summaryrefslogtreecommitdiff
path: root/src/video_core/shader/shader.cpp
diff options
context:
space:
mode:
authorGravatar bunnei2015-08-15 18:26:12 -0400
committerGravatar bunnei2015-08-15 18:26:12 -0400
commitd852c4ecc71f800b552978362947a5eb5a524694 (patch)
tree9ea9a86da4027126914e69b12c24d2849fdb2c2d /src/video_core/shader/shader.cpp
parentMerge pull request #1027 from lioncash/debugger (diff)
parentShader: Use a POD struct for registers. (diff)
downloadyuzu-d852c4ecc71f800b552978362947a5eb5a524694.tar.gz
yuzu-d852c4ecc71f800b552978362947a5eb5a524694.tar.xz
yuzu-d852c4ecc71f800b552978362947a5eb5a524694.zip
Merge pull request #1002 from bunnei/shader-jit
Vertex Shader JIT for X86-64
Diffstat (limited to 'src/video_core/shader/shader.cpp')
-rw-r--r--src/video_core/shader/shader.cpp145
1 files changed, 145 insertions, 0 deletions
diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp
new file mode 100644
index 000000000..6a27a8015
--- /dev/null
+++ b/src/video_core/shader/shader.cpp
@@ -0,0 +1,145 @@
1// Copyright 2015 Citra Emulator Project
2// Licensed under GPLv2 or any later version
3// Refer to the license.txt file included.
4
5#include <memory>
6#include <unordered_map>
7
8#include "common/hash.h"
9#include "common/make_unique.h"
10#include "common/profiler.h"
11
12#include "video_core/debug_utils/debug_utils.h"
13#include "video_core/pica.h"
14#include "video_core/video_core.h"
15
16#include "shader.h"
17#include "shader_interpreter.h"
18
19#ifdef ARCHITECTURE_x86_64
20#include "shader_jit_x64.h"
21#endif // ARCHITECTURE_x86_64
22
23namespace Pica {
24
25namespace Shader {
26
27#ifdef ARCHITECTURE_x86_64
28static std::unordered_map<u64, CompiledShader*> shader_map;
29static JitCompiler jit;
30static CompiledShader* jit_shader;
31#endif // ARCHITECTURE_x86_64
32
33void Setup(UnitState& state) {
34#ifdef ARCHITECTURE_x86_64
35 if (VideoCore::g_shader_jit_enabled) {
36 u64 cache_key = (Common::ComputeHash64(&g_state.vs.program_code, sizeof(g_state.vs.program_code)) ^
37 Common::ComputeHash64(&g_state.vs.swizzle_data, sizeof(g_state.vs.swizzle_data)) ^
38 g_state.regs.vs.main_offset);
39
40 auto iter = shader_map.find(cache_key);
41 if (iter != shader_map.end()) {
42 jit_shader = iter->second;
43 } else {
44 jit_shader = jit.Compile();
45 shader_map.emplace(cache_key, jit_shader);
46 }
47 }
48#endif // ARCHITECTURE_x86_64
49}
50
51void Shutdown() {
52 shader_map.clear();
53}
54
55static Common::Profiling::TimingCategory shader_category("Vertex Shader");
56
57OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes) {
58 auto& config = g_state.regs.vs;
59 auto& setup = g_state.vs;
60
61 Common::Profiling::ScopeTimer timer(shader_category);
62
63 state.program_counter = config.main_offset;
64 state.debug.max_offset = 0;
65 state.debug.max_opdesc_id = 0;
66
67 // Setup input register table
68 const auto& attribute_register_map = config.input_register_map;
69
70 if (num_attributes > 0) state.registers.input[attribute_register_map.attribute0_register] = input.attr[0];
71 if (num_attributes > 1) state.registers.input[attribute_register_map.attribute1_register] = input.attr[1];
72 if (num_attributes > 2) state.registers.input[attribute_register_map.attribute2_register] = input.attr[2];
73 if (num_attributes > 3) state.registers.input[attribute_register_map.attribute3_register] = input.attr[3];
74 if (num_attributes > 4) state.registers.input[attribute_register_map.attribute4_register] = input.attr[4];
75 if (num_attributes > 5) state.registers.input[attribute_register_map.attribute5_register] = input.attr[5];
76 if (num_attributes > 6) state.registers.input[attribute_register_map.attribute6_register] = input.attr[6];
77 if (num_attributes > 7) state.registers.input[attribute_register_map.attribute7_register] = input.attr[7];
78 if (num_attributes > 8) state.registers.input[attribute_register_map.attribute8_register] = input.attr[8];
79 if (num_attributes > 9) state.registers.input[attribute_register_map.attribute9_register] = input.attr[9];
80 if (num_attributes > 10) state.registers.input[attribute_register_map.attribute10_register] = input.attr[10];
81 if (num_attributes > 11) state.registers.input[attribute_register_map.attribute11_register] = input.attr[11];
82 if (num_attributes > 12) state.registers.input[attribute_register_map.attribute12_register] = input.attr[12];
83 if (num_attributes > 13) state.registers.input[attribute_register_map.attribute13_register] = input.attr[13];
84 if (num_attributes > 14) state.registers.input[attribute_register_map.attribute14_register] = input.attr[14];
85 if (num_attributes > 15) state.registers.input[attribute_register_map.attribute15_register] = input.attr[15];
86
87 state.conditional_code[0] = false;
88 state.conditional_code[1] = false;
89
90#ifdef ARCHITECTURE_x86_64
91 if (VideoCore::g_shader_jit_enabled)
92 jit_shader(&state.registers);
93 else
94 RunInterpreter(state);
95#else
96 RunInterpreter(state);
97#endif // ARCHITECTURE_x86_64
98
99#if PICA_DUMP_SHADERS
100 DebugUtils::DumpShader(setup.program_code.data(), state.debug.max_offset, setup.swizzle_data.data(),
101 state.debug.max_opdesc_id, config.main_offset,
102 g_state.regs.vs_output_attributes); // TODO: Don't hardcode VS here
103#endif
104
105 // Setup output data
106 OutputVertex ret;
107 // TODO(neobrain): Under some circumstances, up to 16 attributes may be output. We need to
108 // figure out what those circumstances are and enable the remaining outputs then.
109 for (int i = 0; i < 7; ++i) {
110 const auto& output_register_map = g_state.regs.vs_output_attributes[i]; // TODO: Don't hardcode VS here
111
112 u32 semantics[4] = {
113 output_register_map.map_x, output_register_map.map_y,
114 output_register_map.map_z, output_register_map.map_w
115 };
116
117 for (int comp = 0; comp < 4; ++comp) {
118 float24* out = ((float24*)&ret) + semantics[comp];
119 if (semantics[comp] != Regs::VSOutputAttributes::INVALID) {
120 *out = state.registers.output[i][comp];
121 } else {
122 // Zero output so that attributes which aren't output won't have denormals in them,
123 // which would slow us down later.
124 memset(out, 0, sizeof(*out));
125 }
126 }
127 }
128
129 // The hardware takes the absolute and saturates vertex colors like this, *before* doing interpolation
130 for (int i = 0; i < 4; ++i) {
131 ret.color[i] = float24::FromFloat32(
132 std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f));
133 }
134
135 LOG_TRACE(Render_Software, "Output vertex: pos (%.2f, %.2f, %.2f, %.2f), col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f)",
136 ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(),
137 ret.color.x.ToFloat32(), ret.color.y.ToFloat32(), ret.color.z.ToFloat32(), ret.color.w.ToFloat32(),
138 ret.tc0.u().ToFloat32(), ret.tc0.v().ToFloat32());
139
140 return ret;
141}
142
143} // namespace Shader
144
145} // namespace Pica