Merge pull request #11192 from unknownbrackets/stencil

Correct stencil upload bugs

GPU/Common/FramebufferCommon.h

@@ -86,10 +86,13 @@ struct VirtualFramebuffer {
 	// width/height: The detected size of the current framebuffer.
 	u16 width;
 	u16 height;
-	// renderWidth/renderHeight: The actual size we render at. May be scaled to render at higher resolutions.
+	// renderWidth/renderHeight: The scaled size we render at. May be scaled to render at higher resolutions.
+	// The physical buffer may be larger than renderWidth/renderHeight.
 	u16 renderWidth;
 	u16 renderHeight;
-	// bufferWidth/bufferHeight: The actual (but non scaled) size of the buffer we render to. May only be bigger than width/height.
+	// bufferWidth/bufferHeight: The pre-scaling size of the buffer itself. May only be bigger than width/height.
+	// Actual physical buffer is this size times the render resolution multiplier.
+	// The buffer may be used to render a width or height from 0 to these values without being recreated.
 	u16 bufferWidth;
 	u16 bufferHeight;
 

GPU/GLES/StencilBufferGLES.cpp

@@ -126,12 +126,15 @@ bool FramebufferManagerGLES::NotifyStencilUpload(u32 addr, int size, bool skipZe
 		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::Semantic> semantics;
+		semantics.push_back({ 0, "a_position" });
+		semantics.push_back({ 1, "a_texcoord0" });
 		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);
+		stencilUploadProgram_ = render_->CreateProgram(shaders, semantics, queries, inits, false);
 		for (auto iter : shaders) {
 			render_->DeleteShader(iter);
 		}
@@ -146,27 +149,28 @@ bool FramebufferManagerGLES::NotifyStencilUpload(u32 addr, int size, bool skipZe
 
 	// 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) {
+	if (dstBuffer->width == dstBuffer->renderWidth || !dstBuffer->fbo) {
 		useBlit = false;
 	}
-	u16 w = useBlit ? dstBuffer->bufferWidth : dstBuffer->renderWidth;
-	u16 h = useBlit ? dstBuffer->bufferHeight : dstBuffer->renderHeight;
+	u16 w = useBlit ? dstBuffer->width : dstBuffer->renderWidth;
+	u16 h = useBlit ? dstBuffer->height : dstBuffer->renderHeight;
 
 	Draw::Framebuffer *blitFBO = nullptr;
 	if (useBlit) {
 		blitFBO = GetTempFBO(TempFBO::COPY, 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 });
+		draw_->BindFramebufferAsRenderTarget(dstBuffer->fbo, { Draw::RPAction::KEEP, Draw::RPAction::KEEP, Draw::RPAction::DONT_CARE });
 	}
 	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);
+	MakePixelTexture(src, dstBuffer->format, dstBuffer->fb_stride, dstBuffer->width, dstBuffer->height, u1, v1);
 	textureCacheGL_->ForgetLastTexture();
 
 	// We must bind the program after starting the render pass, and set the color mask after clearing.
+	render_->SetScissor({ 0, 0, w, h });
 	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);

GPU/Vulkan/StencilBufferVulkan.cpp

@@ -27,6 +27,8 @@
 #include "GPU/Vulkan/TextureCacheVulkan.h"
 #include "GPU/Vulkan/VulkanUtil.h"
 
+// This shader references gl_FragDepth to prevent early fragment tests.
+// They shouldn't happen since it uses discard, but Adreno detects that incorrectly - see #10634.
 static const char *stencil_fs = R"(#version 450
 #extension GL_ARB_separate_shader_objects : enable
 #extension GL_ARB_shading_language_420pack : enable
@@ -38,6 +40,8 @@ layout (location = 0) in vec2 v_texcoord0;
 layout (location = 0) out vec4 fragColor0;
 
 void main() {
+	gl_FragDepth = 0.0;
+
 	if (u_stencilValue == 0) {
 		fragColor0 = vec4(0.0);
 	} else {