[x64] Introduce FMA3 instructions on scalar data elements.

src/x64/assembler-x64.h

 // Copyright (c) 1994-2006 Sun Microsystems Inc.
 // All Rights Reserved.
 //
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
 //
 // - Redistributions of source code must retain the above copyright notice,
 // this list of conditions and the following disclaimer.
 //
@@ skipping 996 matching lines @@
   void fyl2x();
   void f2xm1();
   void fscale();
   void fninit();
 
   void frndint();
 
   void sahf();
 
   // SSE instructions
+  void addss(XMMRegister dst, XMMRegister src);
+  void addss(XMMRegister dst, const Operand& src);
+  void subss(XMMRegister dst, XMMRegister src);
+  void subss(XMMRegister dst, const Operand& src);
+  void mulss(XMMRegister dst, XMMRegister src);
+  void mulss(XMMRegister dst, const Operand& src);
+  void divss(XMMRegister dst, XMMRegister src);
+  void divss(XMMRegister dst, const Operand& src);
+
+  void ucomiss(XMMRegister dst, XMMRegister src);
+  void ucomiss(XMMRegister dst, const Operand& src);
   void movaps(XMMRegister dst, XMMRegister src);
   void movss(XMMRegister dst, const Operand& src);
   void movss(const Operand& dst, XMMRegister src);
   void shufps(XMMRegister dst, XMMRegister src, byte imm8);
 
   void cvttss2si(Register dst, const Operand& src);
   void cvttss2si(Register dst, XMMRegister src);
   void cvtlsi2ss(XMMRegister dst, Register src);
 
   void andps(XMMRegister dst, XMMRegister src);
@@ skipping 89 matching lines @@
 
   enum RoundingMode {
     kRoundToNearest = 0x0,
     kRoundDown      = 0x1,
     kRoundUp        = 0x2,
     kRoundToZero    = 0x3
   };
 
   void roundsd(XMMRegister dst, XMMRegister src, RoundingMode mode);
 
+  // AVX instruction
+  void vfmadd132sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0x99, dst, src1, src2);
+  }
+  void vfmadd213sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0xa9, dst, src1, src2);
+  }
+  void vfmadd231sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0xb9, dst, src1, src2);
+  }
+  void vfmadd132sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0x99, dst, src1, src2);
+  }
+  void vfmadd213sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0xa9, dst, src1, src2);
+  }
+  void vfmadd231sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0xb9, dst, src1, src2);
+  }
+  void vfmsub132sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0x9b, dst, src1, src2);
+  }
+  void vfmsub213sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0xab, dst, src1, src2);
+  }
+  void vfmsub231sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0xbb, dst, src1, src2);
+  }
+  void vfmsub132sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0x9b, dst, src1, src2);
+  }
+  void vfmsub213sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0xab, dst, src1, src2);
+  }
+  void vfmsub231sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0xbb, dst, src1, src2);
+  }
+  void vfnmadd132sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0x9d, dst, src1, src2);
+  }
+  void vfnmadd213sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0xad, dst, src1, src2);
+  }
+  void vfnmadd231sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0xbd, dst, src1, src2);
+  }
+  void vfnmadd132sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0x9d, dst, src1, src2);
+  }
+  void vfnmadd213sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0xad, dst, src1, src2);
+  }
+  void vfnmadd231sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0xbd, dst, src1, src2);
+  }
+  void vfnmsub132sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0x9f, dst, src1, src2);
+  }
+  void vfnmsub213sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0xaf, dst, src1, src2);
+  }
+  void vfnmsub231sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmasd(0xbf, dst, src1, src2);
+  }
+  void vfnmsub132sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0x9f, dst, src1, src2);
+  }
+  void vfnmsub213sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0xaf, dst, src1, src2);
+  }
+  void vfnmsub231sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmasd(0xbf, dst, src1, src2);
+  }
+  void vfmasd(byte op, XMMRegister dst, XMMRegister src1, XMMRegister src2);
+  void vfmasd(byte op, XMMRegister dst, XMMRegister src1, const Operand& src2);
+
+  void vfmadd132ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0x99, dst, src1, src2);
+  }
+  void vfmadd213ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0xa9, dst, src1, src2);
+  }
+  void vfmadd231ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0xb9, dst, src1, src2);
+  }
+  void vfmadd132ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0x99, dst, src1, src2);
+  }
+  void vfmadd213ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0xa9, dst, src1, src2);
+  }
+  void vfmadd231ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0xb9, dst, src1, src2);
+  }
+  void vfmsub132ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0x9b, dst, src1, src2);
+  }
+  void vfmsub213ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0xab, dst, src1, src2);
+  }
+  void vfmsub231ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0xbb, dst, src1, src2);
+  }
+  void vfmsub132ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0x9b, dst, src1, src2);
+  }
+  void vfmsub213ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0xab, dst, src1, src2);
+  }
+  void vfmsub231ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0xbb, dst, src1, src2);
+  }
+  void vfnmadd132ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0x9d, dst, src1, src2);
+  }
+  void vfnmadd213ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0xad, dst, src1, src2);
+  }
+  void vfnmadd231ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0xbd, dst, src1, src2);
+  }
+  void vfnmadd132ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0x9d, dst, src1, src2);
+  }
+  void vfnmadd213ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0xad, dst, src1, src2);
+  }
+  void vfnmadd231ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0xbd, dst, src1, src2);
+  }
+  void vfnmsub132ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0x9f, dst, src1, src2);
+  }
+  void vfnmsub213ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0xaf, dst, src1, src2);
+  }
+  void vfnmsub231ss(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
+    vfmass(0xbf, dst, src1, src2);
+  }
+  void vfnmsub132ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0x9f, dst, src1, src2);
+  }
+  void vfnmsub213ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0xaf, dst, src1, src2);
+  }
+  void vfnmsub231ss(XMMRegister dst, XMMRegister src1, const Operand& src2) {
+    vfmass(0xbf, dst, src1, src2);
+  }
+  void vfmass(byte op, XMMRegister dst, XMMRegister src1, XMMRegister src2);
+  void vfmass(byte op, XMMRegister dst, XMMRegister src1, const Operand& src2);
+
   // Debugging
   void Print();
 
   // Check the code size generated from label to here.
   int SizeOfCodeGeneratedSince(Label* label) {
     return pc_offset() - label->pos();
   }
 
   // Mark address of the ExitJSFrame code.
   void RecordJSReturn();
@@ skipping 173 matching lines @@
   template<class P1, class P2>
   void emit_rex(P1 p1, P2 p2, int size) {
     if (size == kInt64Size) {
       emit_rex_64(p1, p2);
     } else {
       DCHECK(size == kInt32Size);
       emit_optional_rex_32(p1, p2);
     }
   }
 
+  // Emit vex prefix
+  void emit_vex2_byte0() { emit(0xc5); }
+  void emit_vex2_byte1(XMMRegister reg, XMMRegister v, byte lpp);
+  void emit_vex3_byte0() { emit(0xc4); }
+  void emit_vex3_byte1(XMMRegister reg, XMMRegister rm, byte m);
+  void emit_vex3_byte1(XMMRegister reg, const Operand& rm, byte m);
+  void emit_vex3_byte2(byte w, XMMRegister v, byte lpp);
+
   // Emit the ModR/M byte, and optionally the SIB byte and
   // 1- or 4-byte offset for a memory operand.  Also encodes
   // the second operand of the operation, a register or operation
   // subcode, into the reg field of the ModR/M byte.
   void emit_operand(Register reg, const Operand& adr) {
     emit_operand(reg.low_bits(), adr);
   }
 
   // Emit the ModR/M byte, and optionally the SIB byte and
   // 1- or 4-byte offset for a memory operand.  Also used to encode
@@ skipping 297 matching lines @@
  private:
   Assembler* assembler_;
 #ifdef DEBUG
   int space_before_;
 #endif
 };
 
 } }  // namespace v8::internal
 
 #endif  // V8_X64_ASSEMBLER_X64_H_