From bd674c47b66b62bfb00b8a2b1f57ee59092be2b8 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Wed, 24 Aug 2022 11:01:57 +0200
Subject: [PATCH] OpenGL: Fix regular depal (shader depal still worked) that
 broke recently

---
 Common/GPU/OpenGL/GLQueueRunner.cpp | 12 +++++++-----
 Common/GPU/OpenGL/thin3d_gl.cpp     |  2 +-
 GPU/Common/Draw2D.cpp               |  5 ++++-
 GPU/Common/Draw2D.h                 |  3 ++-
 GPU/Common/TextureCacheCommon.cpp   | 12 ++++++------
 GPU/Common/TextureShaderCommon.cpp  |  1 +
 6 files changed, 21 insertions(+), 14 deletions(-)

diff --git a/Common/GPU/OpenGL/GLQueueRunner.cpp b/Common/GPU/OpenGL/GLQueueRunner.cpp
index dd28d3cfb4..3d3c30e433 100644
--- a/Common/GPU/OpenGL/GLQueueRunner.cpp
+++ b/Common/GPU/OpenGL/GLQueueRunner.cpp
@@ -1035,15 +1035,17 @@ void GLQueueRunner::PerformRenderPass(const GLRStep &step, bool first, bool last
 				activeSlot = slot;
 			}
 			if (c.bind_fb_texture.aspect == GL_COLOR_BUFFER_BIT) {
-				if (curTex[slot] != &c.bind_fb_texture.framebuffer->color_texture)
+				if (curTex[slot] != &c.bind_fb_texture.framebuffer->color_texture) {
 					glBindTexture(GL_TEXTURE_2D, c.bind_fb_texture.framebuffer->color_texture.texture);
-				curTex[slot] = &c.bind_fb_texture.framebuffer->color_texture;
+					curTex[slot] = &c.bind_fb_texture.framebuffer->color_texture;
+				}
 			} else if (c.bind_fb_texture.aspect == GL_DEPTH_BUFFER_BIT) {
-				if (curTex[slot] != &c.bind_fb_texture.framebuffer->z_stencil_texture)
+				if (curTex[slot] != &c.bind_fb_texture.framebuffer->z_stencil_texture) {
 					glBindTexture(GL_TEXTURE_2D, c.bind_fb_texture.framebuffer->z_stencil_texture.texture);
-				curTex[slot] = &c.bind_fb_texture.framebuffer->z_stencil_texture;
+					curTex[slot] = &c.bind_fb_texture.framebuffer->z_stencil_texture;
+				}
 			} else {
-				// TODO: Stencil texturing?
+				// Can't texture from stencil buffers.
 				curTex[slot] = nullptr;
 			}
 			CHECK_GL_ERROR_IF_DEBUG();
diff --git a/Common/GPU/OpenGL/thin3d_gl.cpp b/Common/GPU/OpenGL/thin3d_gl.cpp
index e26a666748..5391fe2198 100644
--- a/Common/GPU/OpenGL/thin3d_gl.cpp
+++ b/Common/GPU/OpenGL/thin3d_gl.cpp
@@ -1214,7 +1214,7 @@ bool OpenGLPipeline::LinkShaders() {
 	}
 	std::vector<GLRProgram::Initializer> initialize;
 	for (int i = 0; i < MAX_TEXTURE_SLOTS; ++i) {
-		if (i < queries.size()) {
+		if (i < samplers_.size()) {
 			initialize.push_back({ &samplerLocs_[i], 0, i });
 		} else {
 			samplerLocs_[i] = -1;
diff --git a/GPU/Common/Draw2D.cpp b/GPU/Common/Draw2D.cpp
index b9ac114f92..c1aefbce20 100644
--- a/GPU/Common/Draw2D.cpp
+++ b/GPU/Common/Draw2D.cpp
@@ -230,7 +230,10 @@ Draw2DPipeline *Draw2D::Create2DPipeline(std::function<Draw2DPipelineInfo (Shade
 		{ draw2DVs_, fs },
 		inputLayout,
 		depthStencil,
-		blend, rasterNoCull, &draw2DUBDesc,
+		blend,
+		rasterNoCull,
+		&draw2DUBDesc,
+		info.samplers.is_empty() ? samplers : info.samplers,
 	};
 
 	Draw::Pipeline *pipeline = draw_->CreateGraphicsPipeline(pipelineDesc);
diff --git a/GPU/Common/Draw2D.h b/GPU/Common/Draw2D.h
index b9516e8d39..ad489dfce6 100644
--- a/GPU/Common/Draw2D.h
+++ b/GPU/Common/Draw2D.h
@@ -1,6 +1,7 @@
 #pragma once
 
 #include "GPU/GPU.h"
+#include "Common/GPU/Shader.h"
 
 // For framebuffer copies and similar things that just require passthrough.
 struct Draw2DVertex {
@@ -32,7 +33,7 @@ inline RasterChannel Draw2DSourceChannel(Draw2DShader shader) {
 struct Draw2DPipelineInfo {
 	RasterChannel readChannel;
 	RasterChannel writeChannel;
-	bool secondTexture;
+	Slice<SamplerDef> samplers;
 };
 
 struct Draw2DPipeline {
diff --git a/GPU/Common/TextureCacheCommon.cpp b/GPU/Common/TextureCacheCommon.cpp
index aaa38bb7c7..3c26ba9544 100644
--- a/GPU/Common/TextureCacheCommon.cpp
+++ b/GPU/Common/TextureCacheCommon.cpp
@@ -1890,13 +1890,14 @@ void TextureCacheCommon::ApplyTextureFramebuffer(VirtualFramebuffer *framebuffer
 
 	bool depth = channel == RASTER_DEPTH;
 	bool need_depalettize = CanDepalettize(texFormat, depth ? GE_FORMAT_DEPTH16 : framebuffer->drawnFormat);
-	bool useShaderDepal = framebufferManager_->GetCurrentRenderVFB() != framebuffer && !depth && !gstate_c.curTextureIs3D;
+
+	// Shader depal is not supported during 3D texturing or depth texturing, and requires 32-bit integer instructions in the shader.
+	bool useShaderDepal = framebufferManager_->GetCurrentRenderVFB() != framebuffer &&
+		!depth &&
+		!gstate_c.curTextureIs3D &&
+		draw_->GetDeviceCaps().fragmentShaderInt32Supported;
 
 	// TODO: Implement shader depal in the fragment shader generator for D3D11 at least.
-	if (!draw_->GetDeviceCaps().fragmentShaderInt32Supported) {
-		useShaderDepal = false;
-	}
-
 	switch (draw_->GetShaderLanguageDesc().shaderLanguage) {
 	case ShaderLanguage::HLSL_D3D11:
 	case ShaderLanguage::HLSL_D3D9:
@@ -1915,7 +1916,6 @@ void TextureCacheCommon::ApplyTextureFramebuffer(VirtualFramebuffer *framebuffer
 		smoothedDepal = CanUseSmoothDepal(gstate, framebuffer->drawnFormat, clutTexture.rampLength);
 
 		if (useShaderDepal) {
-
 			// Very icky conflation here of native and thin3d rendering. This will need careful work per backend in BindAsClutTexture.
 			BindAsClutTexture(clutTexture.texture);
 
diff --git a/GPU/Common/TextureShaderCommon.cpp b/GPU/Common/TextureShaderCommon.cpp
index 694a646131..8fb8cd6fe9 100644
--- a/GPU/Common/TextureShaderCommon.cpp
+++ b/GPU/Common/TextureShaderCommon.cpp
@@ -213,6 +213,7 @@ Draw2DPipeline *TextureShaderCache::GetDepalettizeShader(uint32_t clutMode, GETe
 		return Draw2DPipelineInfo{
 			config.bufferFormat == GE_FORMAT_DEPTH16 ? RASTER_DEPTH : RASTER_COLOR,
 			RASTER_COLOR,
+			samplers
 		};
 	});
 	delete[] buffer;