5 files changed, 26 insertions, 16 deletions

diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 4f137e693..0c403e16a 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -728,6 +728,7 @@ public:
                 u32 frag_color_clamp;
 
                 union {
+                    BitField<0, 1, u32> y_negate;
                     BitField<4, 1, u32> triangle_rast_flip;
                 } screen_y_control;
 
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 97b9028c5..a9d29501f 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -845,7 +845,8 @@ private:
             // vertex shader, and what's the value of the fourth element when inside a Tess Eval
             // shader.
             ASSERT(stage == Maxwell3D::Regs::ShaderStage::Vertex);
-            return "vec4(0, 0, uintBitsToFloat(instance_id.x), uintBitsToFloat(gl_VertexID))";
+            // Config pack's first value is instance_id.
+            return "vec4(0, 0, uintBitsToFloat(config_pack[0]), uintBitsToFloat(gl_VertexID))";
         case Attribute::Index::FrontFacing:
             // TODO(Subv): Find out what the values are for the other elements.
             ASSERT(stage == Maxwell3D::Regs::ShaderStage::Fragment);
@@ -3507,6 +3508,11 @@ private:
                     regs.SetRegisterToInteger(instr.gpr0, false, 0, "0u", 1, 1);
                     break;
                 }
+                case Tegra::Shader::SystemVariable::Ydirection: {
+                    // Config pack's third value is Y_NEGATE's state.
+                    regs.SetRegisterToFloat(instr.gpr0, 0, "uintBitsToFloat(config_pack[2])", 1, 1);
+                    break;
+                }
                 default: {
                     UNIMPLEMENTED_MSG("Unhandled system move: {}",
                                       static_cast<u32>(instr.sys20.Value()));
diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp
index eea090e52..23ed91e27 100644
--- a/src/video_core/renderer_opengl/gl_shader_gen.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp
@@ -24,8 +24,7 @@ layout (location = 0) out vec4 position;
 
 layout(std140) uniform vs_config {
     vec4 viewport_flip;
-    uvec4 instance_id;
-    uvec4 flip_stage;
+    uvec4 config_pack; // instance_id, flip_stage, y_direction, padding
     uvec4 alpha_test;
 };
 )";
@@ -63,7 +62,8 @@ void main() {
     out += R"(
 
     // Check if the flip stage is VertexB
-    if (flip_stage[0] == 1) {
+    // Config pack's second value is flip_stage
+    if (config_pack[1] == 1) {
         // Viewport can be flipped, which is unsupported by glViewport
         position.xy *= viewport_flip.xy;
     }
@@ -71,7 +71,7 @@ void main() {
 
     // TODO(bunnei): This is likely a hack, position.w should be interpolated as 1.0
     // For now, this is here to bring order in lieu of proper emulation
-    if (flip_stage[0] == 1) {
+    if (config_pack[1] == 1) {
         position.w = 1.0;
     }
 }
@@ -101,8 +101,7 @@ layout (location = 0) out vec4 position;
 
 layout (std140) uniform gs_config {
     vec4 viewport_flip;
-    uvec4 instance_id;
-    uvec4 flip_stage;
+    uvec4 config_pack; // instance_id, flip_stage, y_direction, padding
     uvec4 alpha_test;
 };
 
@@ -139,8 +138,7 @@ layout (location = 0) in vec4 position;
 
 layout (std140) uniform fs_config {
     vec4 viewport_flip;
-    uvec4 instance_id;
-    uvec4 flip_stage;
+    uvec4 config_pack; // instance_id, flip_stage, y_direction, padding
     uvec4 alpha_test;
 };
 
diff --git a/src/video_core/renderer_opengl/gl_shader_manager.cpp b/src/video_core/renderer_opengl/gl_shader_manager.cpp
index 8b8869ecb..6a30c28d2 100644
--- a/src/video_core/renderer_opengl/gl_shader_manager.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_manager.cpp
@@ -27,16 +27,18 @@ void MaxwellUniformData::SetFromRegs(const Maxwell3D::State::ShaderStageInfo& sh
     alpha_test.func = func;
     alpha_test.ref = regs.alpha_test_ref;
 
-    // We only assign the instance to the first component of the vector, the rest is just padding.
-    instance_id[0] = state.current_instance;
+    instance_id = state.current_instance;
 
     // Assign in which stage the position has to be flipped
     // (the last stage before the fragment shader).
     if (gpu.regs.shader_config[static_cast<u32>(Maxwell3D::Regs::ShaderProgram::Geometry)].enable) {
-        flip_stage[0] = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::Geometry);
+        flip_stage = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::Geometry);
     } else {
-        flip_stage[0] = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::VertexB);
+        flip_stage = static_cast<u32>(Maxwell3D::Regs::ShaderProgram::VertexB);
     }
+
+    // Y_NEGATE controls what value S2R returns for the Y_DIRECTION system value.
+    y_direction = regs.screen_y_control.y_negate == 0 ? 1.f : -1.f;
 }
 
 } // namespace OpenGL::GLShader
diff --git a/src/video_core/renderer_opengl/gl_shader_manager.h b/src/video_core/renderer_opengl/gl_shader_manager.h
index 9a5d7e289..b757f5f44 100644
--- a/src/video_core/renderer_opengl/gl_shader_manager.h
+++ b/src/video_core/renderer_opengl/gl_shader_manager.h
@@ -21,8 +21,11 @@ using Tegra::Engines::Maxwell3D;
 struct MaxwellUniformData {
     void SetFromRegs(const Maxwell3D::State::ShaderStageInfo& shader_stage);
     alignas(16) GLvec4 viewport_flip;
-    alignas(16) GLuvec4 instance_id;
-    alignas(16) GLuvec4 flip_stage;
+    struct alignas(16) {
+        GLuint instance_id;
+        GLuint flip_stage;
+        GLfloat y_direction;
+    };
     struct alignas(16) {
         GLuint enabled;
         GLuint func;
@@ -30,7 +33,7 @@ struct MaxwellUniformData {
         GLuint padding;
     } alpha_test;
 };
-static_assert(sizeof(MaxwellUniformData) == 64, "MaxwellUniformData structure size is incorrect");
+static_assert(sizeof(MaxwellUniformData) == 48, "MaxwellUniformData structure size is incorrect");
 static_assert(sizeof(MaxwellUniformData) < 16384,
               "MaxwellUniformData structure must be less than 16kb as per the OpenGL spec");