10 files changed, 93 insertions, 38 deletions

diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h b/src/shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h
index 3da37a2bb..fd73f656c 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h
@@ -46,7 +46,8 @@ inline IR::FmzMode CastFmzMode(FmzMode fmz_mode) {
     case FmzMode::FTZ:
         return IR::FmzMode::FTZ;
     case FmzMode::FMZ:
-        return IR::FmzMode::FMZ;
+        // FMZ is manually handled in the instruction
+        return IR::FmzMode::FTZ;
     case FmzMode::INVALIDFMZ3:
         break;
     }
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_add.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_add.cpp
index 219ffcc6a..76a807d4e 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_add.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_add.cpp
@@ -53,7 +53,7 @@ void FADD(TranslatorVisitor& v, u64 insn, const IR::F32& src_b) {
 } // Anonymous namespace
 
 void TranslatorVisitor::FADD_reg(u64 insn) {
-    FADD(*this, insn, GetRegFloat20(insn));
+    FADD(*this, insn, GetFloatReg20(insn));
 }
 
 void TranslatorVisitor::FADD_cbuf(u64 insn) {
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp
index 758700d3c..c2ca0873b 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp
@@ -51,7 +51,7 @@ void FFMA(TranslatorVisitor& v, u64 insn, const IR::F32& src_b, const IR::F32& s
 } // Anonymous namespace
 
 void TranslatorVisitor::FFMA_reg(u64 insn) {
-    FFMA(*this, insn, GetRegFloat20(insn), GetRegFloat39(insn));
+    FFMA(*this, insn, GetFloatReg20(insn), GetFloatReg39(insn));
 }
 
 void TranslatorVisitor::FFMA_rc(u64) {
@@ -59,7 +59,7 @@ void TranslatorVisitor::FFMA_rc(u64) {
 }
 
 void TranslatorVisitor::FFMA_cr(u64 insn) {
-    FFMA(*this, insn, GetFloatCbuf(insn), GetRegFloat39(insn));
+    FFMA(*this, insn, GetFloatCbuf(insn), GetFloatReg39(insn));
 }
 
 void TranslatorVisitor::FFMA_imm(u64) {
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multi_function.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multi_function.cpp
index ba005fbf4..2f8605619 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multi_function.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multi_function.cpp
@@ -10,7 +10,7 @@
 
 namespace Shader::Maxwell {
 namespace {
-enum class Operation {
+enum class Operation : u64 {
     Cos = 0,
     Sin = 1,
     Ex2 = 2,    // Base 2 exponent
@@ -39,11 +39,11 @@ void TranslatorVisitor::MUFU(u64 insn) {
     IR::F32 value{[&]() -> IR::F32 {
         switch (mufu.operation) {
         case Operation::Cos:
-            return ir.FPCosNotReduced(op_a);
+            return ir.FPCos(op_a);
         case Operation::Sin:
-            return ir.FPSinNotReduced(op_a);
+            return ir.FPSin(op_a);
         case Operation::Ex2:
-            return ir.FPExp2NotReduced(op_a);
+            return ir.FPExp2(op_a);
         case Operation::Lg2:
             return ir.FPLog2(op_a);
         case Operation::Rcp:
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp
index 5c38d3fc1..edf2cadae 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp
@@ -55,9 +55,6 @@ void FMUL(TranslatorVisitor& v, u64 insn, const IR::F32& src_b, FmzMode fmz_mode
     if (cc) {
         throw NotImplementedException("FMUL CC");
     }
-    if (sat) {
-        throw NotImplementedException("FMUL SAT");
-    }
     IR::F32 op_a{v.F(fmul.src_a)};
     if (scale != Scale::None) {
         if (fmz_mode != FmzMode::FTZ || fp_rounding != FpRounding::RN) {
@@ -71,7 +68,20 @@ void FMUL(TranslatorVisitor& v, u64 insn, const IR::F32& src_b, FmzMode fmz_mode
         .rounding{CastFpRounding(fp_rounding)},
         .fmz_mode{CastFmzMode(fmz_mode)},
     };
-    v.F(fmul.dest_reg, v.ir.FPMul(op_a, op_b, fp_control));
+    IR::F32 value{v.ir.FPMul(op_a, op_b, fp_control)};
+    if (fmz_mode == FmzMode::FMZ && !sat) {
+        // Do not implement FMZ if SAT is enabled, as it does the logic for us.
+        // On D3D9 mode, anything * 0 is zero, even NAN and infinity
+        const IR::F32 zero{v.ir.Imm32(0.0f)};
+        const IR::U1 zero_a{v.ir.FPEqual(op_a, zero)};
+        const IR::U1 zero_b{v.ir.FPEqual(op_b, zero)};
+        const IR::U1 any_zero{v.ir.LogicalOr(zero_a, zero_b)};
+        value = IR::F32{v.ir.Select(any_zero, zero, value)};
+    }
+    if (sat) {
+        value = v.ir.FPSaturate(value);
+    }
+    v.F(fmul.dest_reg, value);
 }
 
 void FMUL(TranslatorVisitor& v, u64 insn, const IR::F32& src_b) {
@@ -83,27 +93,33 @@ void FMUL(TranslatorVisitor& v, u64 insn, const IR::F32& src_b) {
         BitField<47, 1, u64> cc;
         BitField<48, 1, u64> neg_b;
         BitField<50, 1, u64> sat;
-    } fmul{insn};
-
+    } const fmul{insn};
     FMUL(v, insn, src_b, fmul.fmz, fmul.fp_rounding, fmul.scale, fmul.sat != 0, fmul.cc != 0,
          fmul.neg_b != 0);
 }
 } // Anonymous namespace
 
 void TranslatorVisitor::FMUL_reg(u64 insn) {
-    return FMUL(*this, insn, GetRegFloat20(insn));
+    return FMUL(*this, insn, GetFloatReg20(insn));
 }
 
-void TranslatorVisitor::FMUL_cbuf(u64) {
-    throw NotImplementedException("FMUL (cbuf)");
+void TranslatorVisitor::FMUL_cbuf(u64 insn) {
+    return FMUL(*this, insn, GetFloatCbuf(insn));
 }
 
-void TranslatorVisitor::FMUL_imm(u64) {
-    throw NotImplementedException("FMUL (imm)");
+void TranslatorVisitor::FMUL_imm(u64 insn) {
+    return FMUL(*this, insn, GetFloatImm20(insn));
 }
 
-void TranslatorVisitor::FMUL32I(u64) {
-    throw NotImplementedException("FMUL32I");
+void TranslatorVisitor::FMUL32I(u64 insn) {
+    union {
+        u64 raw;
+        BitField<52, 1, u64> cc;
+        BitField<53, 2, FmzMode> fmz;
+        BitField<55, 1, u64> sat;
+    } const fmul32i{insn};
+    FMUL(*this, insn, GetFloatImm32(insn), fmul32i.fmz, FpRounding::RN, Scale::None,
+         fmul32i.sat != 0, fmul32i.cc != 0, false);
 }
 
 } // namespace Shader::Maxwell
\ No newline at end of file
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_range_reduction.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_range_reduction.cpp
new file mode 100644
index 000000000..f91b93fad
--- /dev/null
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_range_reduction.cpp
@@ -0,0 +1,41 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+enum class Mode : u64 {
+    SINCOS,
+    EX2,
+};
+
+void RRO(TranslatorVisitor& v, u64 insn, const IR::F32& src) {
+    union {
+        u64 raw;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<39, 1, Mode> mode;
+        BitField<45, 1, u64> neg;
+        BitField<49, 1, u64> abs;
+    } const rro{insn};
+
+    v.F(rro.dest_reg, v.ir.FPAbsNeg(src, rro.abs != 0, rro.neg != 0));
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::RRO_reg(u64 insn) {
+    RRO(*this, insn, GetFloatReg20(insn));
+}
+
+void TranslatorVisitor::RRO_cbuf(u64 insn) {
+    RRO(*this, insn, GetFloatCbuf(insn));
+}
+
+void TranslatorVisitor::RRO_imm(u64) {
+    throw NotImplementedException("RRO (imm)");
+}
+
+} // namespace Shader::Maxwell
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp
index 165d475b9..a5a0e1a9b 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp
@@ -48,11 +48,11 @@ IR::U32 TranslatorVisitor::GetReg39(u64 insn) {
     return X(reg.index);
 }
 
-IR::F32 TranslatorVisitor::GetRegFloat20(u64 insn) {
+IR::F32 TranslatorVisitor::GetFloatReg20(u64 insn) {
     return ir.BitCast<IR::F32>(GetReg20(insn));
 }
 
-IR::F32 TranslatorVisitor::GetRegFloat39(u64 insn) {
+IR::F32 TranslatorVisitor::GetFloatReg39(u64 insn) {
     return ir.BitCast<IR::F32>(GetReg39(insn));
 }
 
@@ -110,6 +110,14 @@ IR::U32 TranslatorVisitor::GetImm32(u64 insn) {
     return ir.Imm32(static_cast<u32>(imm.value));
 }
 
+IR::F32 TranslatorVisitor::GetFloatImm32(u64 insn) {
+    union {
+        u64 raw;
+        BitField<20, 32, u64> value;
+    } const imm{insn};
+    return ir.Imm32(Common::BitCast<f32>(static_cast<u32>(imm.value)));
+}
+
 void TranslatorVisitor::SetZFlag(const IR::U1& value) {
     ir.SetZFlag(value);
 }
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/impl.h b/src/shader_recompiler/frontend/maxwell/translate/impl/impl.h
index 4d4cf2ebf..4e722e205 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/impl.h
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/impl.h
@@ -304,8 +304,8 @@ public:
     [[nodiscard]] IR::U32 GetReg8(u64 insn);
     [[nodiscard]] IR::U32 GetReg20(u64 insn);
     [[nodiscard]] IR::U32 GetReg39(u64 insn);
-    [[nodiscard]] IR::F32 GetRegFloat20(u64 insn);
-    [[nodiscard]] IR::F32 GetRegFloat39(u64 insn);
+    [[nodiscard]] IR::F32 GetFloatReg20(u64 insn);
+    [[nodiscard]] IR::F32 GetFloatReg39(u64 insn);
 
     [[nodiscard]] IR::U32 GetCbuf(u64 insn);
     [[nodiscard]] IR::F32 GetFloatCbuf(u64 insn);
@@ -314,6 +314,7 @@ public:
     [[nodiscard]] IR::F32 GetFloatImm20(u64 insn);
 
     [[nodiscard]] IR::U32 GetImm32(u64 insn);
+    [[nodiscard]] IR::F32 GetFloatImm32(u64 insn);
 
     void SetZFlag(const IR::U1& value);
     void SetSFlag(const IR::U1& value);
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_shift_left.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_shift_left.cpp
index d8a5158b5..20af68852 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_shift_left.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_shift_left.cpp
@@ -50,7 +50,7 @@ void SHL(TranslatorVisitor& v, u64 insn, const IR::U32& unsafe_shift) {
         //
         const IR::U1 is_safe{v.ir.ILessThan(unsafe_shift, v.ir.Imm32(32), false)};
         const IR::U32 unsafe_result{v.ir.ShiftLeftLogical(base, unsafe_shift)};
-        result = v.ir.Select(is_safe, unsafe_result, v.ir.Imm32(0));
+        result = IR::U32{v.ir.Select(is_safe, unsafe_result, v.ir.Imm32(0))};
     }
     v.X(shl.dest_reg, result);
 }
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
index 628cf1c14..4114e10be 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
@@ -721,18 +721,6 @@ void TranslatorVisitor::RET(u64) {
     ThrowNotImplemented(Opcode::RET);
 }
 
-void TranslatorVisitor::RRO_reg(u64) {
-    ThrowNotImplemented(Opcode::RRO_reg);
-}
-
-void TranslatorVisitor::RRO_cbuf(u64) {
-    ThrowNotImplemented(Opcode::RRO_cbuf);
-}
-
-void TranslatorVisitor::RRO_imm(u64) {
-    ThrowNotImplemented(Opcode::RRO_imm);
-}
-
 void TranslatorVisitor::RTT(u64) {
     ThrowNotImplemented(Opcode::RTT);
 }