void FramebufferManagerCommon::NotifyBlockTransferAfter(u32 dstBasePtr, int dstStride, int dstX, int dstY, u32 srcBasePtr, int srcStride, int srcX, int srcY, int width, int height, int bpp, u32 skipDrawReason) {
// A few games use this INSTEAD of actually drawing the video image to the screen, they just blast it to
// the backbuffer. Detect this and have the framebuffermanager draw the pixels.
u32 backBuffer = PrevDisplayFramebufAddr();
u32 displayBuffer = DisplayFramebufAddr();
// TODO: Is this not handled by upload? Should we check !dstBuffer to avoid a double copy?
if (((backBuffer != 0 && dstBasePtr == backBuffer) ||
(displayBuffer != 0 && dstBasePtr == displayBuffer)) &&
dstStride == 512 && height == 272 && !useBufferedRendering_) {
FlushBeforeCopy();
DrawFramebufferToOutput(Memory::GetPointerUnchecked(dstBasePtr), displayFormat_, 512, false);
}
if (MayIntersectFramebuffer(srcBasePtr) || MayIntersectFramebuffer(dstBasePtr)) {
VirtualFramebuffer *dstBuffer = 0;
VirtualFramebuffer *srcBuffer = 0;
int srcWidth = width;
int srcHeight = height;
int dstWidth = width;
int dstHeight = height;
FindTransferFramebuffers(dstBuffer, srcBuffer, dstBasePtr, dstStride, dstX, dstY, srcBasePtr, srcStride, srcX, srcY, srcWidth, srcHeight, dstWidth, dstHeight, bpp);
if (!useBufferedRendering_ && currentRenderVfb_ != dstBuffer) {
return;
}
if (dstBuffer && !srcBuffer) {
WARN_LOG_ONCE(btu, G3D, "Block transfer upload %08x -> %08x", srcBasePtr, dstBasePtr);
if (g_Config.bBlockTransferGPU) {
FlushBeforeCopy();
const u8 *srcBase = Memory::GetPointerUnchecked(srcBasePtr) + (srcX + srcY * srcStride) * bpp;
int dstBpp = dstBuffer->format == GE_FORMAT_8888 ? 4 : 2;
float dstXFactor = (float)bpp / dstBpp;
if (dstWidth > dstBuffer->width || dstHeight > dstBuffer->height) {
// The buffer isn't big enough, and we have a clear hint of size. Resize.
// This happens in Valkyrie Profile when uploading video at the ending.
ResizeFramebufFBO(dstBuffer, dstWidth, dstHeight, false, true);
}
DrawPixels(dstBuffer, static_cast<int>(dstX * dstXFactor), dstY, srcBase, dstBuffer->format, static_cast<int>(srcStride * dstXFactor), static_cast<int>(dstWidth * dstXFactor), dstHeight);
SetColorUpdated(dstBuffer, skipDrawReason);
RebindFramebuffer();
}
}
}
}
bool FramebufferManagerCommon::NotifyBlockTransferBefore(u32 dstBasePtr, int dstStride, int dstX, int dstY, u32 srcBasePtr, int srcStride, int srcX, int srcY, int width, int height, int bpp) {
if (!useBufferedRendering_ || updateVRAM_) {
return false;
}
// Skip checking if there's no framebuffers in that area.
if (!MayIntersectFramebuffer(srcBasePtr) && !MayIntersectFramebuffer(dstBasePtr)) {
return false;
}
VirtualFramebuffer *dstBuffer = 0;
VirtualFramebuffer *srcBuffer = 0;
int srcWidth = width;
int srcHeight = height;
int dstWidth = width;
int dstHeight = height;
FindTransferFramebuffers(dstBuffer, srcBuffer, dstBasePtr, dstStride, dstX, dstY, srcBasePtr, srcStride, srcX, srcY, srcWidth, srcHeight, dstWidth, dstHeight, bpp);
if (dstBuffer && srcBuffer) {
if (srcBuffer == dstBuffer) {
if (srcX != dstX || srcY != dstY) {
WARN_LOG_ONCE(dstsrc, G3D, "Intra-buffer block transfer %08x -> %08x", srcBasePtr, dstBasePtr);
if (g_Config.bBlockTransferGPU) {
FlushBeforeCopy();
BlitFramebuffer(dstBuffer, dstX, dstY, srcBuffer, srcX, srcY, dstWidth, dstHeight, bpp);
RebindFramebuffer();
SetColorUpdated(dstBuffer);
return true;
}
} else {
// Ignore, nothing to do. Tales of Phantasia X does this by accident.
if (g_Config.bBlockTransferGPU) {
return true;
}
}
} else {
WARN_LOG_ONCE(dstnotsrc, G3D, "Inter-buffer block transfer %08x -> %08x", srcBasePtr, dstBasePtr);
// Just do the blit!
if (g_Config.bBlockTransferGPU) {
FlushBeforeCopy();
BlitFramebuffer(dstBuffer, dstX, dstY, srcBuffer, srcX, srcY, dstWidth, dstHeight, bpp);
RebindFramebuffer();
SetColorUpdated(dstBuffer);
return true; // No need to actually do the memory copy behind, probably.
}
}
return false;
} else if (dstBuffer) {
// Here we should just draw the pixels into the buffer. Copy first.
return false;
} else if (srcBuffer) {
WARN_LOG_ONCE(btd, G3D, "Block transfer download %08x -> %08x", srcBasePtr, dstBasePtr);
FlushBeforeCopy();
if (g_Config.bBlockTransferGPU && !srcBuffer->memoryUpdated) {
const int srcBpp = srcBuffer->format == GE_FORMAT_8888 ? 4 : 2;
const float srcXFactor = (float)bpp / srcBpp;
const bool tooTall = srcY + srcHeight > srcBuffer->bufferHeight;
if (srcHeight <= 0 || (tooTall && srcY != 0)) {
WARN_LOG_ONCE(btdheight, G3D, "Block transfer download %08x -> %08x skipped, %d+%d is taller than %d", srcBasePtr, dstBasePtr, srcY, srcHeight, srcBuffer->bufferHeight);
} else {
if (tooTall)
WARN_LOG_ONCE(btdheight, G3D, "Block transfer download %08x -> %08x dangerous, %d+%d is taller than %d", srcBasePtr, dstBasePtr, srcY, srcHeight, srcBuffer->bufferHeight);
ReadFramebufferToMemory(srcBuffer, true, static_cast<int>(srcX * srcXFactor), srcY, static_cast<int>(srcWidth * srcXFactor), srcHeight);
}
}
return false; // Let the bit copy happen
} else {
return false;
}
}
bool FramebufferManagerGLES::NotifyStencilUpload(u32 addr, int size, bool skipZero) {
if (!MayIntersectFramebuffer(addr)) {
return false;
}
VirtualFramebuffer *dstBuffer = 0;
for (size_t i = 0; i < vfbs_.size(); ++i) {
VirtualFramebuffer *vfb = vfbs_[i];
if (MaskedEqual(vfb->fb_address, addr)) {
dstBuffer = vfb;
}
}
if (!dstBuffer) {
return false;
}
int values = 0;
u8 usedBits = 0;
const u8 *src = Memory::GetPointer(addr);
if (!src)
return false;
switch (dstBuffer->format) {
case GE_FORMAT_565:
// Well, this doesn't make much sense.
return false;
case GE_FORMAT_5551:
usedBits = StencilBits5551(src, dstBuffer->fb_stride * dstBuffer->bufferHeight);
values = 2;
break;
case GE_FORMAT_4444:
usedBits = StencilBits4444(src, dstBuffer->fb_stride * dstBuffer->bufferHeight);
values = 16;
break;
case GE_FORMAT_8888:
usedBits = StencilBits8888(src, dstBuffer->fb_stride * dstBuffer->bufferHeight);
values = 256;
break;
case GE_FORMAT_INVALID:
// Impossible.
break;
}
if (usedBits == 0) {
if (skipZero) {
// Common when creating buffers, it's already 0. We're done.
return false;
}
// Let's not bother with the shader if it's just zero.
if (dstBuffer->fbo) {
draw_->BindFramebufferAsRenderTarget(dstBuffer->fbo, { Draw::RPAction::KEEP, Draw::RPAction::KEEP, Draw::RPAction::CLEAR });
}
render_->Clear(0, 0, 0, GL_STENCIL_BUFFER_BIT | GL_COLOR_BUFFER_BIT, 0x8, 0, 0, 0, 0);
gstate_c.Dirty(DIRTY_BLEND_STATE | DIRTY_VIEWPORTSCISSOR_STATE);
return true;
}
if (!stencilUploadProgram_) {
std::string errorString;
static std::string vs_code, fs_code;
vs_code = ApplyGLSLPrelude(stencil_vs, GL_VERTEX_SHADER);
fs_code = ApplyGLSLPrelude(stencil_fs, GL_FRAGMENT_SHADER);
std::vector<GLRShader *> shaders;
shaders.push_back(render_->CreateShader(GL_VERTEX_SHADER, vs_code, "stencil"));
shaders.push_back(render_->CreateShader(GL_FRAGMENT_SHADER, fs_code, "stencil"));
std::vector<GLRProgram::UniformLocQuery> queries;
queries.push_back({ &u_stencilUploadTex, "tex" });
queries.push_back({ &u_stencilValue, "u_stencilValue" });
std::vector<GLRProgram::Initializer> inits;
inits.push_back({ &u_stencilUploadTex, 0, 0 });
stencilUploadProgram_ = render_->CreateProgram(shaders, {}, queries, inits, false);
for (auto iter : shaders) {
render_->DeleteShader(iter);
}
if (!stencilUploadProgram_) {
ERROR_LOG_REPORT(G3D, "Failed to compile stencilUploadProgram! This shouldn't happen.\n%s", errorString.c_str());
}
}
shaderManagerGL_->DirtyLastShader();
bool useBlit = gstate_c.Supports(GPU_SUPPORTS_ARB_FRAMEBUFFER_BLIT | GPU_SUPPORTS_NV_FRAMEBUFFER_BLIT);
// Our fragment shader (and discard) is slow. Since the source is 1x, we can stencil to 1x.
// Then after we're done, we'll just blit it across and stretch it there.
if (dstBuffer->bufferWidth == dstBuffer->renderWidth || !dstBuffer->fbo) {
useBlit = false;
}
u16 w = useBlit ? dstBuffer->bufferWidth : dstBuffer->renderWidth;
u16 h = useBlit ? dstBuffer->bufferHeight : dstBuffer->renderHeight;
Draw::Framebuffer *blitFBO = nullptr;
if (useBlit) {
blitFBO = GetTempFBO(w, h, Draw::FBO_8888);
draw_->BindFramebufferAsRenderTarget(blitFBO, { Draw::RPAction::DONT_CARE, Draw::RPAction::DONT_CARE, Draw::RPAction::DONT_CARE });
} else if (dstBuffer->fbo) {
draw_->BindFramebufferAsRenderTarget(dstBuffer->fbo, { Draw::RPAction::KEEP, Draw::RPAction::KEEP, Draw::RPAction::CLEAR });
}
render_->SetViewport({ 0, 0, (float)w, (float)h, 0.0f, 1.0f });
float u1 = 1.0f;
float v1 = 1.0f;
MakePixelTexture(src, dstBuffer->format, dstBuffer->fb_stride, dstBuffer->bufferWidth, dstBuffer->bufferHeight, u1, v1);
textureCacheGL_->ForgetLastTexture();
// We must bind the program after starting the render pass, and set the color mask after clearing.
render_->SetDepth(false, false, GL_ALWAYS);
render_->Clear(0, 0, 0, GL_COLOR_BUFFER_BIT | GL_STENCIL_BUFFER_BIT, 0x8, 0, 0, 0, 0);
render_->SetStencilFunc(GL_TRUE, GL_ALWAYS, 0xFF, 0xFF);
render_->SetRaster(false, GL_CCW, GL_FRONT, GL_FALSE);
render_->BindProgram(stencilUploadProgram_);
render_->SetNoBlendAndMask(0x8);
for (int i = 1; i < values; i += i) {
if (!(usedBits & i)) {
// It's already zero, let's skip it.
continue;
}
if (dstBuffer->format == GE_FORMAT_4444) {
render_->SetStencilOp((i << 4) | i, GL_REPLACE, GL_REPLACE, GL_REPLACE);
render_->SetUniformF1(&u_stencilValue, i * (16.0f / 255.0f));
} else if (dstBuffer->format == GE_FORMAT_5551) {
render_->SetStencilOp(0xFF, GL_REPLACE, GL_REPLACE, GL_REPLACE);
render_->SetUniformF1(&u_stencilValue, i * (128.0f / 255.0f));
} else {
render_->SetStencilOp(i, GL_REPLACE, GL_REPLACE, GL_REPLACE);
render_->SetUniformF1(&u_stencilValue, i * (1.0f / 255.0f));
}
DrawActiveTexture(0, 0, dstBuffer->width, dstBuffer->height, dstBuffer->bufferWidth, dstBuffer->bufferHeight, 0.0f, 0.0f, u1, v1, ROTATION_LOCKED_HORIZONTAL, DRAWTEX_NEAREST | DRAWTEX_KEEP_STENCIL_ALPHA);
}
if (useBlit) {
draw_->BlitFramebuffer(blitFBO, 0, 0, w, h, dstBuffer->fbo, 0, 0, dstBuffer->renderWidth, dstBuffer->renderHeight, Draw::FB_STENCIL_BIT, Draw::FB_BLIT_NEAREST);
}
gstate_c.Dirty(DIRTY_BLEND_STATE | DIRTY_RASTER_STATE | DIRTY_DEPTHSTENCIL_STATE | DIRTY_VIEWPORTSCISSOR_STATE);
RebindFramebuffer();
return true;
}