Merge pull request #1519 from ReinUsesLisp/vsetp

gl_shader_decompiler: Implement VSETP

src/video_core/engines/shader_bytecode.h

@@ -214,7 +214,7 @@ enum class IMinMaxExchange : u64 {
     XHi = 3,
 };
 
-enum class VmadType : u64 {
+enum class VideoType : u64 {
     Size16_Low = 0,
     Size16_High = 1,
     Size32 = 2,
@@ -782,6 +782,14 @@ union Instruction {
         BitField<45, 2, PredOperation> op;
     } psetp;
 
+    union {
+        BitField<43, 4, PredCondition> cond;
+        BitField<45, 2, PredOperation> op;
+        BitField<3, 3, u64> pred3;
+        BitField<0, 3, u64> pred0;
+        BitField<39, 3, u64> pred39;
+    } vsetp;
+
     union {
         BitField<12, 3, u64> pred12;
         BitField<15, 1, u64> neg_pred12;
@@ -1154,15 +1162,17 @@ union Instruction {
     union {
         BitField<48, 1, u64> signed_a;
         BitField<38, 1, u64> is_byte_chunk_a;
-        BitField<36, 2, VmadType> type_a;
+        BitField<36, 2, VideoType> type_a;
         BitField<36, 2, u64> byte_height_a;
 
         BitField<49, 1, u64> signed_b;
         BitField<50, 1, u64> use_register_b;
         BitField<30, 1, u64> is_byte_chunk_b;
-        BitField<28, 2, VmadType> type_b;
+        BitField<28, 2, VideoType> type_b;
         BitField<28, 2, u64> byte_height_b;
+    } video;
 
+    union {
         BitField<51, 2, VmadShr> shr;
         BitField<55, 1, u64> saturate; // Saturates the result (a * b + c)
         BitField<47, 1, u64> cc;
@@ -1238,6 +1248,7 @@ public:
         OUT_R, // Emit vertex/primitive
         ISBERD,
         VMAD,
+        VSETP,
         FFMA_IMM, // Fused Multiply and Add
         FFMA_CR,
         FFMA_RC,
@@ -1495,6 +1506,7 @@ private:
             INST("1111101111100---", Id::OUT_R, Type::Trivial, "OUT_R"),
             INST("1110111111010---", Id::ISBERD, Type::Trivial, "ISBERD"),
             INST("01011111--------", Id::VMAD, Type::Trivial, "VMAD"),
+            INST("0101000011110---", Id::VSETP, Type::Trivial, "VSETP"),
             INST("0011001-1-------", Id::FFMA_IMM, Type::Ffma, "FFMA_IMM"),
             INST("010010011-------", Id::FFMA_CR, Type::Ffma, "FFMA_CR"),
             INST("010100011-------", Id::FFMA_RC, Type::Ffma, "FFMA_RC"),

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -1312,6 +1312,63 @@ private:
         }
     }
 
+    /// Unpacks a video instruction operand (e.g. VMAD).
+    std::string GetVideoOperand(const std::string& op, bool is_chunk, bool is_signed,
+                                Tegra::Shader::VideoType 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 std::string zero = "0";
+
+            switch (type) {
+            case Tegra::Shader::VideoType::Size16_Low:
+                return '(' + op + " & 0xffff)";
+            case Tegra::Shader::VideoType::Size16_High:
+                return '(' + op + " >> 16)";
+            case Tegra::Shader::VideoType::Size32:
+                // TODO(Rodrigo): From my hardware tests it becomes a bit "mad" when
+                // this type is used (1 * 1 + 0 == 0x5b800000). Until a better
+                // explanation is found: assert.
+                UNIMPLEMENTED();
+                return zero;
+            case Tegra::Shader::VideoType::Invalid:
+                UNREACHABLE_MSG("Invalid instruction encoding");
+                return zero;
+            default:
+                UNREACHABLE();
+                return zero;
+            }
+        }();
+
+        if (is_signed) {
+            return "int(" + value + ')';
+        }
+        return value;
+    };
+
+    /// Gets the A operand for a video instruction.
+    std::string GetVideoOperandA(Instruction instr) {
+        return GetVideoOperand(regs.GetRegisterAsInteger(instr.gpr8, 0, false),
+                               instr.video.is_byte_chunk_a != 0, instr.video.signed_a,
+                               instr.video.type_a, instr.video.byte_height_a);
+    }
+
+    /// Gets the B operand for a video instruction.
+    std::string GetVideoOperandB(Instruction instr) {
+        if (instr.video.use_register_b) {
+            return GetVideoOperand(regs.GetRegisterAsInteger(instr.gpr20, 0, false),
+                                   instr.video.is_byte_chunk_b != 0, instr.video.signed_b,
+                                   instr.video.type_b, instr.video.byte_height_b);
+        } else {
+            return '(' +
+                   std::to_string(instr.video.signed_b ? static_cast<s16>(instr.alu.GetImm20_16())
+                                                       : instr.alu.GetImm20_16()) +
+                   ')';
+        }
+    }
+
     /**
      * Compiles a single instruction from Tegra to GLSL.
      * @param offset the offset of the Tegra shader instruction.
@@ -3321,87 +3378,51 @@ private:
                 break;
             }
             case OpCode::Id::VMAD: {
-                const bool signed_a = instr.vmad.signed_a == 1;
+                const bool result_signed = instr.video.signed_a == 1 || instr.video.signed_b == 1;
-                const bool signed_b = instr.vmad.signed_b == 1;
+                const std::string op_a = GetVideoOperandA(instr);
-                const bool result_signed = signed_a || signed_b;
+                const std::string op_b = GetVideoOperandB(instr);
-                boost::optional<std::string> forced_result;
-
-                auto Unpack = [&](const std::string& op, bool is_chunk, bool is_signed,
-                                  Tegra::Shader::VmadType type, u64 byte_height) {
-                    const std::string value = [&]() {
-                        if (!is_chunk) {
-                            const auto shift = static_cast<u32>(byte_height * 8);
-                            return "((" + op + " >> " + std::to_string(shift) + ") & 0xff)";
-                        }
-                        const std::string zero = "0";
-
-                        switch (type) {
-                        case Tegra::Shader::VmadType::Size16_Low:
-                            return '(' + op + " & 0xffff)";
-                        case Tegra::Shader::VmadType::Size16_High:
-                            return '(' + op + " >> 16)";
-                        case Tegra::Shader::VmadType::Size32:
-                            // TODO(Rodrigo): From my hardware tests it becomes a bit "mad" when
-                            // this type is used (1 * 1 + 0 == 0x5b800000). Until a better
-                            // explanation is found: assert.
-                            UNREACHABLE_MSG("Unimplemented");
-                            return zero;
-                        case Tegra::Shader::VmadType::Invalid:
-                            // Note(Rodrigo): This flag is invalid according to nvdisasm. From my
-                            // testing (even though it's invalid) this makes the whole instruction
-                            // assign zero to target register.
-                            forced_result = boost::make_optional(zero);
-                            return zero;
-                        default:
-                            UNREACHABLE();
-                            return zero;
-                        }
-                    }();
-
-                    if (is_signed) {
-                        return "int(" + value + ')';
-                    }
-                    return value;
-                };
-
-                const std::string op_a = Unpack(regs.GetRegisterAsInteger(instr.gpr8, 0, false),
-                                                instr.vmad.is_byte_chunk_a != 0, signed_a,
-                                                instr.vmad.type_a, instr.vmad.byte_height_a);
-
-                std::string op_b;
-                if (instr.vmad.use_register_b) {
-                    op_b = Unpack(regs.GetRegisterAsInteger(instr.gpr20, 0, false),
-                                  instr.vmad.is_byte_chunk_b != 0, signed_b, instr.vmad.type_b,
-                                  instr.vmad.byte_height_b);
-                } else {
-                    op_b = '(' +
-                           std::to_string(signed_b ? static_cast<s16>(instr.alu.GetImm20_16())
-                                                   : instr.alu.GetImm20_16()) +
-                           ')';
-                }
-
                 const std::string op_c = regs.GetRegisterAsInteger(instr.gpr39, 0, result_signed);
 
-                std::string result;
+                std::string result = '(' + op_a + " * " + op_b + " + " + op_c + ')';
-                if (forced_result) {
-                    result = *forced_result;
-                } else {
-                    result = '(' + op_a + " * " + op_b + " + " + op_c + ')';
 
-                    switch (instr.vmad.shr) {
+                switch (instr.vmad.shr) {
-                    case Tegra::Shader::VmadShr::Shr7:
+                case Tegra::Shader::VmadShr::Shr7:
-                        result = '(' + result + " >> 7)";
+                    result = '(' + result + " >> 7)";
-                        break;
+                    break;
-                    case Tegra::Shader::VmadShr::Shr15:
+                case Tegra::Shader::VmadShr::Shr15:
-                        result = '(' + result + " >> 15)";
+                    result = '(' + result + " >> 15)";
-                        break;
+                    break;
-                    }
                 }
+
                 regs.SetRegisterToInteger(instr.gpr0, result_signed, 1, result, 1, 1,
                                           instr.vmad.saturate == 1, 0, Register::Size::Word,
                                           instr.vmad.cc);
                 break;
             }
+            case OpCode::Id::VSETP: {
+                const std::string op_a = GetVideoOperandA(instr);
+                const std::string op_b = GetVideoOperandB(instr);
+
+                // We can't use the constant predicate as destination.
+                ASSERT(instr.vsetp.pred3 != static_cast<u64>(Pred::UnusedIndex));
+
+                const std::string second_pred = GetPredicateCondition(instr.vsetp.pred39, false);
+
+                const std::string combiner = GetPredicateCombiner(instr.vsetp.op);
+
+                const std::string predicate = GetPredicateComparison(instr.vsetp.cond, op_a, op_b);
+                // Set the primary predicate to the result of Predicate OP SecondPredicate
+                SetPredicate(instr.vsetp.pred3,
+                             '(' + predicate + ") " + combiner + " (" + second_pred + ')');
+
+                if (instr.vsetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) {
+                    // Set the secondary predicate to the result of !Predicate OP SecondPredicate,
+                    // if enabled
+                    SetPredicate(instr.vsetp.pred0,
+                                 "!(" + predicate + ") " + combiner + " (" + second_pred + ')');
+                }
+                break;
+            }
             default: {
                 LOG_CRITICAL(HW_GPU, "Unhandled instruction: {}", opcode->GetName());
                 UNREACHABLE();