Add missing Location::kNoOutputOverlap

This can save some ParallelMove instructions.

For x86(_64) all FPToFP intrinsics can add it.

For RISC-V, MathSqrt can add it but the ones that call
GenDoubleRound can't. Also, we can add it for MathMultiplyHigh.

Test: art/test/testrunner/testrunner.py --host --64 -b --optimizing
Test: LUCI run https://ci.chromium.org/b/8731845964396026257
Change-Id: I28e13caf84cd850566538efbd285c0264ce80a1a

3 files changed, 10 insertions, 6 deletions

diff --git a/compiler/optimizing/intrinsics_riscv64.cc b/compiler/optimizing/intrinsics_riscv64.cc
index 009fedeb6d..7fc5cfb764 100644
--- a/compiler/optimizing/intrinsics_riscv64.cc
+++ b/compiler/optimizing/intrinsics_riscv64.cc
@@ -19,6 +19,7 @@
 #include "code_generator_riscv64.h"
 #include "intrinsic_objects.h"
 #include "intrinsics_utils.h"
+#include "optimizing/locations.h"
 #include "well_known_classes.h"
 
 namespace art HIDDEN {
@@ -160,11 +161,13 @@ static void CreateFpFpFpToFpNoOverlapLocations(ArenaAllocator* allocator, HInvok
   locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
 }
 
-static void CreateFPToFPLocations(ArenaAllocator* allocator, HInvoke* invoke) {
+static void CreateFPToFPLocations(ArenaAllocator* allocator,
+                                  HInvoke* invoke,
+                                  Location::OutputOverlap overlaps = Location::kOutputOverlap) {
   LocationSummary* locations =
       new (allocator) LocationSummary(invoke, LocationSummary::kNoCall, kIntrinsified);
   locations->SetInAt(0, Location::RequiresFpuRegister());
-  locations->SetOut(Location::RequiresFpuRegister());
+  locations->SetOut(Location::RequiresFpuRegister(), overlaps);
 }
 
 void IntrinsicLocationsBuilderRISCV64::VisitDoubleDoubleToRawLongBits(HInvoke* invoke) {
@@ -5329,7 +5332,7 @@ void IntrinsicCodeGeneratorRISCV64::VisitMathTanh(HInvoke* invoke) {
 }
 
 void IntrinsicLocationsBuilderRISCV64::VisitMathSqrt(HInvoke* invoke) {
-  CreateFPToFPLocations(allocator_, invoke);
+  CreateFPToFPLocations(allocator_, invoke, Location::kNoOutputOverlap);
 }
 
 void IntrinsicCodeGeneratorRISCV64::VisitMathSqrt(HInvoke* invoke) {
@@ -5452,7 +5455,7 @@ void IntrinsicLocationsBuilderRISCV64::VisitMathMultiplyHigh(HInvoke* invoke) {
       new (allocator_) LocationSummary(invoke, LocationSummary::kNoCall, kIntrinsified);
   locations->SetInAt(0, Location::RequiresRegister());
   locations->SetInAt(1, Location::RequiresRegister());
-  locations->SetOut(Location::RequiresRegister());
+  locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
 }
 
 void IntrinsicCodeGeneratorRISCV64::VisitMathMultiplyHigh(HInvoke* invoke) {
diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index f71689230d..952fb855be 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -310,7 +310,7 @@ static void CreateFPToFPLocations(ArenaAllocator* allocator, HInvoke* invoke) {
   LocationSummary* locations =
       new (allocator) LocationSummary(invoke, LocationSummary::kNoCall, kIntrinsified);
   locations->SetInAt(0, Location::RequiresFpuRegister());
-  locations->SetOut(Location::RequiresFpuRegister());
+  locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
 }
 
 void IntrinsicLocationsBuilderX86::VisitMathSqrt(HInvoke* invoke) {
diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index 5e7c3a9611..85c33d312f 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -37,6 +37,7 @@
 #include "mirror/string.h"
 #include "optimizing/code_generator.h"
 #include "optimizing/data_type.h"
+#include "optimizing/locations.h"
 #include "scoped_thread_state_change-inl.h"
 #include "thread-current-inl.h"
 #include "utils/x86_64/assembler_x86_64.h"
@@ -329,7 +330,7 @@ static void CreateFPToFPLocations(ArenaAllocator* allocator, HInvoke* invoke) {
   LocationSummary* locations =
       new (allocator) LocationSummary(invoke, LocationSummary::kNoCall, kIntrinsified);
   locations->SetInAt(0, Location::RequiresFpuRegister());
-  locations->SetOut(Location::RequiresFpuRegister());
+  locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
 }
 
 void IntrinsicLocationsBuilderX86_64::VisitMathSqrt(HInvoke* invoke) {