8265491: Math Signum optimization for x86

Reviewed-by: jiefu, jbhateja, neliasso

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

@@ -1045,6 +1045,35 @@ void C2_MacroAssembler::evminmax_fp(int opcode, BasicType elem_bt,
   }
 }
 
+// Float/Double signum
+void C2_MacroAssembler::signum_fp(int opcode, XMMRegister dst,
+                                  XMMRegister zero, XMMRegister one,
+                                  Register scratch) {
+  assert(opcode == Op_SignumF || opcode == Op_SignumD, "sanity");
+
+  Label DONE_LABEL;
+
+  if (opcode == Op_SignumF) {
+    assert(UseSSE > 0, "required");
+    ucomiss(dst, zero);
+    jcc(Assembler::equal, DONE_LABEL);    // handle special case +0.0/-0.0, if argument is +0.0/-0.0, return argument
+    jcc(Assembler::parity, DONE_LABEL);   // handle special case NaN, if argument NaN, return NaN
+    movflt(dst, one);
+    jcc(Assembler::above, DONE_LABEL);
+    xorps(dst, ExternalAddress(StubRoutines::x86::vector_float_sign_flip()), scratch);
+  } else if (opcode == Op_SignumD) {
+    assert(UseSSE > 1, "required");
+    ucomisd(dst, zero);
+    jcc(Assembler::equal, DONE_LABEL);    // handle special case +0.0/-0.0, if argument is +0.0/-0.0, return argument
+    jcc(Assembler::parity, DONE_LABEL);   // handle special case NaN, if argument NaN, return NaN
+    movdbl(dst, one);
+    jcc(Assembler::above, DONE_LABEL);
+    xorpd(dst, ExternalAddress(StubRoutines::x86::vector_double_sign_flip()), scratch);
+  }
+
+  bind(DONE_LABEL);
+}
+
 void C2_MacroAssembler::vextendbw(bool sign, XMMRegister dst, XMMRegister src) {
   if (sign) {
     pmovsxbw(dst, src);

src/hotspot/cpu/x86/c2_MacroAssembler_x86.hpp

@@ -89,6 +89,10 @@ public:
                    KRegister ktmp, XMMRegister atmp, XMMRegister btmp,
                    int vlen_enc);
 
+  void signum_fp(int opcode, XMMRegister dst,
+                 XMMRegister zero, XMMRegister one,
+                 Register scratch);
+
   void vextendbw(bool sign, XMMRegister dst, XMMRegister src, int vector_len);
   void vextendbw(bool sign, XMMRegister dst, XMMRegister src);
   void vextendbd(bool sign, XMMRegister dst, XMMRegister src, int vector_len);

src/hotspot/cpu/x86/vm_version_x86.cpp

@@ -1702,6 +1702,9 @@ void VM_Version::get_processor_features() {
     }
   }
 #endif // !PRODUCT
+  if (FLAG_IS_DEFAULT(UseSignumIntrinsic)) {
+      FLAG_SET_DEFAULT(UseSignumIntrinsic, true);
+  }
 }
 
 void VM_Version::print_platform_virtualization_info(outputStream* st) {

src/hotspot/cpu/x86/x86.ad

@@ -1599,6 +1599,16 @@ const bool Matcher::match_rule_supported(int opcode) {
       }
       break;
 #endif // !LP64
+    case Op_SignumF:
+      if (UseSSE < 1) {
+        return false;
+      }
+      break;
+    case Op_SignumD:
+      if (UseSSE < 2) {
+        return false;
+      }
+      break;
   }
   return true;  // Match rules are supported by default.
 }
@@ -5779,6 +5789,30 @@ instruct evminmaxFP_reg_eavx(vec dst, vec a, vec b, vec atmp, vec btmp, kReg ktm
   ins_pipe( pipe_slow );
 %}
 
+// --------------------------------- Signum ---------------------------
+
+instruct signumF_reg(regF dst, regF zero, regF one, rRegP scratch, rFlagsReg cr) %{
+  match(Set dst (SignumF dst (Binary zero one)));
+  effect(TEMP scratch, KILL cr);
+  format %{ "signumF $dst, $dst\t! using $scratch as TEMP" %}
+  ins_encode %{
+    int opcode = this->ideal_Opcode();
+    __ signum_fp(opcode, $dst$$XMMRegister, $zero$$XMMRegister, $one$$XMMRegister, $scratch$$Register);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct signumD_reg(regD dst, regD zero, regD one, rRegP scratch, rFlagsReg cr) %{
+  match(Set dst (SignumD dst (Binary zero one)));
+  effect(TEMP scratch, KILL cr);
+  format %{ "signumD $dst, $dst\t! using $scratch as TEMP" %}
+  ins_encode %{
+    int opcode = this->ideal_Opcode();
+    __ signum_fp(opcode, $dst$$XMMRegister, $zero$$XMMRegister, $one$$XMMRegister, $scratch$$Register);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 // --------------------------------- Sqrt --------------------------------------
 
 instruct vsqrtF_reg(vec dst, vec src) %{

src/hotspot/share/opto/library_call.cpp

@@ -1737,8 +1737,8 @@ bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
   case vmIntrinsics::_dpow:      return inline_math_pow();
   case vmIntrinsics::_dcopySign: return inline_double_math(id);
   case vmIntrinsics::_fcopySign: return inline_math(id);
-  case vmIntrinsics::_dsignum: return inline_double_math(id);
+  case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
-  case vmIntrinsics::_fsignum: return inline_math(id);
+  case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
 
    // These intrinsics are not yet correctly implemented
   case vmIntrinsics::_datan2:

test/hotspot/jtreg/compiler/intrinsics/math/TestSignumIntrinsic.java

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2020, 2021, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2020, BELLSOFT. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -25,7 +25,6 @@
 /*
  * @test
  * @summary Test compiler intrinsics for signum
- * @requires os.arch=="aarch64"
  * @library /test/lib
  *
  * @run main/othervm
@@ -100,7 +99,7 @@ public class TestSignumIntrinsic {
             float arg = fcase[0];
             float expected = fcase[1];
             float calculated = Math.signum(arg);
-            Asserts.assertEQ(expected, calculated, "Unexpected float result");
+            Asserts.assertEQ(expected, calculated, "Unexpected float result from " + arg);
             accum += calculated;
         }
         return accum;
@@ -112,7 +111,7 @@ public class TestSignumIntrinsic {
             double arg = dcase[0];
             double expected = dcase[1];
             double calculated = Math.signum(arg);
-            Asserts.assertEQ(expected, calculated, "Unexpected double result");
+            Asserts.assertEQ(expected, calculated, "Unexpected double result from " + arg);
             accum += calculated;
         }
         return accum;

test/micro/org/openjdk/bench/vm/compiler/Signum.java

@@ -0,0 +1,142 @@
+/*
+ * Copyright (c) Intel, 2021 All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+package org.openjdk.bench.vm.compiler;
+
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.BenchmarkMode;
+import org.openjdk.jmh.annotations.Fork;
+import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Mode;
+import org.openjdk.jmh.annotations.OperationsPerInvocation;
+import org.openjdk.jmh.annotations.OutputTimeUnit;
+import org.openjdk.jmh.annotations.Scope;
+import org.openjdk.jmh.annotations.State;
+import org.openjdk.jmh.annotations.Warmup;
+import org.openjdk.jmh.infra.Blackhole;
+
+import java.util.concurrent.TimeUnit;
+
+@BenchmarkMode(Mode.AverageTime)
+@OutputTimeUnit(TimeUnit.NANOSECONDS)
+@State(Scope.Thread)
+@Warmup(iterations = 10, time = 500, timeUnit = TimeUnit.MILLISECONDS)
+@Measurement(iterations = 5, time = 1000, timeUnit = TimeUnit.MILLISECONDS)
+@Fork(3)
+public class Signum {
+
+    private final int ITERATIONS = 15000;
+
+    private double doubleValue = 1D;
+    private float floatValue = 1F;
+
+    private static final float[] float_values = {
+        123.4f,
+        -56.7f,
+        7e30f,
+        -0.3e30f,
+        Float.MAX_VALUE,
+        -Float.MAX_VALUE,
+        Float.MIN_VALUE,
+        -Float.MIN_VALUE,
+        0.0f,
+        -0.0f,
+        Float.POSITIVE_INFINITY,
+        Float.NEGATIVE_INFINITY,
+        Float.NaN,
+        Float.MIN_NORMAL,
+        -Float.MIN_NORMAL,
+        0x0.0002P-126f,
+        -0x0.0002P-126f
+    };
+
+    private static final double[] double_values = {
+        123.4d,
+        -56.7d,
+        7e30d,
+        -0.3e30d,
+        Double.MAX_VALUE,
+        -Double.MAX_VALUE,
+        Double.MIN_VALUE,
+        -Double.MIN_VALUE,
+        0.0d,
+        -0.0d,
+        Double.POSITIVE_INFINITY,
+        Double.NEGATIVE_INFINITY,
+        Double.NaN,
+        Double.MIN_NORMAL,
+        -Double.MIN_NORMAL,
+        0x0.00000001P-1022,
+        -0x0.00000001P-1022,
+    };
+
+    private static double Signum_Kernel(double data)
+    {
+        return Math.signum(data);
+    }
+
+    private static float Signum_Kernel(float data)
+    {
+        return Math.signum(data);
+    }
+
+    @Benchmark
+    @OperationsPerInvocation(ITERATIONS * 17)
+    public void _1_signumFloatTest(Blackhole bh) {
+        for (int i = 0; i < ITERATIONS; i++) {
+            for (float f : float_values) {
+                bh.consume(Signum_Kernel(f));
+            }
+        }
+    }
+
+    @Benchmark
+    @OperationsPerInvocation(ITERATIONS * 17)
+    public void _2_overheadFloat(Blackhole bh) {
+        for (int i = 0; i < ITERATIONS; i++) {
+            for (float f : float_values) {
+                bh.consume(f);
+            }
+        }
+    }
+
+    @Benchmark
+    @OperationsPerInvocation(ITERATIONS * 17)
+    public void _3_signumDoubleTest(Blackhole bh) {
+        for (int i = 0; i < ITERATIONS; i++) {
+            for (double d : double_values) {
+                bh.consume(Signum_Kernel(d));
+            }
+        }
+    }
+
+    @Benchmark
+    @OperationsPerInvocation(ITERATIONS * 17)
+    public void _4_overheadDouble(Blackhole bh) {
+        for (int i = 0; i < ITERATIONS; i++) {
+            for (double d : double_values) {
+                bh.consume(d);
+            }
+        }
+    }
+}