Merge pull request #11553 from hrydgard/framebuffers-outside-vram

Remove constraint that virtual framebuffers have to represent VRAM.

GPU/Common/FramebufferCommon.cpp

@@ -172,24 +172,20 @@ void FramebufferManagerCommon::SetDisplayFramebuffer(u32 framebuf, u32 stride, G
 }
 
 VirtualFramebuffer *FramebufferManagerCommon::GetVFBAt(u32 addr) {
+	addr &= 0x3FFFFFFF;
 	VirtualFramebuffer *match = nullptr;
 	for (size_t i = 0; i < vfbs_.size(); ++i) {
 		VirtualFramebuffer *v = vfbs_[i];
-		if (MaskedEqual(v->fb_address, addr)) {
+		if (v->fb_address == addr) {
 			// Could check w too but whatever
 			if (match == nullptr || match->last_frame_render < v->last_frame_render) {
 				match = v;
 			}
 		}
 	}
-
 	return match;
 }
 
-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);
 }
@@ -253,11 +249,10 @@ void FramebufferManagerCommon::EstimateDrawingSize(u32 fb_address, GEBufferForma
 
 	if (viewport_width != region_width) {
 		// The majority of the time, these are equal.  If not, let's check what we know.
-		const u32 fb_normalized_address = fb_address | 0x44000000;
 		u32 nearest_address = 0xFFFFFFFF;
 		for (size_t i = 0; i < vfbs_.size(); ++i) {
-			const u32 other_address = vfbs_[i]->fb_address | 0x44000000;
-			if (other_address > fb_normalized_address && other_address < nearest_address) {
+			const u32 other_address = vfbs_[i]->fb_address & 0x3FFFFFFF;
+			if (other_address > fb_address && other_address < nearest_address) {
 				nearest_address = other_address;
 			}
 		}
@@ -266,7 +261,7 @@ void FramebufferManagerCommon::EstimateDrawingSize(u32 fb_address, GEBufferForma
 		// This catches some cases where we can know this.
 		// Hmm.  The problem is that we could only catch it for the first of two buffers...
 		const u32 bpp = fb_format == GE_FORMAT_8888 ? 4 : 2;
-		int avail_height = (nearest_address - fb_normalized_address) / (fb_stride * bpp);
+		int avail_height = (nearest_address - fb_address) / (fb_stride * bpp);
 		if (avail_height < drawing_height && avail_height == region_height) {
 			drawing_width = std::min(region_width, fb_stride);
 			drawing_height = avail_height;
@@ -282,11 +277,10 @@ void FramebufferManagerCommon::EstimateDrawingSize(u32 fb_address, GEBufferForma
 }
 
 void GetFramebufferHeuristicInputs(FramebufferHeuristicParams *params, const GPUgstate &gstate) {
-	params->fb_addr = gstate.getFrameBufAddress();
-	params->fb_address = gstate.getFrameBufRawAddress();
+	params->fb_address = (gstate.getFrameBufRawAddress() & 0x3FFFFFFF) | 0x04000000;  // GetFramebufferHeuristicInputs is only called from rendering, and thus, it's VRAM.
 	params->fb_stride = gstate.FrameBufStride();
 
-	params->z_address = gstate.getDepthBufRawAddress();
+	params->z_address = (gstate.getDepthBufRawAddress() & 0x3FFFFFFF) | 0x04000000;
 	params->z_stride = gstate.DepthBufStride();
 
 	params->fmt = gstate.FrameBufFormat();
@@ -440,9 +434,9 @@ VirtualFramebuffer *FramebufferManagerCommon::DoSetRenderFrameBuffer(const Frame
 		SetColorUpdated(vfb, skipDrawReason);
 
 		u32 byteSize = FramebufferByteSize(vfb);
-		u32 fb_address_mem = (params.fb_address & 0x3FFFFFFF) | 0x04000000;
-		if (Memory::IsVRAMAddress(fb_address_mem) && fb_address_mem + byteSize > framebufRangeEnd_) {
-			framebufRangeEnd_ = fb_address_mem + byteSize;
+		// FB heuristics always produce an address in VRAM (this is during rendering) so we don't need to poke in the 0x04000000 flag here.
+		if (Memory::IsVRAMAddress(params.fb_address) && params.fb_address + byteSize > framebufRangeEnd_) {
+			framebufRangeEnd_ = params.fb_address + byteSize;
 		}
 
 		ResizeFramebufFBO(vfb, drawing_width, drawing_height, true);
@@ -456,8 +450,8 @@ VirtualFramebuffer *FramebufferManagerCommon::DoSetRenderFrameBuffer(const Frame
 		currentRenderVfb_ = vfb;
 
 		if (useBufferedRendering_ && !g_Config.bDisableSlowFramebufEffects) {
-			gpu->PerformMemoryUpload(fb_address_mem, byteSize);
-			NotifyStencilUpload(fb_address_mem, byteSize, true);
+			gpu->PerformMemoryUpload(params.fb_address, byteSize);
+			NotifyStencilUpload(params.fb_address, byteSize, true);
 			// TODO: Is it worth trying to upload the depth buffer?
 		}
 
@@ -680,7 +674,8 @@ void FramebufferManagerCommon::NotifyVideoUpload(u32 addr, int size, int width,
 }
 
 void FramebufferManagerCommon::UpdateFromMemory(u32 addr, int size, bool safe) {
-	addr &= ~0x40000000;
+	// Take off the uncached flag from the address. Not to be confused with the start of VRAM.
+	addr &= 0x3FFFFFFF;
 	// TODO: Could go through all FBOs, but probably not important?
 	// TODO: Could also check for inner changes, but video is most important.
 	bool isDisplayBuf = addr == DisplayFramebufAddr() || addr == PrevDisplayFramebufAddr();
@@ -691,7 +686,7 @@ void FramebufferManagerCommon::UpdateFromMemory(u32 addr, int size, bool safe) {
 
 		for (size_t i = 0; i < vfbs_.size(); ++i) {
 			VirtualFramebuffer *vfb = vfbs_[i];
-			if (MaskedEqual(vfb->fb_address, addr)) {
+			if (vfb->fb_address == addr) {
 				FlushBeforeCopy();
 
 				if (useBufferedRendering_ && vfb->fbo) {
@@ -700,7 +695,7 @@ void FramebufferManagerCommon::UpdateFromMemory(u32 addr, int size, bool safe) {
 						// If we're not rendering to it, format may be wrong.  Use displayFormat_ instead.
 						fmt = displayFormat_;
 					}
-					DrawPixels(vfb, 0, 0, Memory::GetPointer(addr | 0x04000000), fmt, vfb->fb_stride, vfb->width, vfb->height);
+					DrawPixels(vfb, 0, 0, Memory::GetPointer(addr), fmt, vfb->fb_stride, vfb->width, vfb->height);
 					SetColorUpdated(vfb, gstate_c.skipDrawReason);
 				} else {
 					INFO_LOG(FRAMEBUF, "Invalidating FBO for %08x (%i x %i x %i)", vfb->fb_address, vfb->width, vfb->height, vfb->format);
@@ -866,11 +861,13 @@ void FramebufferManagerCommon::CopyDisplayToOutput() {
 
 	VirtualFramebuffer *vfb = GetVFBAt(displayFramebufPtr_);
 	if (!vfb) {
-		// Let's search for a framebuf within this range.
-		const u32 addr = (displayFramebufPtr_ & 0x03FFFFFF) | 0x04000000;
+		// Let's search for a framebuf within this range. Note that we also look for
+		// "framebuffers" sitting in RAM so we only take off the kernel and uncached bits of the address
+		// when comparing.
+		const u32 addr = displayFramebufPtr_ & 0x3FFFFFFF;
 		for (size_t i = 0; i < vfbs_.size(); ++i) {
 			VirtualFramebuffer *v = vfbs_[i];
-			const u32 v_addr = (v->fb_address & 0x03FFFFFF) | 0x04000000;
+			const u32 v_addr = v->fb_address & 0x3FFFFFFF;
 			const u32 v_size = FramebufferByteSize(v);
 			if (addr >= v_addr && addr < v_addr + v_size) {
 				const u32 dstBpp = v->format == GE_FORMAT_8888 ? 4 : 2;
@@ -1241,7 +1238,8 @@ bool FramebufferManagerCommon::NotifyFramebufferCopy(u32 src, u32 dst, int size,
 			continue;
 		}
 
-		const u32 vfb_address = (0x04000000 | vfb->fb_address) & 0x3FFFFFFF;
+		// We only remove the kernel and uncached bits when comparing.
+		const u32 vfb_address = vfb->fb_address & 0x3FFFFFFF;
 		const u32 vfb_size = FramebufferByteSize(vfb);
 		const u32 vfb_bpp = vfb->format == GE_FORMAT_8888 ? 4 : 2;
 		const u32 vfb_byteStride = vfb->fb_stride * vfb_bpp;
@@ -1352,7 +1350,7 @@ void FramebufferManagerCommon::FindTransferFramebuffers(VirtualFramebuffer *&dst
 
 	for (size_t i = 0; i < vfbs_.size(); ++i) {
 		VirtualFramebuffer *vfb = vfbs_[i];
-		const u32 vfb_address = (0x04000000 | vfb->fb_address) & 0x3FFFFFFF;
+		const u32 vfb_address = vfb->fb_address & 0x3FFFFFFF;
 		const u32 vfb_size = FramebufferByteSize(vfb);
 		const u32 vfb_bpp = vfb->format == GE_FORMAT_8888 ? 4 : 2;
 		const u32 vfb_byteStride = vfb->fb_stride * vfb_bpp;
@@ -1915,7 +1913,7 @@ bool FramebufferManagerCommon::GetFramebuffer(u32 fb_address, int fb_stride, GEB
 
 	if (!vfb) {
 		// If there's no vfb and we're drawing there, must be memory?
-		buffer = GPUDebugBuffer(Memory::GetPointer(fb_address | 0x04000000), fb_stride, 512, format);
+		buffer = GPUDebugBuffer(Memory::GetPointer(fb_address), fb_stride, 512, format);
 		return true;
 	}
 
@@ -1969,7 +1967,7 @@ bool FramebufferManagerCommon::GetDepthbuffer(u32 fb_address, int fb_stride, u32
 
 	if (!vfb) {
 		// If there's no vfb and we're drawing there, must be memory?
-		buffer = GPUDebugBuffer(Memory::GetPointer(z_address | 0x04000000), z_stride, 512, GPU_DBG_FORMAT_16BIT);
+		buffer = GPUDebugBuffer(Memory::GetPointer(z_address), z_stride, 512, GPU_DBG_FORMAT_16BIT);
 		return true;
 	}
 
@@ -2005,7 +2003,7 @@ bool FramebufferManagerCommon::GetStencilbuffer(u32 fb_address, int fb_stride, G
 	if (!vfb) {
 		// If there's no vfb and we're drawing there, must be memory?
 		// TODO: Actually get the stencil.
-		buffer = GPUDebugBuffer(Memory::GetPointer(fb_address | 0x04000000), fb_stride, 512, GPU_DBG_FORMAT_8888);
+		buffer = GPUDebugBuffer(Memory::GetPointer(fb_address), fb_stride, 512, GPU_DBG_FORMAT_8888);
 		return true;
 	}
 
@@ -2057,7 +2055,7 @@ void FramebufferManagerCommon::PackFramebufferSync_(VirtualFramebuffer *vfb, int
 		return;
 	}
 
-	const u32 fb_address = (0x04000000) | vfb->fb_address;
+	const u32 fb_address = vfb->fb_address & 0x3FFFFFFF;
 
 	Draw::DataFormat destFormat = GEFormatToThin3D(vfb->format);
 	const int dstBpp = (int)DataFormatSizeInBytes(destFormat);

GPU/Common/FramebufferCommon.h

@@ -119,7 +119,6 @@ struct VirtualFramebuffer {
 };
 
 struct FramebufferHeuristicParams {
-	u32 fb_addr;
 	u32 fb_address;
 	int fb_stride;
 	u32 z_address;
@@ -243,10 +242,10 @@ public:
 	size_t NumVFBs() const { return vfbs_.size(); }
 
 	u32 PrevDisplayFramebufAddr() {
-		return prevDisplayFramebuf_ ? (0x04000000 | prevDisplayFramebuf_->fb_address) : 0;
+		return prevDisplayFramebuf_ ? prevDisplayFramebuf_->fb_address : 0;
 	}
 	u32 DisplayFramebufAddr() {
-		return displayFramebuf_ ? (0x04000000 | displayFramebuf_->fb_address) : 0;
+		return displayFramebuf_ ? displayFramebuf_->fb_address : 0;
 	}
 
 	u32 DisplayFramebufStride() {
@@ -332,7 +331,6 @@ protected:
 
 	void EstimateDrawingSize(u32 fb_address, GEBufferFormat fb_format, int viewport_width, int viewport_height, int region_width, int region_height, int scissor_width, int scissor_height, int fb_stride, int &drawing_width, int &drawing_height);
 	u32 FramebufferByteSize(const VirtualFramebuffer *vfb) const;
-	static bool MaskedEqual(u32 addr1, u32 addr2);
 
 	void NotifyRenderFramebufferCreated(VirtualFramebuffer *vfb);
 	void NotifyRenderFramebufferUpdated(VirtualFramebuffer *vfb, bool vfbFormatChanged);

GPU/Common/TextureCacheCommon.cpp

@@ -635,9 +635,10 @@ void TextureCacheCommon::HandleTextureChange(TexCacheEntry *const entry, const c
 }
 
 void TextureCacheCommon::NotifyFramebuffer(u32 address, VirtualFramebuffer *framebuffer, FramebufferNotification msg) {
-	// Must be in VRAM so | 0x04000000 it is.  Also, ignore memory mirrors.
+	// Mask to ignore the Z memory mirrors if the address is in VRAM.
 	// These checks are mainly to reduce scanning all textures.
-	const u32 addr = (address | 0x04000000) & 0x3F9FFFFF;
+	const u32 mirrorMask = 0x00600000;
+	const u32 addr = Memory::IsVRAMAddress(address) ? (address & ~mirrorMask) : address;
 	const u32 bpp = framebuffer->format == GE_FORMAT_8888 ? 4 : 2;
 	const u64 cacheKey = (u64)addr << 32;
 	// If it has a clut, those are the low 32 bits, so it'll be inside this range.
@@ -749,10 +750,13 @@ bool TextureCacheCommon::AttachFramebuffer(TexCacheEntry *entry, u32 address, Vi
 
 	AttachedFramebufferInfo fbInfo = { 0 };
 
-	const u64 mirrorMask = 0x00600000;
-	// Must be in VRAM so | 0x04000000 it is.  Also, ignore memory mirrors.
-	const u32 addr = (address | 0x04000000) & 0x3FFFFFFF & ~mirrorMask;
-	const u32 texaddr = ((entry->addr + texaddrOffset) & ~mirrorMask);
+	const u32 mirrorMask = 0x00600000;
+	u32 addr = address & 0x3FFFFFFF;
+	u32 texaddr = entry->addr + texaddrOffset;
+	if (entry->addr & 0x04000000) {
+		addr &= ~mirrorMask;
+		texaddr &= ~mirrorMask;
+	}
 	const bool noOffset = texaddr == addr;
 	const bool exactMatch = noOffset && entry->format < 4;
 	const u32 w = 1 << ((entry->dim >> 0) & 0xf);
@@ -990,7 +994,7 @@ void TextureCacheCommon::LoadClut(u32 clutAddr, u32 loadBytes) {
 			clutRenderOffset_ = MAX_CLUT_OFFSET;
 			for (size_t i = 0, n = fbCache_.size(); i < n; ++i) {
 				auto framebuffer = fbCache_[i];
-				const u32 fb_address = framebuffer->fb_address | 0x04000000;
+				const u32 fb_address = framebuffer->fb_address & 0x3FFFFFFF;
 				const u32 bpp = framebuffer->drawnFormat == GE_FORMAT_8888 ? 4 : 2;
 				u32 offset = clutFramebufAddr - fb_address;
 

GPU/D3D11/FramebufferManagerD3D11.cpp

@@ -687,7 +687,7 @@ void FramebufferManagerD3D11::PackDepthbuffer(VirtualFramebuffer *vfb, int x, in
 		return;
 	}
 
-	const u32 z_address = (0x04000000) | vfb->z_address;
+	const u32 z_address = vfb->z_address;
 	// TODO
 }
 

GPU/D3D11/StencilBufferD3D11.cpp

@@ -71,6 +71,7 @@ VS_OUT main(VS_IN In) {
 
 // TODO : If SV_StencilRef is available (D3D11.3) then this can be done in a single pass.
 bool FramebufferManagerD3D11::NotifyStencilUpload(u32 addr, int size, bool skipZero) {
+	addr &= 0x3FFFFFFF;
 	if (!MayIntersectFramebuffer(addr)) {
 		return false;
 	}
@@ -78,7 +79,7 @@ bool FramebufferManagerD3D11::NotifyStencilUpload(u32 addr, int size, bool skipZ
 	VirtualFramebuffer *dstBuffer = 0;
 	for (size_t i = 0; i < vfbs_.size(); ++i) {
 		VirtualFramebuffer *vfb = vfbs_[i];
-		if (MaskedEqual(vfb->fb_address, addr)) {
+		if (vfb->fb_address == addr) {
 			dstBuffer = vfb;
 		}
 	}

GPU/Directx9/FramebufferDX9.cpp

@@ -585,7 +585,7 @@ static const D3DVERTEXELEMENT9 g_FramebufferVertexElements[] = {
 			return;
 		}
 
-		const u32 fb_address = (0x04000000) | vfb->fb_address;
+		const u32 fb_address = vfb->fb_address & 0x3FFFFFFF;
 		const int dstBpp = vfb->format == GE_FORMAT_8888 ? 4 : 2;
 
 		// We always need to convert from the framebuffer native format.
@@ -627,7 +627,7 @@ static const D3DVERTEXELEMENT9 g_FramebufferVertexElements[] = {
 		}
 
 		// We always read the depth buffer in 24_8 format.
-		const u32 z_address = (0x04000000) | vfb->z_address;
+		const u32 z_address = vfb->z_address;
 
 		DEBUG_LOG(FRAMEBUF, "Reading depthbuffer to mem at %08x for vfb=%08x", z_address, vfb->fb_address);
 
@@ -732,7 +732,7 @@ static const D3DVERTEXELEMENT9 g_FramebufferVertexElements[] = {
 
 		if (!vfb) {
 			// If there's no vfb and we're drawing there, must be memory?
-			buffer = GPUDebugBuffer(Memory::GetPointer(fb_address | 0x04000000), fb_stride, 512, fb_format);
+			buffer = GPUDebugBuffer(Memory::GetPointer(fb_address), fb_stride, 512, fb_format);
 			return true;
 		}
 		LPDIRECT3DSURFACE9 renderTarget = vfb->fbo ? (LPDIRECT3DSURFACE9)draw_->GetFramebufferAPITexture(vfb->fbo, Draw::FB_COLOR_BIT | Draw::FB_SURFACE_BIT, 0) : nullptr;
@@ -809,7 +809,7 @@ static const D3DVERTEXELEMENT9 g_FramebufferVertexElements[] = {
 
 		if (!vfb) {
 			// If there's no vfb and we're drawing there, must be memory?
-			buffer = GPUDebugBuffer(Memory::GetPointer(z_address | 0x04000000), z_stride, 512, GPU_DBG_FORMAT_16BIT);
+			buffer = GPUDebugBuffer(Memory::GetPointer(z_address), z_stride, 512, GPU_DBG_FORMAT_16BIT);
 			return true;
 		}
 
@@ -847,7 +847,7 @@ static const D3DVERTEXELEMENT9 g_FramebufferVertexElements[] = {
 
 		if (!vfb) {
 			// If there's no vfb and we're drawing there, must be memory?
-			buffer = GPUDebugBuffer(Memory::GetPointer(vfb->z_address | 0x04000000), vfb->z_stride, 512, GPU_DBG_FORMAT_16BIT);
+			buffer = GPUDebugBuffer(Memory::GetPointer(vfb->z_address), vfb->z_stride, 512, GPU_DBG_FORMAT_16BIT);
 			return true;
 		}
 

GPU/Directx9/StencilBufferDX9.cpp

@@ -65,6 +65,7 @@ static const char *stencil_vs =
 "}\n";
 
 bool FramebufferManagerDX9::NotifyStencilUpload(u32 addr, int size, bool skipZero) {
+	addr &= 0x3FFFFFFF;
 	if (!MayIntersectFramebuffer(addr)) {
 		return false;
 	}
@@ -72,7 +73,7 @@ bool FramebufferManagerDX9::NotifyStencilUpload(u32 addr, int size, bool skipZer
 	VirtualFramebuffer *dstBuffer = 0;
 	for (size_t i = 0; i < vfbs_.size(); ++i) {
 		VirtualFramebuffer *vfb = vfbs_[i];
-		if (MaskedEqual(vfb->fb_address, addr)) {
+		if (vfb->fb_address == addr) {
 			dstBuffer = vfb;
 		}
 	}

GPU/GLES/DrawEngineGLES.cpp

@@ -311,7 +311,7 @@ void DrawEngineGLES::DoFlush() {
 		gstate_c.Clean(DIRTY_TEXTURE_IMAGE | DIRTY_TEXTURE_PARAMS);
 		textureNeedsApply = true;
 	} else if (gstate.getTextureAddress(0) == ((gstate.getFrameBufRawAddress() | 0x04000000) & 0x3FFFFFFF)) {
-		// This catches the case of clearing a texture.
+		// This catches the case of clearing a texture. (#10957)
 		gstate_c.Dirty(DIRTY_TEXTURE_IMAGE);
 	}
 

GPU/GLES/StencilBufferGLES.cpp

@@ -62,6 +62,7 @@ static const char *stencil_vs =
 "}\n";
 
 bool FramebufferManagerGLES::NotifyStencilUpload(u32 addr, int size, bool skipZero) {
+	addr &= 0x3FFFFFFF;
 	if (!MayIntersectFramebuffer(addr)) {
 		return false;
 	}
@@ -69,7 +70,7 @@ bool FramebufferManagerGLES::NotifyStencilUpload(u32 addr, int size, bool skipZe
 	VirtualFramebuffer *dstBuffer = 0;
 	for (size_t i = 0; i < vfbs_.size(); ++i) {
 		VirtualFramebuffer *vfb = vfbs_[i];
-		if (MaskedEqual(vfb->fb_address, addr)) {
+		if (vfb->fb_address == addr) {
 			dstBuffer = vfb;
 		}
 	}

GPU/Vulkan/StencilBufferVulkan.cpp

@@ -97,6 +97,7 @@ void main() {
 // messing about with bitplane textures and the like. Or actually, maybe not... Let's start with
 // the traditional approach.
 bool FramebufferManagerVulkan::NotifyStencilUpload(u32 addr, int size, bool skipZero) {
+	addr &= 0x3FFFFFFF;
 	if (!MayIntersectFramebuffer(addr)) {
 		return false;
 	}
@@ -104,7 +105,7 @@ bool FramebufferManagerVulkan::NotifyStencilUpload(u32 addr, int size, bool skip
 	VirtualFramebuffer *dstBuffer = 0;
 	for (size_t i = 0; i < vfbs_.size(); ++i) {
 		VirtualFramebuffer *vfb = vfbs_[i];
-		if (MaskedEqual(vfb->fb_address, addr)) {
+		if (vfb->fb_address == addr) {
 			dstBuffer = vfb;
 		}
 	}