aboutsummaryrefslogtreecommitdiff
path: root/src/video_core/shader/shader.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/shader/shader.cpp')
-rw-r--r--src/video_core/shader/shader.cpp112
1 files changed, 40 insertions, 72 deletions
diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp
index 509558fc0..65dcc9156 100644
--- a/src/video_core/shader/shader.cpp
+++ b/src/video_core/shader/shader.cpp
@@ -2,63 +2,53 @@
// 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/make_unique.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 void ClearCache() {
- shader_map.clear();
- jit.Clear();
- LOG_INFO(HW_GPU, "Shader JIT cache cleared");
-}
+static std::unordered_map<u64, std::unique_ptr<JitShader>> shader_map;
+static const JitShader* jit_shader;
#endif // ARCHITECTURE_x86_64
-void Setup(UnitState<false>& state) {
+void 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
@@ -66,17 +56,15 @@ void Setup(UnitState<false>& state) {
void Shutdown() {
#ifdef ARCHITECTURE_x86_64
- ClearCache();
+ shader_map.clear();
#endif // ARCHITECTURE_x86_64
}
-static Common::Profiling::TimingCategory shader_category("Vertex Shader");
MICROPROFILE_DEFINE(GPU_VertexShader, "GPU", "Vertex Shader", MP_RGB(50, 50, 240));
OutputVertex Run(UnitState<false>& state, const InputVertex& input, int num_attributes) {
auto& config = g_state.regs.vs;
- Common::Profiling::ScopeTimer timer(shader_category);
MICROPROFILE_SCOPE(GPU_VertexShader);
state.program_counter = config.main_offset;
@@ -86,31 +74,15 @@ OutputVertex Run(UnitState<false>& state, const InputVertex& input, int num_attr
// 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(&state.registers, g_state.regs.vs.main_offset);
else
RunInterpreter(state);
#else
@@ -121,15 +93,23 @@ OutputVertex Run(UnitState<false>& state, const InputVertex& input, int num_attr
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
+ unsigned index = 0;
+ for (unsigned i = 0; i < 7; ++i) {
+
+ if (index >= g_state.regs.vs_output_total)
+ break;
+
+ if ((g_state.regs.vs.output_mask & (1 << i)) == 0)
+ continue;
+
+ const auto& output_register_map = g_state.regs.vs_output_attributes[index]; // 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) {
+ for (unsigned comp = 0; comp < 4; ++comp) {
float24* out = ((float24*)&ret) + semantics[comp];
if (semantics[comp] != Regs::VSOutputAttributes::INVALID) {
*out = state.registers.output[i][comp];
@@ -139,10 +119,12 @@ OutputVertex Run(UnitState<false>& state, const InputVertex& input, int num_attr
memset(out, 0, sizeof(*out));
}
}
+
+ index++;
}
// The hardware takes the absolute and saturates vertex colors like this, *before* doing interpolation
- for (int i = 0; i < 4; ++i) {
+ for (unsigned i = 0; i < 4; ++i) {
ret.color[i] = float24::FromFloat32(
std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f));
}
@@ -170,22 +152,8 @@ 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;
generated by cgit v1.2.3 (git 2.25.1) at 2025-12-13 04:40:10 +0000