From 305a05f8205ba35422e2106de5c65ec2e1a8fe5f Mon Sep 17 00:00:00 2001 From: Subv Date: Sat, 11 Aug 2018 15:55:11 -0500 Subject: GPU/Shaders: Implemented SSY and SYNC as a way to modify control flow during shader execution. SSY sets the target label to jump to when the SYNC instruction is executed. --- .../renderer_opengl/gl_shader_decompiler.cpp | 31 +++++++++++++++++----- 1 file changed, 25 insertions(+), 6 deletions(-) (limited to 'src') diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 32f06f409..532a47037 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -141,6 +141,15 @@ private: ExitMethod jmp = Scan(target, end, labels); return exit_method = ParallelExit(no_jmp, jmp); } + case OpCode::Id::SSY: { + // The SSY instruction uses a similar encoding as the BRA instruction. + ASSERT_MSG(instr.bra.constant_buffer == 0, + "Constant buffer SSY is not supported"); + u32 target = offset + instr.bra.GetBranchTarget(); + labels.insert(target); + // Continue scanning for an exit method. + break; + } } } } @@ -1668,16 +1677,25 @@ private: break; } case OpCode::Id::SSY: { - // The SSY opcode tells the GPU where to re-converge divergent execution paths, we - // can ignore this when generating GLSL code. + // The SSY opcode tells the GPU where to re-converge divergent execution paths, it + // sets the target of the jump that the SYNC instruction will make. The SSY opcode + // has a similar structure to the BRA opcode. + ASSERT_MSG(instr.bra.constant_buffer == 0, "Constant buffer SSY is not supported"); + + u32 target = offset + instr.bra.GetBranchTarget(); + shader.AddLine("ssy_target = " + std::to_string(target) + "u;"); break; } - case OpCode::Id::SYNC: + case OpCode::Id::SYNC: { + // The SYNC opcode jumps to the address previously set by the SSY opcode ASSERT(instr.flow.cond == Tegra::Shader::FlowCondition::Always); + shader.AddLine("{ jmp_to = ssy_target; break; }"); + break; + } case OpCode::Id::DEPBAR: { - // TODO(Subv): Find out if we actually have to care about these instructions or if + // TODO(Subv): Find out if we actually have to care about this instruction or if // the GLSL compiler takes care of that for us. - LOG_WARNING(HW_GPU, "DEPBAR/SYNC instruction is stubbed"); + LOG_WARNING(HW_GPU, "DEPBAR instruction is stubbed"); break; } default: { @@ -1742,6 +1760,7 @@ private: } else { labels.insert(subroutine.begin); shader.AddLine("uint jmp_to = " + std::to_string(subroutine.begin) + "u;"); + shader.AddLine("uint ssy_target = 0u;"); shader.AddLine("while (true) {"); ++shader.scope; @@ -1757,7 +1776,7 @@ private: u32 compile_end = CompileRange(label, next_label); if (compile_end > next_label && compile_end != PROGRAM_END) { // This happens only when there is a label inside a IF/LOOP block - shader.AddLine("{ jmp_to = " + std::to_string(compile_end) + "u; break; }"); + shader.AddLine(" jmp_to = " + std::to_string(compile_end) + "u; break; }"); labels.emplace(compile_end); } -- cgit v1.2.3 From c1ad9738810d312ec38beccd709ce0853641a94a Mon Sep 17 00:00:00 2001 From: Subv Date: Sat, 11 Aug 2018 16:00:14 -0500 Subject: GPU/Shader: Don't predicate instructions that don't have a predicate field (SSY). --- src/video_core/engines/shader_bytecode.h | 7 +++++++ src/video_core/renderer_opengl/gl_shader_decompiler.cpp | 8 ++++++-- 2 files changed, 13 insertions(+), 2 deletions(-) (limited to 'src') diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h index 3d4557b7e..42c6ea5fd 100644 --- a/src/video_core/engines/shader_bytecode.h +++ b/src/video_core/engines/shader_bytecode.h @@ -598,6 +598,13 @@ public: Unknown, }; + /// Returns whether an opcode has an execution predicate field or not (ie, whether it can be + /// conditionally executed). + static bool IsPredicatedInstruction(Id opcode) { + // TODO(Subv): Add the rest of unpredicated instructions. + return opcode != Id::SSY; + } + class Matcher { public: Matcher(const char* const name, u16 mask, u16 expected, OpCode::Id id, OpCode::Type type) diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 532a47037..8954deb81 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -837,7 +837,11 @@ private: ASSERT_MSG(instr.pred.full_pred != Pred::NeverExecute, "NeverExecute predicate not implemented"); - if (instr.pred.pred_index != static_cast(Pred::UnusedIndex)) { + // Some instructions (like SSY) don't have a predicate field, they are always + // unconditionally executed. + bool can_be_predicated = OpCode::IsPredicatedInstruction(opcode->GetId()); + + if (can_be_predicated && instr.pred.pred_index != static_cast(Pred::UnusedIndex)) { shader.AddLine("if (" + GetPredicateCondition(instr.pred.pred_index, instr.negate_pred != 0) + ')'); @@ -1709,7 +1713,7 @@ private: } // Close the predicate condition scope. - if (instr.pred.pred_index != static_cast(Pred::UnusedIndex)) { + if (can_be_predicated && instr.pred.pred_index != static_cast(Pred::UnusedIndex)) { --shader.scope; shader.AddLine('}'); } -- cgit v1.2.3