4 files changed, 278 insertions, 184 deletions

diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp
new file mode 100644
index 000000000..e397e8e03
--- /dev/null
+++ b/src/video_core/shader/shader.cpp
@@ -0,0 +1,105 @@
+// Copyright 2015 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/logging/log.h"
+#include "common/profiler.h"
+
+#include "video_core/debug_utils/debug_utils.h"
+#include "video_core/pica.h"
+
+#include "shader.h"
+#include "shader_interpreter.h"
+
+namespace Pica {
+
+namespace Shader {
+
+void Setup(UnitState& state) {
+    // TODO(bunnei): This will be used by the JIT in a subsequent commit
+}
+
+static Common::Profiling::TimingCategory shader_category("Vertex Shader");
+
+OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes) {
+    auto& config = g_state.regs.vs;
+    auto& setup = g_state.vs;
+
+    Common::Profiling::ScopeTimer timer(shader_category);
+
+    state.program_counter = config.main_offset;
+    state.debug.max_offset = 0;
+    state.debug.max_opdesc_id = 0;
+
+    // Setup input register table
+    const auto& attribute_register_map = config.input_register_map;
+
+    if (num_attributes > 0) state.input_registers[attribute_register_map.attribute0_register] = input.attr[0];
+    if (num_attributes > 1) state.input_registers[attribute_register_map.attribute1_register] = input.attr[1];
+    if (num_attributes > 2) state.input_registers[attribute_register_map.attribute2_register] = input.attr[2];
+    if (num_attributes > 3) state.input_registers[attribute_register_map.attribute3_register] = input.attr[3];
+    if (num_attributes > 4) state.input_registers[attribute_register_map.attribute4_register] = input.attr[4];
+    if (num_attributes > 5) state.input_registers[attribute_register_map.attribute5_register] = input.attr[5];
+    if (num_attributes > 6) state.input_registers[attribute_register_map.attribute6_register] = input.attr[6];
+    if (num_attributes > 7) state.input_registers[attribute_register_map.attribute7_register] = input.attr[7];
+    if (num_attributes > 8) state.input_registers[attribute_register_map.attribute8_register] = input.attr[8];
+    if (num_attributes > 9) state.input_registers[attribute_register_map.attribute9_register] = input.attr[9];
+    if (num_attributes > 10) state.input_registers[attribute_register_map.attribute10_register] = input.attr[10];
+    if (num_attributes > 11) state.input_registers[attribute_register_map.attribute11_register] = input.attr[11];
+    if (num_attributes > 12) state.input_registers[attribute_register_map.attribute12_register] = input.attr[12];
+    if (num_attributes > 13) state.input_registers[attribute_register_map.attribute13_register] = input.attr[13];
+    if (num_attributes > 14) state.input_registers[attribute_register_map.attribute14_register] = input.attr[14];
+    if (num_attributes > 15) state.input_registers[attribute_register_map.attribute15_register] = input.attr[15];
+
+    state.conditional_code[0] = false;
+    state.conditional_code[1] = false;
+
+    RunInterpreter(state);
+
+#if PICA_DUMP_SHADERS
+    DebugUtils::DumpShader(setup.program_code.data(), state.debug.max_offset, setup.swizzle_data.data(),
+        state.debug.max_opdesc_id, config.main_offset,
+        g_state.regs.vs_output_attributes); // TODO: Don't hardcode VS here
+#endif
+
+    // Setup output data
+    OutputVertex ret;
+    // TODO(neobrain): Under some circumstances, up to 16 attributes may be output. We need to
+    // figure out what those circumstances are and enable the remaining outputs then.
+    for (int i = 0; i < 7; ++i) {
+        const auto& output_register_map = g_state.regs.vs_output_attributes[i]; // TODO: Don't hardcode VS here
+
+        u32 semantics[4] = {
+            output_register_map.map_x, output_register_map.map_y,
+            output_register_map.map_z, output_register_map.map_w
+        };
+
+        for (int comp = 0; comp < 4; ++comp) {
+            float24* out = ((float24*)&ret) + semantics[comp];
+            if (semantics[comp] != Regs::VSOutputAttributes::INVALID) {
+                *out = state.output_registers[i][comp];
+            } else {
+                // Zero output so that attributes which aren't output won't have denormals in them,
+                // which would slow us down later.
+                memset(out, 0, sizeof(*out));
+            }
+        }
+    }
+
+    // The hardware takes the absolute and saturates vertex colors like this, *before* doing interpolation
+    for (int i = 0; i < 4; ++i) {
+        ret.color[i] = float24::FromFloat32(
+            std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f));
+    }
+
+    LOG_TRACE(Render_Software, "Output vertex: pos (%.2f, %.2f, %.2f, %.2f), col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f)",
+        ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(),
+        ret.color.x.ToFloat32(), ret.color.y.ToFloat32(), ret.color.z.ToFloat32(), ret.color.w.ToFloat32(),
+        ret.tc0.u().ToFloat32(), ret.tc0.v().ToFloat32());
+
+    return ret;
+}
+
+} // namespace Shader
+
+} // namespace Pica
diff --git a/src/video_core/shader/shader.h b/src/video_core/shader/shader.h
new file mode 100644
index 000000000..38c00768d
--- /dev/null
+++ b/src/video_core/shader/shader.h
@@ -0,0 +1,163 @@
+// Copyright 2015 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <boost/container/static_vector.hpp>
+#include <nihstro/shader_binary.h>
+
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+#include "common/vector_math.h"
+
+#include "video_core/pica.h"
+
+using nihstro::RegisterType;
+using nihstro::SourceRegister;
+using nihstro::DestRegister;
+
+namespace Pica {
+
+namespace Shader {
+
+struct InputVertex {
+    Math::Vec4<float24> attr[16];
+};
+
+struct OutputVertex {
+    OutputVertex() = default;
+
+    // VS output attributes
+    Math::Vec4<float24> pos;
+    Math::Vec4<float24> dummy; // quaternions (not implemented, yet)
+    Math::Vec4<float24> color;
+    Math::Vec2<float24> tc0;
+    Math::Vec2<float24> tc1;
+    float24 pad[6];
+    Math::Vec2<float24> tc2;
+
+    // Padding for optimal alignment
+    float24 pad2[4];
+
+    // Attributes used to store intermediate results
+
+    // position after perspective divide
+    Math::Vec3<float24> screenpos;
+    float24 pad3;
+
+    // Linear interpolation
+    // factor: 0=this, 1=vtx
+    void Lerp(float24 factor, const OutputVertex& vtx) {
+        pos = pos * factor + vtx.pos * (float24::FromFloat32(1) - factor);
+
+        // TODO: Should perform perspective correct interpolation here...
+        tc0 = tc0 * factor + vtx.tc0 * (float24::FromFloat32(1) - factor);
+        tc1 = tc1 * factor + vtx.tc1 * (float24::FromFloat32(1) - factor);
+        tc2 = tc2 * factor + vtx.tc2 * (float24::FromFloat32(1) - factor);
+
+        screenpos = screenpos * factor + vtx.screenpos * (float24::FromFloat32(1) - factor);
+
+        color = color * factor + vtx.color * (float24::FromFloat32(1) - factor);
+    }
+
+    // Linear interpolation
+    // factor: 0=v0, 1=v1
+    static OutputVertex Lerp(float24 factor, const OutputVertex& v0, const OutputVertex& v1) {
+        OutputVertex ret = v0;
+        ret.Lerp(factor, v1);
+        return ret;
+    }
+};
+static_assert(std::is_pod<OutputVertex>::value, "Structure is not POD");
+static_assert(sizeof(OutputVertex) == 32 * sizeof(float), "OutputVertex has invalid size");
+
+/**
+ * This structure contains the state information that needs to be unique for a shader unit. The 3DS
+ * has four shader units that process shaders in parallel. At the present, Citra only implements a
+ * single shader unit that processes all shaders serially. Putting the state information in a struct
+ * here will make it easier for us to parallelize the shader processing later.
+ */
+struct UnitState {
+    // The registers are accessed by the shader JIT using SSE instructions, and are therefore
+    // required to be 16-byte aligned.
+    Math::Vec4<float24> MEMORY_ALIGNED16(input_registers[16]);
+    Math::Vec4<float24> MEMORY_ALIGNED16(output_registers[16]);
+    Math::Vec4<float24> MEMORY_ALIGNED16(temporary_registers[16]);
+
+    u32 program_counter;
+    bool conditional_code[2];
+
+    // Two Address registers and one loop counter
+    // TODO: How many bits do these actually have?
+    s32 address_registers[3];
+
+    enum {
+        INVALID_ADDRESS = 0xFFFFFFFF
+    };
+
+    struct CallStackElement {
+        u32 final_address;  // Address upon which we jump to return_address
+        u32 return_address; // Where to jump when leaving scope
+        u8 repeat_counter;  // How often to repeat until this call stack element is removed
+        u8 loop_increment;  // Which value to add to the loop counter after an iteration
+                            // TODO: Should this be a signed value? Does it even matter?
+        u32 loop_address;   // The address where we'll return to after each loop iteration
+    };
+
+    // TODO: Is there a maximal size for this?
+    boost::container::static_vector<CallStackElement, 16> call_stack;
+
+    struct {
+        u32 max_offset; // maximum program counter ever reached
+        u32 max_opdesc_id; // maximum swizzle pattern index ever used
+    } debug;
+
+    static int InputOffset(const SourceRegister& reg) {
+        switch (reg.GetRegisterType()) {
+        case RegisterType::Input:
+            return (int)offsetof(UnitState, input_registers) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
+
+        case RegisterType::Temporary:
+            return (int)offsetof(UnitState, temporary_registers) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
+
+        default:
+            UNREACHABLE();
+            return 0;
+        }
+    }
+
+    static int OutputOffset(const DestRegister& reg) {
+        switch (reg.GetRegisterType()) {
+        case RegisterType::Output:
+            return (int)offsetof(UnitState, output_registers) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
+
+        case RegisterType::Temporary:
+            return (int)offsetof(UnitState, temporary_registers) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
+
+        default:
+            UNREACHABLE();
+            return 0;
+        }
+    }
+};
+
+/**
+ * Performs any shader unit setup that only needs to happen once per shader (as opposed to once per
+ * vertex, which would happen within the `Run` function).
+ * @param state Shader unit state, must be setup per shader and per shader unit
+ */
+void Setup(UnitState& state);
+
+/**
+ * Runs the currently setup shader
+ * @param state Shader unit state, must be setup per shader and per shader unit
+ * @param input Input vertex into the shader
+ * @param num_attributes The number of vertex shader attributes
+ * @return The output vertex, after having been processed by the vertex shader
+ */
+OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes);
+
+} // namespace Shader
+
+} // namespace Pica
diff --git a/src/video_core/shader/shader_interpreter.cpp b/src/video_core/shader/shader_interpreter.cpp
index 369883225..eb48e7053 100644
--- a/src/video_core/shader/shader_interpreter.cpp
+++ b/src/video_core/shader/shader_interpreter.cpp
@@ -2,18 +2,14 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <boost/container/static_vector.hpp>
-#include <boost/range/algorithm.hpp>
-
 #include <common/file_util.h>
 
 #include <nihstro/shader_bytecode.h>
 
-#include "common/profiler.h"
-
 #include "video_core/pica.h"
-#include "video_core/shader/shader_interpreter.h"
-#include "video_core/debug_utils/debug_utils.h"
+
+#include "shader.h"
+#include "shader_interpreter.h"
 
 using nihstro::OpCode;
 using nihstro::Instruction;
@@ -25,42 +21,7 @@ namespace Pica {
 
 namespace Shader {
 
-struct ShaderState {
-    u32 program_counter;
-
-    const float24* input_register_table[16];
-    Math::Vec4<float24> output_registers[16];
-
-    Math::Vec4<float24> temporary_registers[16];
-    bool conditional_code[2];
-
-    // Two Address registers and one loop counter
-    // TODO: How many bits do these actually have?
-    s32 address_registers[3];
-
-    enum {
-        INVALID_ADDRESS = 0xFFFFFFFF
-    };
-
-    struct CallStackElement {
-        u32 final_address;  // Address upon which we jump to return_address
-        u32 return_address; // Where to jump when leaving scope
-        u8 repeat_counter;  // How often to repeat until this call stack element is removed
-        u8 loop_increment;  // Which value to add to the loop counter after an iteration
-                            // TODO: Should this be a signed value? Does it even matter?
-        u32 loop_address;   // The address where we'll return to after each loop iteration
-    };
-
-    // TODO: Is there a maximal size for this?
-    boost::container::static_vector<CallStackElement, 16> call_stack;
-
-    struct {
-        u32 max_offset; // maximum program counter ever reached
-        u32 max_opdesc_id; // maximum swizzle pattern index ever used
-    } debug;
-};
-
-static void ProcessShaderCode(ShaderState& state) {
+void RunInterpreter(UnitState& state) {
     const auto& uniforms = g_state.vs.uniforms;
     const auto& swizzle_data = g_state.vs.swizzle_data;
     const auto& program_code = g_state.vs.program_code;
@@ -90,7 +51,7 @@ static void ProcessShaderCode(ShaderState& state) {
         const Instruction instr = { program_code[state.program_counter] };
         const SwizzlePattern swizzle = { swizzle_data[instr.common.operand_desc_id] };
 
-        static auto call = [](ShaderState& state, u32 offset, u32 num_instructions,
+        static auto call = [](UnitState& state, u32 offset, u32 num_instructions,
                               u32 return_offset, u8 repeat_count, u8 loop_increment) {
             state.program_counter = offset - 1; // -1 to make sure when incrementing the PC we end up at the correct offset
             ASSERT(state.call_stack.size() < state.call_stack.capacity());
@@ -101,7 +62,7 @@ static void ProcessShaderCode(ShaderState& state) {
         auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* {
             switch (source_reg.GetRegisterType()) {
             case RegisterType::Input:
-                return state.input_register_table[source_reg.GetIndex()];
+                return &state.input_registers[source_reg.GetIndex()].x;
 
             case RegisterType::Temporary:
                 return &state.temporary_registers[source_reg.GetIndex()].x;
@@ -413,7 +374,7 @@ static void ProcessShaderCode(ShaderState& state) {
 
         default:
         {
-            static auto evaluate_condition = [](const ShaderState& state, bool refx, bool refy, Instruction::FlowControlType flow_control) {
+            static auto evaluate_condition = [](const UnitState& state, bool refx, bool refy, Instruction::FlowControlType flow_control) {
                 bool results[2] = { refx == state.conditional_code[0],
                                     refy == state.conditional_code[1] };
 
@@ -542,88 +503,6 @@ static void ProcessShaderCode(ShaderState& state) {
     }
 }
 
-static Common::Profiling::TimingCategory shader_category("Vertex Shader");
-
-OutputVertex RunShader(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const State::ShaderSetup& setup) {
-    Common::Profiling::ScopeTimer timer(shader_category);
-
-    ShaderState state;
-
-    state.program_counter = config.main_offset;
-    state.debug.max_offset = 0;
-    state.debug.max_opdesc_id = 0;
-
-    // Setup input register table
-    const auto& attribute_register_map = config.input_register_map;
-    float24 dummy_register;
-    boost::fill(state.input_register_table, &dummy_register);
-
-    if (num_attributes > 0) state.input_register_table[attribute_register_map.attribute0_register] = &input.attr[0].x;
-    if (num_attributes > 1) state.input_register_table[attribute_register_map.attribute1_register] = &input.attr[1].x;
-    if (num_attributes > 2) state.input_register_table[attribute_register_map.attribute2_register] = &input.attr[2].x;
-    if (num_attributes > 3) state.input_register_table[attribute_register_map.attribute3_register] = &input.attr[3].x;
-    if (num_attributes > 4) state.input_register_table[attribute_register_map.attribute4_register] = &input.attr[4].x;
-    if (num_attributes > 5) state.input_register_table[attribute_register_map.attribute5_register] = &input.attr[5].x;
-    if (num_attributes > 6) state.input_register_table[attribute_register_map.attribute6_register] = &input.attr[6].x;
-    if (num_attributes > 7) state.input_register_table[attribute_register_map.attribute7_register] = &input.attr[7].x;
-    if (num_attributes > 8) state.input_register_table[attribute_register_map.attribute8_register] = &input.attr[8].x;
-    if (num_attributes > 9) state.input_register_table[attribute_register_map.attribute9_register] = &input.attr[9].x;
-    if (num_attributes > 10) state.input_register_table[attribute_register_map.attribute10_register] = &input.attr[10].x;
-    if (num_attributes > 11) state.input_register_table[attribute_register_map.attribute11_register] = &input.attr[11].x;
-    if (num_attributes > 12) state.input_register_table[attribute_register_map.attribute12_register] = &input.attr[12].x;
-    if (num_attributes > 13) state.input_register_table[attribute_register_map.attribute13_register] = &input.attr[13].x;
-    if (num_attributes > 14) state.input_register_table[attribute_register_map.attribute14_register] = &input.attr[14].x;
-    if (num_attributes > 15) state.input_register_table[attribute_register_map.attribute15_register] = &input.attr[15].x;
-
-    state.conditional_code[0] = false;
-    state.conditional_code[1] = false;
-
-    ProcessShaderCode(state);
-#if PICA_DUMP_SHADERS
-    DebugUtils::DumpShader(setup.program_code.data(), state.debug.max_offset, setup.swizzle_data.data(),
-                           state.debug.max_opdesc_id, config.main_offset,
-                           g_state.regs.vs_output_attributes); // TODO: Don't hardcode VS here
-#endif
-
-    // Setup output data
-    OutputVertex ret;
-    // TODO(neobrain): Under some circumstances, up to 16 attributes may be output. We need to
-    // figure out what those circumstances are and enable the remaining outputs then.
-    for (int i = 0; i < 7; ++i) {
-        const auto& output_register_map = g_state.regs.vs_output_attributes[i]; // TODO: Don't hardcode VS here
-
-        u32 semantics[4] = {
-            output_register_map.map_x, output_register_map.map_y,
-            output_register_map.map_z, output_register_map.map_w
-        };
-
-        for (int comp = 0; comp < 4; ++comp) {
-            float24* out = ((float24*)&ret) + semantics[comp];
-            if (semantics[comp] != Regs::VSOutputAttributes::INVALID) {
-                *out = state.output_registers[i][comp];
-            } else {
-                // Zero output so that attributes which aren't output won't have denormals in them,
-                // which would slow us down later.
-                memset(out, 0, sizeof(*out));
-            }
-        }
-    }
-
-    // The hardware takes the absolute and saturates vertex colors like this, *before* doing interpolation
-    for (int i = 0; i < 4; ++i) {
-        ret.color[i] = float24::FromFloat32(
-            std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f));
-    }
-
-    LOG_TRACE(Render_Software, "Output vertex: pos (%.2f, %.2f, %.2f, %.2f), col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f)",
-        ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(),
-        ret.color.x.ToFloat32(), ret.color.y.ToFloat32(), ret.color.z.ToFloat32(), ret.color.w.ToFloat32(),
-        ret.tc0.u().ToFloat32(), ret.tc0.v().ToFloat32());
-
-    return ret;
-}
-
-
 } // namespace
 
 } // namespace
diff --git a/src/video_core/shader/shader_interpreter.h b/src/video_core/shader/shader_interpreter.h
index f2900bfc6..ad6e58e39 100644
--- a/src/video_core/shader/shader_interpreter.h
+++ b/src/video_core/shader/shader_interpreter.h
@@ -4,68 +4,15 @@
 
 #pragma once
 
-#include <type_traits>
-
-#include "common/vector_math.h"
-
 #include "video_core/pica.h"
 
+#include "shader.h"
+
 namespace Pica {
 
 namespace Shader {
 
-struct InputVertex {
-    Math::Vec4<float24> attr[16];
-};
-
-struct OutputVertex {
-    OutputVertex() = default;
-
-    // VS output attributes
-    Math::Vec4<float24> pos;
-    Math::Vec4<float24> dummy; // quaternions (not implemented, yet)
-    Math::Vec4<float24> color;
-    Math::Vec2<float24> tc0;
-    Math::Vec2<float24> tc1;
-    float24 pad[6];
-    Math::Vec2<float24> tc2;
-
-    // Padding for optimal alignment
-    float24 pad2[4];
-
-    // Attributes used to store intermediate results
-
-    // position after perspective divide
-    Math::Vec3<float24> screenpos;
-    float24 pad3;
-
-    // Linear interpolation
-    // factor: 0=this, 1=vtx
-    void Lerp(float24 factor, const OutputVertex& vtx) {
-        pos = pos * factor + vtx.pos * (float24::FromFloat32(1) - factor);
-
-        // TODO: Should perform perspective correct interpolation here...
-        tc0 = tc0 * factor + vtx.tc0 * (float24::FromFloat32(1) - factor);
-        tc1 = tc1 * factor + vtx.tc1 * (float24::FromFloat32(1) - factor);
-        tc2 = tc2 * factor + vtx.tc2 * (float24::FromFloat32(1) - factor);
-
-        screenpos = screenpos * factor + vtx.screenpos * (float24::FromFloat32(1) - factor);
-
-        color = color * factor + vtx.color * (float24::FromFloat32(1) - factor);
-    }
-
-    // Linear interpolation
-    // factor: 0=v0, 1=v1
-    static OutputVertex Lerp(float24 factor, const OutputVertex& v0, const OutputVertex& v1) {
-        OutputVertex ret = v0;
-        ret.Lerp(factor, v1);
-        return ret;
-    }
-};
-static_assert(std::is_pod<OutputVertex>::value, "Structure is not POD");
-static_assert(sizeof(OutputVertex) == 32 * sizeof(float), "OutputVertex has invalid size");
-
-OutputVertex RunShader(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const State::ShaderSetup& setup);
+void RunInterpreter(UnitState& state);
 
 } // namespace