ART: Add CRC32.udate(int,int) intrinsic for ARM64

Use crc32 instructions for java.util.zip.CRC32.update(int,int).

Note that CRC32 is an optional feature on ARMv8, this intrinsic
is only enabled for devices with CRC32 intruction support.

Original author: tim.zhang@linaro.org

Performance improvements in CRC32Bench.UpdateInt:
Pixel 2: 22.8x
Nexus 6P:
  little core: 28.3x
  big core   : 21.6x

Test: m test-art-target-gtest
Test: testrunner.py --target --optimizing --jit --interpreter
Test: 580-crc32

Change-Id: I1a9bc2befd2934b04103a27ce05806e919874070

6 files changed, 40 insertions, 0 deletions

diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index 7684dc79f2..6d04b0e9d9 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -2916,6 +2916,40 @@ void IntrinsicLocationsBuilderARM64::VisitReachabilityFence(HInvoke* invoke) {
 
 void IntrinsicCodeGeneratorARM64::VisitReachabilityFence(HInvoke* invoke ATTRIBUTE_UNUSED) { }
 
+void IntrinsicLocationsBuilderARM64::VisitCRC32Update(HInvoke* invoke) {
+  if (!codegen_->GetInstructionSetFeatures().HasCRC()) {
+    return;
+  }
+
+  LocationSummary* locations = new (allocator_) LocationSummary(invoke,
+                                                                LocationSummary::kNoCall,
+                                                                kIntrinsified);
+
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetInAt(1, Location::RequiresRegister());
+  locations->SetOut(Location::RequiresRegister());
+}
+
+// Lower the invoke of CRC32.update(int crc, int b).
+void IntrinsicCodeGeneratorARM64::VisitCRC32Update(HInvoke* invoke) {
+  DCHECK(codegen_->GetInstructionSetFeatures().HasCRC());
+
+  MacroAssembler* masm = GetVIXLAssembler();
+
+  Register crc = InputRegisterAt(invoke, 0);
+  Register val = InputRegisterAt(invoke, 1);
+  Register out = OutputRegister(invoke);
+
+  // The general algorithm of the CRC32 calculation is:
+  //   crc = ~crc
+  //   result = crc32_for_byte(crc, b)
+  //   crc = ~result
+  // It is directly lowered to three instructions.
+  __ Mvn(out, crc);
+  __ Crc32b(out, out, val);
+  __ Mvn(out, out);
+}
+
 UNIMPLEMENTED_INTRINSIC(ARM64, ReferenceGetReferent)
 
 UNIMPLEMENTED_INTRINSIC(ARM64, StringStringIndexOf);
diff --git a/compiler/optimizing/intrinsics_arm_vixl.cc b/compiler/optimizing/intrinsics_arm_vixl.cc
index 38e4c8968a..0c463a280e 100644
--- a/compiler/optimizing/intrinsics_arm_vixl.cc
+++ b/compiler/optimizing/intrinsics_arm_vixl.cc
@@ -3059,6 +3059,7 @@ UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathRoundDouble)   // Could be done by changing
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, UnsafeCASLong)     // High register pressure.
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, SystemArrayCopyChar)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, ReferenceGetReferent)
+UNIMPLEMENTED_INTRINSIC(ARMVIXL, CRC32Update)
 
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringStringIndexOf);
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringStringIndexOfAfter);
diff --git a/compiler/optimizing/intrinsics_mips.cc b/compiler/optimizing/intrinsics_mips.cc
index 6f7f5e49c1..21fb7d7f1c 100644
--- a/compiler/optimizing/intrinsics_mips.cc
+++ b/compiler/optimizing/intrinsics_mips.cc
@@ -2696,6 +2696,8 @@ UNIMPLEMENTED_INTRINSIC(MIPS, UnsafeCASLong)
 UNIMPLEMENTED_INTRINSIC(MIPS, ReferenceGetReferent)
 UNIMPLEMENTED_INTRINSIC(MIPS, SystemArrayCopy)
 
+UNIMPLEMENTED_INTRINSIC(MIPS, CRC32Update)
+
 UNIMPLEMENTED_INTRINSIC(MIPS, StringStringIndexOf);
 UNIMPLEMENTED_INTRINSIC(MIPS, StringStringIndexOfAfter);
 UNIMPLEMENTED_INTRINSIC(MIPS, StringBufferAppend);
diff --git a/compiler/optimizing/intrinsics_mips64.cc b/compiler/optimizing/intrinsics_mips64.cc
index 2eb252908c..4b86f5d423 100644
--- a/compiler/optimizing/intrinsics_mips64.cc
+++ b/compiler/optimizing/intrinsics_mips64.cc
@@ -2346,6 +2346,7 @@ void IntrinsicCodeGeneratorMIPS64::VisitReachabilityFence(HInvoke* invoke ATTRIB
 
 UNIMPLEMENTED_INTRINSIC(MIPS64, ReferenceGetReferent)
 UNIMPLEMENTED_INTRINSIC(MIPS64, SystemArrayCopy)
+UNIMPLEMENTED_INTRINSIC(MIPS64, CRC32Update)
 
 UNIMPLEMENTED_INTRINSIC(MIPS64, StringStringIndexOf);
 UNIMPLEMENTED_INTRINSIC(MIPS64, StringStringIndexOfAfter);
diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index 3504d7a6f8..6dd4681847 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -2967,6 +2967,7 @@ UNIMPLEMENTED_INTRINSIC(X86, IntegerHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(X86, LongHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(X86, IntegerLowestOneBit)
 UNIMPLEMENTED_INTRINSIC(X86, LongLowestOneBit)
+UNIMPLEMENTED_INTRINSIC(X86, CRC32Update)
 
 UNIMPLEMENTED_INTRINSIC(X86, StringStringIndexOf);
 UNIMPLEMENTED_INTRINSIC(X86, StringStringIndexOfAfter);
diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index 96f6eaaf33..7db26dc9be 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -2732,6 +2732,7 @@ void IntrinsicCodeGeneratorX86_64::VisitReachabilityFence(HInvoke* invoke ATTRIB
 UNIMPLEMENTED_INTRINSIC(X86_64, ReferenceGetReferent)
 UNIMPLEMENTED_INTRINSIC(X86_64, FloatIsInfinite)
 UNIMPLEMENTED_INTRINSIC(X86_64, DoubleIsInfinite)
+UNIMPLEMENTED_INTRINSIC(X86_64, CRC32Update)
 
 UNIMPLEMENTED_INTRINSIC(X86_64, StringStringIndexOf);
 UNIMPLEMENTED_INTRINSIC(X86_64, StringStringIndexOfAfter);