Postshader: Let shaders use the previous frame

Common/GPU/D3D11/thin3d_d3d11.cpp

@@ -26,6 +26,8 @@
 
 namespace Draw {
 
+static constexpr int MAX_BOUND_TEXTURES = 8;
+
 // A problem is that we can't get the D3Dcompiler.dll without using a later SDK than 7.1, which was the last that
 // supported XP. A possible solution might be here:
 // https://tedwvc.wordpress.com/2014/01/01/how-to-target-xp-with-vc2012-or-vc2013-and-continue-to-use-the-windows-8-x-sdk/
@@ -1162,6 +1164,7 @@ void D3D11DrawContext::UpdateBuffer(Buffer *buffer, const uint8_t *data, size_t
 }
 
 void D3D11DrawContext::BindVertexBuffers(int start, int count, Buffer **buffers, int *offsets) {
+	_assert_(start + count <= ARRAY_SIZE(nextVertexBuffers_));
 	// Lazy application
 	for (int i = 0; i < count; i++) {
 		D3D11Buffer *buf = (D3D11Buffer *)buffers[i];
@@ -1329,6 +1332,7 @@ Framebuffer *D3D11DrawContext::CreateFramebuffer(const FramebufferDesc &desc) {
 void D3D11DrawContext::BindTextures(int start, int count, Texture **textures) {
 	// Collect the resource views from the textures.
 	ID3D11ShaderResourceView *views[8];
+	_assert_(start + count <= ARRAY_SIZE(views));
 	for (int i = 0; i < count; i++) {
 		D3D11Texture *tex = (D3D11Texture *)textures[i];
 		views[i] = tex ? tex->view : nullptr;
@@ -1338,6 +1342,7 @@ void D3D11DrawContext::BindTextures(int start, int count, Texture **textures) {
 
 void D3D11DrawContext::BindSamplerStates(int start, int count, SamplerState **states) {
 	ID3D11SamplerState *samplers[8];
+	_assert_(start + count <= ARRAY_SIZE(samplers));
 	for (int i = 0; i < count; i++) {
 		D3D11SamplerState *samp = (D3D11SamplerState *)states[i];
 		samplers[i] = samp->ss;
@@ -1613,6 +1618,7 @@ void D3D11DrawContext::BindFramebufferAsRenderTarget(Framebuffer *fbo, const Ren
 }
 
 void D3D11DrawContext::BindFramebufferAsTexture(Framebuffer *fbo, int binding, FBChannel channelBit, int attachment) {
+	_assert_(binding < MAX_BOUND_TEXTURES);
 	D3D11Framebuffer *fb = (D3D11Framebuffer *)fbo;
 	switch (channelBit) {
 	case FBChannel::FB_COLOR_BIT:

Common/GPU/D3D9/thin3d_d3d9.cpp

@@ -28,6 +28,8 @@
 
 namespace Draw {
 
+static constexpr int MAX_BOUND_TEXTURES = 8;
+
 // Could be declared as u8
 static const D3DCMPFUNC compareToD3D9[] = {
 	D3DCMP_NEVER,
@@ -525,12 +527,14 @@ class D3D9Context : public DrawContext {
 
 	void BindTextures(int start, int count, Texture **textures) override;
 	void BindSamplerStates(int start, int count, SamplerState **states) override {
+		_assert_(start + count <= MAX_BOUND_TEXTURES);
 		for (int i = 0; i < count; ++i) {
 			D3D9SamplerState *s = static_cast<D3D9SamplerState *>(states[i]);
 			s->Apply(device_, start + i);
 		}
 	}
 	void BindVertexBuffers(int start, int count, Buffer **buffers, int *offsets) override {
+		_assert_(start + count <= ARRAY_SIZE(curVBuffers_));
 		for (int i = 0; i < count; i++) {
 			curVBuffers_[i + start] = (D3D9Buffer *)buffers[i];
 			curVBufferOffsets_[i + start] = offsets ? offsets[i] : 0;
@@ -785,6 +789,7 @@ Texture *D3D9Context::CreateTexture(const TextureDesc &desc) {
 }
 
 void D3D9Context::BindTextures(int start, int count, Texture **textures) {
+	_assert_(start + count <= MAX_BOUND_TEXTURES);
 	for (int i = start; i < start + count; i++) {
 		D3D9Texture *tex = static_cast<D3D9Texture *>(textures[i - start]);
 		if (tex) {
@@ -1182,6 +1187,7 @@ uintptr_t D3D9Context::GetFramebufferAPITexture(Framebuffer *fbo, int channelBit
 }
 
 void D3D9Context::BindFramebufferAsTexture(Framebuffer *fbo, int binding, FBChannel channelBit, int color) {
+	_assert_(binding < MAX_BOUND_TEXTURES);
 	D3D9Framebuffer *fb = (D3D9Framebuffer *)fbo;
 	switch (channelBit) {
 	case FB_DEPTH_BIT:

Common/GPU/OpenGL/thin3d_gl.cpp

@@ -368,9 +368,7 @@ class OpenGLContext : public DrawContext {
 	void GetFramebufferDimensions(Framebuffer *fbo, int *w, int *h) override;
 
 	void BindSamplerStates(int start, int count, SamplerState **states) override {
-		if (start + count > MAX_TEXTURE_SLOTS) {
-			return;
-		}
+		_assert_(start + count <= MAX_TEXTURE_SLOTS);
 		for (int i = 0; i < count; i++) {
 			int index = i + start;
 			boundSamplers_[index] = static_cast<OpenGLSamplerState *>(states[i]);
@@ -402,6 +400,7 @@ class OpenGLContext : public DrawContext {
 	void BindTextures(int start, int count, Texture **textures) override;
 	void BindPipeline(Pipeline *pipeline) override;
 	void BindVertexBuffers(int start, int count, Buffer **buffers, int *offsets) override {
+		_assert_(start + count <= ARRAY_SIZE(curVBuffers_));
 		for (int i = 0; i < count; i++) {
 			curVBuffers_[i + start] = (OpenGLBuffer *)buffers[i];
 			curVBufferOffsets_[i + start] = offsets ? offsets[i] : 0;
@@ -1070,9 +1069,7 @@ Pipeline *OpenGLContext::CreateGraphicsPipeline(const PipelineDesc &desc) {
 }
 
 void OpenGLContext::BindTextures(int start, int count, Texture **textures) {
-	if (start + count > MAX_TEXTURE_SLOTS) {
-		return;
-	}
+	_assert_(start + count <= MAX_TEXTURE_SLOTS);
 	for (int i = start; i < start + count; i++) {
 		OpenGLTexture *glTex = static_cast<OpenGLTexture *>(textures[i - start]);
 		if (!glTex) {
@@ -1151,12 +1148,14 @@ bool OpenGLPipeline::LinkShaders() {
 	std::vector<GLRProgram::UniformLocQuery> queries;
 	queries.push_back({ &samplerLocs_[0], "sampler0" });
 	queries.push_back({ &samplerLocs_[1], "sampler1" });
+	queries.push_back({ &samplerLocs_[2], "sampler2" });
 	for (size_t i = 0; i < dynamicUniforms.uniforms.size(); ++i) {
 		queries.push_back({ &dynamicUniformLocs_[i], dynamicUniforms.uniforms[i].name });
 	}
 	std::vector<GLRProgram::Initializer> initialize;
 	initialize.push_back({ &samplerLocs_[0], 0, 0 });
 	initialize.push_back({ &samplerLocs_[1], 0, 1 });
+	initialize.push_back({ &samplerLocs_[2], 0, 2 });
 	program_ = render_->CreateProgram(linkShaders, semantics, queries, initialize, false);
 	return true;
 }
@@ -1354,6 +1353,7 @@ bool OpenGLContext::BlitFramebuffer(Framebuffer *fbsrc, int srcX1, int srcY1, in
 
 void OpenGLContext::BindFramebufferAsTexture(Framebuffer *fbo, int binding, FBChannel channelBit, int color) {
 	OpenGLFramebuffer *fb = (OpenGLFramebuffer *)fbo;
+	_assert_(binding < MAX_TEXTURE_SLOTS);
 
 	GLuint aspect = 0;
 	if (channelBit & FB_COLOR_BIT) {

Common/GPU/ShaderTranslation.cpp

@@ -85,6 +85,7 @@ cbuffer data : register(b0) {
 	float2 u_texelDelta;
 	float2 u_pixelDelta;
 	float4 u_time;
+	float4 u_timeDelta;
 	float4 u_setting;
 	float u_video;
 };
@@ -101,6 +102,7 @@ layout (std140, set = 0, binding = 0) uniform Data {
 	vec2 u_texelDelta;
 	vec2 u_pixelDelta;
 	vec4 u_time;
+	vec4 u_timeDelta;
 	vec4 u_setting;
 	float u_video;
 };
@@ -111,8 +113,9 @@ float4 gl_HalfPixel : register(c0);
 float2 u_texelDelta : register(c1);
 float2 u_pixelDelta : register(c2);
 float4 u_time : register(c3);
-float4 u_setting : register(c4);
-float u_video : register(c5);
+float4 u_timeDelta : register(c4);
+float4 u_setting : register(c5);
+float u_video : register(c6);
 )";
 
 // SPIRV-Cross' HLSL output has some deficiencies we need to work around.
@@ -135,12 +138,9 @@ std::string Postprocess(std::string code, ShaderLanguage lang, ShaderStage stage
 	std::string line;
 	std::stringstream instream(code);
 	while (std::getline(instream, line)) {
-		if (line == "uniform sampler2D sampler0;" && lang == HLSL_D3D9) {
-			out << "sampler2D sampler0 : register(s0);\n";
-			continue;
-		}
-		if (line == "uniform sampler2D sampler1;" && lang == HLSL_D3D9) {
-			out << "sampler2D sampler1 : register(s1);\n";
+		int num;
+		if (lang == HLSL_D3D9 && sscanf(line.c_str(), "uniform sampler2D sampler%d;", &num) == 1) {
+			out << "sampler2D sampler" << num << " : register(s" << num << ");\n";
 			continue;
 		}
 		if (line.find("uniform float") != std::string::npos) {
@@ -184,7 +184,9 @@ bool ConvertToVulkanGLSL(std::string *dest, TranslatedShaderMetadata *destMetada
 		if (line.find("uniform bool") != std::string::npos) {
 			continue;
 		} else if (line.find("uniform sampler2D") == 0) {
-			if (line.find("sampler0") != line.npos)
+			if (sscanf(line.c_str(), "uniform sampler2D sampler%d", &num) == 1)
+				line = StringFromFormat("layout(set = 0, binding = %d) ", num + 1) + line;
+			else if (line.find("sampler0") != line.npos)
 				line = "layout(set = 0, binding = 1) " + line;
 			else
 				line = "layout(set = 0, binding = 2) " + line;
@@ -299,8 +301,11 @@ bool TranslateShader(std::string *dest, ShaderLanguage destLang, const ShaderLan
 
 		int i = 0;
 		for (auto &resource : resources.sampled_images) {
-			// int location = hlsl.get_decoration(resource.id, spv::DecorationLocation);
-			hlsl.set_decoration(resource.id, spv::DecorationLocation, i);
+			const std::string &name = hlsl.get_name(resource.id);
+			int num;
+			if (sscanf(name.c_str(), "sampler%d", &num) != 1)
+				num = i;
+			hlsl.set_decoration(resource.id, spv::DecorationBinding, num);
 			i++;
 		}
 		spirv_cross::CompilerHLSL::Options options{};

Common/GPU/Vulkan/thin3d_vulkan.cpp

@@ -295,7 +295,7 @@ class VKBuffer;
 class VKSamplerState;
 
 enum {
-	MAX_BOUND_TEXTURES = 2
+	MAX_BOUND_TEXTURES = MAX_TEXTURE_SLOTS,
 };
 
 struct DescriptorSetKey {
@@ -416,6 +416,7 @@ class VKContext : public DrawContext {
 
 	// TODO: Make VKBuffers proper buffers, and do a proper binding model. This is just silly.
 	void BindVertexBuffers(int start, int count, Buffer **buffers, int *offsets) override {
+		_assert_(start + count <= ARRAY_SIZE(curVBuffers_));
 		for (int i = 0; i < count; i++) {
 			curVBuffers_[i + start] = (VKBuffer *)buffers[i];
 			curVBufferOffsets_[i + start] = offsets ? offsets[i] : 0;
@@ -689,6 +690,7 @@ RasterState *VKContext::CreateRasterState(const RasterStateDesc &desc) {
 }
 
 void VKContext::BindSamplerStates(int start, int count, SamplerState **state) {
+	_assert_(start + count <= MAX_BOUND_TEXTURES);
 	for (int i = start; i < start + count; i++) {
 		boundSamplers_[i] = (VKSamplerState *)state[i - start];
 	}
@@ -1271,6 +1273,7 @@ void VKContext::UpdateBuffer(Buffer *buffer, const uint8_t *data, size_t offset,
 }
 
 void VKContext::BindTextures(int start, int count, Texture **textures) {
+	_assert_(start + count <= MAX_BOUND_TEXTURES);
 	for (int i = start; i < start + count; i++) {
 		boundTextures_[i] = static_cast<VKTexture *>(textures[i - start]);
 		boundImageView_[i] = boundTextures_[i] ? boundTextures_[i]->GetImageView() : GetNullTexture()->GetImageView();
@@ -1544,6 +1547,7 @@ void VKContext::BindFramebufferAsRenderTarget(Framebuffer *fbo, const RenderPass
 
 void VKContext::BindFramebufferAsTexture(Framebuffer *fbo, int binding, FBChannel channelBit, int attachment) {
 	VKFramebuffer *fb = (VKFramebuffer *)fbo;
+	_assert_(binding < MAX_BOUND_TEXTURES);
 
 	// TODO: There are cases where this is okay, actually. But requires layout transitions and stuff -
 	// we're not ready for this.

Common/GPU/thin3d.h

@@ -278,7 +278,7 @@ enum class Event {
 	PRESENTED,
 };
 
-constexpr uint32_t MAX_TEXTURE_SLOTS = 2;
+constexpr uint32_t MAX_TEXTURE_SLOTS = 3;
 
 struct FramebufferDesc {
 	int width;

Core/Font/PGF.cpp

@@ -230,14 +230,14 @@ bool PGF::ReadPtr(const u8 *ptr, size_t dataSize) {
 
 	const u8 *uptr = (const u8 *)wptr;
 
-	if (uptr >= startPtr + dataSize) {
-		return false;
-	}
-
 	int shadowCharMapSize = ((header.shadowMapLength * header.shadowMapBpe + 31) & ~31) / 8;
 	const u8 *shadowCharMap = uptr;
 	uptr += shadowCharMapSize;
 
+	if (uptr < startPtr || uptr >= startPtr + dataSize) {
+		return false;
+	}
+
 	const u16_le *sptr = (const u16_le *)uptr;
 	if (header.revision == 3) {
 		charmapCompressionTable1[0].resize(rev3extra.compCharMapLength1);
@@ -257,10 +257,6 @@ bool PGF::ReadPtr(const u8 *ptr, size_t dataSize) {
 
 	uptr = (const u8 *)sptr;
 
-	if (uptr >= startPtr + dataSize) {
-		return false;
-	}
-
 	int charMapSize = ((header.charMapLength * header.charMapBpe + 31) & ~31) / 8;
 	const u8 *charMap = uptr;
 	uptr += charMapSize;
@@ -269,6 +265,10 @@ bool PGF::ReadPtr(const u8 *ptr, size_t dataSize) {
 	const u8 *charPointerTable = uptr;
 	uptr += charPointerSize;
 
+	if (uptr < startPtr || uptr >= startPtr + dataSize) {
+		return false;
+	}
+
 	// PGF Fontdata.
 	u32 fontDataOffset = (u32)(uptr - startPtr);
 

GPU/Common/PostShader.cpp

@@ -127,6 +127,7 @@ void LoadPostShaderInfo(const std::vector<Path> &directories) {
 					section.Get("Upscaling", &info.isUpscalingFilter, false);
 					section.Get("SSAA", &info.SSAAFilterLevel, 0);
 					section.Get("60fps", &info.requires60fps, false);
+					section.Get("UsePreviousFrame", &info.usePreviousFrame, false);
 
 					if (info.parent == "Off")
 						info.parent = "";

GPU/Common/PostShader.h

@@ -43,6 +43,8 @@ struct ShaderInfo {
 	int SSAAFilterLevel;
 	// Force constant/max refresh for animated filters
 	bool requires60fps;
+	// Takes previous frame as input (for blending effects.)
+	bool usePreviousFrame;
 
 	struct Setting {
 		std::string name;