Improve String.indexOf() intrinsics.

If the code point input is a char, we don't need the slow
path. Also improve the slow-path check (if we do need it)
on arm and arm64 to avoid loading 0xffff into a register.

Bug: 28330359
Change-Id: Ie6514c16126717bb0b11e3c7ab2b60eaa70fed4c

compiler/optimizing/intrinsics_arm.cc

@@ -1115,15 +1115,15 @@ static void GenerateVisitStringIndexOf(HInvoke* invoke,
                                        ArenaAllocator* allocator,
                                        bool start_at_zero) {
   LocationSummary* locations = invoke->GetLocations();
-  Register tmp_reg = locations->GetTemp(0).AsRegister<Register>();
 
   // Note that the null check must have been done earlier.
   DCHECK(!invoke->CanDoImplicitNullCheckOn(invoke->InputAt(0)));
 
   // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
-  // or directly dispatch if we have a constant.
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCode* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
     if (static_cast<uint32_t>(invoke->InputAt(1)->AsIntConstant()->GetValue()) >
         std::numeric_limits<uint16_t>::max()) {
       // Always needs the slow-path. We could directly dispatch to it, but this case should be
@@ -1134,16 +1134,18 @@ static void GenerateVisitStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     Register char_reg = locations->InAt(1).AsRegister<Register>();
-    __ LoadImmediate(tmp_reg, std::numeric_limits<uint16_t>::max());
-    __ cmp(char_reg, ShifterOperand(tmp_reg));
+    // 0xffff is not modified immediate but 0x10000 is, so use `>= 0x10000` instead of `> 0xffff`.
+    __ cmp(char_reg,
+           ShifterOperand(static_cast<uint32_t>(std::numeric_limits<uint16_t>::max()) + 1));
     slow_path = new (allocator) IntrinsicSlowPathARM(invoke);
     codegen->AddSlowPath(slow_path);
-    __ b(slow_path->GetEntryLabel(), HI);
+    __ b(slow_path->GetEntryLabel(), HS);
   }
 
   if (start_at_zero) {
+    Register tmp_reg = locations->GetTemp(0).AsRegister<Register>();
     DCHECK_EQ(tmp_reg, R2);
     // Start-index = 0.
     __ LoadImmediate(tmp_reg, 0);
@@ -1170,7 +1172,7 @@ void IntrinsicLocationsBuilderARM::VisitStringIndexOf(HInvoke* invoke) {
   locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
   locations->SetOut(Location::RegisterLocation(R0));
 
-  // Need a temp for slow-path codepoint compare, and need to send start-index=0.
+  // Need to send start-index=0.
   locations->AddTemp(Location::RegisterLocation(calling_convention.GetRegisterAt(2)));
 }
 
@@ -1190,9 +1192,6 @@ void IntrinsicLocationsBuilderARM::VisitStringIndexOfAfter(HInvoke* invoke) {
   locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
   locations->SetInAt(2, Location::RegisterLocation(calling_convention.GetRegisterAt(2)));
   locations->SetOut(Location::RegisterLocation(R0));
-
-  // Need a temp for slow-path codepoint compare.
-  locations->AddTemp(Location::RequiresRegister());
 }
 
 void IntrinsicCodeGeneratorARM::VisitStringIndexOfAfter(HInvoke* invoke) {

compiler/optimizing/intrinsics_arm64.cc

@@ -1390,15 +1390,15 @@ static void GenerateVisitStringIndexOf(HInvoke* invoke,
                                        ArenaAllocator* allocator,
                                        bool start_at_zero) {
   LocationSummary* locations = invoke->GetLocations();
-  Register tmp_reg = WRegisterFrom(locations->GetTemp(0));
 
   // Note that the null check must have been done earlier.
   DCHECK(!invoke->CanDoImplicitNullCheckOn(invoke->InputAt(0)));
 
   // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
-  // or directly dispatch if we have a constant.
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCodeARM64* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
     if (static_cast<uint32_t>(invoke->InputAt(1)->AsIntConstant()->GetValue()) > 0xFFFFU) {
       // Always needs the slow-path. We could directly dispatch to it, but this case should be
       // rare, so for simplicity just put the full slow-path down and branch unconditionally.
@@ -1408,17 +1408,17 @@ static void GenerateVisitStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     Register char_reg = WRegisterFrom(locations->InAt(1));
-    __ Mov(tmp_reg, 0xFFFF);
-    __ Cmp(char_reg, Operand(tmp_reg));
+    __ Tst(char_reg, 0xFFFF0000);
     slow_path = new (allocator) IntrinsicSlowPathARM64(invoke);
     codegen->AddSlowPath(slow_path);
-    __ B(hi, slow_path->GetEntryLabel());
+    __ B(ne, slow_path->GetEntryLabel());
   }
 
   if (start_at_zero) {
     // Start-index = 0.
+    Register tmp_reg = WRegisterFrom(locations->GetTemp(0));
     __ Mov(tmp_reg, 0);
   }
 
@@ -1442,7 +1442,7 @@ void IntrinsicLocationsBuilderARM64::VisitStringIndexOf(HInvoke* invoke) {
   locations->SetInAt(1, LocationFrom(calling_convention.GetRegisterAt(1)));
   locations->SetOut(calling_convention.GetReturnLocation(Primitive::kPrimInt));
 
-  // Need a temp for slow-path codepoint compare, and need to send start_index=0.
+  // Need to send start_index=0.
   locations->AddTemp(LocationFrom(calling_convention.GetRegisterAt(2)));
 }
 
@@ -1462,9 +1462,6 @@ void IntrinsicLocationsBuilderARM64::VisitStringIndexOfAfter(HInvoke* invoke) {
   locations->SetInAt(1, LocationFrom(calling_convention.GetRegisterAt(1)));
   locations->SetInAt(2, LocationFrom(calling_convention.GetRegisterAt(2)));
   locations->SetOut(calling_convention.GetReturnLocation(Primitive::kPrimInt));
-
-  // Need a temp for slow-path codepoint compare.
-  locations->AddTemp(Location::RequiresRegister());
 }
 
 void IntrinsicCodeGeneratorARM64::VisitStringIndexOfAfter(HInvoke* invoke) {

compiler/optimizing/intrinsics_mips.cc

@@ -2067,10 +2067,11 @@ static void GenerateStringIndexOf(HInvoke* invoke,
   // Note that the null check must have been done earlier.
   DCHECK(!invoke->CanDoImplicitNullCheckOn(invoke->InputAt(0)));
 
-  // Check for code points > 0xFFFF. Either a slow-path check when we
-  // don't know statically, or directly dispatch if we have a constant.
+  // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCodeMIPS* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
     if (!IsUint<16>(invoke->InputAt(1)->AsIntConstant()->GetValue())) {
       // Always needs the slow-path. We could directly dispatch to it,
       // but this case should be rare, so for simplicity just put the
@@ -2081,7 +2082,7 @@ static void GenerateStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     Register char_reg = locations->InAt(1).AsRegister<Register>();
     // The "bltu" conditional branch tests to see if the character value
     // fits in a valid 16-bit (MIPS halfword) value. If it doesn't then

compiler/optimizing/intrinsics_mips64.cc

@@ -1563,10 +1563,11 @@ static void GenerateStringIndexOf(HInvoke* invoke,
   // Note that the null check must have been done earlier.
   DCHECK(!invoke->CanDoImplicitNullCheckOn(invoke->InputAt(0)));
 
-  // Check for code points > 0xFFFF. Either a slow-path check when we
-  // don't know statically, or directly dispatch if we have a constant.
+  // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCodeMIPS64* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
     if (!IsUint<16>(invoke->InputAt(1)->AsIntConstant()->GetValue())) {
       // Always needs the slow-path. We could directly dispatch to it,
       // but this case should be rare, so for simplicity just put the
@@ -1577,7 +1578,7 @@ static void GenerateStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     GpuRegister char_reg = locations->InAt(1).AsRegister<GpuRegister>();
     __ LoadConst32(tmp_reg, std::numeric_limits<uint16_t>::max());
     slow_path = new (allocator) IntrinsicSlowPathMIPS64(invoke);

compiler/optimizing/intrinsics_x86.cc

@@ -1418,9 +1418,10 @@ static void GenerateStringIndexOf(HInvoke* invoke,
   DCHECK_EQ(out, EDI);
 
   // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
-  // or directly dispatch if we have a constant.
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCode* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
     if (static_cast<uint32_t>(invoke->InputAt(1)->AsIntConstant()->GetValue()) >
     std::numeric_limits<uint16_t>::max()) {
       // Always needs the slow-path. We could directly dispatch to it, but this case should be
@@ -1431,7 +1432,7 @@ static void GenerateStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     __ cmpl(search_value, Immediate(std::numeric_limits<uint16_t>::max()));
     slow_path = new (allocator) IntrinsicSlowPathX86(invoke);
     codegen->AddSlowPath(slow_path);

compiler/optimizing/intrinsics_x86_64.cc

@@ -1517,9 +1517,10 @@ static void GenerateStringIndexOf(HInvoke* invoke,
   DCHECK_EQ(out.AsRegister(), RDI);
 
   // Check for code points > 0xFFFF. Either a slow-path check when we don't know statically,
-  // or directly dispatch if we have a constant.
+  // or directly dispatch for a large constant, or omit slow-path for a small constant or a char.
   SlowPathCode* slow_path = nullptr;
-  if (invoke->InputAt(1)->IsIntConstant()) {
+  HInstruction* code_point = invoke->InputAt(1);
+  if (code_point->IsIntConstant()) {
     if (static_cast<uint32_t>(invoke->InputAt(1)->AsIntConstant()->GetValue()) >
     std::numeric_limits<uint16_t>::max()) {
       // Always needs the slow-path. We could directly dispatch to it, but this case should be
@@ -1530,7 +1531,7 @@ static void GenerateStringIndexOf(HInvoke* invoke,
       __ Bind(slow_path->GetExitLabel());
       return;
     }
-  } else {
+  } else if (code_point->GetType() != Primitive::kPrimChar) {
     __ cmpl(search_value, Immediate(std::numeric_limits<uint16_t>::max()));
     slow_path = new (allocator) IntrinsicSlowPathX86_64(invoke);
     codegen->AddSlowPath(slow_path);