ART: Add entrypoint and intrinsic for Math.pow().
MathBenchmarks.java#timePow results on taimen's little cores
fixed at frequency 1401600 with forced JIT compilation:
- before:
- X32: 356.33 (@FastNative), 315.39 (@CriticalNative)
- X64: 357.31 (@FastNative), 315.37 (@CriticalNative)
- after (LICM defeats the benchmark):
- X32: 2.88
- X64: 2.87
- after but with kAllSideEffects to prevent LICM:
- X32: 275.42
- X64: 275.67
Test: Rely on TreeHugger.
Bug: 70727450
Change-Id: Iaa31f70acabbd57c163cfeafe02eed67c1348861
diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index ca1b451..2f8e33f 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -2011,6 +2011,14 @@
GenFPToFPCall(invoke, codegen_, kQuickAtan2);
}
+void IntrinsicLocationsBuilderARM64::VisitMathPow(HInvoke* invoke) {
+ CreateFPFPToFPCallLocations(allocator_, invoke);
+}
+
+void IntrinsicCodeGeneratorARM64::VisitMathPow(HInvoke* invoke) {
+ GenFPToFPCall(invoke, codegen_, kQuickPow);
+}
+
void IntrinsicLocationsBuilderARM64::VisitMathHypot(HInvoke* invoke) {
CreateFPFPToFPCallLocations(allocator_, invoke);
}
diff --git a/compiler/optimizing/intrinsics_arm_vixl.cc b/compiler/optimizing/intrinsics_arm_vixl.cc
index 99b8b5d..830d040 100644
--- a/compiler/optimizing/intrinsics_arm_vixl.cc
+++ b/compiler/optimizing/intrinsics_arm_vixl.cc
@@ -2811,6 +2811,14 @@
GenFPFPToFPCall(invoke, GetAssembler(), codegen_, kQuickAtan2);
}
+void IntrinsicLocationsBuilderARMVIXL::VisitMathPow(HInvoke* invoke) {
+ CreateFPFPToFPCallLocations(allocator_, invoke);
+}
+
+void IntrinsicCodeGeneratorARMVIXL::VisitMathPow(HInvoke* invoke) {
+ GenFPFPToFPCall(invoke, GetAssembler(), codegen_, kQuickPow);
+}
+
void IntrinsicLocationsBuilderARMVIXL::VisitMathHypot(HInvoke* invoke) {
CreateFPFPToFPCallLocations(allocator_, invoke);
}
diff --git a/compiler/optimizing/intrinsics_mips.cc b/compiler/optimizing/intrinsics_mips.cc
index 113c9de..cafa522 100644
--- a/compiler/optimizing/intrinsics_mips.cc
+++ b/compiler/optimizing/intrinsics_mips.cc
@@ -2835,6 +2835,15 @@
GenFPFPToFPCall(invoke, codegen_, kQuickAtan2);
}
+// static double java.lang.Math.pow(double y, double x)
+void IntrinsicLocationsBuilderMIPS::VisitMathPow(HInvoke* invoke) {
+ CreateFPFPToFPCallLocations(allocator_, invoke);
+}
+
+void IntrinsicCodeGeneratorMIPS::VisitMathPow(HInvoke* invoke) {
+ GenFPFPToFPCall(invoke, codegen_, kQuickPow);
+}
+
// static double java.lang.Math.cbrt(double a)
void IntrinsicLocationsBuilderMIPS::VisitMathCbrt(HInvoke* invoke) {
CreateFPToFPCallLocations(allocator_, invoke);
diff --git a/compiler/optimizing/intrinsics_mips64.cc b/compiler/optimizing/intrinsics_mips64.cc
index 521bad2..89f1818 100644
--- a/compiler/optimizing/intrinsics_mips64.cc
+++ b/compiler/optimizing/intrinsics_mips64.cc
@@ -2416,6 +2416,15 @@
GenFPFPToFPCall(invoke, codegen_, kQuickAtan2);
}
+// static double java.lang.Math.pow(double y, double x)
+void IntrinsicLocationsBuilderMIPS64::VisitMathPow(HInvoke* invoke) {
+ CreateFPFPToFPCallLocations(allocator_, invoke);
+}
+
+void IntrinsicCodeGeneratorMIPS64::VisitMathPow(HInvoke* invoke) {
+ GenFPFPToFPCall(invoke, codegen_, kQuickPow);
+}
+
// static double java.lang.Math.cbrt(double a)
void IntrinsicLocationsBuilderMIPS64::VisitMathCbrt(HInvoke* invoke) {
CreateFPToFPCallLocations(allocator_, invoke);
diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index baa410b..46b7f3f 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -1105,6 +1105,14 @@
GenFPToFPCall(invoke, codegen_, kQuickAtan2);
}
+void IntrinsicLocationsBuilderX86::VisitMathPow(HInvoke* invoke) {
+ CreateFPFPToFPCallLocations(allocator_, invoke);
+}
+
+void IntrinsicCodeGeneratorX86::VisitMathPow(HInvoke* invoke) {
+ GenFPToFPCall(invoke, codegen_, kQuickPow);
+}
+
void IntrinsicLocationsBuilderX86::VisitMathHypot(HInvoke* invoke) {
CreateFPFPToFPCallLocations(allocator_, invoke);
}
diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index 6dd8b8e..6483b7c 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -897,6 +897,14 @@
GenFPToFPCall(invoke, codegen_, kQuickAtan2);
}
+void IntrinsicLocationsBuilderX86_64::VisitMathPow(HInvoke* invoke) {
+ CreateFPFPToFPCallLocations(allocator_, invoke);
+}
+
+void IntrinsicCodeGeneratorX86_64::VisitMathPow(HInvoke* invoke) {
+ GenFPToFPCall(invoke, codegen_, kQuickPow);
+}
+
void IntrinsicLocationsBuilderX86_64::VisitMathHypot(HInvoke* invoke) {
CreateFPFPToFPCallLocations(allocator_, invoke);
}
diff --git a/compiler/utils/assembler_thumb_test_expected.cc.inc b/compiler/utils/assembler_thumb_test_expected.cc.inc
index 0a09435..674dc9a 100644
--- a/compiler/utils/assembler_thumb_test_expected.cc.inc
+++ b/compiler/utils/assembler_thumb_test_expected.cc.inc
@@ -153,7 +153,7 @@
" 21c: f8d9 8034 ldr.w r8, [r9, #52] ; 0x34\n",
" 220: 4770 bx lr\n",
" 222: 4660 mov r0, ip\n",
- " 224: f8d9 c2c0 ldr.w ip, [r9, #704] ; 0x2c0\n",
+ " 224: f8d9 c2c4 ldr.w ip, [r9, #708] ; 0x2c4\n",
" 228: 47e0 blx ip\n",
nullptr
};