diff options
Diffstat (limited to 'src/video_core/renderer_opengl')
| -rw-r--r-- | src/video_core/renderer_opengl/gl_shader_decompiler.cpp | 75 |
1 files changed, 47 insertions, 28 deletions
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index f224cb5b2..7a019fc86 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -163,10 +163,11 @@ private: const 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. + case OpCode::Id::SSY: + case OpCode::Id::PBK: { + // The SSY and PBK use a similar encoding as the BRA instruction. ASSERT_MSG(instr.bra.constant_buffer == 0, - "Constant buffer SSY is not supported"); + "Constant buffer branching is not supported"); const u32 target = offset + instr.bra.GetBranchTarget(); labels.insert(target); // Continue scanning for an exit method. @@ -378,8 +379,8 @@ public: * @param reg The destination register to use. * @param elem The element to use for the operation. * @param value The code representing the value to assign. Type has to be half float. - * @param type Half float kind of assignment. - * @param dest_num_components Number of components in the destionation. + * @param merge Half float kind of assignment. + * @param dest_num_components Number of components in the destination. * @param value_num_components Number of components in the value. * @param is_saturated Optional, when True, saturates the provided value. * @param dest_elem Optional, the destination element to use for the operation. @@ -422,6 +423,7 @@ public: * @param reg The destination register to use. * @param elem The element to use for the operation. * @param attribute The input attribute to use as the source value. + * @param input_mode The input mode. * @param vertex The register that decides which vertex to read from (used in GS). */ void SetRegisterToInputAttibute(const Register& reg, u64 elem, Attribute::Index attribute, @@ -951,7 +953,7 @@ private: // Can't assign to the constant predicate. ASSERT(pred != static_cast<u64>(Pred::UnusedIndex)); - const std::string variable = 'p' + std::to_string(pred) + '_' + suffix; + std::string variable = 'p' + std::to_string(pred) + '_' + suffix; shader.AddLine(variable + " = " + value + ';'); declr_predicates.insert(std::move(variable)); } @@ -1058,7 +1060,7 @@ private: /* * Transforms the input string GLSL operand into an unpacked half float pair. * @note This function returns a float type pair instead of a half float pair. This is because - * real half floats are not standarized in GLSL but unpackHalf2x16 (which returns a vec2) is. + * real half floats are not standardized in GLSL but unpackHalf2x16 (which returns a vec2) is. * @param operand Input operand. It has to be an unsigned integer. * @param type How to unpack the unsigned integer to a half float pair. * @param abs Get the absolute value of unpacked half floats. @@ -1232,27 +1234,27 @@ private: } /* - * Emits code to push the input target address to the SSY address stack, incrementing the stack + * Emits code to push the input target address to the flow address stack, incrementing the stack * top. */ - void EmitPushToSSYStack(u32 target) { + void EmitPushToFlowStack(u32 target) { shader.AddLine('{'); ++shader.scope; - shader.AddLine("ssy_stack[ssy_stack_top] = " + std::to_string(target) + "u;"); - shader.AddLine("ssy_stack_top++;"); + shader.AddLine("flow_stack[flow_stack_top] = " + std::to_string(target) + "u;"); + shader.AddLine("flow_stack_top++;"); --shader.scope; shader.AddLine('}'); } /* - * Emits code to pop an address from the SSY address stack, setting the jump address to the + * Emits code to pop an address from the flow address stack, setting the jump address to the * popped address and decrementing the stack top. */ - void EmitPopFromSSYStack() { + void EmitPopFromFlowStack() { shader.AddLine('{'); ++shader.scope; - shader.AddLine("ssy_stack_top--;"); - shader.AddLine("jmp_to = ssy_stack[ssy_stack_top];"); + shader.AddLine("flow_stack_top--;"); + shader.AddLine("jmp_to = flow_stack[flow_stack_top];"); shader.AddLine("break;"); --shader.scope; shader.AddLine('}'); @@ -1479,9 +1481,10 @@ private: break; } case OpCode::Id::FMUL32_IMM: { - regs.SetRegisterToFloat( - instr.gpr0, 0, - regs.GetRegisterAsFloat(instr.gpr8) + " * " + GetImmediate32(instr), 1, 1); + regs.SetRegisterToFloat(instr.gpr0, 0, + regs.GetRegisterAsFloat(instr.gpr8) + " * " + + GetImmediate32(instr), + 1, 1, instr.fmul32.saturate); break; } case OpCode::Id::FADD32I: { @@ -3283,16 +3286,32 @@ private: // 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"); + ASSERT_MSG(instr.bra.constant_buffer == 0, "Constant buffer flow is not supported"); const u32 target = offset + instr.bra.GetBranchTarget(); - EmitPushToSSYStack(target); + EmitPushToFlowStack(target); + break; + } + case OpCode::Id::PBK: { + // PBK pushes to a stack the address where BRK will jump to. This shares stack with + // SSY but using SYNC on a PBK address will kill the shader execution. We don't + // emulate this because it's very unlikely a driver will emit such invalid shader. + ASSERT_MSG(instr.bra.constant_buffer == 0, "Constant buffer PBK is not supported"); + + const u32 target = offset + instr.bra.GetBranchTarget(); + EmitPushToFlowStack(target); break; } 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); - EmitPopFromSSYStack(); + EmitPopFromFlowStack(); + break; + } + case OpCode::Id::BRK: { + // The BRK opcode jumps to the address previously set by the PBK opcode + ASSERT(instr.flow.cond == Tegra::Shader::FlowCondition::Always); + EmitPopFromFlowStack(); break; } case OpCode::Id::DEPBAR: { @@ -3311,8 +3330,8 @@ private: Tegra::Shader::VmadType type, u64 byte_height) { const std::string value = [&]() { if (!is_chunk) { - const auto offset = static_cast<u32>(byte_height * 8); - return "((" + op + " >> " + std::to_string(offset) + ") & 0xff)"; + const auto shift = static_cast<u32>(byte_height * 8); + return "((" + op + " >> " + std::to_string(shift) + ") & 0xff)"; } const std::string zero = "0"; @@ -3446,11 +3465,11 @@ private: labels.insert(subroutine.begin); shader.AddLine("uint jmp_to = " + std::to_string(subroutine.begin) + "u;"); - // TODO(Subv): Figure out the actual depth of the SSY stack, for now it seems - // unlikely that shaders will use 20 nested SSYs. - constexpr u32 SSY_STACK_SIZE = 20; - shader.AddLine("uint ssy_stack[" + std::to_string(SSY_STACK_SIZE) + "];"); - shader.AddLine("uint ssy_stack_top = 0u;"); + // TODO(Subv): Figure out the actual depth of the flow stack, for now it seems + // unlikely that shaders will use 20 nested SSYs and PBKs. + constexpr u32 FLOW_STACK_SIZE = 20; + shader.AddLine("uint flow_stack[" + std::to_string(FLOW_STACK_SIZE) + "];"); + shader.AddLine("uint flow_stack_top = 0u;"); shader.AddLine("while (true) {"); ++shader.scope; |