general: Fixes for Tales of the Abyss (#7381)

* geometry_pipeline: Remove unneeded assert

* Has been hw-tested that gs works correctly even when not in exclusive mode

* pica_core: Propagate output_mask to gs

* Has been hw-tested to occur under the same conditions that other uniforms are shared

* regs_shader: Intialize GPUREG_SH_INPUTBUFFER_CONFIG to default value

* Default value verified on hw. Tales of Abyss does not update the number of vertex attributes for the geometry unit and expects it to be 2

* texture_codec: Align buffer sizes to bpp

* Prevents out of bounds texture reads when launching TOA from the HOME menu

* pica_core: Make default value more clear

src/video_core/pica/geometry_pipeline.cpp

@@ -337,7 +337,6 @@ void GeometryPipeline::Reconfigure() {
     // The following assumes that when geometry shader is in use, the shader unit 3 is configured as
     // a geometry shader unit.
     // TODO: what happens if this is not true?
-    ASSERT(regs.pipeline.gs_unit_exclusive_configuration == 1);
     ASSERT(regs.gs.shader_mode == ShaderRegs::ShaderMode::GS);
     ASSERT(regs.pipeline.use_gs == PipelineRegs::UseGS::Yes);
 

src/video_core/pica/pica_core.cpp

@@ -35,7 +35,7 @@ PicaCore::PicaCore(Memory::MemorySystem& memory_, std::shared_ptr<DebugContext>
                                                                                    gs_unit,
                                                                                    gs_setup},
       shader_engine{CreateEngine(Settings::values.use_shader_jit.GetValue())} {
-    SetFramebufferDefaults();
+    InitializeRegs();
 
     const auto submit_vertex = [this](const AttributeBuffer& buffer) {
         const auto add_triangle = [this](const OutputVertex& v0, const OutputVertex& v1,
@@ -54,7 +54,7 @@ PicaCore::PicaCore(Memory::MemorySystem& memory_, std::shared_ptr<DebugContext>
 
 PicaCore::~PicaCore() = default;
 
-void PicaCore::SetFramebufferDefaults() {
+void PicaCore::InitializeRegs() {
     auto& framebuffer_top = regs.framebuffer_config[0];
     auto& framebuffer_sub = regs.framebuffer_config[1];
 
@@ -77,6 +77,11 @@ void PicaCore::SetFramebufferDefaults() {
     framebuffer_sub.stride = 3 * 240;
     framebuffer_sub.color_format.Assign(PixelFormat::RGB8);
     framebuffer_sub.active_fb = 0;
+
+    // Tales of Abyss expects this register to have the following default values.
+    auto& gs = regs.internal.gs;
+    gs.max_input_attribute_index.Assign(1);
+    gs.shader_mode.Assign(ShaderRegs::ShaderMode::VS);
 }
 
 void PicaCore::BindRasterizer(VideoCore::RasterizerInterface* rasterizer) {
@@ -254,6 +259,13 @@ void PicaCore::WriteInternalReg(u32 id, u32 value, u32 mask) {
         break;
     }
 
+    case PICA_REG_INDEX(vs.output_mask):
+        if (!regs.internal.pipeline.gs_unit_exclusive_configuration &&
+            regs.internal.pipeline.use_gs == PipelineRegs::UseGS::No) {
+            regs.internal.gs.output_mask.Assign(value);
+        }
+        break;
+
     case PICA_REG_INDEX(vs.bool_uniforms):
         vs_setup.WriteUniformBoolReg(regs.internal.vs.bool_uniforms.Value());
         if (!regs.internal.pipeline.gs_unit_exclusive_configuration &&

src/video_core/pica/pica_core.h

@@ -39,7 +39,7 @@ public:
     void ProcessCmdList(PAddr list, u32 size);
 
 private:
-    void SetFramebufferDefaults();
+    void InitializeRegs();
 
     void WriteInternalReg(u32 id, u32 value, u32 mask);
 

src/video_core/pica/regs_shader.h

@@ -35,6 +35,7 @@ struct ShaderRegs {
 
     union {
         // Number of input attributes to shader unit - 1
+        u32 input_buffer_config;
         BitField<0, 4, u32> max_input_attribute_index;
         BitField<8, 8, u32> input_to_uniform;
         BitField<24, 8, ShaderMode> shader_mode;

src/video_core/rasterizer_cache/texture_codec.h

@@ -341,8 +341,8 @@ static constexpr void MortonCopy(u32 width, u32 height, u32 start_offset, u32 en
  */
 template <bool decode, PixelFormat format, bool converted = false>
 static constexpr void LinearCopy(std::span<u8> src_buffer, std::span<u8> dst_buffer) {
-    const std::size_t src_size = src_buffer.size();
-    const std::size_t dst_size = dst_buffer.size();
+    std::size_t src_size = src_buffer.size();
+    std::size_t dst_size = dst_buffer.size();
 
     if constexpr (converted) {
         constexpr u32 encoded_bytes_per_pixel = GetFormatBpp(format) / 8;
@@ -352,6 +352,9 @@ static constexpr void LinearCopy(std::span<u8> src_buffer, std::span<u8> dst_buf
         constexpr u32 dst_bytes_per_pixel =
             decode ? decoded_bytes_per_pixel : encoded_bytes_per_pixel;
 
+        src_size = Common::AlignDown(src_size, src_bytes_per_pixel);
+        dst_size = Common::AlignDown(dst_size, dst_bytes_per_pixel);
+
         for (std::size_t src_index = 0, dst_index = 0; src_index < src_size && dst_index < dst_size;
              src_index += src_bytes_per_pixel, dst_index += dst_bytes_per_pixel) {
             const auto src_pixel = src_buffer.subspan(src_index, src_bytes_per_pixel);