
// Possibly a render-to-offset.
const u32 bpp = v->format == GE_FORMAT_8888 ? 4 : 2;
const int x_offset = (fb_address - v->fb_address) / bpp;
if (v->format == fmt && v->fb_stride == fb_stride && x_offset < fb_stride && v->height >= drawing_height) {
WARN_LOG_REPORT_ONCE(renderoffset, HLE, "Rendering to framebuffer offset: %08x +%dx%d", v->fb_address, x_offset, 0);
vfb = v;
gstate_c.cutRTOffsetX = x_offset;
vfb->width = std::max((int)vfb->width, x_offset + drawing_width);
// To prevent the newSize code from being confused.
drawing_width += x_offset;
break;
}
}
}
if (vfb) {
if ((drawing_width != vfb->bufferWidth || drawing_height != vfb->bufferHeight)) {
// Even if it's not newly wrong, if this is larger we need to resize up.
if (vfb->width > vfb->bufferWidth || vfb->height > vfb->bufferHeight) {
ResizeFramebufFBO(vfb, vfb->width, vfb->height);
} else if (vfb->newWidth != drawing_width || vfb->newHeight != drawing_height) {
// If it's newly wrong, or changing every frame, just keep track.
vfb->newWidth = drawing_width;
vfb->newHeight = drawing_height;
vfb->lastFrameNewSize = gpuStats.numFlips;
} else if (vfb->lastFrameNewSize + FBO_OLD_AGE < gpuStats.numFlips) {
// Okay, it's changed for a while (and stayed that way.) Let's start over.
// But only if we really need to, to avoid blinking.
bool needsRecreate = vfb->bufferWidth > fb_stride;
needsRecreate = needsRecreate || vfb->newWidth > vfb->bufferWidth || vfb->newWidth * 2 < vfb->bufferWidth;
needsRecreate = needsRecreate || vfb->newHeight > vfb->newHeight || vfb->newHeight * 2 < vfb->newHeight;
if (needsRecreate) {
ResizeFramebufFBO(vfb, vfb->width, vfb->height, true);
}
}
} else {
// It's not different, let's keep track of that too.
vfb->lastFrameNewSize = gpuStats.numFlips;
}
}
float renderWidthFactor = (float)PSP_CoreParameter().renderWidth / 480.0f;
float renderHeightFactor = (float)PSP_CoreParameter().renderHeight / 272.0f;
if (hackForce04154000Download_ && fb_address == 0x00154000) {
renderWidthFactor = 1.0;
renderHeightFactor = 1.0;
}
// None found? Create one.
if (!vfb) {
vfb = new VirtualFramebuffer();
vfb->fbo = 0;
vfb->fb_address = fb_address;
vfb->fb_stride = fb_stride;
vfb->z_address = z_address;
vfb->z_stride = z_stride;
vfb->width = drawing_width;
vfb->height = drawing_height;
vfb->newWidth = drawing_width;
vfb->newHeight = drawing_height;
vfb->lastFrameNewSize = gpuStats.numFlips;
vfb->renderWidth = (u16)(drawing_width * renderWidthFactor);
vfb->renderHeight = (u16)(drawing_height * renderHeightFactor);
vfb->bufferWidth = drawing_width;
vfb->bufferHeight = drawing_height;
vfb->format = fmt;
vfb->drawnWidth = 0;
vfb->drawnHeight = 0;
vfb->drawnFormat = fmt;
vfb->usageFlags = FB_USAGE_RENDERTARGET;
SetColorUpdated(vfb);
vfb->depthUpdated = false;
u32 byteSize = FramebufferByteSize(vfb);
u32 fb_address_mem = (fb_address & 0x3FFFFFFF) | 0x04000000;
if (Memory::IsVRAMAddress(fb_address_mem) && fb_address_mem + byteSize > framebufRangeEnd_) {
framebufRangeEnd_ = fb_address_mem + byteSize;
}
ResizeFramebufFBO(vfb, drawing_width, drawing_height, true);
NotifyRenderFramebufferCreated(vfb);
if (useBufferedRendering_ && !updateVRAM_ && !g_Config.bDisableSlowFramebufEffects) {
u32 byteSize = FramebufferByteSize(vfb);
u32 fb_address_mem = (vfb->fb_address & 0x3FFFFFFF) | 0x04000000;
gpu->PerformMemoryUpload(fb_address_mem, byteSize);
NotifyStencilUpload(fb_address_mem, byteSize, true);
// TODO: Is it worth trying to upload the depth buffer?
}
INFO_LOG(SCEGE, "Creating FBO for %08x : %i x %i x %i", vfb->fb_address, vfb->width, vfb->height, vfb->format);
vfb->last_frame_render = gpuStats.numFlips;
vfb->last_frame_used = 0;
vfb->last_frame_attached = 0;
frameLastFramebufUsed_ = gpuStats.numFlips;
vfbs_.push_back(vfb);
currentRenderVfb_ = vfb;
// Let's check for depth buffer overlap. Might be interesting.
bool sharingReported = false;
bool writingDepth = true;
// Technically, it may write depth later, but we're trying to detect it only when it's really true.
if (gstate.isModeClear()) {
writingDepth = !gstate.isClearModeDepthMask() && gstate.isDepthWriteEnabled();
} else {
writingDepth = gstate.isDepthWriteEnabled();
}
for (size_t i = 0, end = vfbs_.size(); i < end; ++i) {
if (vfbs_[i]->z_stride != 0 && fb_address == vfbs_[i]->z_address) {
// If it's clearing it, most likely it just needs more video memory.
// Technically it could write something interesting and the other might not clear, but that's not likely.
if (!gstate.isModeClear() || !gstate.isClearModeColorMask() || !gstate.isClearModeAlphaMask()) {
if (fb_address != z_address && vfbs_[i]->fb_address != vfbs_[i]->z_address) {
WARN_LOG_REPORT(SCEGE, "FBO created from existing depthbuffer as color, %08x/%08x and %08x/%08x", fb_address, z_address, vfbs_[i]->fb_address, vfbs_[i]->z_address);
}
}
} else if (z_stride != 0 && z_address == vfbs_[i]->fb_address) {
// If it's clearing it, then it's probably just the reverse of the above case.
if (writingDepth) {
WARN_LOG_REPORT(SCEGE, "FBO using existing buffer as depthbuffer, %08x/%08x and %08x/%08x", fb_address, z_address, vfbs_[i]->fb_address, vfbs_[i]->z_address);
}
} else if (vfbs_[i]->z_stride != 0 && z_address == vfbs_[i]->z_address && fb_address != vfbs_[i]->fb_address && !sharingReported) {
// This happens a lot, but virtually always it's cleared.
// It's possible the other might not clear, but when every game is reported it's not useful.
if (writingDepth) {
WARN_LOG_REPORT(SCEGE, "FBO reusing depthbuffer, %08x/%08x and %08x/%08x", fb_address, z_address, vfbs_[i]->fb_address, vfbs_[i]->z_address);
sharingReported = true;
}
}
}
// We already have it!
} else if (vfb != currentRenderVfb_) {
// Use it as a render target.
DEBUG_LOG(SCEGE, "Switching render target to FBO for %08x: %i x %i x %i ", vfb->fb_address, vfb->width, vfb->height, vfb->format);
vfb->usageFlags |= FB_USAGE_RENDERTARGET;
vfb->last_frame_render = gpuStats.numFlips;
frameLastFramebufUsed_ = gpuStats.numFlips;
vfb->dirtyAfterDisplay = true;
if ((gstate_c.skipDrawReason & SKIPDRAW_SKIPFRAME) == 0)
vfb->reallyDirtyAfterDisplay = true;
VirtualFramebuffer *prev = currentRenderVfb_;
currentRenderVfb_ = vfb;
NotifyRenderFramebufferSwitched(prev, vfb);
} else {
vfb->last_frame_render = gpuStats.numFlips;
frameLastFramebufUsed_ = gpuStats.numFlips;
vfb->dirtyAfterDisplay = true;
if ((gstate_c.skipDrawReason & SKIPDRAW_SKIPFRAME) == 0)
vfb->reallyDirtyAfterDisplay = true;
NotifyRenderFramebufferUpdated(vfb, vfbFormatChanged);
}
gstate_c.curRTWidth = vfb->width;
gstate_c.curRTHeight = vfb->height;
gstate_c.curRTRenderWidth = vfb->renderWidth;
gstate_c.curRTRenderHeight = vfb->renderHeight;
}

34
GPU/Common/FramebufferCommon.h
View file

@ -93,7 +93,12 @@ public:
FramebufferManagerCommon();
virtual ~FramebufferManagerCommon();
virtual void DoSetRenderFrameBuffer() = 0;
void BeginFrame();
virtual bool NotifyFramebufferCopy(u32 src, u32 dest, int size, bool isMemset = false) = 0;
virtual bool NotifyStencilUpload(u32 addr, int size, bool skipZero = false) = 0;
void DoSetRenderFrameBuffer();
void SetRenderFrameBuffer() {
// Inlining this part since it's so frequent.
if (!gstate_c.framebufChanged && currentRenderVfb_) {
@ -138,6 +143,12 @@ public:
return true;
}
// TODO: Break out into some form of FBO manager
VirtualFramebuffer *GetVFBAt(u32 addr);
VirtualFramebuffer *GetDisplayVFB() {
return GetVFBAt(displayFramebufPtr_);
}
int GetRenderWidth() const { return currentRenderVfb_ ? currentRenderVfb_->renderWidth : 480; }
int GetRenderHeight() const { return currentRenderVfb_ ? currentRenderVfb_->renderHeight : 272; }
int GetTargetWidth() const { return currentRenderVfb_ ? currentRenderVfb_->width : 480; }
@ -149,6 +160,17 @@ public:
protected:
void EstimateDrawingSize(int &drawing_width, int &drawing_height);
u32 FramebufferByteSize(const VirtualFramebuffer *vfb) const;
static bool MaskedEqual(u32 addr1, u32 addr2);
virtual void DecimateFBOs() = 0;
virtual void DestroyFramebuf(VirtualFramebuffer *vfb) = 0;
virtual void ResizeFramebufFBO(VirtualFramebuffer *vfb, u16 w, u16 h, bool force = false) = 0;
virtual void NotifyRenderFramebufferCreated(VirtualFramebuffer *vfb) = 0;
virtual void NotifyRenderFramebufferSwitched(VirtualFramebuffer *prevVfb, VirtualFramebuffer *vfb) = 0;
virtual void NotifyRenderFramebufferUpdated(VirtualFramebuffer *vfb, bool vfbFormatChanged) = 0;
bool ShouldDownloadFramebuffer(const VirtualFramebuffer *vfb) const;
void SetColorUpdated(VirtualFramebuffer *dstBuffer) {
dstBuffer->memoryUpdated = false;
@ -178,5 +200,15 @@ protected:
// The range of PSP memory that may contain FBOs. So we can skip iterating.
u32 framebufRangeEnd_;
bool useBufferedRendering_;
bool updateVRAM_;
std::vector<VirtualFramebuffer *> vfbs_;
bool hackForce04154000Download_;
// Aggressively delete unused FBOs to save gpu memory.
enum {
FBO_OLD_AGE = 5,
};
};

389
GPU/Directx9/FramebufferDX9.cpp
View file

@ -35,16 +35,6 @@
#include <algorithm>
namespace DX9 {
// Aggressively delete unused FBO:s to save gpu memory.
enum {
FBO_OLD_AGE = 5,
};
static bool MaskedEqual(u32 addr1, u32 addr2) {
return (addr1 & 0x03FFFFFF) == (addr2 & 0x03FFFFFF);
}
inline u16 RGBA8888toRGB565(u32 px) {
return ((px >> 3) & 0x001F) | ((px >> 5) & 0x07E0) | ((px >> 8) & 0xF800);
}
@ -136,11 +126,11 @@ namespace DX9 {
usage = D3DUSAGE_DYNAMIC;
}
HRESULT hr = pD3Ddevice->CreateTexture(512, 272, 1, usage, D3DFMT(D3DFMT_A8R8G8B8), pool, &drawPixelsTex_, NULL);
if (!hr) {
if (FAILED(hr)) {
drawPixelsTex_ = nullptr;
ERROR_LOG(G3D, "Failed to create drawpixels texture");
}
useBufferedRendering_ = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;
BeginFrame();
}
FramebufferManagerDX9::~FramebufferManagerDX9() {
@ -294,26 +284,6 @@ namespace DX9 {
pD3Ddevice->DrawPrimitiveUP(D3DPT_TRIANGLEFAN, 2, coord, 5 * sizeof(float));
}
VirtualFramebuffer *FramebufferManagerDX9::GetVFBAt(u32 addr) {
VirtualFramebuffer *match = NULL;
for (size_t i = 0; i < vfbs_.size(); ++i) {
VirtualFramebuffer *v = vfbs_[i];
if (MaskedEqual(v->fb_address, addr) && v->format == displayFormat_ && v->width >= 480) {
// Could check w too but whatever
if (match == NULL || match->last_frame_render < v->last_frame_render) {
match = v;
}
}
}
if (match != NULL) {
return match;
}
DEBUG_LOG(SCEGE, "Finding no FBO matching address %08x", addr);
return 0;
}
void FramebufferManagerDX9::DestroyFramebuf(VirtualFramebuffer *v) {
textureCache_->NotifyFramebuffer(v->fb_address, v, NOTIFY_FB_DESTROYED);
if (v->fbo) {
@ -334,221 +304,171 @@ namespace DX9 {
delete v;
}
void FramebufferManagerDX9::DoSetRenderFrameBuffer() {
#if 0
if (g_Config.iRenderingMode != 0 && g_Config.bWipeFramebufferAlpha && currentRenderVfb_) {
// Hack is enabled, and there was a previous framebuffer.
// Before we switch, let's do a series of trickery to copy one bit of stencil to
// destination alpha. Or actually, this is just a bunch of hackery attempts on Wipeout.
// Ignore for now.
/*
glstate.depthTest.disable();
glstate.colorMask.set(GL_FALSE, GL_FALSE, GL_FALSE, GL_TRUE);
glstate.stencilTest.enable();
glstate.stencilOp.set(GL_KEEP, GL_KEEP, GL_KEEP); // don't modify stencil?
glstate.stencilFunc.set(GL_GEQUAL, 0xFE, 0xFF);
DrawPlainColor(0x00000000);
//glstate.stencilFunc.set(GL_LESS, 0x80, 0xFF);
//DrawPlainColor(0xFF000000);
glstate.stencilTest.disable();
glstate.colorMask.set(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
*/
void FramebufferManagerDX9::ResizeFramebufFBO(VirtualFramebuffer *vfb, u16 w, u16 h, bool force) {
float renderWidthFactor = (float)vfb->renderWidth / (float)vfb->bufferWidth;
float renderHeightFactor = (float)vfb->renderHeight / (float)vfb->bufferHeight;
VirtualFramebuffer old = *vfb;
glstate.depthTest.disable();
glstate.colorMask.set(GL_FALSE, GL_FALSE, GL_FALSE, GL_TRUE);
DrawPlainColor(0x00000000);
shaderManager_->DirtyLastShader(); // dirty lastShader_
}
#endif
gstate_c.framebufChanged = false;
// Get parameters
u32 fb_address = gstate.getFrameBufRawAddress();
int fb_stride = gstate.fbwidth & 0x3C0;
u32 z_address = gstate.getDepthBufRawAddress();
int z_stride = gstate.zbwidth & 0x3C0;
// Yeah this is not completely right. but it'll do for now.
//int drawing_width = ((gstate.region2) & 0x3FF) + 1;
//int drawing_height = ((gstate.region2 >> 10) & 0x3FF) + 1;
GEBufferFormat fmt = gstate.FrameBufFormat();
// As there are no clear "framebuffer width" and "framebuffer height" registers,
// we need to infer the size of the current framebuffer somehow.
int drawing_width, drawing_height;
EstimateDrawingSize(drawing_width, drawing_height);
int buffer_width = drawing_width;
int buffer_height = drawing_height;
// Find a matching framebuffer
VirtualFramebuffer *vfb = 0;
for (size_t i = 0; i < vfbs_.size(); ++i) {
VirtualFramebuffer *v = vfbs_[i];
if (MaskedEqual(v->fb_address, fb_address) && v->width >= drawing_width && v->height >= drawing_height) {
// Let's not be so picky for now. Let's say this is the one.
vfb = v;
// Update fb stride in case it changed
vfb->fb_stride = fb_stride;
v->format = fmt;
break;
if (force) {
vfb->bufferWidth = w;
vfb->bufferHeight = h;
} else {
if (vfb->bufferWidth >= w && vfb->bufferHeight >= h) {
return;
}
// In case it gets thin and wide, don't resize down either side.
vfb->bufferWidth = std::max(vfb->bufferWidth, w);
vfb->bufferHeight = std::max(vfb->bufferHeight, h);
}
float renderWidthFactor = (float)PSP_CoreParameter().renderWidth / 480.0f;
float renderHeightFactor = (float)PSP_CoreParameter().renderHeight / 272.0f;
vfb->renderWidth = vfb->bufferWidth * renderWidthFactor;
vfb->renderHeight = vfb->bufferHeight * renderHeightFactor;
// None found? Create one.
if (!vfb) {
gstate_c.textureChanged = true;
vfb = new VirtualFramebuffer();
vfb->fbo = 0;
vfb->fb_address = fb_address;
vfb->fb_stride = fb_stride;
vfb->z_address = z_address;
vfb->z_stride = z_stride;
vfb->width = drawing_width;
vfb->height = drawing_height;
vfb->renderWidth = (u16)(drawing_width * renderWidthFactor);
vfb->renderHeight = (u16)(drawing_height * renderHeightFactor);
vfb->bufferWidth = buffer_width;
vfb->bufferHeight = buffer_height;
vfb->format = fmt;
vfb->usageFlags = FB_USAGE_RENDERTARGET;
vfb->dirtyAfterDisplay = true;
if ((gstate_c.skipDrawReason & SKIPDRAW_SKIPFRAME) == 0)
vfb->reallyDirtyAfterDisplay = true;
vfb->memoryUpdated = false;
bool trueColor = g_Config.bTrueColor;
if (hackForce04154000Download_ && vfb->fb_address == 0x00154000) {
trueColor = true;
}
if (g_Config.bTrueColor) {
if (trueColor) {
vfb->colorDepth = FBO_8888;
} else {
switch (vfb->format) {
case GE_FORMAT_4444:
vfb->colorDepth = FBO_4444;
break;
case GE_FORMAT_5551:
vfb->colorDepth = FBO_5551;
break;
case GE_FORMAT_565:
vfb->colorDepth = FBO_565;
break;
case GE_FORMAT_8888:
default:
vfb->colorDepth = FBO_8888;
} else {
switch (fmt) {
case GE_FORMAT_4444:
vfb->colorDepth = FBO_4444;
break;
case GE_FORMAT_5551:
vfb->colorDepth = FBO_5551;
break;
case GE_FORMAT_565:
vfb->colorDepth = FBO_565;
break;
case GE_FORMAT_8888:
vfb->colorDepth = FBO_8888;
break;
default:
vfb->colorDepth = FBO_8888;
break;
break;
}
}
textureCache_->ForgetLastTexture();
fbo_unbind();
if (!useBufferedRendering_) {
if (vfb->fbo) {
fbo_destroy(vfb->fbo);
vfb->fbo = 0;
}
return;
}
vfb->fbo = fbo_create(vfb->renderWidth, vfb->renderHeight, 1, true, (FBOColorDepth)vfb->colorDepth);
if (old.fbo) {
INFO_LOG(SCEGE, "Resizing FBO for %08x : %i x %i x %i", vfb->fb_address, w, h, vfb->format);
if (vfb->fbo) {
ClearBuffer();
if (!g_Config.bDisableSlowFramebufEffects) {
// TODO
//BlitFramebuffer_(vfb, 0, 0, &old, 0, 0, std::min(vfb->bufferWidth, vfb->width), std::min(vfb->height, vfb->bufferHeight), 0);
}
}
fbo_destroy(old.fbo);
if (vfb->fbo) {
fbo_bind_as_render_target(vfb->fbo);
}
}
if (useBufferedRendering_) {
vfb->fbo = fbo_create(vfb->renderWidth, vfb->renderHeight, 1, true, (FBOColorDepth)vfb->colorDepth);
if (vfb->fbo) {
fbo_bind_as_render_target(vfb->fbo);
} else {
ERROR_LOG(SCEGE, "Error creating FBO! %i x %i", vfb->renderWidth, vfb->renderHeight);
}
if (!vfb->fbo) {
ERROR_LOG(SCEGE, "Error creating FBO! %i x %i", vfb->renderWidth, vfb->renderHeight);
}
}
void FramebufferManagerDX9::NotifyRenderFramebufferCreated(VirtualFramebuffer *vfb) {
if (!useBufferedRendering_) {
fbo_unbind();
// Let's ignore rendering to targets that have not (yet) been displayed.
gstate_c.skipDrawReason |= SKIPDRAW_NON_DISPLAYED_FB;
}
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_CREATED);
ClearBuffer();
// TODO (in Common)
//if (useBufferedRendering_ && !updateVRAM_ && !g_Config.bDisableSlowFramebufEffects) {
// u32 byteSize = FramebufferByteSize(vfb);
// u32 fb_address_mem = (vfb->fb_address & 0x3FFFFFFF) | 0x04000000;
// gpu->PerformMemoryUpload(fb_address_mem, byteSize);
// NotifyStencilUpload(fb_address_mem, byteSize, true);
// // TODO: Is it worth trying to upload the depth buffer?
//}
// ugly...
if (gstate_c.curRTWidth != vfb->width || gstate_c.curRTHeight != vfb->height) {
shaderManager_->DirtyUniform(DIRTY_PROJTHROUGHMATRIX);
}
}
void FramebufferManagerDX9::NotifyRenderFramebufferSwitched(VirtualFramebuffer *prevVfb, VirtualFramebuffer *vfb) {
if (ShouldDownloadFramebuffer(vfb) && !vfb->memoryUpdated) {
// TODO
//ReadFramebufferToMemory(vfb, true, 0, 0, vfb->width, vfb->height);
}
textureCache_->ForgetLastTexture();
if (useBufferedRendering_) {
if (vfb->fbo) {
fbo_bind_as_render_target(vfb->fbo);
} else {
// wtf? This should only happen very briefly when toggling bBufferedRendering
fbo_unbind();
// Let's ignore rendering to targets that have not (yet) been displayed.
}
} else {
if (vfb->fbo) {
// wtf? This should only happen very briefly when toggling bBufferedRendering
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_DESTROYED);
fbo_destroy(vfb->fbo);
vfb->fbo = 0;
}
fbo_unbind();
// Let's ignore rendering to targets that have not (yet) been displayed.
if (vfb->usageFlags & FB_USAGE_DISPLAYED_FRAMEBUFFER) {
gstate_c.skipDrawReason &= ~SKIPDRAW_NON_DISPLAYED_FB;
} else {
gstate_c.skipDrawReason |= SKIPDRAW_NON_DISPLAYED_FB;
}
}
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_UPDATED);
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_CREATED);
vfb->last_frame_render = gpuStats.numFlips;
vfb->last_frame_used = 0;
vfb->last_frame_attached = 0;
frameLastFramebufUsed_ = gpuStats.numFlips;
vfbs_.push_back(vfb);
ClearBuffer();
currentRenderVfb_ = vfb;
INFO_LOG(SCEGE, "Creating FBO for %08x : %i x %i x %i", vfb->fb_address, vfb->width, vfb->height, vfb->format);
// Let's check for depth buffer overlap. Might be interesting.
bool sharingReported = false;
for (size_t i = 0, end = vfbs_.size(); i < end; ++i) {
if (MaskedEqual(fb_address, vfbs_[i]->z_address)) {
WARN_LOG_REPORT(SCEGE, "FBO created from existing depthbuffer (unsupported), %08x/%08x and %08x/%08x", fb_address, z_address, vfbs_[i]->fb_address, vfbs_[i]->z_address);
} else if (MaskedEqual(z_address, vfbs_[i]->fb_address)) {
WARN_LOG_REPORT(SCEGE, "FBO using other buffer as depthbuffer (unsupported), %08x/%08x and %08x/%08x", fb_address, z_address, vfbs_[i]->fb_address, vfbs_[i]->z_address);
} else if (MaskedEqual(z_address, vfbs_[i]->z_address) && fb_address != vfbs_[i]->fb_address && !sharingReported) {
WARN_LOG_REPORT(SCEGE, "FBO sharing existing depthbuffer (unsupported), %08x/%08x and %08x/%08x", fb_address, z_address, vfbs_[i]->fb_address, vfbs_[i]->z_address);
sharingReported = true;
}
}
// We already have it!
} else if (vfb != currentRenderVfb_) {
bool updateVRAM = !(g_Config.iRenderingMode == FB_NON_BUFFERED_MODE || g_Config.iRenderingMode == FB_BUFFERED_MODE);
if (updateVRAM && !vfb->memoryUpdated) {
ReadFramebufferToMemory(vfb, true);
}
// Use it as a render target.
DEBUG_LOG(SCEGE, "Switching render target to FBO for %08x: %i x %i x %i ", vfb->fb_address, vfb->width, vfb->height, vfb->format);
vfb->usageFlags |= FB_USAGE_RENDERTARGET;
gstate_c.textureChanged = true;
vfb->last_frame_render = gpuStats.numFlips;
frameLastFramebufUsed_ = gpuStats.numFlips;
vfb->dirtyAfterDisplay = true;
if ((gstate_c.skipDrawReason & SKIPDRAW_SKIPFRAME) == 0)
vfb->reallyDirtyAfterDisplay = true;
vfb->memoryUpdated = false;
if (useBufferedRendering_) {
if (vfb->fbo) {
fbo_bind_as_render_target(vfb->fbo);
} else {
// wtf? This should only happen very briefly when toggling bBufferedRendering
fbo_unbind();
}
} else {
if (vfb->fbo) {
// wtf? This should only happen very briefly when toggling bBufferedRendering
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_DESTROYED);
fbo_destroy(vfb->fbo);
vfb->fbo = 0;
}
fbo_unbind();
// Let's ignore rendering to targets that have not (yet) been displayed.
if (vfb->usageFlags & FB_USAGE_DISPLAYED_FRAMEBUFFER) {
gstate_c.skipDrawReason &= ~SKIPDRAW_NON_DISPLAYED_FB;
} else {
gstate_c.skipDrawReason |= SKIPDRAW_NON_DISPLAYED_FB;
}
}
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_UPDATED);
#if 0
// Some tiled mobile GPUs benefit IMMENSELY from clearing an FBO before rendering
// to it. This broke stuff before, so now it only clears on the first use of an
// FBO in a frame. This means that some games won't be able to avoid the on-some-GPUs
// performance-crushing framebuffer reloads from RAM, but we'll have to live with that.
if (vfb->last_frame_render != gpuStats.numFlips) {
ClearBuffer();
}
#endif
currentRenderVfb_ = vfb;
} else {
vfb->last_frame_render = gpuStats.numFlips;
frameLastFramebufUsed_ = gpuStats.numFlips;
vfb->dirtyAfterDisplay = true;
if ((gstate_c.skipDrawReason & SKIPDRAW_SKIPFRAME) == 0)
vfb->reallyDirtyAfterDisplay = true;
// Copy depth pixel value from the read framebuffer to the draw framebuffer
if (prevVfb && !g_Config.bDisableSlowFramebufEffects) {
// TODO
//BlitFramebufferDepth(prevVfb, vfb);
}
if (vfb->drawnFormat != vfb->format) {
// TODO: Might ultimately combine this with the resize step in DoSetRenderFrameBuffer().
// TODO
//ReformatFramebufferFrom(vfb, vfb->drawnFormat);
}
// ugly...
if (gstate_c.curRTWidth != vfb->width || gstate_c.curRTHeight != vfb->height) {
shaderManager_->DirtyUniform(DIRTY_PROJTHROUGHMATRIX);
}
}
void FramebufferManagerDX9::NotifyRenderFramebufferUpdated(VirtualFramebuffer *vfb, bool vfbFormatChanged) {
if (vfbFormatChanged) {
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_UPDATED);
if (vfb->drawnFormat != vfb->format) {
// TODO
//ReformatFramebufferFrom(vfb, vfb->drawnFormat);
}
}
// ugly...
if (gstate_c.curRTWidth != vfb->width || gstate_c.curRTHeight != vfb->height) {
shaderManager_->DirtyUniform(DIRTY_PROJTHROUGHMATRIX);
gstate_c.curRTWidth = vfb->width;
gstate_c.curRTHeight = vfb->height;
}
}
@ -896,12 +816,6 @@ namespace DX9 {
resized_ = false;
}
void FramebufferManagerDX9::BeginFrame() {
DecimateFBOs();
currentRenderVfb_ = 0;
useBufferedRendering_ = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;
}
std::vector<FramebufferInfo> FramebufferManagerDX9::GetFramebufferList() {
std::vector<FramebufferInfo> list;
@ -922,7 +836,7 @@ namespace DX9 {
}
// MotoGP workaround
void FramebufferManagerDX9::NotifyFramebufferCopy(u32 src, u32 dest, int size) {
bool FramebufferManagerDX9::NotifyFramebufferCopy(u32 src, u32 dest, int size, bool isMemset) {
for (size_t i = 0; i < vfbs_.size(); i++) {
// This size fits for MotoGP. Might want to make this more flexible for other games if they do the same.
if ((vfbs_[i]->fb_address | 0x04000000) == src && size == 512 * 272 * 2) {
@ -930,6 +844,13 @@ namespace DX9 {
knownFramebufferCopies_.insert(std::pair<u32, u32>(src, dest));
}
}
// TODO
return false;
}
bool FramebufferManagerDX9::NotifyStencilUpload(u32 addr, int size, bool skipZero) {
// TODO
return false;
}
void FramebufferManagerDX9::DecimateFBOs() {

20
GPU/Directx9/FramebufferDX9.h
View file

@ -60,29 +60,22 @@ public:
void DrawActiveTexture(LPDIRECT3DTEXTURE9 tex, float x, float y, float w, float h, float destW, float destH, bool flip = false, float uscale = 1.0f, float vscale = 1.0f);
void DestroyAllFBOs();
void DecimateFBOs();
void BeginFrame();
void EndFrame();
void Resized();
void DeviceLost();
void CopyDisplayToOutput();
virtual void DoSetRenderFrameBuffer() override; // Uses parameters computed from gstate
void UpdateFromMemory(u32 addr, int size, bool safe);
void ReadFramebufferToMemory(VirtualFramebuffer *vfb, bool sync = true);
// TODO: Break out into some form of FBO manager
VirtualFramebuffer *GetVFBAt(u32 addr);
VirtualFramebuffer *GetDisplayVFB() {
return GetVFBAt(displayFramebufPtr_);
}
std::vector<FramebufferInfo> GetFramebufferList();
void NotifyFramebufferCopy(u32 src, u32 dest, int size);
bool NotifyFramebufferCopy(u32 src, u32 dest, int size, bool isMemset = false);
bool NotifyStencilUpload(u32 addr, int size, bool skipZero = false);
void DestroyFramebuf(VirtualFramebuffer *vfb);
void ResizeFramebufFBO(VirtualFramebuffer *vfb, u16 w, u16 h, bool force = false);
bool GetCurrentFramebuffer(GPUDebugBuffer &buffer);
bool GetCurrentDepthbuffer(GPUDebugBuffer &buffer);
@ -93,6 +86,12 @@ protected:
virtual void ClearBuffer() override;
virtual void ClearDepthBuffer() override;
virtual void NotifyRenderFramebufferCreated(VirtualFramebuffer *vfb) override;
virtual void NotifyRenderFramebufferSwitched(VirtualFramebuffer *prevVfb, VirtualFramebuffer *vfb) override;
virtual void NotifyRenderFramebufferUpdated(VirtualFramebuffer *vfb, bool vfbFormatChanged) override;
virtual void DecimateFBOs() override;
private:
void CompileDraw2DProgram();
void DestroyDraw2DProgram();
@ -120,7 +119,6 @@ private:
std::vector<FBO *> extraFBOs_;
bool resized_;
bool useBufferedRendering_;
std::vector<VirtualFramebuffer *> bvfbs_; // blitting FBOs

397
GPU/GLES/Framebuffer.cpp
View file

@ -97,15 +97,6 @@ static const char color_vs[] =
" gl_Position = a_position;\n"
"}\n";
// Aggressively delete unused FBO:s to save gpu memory.
enum {
FBO_OLD_AGE = 5,
};
bool FramebufferManager::MaskedEqual(u32 addr1, u32 addr2) {
return (addr1 & 0x03FFFFFF) == (addr2 & 0x03FFFFFF);
}
inline u16 RGBA8888toRGB565(u32 px) {
return ((px >> 3) & 0x001F) | ((px >> 5) & 0x07E0) | ((px >> 8) & 0xF800);
}
@ -640,26 +631,6 @@ void FramebufferManager::DrawActiveTexture(GLuint texture, float x, float y, flo
glsl_unbind();
}
VirtualFramebuffer *FramebufferManager::GetVFBAt(u32 addr) {
VirtualFramebuffer *match = NULL;
for (size_t i = 0; i < vfbs_.size(); ++i) {
VirtualFramebuffer *v = vfbs_[i];
if (MaskedEqual(v->fb_address, addr)) {
// Could check w too but whatever
if (match == NULL || match->last_frame_render < v->last_frame_render) {
match = v;
}
}
}
if (match != NULL) {
return match;
}
DEBUG_LOG(SCEGE, "Finding no FBO matching address %08x", addr);
return 0;
}
void FramebufferManager::DestroyFramebuf(VirtualFramebuffer *v) {
textureCache_->NotifyFramebuffer(v->fb_address, v, NOTIFY_FB_DESTROYED);
if (v->fbo) {
@ -765,300 +736,93 @@ void FramebufferManager::ResizeFramebufFBO(VirtualFramebuffer *vfb, u16 w, u16 h
}
}
void FramebufferManager::DoSetRenderFrameBuffer() {
/*
if (useBufferedRendering_ && currentRenderVfb_) {
// Hack is enabled, and there was a previous framebuffer.
// Before we switch, let's do a series of trickery to copy one bit of stencil to
// destination alpha. Or actually, this is just a bunch of hackery attempts on Wipeout.
// Ignore for now.
glstate.depthTest.disable();
glstate.colorMask.set(GL_FALSE, GL_FALSE, GL_FALSE, GL_TRUE);
glstate.stencilTest.enable();
glstate.stencilOp.set(GL_KEEP, GL_KEEP, GL_KEEP); // don't modify stencil§
glstate.stencilFunc.set(GL_GEQUAL, 0xFE, 0xFF);
DrawPlainColor(0x00000000);
//glstate.stencilFunc.set(GL_LESS, 0x80, 0xFF);
//DrawPlainColor(0xFF000000);
glstate.stencilTest.disable();
glstate.colorMask.set(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
glstate.depthTest.disable();
glstate.colorMask.set(GL_FALSE, GL_FALSE, GL_FALSE, GL_TRUE);
DrawPlainColor(0x00000000);
shaderManager_->DirtyLastShader(); // dirty lastShader_
}
*/
gstate_c.framebufChanged = false;
// Get parameters
const u32 fb_address = gstate.getFrameBufRawAddress();
const int fb_stride = gstate.FrameBufStride();
const u32 z_address = gstate.getDepthBufRawAddress();
const int z_stride = gstate.DepthBufStride();
GEBufferFormat fmt = gstate.FrameBufFormat();
// As there are no clear "framebuffer width" and "framebuffer height" registers,
// we need to infer the size of the current framebuffer somehow.
int drawing_width, drawing_height;
EstimateDrawingSize(drawing_width, drawing_height);
gstate_c.cutRTOffsetX = 0;
bool vfbFormatChanged = false;
GEBufferFormat vfbOldFormat = GE_FORMAT_INVALID;
// Find a matching framebuffer
VirtualFramebuffer *vfb = 0;
size_t i;
for (i = 0; i < vfbs_.size(); ++i) {
VirtualFramebuffer *v = vfbs_[i];
if (v->fb_address == fb_address) {
vfb = v;
// Update fb stride in case it changed
if (vfb->fb_stride != fb_stride || vfb->format != fmt) {
vfbOldFormat = vfb->format;
vfbFormatChanged = true;
vfb->fb_stride = fb_stride;
vfb->format = fmt;
}
// In throughmode, a higher height could be used. Let's avoid shrinking the buffer.
if (gstate.isModeThrough() && (int)vfb->width < fb_stride) {
vfb->width = std::max((int)vfb->width, drawing_width);
vfb->height = std::max((int)vfb->height, drawing_height);
} else {
vfb->width = drawing_width;
vfb->height = drawing_height;
}
break;
} else if (v->fb_address < fb_address && v->fb_address + v->fb_stride * 4 > fb_address) {
// Possibly a render-to-offset.
const u32 bpp = v->format == GE_FORMAT_8888 ? 4 : 2;
const int x_offset = (fb_address - v->fb_address) / bpp;
if (v->format == fmt && v->fb_stride == fb_stride && x_offset < fb_stride && v->height >= drawing_height) {
WARN_LOG_REPORT_ONCE(renderoffset, HLE, "Rendering to framebuffer offset: %08x +%dx%d", v->fb_address, x_offset, 0);
vfb = v;
gstate_c.cutRTOffsetX = x_offset;
vfb->width = std::max((int)vfb->width, x_offset + drawing_width);
// To prevent the newSize code from being confused.
drawing_width += x_offset;
break;
}
}
void FramebufferManager::NotifyRenderFramebufferCreated(VirtualFramebuffer *vfb) {
if (!useBufferedRendering_) {
fbo_unbind();
// Let's ignore rendering to targets that have not (yet) been displayed.
gstate_c.skipDrawReason |= SKIPDRAW_NON_DISPLAYED_FB;
}
if (vfb) {
if ((drawing_width != vfb->bufferWidth || drawing_height != vfb->bufferHeight)) {
// Even if it's not newly wrong, if this is larger we need to resize up.
if (vfb->width > vfb->bufferWidth || vfb->height > vfb->bufferHeight) {
ResizeFramebufFBO(vfb, vfb->width, vfb->height);
} else if (vfb->newWidth != drawing_width || vfb->newHeight != drawing_height) {
// If it's newly wrong, or changing every frame, just keep track.
vfb->newWidth = drawing_width;
vfb->newHeight = drawing_height;
vfb->lastFrameNewSize = gpuStats.numFlips;
} else if (vfb->lastFrameNewSize + FBO_OLD_AGE < gpuStats.numFlips) {
// Okay, it's changed for a while (and stayed that way.) Let's start over.
// But only if we really need to, to avoid blinking.
bool needsRecreate = vfb->bufferWidth > fb_stride;
needsRecreate = needsRecreate || vfb->newWidth > vfb->bufferWidth || vfb->newWidth * 2 < vfb->bufferWidth;
needsRecreate = needsRecreate || vfb->newHeight > vfb->newHeight || vfb->newHeight * 2 < vfb->newHeight;
if (needsRecreate) {
ResizeFramebufFBO(vfb, vfb->width, vfb->height, true);
}
}
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_CREATED);
// Some AMD drivers crash if we don't clear the buffer first?
glDisable(GL_DITHER); // why?
ClearBuffer();
// ugly...
if (gstate_c.curRTWidth != vfb->width || gstate_c.curRTHeight != vfb->height) {
shaderManager_->DirtyUniform(DIRTY_PROJTHROUGHMATRIX);
}
}
void FramebufferManager::NotifyRenderFramebufferSwitched(VirtualFramebuffer *prevVfb, VirtualFramebuffer *vfb) {
if (ShouldDownloadFramebuffer(vfb) && !vfb->memoryUpdated) {
ReadFramebufferToMemory(vfb, true, 0, 0, vfb->width, vfb->height);
}
textureCache_->ForgetLastTexture();
if (useBufferedRendering_) {
if (vfb->fbo) {
fbo_bind_as_render_target(vfb->fbo);
} else {
// It's not different, let's keep track of that too.
vfb->lastFrameNewSize = gpuStats.numFlips;
}
}
float renderWidthFactor = (float)PSP_CoreParameter().renderWidth / 480.0f;
float renderHeightFactor = (float)PSP_CoreParameter().renderHeight / 272.0f;
if (hackForce04154000Download_ && fb_address == 0x00154000) {
renderWidthFactor = 1.0;
renderHeightFactor = 1.0;
}
// None found? Create one.
if (!vfb) {
vfb = new VirtualFramebuffer();
vfb->fbo = 0;
vfb->fb_address = fb_address;
vfb->fb_stride = fb_stride;
vfb->z_address = z_address;
vfb->z_stride = z_stride;
vfb->width = drawing_width;
vfb->height = drawing_height;
vfb->newWidth = drawing_width;
vfb->newHeight = drawing_height;
vfb->lastFrameNewSize = gpuStats.numFlips;
vfb->renderWidth = (u16)(drawing_width * renderWidthFactor);
vfb->renderHeight = (u16)(drawing_height * renderHeightFactor);
vfb->bufferWidth = drawing_width;
vfb->bufferHeight = drawing_height;
vfb->format = fmt;
vfb->drawnWidth = 0;
vfb->drawnHeight = 0;
vfb->drawnFormat = fmt;
vfb->usageFlags = FB_USAGE_RENDERTARGET;
SetColorUpdated(vfb);
vfb->depthUpdated = false;
ResizeFramebufFBO(vfb, drawing_width, drawing_height, true);
if (!useBufferedRendering_) {
// wtf? This should only happen very briefly when toggling bBufferedRendering
fbo_unbind();
// Let's ignore rendering to targets that have not (yet) been displayed.
gstate_c.skipDrawReason |= SKIPDRAW_NON_DISPLAYED_FB;
}
INFO_LOG(SCEGE, "Creating FBO for %08x : %i x %i x %i", vfb->fb_address, vfb->width, vfb->height, vfb->format);
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_CREATED);
vfb->last_frame_render = gpuStats.numFlips;
vfb->last_frame_used = 0;
vfb->last_frame_attached = 0;
frameLastFramebufUsed_ = gpuStats.numFlips;
vfbs_.push_back(vfb);
glDisable(GL_DITHER); // why?
currentRenderVfb_ = vfb;
u32 byteSize = FramebufferByteSize(vfb);
u32 fb_address_mem = (fb_address & 0x3FFFFFFF) | 0x04000000;
if (Memory::IsVRAMAddress(fb_address_mem) && fb_address_mem + byteSize > framebufRangeEnd_) {
framebufRangeEnd_ = fb_address_mem + byteSize;
}
// Some AMD drivers crash if we don't clear the buffer first?
ClearBuffer();
if (useBufferedRendering_ && !updateVRAM_ && !g_Config.bDisableSlowFramebufEffects) {
gpu->PerformMemoryUpload(fb_address_mem, byteSize);
NotifyStencilUpload(fb_address_mem, byteSize, true);
// TODO: Is it worth trying to upload the depth buffer?
}
// Let's check for depth buffer overlap. Might be interesting.
bool sharingReported = false;
bool writingDepth = true;
// Technically, it may write depth later, but we're trying to detect it only when it's really true.
if (gstate.isModeClear()) {
writingDepth = !gstate.isClearModeDepthMask() && gstate.isDepthWriteEnabled();
} else {
writingDepth = gstate.isDepthWriteEnabled();
}
for (size_t i = 0, end = vfbs_.size(); i < end; ++i) {
if (vfbs_[i]->z_stride != 0 && fb_address == vfbs_[i]->z_address) {
// If it's clearing it, most likely it just needs more video memory.
// Technically it could write something interesting and the other might not clear, but that's not likely.
if (!gstate.isModeClear() || !gstate.isClearModeColorMask() || !gstate.isClearModeAlphaMask()) {
if (fb_address != z_address && vfbs_[i]->fb_address != vfbs_[i]->z_address) {
WARN_LOG_REPORT(SCEGE, "FBO created from existing depthbuffer as color, %08x/%08x and %08x/%08x", fb_address, z_address, vfbs_[i]->fb_address, vfbs_[i]->z_address);
}
}
} else if (z_stride != 0 && z_address == vfbs_[i]->fb_address) {
// If it's clearing it, then it's probably just the reverse of the above case.
if (writingDepth) {
WARN_LOG_REPORT(SCEGE, "FBO using existing buffer as depthbuffer, %08x/%08x and %08x/%08x", fb_address, z_address, vfbs_[i]->fb_address, vfbs_[i]->z_address);
}
} else if (vfbs_[i]->z_stride != 0 && z_address == vfbs_[i]->z_address && fb_address != vfbs_[i]->fb_address && !sharingReported) {
// This happens a lot, but virtually always it's cleared.
// It's possible the other might not clear, but when every game is reported it's not useful.
if (writingDepth) {
WARN_LOG_REPORT(SCEGE, "FBO reusing depthbuffer, %08x/%08x and %08x/%08x", fb_address, z_address, vfbs_[i]->fb_address, vfbs_[i]->z_address);
sharingReported = true;
}
}
}
// We already have it!
} else if (vfb != currentRenderVfb_) {
if (ShouldDownloadFramebuffer(vfb) && !vfb->memoryUpdated) {
ReadFramebufferToMemory(vfb, true, 0, 0, vfb->width, vfb->height);
}
// Use it as a render target.
DEBUG_LOG(SCEGE, "Switching render target to FBO for %08x: %i x %i x %i ", vfb->fb_address, vfb->width, vfb->height, vfb->format);
vfb->usageFlags |= FB_USAGE_RENDERTARGET;
textureCache_->ForgetLastTexture();
vfb->last_frame_render = gpuStats.numFlips;
frameLastFramebufUsed_ = gpuStats.numFlips;
vfb->dirtyAfterDisplay = true;
if ((gstate_c.skipDrawReason & SKIPDRAW_SKIPFRAME) == 0)
vfb->reallyDirtyAfterDisplay = true;
if (useBufferedRendering_) {
if (vfb->fbo) {
fbo_bind_as_render_target(vfb->fbo);
} else {
// wtf? This should only happen very briefly when toggling bBufferedRendering
fbo_unbind();
}
} else {
if (vfb->fbo) {
// wtf? This should only happen very briefly when toggling bBufferedRendering
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_DESTROYED);
fbo_destroy(vfb->fbo);
vfb->fbo = 0;
}
fbo_unbind();
// Let's ignore rendering to targets that have not (yet) been displayed.
if (vfb->usageFlags & FB_USAGE_DISPLAYED_FRAMEBUFFER) {
gstate_c.skipDrawReason &= ~SKIPDRAW_NON_DISPLAYED_FB;
} else {
gstate_c.skipDrawReason |= SKIPDRAW_NON_DISPLAYED_FB;
}
}
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_UPDATED);
#ifdef USING_GLES2
// Some tiled mobile GPUs benefit IMMENSELY from clearing an FBO before rendering
// to it. This broke stuff before, so now it only clears on the first use of an
// FBO in a frame. This means that some games won't be able to avoid the on-some-GPUs
// performance-crushing framebuffer reloads from RAM, but we'll have to live with that.
if (vfb->last_frame_render != gpuStats.numFlips) {
ClearBuffer();
}
#endif
// Copy depth pixel value from the read framebuffer to the draw framebuffer
if (currentRenderVfb_ && !g_Config.bDisableSlowFramebufEffects) {
BlitFramebufferDepth(currentRenderVfb_, vfb);
}
currentRenderVfb_ = vfb;
if (vfbFormatChanged && vfbOldFormat != vfb->format) {
// TODO: Might ultimately combine this with the resize step above.
ReformatFramebufferFrom(vfb, vfbOldFormat);
}
} else {
if (vfbFormatChanged) {
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_UPDATED);
if (vfbOldFormat != vfb->format) {
ReformatFramebufferFrom(vfb, vfbOldFormat);
}
if (vfb->fbo) {
// wtf? This should only happen very briefly when toggling bBufferedRendering
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_DESTROYED);
fbo_destroy(vfb->fbo);
vfb->fbo = 0;
}
fbo_unbind();
vfb->last_frame_render = gpuStats.numFlips;
frameLastFramebufUsed_ = gpuStats.numFlips;
vfb->dirtyAfterDisplay = true;
if ((gstate_c.skipDrawReason & SKIPDRAW_SKIPFRAME) == 0)
vfb->reallyDirtyAfterDisplay = true;
// Let's ignore rendering to targets that have not (yet) been displayed.
if (vfb->usageFlags & FB_USAGE_DISPLAYED_FRAMEBUFFER) {
gstate_c.skipDrawReason &= ~SKIPDRAW_NON_DISPLAYED_FB;
} else {
gstate_c.skipDrawReason |= SKIPDRAW_NON_DISPLAYED_FB;
}
}
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_UPDATED);
#ifdef USING_GLES2
// Some tiled mobile GPUs benefit IMMENSELY from clearing an FBO before rendering
// to it. This broke stuff before, so now it only clears on the first use of an
// FBO in a frame. This means that some games won't be able to avoid the on-some-GPUs
// performance-crushing framebuffer reloads from RAM, but we'll have to live with that.
if (vfb->last_frame_render != gpuStats.numFlips) {
ClearBuffer();
}
#endif
// Copy depth pixel value from the read framebuffer to the draw framebuffer
if (prevVfb && !g_Config.bDisableSlowFramebufEffects) {
BlitFramebufferDepth(prevVfb, vfb);
}
if (vfb->drawnFormat != vfb->format) {
// TODO: Might ultimately combine this with the resize step in DoSetRenderFrameBuffer().
ReformatFramebufferFrom(vfb, vfb->drawnFormat);
}
// ugly...
if (gstate_c.curRTWidth != vfb->width || gstate_c.curRTHeight != vfb->height) {
shaderManager_->DirtyUniform(DIRTY_PROJTHROUGHMATRIX);
}
}
void FramebufferManager::NotifyRenderFramebufferUpdated(VirtualFramebuffer *vfb, bool vfbFormatChanged) {
if (vfbFormatChanged) {
textureCache_->NotifyFramebuffer(vfb->fb_address, vfb, NOTIFY_FB_UPDATED);
if (vfb->drawnFormat != vfb->format) {
ReformatFramebufferFrom(vfb, vfb->drawnFormat);
}
}
// ugly...
if (gstate_c.curRTWidth != vfb->width || gstate_c.curRTHeight != vfb->height) {
shaderManager_->DirtyUniform(DIRTY_PROJTHROUGHMATRIX);
gstate_c.curRTWidth = vfb->width;
gstate_c.curRTHeight = vfb->height;
}
gstate_c.curRTRenderWidth = vfb->renderWidth;
gstate_c.curRTRenderHeight = vfb->renderHeight;
}
void FramebufferManager::SetLineWidth() {
@ -1342,10 +1106,6 @@ void FramebufferManager::CopyDisplayToOutput() {
}
}
inline bool FramebufferManager::ShouldDownloadFramebuffer(const VirtualFramebuffer *vfb) const {
return updateVRAM_ || (hackForce04154000Download_ && vfb->fb_address == 0x00154000);
}
inline bool FramebufferManager::ShouldDownloadUsingCPU(const VirtualFramebuffer *vfb) const {
#ifndef USING_GLES2
bool useCPU = g_Config.iRenderingMode == FB_READFBOMEMORY_CPU;
@ -1945,13 +1705,6 @@ void FramebufferManager::DeviceLost() {
resized_ = false;
}
void FramebufferManager::BeginFrame() {
DecimateFBOs();
currentRenderVfb_ = 0;
useBufferedRendering_ = g_Config.iRenderingMode != FB_NON_BUFFERED_MODE;
updateVRAM_ = !(g_Config.iRenderingMode == FB_NON_BUFFERED_MODE || g_Config.iRenderingMode == FB_BUFFERED_MODE);
}
std::vector<FramebufferInfo> FramebufferManager::GetFramebufferList() {
std::vector<FramebufferInfo> list;
@ -2185,10 +1938,6 @@ bool FramebufferManager::NotifyFramebufferCopy(u32 src, u32 dst, int size, bool
}
}
u32 FramebufferManager::FramebufferByteSize(const VirtualFramebuffer *vfb) const {
return vfb->fb_stride * vfb->height * (vfb->format == GE_FORMAT_8888 ? 4 : 2);
}
void FramebufferManager::FindTransferFramebuffers(VirtualFramebuffer *&dstBuffer, VirtualFramebuffer *&srcBuffer, u32 dstBasePtr, int dstStride, int &dstX, int &dstY, u32 srcBasePtr, int srcStride, int &srcX, int &srcY, int &srcWidth, int &srcHeight, int &dstWidth, int &dstHeight, int bpp) const {
u32 dstYOffset = -1;
u32 dstXOffset = -1;

23
GPU/GLES/Framebuffer.h
View file

@ -83,15 +83,12 @@ public:
void DrawPlainColor(u32 color);
void DestroyAllFBOs();
void DecimateFBOs();
void Init();
void BeginFrame();
void EndFrame();
void Resized();
void DeviceLost();
void CopyDisplayToOutput();
virtual void DoSetRenderFrameBuffer() override; // Uses parameters computed from gstate
void UpdateFromMemory(u32 addr, int size, bool safe);
void SetLineWidth();
void ReformatFramebufferFrom(VirtualFramebuffer *vfb, GEBufferFormat old);
@ -110,12 +107,6 @@ public:
// Reads a rectangular subregion of a framebuffer to the right position in its backing memory.
void ReadFramebufferToMemory(VirtualFramebuffer *vfb, bool sync, int x, int y, int w, int h);
// TODO: Break out into some form of FBO manager
VirtualFramebuffer *GetVFBAt(u32 addr);
VirtualFramebuffer *GetDisplayVFB() {
return GetVFBAt(displayFramebufPtr_);
}
std::vector<FramebufferInfo> GetFramebufferList();
bool NotifyFramebufferCopy(u32 src, u32 dest, int size, bool isMemset = false);
@ -137,19 +128,21 @@ protected:
virtual void ClearBuffer() override;
virtual void ClearDepthBuffer() override;
virtual void NotifyRenderFramebufferCreated(VirtualFramebuffer *vfb) override;
virtual void NotifyRenderFramebufferSwitched(VirtualFramebuffer *prevVfb, VirtualFramebuffer *vfb) override;
virtual void NotifyRenderFramebufferUpdated(VirtualFramebuffer *vfb, bool vfbFormatChanged) override;
virtual void DecimateFBOs() override;
private:
void CompileDraw2DProgram();
void DestroyDraw2DProgram();
void FlushBeforeCopy();
void FindTransferFramebuffers(VirtualFramebuffer *&dstBuffer, VirtualFramebuffer *&srcBuffer, u32 dstBasePtr, int dstStride, int &dstX, int &dstY, u32 srcBasePtr, int srcStride, int &srcX, int &srcY, int &srcWidth, int &srcHeight, int &dstWidth, int &dstHeight, int bpp) const;
u32 FramebufferByteSize(const VirtualFramebuffer *vfb) const;
void SetNumExtraFBOs(int num);
static bool MaskedEqual(u32 addr1, u32 addr2);
inline bool ShouldDownloadFramebuffer(const VirtualFramebuffer *vfb) const;
inline bool ShouldDownloadUsingCPU(const VirtualFramebuffer *vfb) const;
// Used by ReadFramebufferToMemory and later framebuffer block copies
@ -184,12 +177,8 @@ private:
std::vector<FBO *> extraFBOs_;
bool resized_;
bool useBufferedRendering_;
bool updateVRAM_;
bool gameUsesSequentialCopies_;
bool hackForce04154000Download_;
struct TempFBO {
FBO *fbo;
int last_frame_used;

4
Windows/D3D9Base.cpp
View file

@ -60,12 +60,11 @@ bool D3D9_Init(HWND hWnd, bool windowed, std::string *error_message) {
HMODULE hD3D9 = LoadLibrary(TEXT("d3d9.dll"));
g_pfnCreate9ex = (DIRECT3DCREATE9EX)GetProcAddress(hD3D9, "Direct3DCreate9Ex");
has9Ex = (g_pfnCreate9ex != NULL);
FreeLibrary(hD3D9);
has9Ex = false;
if (has9Ex) {
HRESULT result = Direct3DCreate9Ex(D3D_SDK_VERSION, &d3dEx);
HRESULT result = g_pfnCreate9ex(D3D_SDK_VERSION, &d3dEx);
d3d = d3dEx;
if (FAILED(result)) {
ELOG("Failed to create D3D9Ex context");
@ -78,6 +77,7 @@ bool D3D9_Init(HWND hWnd, bool windowed, std::string *error_message) {
return false;
}
}
FreeLibrary(hD3D9);
D3DCAPS9 d3dCaps;