OPCODE_RECIP

src/xenia/cpu/backend/x64/x64_sequences.cc

@@ -4863,6 +4863,28 @@ struct RSQRT_V128 : Sequence<RSQRT_V128, I<OPCODE_RSQRT, V128Op, V128Op>> {
 };
 EMITTER_OPCODE_TABLE(OPCODE_RSQRT, RSQRT_F32, RSQRT_F64, RSQRT_V128);
 
+// ============================================================================
+// OPCODE_RECIP
+// ============================================================================
+struct RECIP_F32 : Sequence<RECIP_F32, I<OPCODE_RECIP, F32Op, F32Op>> {
+  static void Emit(X64Emitter& e, const EmitArgType& i) {
+    e.vrcpss(i.dest, i.src1);
+  }
+};
+struct RECIP_F64 : Sequence<RECIP_F64, I<OPCODE_RECIP, F64Op, F64Op>> {
+  static void Emit(X64Emitter& e, const EmitArgType& i) {
+    e.vcvtsd2ss(i.dest, i.src1);
+    e.vrcpss(i.dest, i.dest);
+    e.vcvtss2sd(i.dest, i.dest);
+  }
+};
+struct RECIP_V128 : Sequence<RECIP_V128, I<OPCODE_RECIP, V128Op, V128Op>> {
+  static void Emit(X64Emitter& e, const EmitArgType& i) {
+    e.vrcpps(i.dest, i.src1);
+  }
+};
+EMITTER_OPCODE_TABLE(OPCODE_RECIP, RECIP_F32, RECIP_F64, RECIP_V128);
+
 // ============================================================================
 // OPCODE_POW2
 // ============================================================================
@@ -7582,6 +7604,7 @@ void RegisterSequences() {
   Register_OPCODE_ABS();
   Register_OPCODE_SQRT();
   Register_OPCODE_RSQRT();
+  Register_OPCODE_RECIP();
   Register_OPCODE_POW2();
   Register_OPCODE_LOG2();
   Register_OPCODE_DOT_PRODUCT_3();

src/xenia/cpu/compiler/passes/constant_propagation_pass.cc

@@ -487,7 +487,13 @@ bool ConstantPropagationPass::Run(HIRBuilder* builder) {
             i->Remove();
           }
           break;
-
+        case OPCODE_RECIP:
+          if (i->src1.value->IsConstant()) {
+            v->set_from(i->src1.value);
+            v->Recip();
+            i->Remove();
+          }
+          break;
         case OPCODE_AND:
           if (i->src1.value->IsConstant() && i->src2.value->IsConstant()) {
             v->set_from(i->src1.value);

src/xenia/cpu/hir/hir_builder.cc

@@ -1659,6 +1659,15 @@ Value* HIRBuilder::RSqrt(Value* value) {
   return i->dest;
 }
 
+Value* HIRBuilder::Recip(Value* value) {
+  ASSERT_FLOAT_OR_VECTOR_TYPE(value);
+
+  Instr* i = AppendInstr(OPCODE_RECIP_info, 0, AllocValue(value->type));
+  i->set_src1(value);
+  i->src2.value = i->src3.value = NULL;
+  return i->dest;
+}
+
 Value* HIRBuilder::Pow2(Value* value) {
   ASSERT_FLOAT_OR_VECTOR_TYPE(value);
 

src/xenia/cpu/hir/hir_builder.h

@@ -196,6 +196,7 @@ class HIRBuilder {
   Value* Abs(Value* value);
   Value* Sqrt(Value* value);
   Value* RSqrt(Value* value);
+  Value* Recip(Value* value);
   Value* Pow2(Value* value);
   Value* Log2(Value* value);
   Value* DotProduct3(Value* value1, Value* value2);

src/xenia/cpu/hir/opcodes.h

@@ -194,6 +194,7 @@ enum Opcode {
   OPCODE_ABS,
   OPCODE_SQRT,
   OPCODE_RSQRT,
+  OPCODE_RECIP,
   OPCODE_POW2,
   OPCODE_LOG2,
   OPCODE_DOT_PRODUCT_3,

src/xenia/cpu/hir/opcodes.inl

@@ -470,6 +470,12 @@ DEFINE_OPCODE(
     OPCODE_SIG_V_V,
     0)
 
+DEFINE_OPCODE(
+    OPCODE_RECIP,
+    "recip",
+    OPCODE_SIG_V_V,
+    0)
+
 DEFINE_OPCODE(
     OPCODE_POW2,
     "pow2",

src/xenia/cpu/hir/value.cc

@@ -629,6 +629,25 @@ void Value::RSqrt() {
   }
 }
 
+void Value::Recip() {
+  switch (type) {
+    case FLOAT32_TYPE:
+      constant.f32 = 1.0f / constant.f32;
+      break;
+    case FLOAT64_TYPE:
+      constant.f64 = 1.0f / constant.f64;
+      break;
+    case VEC128_TYPE:
+      for (int i = 0; i < 4; i++) {
+        constant.v128.f32[i] = 1.0f / constant.v128.f32[i];
+      }
+      break;
+    default:
+      assert_unhandled_case(type);
+      break;
+  }
+}
+
 void Value::And(Value* other) {
   assert_true(type == other->type);
   switch (type) {

src/xenia/cpu/hir/value.h

@@ -487,6 +487,7 @@ class Value {
   void Abs();
   void Sqrt();
   void RSqrt();
+  void Recip();
   void And(Value* other);
   void Or(Value* other);
   void Xor(Value* other);

src/xenia/cpu/ppc/ppc_emit_altivec.cc

@@ -1210,8 +1210,7 @@ int InstrEmit_vpermwi128(PPCHIRBuilder& f, const InstrData& i) {
 
 int InstrEmit_vrefp_(PPCHIRBuilder& f, uint32_t vd, uint32_t vb) {
   // (VD) <- 1/(VB)
-  vec128_t one = vec128f(1.0f);
+  Value* v = f.Recip(f.LoadVR(vb));
-  Value* v = f.Div(f.LoadConstantVec128(one), f.LoadVR(vb));
   f.StoreVR(vd, v);
   return 0;
 }