diff --git a/src/video_core/shader/shader.cpp b/src/video_core/shader/shader.cpp index 5e8930476..509558fc0 100644 --- a/src/video_core/shader/shader.cpp +++ b/src/video_core/shader/shader.cpp @@ -32,6 +32,12 @@ namespace Shader { static std::unordered_map 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"); +} #endif // ARCHITECTURE_x86_64 void Setup(UnitState& state) { @@ -45,6 +51,12 @@ void Setup(UnitState& state) { if (iter != shader_map.end()) { jit_shader = iter->second; } 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); } @@ -54,7 +66,7 @@ void Setup(UnitState& state) { void Shutdown() { #ifdef ARCHITECTURE_x86_64 - shader_map.clear(); + ClearCache(); #endif // ARCHITECTURE_x86_64 } @@ -135,7 +147,7 @@ OutputVertex Run(UnitState& state, const InputVertex& input, int num_attr std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f)); } - LOG_TRACE(Render_Software, "Output vertex: pos(%.2f, %.2f, %.2f, %.2f), quat(%.2f, %.2f, %.2f, %.2f), " + LOG_TRACE(HW_GPU, "Output vertex: pos(%.2f, %.2f, %.2f, %.2f), quat(%.2f, %.2f, %.2f, %.2f), " "col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f), view(%.2f, %.2f, %.2f)", ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(), ret.quat.x.ToFloat32(), ret.quat.y.ToFloat32(), ret.quat.z.ToFloat32(), ret.quat.w.ToFloat32(), diff --git a/src/video_core/shader/shader_jit_x64.cpp b/src/video_core/shader/shader_jit_x64.cpp index 443bcbbef..dffe051ef 100644 --- a/src/video_core/shader/shader_jit_x64.cpp +++ b/src/video_core/shader/shader_jit_x64.cpp @@ -645,7 +645,8 @@ void JitCompiler::Compile_MAD(Instruction instr) { } void JitCompiler::Compile_IF(Instruction instr) { - ASSERT_MSG(instr.flow_control.dest_offset > *offset_ptr, "Backwards if-statements not supported"); + ASSERT_MSG(instr.flow_control.dest_offset > *offset_ptr, "Backwards if-statements (%d -> %d) not supported", + *offset_ptr, instr.flow_control.dest_offset.Value()); // Evaluate the "IF" condition if (instr.opcode.Value() == OpCode::Id::IFU) { @@ -676,7 +677,8 @@ void JitCompiler::Compile_IF(Instruction instr) { } void JitCompiler::Compile_LOOP(Instruction instr) { - ASSERT_MSG(instr.flow_control.dest_offset > *offset_ptr, "Backwards loops not supported"); + 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"); looping = true; @@ -704,7 +706,8 @@ void JitCompiler::Compile_LOOP(Instruction instr) { } void JitCompiler::Compile_JMP(Instruction instr) { - ASSERT_MSG(instr.flow_control.dest_offset > *offset_ptr, "Backwards jumps not supported"); + ASSERT_MSG(instr.flow_control.dest_offset > *offset_ptr, "Backwards jumps (%d -> %d) not supported", + *offset_ptr, instr.flow_control.dest_offset.Value()); if (instr.opcode.Value() == OpCode::Id::JMPC) Compile_EvaluateCondition(instr); @@ -748,7 +751,7 @@ void JitCompiler::Compile_NextInstr(unsigned* offset) { } else { // Unhandled instruction LOG_CRITICAL(HW_GPU, "Unhandled instruction: 0x%02x (0x%08x)", - instr.opcode.Value().EffectiveOpCode(), instr.hex); + instr.opcode.Value().EffectiveOpCode(), instr.hex); } } @@ -787,7 +790,7 @@ CompiledShader* JitCompiler::Compile() { } JitCompiler::JitCompiler() { - AllocCodeSpace(1024 * 1024 * 4); + AllocCodeSpace(jit_cache_size); } void JitCompiler::Clear() { diff --git a/src/video_core/shader/shader_jit_x64.h b/src/video_core/shader/shader_jit_x64.h index 5ad2d9606..5357c964b 100644 --- a/src/video_core/shader/shader_jit_x64.h +++ b/src/video_core/shader/shader_jit_x64.h @@ -19,6 +19,11 @@ 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); /**