aboutsummaryrefslogtreecommitdiff
path: root/src/video_core/shader
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/shader')
-rw-r--r--src/video_core/shader/shader.cpp116
-rw-r--r--src/video_core/shader/shader.h137
-rw-r--r--src/video_core/shader/shader_interpreter.cpp86
-rw-r--r--src/video_core/shader/shader_interpreter.h6
-rw-r--r--src/video_core/shader/shader_jit_x64.cpp344
-rw-r--r--src/video_core/shader/shader_jit_x64.h63
6 files changed, 422 insertions, 330 deletions
diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp
index 78d295c76..161097610 100644
--- a/src/video_core/shader/shader.cpp
+++ b/src/video_core/shader/shader.cpp
@@ -2,118 +2,91 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
-#include <memory>
+#include <atomic>
+#include <cmath>
+#include <cstring>
#include <unordered_map>
+#include <utility>
#include <boost/range/algorithm/fill.hpp>
+#include "common/bit_field.h"
#include "common/hash.h"
+#include "common/logging/log.h"
#include "common/microprofile.h"
-#include "common/profiler.h"
-#include "video_core/debug_utils/debug_utils.h"
#include "video_core/pica.h"
#include "video_core/pica_state.h"
-#include "video_core/video_core.h"
-
-#include "shader.h"
-#include "shader_interpreter.h"
+#include "video_core/shader/shader.h"
+#include "video_core/shader/shader_interpreter.h"
#ifdef ARCHITECTURE_x86_64
-#include "shader_jit_x64.h"
+#include "video_core/shader/shader_jit_x64.h"
#endif // ARCHITECTURE_x86_64
+#include "video_core/video_core.h"
+
namespace Pica {
namespace Shader {
#ifdef ARCHITECTURE_x86_64
-static std::unordered_map<u64, CompiledShader*> shader_map;
-static JitCompiler jit;
-static CompiledShader* jit_shader;
+static std::unordered_map<u64, std::unique_ptr<JitShader>> shader_map;
+static const JitShader* jit_shader;
+#endif // ARCHITECTURE_x86_64
-static void ClearCache() {
+void ClearCache() {
+#ifdef ARCHITECTURE_x86_64
shader_map.clear();
- jit.Clear();
- LOG_INFO(HW_GPU, "Shader JIT cache cleared");
-}
#endif // ARCHITECTURE_x86_64
+}
-void Setup(UnitState<false>& state) {
+void ShaderSetup::Setup() {
#ifdef ARCHITECTURE_x86_64
if (VideoCore::g_shader_jit_enabled) {
u64 cache_key = (Common::ComputeHash64(&g_state.vs.program_code, sizeof(g_state.vs.program_code)) ^
- Common::ComputeHash64(&g_state.vs.swizzle_data, sizeof(g_state.vs.swizzle_data)) ^
- g_state.regs.vs.main_offset);
+ Common::ComputeHash64(&g_state.vs.swizzle_data, sizeof(g_state.vs.swizzle_data)));
auto iter = shader_map.find(cache_key);
if (iter != shader_map.end()) {
- jit_shader = iter->second;
+ jit_shader = iter->second.get();
} else {
- // Check if remaining JIT code space is enough for at least one more (massive) shader
- if (jit.GetSpaceLeft() < jit_shader_size) {
- // If not, clear the cache of all previously compiled shaders
- ClearCache();
- }
-
- jit_shader = jit.Compile();
- shader_map.emplace(cache_key, jit_shader);
+ auto shader = std::make_unique<JitShader>();
+ shader->Compile();
+ jit_shader = shader.get();
+ shader_map[cache_key] = std::move(shader);
}
}
#endif // ARCHITECTURE_x86_64
}
-void Shutdown() {
-#ifdef ARCHITECTURE_x86_64
- ClearCache();
-#endif // ARCHITECTURE_x86_64
-}
-
-static Common::Profiling::TimingCategory shader_category("Vertex Shader");
-MICROPROFILE_DEFINE(GPU_VertexShader, "GPU", "Vertex Shader", MP_RGB(50, 50, 240));
+MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240));
-OutputVertex Run(UnitState<false>& state, const InputVertex& input, int num_attributes) {
+OutputVertex ShaderSetup::Run(UnitState<false>& state, const InputVertex& input, int num_attributes) {
auto& config = g_state.regs.vs;
+ auto& setup = g_state.vs;
- Common::Profiling::ScopeTimer timer(shader_category);
- MICROPROFILE_SCOPE(GPU_VertexShader);
+ MICROPROFILE_SCOPE(GPU_Shader);
- 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;
- // TODO: Instead of this cumbersome logic, just load the input data directly like
- // for (int attr = 0; attr < num_attributes; ++attr) { input_attr[0] = state.registers.input[attribute_register_map.attribute0_register]; }
- if (num_attributes > 0) state.registers.input[attribute_register_map.attribute0_register] = input.attr[0];
- if (num_attributes > 1) state.registers.input[attribute_register_map.attribute1_register] = input.attr[1];
- if (num_attributes > 2) state.registers.input[attribute_register_map.attribute2_register] = input.attr[2];
- if (num_attributes > 3) state.registers.input[attribute_register_map.attribute3_register] = input.attr[3];
- if (num_attributes > 4) state.registers.input[attribute_register_map.attribute4_register] = input.attr[4];
- if (num_attributes > 5) state.registers.input[attribute_register_map.attribute5_register] = input.attr[5];
- if (num_attributes > 6) state.registers.input[attribute_register_map.attribute6_register] = input.attr[6];
- if (num_attributes > 7) state.registers.input[attribute_register_map.attribute7_register] = input.attr[7];
- if (num_attributes > 8) state.registers.input[attribute_register_map.attribute8_register] = input.attr[8];
- if (num_attributes > 9) state.registers.input[attribute_register_map.attribute9_register] = input.attr[9];
- if (num_attributes > 10) state.registers.input[attribute_register_map.attribute10_register] = input.attr[10];
- if (num_attributes > 11) state.registers.input[attribute_register_map.attribute11_register] = input.attr[11];
- if (num_attributes > 12) state.registers.input[attribute_register_map.attribute12_register] = input.attr[12];
- if (num_attributes > 13) state.registers.input[attribute_register_map.attribute13_register] = input.attr[13];
- if (num_attributes > 14) state.registers.input[attribute_register_map.attribute14_register] = input.attr[14];
- if (num_attributes > 15) state.registers.input[attribute_register_map.attribute15_register] = input.attr[15];
+ for (unsigned i = 0; i < num_attributes; i++)
+ state.registers.input[attribute_register_map.GetRegisterForAttribute(i)] = input.attr[i];
state.conditional_code[0] = false;
state.conditional_code[1] = false;
#ifdef ARCHITECTURE_x86_64
if (VideoCore::g_shader_jit_enabled)
- jit_shader(&state.registers);
+ jit_shader->Run(setup, state, config.main_offset);
else
- RunInterpreter(state);
+ RunInterpreter(setup, state, config.main_offset);
#else
- RunInterpreter(state);
+ RunInterpreter(setup, state, config.main_offset);
#endif // ARCHITECTURE_x86_64
// Setup output data
@@ -167,10 +140,9 @@ OutputVertex Run(UnitState<false>& state, const InputVertex& input, int num_attr
return ret;
}
-DebugData<true> ProduceDebugInfo(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const ShaderSetup& setup) {
+DebugData<true> ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const ShaderSetup& setup) {
UnitState<true> state;
- state.program_counter = config.main_offset;
state.debug.max_offset = 0;
state.debug.max_opdesc_id = 0;
@@ -179,27 +151,13 @@ DebugData<true> ProduceDebugInfo(const InputVertex& input, int num_attributes, c
float24 dummy_register;
boost::fill(state.registers.input, &dummy_register);
- if (num_attributes > 0) state.registers.input[attribute_register_map.attribute0_register] = &input.attr[0].x;
- if (num_attributes > 1) state.registers.input[attribute_register_map.attribute1_register] = &input.attr[1].x;
- if (num_attributes > 2) state.registers.input[attribute_register_map.attribute2_register] = &input.attr[2].x;
- if (num_attributes > 3) state.registers.input[attribute_register_map.attribute3_register] = &input.attr[3].x;
- if (num_attributes > 4) state.registers.input[attribute_register_map.attribute4_register] = &input.attr[4].x;
- if (num_attributes > 5) state.registers.input[attribute_register_map.attribute5_register] = &input.attr[5].x;
- if (num_attributes > 6) state.registers.input[attribute_register_map.attribute6_register] = &input.attr[6].x;
- if (num_attributes > 7) state.registers.input[attribute_register_map.attribute7_register] = &input.attr[7].x;
- if (num_attributes > 8) state.registers.input[attribute_register_map.attribute8_register] = &input.attr[8].x;
- if (num_attributes > 9) state.registers.input[attribute_register_map.attribute9_register] = &input.attr[9].x;
- if (num_attributes > 10) state.registers.input[attribute_register_map.attribute10_register] = &input.attr[10].x;
- if (num_attributes > 11) state.registers.input[attribute_register_map.attribute11_register] = &input.attr[11].x;
- if (num_attributes > 12) state.registers.input[attribute_register_map.attribute12_register] = &input.attr[12].x;
- if (num_attributes > 13) state.registers.input[attribute_register_map.attribute13_register] = &input.attr[13].x;
- if (num_attributes > 14) state.registers.input[attribute_register_map.attribute14_register] = &input.attr[14].x;
- if (num_attributes > 15) state.registers.input[attribute_register_map.attribute15_register] = &input.attr[15].x;
+ for (unsigned i = 0; i < num_attributes; i++)
+ state.registers.input[attribute_register_map.GetRegisterForAttribute(i)] = input.attr[i];
state.conditional_code[0] = false;
state.conditional_code[1] = false;
- RunInterpreter(state);
+ RunInterpreter(setup, state, config.main_offset);
return state.debug;
}
diff --git a/src/video_core/shader/shader.h b/src/video_core/shader/shader.h
index 7af8f1fa1..84898f21c 100644
--- a/src/video_core/shader/shader.h
+++ b/src/video_core/shader/shader.h
@@ -4,17 +4,23 @@
#pragma once
+#include <array>
+#include <cstddef>
+#include <memory>
+#include <type_traits>
#include <vector>
#include <boost/container/static_vector.hpp>
-#include <nihstro/shader_binary.h>
+#include <nihstro/shader_bytecode.h>
+#include "common/assert.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/vector_math.h"
#include "video_core/pica.h"
+#include "video_core/pica_types.h"
using nihstro::RegisterType;
using nihstro::SourceRegister;
@@ -25,7 +31,7 @@ namespace Pica {
namespace Shader {
struct InputVertex {
- Math::Vec4<float24> attr[16];
+ alignas(16) Math::Vec4<float24> attr[16];
};
struct OutputVertex {
@@ -37,7 +43,8 @@ struct OutputVertex {
Math::Vec4<float24> color;
Math::Vec2<float24> tc0;
Math::Vec2<float24> tc1;
- INSERT_PADDING_WORDS(2);
+ float24 tc0_w;
+ INSERT_PADDING_WORDS(1);
Math::Vec3<float24> view;
INSERT_PADDING_WORDS(1);
Math::Vec2<float24> tc2;
@@ -77,23 +84,6 @@ struct OutputVertex {
static_assert(std::is_pod<OutputVertex>::value, "Structure is not POD");
static_assert(sizeof(OutputVertex) == 32 * sizeof(float), "OutputVertex has invalid size");
-/// Vertex shader memory
-struct ShaderSetup {
- struct {
- // The float uniforms are accessed by the shader JIT using SSE instructions, and are
- // therefore required to be 16-byte aligned.
- alignas(16) Math::Vec4<float24> f[96];
-
- std::array<bool, 16> b;
- std::array<Math::Vec4<u8>, 4> i;
- } uniforms;
-
- Math::Vec4<float24> default_attributes[16];
-
- std::array<u32, 1024> program_code;
- std::array<u32, 1024> swizzle_data;
-};
-
// Helper structure used to keep track of data useful for inspection of shader emulation
template<bool full_debugging>
struct DebugData;
@@ -282,38 +272,21 @@ struct UnitState {
} registers;
static_assert(std::is_pod<Registers>::value, "Structure is not POD");
- 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;
-
DebugData<Debug> debug;
static size_t InputOffset(const SourceRegister& reg) {
switch (reg.GetRegisterType()) {
case RegisterType::Input:
- return offsetof(UnitState::Registers, input) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
+ return offsetof(UnitState, registers.input) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
case RegisterType::Temporary:
- return offsetof(UnitState::Registers, temporary) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
+ return offsetof(UnitState, registers.temporary) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
default:
UNREACHABLE();
@@ -324,10 +297,10 @@ struct UnitState {
static size_t OutputOffset(const DestRegister& reg) {
switch (reg.GetRegisterType()) {
case RegisterType::Output:
- return offsetof(UnitState::Registers, output) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
+ return offsetof(UnitState, registers.output) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
case RegisterType::Temporary:
- return offsetof(UnitState::Registers, temporary) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
+ return offsetof(UnitState, registers.temporary) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
default:
UNREACHABLE();
@@ -336,34 +309,66 @@ struct UnitState {
}
};
-/**
- * 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<false>& state);
+/// Clears the shader cache
+void ClearCache();
-/// Performs any cleanup when the emulator is shutdown
-void Shutdown();
+struct ShaderSetup {
-/**
- * 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<false>& state, const InputVertex& input, int num_attributes);
+ struct {
+ // The float uniforms are accessed by the shader JIT using SSE instructions, and are
+ // therefore required to be 16-byte aligned.
+ alignas(16) Math::Vec4<float24> f[96];
-/**
- * Produce debug information based on the given shader and input vertex
- * @param input Input vertex into the shader
- * @param num_attributes The number of vertex shader attributes
- * @param config Configuration object for the shader pipeline
- * @param setup Setup object for the shader pipeline
- * @return Debug information for this shader with regards to the given vertex
- */
-DebugData<true> ProduceDebugInfo(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const ShaderSetup& setup);
+ std::array<bool, 16> b;
+ std::array<Math::Vec4<u8>, 4> i;
+ } uniforms;
+
+ static size_t UniformOffset(RegisterType type, unsigned index) {
+ switch (type) {
+ case RegisterType::FloatUniform:
+ return offsetof(ShaderSetup, uniforms.f) + index*sizeof(Math::Vec4<float24>);
+
+ case RegisterType::BoolUniform:
+ return offsetof(ShaderSetup, uniforms.b) + index*sizeof(bool);
+
+ case RegisterType::IntUniform:
+ return offsetof(ShaderSetup, uniforms.i) + index*sizeof(Math::Vec4<u8>);
+
+ default:
+ UNREACHABLE();
+ return 0;
+ }
+ }
+
+ std::array<u32, 1024> program_code;
+ std::array<u32, 1024> swizzle_data;
+
+ /**
+ * 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).
+ */
+ void Setup();
+
+ /**
+ * 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<false>& state, const InputVertex& input, int num_attributes);
+
+ /**
+ * Produce debug information based on the given shader and input vertex
+ * @param input Input vertex into the shader
+ * @param num_attributes The number of vertex shader attributes
+ * @param config Configuration object for the shader pipeline
+ * @param setup Setup object for the shader pipeline
+ * @return Debug information for this shader with regards to the given vertex
+ */
+ DebugData<true> ProduceDebugInfo(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const ShaderSetup& setup);
+
+};
} // namespace Shader
diff --git a/src/video_core/shader/shader_interpreter.cpp b/src/video_core/shader/shader_interpreter.cpp
index 9b978583e..714e8bfd5 100644
--- a/src/video_core/shader/shader_interpreter.cpp
+++ b/src/video_core/shader/shader_interpreter.cpp
@@ -2,12 +2,20 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
+#include <algorithm>
+#include <array>
+#include <cmath>
#include <numeric>
+
#include <nihstro/shader_bytecode.h>
-#include "common/file_util.h"
-#include "video_core/pica.h"
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "common/logging/log.h"
+#include "common/vector_math.h"
+
#include "video_core/pica_state.h"
+#include "video_core/pica_types.h"
#include "video_core/shader/shader.h"
#include "video_core/shader/shader_interpreter.h"
@@ -21,8 +29,24 @@ namespace Pica {
namespace Shader {
+constexpr u32 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
+};
+
template<bool Debug>
-void RunInterpreter(UnitState<Debug>& state) {
+void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned offset) {
+ // TODO: Is there a maximal size for this?
+ boost::container::static_vector<CallStackElement, 16> call_stack;
+
+ u32 program_counter = offset;
+
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;
@@ -33,16 +57,16 @@ void RunInterpreter(UnitState<Debug>& state) {
unsigned iteration = 0;
bool exit_loop = false;
while (!exit_loop) {
- if (!state.call_stack.empty()) {
- auto& top = state.call_stack.back();
- if (state.program_counter == top.final_address) {
+ if (!call_stack.empty()) {
+ auto& top = call_stack.back();
+ if (program_counter == top.final_address) {
state.address_registers[2] += top.loop_increment;
if (top.repeat_counter-- == 0) {
- state.program_counter = top.return_address;
- state.call_stack.pop_back();
+ program_counter = top.return_address;
+ call_stack.pop_back();
} else {
- state.program_counter = top.loop_address;
+ program_counter = top.loop_address;
}
// TODO: Is "trying again" accurate to hardware?
@@ -50,20 +74,20 @@ void RunInterpreter(UnitState<Debug>& state) {
}
}
- const Instruction instr = { program_code[state.program_counter] };
+ const Instruction instr = { program_code[program_counter] };
const SwizzlePattern swizzle = { swizzle_data[instr.common.operand_desc_id] };
- static auto call = [](UnitState<Debug>& state, u32 offset, u32 num_instructions,
+ static auto call = [&program_counter, &call_stack](UnitState<Debug>& 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());
- state.call_stack.push_back({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset });
+ program_counter = offset - 1; // -1 to make sure when incrementing the PC we end up at the correct offset
+ ASSERT(call_stack.size() < call_stack.capacity());
+ call_stack.push_back({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset });
};
- Record<DebugDataRecord::CUR_INSTR>(state.debug, iteration, state.program_counter);
+ Record<DebugDataRecord::CUR_INSTR>(state.debug, iteration, program_counter);
if (iteration > 0)
- Record<DebugDataRecord::NEXT_INSTR>(state.debug, iteration - 1, state.program_counter);
+ Record<DebugDataRecord::NEXT_INSTR>(state.debug, iteration - 1, program_counter);
- state.debug.max_offset = std::max<u32>(state.debug.max_offset, 1 + state.program_counter);
+ state.debug.max_offset = std::max<u32>(state.debug.max_offset, 1 + program_counter);
auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* {
switch (source_reg.GetRegisterType()) {
@@ -511,7 +535,7 @@ void RunInterpreter(UnitState<Debug>& state) {
case OpCode::Id::JMPC:
Record<DebugDataRecord::COND_CMP_IN>(state.debug, iteration, state.conditional_code);
if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) {
- state.program_counter = instr.flow_control.dest_offset - 1;
+ program_counter = instr.flow_control.dest_offset - 1;
}
break;
@@ -519,7 +543,7 @@ void RunInterpreter(UnitState<Debug>& state) {
Record<DebugDataRecord::COND_BOOL_IN>(state.debug, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
if (uniforms.b[instr.flow_control.bool_uniform_id] == !(instr.flow_control.num_instructions & 1)) {
- state.program_counter = instr.flow_control.dest_offset - 1;
+ program_counter = instr.flow_control.dest_offset - 1;
}
break;
@@ -527,7 +551,7 @@ void RunInterpreter(UnitState<Debug>& state) {
call(state,
instr.flow_control.dest_offset,
instr.flow_control.num_instructions,
- state.program_counter + 1, 0, 0);
+ program_counter + 1, 0, 0);
break;
case OpCode::Id::CALLU:
@@ -536,7 +560,7 @@ void RunInterpreter(UnitState<Debug>& state) {
call(state,
instr.flow_control.dest_offset,
instr.flow_control.num_instructions,
- state.program_counter + 1, 0, 0);
+ program_counter + 1, 0, 0);
}
break;
@@ -546,7 +570,7 @@ void RunInterpreter(UnitState<Debug>& state) {
call(state,
instr.flow_control.dest_offset,
instr.flow_control.num_instructions,
- state.program_counter + 1, 0, 0);
+ program_counter + 1, 0, 0);
}
break;
@@ -557,8 +581,8 @@ void RunInterpreter(UnitState<Debug>& state) {
Record<DebugDataRecord::COND_BOOL_IN>(state.debug, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
if (uniforms.b[instr.flow_control.bool_uniform_id]) {
call(state,
- state.program_counter + 1,
- instr.flow_control.dest_offset - state.program_counter - 1,
+ program_counter + 1,
+ instr.flow_control.dest_offset - program_counter - 1,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
} else {
call(state,
@@ -576,8 +600,8 @@ void RunInterpreter(UnitState<Debug>& state) {
Record<DebugDataRecord::COND_CMP_IN>(state.debug, iteration, state.conditional_code);
if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) {
call(state,
- state.program_counter + 1,
- instr.flow_control.dest_offset - state.program_counter - 1,
+ program_counter + 1,
+ instr.flow_control.dest_offset - program_counter - 1,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
} else {
call(state,
@@ -599,8 +623,8 @@ void RunInterpreter(UnitState<Debug>& state) {
Record<DebugDataRecord::LOOP_INT_IN>(state.debug, iteration, loop_param);
call(state,
- state.program_counter + 1,
- instr.flow_control.dest_offset - state.program_counter + 1,
+ program_counter + 1,
+ instr.flow_control.dest_offset - program_counter + 1,
instr.flow_control.dest_offset + 1,
loop_param.x,
loop_param.z);
@@ -617,14 +641,14 @@ void RunInterpreter(UnitState<Debug>& state) {
}
}
- ++state.program_counter;
+ ++program_counter;
++iteration;
}
}
// Explicit instantiation
-template void RunInterpreter(UnitState<false>& state);
-template void RunInterpreter(UnitState<true>& state);
+template void RunInterpreter(const ShaderSetup& setup, UnitState<false>& state, unsigned offset);
+template void RunInterpreter(const ShaderSetup& setup, UnitState<true>& state, unsigned offset);
} // namespace
diff --git a/src/video_core/shader/shader_interpreter.h b/src/video_core/shader/shader_interpreter.h
index 294bca50e..bb3ce1c6e 100644
--- a/src/video_core/shader/shader_interpreter.h
+++ b/src/video_core/shader/shader_interpreter.h
@@ -4,14 +4,14 @@
#pragma once
-#include "video_core/shader/shader.h"
-
namespace Pica {
namespace Shader {
+template <bool Debug> struct UnitState;
+
template<bool Debug>
-void RunInterpreter(UnitState<Debug>& state);
+void RunInterpreter(const ShaderSetup& setup, UnitState<Debug>& state, unsigned offset);
} // namespace
diff --git a/src/video_core/shader/shader_jit_x64.cpp b/src/video_core/shader/shader_jit_x64.cpp
index dffe051ef..43e7e6b4c 100644
--- a/src/video_core/shader/shader_jit_x64.cpp
+++ b/src/video_core/shader/shader_jit_x64.cpp
@@ -2,8 +2,16 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
-#include <smmintrin.h>
+#include <algorithm>
+#include <cmath>
+#include <cstdint>
+#include <xmmintrin.h>
+#include <nihstro/shader_bytecode.h>
+
+#include "common/assert.h"
+#include "common/logging/log.h"
+#include "common/vector_math.h"
#include "common/x64/abi.h"
#include "common/x64/cpu_detect.h"
#include "common/x64/emitter.h"
@@ -12,6 +20,7 @@
#include "shader_jit_x64.h"
#include "video_core/pica_state.h"
+#include "video_core/pica_types.h"
namespace Pica {
@@ -19,73 +28,73 @@ namespace Shader {
using namespace Gen;
-typedef void (JitCompiler::*JitFunction)(Instruction instr);
+typedef void (JitShader::*JitFunction)(Instruction instr);
const JitFunction instr_table[64] = {
- &JitCompiler::Compile_ADD, // add
- &JitCompiler::Compile_DP3, // dp3
- &JitCompiler::Compile_DP4, // dp4
- &JitCompiler::Compile_DPH, // dph
+ &JitShader::Compile_ADD, // add
+ &JitShader::Compile_DP3, // dp3
+ &JitShader::Compile_DP4, // dp4
+ &JitShader::Compile_DPH, // dph
nullptr, // unknown
- &JitCompiler::Compile_EX2, // ex2
- &JitCompiler::Compile_LG2, // lg2
+ &JitShader::Compile_EX2, // ex2
+ &JitShader::Compile_LG2, // lg2
nullptr, // unknown
- &JitCompiler::Compile_MUL, // mul
- &JitCompiler::Compile_SGE, // sge
- &JitCompiler::Compile_SLT, // slt
- &JitCompiler::Compile_FLR, // flr
- &JitCompiler::Compile_MAX, // max
- &JitCompiler::Compile_MIN, // min
- &JitCompiler::Compile_RCP, // rcp
- &JitCompiler::Compile_RSQ, // rsq
+ &JitShader::Compile_MUL, // mul
+ &JitShader::Compile_SGE, // sge
+ &JitShader::Compile_SLT, // slt
+ &JitShader::Compile_FLR, // flr
+ &JitShader::Compile_MAX, // max
+ &JitShader::Compile_MIN, // min
+ &JitShader::Compile_RCP, // rcp
+ &JitShader::Compile_RSQ, // rsq
nullptr, // unknown
nullptr, // unknown
- &JitCompiler::Compile_MOVA, // mova
- &JitCompiler::Compile_MOV, // mov
+ &JitShader::Compile_MOVA, // mova
+ &JitShader::Compile_MOV, // mov
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
- &JitCompiler::Compile_DPH, // dphi
+ &JitShader::Compile_DPH, // dphi
nullptr, // unknown
- &JitCompiler::Compile_SGE, // sgei
- &JitCompiler::Compile_SLT, // slti
+ &JitShader::Compile_SGE, // sgei
+ &JitShader::Compile_SLT, // slti
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
nullptr, // unknown
- &JitCompiler::Compile_NOP, // nop
- &JitCompiler::Compile_END, // end
+ &JitShader::Compile_NOP, // nop
+ &JitShader::Compile_END, // end
nullptr, // break
- &JitCompiler::Compile_CALL, // call
- &JitCompiler::Compile_CALLC, // callc
- &JitCompiler::Compile_CALLU, // callu
- &JitCompiler::Compile_IF, // ifu
- &JitCompiler::Compile_IF, // ifc
- &JitCompiler::Compile_LOOP, // loop
+ &JitShader::Compile_CALL, // call
+ &JitShader::Compile_CALLC, // callc
+ &JitShader::Compile_CALLU, // callu
+ &JitShader::Compile_IF, // ifu
+ &JitShader::Compile_IF, // ifc
+ &JitShader::Compile_LOOP, // loop
nullptr, // emit
nullptr, // sete
- &JitCompiler::Compile_JMP, // jmpc
- &JitCompiler::Compile_JMP, // jmpu
- &JitCompiler::Compile_CMP, // cmp
- &JitCompiler::Compile_CMP, // cmp
- &JitCompiler::Compile_MAD, // madi
- &JitCompiler::Compile_MAD, // madi
- &JitCompiler::Compile_MAD, // madi
- &JitCompiler::Compile_MAD, // madi
- &JitCompiler::Compile_MAD, // madi
- &JitCompiler::Compile_MAD, // madi
- &JitCompiler::Compile_MAD, // madi
- &JitCompiler::Compile_MAD, // madi
- &JitCompiler::Compile_MAD, // mad
- &JitCompiler::Compile_MAD, // mad
- &JitCompiler::Compile_MAD, // mad
- &JitCompiler::Compile_MAD, // mad
- &JitCompiler::Compile_MAD, // mad
- &JitCompiler::Compile_MAD, // mad
- &JitCompiler::Compile_MAD, // mad
- &JitCompiler::Compile_MAD, // mad
+ &JitShader::Compile_JMP, // jmpc
+ &JitShader::Compile_JMP, // jmpu
+ &JitShader::Compile_CMP, // cmp
+ &JitShader::Compile_CMP, // cmp
+ &JitShader::Compile_MAD, // madi
+ &JitShader::Compile_MAD, // madi
+ &JitShader::Compile_MAD, // madi
+ &JitShader::Compile_MAD, // madi
+ &JitShader::Compile_MAD, // madi
+ &JitShader::Compile_MAD, // madi
+ &JitShader::Compile_MAD, // madi
+ &JitShader::Compile_MAD, // madi
+ &JitShader::Compile_MAD, // mad
+ &JitShader::Compile_MAD, // mad
+ &JitShader::Compile_MAD, // mad
+ &JitShader::Compile_MAD, // mad
+ &JitShader::Compile_MAD, // mad
+ &JitShader::Compile_MAD, // mad
+ &JitShader::Compile_MAD, // mad
+ &JitShader::Compile_MAD, // mad
};
// The following is used to alias some commonly used registers. Generally, RAX-RDX and XMM0-XMM3 can
@@ -93,7 +102,7 @@ const JitFunction instr_table[64] = {
// purposes, as documented below:
/// Pointer to the uniform memory
-static const X64Reg UNIFORMS = R9;
+static const X64Reg SETUP = R9;
/// The two 32-bit VS address offset registers set by the MOVA instruction
static const X64Reg ADDROFFS_REG_0 = R10;
static const X64Reg ADDROFFS_REG_1 = R11;
@@ -108,7 +117,7 @@ static const X64Reg COND0 = R13;
/// Result of the previous CMP instruction for the Y-component comparison
static const X64Reg COND1 = R14;
/// Pointer to the UnitState instance for the current VS unit
-static const X64Reg REGISTERS = R15;
+static const X64Reg STATE = R15;
/// SIMD scratch register
static const X64Reg SCRATCH = XMM0;
/// Loaded with the first swizzled source register, otherwise can be used as a scratch register
@@ -127,7 +136,7 @@ static const X64Reg NEGBIT = XMM15;
// State registers that must not be modified by external functions calls
// Scratch registers, e.g., SRC1 and SCRATCH, have to be saved on the side if needed
static const BitSet32 persistent_regs = {
- UNIFORMS, REGISTERS, // Pointers to register blocks
+ SETUP, STATE, // Pointers to register blocks
ADDROFFS_REG_0, ADDROFFS_REG_1, LOOPCOUNT_REG, COND0, COND1, // Cached registers
ONE+16, NEGBIT+16, // Constants
};
@@ -138,21 +147,40 @@ static const u8 NO_SRC_REG_SWIZZLE = 0x1b;
static const u8 NO_DEST_REG_MASK = 0xf;
/**
+ * Get the vertex shader instruction for a given offset in the current shader program
+ * @param offset Offset in the current shader program of the instruction
+ * @return Instruction at the specified offset
+ */
+static Instruction GetVertexShaderInstruction(size_t offset) {
+ return { g_state.vs.program_code[offset] };
+}
+
+static void LogCritical(const char* msg) {
+ LOG_CRITICAL(HW_GPU, "%s", msg);
+}
+
+void JitShader::Compile_Assert(bool condition, const char* msg) {
+ if (!condition) {
+ ABI_CallFunctionP(reinterpret_cast<const void*>(LogCritical), const_cast<char*>(msg));
+ }
+}
+
+/**
* Loads and swizzles a source register into the specified XMM register.
* @param instr VS instruction, used for determining how to load the source register
* @param src_num Number indicating which source register to load (1 = src1, 2 = src2, 3 = src3)
* @param src_reg SourceRegister object corresponding to the source register to load
* @param dest Destination XMM register to store the loaded, swizzled source register
*/
-void JitCompiler::Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRegister src_reg, X64Reg dest) {
+void JitShader::Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRegister src_reg, X64Reg dest) {
X64Reg src_ptr;
size_t src_offset;
if (src_reg.GetRegisterType() == RegisterType::FloatUniform) {
- src_ptr = UNIFORMS;
- src_offset = src_reg.GetIndex() * sizeof(float24) * 4;
+ src_ptr = SETUP;
+ src_offset = ShaderSetup::UniformOffset(RegisterType::FloatUniform, src_reg.GetIndex());
} else {
- src_ptr = REGISTERS;
+ src_ptr = STATE;
src_offset = UnitState<false>::InputOffset(src_reg);
}
@@ -216,7 +244,7 @@ void JitCompiler::Compile_SwizzleSrc(Instruction instr, unsigned src_num, Source
}
}
-void JitCompiler::Compile_DestEnable(Instruction instr,X64Reg src) {
+void JitShader::Compile_DestEnable(Instruction instr,X64Reg src) {
DestRegister dest;
unsigned operand_desc_id;
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MAD ||
@@ -236,11 +264,11 @@ void JitCompiler::Compile_DestEnable(Instruction instr,X64Reg src) {
// If all components are enabled, write the result to the destination register
if (swiz.dest_mask == NO_DEST_REG_MASK) {
// Store dest back to memory
- MOVAPS(MDisp(REGISTERS, dest_offset_disp), src);
+ MOVAPS(MDisp(STATE, dest_offset_disp), src);
} else {
// Not all components are enabled, so mask the result when storing to the destination register...
- MOVAPS(SCRATCH, MDisp(REGISTERS, dest_offset_disp));
+ MOVAPS(SCRATCH, MDisp(STATE, dest_offset_disp));
if (Common::GetCPUCaps().sse4_1) {
u8 mask = ((swiz.dest_mask & 1) << 3) | ((swiz.dest_mask & 8) >> 3) | ((swiz.dest_mask & 2) << 1) | ((swiz.dest_mask & 4) >> 1);
@@ -259,11 +287,11 @@ void JitCompiler::Compile_DestEnable(Instruction instr,X64Reg src) {
}
// Store dest back to memory
- MOVAPS(MDisp(REGISTERS, dest_offset_disp), SCRATCH);
+ MOVAPS(MDisp(STATE, dest_offset_disp), SCRATCH);
}
}
-void JitCompiler::Compile_SanitizedMul(Gen::X64Reg src1, Gen::X64Reg src2, Gen::X64Reg scratch) {
+void JitShader::Compile_SanitizedMul(Gen::X64Reg src1, Gen::X64Reg src2, Gen::X64Reg scratch) {
MOVAPS(scratch, R(src1));
CMPPS(scratch, R(src2), CMP_ORD);
@@ -276,7 +304,7 @@ void JitCompiler::Compile_SanitizedMul(Gen::X64Reg src1, Gen::X64Reg src2, Gen::
ANDPS(src1, R(scratch));
}
-void JitCompiler::Compile_EvaluateCondition(Instruction instr) {
+void JitShader::Compile_EvaluateCondition(Instruction instr) {
// Note: NXOR is used below to check for equality
switch (instr.flow_control.op) {
case Instruction::FlowControlType::Or:
@@ -307,23 +335,23 @@ void JitCompiler::Compile_EvaluateCondition(Instruction instr) {
}
}
-void JitCompiler::Compile_UniformCondition(Instruction instr) {
- int offset = offsetof(decltype(g_state.vs.uniforms), b) + (instr.flow_control.bool_uniform_id * sizeof(bool));
- CMP(sizeof(bool) * 8, MDisp(UNIFORMS, offset), Imm8(0));
+void JitShader::Compile_UniformCondition(Instruction instr) {
+ int offset = ShaderSetup::UniformOffset(RegisterType::BoolUniform, instr.flow_control.bool_uniform_id);
+ CMP(sizeof(bool) * 8, MDisp(SETUP, offset), Imm8(0));
}
-BitSet32 JitCompiler::PersistentCallerSavedRegs() {
+BitSet32 JitShader::PersistentCallerSavedRegs() {
return persistent_regs & ABI_ALL_CALLER_SAVED;
}
-void JitCompiler::Compile_ADD(Instruction instr) {
+void JitShader::Compile_ADD(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
ADDPS(SRC1, R(SRC2));
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_DP3(Instruction instr) {
+void JitShader::Compile_DP3(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
@@ -342,7 +370,7 @@ void JitCompiler::Compile_DP3(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_DP4(Instruction instr) {
+void JitShader::Compile_DP4(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
@@ -359,7 +387,7 @@ void JitCompiler::Compile_DP4(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_DPH(Instruction instr) {
+void JitShader::Compile_DPH(Instruction instr) {
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::DPHI) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
@@ -391,7 +419,7 @@ void JitCompiler::Compile_DPH(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_EX2(Instruction instr) {
+void JitShader::Compile_EX2(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
MOVSS(XMM0, R(SRC1));
@@ -404,7 +432,7 @@ void JitCompiler::Compile_EX2(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_LG2(Instruction instr) {
+void JitShader::Compile_LG2(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
MOVSS(XMM0, R(SRC1));
@@ -417,14 +445,14 @@ void JitCompiler::Compile_LG2(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_MUL(Instruction instr) {
+void JitShader::Compile_MUL(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_SGE(Instruction instr) {
+void JitShader::Compile_SGE(Instruction instr) {
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::SGEI) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
@@ -439,7 +467,7 @@ void JitCompiler::Compile_SGE(Instruction instr) {
Compile_DestEnable(instr, SRC2);
}
-void JitCompiler::Compile_SLT(Instruction instr) {
+void JitShader::Compile_SLT(Instruction instr) {
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::SLTI) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
@@ -454,7 +482,7 @@ void JitCompiler::Compile_SLT(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_FLR(Instruction instr) {
+void JitShader::Compile_FLR(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
if (Common::GetCPUCaps().sse4_1) {
@@ -467,7 +495,7 @@ void JitCompiler::Compile_FLR(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_MAX(Instruction instr) {
+void JitShader::Compile_MAX(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
// SSE semantics match PICA200 ones: In case of NaN, SRC2 is returned.
@@ -475,7 +503,7 @@ void JitCompiler::Compile_MAX(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_MIN(Instruction instr) {
+void JitShader::Compile_MIN(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
// SSE semantics match PICA200 ones: In case of NaN, SRC2 is returned.
@@ -483,7 +511,7 @@ void JitCompiler::Compile_MIN(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_MOVA(Instruction instr) {
+void JitShader::Compile_MOVA(Instruction instr) {
SwizzlePattern swiz = { g_state.vs.swizzle_data[instr.common.operand_desc_id] };
if (!swiz.DestComponentEnabled(0) && !swiz.DestComponentEnabled(1)) {
@@ -528,12 +556,12 @@ void JitCompiler::Compile_MOVA(Instruction instr) {
}
}
-void JitCompiler::Compile_MOV(Instruction instr) {
+void JitShader::Compile_MOV(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_RCP(Instruction instr) {
+void JitShader::Compile_RCP(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
// TODO(bunnei): RCPSS is a pretty rough approximation, this might cause problems if Pica
@@ -544,7 +572,7 @@ void JitCompiler::Compile_RCP(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_RSQ(Instruction instr) {
+void JitShader::Compile_RSQ(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
// TODO(bunnei): RSQRTSS is a pretty rough approximation, this might cause problems if Pica
@@ -555,36 +583,41 @@ void JitCompiler::Compile_RSQ(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_NOP(Instruction instr) {
+void JitShader::Compile_NOP(Instruction instr) {
}
-void JitCompiler::Compile_END(Instruction instr) {
+void JitShader::Compile_END(Instruction instr) {
ABI_PopRegistersAndAdjustStack(ABI_ALL_CALLEE_SAVED, 8);
RET();
}
-void JitCompiler::Compile_CALL(Instruction instr) {
- unsigned offset = instr.flow_control.dest_offset;
- while (offset < (instr.flow_control.dest_offset + instr.flow_control.num_instructions)) {
- Compile_NextInstr(&offset);
- }
+void JitShader::Compile_CALL(Instruction instr) {
+ // Push offset of the return
+ PUSH(64, Imm32(instr.flow_control.dest_offset + instr.flow_control.num_instructions));
+
+ // Call the subroutine
+ FixupBranch b = CALL();
+ fixup_branches.push_back({ b, instr.flow_control.dest_offset });
+
+ // Skip over the return offset that's on the stack
+ ADD(64, R(RSP), Imm32(8));
}
-void JitCompiler::Compile_CALLC(Instruction instr) {
+void JitShader::Compile_CALLC(Instruction instr) {
Compile_EvaluateCondition(instr);
FixupBranch b = J_CC(CC_Z, true);
Compile_CALL(instr);
SetJumpTarget(b);
}
-void JitCompiler::Compile_CALLU(Instruction instr) {
+void JitShader::Compile_CALLU(Instruction instr) {
Compile_UniformCondition(instr);
FixupBranch b = J_CC(CC_Z, true);
Compile_CALL(instr);
SetJumpTarget(b);
}
-void JitCompiler::Compile_CMP(Instruction instr) {
+void JitShader::Compile_CMP(Instruction instr) {
using Op = Instruction::Common::CompareOpType::Op;
Op op_x = instr.common.compare_op.x;
Op op_y = instr.common.compare_op.y;
@@ -627,7 +660,7 @@ void JitCompiler::Compile_CMP(Instruction instr) {
SHR(64, R(COND1), Imm8(63));
}
-void JitCompiler::Compile_MAD(Instruction instr) {
+void JitShader::Compile_MAD(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.mad.src1, SRC1);
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::MADI) {
@@ -644,9 +677,8 @@ void JitCompiler::Compile_MAD(Instruction instr) {
Compile_DestEnable(instr, SRC1);
}
-void JitCompiler::Compile_IF(Instruction instr) {
- ASSERT_MSG(instr.flow_control.dest_offset > *offset_ptr, "Backwards if-statements (%d -> %d) not supported",
- *offset_ptr, instr.flow_control.dest_offset.Value());
+void JitShader::Compile_IF(Instruction instr) {
+ Compile_Assert(instr.flow_control.dest_offset >= program_counter, "Backwards if-statements not supported");
// Evaluate the "IF" condition
if (instr.opcode.Value() == OpCode::Id::IFU) {
@@ -676,15 +708,14 @@ void JitCompiler::Compile_IF(Instruction instr) {
SetJumpTarget(b2);
}
-void JitCompiler::Compile_LOOP(Instruction instr) {
- ASSERT_MSG(instr.flow_control.dest_offset > *offset_ptr, "Backwards loops (%d -> %d) not supported",
- *offset_ptr, instr.flow_control.dest_offset.Value());
- ASSERT_MSG(!looping, "Nested loops not supported");
+void JitShader::Compile_LOOP(Instruction instr) {
+ Compile_Assert(instr.flow_control.dest_offset >= program_counter, "Backwards loops not supported");
+ Compile_Assert(!looping, "Nested loops not supported");
looping = true;
- int offset = offsetof(decltype(g_state.vs.uniforms), i) + (instr.flow_control.int_uniform_id * sizeof(Math::Vec4<u8>));
- MOV(32, R(LOOPCOUNT), MDisp(UNIFORMS, offset));
+ int offset = ShaderSetup::UniformOffset(RegisterType::IntUniform, instr.flow_control.int_uniform_id);
+ MOV(32, R(LOOPCOUNT), MDisp(SETUP, offset));
MOV(32, R(LOOPCOUNT_REG), R(LOOPCOUNT));
SHR(32, R(LOOPCOUNT_REG), Imm8(8));
AND(32, R(LOOPCOUNT_REG), Imm32(0xff)); // Y-component is the start
@@ -705,10 +736,7 @@ void JitCompiler::Compile_LOOP(Instruction instr) {
looping = false;
}
-void JitCompiler::Compile_JMP(Instruction instr) {
- ASSERT_MSG(instr.flow_control.dest_offset > *offset_ptr, "Backwards jumps (%d -> %d) not supported",
- *offset_ptr, instr.flow_control.dest_offset.Value());
-
+void JitShader::Compile_JMP(Instruction instr) {
if (instr.opcode.Value() == OpCode::Id::JMPC)
Compile_EvaluateCondition(instr);
else if (instr.opcode.Value() == OpCode::Id::JMPU)
@@ -718,30 +746,38 @@ void JitCompiler::Compile_JMP(Instruction instr) {
bool inverted_condition = (instr.opcode.Value() == OpCode::Id::JMPU) &&
(instr.flow_control.num_instructions & 1);
+
FixupBranch b = J_CC(inverted_condition ? CC_Z : CC_NZ, true);
+ fixup_branches.push_back({ b, instr.flow_control.dest_offset });
+}
- Compile_Block(instr.flow_control.dest_offset);
+void JitShader::Compile_Block(unsigned end) {
+ while (program_counter < end) {
+ Compile_NextInstr();
+ }
+}
+
+void JitShader::Compile_Return() {
+ // Peek return offset on the stack and check if we're at that offset
+ MOV(64, R(RAX), MDisp(RSP, 8));
+ CMP(32, R(RAX), Imm32(program_counter));
+ // If so, jump back to before CALL
+ FixupBranch b = J_CC(CC_NZ, true);
+ RET();
SetJumpTarget(b);
}
-void JitCompiler::Compile_Block(unsigned end) {
- // Save current offset pointer
- unsigned* prev_offset_ptr = offset_ptr;
- unsigned offset = *prev_offset_ptr;
+void JitShader::Compile_NextInstr() {
+ if (std::binary_search(return_offsets.begin(), return_offsets.end(), program_counter)) {
+ Compile_Return();
+ }
- while (offset < end)
- Compile_NextInstr(&offset);
+ ASSERT_MSG(code_ptr[program_counter] == nullptr, "Tried to compile already compiled shader location!");
+ code_ptr[program_counter] = GetCodePtr();
- // Restore current offset pointer
- offset_ptr = prev_offset_ptr;
- *offset_ptr = offset;
-}
+ Instruction instr = GetVertexShaderInstruction(program_counter++);
-void JitCompiler::Compile_NextInstr(unsigned* offset) {
- offset_ptr = offset;
-
- Instruction instr = *(Instruction*)&g_state.vs.program_code[(*offset_ptr)++];
OpCode::Id opcode = instr.opcode.Value();
auto instr_func = instr_table[static_cast<unsigned>(opcode)];
@@ -755,15 +791,43 @@ void JitCompiler::Compile_NextInstr(unsigned* offset) {
}
}
-CompiledShader* JitCompiler::Compile() {
- const u8* start = GetCodePtr();
- unsigned offset = g_state.regs.vs.main_offset;
+void JitShader::FindReturnOffsets() {
+ return_offsets.clear();
+
+ for (size_t offset = 0; offset < g_state.vs.program_code.size(); ++offset) {
+ Instruction instr = GetVertexShaderInstruction(offset);
+
+ switch (instr.opcode.Value()) {
+ case OpCode::Id::CALL:
+ case OpCode::Id::CALLC:
+ case OpCode::Id::CALLU:
+ return_offsets.push_back(instr.flow_control.dest_offset + instr.flow_control.num_instructions);
+ break;
+ default:
+ break;
+ }
+ }
+
+ // Sort for efficient binary search later
+ std::sort(return_offsets.begin(), return_offsets.end());
+}
+
+void JitShader::Compile() {
+ // Reset flow control state
+ program = (CompiledShader*)GetCodePtr();
+ program_counter = 0;
+ looping = false;
+ code_ptr.fill(nullptr);
+ fixup_branches.clear();
+
+ // Find all `CALL` instructions and identify return locations
+ FindReturnOffsets();
// The stack pointer is 8 modulo 16 at the entry of a procedure
ABI_PushRegistersAndAdjustStack(ABI_ALL_CALLEE_SAVED, 8);
- MOV(PTRBITS, R(REGISTERS), R(ABI_PARAM1));
- MOV(PTRBITS, R(UNIFORMS), ImmPtr(&g_state.vs.uniforms));
+ MOV(PTRBITS, R(SETUP), R(ABI_PARAM1));
+ MOV(PTRBITS, R(STATE), R(ABI_PARAM2));
// Zero address/loop registers
XOR(64, R(ADDROFFS_REG_0), R(ADDROFFS_REG_0));
@@ -780,21 +844,31 @@ CompiledShader* JitCompiler::Compile() {
MOV(PTRBITS, R(RAX), ImmPtr(&neg));
MOVAPS(NEGBIT, MatR(RAX));
- looping = false;
+ // Jump to start of the shader program
+ JMPptr(R(ABI_PARAM3));
+
+ // Compile entire program
+ Compile_Block(static_cast<unsigned>(g_state.vs.program_code.size()));
- while (offset < g_state.vs.program_code.size()) {
- Compile_NextInstr(&offset);
+ // Set the target for any incomplete branches now that the entire shader program has been emitted
+ for (const auto& branch : fixup_branches) {
+ SetJumpTarget(branch.first, code_ptr[branch.second]);
}
- return (CompiledShader*)start;
-}
+ // Free memory that's no longer needed
+ return_offsets.clear();
+ return_offsets.shrink_to_fit();
+ fixup_branches.clear();
+ fixup_branches.shrink_to_fit();
+
+ uintptr_t size = reinterpret_cast<uintptr_t>(GetCodePtr()) - reinterpret_cast<uintptr_t>(program);
+ ASSERT_MSG(size <= MAX_SHADER_SIZE, "Compiled a shader that exceeds the allocated size!");
-JitCompiler::JitCompiler() {
- AllocCodeSpace(jit_cache_size);
+ LOG_DEBUG(HW_GPU, "Compiled shader size=%lu", size);
}
-void JitCompiler::Clear() {
- ClearCodeSpace();
+JitShader::JitShader() {
+ AllocCodeSpace(MAX_SHADER_SIZE);
}
} // namespace Shader
diff --git a/src/video_core/shader/shader_jit_x64.h b/src/video_core/shader/shader_jit_x64.h
index 5357c964b..5468459d4 100644
--- a/src/video_core/shader/shader_jit_x64.h
+++ b/src/video_core/shader/shader_jit_x64.h
@@ -4,11 +4,17 @@
#pragma once
+#include <array>
+#include <cstddef>
+#include <utility>
+#include <vector>
+
#include <nihstro/shader_bytecode.h>
+#include "common/bit_set.h"
+#include "common/common_types.h"
#include "common/x64/emitter.h"
-#include "video_core/pica.h"
#include "video_core/shader/shader.h"
using nihstro::Instruction;
@@ -19,24 +25,22 @@ namespace Pica {
namespace Shader {
-/// Memory needed to be available to compile the next shader (otherwise, clear the cache)
-constexpr size_t jit_shader_size = 1024 * 512;
-/// Memory allocated for the JIT code space cache
-constexpr size_t jit_cache_size = 1024 * 1024 * 8;
-
-using CompiledShader = void(void* registers);
+/// Memory allocated for each compiled shader (64Kb)
+constexpr size_t MAX_SHADER_SIZE = 1024 * 64;
/**
* This class implements the shader JIT compiler. It recompiles a Pica shader program into x86_64
* code that can be executed on the host machine directly.
*/
-class JitCompiler : public Gen::XCodeBlock {
+class JitShader : public Gen::XCodeBlock {
public:
- JitCompiler();
+ JitShader();
- CompiledShader* Compile();
+ void Run(const ShaderSetup& setup, UnitState<false>& state, unsigned offset) const {
+ program(&setup, &state, code_ptr[offset]);
+ }
- void Clear();
+ void Compile();
void Compile_ADD(Instruction instr);
void Compile_DP3(Instruction instr);
@@ -66,8 +70,9 @@ public:
void Compile_MAD(Instruction instr);
private:
+
void Compile_Block(unsigned end);
- void Compile_NextInstr(unsigned* offset);
+ void Compile_NextInstr();
void Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRegister src_reg, Gen::X64Reg dest);
void Compile_DestEnable(Instruction instr, Gen::X64Reg dest);
@@ -81,13 +86,39 @@ private:
void Compile_EvaluateCondition(Instruction instr);
void Compile_UniformCondition(Instruction instr);
+ /**
+ * Emits the code to conditionally return from a subroutine envoked by the `CALL` instruction.
+ */
+ void Compile_Return();
+
BitSet32 PersistentCallerSavedRegs();
- /// Pointer to the variable that stores the current Pica code offset. Used to handle nested code blocks.
- unsigned* offset_ptr = nullptr;
+ /**
+ * Assertion evaluated at compile-time, but only triggered if executed at runtime.
+ * @param msg Message to be logged if the assertion fails.
+ */
+ void Compile_Assert(bool condition, const char* msg);
+
+ /**
+ * Analyzes the entire shader program for `CALL` instructions before emitting any code,
+ * identifying the locations where a return needs to be inserted.
+ */
+ void FindReturnOffsets();
+
+ /// Mapping of Pica VS instructions to pointers in the emitted code
+ std::array<const u8*, 1024> code_ptr;
+
+ /// Offsets in code where a return needs to be inserted
+ std::vector<unsigned> return_offsets;
+
+ unsigned program_counter = 0; ///< Offset of the next instruction to decode
+ bool looping = false; ///< True if compiling a loop, used to check for nested loops
+
+ /// Branches that need to be fixed up once the entire shader program is compiled
+ std::vector<std::pair<Gen::FixupBranch, unsigned>> fixup_branches;
- /// Set to true if currently in a loop, used to check for the existence of nested loops
- bool looping = false;
+ using CompiledShader = void(const void* setup, void* state, const u8* start_addr);
+ CompiledShader* program = nullptr;
};
} // Shader
generated by cgit v1.2.3 (git 2.25.1) at 2025-12-28 02:06:31 +0000