shader: Add support for fp16 comparisons and misc fixes

7 files changed, 28 insertions, 14 deletions

diff --git a/src/shader_recompiler/frontend/ir/ir_emitter.cpp b/src/shader_recompiler/frontend/ir/ir_emitter.cpp
index 652f6949e..1eda95071 100644
--- a/src/shader_recompiler/frontend/ir/ir_emitter.cpp
+++ b/src/shader_recompiler/frontend/ir/ir_emitter.cpp
@@ -895,15 +895,30 @@ U1 IREmitter::FPGreaterThanEqual(const F16F32F64& lhs, const F16F32F64& rhs, FpC
     }
 }
 
-U1 IREmitter::FPIsNan(const F32& value) {
-    return Inst<U1>(Opcode::FPIsNan32, value);
+U1 IREmitter::FPIsNan(const F16F32F64& value) {
+    switch (value.Type()) {
+    case Type::F16:
+        return Inst<U1>(Opcode::FPIsNan16, value);
+    case Type::F32:
+        return Inst<U1>(Opcode::FPIsNan32, value);
+    case Type::F64:
+        return Inst<U1>(Opcode::FPIsNan64, value);
+    default:
+        ThrowInvalidType(value.Type());
+    }
 }
 
-U1 IREmitter::FPOrdered(const F32& lhs, const F32& rhs) {
+U1 IREmitter::FPOrdered(const F16F32F64& lhs, const F16F32F64& rhs) {
+    if (lhs.Type() != rhs.Type()) {
+        throw InvalidArgument("Mismatching types {} and {}", lhs.Type(), rhs.Type());
+    }
     return LogicalAnd(LogicalNot(FPIsNan(lhs)), LogicalNot(FPIsNan(rhs)));
 }
 
-U1 IREmitter::FPUnordered(const F32& lhs, const F32& rhs) {
+U1 IREmitter::FPUnordered(const F16F32F64& lhs, const F16F32F64& rhs) {
+    if (lhs.Type() != rhs.Type()) {
+        throw InvalidArgument("Mismatching types {} and {}", lhs.Type(), rhs.Type());
+    }
     return LogicalOr(FPIsNan(lhs), FPIsNan(rhs));
 }
 
diff --git a/src/shader_recompiler/frontend/ir/ir_emitter.h b/src/shader_recompiler/frontend/ir/ir_emitter.h
index 8edb11154..ab4537d88 100644
--- a/src/shader_recompiler/frontend/ir/ir_emitter.h
+++ b/src/shader_recompiler/frontend/ir/ir_emitter.h
@@ -161,9 +161,9 @@ public:
                                      FpControl control = {}, bool ordered = true);
     [[nodiscard]] U1 FPGreaterThanEqual(const F16F32F64& lhs, const F16F32F64& rhs,
                                         FpControl control = {}, bool ordered = true);
-    [[nodiscard]] U1 FPIsNan(const F32& value);
-    [[nodiscard]] U1 FPOrdered(const F32& lhs, const F32& rhs);
-    [[nodiscard]] U1 FPUnordered(const F32& lhs, const F32& rhs);
+    [[nodiscard]] U1 FPIsNan(const F16F32F64& value);
+    [[nodiscard]] U1 FPOrdered(const F16F32F64& lhs, const F16F32F64& rhs);
+    [[nodiscard]] U1 FPUnordered(const F16F32F64& lhs, const F16F32F64& rhs);
     [[nodiscard]] F32F64 FPMax(const F32F64& lhs, const F32F64& rhs, FpControl control = {});
     [[nodiscard]] F32F64 FPMin(const F32F64& lhs, const F32F64& rhs, FpControl control = {});
 
diff --git a/src/shader_recompiler/frontend/ir/opcodes.inc b/src/shader_recompiler/frontend/ir/opcodes.inc
index 8471db7b9..884eea7a8 100644
--- a/src/shader_recompiler/frontend/ir/opcodes.inc
+++ b/src/shader_recompiler/frontend/ir/opcodes.inc
@@ -236,7 +236,9 @@ OPCODE(FPOrdGreaterThanEqual64,                             U1,             F64,
 OPCODE(FPUnordGreaterThanEqual16,                           U1,             F16,            F16,                                                            )
 OPCODE(FPUnordGreaterThanEqual32,                           U1,             F32,            F32,                                                            )
 OPCODE(FPUnordGreaterThanEqual64,                           U1,             F64,            F64,                                                            )
+OPCODE(FPIsNan16,                                           U1,             F16,                                                                            )
 OPCODE(FPIsNan32,                                           U1,             F32,                                                                            )
+OPCODE(FPIsNan64,                                           U1,             F64,                                                                            )
 
 // Integer operations
 OPCODE(IAdd32,                                              U32,            U32,            U32,                                                            )
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_add.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_add.cpp
index 19e3401ca..03e7bf047 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_add.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_add.cpp
@@ -6,7 +6,6 @@
 
 namespace Shader::Maxwell {
 namespace {
-
 void HADD2(TranslatorVisitor& v, u64 insn, Merge merge, bool ftz, bool sat, bool abs_a, bool neg_a,
            Swizzle swizzle_a, bool abs_b, bool neg_b, Swizzle swizzle_b, const IR::U32& src_b) {
     union {
@@ -66,7 +65,7 @@ void HADD2(TranslatorVisitor& v, u64 insn, bool sat, bool abs_b, bool neg_b, Swi
     HADD2(v, insn, hadd2.merge, hadd2.ftz != 0, sat, hadd2.abs_a != 0, hadd2.neg_a != 0,
           hadd2.swizzle_a, abs_b, neg_b, swizzle_b, src_b);
 }
-} // namespace
+} // Anonymous namespace
 
 void TranslatorVisitor::HADD2_reg(u64 insn) {
     union {
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_fused_multiply_add.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_fused_multiply_add.cpp
index 2f3996274..8b234bd6a 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_fused_multiply_add.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_fused_multiply_add.cpp
@@ -6,7 +6,6 @@
 
 namespace Shader::Maxwell {
 namespace {
-
 void HFMA2(TranslatorVisitor& v, u64 insn, Merge merge, Swizzle swizzle_a, bool neg_b, bool neg_c,
            Swizzle swizzle_b, Swizzle swizzle_c, const IR::U32& src_b, const IR::U32& src_c,
            bool sat, HalfPrecision precision) {
@@ -85,8 +84,7 @@ void HFMA2(TranslatorVisitor& v, u64 insn, bool neg_b, bool neg_c, Swizzle swizz
     HFMA2(v, insn, hfma2.merge, hfma2.swizzle_a, neg_b, neg_c, swizzle_b, swizzle_c, src_b, src_c,
           sat, precision);
 }
-
-} // namespace
+} // Anonymous namespace
 
 void TranslatorVisitor::HFMA2_reg(u64 insn) {
     union {
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_multiply.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_multiply.cpp
index ff34a8c8f..2451a6ef6 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_multiply.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_multiply.cpp
@@ -6,7 +6,6 @@
 
 namespace Shader::Maxwell {
 namespace {
-
 void HMUL2(TranslatorVisitor& v, u64 insn, Merge merge, bool sat, bool abs_a, bool neg_a,
            Swizzle swizzle_a, bool abs_b, bool neg_b, Swizzle swizzle_b, const IR::U32& src_b,
            HalfPrecision precision) {
@@ -79,7 +78,7 @@ void HMUL2(TranslatorVisitor& v, u64 insn, bool sat, bool abs_a, bool neg_a, boo
     HMUL2(v, insn, hmul2.merge, sat, abs_a, neg_a, hmul2.swizzle_a, abs_b, neg_b, swizzle_b, src_b,
           hmul2.precision);
 }
-} // namespace
+} // Anonymous namespace
 
 void TranslatorVisitor::HMUL2_reg(u64 insn) {
     union {
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_set.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_set.cpp
index 1d28c0531..7f1f4b88c 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_set.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_set.cpp
@@ -76,6 +76,7 @@ void TranslatorVisitor::HSET2_reg(u64 insn) {
         BitField<35, 4, FPCompareOp> compare_op;
         BitField<28, 2, Swizzle> swizzle_b;
     } const hset2{insn};
+
     HSET2(*this, insn, GetReg20(insn), hset2.bf != 0, hset2.ftz != 0, hset2.neg_b != 0,
           hset2.abs_b != 0, hset2.compare_op, hset2.swizzle_b);
 }