commit 6545ee31288b72e117450ea77b07f3593bff6180
author Shalini Salomi Bodapati <shalini.salomi.bodapati@intel.com> Tue Nov 02 20:01:06 2021 +0530
committer Vladimir Marko <vmarko@google.com> Thu Nov 11 11:32:25 2021 +0000
tree 60085d1b253758ba32d8b585e493f634d95e422a
parent a0232ad060c65d0ac3e022c9f721b41c02fd0b71

Support Math.fma Intrinsic for x86 and x86_64

Intrinsic Implementation is 2287x faster than base
Time for base: 188.25s
Time for patch: 0.0822s

Test: ./run-test --host --64 --optimizing 082-inline-execute
Test: ./run-test --host --64 --optimizing --instruction-set-features sse4.1 082-inline-execute
Test: ./run-test --host --64 --optimizing --instruction-set-features sse4.1,avx2 082-inline-execute
Signed-off-by: Shalini Salomi Bodapati <shalini.salomi.bodapati@intel.com>
Change-Id: I68b96a35f41f3bb23d7e26d87fee1dd2a5ebf6a6

compiler/utils/x86/assembler_x86.cc

diff --git a/compiler/utils/x86/assembler_x86.cc b/compiler/utils/x86/assembler_x86.cc
index b6708de..4b64e92 100644
--- a/compiler/utils/x86/assembler_x86.cc
+++ b/compiler/utils/x86/assembler_x86.cc
@@ -783,6 +783,43 @@
   EmitXmmRegisterOperand(dst, src2);
 }
 
+void X86Assembler::vfmadd213ss(XmmRegister acc, XmmRegister left, XmmRegister right) {
+  DCHECK(CpuHasAVXorAVX2FeatureFlag());
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  uint8_t ByteZero = 0x00, ByteOne = 0x00, ByteTwo = 0x00;
+  ByteZero = EmitVexPrefixByteZero(/*is_twobyte_form=*/ false);
+  X86ManagedRegister vvvv_reg = X86ManagedRegister::FromXmmRegister(left);
+  ByteOne = EmitVexPrefixByteOne(/*R=*/ false,
+                                 /*X=*/ false,
+                                 /*B=*/ false,
+                                 SET_VEX_M_0F_38);
+  ByteTwo = EmitVexPrefixByteTwo(/*W=*/ false, vvvv_reg, SET_VEX_L_128, SET_VEX_PP_66);
+
+  EmitUint8(ByteZero);
+  EmitUint8(ByteOne);
+  EmitUint8(ByteTwo);
+  EmitUint8(0xA9);
+  EmitXmmRegisterOperand(acc, right);
+}
+
+void X86Assembler::vfmadd213sd(XmmRegister acc, XmmRegister left, XmmRegister right) {
+  DCHECK(CpuHasAVXorAVX2FeatureFlag());
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  uint8_t ByteZero = 0x00, ByteOne = 0x00, ByteTwo = 0x00;
+  ByteZero = EmitVexPrefixByteZero(/*is_twobyte_form=*/ false);
+  X86ManagedRegister vvvv_reg = X86ManagedRegister::FromXmmRegister(left);
+  ByteOne = EmitVexPrefixByteOne(/*R=*/ false,
+                                 /*X=*/ false,
+                                 /*B=*/ false,
+                                 SET_VEX_M_0F_38);
+  ByteTwo = EmitVexPrefixByteTwo(/*W=*/ true, vvvv_reg, SET_VEX_L_128, SET_VEX_PP_66);
+
+  EmitUint8(ByteZero);
+  EmitUint8(ByteOne);
+  EmitUint8(ByteTwo);
+  EmitUint8(0xA9);
+  EmitXmmRegisterOperand(acc, right);
+}
 
 void X86Assembler::movapd(XmmRegister dst, XmmRegister src) {
   if (CpuHasAVXorAVX2FeatureFlag()) {

compiler/utils/x86/assembler_x86.h

diff --git a/compiler/utils/x86/assembler_x86.h b/compiler/utils/x86/assembler_x86.h
index f6e7fbc..dc2427d 100644
--- a/compiler/utils/x86/assembler_x86.h
+++ b/compiler/utils/x86/assembler_x86.h
@@ -492,6 +492,9 @@
   void vsubpd(XmmRegister dst, XmmRegister add_left, XmmRegister add_right);
   void vaddpd(XmmRegister dst, XmmRegister add_left, XmmRegister add_right);
 
+  void vfmadd213ss(XmmRegister acc, XmmRegister left, XmmRegister right);
+  void vfmadd213sd(XmmRegister acc, XmmRegister left, XmmRegister right);
+
   void movapd(XmmRegister dst, XmmRegister src);     // move
   void movapd(XmmRegister dst, const Address& src);  // load aligned
   void movupd(XmmRegister dst, const Address& src);  // load unaligned

compiler/utils/x86/assembler_x86_test.cc

diff --git a/compiler/utils/x86/assembler_x86_test.cc b/compiler/utils/x86/assembler_x86_test.cc
index 92ce788..5db2b65 100644
--- a/compiler/utils/x86/assembler_x86_test.cc
+++ b/compiler/utils/x86/assembler_x86_test.cc
@@ -1073,6 +1073,18 @@
       RepeatFFF(&x86::X86Assembler::vpmaddwd, "vpmaddwd %{reg3}, %{reg2}, %{reg1}"), "vpmaddwd");
 }
 
+TEST_F(AssemblerX86AVXTest, VFMadd213SS) {
+  DriverStr(
+      RepeatFFF(&x86::X86Assembler::vfmadd213ss,
+                "vfmadd213ss %{reg3}, %{reg2}, %{reg1}"), "vfmadd213ss");
+}
+
+TEST_F(AssemblerX86AVXTest, VFMadd213SD) {
+  DriverStr(
+      RepeatFFF(&x86::X86Assembler::vfmadd213sd,
+                "vfmadd213sd %{reg3}, %{reg2}, %{reg1}"), "vfmadd213sd");
+}
+
 TEST_F(AssemblerX86Test, PHAddW) {
   DriverStr(RepeatFF(&x86::X86Assembler::phaddw, "phaddw %{reg2}, %{reg1}"), "phaddw");
 }