3 files changed, 108 insertions, 16 deletions

diff --git a/appveyor.yml b/appveyor.yml
index 4f928adb5..72cda26a7 100644
--- a/appveyor.yml
+++ b/appveyor.yml
@@ -53,7 +53,7 @@ build_script:
           # https://www.appveyor.com/docs/build-phase
           msbuild msvc_build/yuzu.sln /maxcpucount /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll"
         } else {
-          C:\msys64\usr\bin\bash.exe -lc 'mingw32-make -j4 -C mingw_build/ 2>&1'
+          C:\msys64\usr\bin\bash.exe -lc 'mingw32-make -C mingw_build/ 2>&1'
         }
 
 after_build:
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index 93654eb66..4eb507325 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -237,9 +237,22 @@ union Instruction {
             std::memcpy(&result, &imm, sizeof(imm));
             return result;
         }
+
+        s32 GetSignedImm20_20() const {
+            u32 immediate = static_cast<u32>(imm20_19 | (negate_imm << 19));
+            // Sign extend the 20-bit value.
+            u32 mask = 1U << (20 - 1);
+            return static_cast<s32>((immediate ^ mask) - mask);
+        }
     } alu;
 
     union {
+        BitField<39, 5, u64> shift_amount;
+        BitField<48, 1, u64> negate_b;
+        BitField<49, 1, u64> negate_a;
+    } iscadd;
+
+    union {
         BitField<48, 1, u64> negate_b;
         BitField<49, 1, u64> negate_c;
     } ffma;
@@ -328,15 +341,16 @@ union Instruction {
     } texs;
 
     union {
-        BitField<20, 5, u64> target;
+        BitField<20, 24, u64> target;
         BitField<5, 1, u64> constant_buffer;
 
         s32 GetBranchTarget() const {
             // Sign extend the branch target offset
-            u32 mask = 1U << (5 - 1);
+            u32 mask = 1U << (24 - 1);
             u32 value = static_cast<u32>(target);
-            // The branch offset is relative to the next instruction, so add 1 to it.
-            return static_cast<s32>((value ^ mask) - mask) + 1;
+            // The branch offset is relative to the next instruction and is stored in bytes, so
+            // divide it by the size of an instruction and add 1 to it.
+            return static_cast<s32>((value ^ mask) - mask) / sizeof(Instruction) + 1;
         }
     } bra;
 
@@ -378,6 +392,9 @@ public:
         FMUL_R,
         FMUL_IMM,
         FMUL32_IMM,
+        ISCADD_C, // Scale and Add
+        ISCADD_R,
+        ISCADD_IMM,
         MUFU,  // Multi-Function Operator
         RRO_C, // Range Reduction Operator
         RRO_R,
@@ -399,6 +416,9 @@ public:
         MOV_R,
         MOV_IMM,
         MOV32_IMM,
+        SHL_C,
+        SHL_R,
+        SHL_IMM,
         SHR_C,
         SHR_R,
         SHR_IMM,
@@ -421,6 +441,8 @@ public:
         Trivial,
         Arithmetic,
         Logic,
+        Shift,
+        ScaledAdd,
         Ffma,
         Flow,
         Memory,
@@ -544,6 +566,9 @@ private:
             INST("0101110001101---", Id::FMUL_R, Type::Arithmetic, "FMUL_R"),
             INST("0011100-01101---", Id::FMUL_IMM, Type::Arithmetic, "FMUL_IMM"),
             INST("00011110--------", Id::FMUL32_IMM, Type::Arithmetic, "FMUL32_IMM"),
+            INST("0100110000011---", Id::ISCADD_C, Type::ScaledAdd, "ISCADD_C"),
+            INST("0101110000011---", Id::ISCADD_R, Type::ScaledAdd, "ISCADD_R"),
+            INST("0011100-00011---", Id::ISCADD_IMM, Type::ScaledAdd, "ISCADD_IMM"),
             INST("0101000010000---", Id::MUFU, Type::Arithmetic, "MUFU"),
             INST("0100110010010---", Id::RRO_C, Type::Arithmetic, "RRO_C"),
             INST("0101110010010---", Id::RRO_R, Type::Arithmetic, "RRO_R"),
@@ -558,13 +583,16 @@ private:
             INST("0101110010011---", Id::MOV_R, Type::Arithmetic, "MOV_R"),
             INST("0011100-10011---", Id::MOV_IMM, Type::Arithmetic, "MOV_IMM"),
             INST("000000010000----", Id::MOV32_IMM, Type::Arithmetic, "MOV32_IMM"),
-            INST("0100110000101---", Id::SHR_C, Type::Arithmetic, "SHR_C"),
-            INST("0101110000101---", Id::SHR_R, Type::Arithmetic, "SHR_R"),
-            INST("0011100-00101---", Id::SHR_IMM, Type::Arithmetic, "SHR_IMM"),
             INST("0100110001100---", Id::FMNMX_C, Type::Arithmetic, "FMNMX_C"),
             INST("0101110001100---", Id::FMNMX_R, Type::Arithmetic, "FMNMX_R"),
             INST("0011100-01100---", Id::FMNMX_IMM, Type::Arithmetic, "FMNMX_IMM"),
             INST("000001----------", Id::LOP32I, Type::Logic, "LOP32I"),
+            INST("0100110001001---", Id::SHL_C, Type::Shift, "SHL_C"),
+            INST("0101110001001---", Id::SHL_R, Type::Shift, "SHL_R"),
+            INST("0011100-01001---", Id::SHL_IMM, Type::Shift, "SHL_IMM"),
+            INST("0100110000101---", Id::SHR_C, Type::Shift, "SHR_C"),
+            INST("0101110000101---", Id::SHR_R, Type::Shift, "SHR_R"),
+            INST("0011100-00101---", Id::SHR_IMM, Type::Shift, "SHR_IMM"),
             INST("0100110011100---", Id::I2I_C, Type::Conversion, "I2I_C"),
             INST("0101110011100---", Id::I2I_R, Type::Conversion, "I2I_R"),
             INST("01110001-1000---", Id::I2I_IMM, Type::Conversion, "I2I_IMM"),
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 4b14bb47e..4a41e7798 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -115,7 +115,16 @@ private:
             if (const auto opcode = OpCode::Decode(instr)) {
                 switch (opcode->GetId()) {
                 case OpCode::Id::EXIT: {
-                    return exit_method = ExitMethod::AlwaysEnd;
+                    // The EXIT instruction can be predicated, which means that the shader can
+                    // conditionally end on this instruction. We have to consider the case where the
+                    // condition is not met and check the exit method of that other basic block.
+                    using Tegra::Shader::Pred;
+                    if (instr.pred.pred_index == static_cast<u64>(Pred::UnusedIndex)) {
+                        return exit_method = ExitMethod::AlwaysEnd;
+                    } else {
+                        ExitMethod not_met = Scan(offset + 1, end, labels);
+                        return exit_method = ParallelExit(ExitMethod::AlwaysEnd, not_met);
+                    }
                 }
                 case OpCode::Id::BRA: {
                     u32 target = offset + instr.bra.GetBranchTarget();
@@ -645,9 +654,9 @@ private:
     std::string GetPredicateComparison(Tegra::Shader::PredCondition condition) const {
         using Tegra::Shader::PredCondition;
         static const std::unordered_map<PredCondition, const char*> PredicateComparisonStrings = {
-            {PredCondition::LessThan, "<"},      {PredCondition::Equal, "=="},
-            {PredCondition::LessEqual, "<="},    {PredCondition::GreaterThan, ">"},
-            {PredCondition::GreaterEqual, ">="},
+            {PredCondition::LessThan, "<"},   {PredCondition::Equal, "=="},
+            {PredCondition::LessEqual, "<="}, {PredCondition::GreaterThan, ">"},
+            {PredCondition::NotEqual, "!="},  {PredCondition::GreaterEqual, ">="},
         };
 
         auto comparison = PredicateComparisonStrings.find(condition);
@@ -884,6 +893,59 @@ private:
             }
             break;
         }
+
+        case OpCode::Type::Shift: {
+            std::string op_a = regs.GetRegisterAsInteger(instr.gpr8, 0, false);
+            std::string op_b;
+
+            if (instr.is_b_imm) {
+                op_b += '(' + std::to_string(instr.alu.GetSignedImm20_20()) + ')';
+            } else {
+                if (instr.is_b_gpr) {
+                    op_b += regs.GetRegisterAsInteger(instr.gpr20);
+                } else {
+                    op_b += regs.GetUniform(instr.uniform, GLSLRegister::Type::Integer);
+                }
+            }
+
+            switch (opcode->GetId()) {
+            case OpCode::Id::SHL_C:
+            case OpCode::Id::SHL_R:
+            case OpCode::Id::SHL_IMM:
+                regs.SetRegisterToInteger(instr.gpr0, true, 0, op_a + " << " + op_b, 1, 1);
+                break;
+            default: {
+                NGLOG_CRITICAL(HW_GPU, "Unhandled shift instruction: {}", opcode->GetName());
+                UNREACHABLE();
+            }
+            }
+            break;
+        }
+
+        case OpCode::Type::ScaledAdd: {
+            std::string op_a = regs.GetRegisterAsInteger(instr.gpr8);
+
+            if (instr.iscadd.negate_a)
+                op_a = '-' + op_a;
+
+            std::string op_b = instr.iscadd.negate_b ? "-" : "";
+
+            if (instr.is_b_imm) {
+                op_b += '(' + std::to_string(instr.alu.GetSignedImm20_20()) + ')';
+            } else {
+                if (instr.is_b_gpr) {
+                    op_b += regs.GetRegisterAsInteger(instr.gpr20);
+                } else {
+                    op_b += regs.GetUniform(instr.uniform, GLSLRegister::Type::Integer);
+                }
+            }
+
+            std::string shift = std::to_string(instr.iscadd.shift_amount.Value());
+
+            regs.SetRegisterToInteger(instr.gpr0, true, 0,
+                                      "((" + op_a + " << " + shift + ") + " + op_b + ')', 1, 1);
+            break;
+        }
         case OpCode::Type::Ffma: {
             std::string op_a = regs.GetRegisterAsFloat(instr.gpr8);
             std::string op_b = instr.ffma.negate_b ? "-" : "";
@@ -1230,9 +1292,6 @@ private:
         default: {
             switch (opcode->GetId()) {
             case OpCode::Id::EXIT: {
-                ASSERT_MSG(instr.pred.pred_index == static_cast<u64>(Pred::UnusedIndex),
-                           "Predicated exits not implemented");
-
                 // Final color output is currently hardcoded to GPR0-3 for fragment shaders
                 if (stage == Maxwell3D::Regs::ShaderStage::Fragment) {
                     shader.AddLine("color.r = " + regs.GetRegisterAsFloat(0) + ';');
@@ -1242,7 +1301,12 @@ private:
                 }
 
                 shader.AddLine("return true;");
-                offset = PROGRAM_END - 1;
+                if (instr.pred.pred_index == static_cast<u64>(Pred::UnusedIndex)) {
+                    // If this is an unconditional exit then just end processing here, otherwise we
+                    // have to account for the possibility of the condition not being met, so
+                    // continue processing the next instruction.
+                    offset = PROGRAM_END - 1;
+                }
                 break;
             }
             case OpCode::Id::KIL: {