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 #16905 from hrydgard/readback-with-stretchblit

Browse source 
Depth readback with built-in stretchblit
This commit is contained in:
Henrik Rydgård 2023-02-04 11:15:34 +01:00 • committed by GitHub
commit 26b26a6ff8
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
15 changed files with 281 additions and 178 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
CMakeLists.txt
View file

@ -1499,7 +1499,7 @@ if(NOT MOBILE_DEVICE)
endif()
set(GPU_GLES
GPU/GLES/DepthBufferGLES.cpp
GPU/GLES/StencilBufferGLES.cpp
GPU/GLES/GPU_GLES.cpp
GPU/GLES/GPU_GLES.h
GPU/GLES/FragmentTestCacheGLES.cpp
@ -1580,6 +1580,7 @@ set(GPU_SOURCES
${GPU_NEON}
GPU/Common/Draw2D.cpp
GPU/Common/Draw2D.h
GPU/Common/DepthBufferCommon.cpp
GPU/Common/TextureShaderCommon.cpp
GPU/Common/TextureShaderCommon.h
GPU/Common/DepalettizeShaderCommon.cpp

189
GPU/GLES/DepthBufferGLES.cpp → GPU/Common/DepthBufferCommon.cpp
View file

@ -21,57 +21,15 @@
#include "Common/LogReporting.h"
#include "Core/ConfigValues.h"
#include "GPU/Common/GPUStateUtils.h"
#include "GPU/GLES/DrawEngineGLES.h"
#include "GPU/GLES/FramebufferManagerGLES.h"
#include "GPU/GLES/ShaderManagerGLES.h"
#include "GPU/GLES/TextureCacheGLES.h"
#include "GPU/Common/DrawEngineCommon.h"
#include "GPU/Common/FramebufferManagerCommon.h"
#include "GPU/Common/TextureCacheCommon.h"
#include "Common/GPU/ShaderWriter.h"
static const char *depth_dl_fs = R"(
#ifdef GL_ES
#ifdef GL_FRAGMENT_PRECISION_HIGH
precision highp float;
#else
precision mediump float;
#endif
#endif
#if __VERSION__ >= 130
#define varying in
#define texture2D texture
#define gl_FragColor fragColor0
out vec4 fragColor0;
#endif
varying vec2 v_texcoord;
uniform vec4 u_depthFactor;
uniform vec4 u_depthShift;
uniform vec4 u_depthTo8;
uniform sampler2D tex;
void main() {
float depth = texture2D(tex, v_texcoord).r;
// At this point, clamped maps [0, 1] to [0, 65535].
float clamped = clamp((depth + u_depthFactor.x) * u_depthFactor.y, 0.0, 1.0);
vec4 enc = u_depthShift * clamped;
enc = floor(mod(enc, 256.0)) * u_depthTo8;
// Let's ignore the bits outside 16 bit precision.
gl_FragColor = enc.yzww;
}
)";
static const char *depth_vs = R"(
#ifdef GL_ES
precision highp float;
#endif
#if __VERSION__ >= 130
#define attribute in
#define varying out
#endif
attribute vec2 a_position;
varying vec2 v_texcoord;
void main() {
v_texcoord = a_position * 2.0;
gl_Position = vec4(v_texcoord * 2.0 - vec2(1.0, 1.0), 0.0, 1.0);
}
)";
static const InputDef vs_inputs[] = {
{ "vec2", "a_position", Draw::SEM_POSITION },
};
struct DepthUB {
float u_depthFactor[4];
@ -79,12 +37,45 @@ struct DepthUB {
float u_depthTo8[4];
};
const UniformDef depthUniforms[] = {
{ "vec4", "u_depthFactor", 0 },
{ "vec4", "u_depthShift", 1},
{ "vec4", "u_depthTo8", 2},
};
const UniformBufferDesc depthUBDesc{ sizeof(DepthUB), {
{ "u_depthFactor", -1, -1, UniformType::FLOAT4, 0 },
{ "u_depthShift", -1, -1, UniformType::FLOAT4, 16 },
{ "u_depthTo8", -1, -1, UniformType::FLOAT4, 32 },
} };
static const SamplerDef samplers[] = {
{ 0, "tex" },
};
static const VaryingDef varyings[] = {
{ "vec2", "v_texcoord", Draw::SEM_TEXCOORD0, 0, "highp" },
};
void GenerateDepthDownloadFs(ShaderWriter &writer) {
writer.DeclareSamplers(samplers);
writer.BeginFSMain(depthUniforms, varyings);
writer.C(" float depth = ").SampleTexture2D("tex", "v_texcoord").C(".r; \n");
// At this point, clamped maps [0, 1] to [0, 65535].
writer.C(" float clamped = clamp((depth + u_depthFactor.x) * u_depthFactor.y, 0.0, 1.0);\n");
writer.C(" vec4 enc = u_depthShift * clamped;\n");
writer.C(" enc = floor(mod(enc, 256.0)) * u_depthTo8;\n");
writer.C(" vec4 outColor = enc.yzww;\n"); // Let's ignore the bits outside 16 bit precision.
writer.EndFSMain("outColor");
}
void GenerateDepthDownloadVs(ShaderWriter &writer) {
writer.BeginVSMain(vs_inputs, Slice<UniformDef>::empty(), varyings);
writer.C("v_texcoord = a_position * 2.0;\n");
writer.C("gl_Position = vec4(v_texcoord * 2.0 - vec2(1.0, 1.0), 0.0, 1.0);");
writer.EndVSMain(varyings);
}
static const char *stencil_dl_fs = R"(
#ifdef GL_ES
#ifdef GL_FRAGMENT_PRECISION_HIGH
@ -131,7 +122,7 @@ static bool SupportsDepthTexturing() {
return gl_extensions.ARB_texture_float;
}
static Draw::Pipeline *CreateReadbackPipeline(Draw::DrawContext *draw, const char *tag, const UniformBufferDesc *ubDesc, const char *fs, const char *fsTag, const char *vs, const char *vsTag) {
Draw::Pipeline *CreateReadbackPipeline(Draw::DrawContext *draw, const char *tag, const UniformBufferDesc *ubDesc, const char *fs, const char *fsTag, const char *vs, const char *vsTag) {
using namespace Draw;
const ShaderLanguageDesc &shaderLanguageDesc = draw->GetShaderLanguageDesc();
@ -173,7 +164,7 @@ static Draw::Pipeline *CreateReadbackPipeline(Draw::DrawContext *draw, const cha
return pipeline;
}
bool FramebufferManagerGLES::ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint16_t *pixels, int pixelsStride) {
bool FramebufferManagerCommon::ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint16_t *pixels, int pixelsStride, int destW, int destH) {
using namespace Draw;
if (!fbo) {
@ -186,33 +177,43 @@ bool FramebufferManagerGLES::ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int
}
// Pixel size always 4 here because we always request float or RGBA.
const u32 bufSize = w * h * 4;
const u32 bufSize = destW * destH * 4;
if (!convBuf_ || convBufSize_ < bufSize) {
delete[] convBuf_;
convBuf_ = new u8[bufSize];
convBufSize_ = bufSize;
}
const bool useColorPath = gl_extensions.IsGLES;
float scaleX = (float)destW / w;
float scaleY = (float)destH / h;
bool useColorPath = gl_extensions.IsGLES || scaleX != 1.0f || scaleY != 1.0f;
bool format16Bit = false;
if (useColorPath) {
if (!depthReadbackPipeline_) {
depthReadbackPipeline_ = CreateReadbackPipeline(draw_, "depth_dl", &depthUBDesc, depth_dl_fs, "depth_dl_fs", depth_vs, "depth_vs");
const ShaderLanguageDesc &shaderLanguageDesc = draw_->GetShaderLanguageDesc();
char depth_dl_fs[1024];
char depth_dl_vs[1024];
ShaderWriter fsWriter(depth_dl_fs, shaderLanguageDesc, ShaderStage::Fragment);
ShaderWriter vsWriter(depth_dl_vs, shaderLanguageDesc, ShaderStage::Vertex);
GenerateDepthDownloadFs(fsWriter);
GenerateDepthDownloadVs(vsWriter);
depthReadbackPipeline_ = CreateReadbackPipeline(draw_, "depth_dl", &depthUBDesc, depth_dl_fs, "depth_dl_fs", depth_dl_vs, "depth_dl_vs");
depthReadbackSampler_ = draw_->CreateSamplerState({});
}
shaderManager_->DirtyLastShader();
auto *blitFBO = GetTempFBO(TempFBO::COPY, fbo->Width(), fbo->Height());
draw_->BindFramebufferAsRenderTarget(blitFBO, { RPAction::DONT_CARE, RPAction::DONT_CARE, RPAction::DONT_CARE }, "ReadbackDepthbufferSync");
Draw::Viewport viewport = { 0.0f, 0.0f, (float)fbo->Width(), (float)fbo->Height(), 0.0f, 1.0f };
Draw::Viewport viewport = { 0.0f, 0.0f, (float)destW, (float)destH, 0.0f, 1.0f };
draw_->SetViewports(1, &viewport);
draw_->BindFramebufferAsTexture(fbo, TEX_SLOT_PSP_TEXTURE, FB_DEPTH_BIT, 0);
draw_->BindSamplerStates(TEX_SLOT_PSP_TEXTURE, 1, &depthReadbackSampler_);
// We must bind the program after starting the render pass.
draw_->SetScissorRect(0, 0, w, h);
draw_->SetScissorRect(0, 0, destW, destH);
draw_->BindPipeline(depthReadbackPipeline_);
DepthUB ub{};
@ -241,7 +242,7 @@ bool FramebufferManagerGLES::ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int
};
draw_->DrawUP(positions, 3);
draw_->CopyFramebufferToMemorySync(blitFBO, FB_COLOR_BIT, x, y, w, h, DataFormat::R8G8B8A8_UNORM, convBuf_, w, "ReadbackDepthbufferSync");
draw_->CopyFramebufferToMemorySync(blitFBO, FB_COLOR_BIT, x * scaleX, y * scaleY, w * scaleX, h * scaleY, DataFormat::R8G8B8A8_UNORM, convBuf_, w, "ReadbackDepthbufferSync");
textureCache_->ForgetLastTexture();
// TODO: Use 4444 so we can copy lines directly (instead of 32 -> 16 on CPU)?
@ -266,7 +267,7 @@ bool FramebufferManagerGLES::ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int
// TODO: Apply this in the shader? May have precision issues if it becomes important to match.
// We downloaded float values directly in this case.
uint16_t *dest = pixels;
const GLfloat *packedf = (GLfloat *)convBuf_;
const float *packedf = (float *)convBuf_;
DepthScaleFactors depthScale = GetDepthScaleFactors();
for (int yp = 0; yp < h; ++yp) {
for (int xp = 0; xp < w; ++xp) {
@ -287,75 +288,3 @@ bool FramebufferManagerGLES::ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int
gstate_c.Dirty(DIRTY_ALL_RENDER_STATE);
return true;
}
// Well, this is not depth, but it's depth/stencil related.
bool FramebufferManagerGLES::ReadbackStencilbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint8_t *pixels, int pixelsStride) {
using namespace Draw;
if (!fbo) {
ERROR_LOG_REPORT_ONCE(vfbfbozero, SCEGE, "ReadbackStencilbufferSync: bad fbo");
return false;
}
const bool useColorPath = gl_extensions.IsGLES;
if (!useColorPath) {
return draw_->CopyFramebufferToMemorySync(fbo, FB_STENCIL_BIT, x, y, w, h, DataFormat::S8, pixels, pixelsStride, "ReadbackStencilbufferSync");
}
// Unsupported below GLES 3.1 or without ARB_stencil_texturing.
// OES_texture_stencil8 is related, but used to specify texture data.
if ((gl_extensions.IsGLES && !gl_extensions.VersionGEThan(3, 1)) && !gl_extensions.ARB_stencil_texturing)
return false;
// Pixel size always 4 here because we always request RGBA back.
const u32 bufSize = w * h * 4;
if (!convBuf_ || convBufSize_ < bufSize) {
delete[] convBuf_;
convBuf_ = new u8[bufSize];
convBufSize_ = bufSize;
}
if (!stencilReadbackPipeline_) {
stencilReadbackPipeline_ = CreateReadbackPipeline(draw_, "stencil_dl", &depthUBDesc, stencil_dl_fs, "stencil_dl_fs", stencil_vs, "stencil_vs");
stencilReadbackSampler_ = draw_->CreateSamplerState({});
}
shaderManager_->DirtyLastShader();
auto *blitFBO = GetTempFBO(TempFBO::COPY, fbo->Width(), fbo->Height());
draw_->BindFramebufferAsRenderTarget(blitFBO, { RPAction::DONT_CARE, RPAction::DONT_CARE, RPAction::DONT_CARE }, "ReadbackStencilbufferSync");
Draw::Viewport viewport = { 0.0f, 0.0f, (float)fbo->Width(), (float)fbo->Height(), 0.0f, 1.0f };
draw_->SetViewports(1, &viewport);
draw_->BindFramebufferAsTexture(fbo, TEX_SLOT_PSP_TEXTURE, FB_STENCIL_BIT, 0);
draw_->BindSamplerStates(TEX_SLOT_PSP_TEXTURE, 1, &stencilReadbackSampler_);
// We must bind the program after starting the render pass.
draw_->SetScissorRect(0, 0, w, h);
draw_->BindPipeline(stencilReadbackPipeline_);
// Fullscreen triangle coordinates.
static const float positions[6] = {
0.0, 0.0,
1.0, 0.0,
0.0, 1.0,
};
draw_->DrawUP(positions, 3);
draw_->CopyFramebufferToMemorySync(blitFBO, FB_COLOR_BIT, x, y, w, h, DataFormat::R8G8B8A8_UNORM, convBuf_, w, "ReadbackStencilbufferSync");
textureCache_->ForgetLastTexture();
// TODO: Use 1/4 width to write all values directly and skip CPU conversion?
uint8_t *dest = pixels;
const u32_le *packed32 = (u32_le *)convBuf_;
for (int yp = 0; yp < h; ++yp) {
for (int xp = 0; xp < w; ++xp) {
dest[xp] = packed32[xp] & 0xFF;
}
dest += pixelsStride;
packed32 += w;
}
gstate_c.Dirty(DIRTY_ALL_RENDER_STATE);
return true;
}

38
GPU/Common/FramebufferManagerCommon.cpp
View file

@ -80,6 +80,7 @@ FramebufferManagerCommon::~FramebufferManagerCommon() {
bvfbs_.clear();
delete presentation_;
delete[] convBuf_;
}
void FramebufferManagerCommon::Init(int msaaLevel) {
@ -2640,17 +2641,26 @@ bool FramebufferManagerCommon::GetDepthbuffer(u32 fb_address, int fb_stride, u32
}
bool flipY = (GetGPUBackend() == GPUBackend::OPENGL && !useBufferedRendering_) ? true : false;
if (gstate_c.Use(GPU_SCALE_DEPTH_FROM_24BIT_TO_16BIT)) {
buffer.Allocate(w, h, GPU_DBG_FORMAT_FLOAT_DIV_256, flipY);
} else {
buffer.Allocate(w, h, GPU_DBG_FORMAT_FLOAT, flipY);
}
// No need to free on failure, that's the caller's job (it likely will reuse a buffer.)
bool retval = draw_->CopyFramebufferToMemorySync(vfb->fbo, Draw::FB_DEPTH_BIT, 0, 0, w, h, Draw::DataFormat::D32F, buffer.GetData(), w, "GetDepthBuffer");
if (!retval) {
// Try ReadbackDepthbufferSync, in case GLES.
bool retval;
if (true) {
// Always use ReadbackDepthbufferSync (while we debug it)
buffer.Allocate(w, h, GPU_DBG_FORMAT_16BIT, flipY);
retval = ReadbackDepthbufferSync(vfb->fbo, 0, 0, w, h, (uint16_t *)buffer.GetData(), w);
retval = ReadbackDepthbufferSync(vfb->fbo, 0, 0, w, h, (uint16_t *)buffer.GetData(), w, w, h);
} else {
// Old code
if (gstate_c.Use(GPU_SCALE_DEPTH_FROM_24BIT_TO_16BIT)) {
buffer.Allocate(w, h, GPU_DBG_FORMAT_FLOAT_DIV_256, flipY);
} else {
buffer.Allocate(w, h, GPU_DBG_FORMAT_FLOAT, flipY);
}
// No need to free on failure, that's the caller's job (it likely will reuse a buffer.)
retval = draw_->CopyFramebufferToMemorySync(vfb->fbo, Draw::FB_DEPTH_BIT, 0, 0, w, h, Draw::DataFormat::D32F, buffer.GetData(), w, "GetDepthBuffer");
if (!retval) {
// Try ReadbackDepthbufferSync, in case GLES.
buffer.Allocate(w, h, GPU_DBG_FORMAT_16BIT, flipY);
retval = ReadbackDepthbufferSync(vfb->fbo, 0, 0, w, h, (uint16_t *)buffer.GetData(), w, w, h);
}
}
// After a readback we'll have flushed and started over, need to dirty a bunch of things to be safe.
@ -2748,7 +2758,7 @@ void FramebufferManagerCommon::ReadbackFramebufferSync(VirtualFramebuffer *vfb,
if (channel == RASTER_DEPTH) {
_assert_msg_(vfb && vfb->z_address != 0 && vfb->z_stride != 0, "Depth buffer invalid");
ReadbackDepthbufferSync(vfb->fbo, x, y, w, h, (uint16_t *)destPtr, stride);
ReadbackDepthbufferSync(vfb->fbo, x, y, w, h, (uint16_t *)destPtr, stride, w, h);
} else {
draw_->CopyFramebufferToMemorySync(vfb->fbo, channel == RASTER_COLOR ? Draw::FB_COLOR_BIT : Draw::FB_DEPTH_BIT, x, y, w, h, destFormat, destPtr, stride, "ReadbackFramebufferSync");
}
@ -2760,12 +2770,6 @@ void FramebufferManagerCommon::ReadbackFramebufferSync(VirtualFramebuffer *vfb,
gpuStats.numReadbacks++;
}
bool FramebufferManagerCommon::ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint16_t *pixels, int pixelsStride) {
Draw::DataFormat destFormat = GEFormatToThin3D(GE_FORMAT_DEPTH16);
// TODO: Apply depth scale factors if we don't have depth clamp.
return draw_->CopyFramebufferToMemorySync(fbo, Draw::FB_DEPTH_BIT, x, y, w, h, destFormat, pixels, pixelsStride, "ReadbackDepthbufferSync");
}
bool FramebufferManagerCommon::ReadbackStencilbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint8_t *pixels, int pixelsStride) {
return draw_->CopyFramebufferToMemorySync(fbo, Draw::FB_DEPTH_BIT, x, y, w, h, Draw::DataFormat::S8, pixels, pixelsStride, "ReadbackStencilbufferSync");
}

6
GPU/Common/FramebufferManagerCommon.h
View file

@ -453,7 +453,7 @@ public:
protected:
virtual void ReadbackFramebufferSync(VirtualFramebuffer *vfb, int x, int y, int w, int h, RasterChannel channel);
// Used for when a shader is required, such as GLES.
virtual bool ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint16_t *pixels, int pixelsStride);
virtual bool ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint16_t *pixels, int pixelsStride, int destW, int destH);
virtual bool ReadbackStencilbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint8_t *pixels, int pixelsStride);
void SetViewport2D(int x, int y, int w, int h);
Draw::Texture *MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height);
@ -601,4 +601,8 @@ protected:
Draw2D draw2D_;
// The fragment shaders are "owned" by the pipelines since they're 1:1.
// Depth readback helper state
u8 *convBuf_ = nullptr;
u32 convBufSize_ = 0;
};

7
GPU/Directx9/FramebufferManagerDX9.cpp
View file

@ -53,7 +53,12 @@ FramebufferManagerDX9::FramebufferManagerDX9(Draw::DrawContext *draw)
FramebufferManagerDX9::~FramebufferManagerDX9() {
}
bool FramebufferManagerDX9::ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint16_t *pixels, int pixelsStride) {
bool FramebufferManagerDX9::ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint16_t *pixels, int pixelsStride, int destW, int destH) {
// Don't yet support stretched readbacks here.
if (destW != w || destH != h) {
return false;
}
// We always read the depth buffer in 24_8 format.
LPDIRECT3DTEXTURE9 tex = (LPDIRECT3DTEXTURE9)draw_->GetFramebufferAPITexture(fbo, Draw::FB_DEPTH_BIT, 0);
if (!tex)

7
GPU/Directx9/FramebufferManagerDX9.h
View file

@ -21,10 +21,6 @@
#include <d3d9.h>
// Keeps track of allocated FBOs.
// Also provides facilities for drawing and later converting raw
// pixel data.
#include "GPU/GPUCommon.h"
#include "GPU/Common/FramebufferManagerCommon.h"
@ -38,5 +34,6 @@ public:
~FramebufferManagerDX9();
protected:
bool ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint16_t *pixels, int pixelsStride) override;
// TODO: The non-color path of FramebufferManagerCommon::ReadbackDepthbufferSync seems to work just as well.
bool ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint16_t *pixels, int pixelsStride, int destW, int destH) override;
};

4
GPU/GLES/FramebufferManagerGLES.cpp
View file

@ -45,10 +45,6 @@ FramebufferManagerGLES::FramebufferManagerGLES(Draw::DrawContext *draw) :
presentation_->SetLanguage(draw_->GetShaderLanguageDesc().shaderLanguage);
}
FramebufferManagerGLES::~FramebufferManagerGLES() {
delete[] convBuf_;
}
void FramebufferManagerGLES::UpdateDownloadTempBuffer(VirtualFramebuffer *nvfb) {
_assert_msg_(nvfb->fbo, "Expecting a valid nvfb in UpdateDownloadTempBuffer");

6
GPU/GLES/FramebufferManagerGLES.h
View file

@ -29,7 +29,6 @@ class GLRProgram;
class FramebufferManagerGLES : public FramebufferManagerCommon {
public:
FramebufferManagerGLES(Draw::DrawContext *draw);
~FramebufferManagerGLES();
void NotifyDisplayResized() override;
@ -37,10 +36,5 @@ public:
protected:
void UpdateDownloadTempBuffer(VirtualFramebuffer *nvfb) override;
bool ReadbackDepthbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint16_t *pixels, int pixelsStride) override;
bool ReadbackStencilbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint8_t *pixels, int pixelsStride) override;
private:
u8 *convBuf_ = nullptr;
u32 convBufSize_ = 0;
};

170
GPU/GLES/StencilBufferGLES.cpp Normal file
View file

@ -0,0 +1,170 @@
// 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 <algorithm>
#include "Common/GPU/OpenGL/GLFeatures.h"
#include "Common/LogReporting.h"
#include "Core/ConfigValues.h"
#include "GPU/Common/GPUStateUtils.h"
#include "GPU/Common/DrawEngineCommon.h"
#include "GPU/Common/TextureCacheCommon.h"
#include "GPU/GLES/FramebufferManagerGLES.h"
#include "Common/GPU/ShaderWriter.h"
static const InputDef vs_inputs[] = {
{ "vec2", "a_position", Draw::SEM_POSITION },
};
struct DepthUB {
float u_depthFactor[4];
float u_depthShift[4];
float u_depthTo8[4];
};
const UniformDef depthUniforms[] = {
{ "vec4", "u_depthFactor", 0 },
{ "vec4", "u_depthShift", 1},
{ "vec4", "u_depthTo8", 2},
};
const UniformBufferDesc depthUBDesc{ sizeof(DepthUB), {
{ "u_depthFactor", -1, -1, UniformType::FLOAT4, 0 },
{ "u_depthShift", -1, -1, UniformType::FLOAT4, 16 },
{ "u_depthTo8", -1, -1, UniformType::FLOAT4, 32 },
} };
static const SamplerDef samplers[] = {
{ 0, "tex" },
};
static const VaryingDef varyings[] = {
{ "vec2", "v_texcoord", Draw::SEM_TEXCOORD0, 0, "highp" },
};
static const char *stencil_dl_fs = R"(
#ifdef GL_ES
#ifdef GL_FRAGMENT_PRECISION_HIGH
precision highp float;
#else
precision mediump float;
#endif
#endif
#if __VERSION__ >= 130
#define varying in
#define texture2D texture
#define gl_FragColor fragColor0
out vec4 fragColor0;
#endif
varying vec2 v_texcoord;
lowp uniform usampler2D tex;
void main() {
uint stencil = texture2D(tex, v_texcoord).r;
float scaled = float(stencil) / 255.0;
gl_FragColor = vec4(scaled, scaled, scaled, scaled);
}
)";
static const char *stencil_vs = R"(
#ifdef GL_ES
precision highp float;
#endif
#if __VERSION__ >= 130
#define attribute in
#define varying out
#endif
attribute vec2 a_position;
varying vec2 v_texcoord;
void main() {
v_texcoord = a_position * 2.0;
gl_Position = vec4(v_texcoord * 2.0 - vec2(1.0, 1.0), 0.0, 1.0);
}
)";
Draw::Pipeline *CreateReadbackPipeline(Draw::DrawContext *draw, const char *tag, const UniformBufferDesc *ubDesc, const char *fs, const char *fsTag, const char *vs, const char *vsTag);
// Well, this is not depth, but it's depth/stencil related.
bool FramebufferManagerGLES::ReadbackStencilbufferSync(Draw::Framebuffer *fbo, int x, int y, int w, int h, uint8_t *pixels, int pixelsStride) {
using namespace Draw;
if (!fbo) {
ERROR_LOG_REPORT_ONCE(vfbfbozero, SCEGE, "ReadbackStencilbufferSync: bad fbo");
return false;
}
const bool useColorPath = gl_extensions.IsGLES;
if (!useColorPath) {
return draw_->CopyFramebufferToMemorySync(fbo, FB_STENCIL_BIT, x, y, w, h, DataFormat::S8, pixels, pixelsStride, "ReadbackStencilbufferSync");
}
// Unsupported below GLES 3.1 or without ARB_stencil_texturing.
// OES_texture_stencil8 is related, but used to specify texture data.
if ((gl_extensions.IsGLES && !gl_extensions.VersionGEThan(3, 1)) && !gl_extensions.ARB_stencil_texturing)
return false;
// Pixel size always 4 here because we always request RGBA back.
const u32 bufSize = w * h * 4;
if (!convBuf_ || convBufSize_ < bufSize) {
delete[] convBuf_;
convBuf_ = new u8[bufSize];
convBufSize_ = bufSize;
}
if (!stencilReadbackPipeline_) {
stencilReadbackPipeline_ = CreateReadbackPipeline(draw_, "stencil_dl", &depthUBDesc, stencil_dl_fs, "stencil_dl_fs", stencil_vs, "stencil_vs");
stencilReadbackSampler_ = draw_->CreateSamplerState({});
}
shaderManager_->DirtyLastShader();
auto *blitFBO = GetTempFBO(TempFBO::COPY, fbo->Width(), fbo->Height());
draw_->BindFramebufferAsRenderTarget(blitFBO, { RPAction::DONT_CARE, RPAction::DONT_CARE, RPAction::DONT_CARE }, "ReadbackStencilbufferSync");
Draw::Viewport viewport = { 0.0f, 0.0f, (float)fbo->Width(), (float)fbo->Height(), 0.0f, 1.0f };
draw_->SetViewports(1, &viewport);
draw_->BindFramebufferAsTexture(fbo, TEX_SLOT_PSP_TEXTURE, FB_STENCIL_BIT, 0);
draw_->BindSamplerStates(TEX_SLOT_PSP_TEXTURE, 1, &stencilReadbackSampler_);
// We must bind the program after starting the render pass.
draw_->SetScissorRect(0, 0, w, h);
draw_->BindPipeline(stencilReadbackPipeline_);
// Fullscreen triangle coordinates.
static const float positions[6] = {
0.0, 0.0,
1.0, 0.0,
0.0, 1.0,
};
draw_->DrawUP(positions, 3);
draw_->CopyFramebufferToMemorySync(blitFBO, FB_COLOR_BIT, x, y, w, h, DataFormat::R8G8B8A8_UNORM, convBuf_, w, "ReadbackStencilbufferSync");
textureCache_->ForgetLastTexture();
// TODO: Use 1/4 width to write all values directly and skip CPU conversion?
uint8_t *dest = pixels;
const u32_le *packed32 = (u32_le *)convBuf_;
for (int yp = 0; yp < h; ++yp) {
for (int xp = 0; xp < w; ++xp) {
dest[xp] = packed32[xp] & 0xFF;
}
dest += pixelsStride;
packed32 += w;
}
gstate_c.Dirty(DIRTY_ALL_RENDER_STATE);
return true;
}

8
GPU/GPU.vcxproj
View file

@ -454,6 +454,7 @@
</ItemGroup>
<ItemGroup>
<ClCompile Include="..\ext\xbrz\xbrz.cpp" />
<ClCompile Include="Common\DepthBufferCommon.cpp" />
<ClCompile Include="Common\TextureShaderCommon.cpp" />
<ClCompile Include="Common\Draw2D.cpp" />
<ClCompile Include="Common\GeometryShaderGenerator.cpp" />
@ -525,12 +526,6 @@
<ClCompile Include="Directx9\DrawEngineDX9.cpp" />
<ClCompile Include="GeDisasm.cpp" />
<ClCompile Include="GeConstants.cpp" />
<ClCompile Include="GLES\DepthBufferGLES.cpp">
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64'">true</ExcludedFromBuild>
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">true</ExcludedFromBuild>
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|ARM64'">true</ExcludedFromBuild>
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|ARM'">true</ExcludedFromBuild>
</ClCompile>
<ClCompile Include="GLES\FragmentTestCacheGLES.cpp">
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64'">true</ExcludedFromBuild>
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">true</ExcludedFromBuild>
@ -561,6 +556,7 @@
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|ARM64'">true</ExcludedFromBuild>
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|ARM'">true</ExcludedFromBuild>
</ClCompile>
<ClCompile Include="GLES\StencilBufferGLES.cpp" />
<ClCompile Include="GLES\TextureCacheGLES.cpp">
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64'">true</ExcludedFromBuild>
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM'">true</ExcludedFromBuild>

9
GPU/GPU.vcxproj.filters
View file

@ -461,9 +461,6 @@
<ClCompile Include="Debugger\Debugger.cpp">
<Filter>Debugger</Filter>
</ClCompile>
<ClCompile Include="GLES\DepthBufferGLES.cpp">
<Filter>GLES</Filter>
</ClCompile>
<ClCompile Include="Debugger\Playback.cpp">
<Filter>Debugger</Filter>
</ClCompile>
@ -518,6 +515,12 @@
<ClCompile Include="Common\GeometryShaderGenerator.cpp">
<Filter>Common</Filter>
</ClCompile>
<ClCompile Include="Common\DepthBufferCommon.cpp">
<Filter>Common</Filter>
</ClCompile>
<ClCompile Include="GLES\StencilBufferGLES.cpp">
<Filter>GLES</Filter>
</ClCompile>
</ItemGroup>
<ItemGroup>
<FxCompile Include="..\assets\shaders\tex_4xbrz.csh">

3
UWP/GPU_UWP/GPU_UWP.vcxproj
View file

@ -441,6 +441,7 @@
<ClInclude Include="targetver.h" />
</ItemGroup>
<ItemGroup>
<ClCompile Include="..\..\GPU\Common\DepthBufferCommon.cpp" />
<ClCompile Include="..\..\GPU\Common\TextureShaderCommon.cpp" />
<ClCompile Include="..\..\GPU\Common\DepalettizeShaderCommon.cpp" />
<ClCompile Include="..\..\GPU\Common\Draw2D.cpp" />
@ -531,4 +532,4 @@
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
</ImportGroup>
</Project>
</Project>

3
UWP/GPU_UWP/GPU_UWP.vcxproj.filters
View file

@ -60,6 +60,7 @@
<ClCompile Include="..\..\GPU\Common\ReinterpretFramebuffer.cpp" />
<ClCompile Include="..\..\GPU\Common\Draw2D.cpp" />
<ClCompile Include="..\..\GPU\Common\TextureShaderCommon.cpp" />
<ClCompile Include="..\..\GPU\Common\DepthBufferCommon.cpp" />
</ItemGroup>
<ItemGroup>
<ClInclude Include="..\..\GPU\Common\DepalettizeShaderCommon.h" />
@ -123,4 +124,4 @@
<ClInclude Include="..\..\GPU\Common\Draw2D.h" />
<ClInclude Include="..\..\GPU\Common\TextureShaderCommon.h" />
</ItemGroup>
</Project>
</Project>

3
android/jni/Android.mk
View file

@ -372,6 +372,7 @@ EXEC_AND_LIB_FILES := \
$(SRC)/GPU/Common/GPUStateUtils.cpp.arm \
$(SRC)/GPU/Common/SoftwareTransformCommon.cpp.arm \
$(SRC)/GPU/Common/ReinterpretFramebuffer.cpp \
$(SRC)/GPU/Common/DepthBufferCommon.cpp \
$(SRC)/GPU/Common/VertexDecoderCommon.cpp.arm \
$(SRC)/GPU/Common/TextureCacheCommon.cpp.arm \
$(SRC)/GPU/Common/TextureScalerCommon.cpp.arm \
@ -392,7 +393,7 @@ EXEC_AND_LIB_FILES := \
$(SRC)/GPU/Debugger/Record.cpp \
$(SRC)/GPU/Debugger/Stepping.cpp \
$(SRC)/GPU/GLES/FramebufferManagerGLES.cpp \
$(SRC)/GPU/GLES/DepthBufferGLES.cpp \
$(SRC)/GPU/GLES/StencilBufferGLES.cpp \
$(SRC)/GPU/GLES/GPU_GLES.cpp.arm \
$(SRC)/GPU/GLES/TextureCacheGLES.cpp.arm \
$(SRC)/GPU/GLES/DrawEngineGLES.cpp.arm \

3
libretro/Makefile.common
View file

@ -391,6 +391,7 @@ SOURCES_CXX += \
$(GPUDIR)/Common/TextureCacheCommon.cpp \
$(GPUDIR)/Common/TextureScalerCommon.cpp \
$(GPUDIR)/Common/SoftwareTransformCommon.cpp \
$(GPUDIR)/Common/DepthBufferCommon.cpp \
$(GPUDIR)/Common/StencilCommon.cpp \
$(GPUDIR)/Software/TransformUnit.cpp \
$(GPUDIR)/Software/SoftGpu.cpp \
@ -409,7 +410,7 @@ SOURCES_CXX += \
$(GPUDIR)/Software/Rasterizer.cpp \
$(GPUDIR)/Software/RasterizerRectangle.cpp \
$(GPUDIR)/Software/RasterizerRegCache.cpp \
$(GPUDIR)/GLES/DepthBufferGLES.cpp \
$(GPUDIR)/GLES/StencilBufferGLES.cpp \
$(GPUDIR)/GLES/DrawEngineGLES.cpp \
$(GPUDIR)/GLES/GPU_GLES.cpp \
$(GPUDIR)/GLES/FragmentTestCacheGLES.cpp \