POW2/LOG2 for vexptefp/vlogefp.

src/alloy/backend/ivm/ivm_intcode.cc

@@ -2539,6 +2539,64 @@ int Translate_RSQRT(TranslationContext& ctx, Instr* i) {
   return DispatchToC(ctx, i, fns[i->src1.value->type]);
 }
 
+uint32_t IntCode_POW2_F32(IntCodeState& ics, const IntCode* i) {
+  ics.rf[i->dest_reg].f32 = (float)pow(2, ics.rf[i->src1_reg].f32);
+  return IA_NEXT;
+}
+uint32_t IntCode_POW2_F64(IntCodeState& ics, const IntCode* i) {
+  ics.rf[i->dest_reg].f64 = pow(2, ics.rf[i->src1_reg].f64);
+  return IA_NEXT;
+}
+uint32_t IntCode_POW2_V128(IntCodeState& ics, const IntCode* i) {
+  const vec128_t& src1 = ics.rf[i->src1_reg].v128;
+  vec128_t& dest = ics.rf[i->dest_reg].v128;
+  for (size_t i = 0; i < 4; i++) {
+    dest.f4[i] = (float)pow(2, src1.f4[i]);
+  }
+  return IA_NEXT;
+}
+int Translate_POW2(TranslationContext& ctx, Instr* i) {
+  static IntCodeFn fns[] = {
+    IntCode_INVALID_TYPE,
+    IntCode_INVALID_TYPE,
+    IntCode_INVALID_TYPE,
+    IntCode_INVALID_TYPE,
+    IntCode_POW2_F32,
+    IntCode_POW2_F64,
+    IntCode_POW2_V128,
+  };
+  return DispatchToC(ctx, i, fns[i->dest->type]);
+}
+
+uint32_t IntCode_LOG2_F32(IntCodeState& ics, const IntCode* i) {
+  ics.rf[i->dest_reg].f32 = log2(ics.rf[i->src1_reg].f32);
+  return IA_NEXT;
+}
+uint32_t IntCode_LOG2_F64(IntCodeState& ics, const IntCode* i) {
+  ics.rf[i->dest_reg].f64 = log2(ics.rf[i->src1_reg].f64);
+  return IA_NEXT;
+}
+uint32_t IntCode_LOG2_V128(IntCodeState& ics, const IntCode* i) {
+  const vec128_t& src1 = ics.rf[i->src1_reg].v128;
+  vec128_t& dest = ics.rf[i->dest_reg].v128;
+  for (size_t i = 0; i < 4; i++) {
+    dest.f4[i] = log2(src1.f4[i]);
+  }
+  return IA_NEXT;
+}
+int Translate_LOG2(TranslationContext& ctx, Instr* i) {
+  static IntCodeFn fns[] = {
+    IntCode_INVALID_TYPE,
+    IntCode_INVALID_TYPE,
+    IntCode_INVALID_TYPE,
+    IntCode_INVALID_TYPE,
+    IntCode_LOG2_F32,
+    IntCode_LOG2_F64,
+    IntCode_LOG2_V128,
+  };
+  return DispatchToC(ctx, i, fns[i->dest->type]);
+}
+
 uint32_t IntCode_AND_I8_I8(IntCodeState& ics, const IntCode* i) {
   ics.rf[i->dest_reg].i8 = ics.rf[i->src1_reg].i8 & ics.rf[i->src2_reg].i8;
   return IA_NEXT;
@@ -3223,6 +3281,8 @@ static const TranslateFn dispatch_table[] = {
   Translate_ABS,
   Translate_SQRT,
   Translate_RSQRT,
+  Translate_POW2,
+  Translate_LOG2,
   Translate_DOT_PRODUCT_3,
   Translate_DOT_PRODUCT_4,
 

src/alloy/backend/x64/lowering/lowering_sequences.cc

@@ -490,6 +490,18 @@ void alloy::backend::x64::lowering::RegisterSequences(LoweringTable* table) {
     return true;
   });
 
+  table->AddSequence(OPCODE_POW2, [](LIRBuilder& lb, Instr*& instr) {
+    // TODO
+    instr = instr->next;
+    return true;
+  });
+
+  table->AddSequence(OPCODE_LOG2, [](LIRBuilder& lb, Instr*& instr) {
+    // TODO
+    instr = instr->next;
+    return true;
+  });
+
   table->AddSequence(OPCODE_DOT_PRODUCT_3, [](LIRBuilder& lb, Instr*& instr) {
     // TODO
     instr = instr->next;

src/alloy/frontend/ppc/ppc_emit_altivec.cc

@@ -703,22 +703,30 @@ XEEMITTER(vctuxs,         0x1000038A, VX  )(PPCHIRBuilder& f, InstrData& i) {
   return 1;
 }
 
+int InstrEmit_vexptefp_(PPCHIRBuilder& f, uint32_t vd, uint32_t vb) {
+  // (VD) <- pow2(VB)
+  Value* v = f.Pow2(f.LoadVR(vb));
+  f.StoreVR(vd, v);
+  return 0;
+}
 XEEMITTER(vexptefp,       0x1000018A, VX  )(PPCHIRBuilder& f, InstrData& i) {
-  XEINSTRNOTIMPLEMENTED();
+  return InstrEmit_vexptefp_(f, i.VX.VD, i.VX.VB);
-  return 1;
 }
 XEEMITTER(vexptefp128,    VX128_3(6, 1712), VX128_3)(PPCHIRBuilder& f, InstrData& i) {
-  XEINSTRNOTIMPLEMENTED();
+  return InstrEmit_vexptefp_(f, VX128_3_VD128, VX128_3_VB128);
-  return 1;
 }
 
+int InstrEmit_vlogefp_(PPCHIRBuilder& f, uint32_t vd, uint32_t vb) {
+  // (VD) <- log2(VB)
+  Value* v = f.Log2(f.LoadVR(vb));
+  f.StoreVR(vd, v);
+  return 0;
+}
 XEEMITTER(vlogefp,        0x100001CA, VX  )(PPCHIRBuilder& f, InstrData& i) {
-  XEINSTRNOTIMPLEMENTED();
+  return InstrEmit_vlogefp_(f, i.VX.VD, i.VX.VB);
-  return 1;
 }
 XEEMITTER(vlogefp128,     VX128_3(6, 1776), VX128_3)(PPCHIRBuilder& f, InstrData& i) {
-  XEINSTRNOTIMPLEMENTED();
+  return InstrEmit_vlogefp_(f, VX128_3_VD128, VX128_3_VB128);
-  return 1;
 }
 
 int InstrEmit_vmaddfp_(PPCHIRBuilder& f, uint32_t vd, uint32_t va, uint32_t vb, uint32_t vc) {
@@ -1158,7 +1166,7 @@ XEEMITTER(vpkd3d128,      VX128_4(6, 1552), VX128_4)(PPCHIRBuilder& f, InstrData
 int InstrEmit_vrefp_(PPCHIRBuilder& f, uint32_t vd, uint32_t vb) {
   // (VD) <- 1/(VB)
   vec128_t one = { 1, 1, 1, 1 };
-  Value* v = f.Div(f.LoadConstant(one), f.LoadVR(vd));
+  Value* v = f.Div(f.LoadConstant(one), f.LoadVR(vb));
   f.StoreVR(vd, v);
   return 0;
 }
@@ -1180,7 +1188,7 @@ XEEMITTER(vrfim128,       VX128_3(6, 816),  VX128_3)(PPCHIRBuilder& f, InstrData
 
 int InstrEmit_vrfin_(PPCHIRBuilder& f, uint32_t vd, uint32_t vb) {
   // (VD) <- RoundToNearest(VB)
-  Value* v = f.Round(f.LoadVR(vd), ROUND_TO_NEAREST);
+  Value* v = f.Round(f.LoadVR(vb), ROUND_TO_NEAREST);
   f.StoreVR(vd, v);
   return 0;
 }

src/alloy/hir/hir_builder.cc

@@ -1267,6 +1267,28 @@ Value* HIRBuilder::RSqrt(Value* value) {
   return i->dest;
 }
 
+Value* HIRBuilder::Pow2(Value* value) {
+  ASSERT_FLOAT_TYPE(value);
+
+  Instr* i = AppendInstr(
+      OPCODE_POW2_info, 0,
+      AllocValue(value->type));
+  i->set_src1(value);
+  i->src2.value = i->src3.value = NULL;
+  return i->dest;
+}
+
+Value* HIRBuilder::Log2(Value* value) {
+  ASSERT_FLOAT_TYPE(value);
+
+  Instr* i = AppendInstr(
+      OPCODE_LOG2_info, 0,
+      AllocValue(value->type));
+  i->set_src1(value);
+  i->src2.value = i->src3.value = NULL;
+  return i->dest;
+}
+
 Value* HIRBuilder::DotProduct3(Value* value1, Value* value2) {
   ASSERT_VECTOR_TYPE(value1);
   ASSERT_VECTOR_TYPE(value2);

src/alloy/hir/hir_builder.h

@@ -161,6 +161,8 @@ public:
   Value* Abs(Value* value);
   Value* Sqrt(Value* value);
   Value* RSqrt(Value* value);
+  Value* Pow2(Value* value);
+  Value* Log2(Value* value);
   Value* DotProduct3(Value* value1, Value* value2);
   Value* DotProduct4(Value* value1, Value* value2);
 

src/alloy/hir/opcodes.h

@@ -142,6 +142,8 @@ enum Opcode {
   OPCODE_ABS,
   OPCODE_SQRT,
   OPCODE_RSQRT,
+  OPCODE_POW2,
+  OPCODE_LOG2,
   OPCODE_DOT_PRODUCT_3,
   OPCODE_DOT_PRODUCT_4,
 

src/alloy/hir/opcodes.inl

@@ -391,6 +391,18 @@ DEFINE_OPCODE(
     OPCODE_SIG_V_V,
     0);
 
+DEFINE_OPCODE(
+    OPCODE_POW2,
+    "pow2",
+    OPCODE_SIG_V_V,
+    0);
+
+DEFINE_OPCODE(
+    OPCODE_LOG2,
+    "log2",
+    OPCODE_SIG_V_V,
+    0);
+
 DEFINE_OPCODE(
     OPCODE_DOT_PRODUCT_3,
     "dot_product_3",