Add several SIMD opcodes to IA32

src/ia32/assembler-ia32.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 22 matching lines @@
 // Copyright 2011 the V8 project authors. All rights reserved.
 
 // A light-weight IA32 Assembler.
 
 #ifndef V8_IA32_ASSEMBLER_IA32_H_
 #define V8_IA32_ASSEMBLER_IA32_H_
 
 #include <deque>
 
 #include "src/assembler.h"
+#include "src/ia32/sse-instr.h"
 #include "src/isolate.h"
 #include "src/utils.h"
 
 namespace v8 {
 namespace internal {
 
 #define GENERAL_REGISTERS(V) \
   V(eax)                     \
   V(ecx)                     \
   V(edx)                     \
@@ skipping 920 matching lines @@
   void addps(XMMRegister dst, const Operand& src);
   void addps(XMMRegister dst, XMMRegister src) { addps(dst, Operand(src)); }
   void subps(XMMRegister dst, const Operand& src);
   void subps(XMMRegister dst, XMMRegister src) { subps(dst, Operand(src)); }
   void mulps(XMMRegister dst, const Operand& src);
   void mulps(XMMRegister dst, XMMRegister src) { mulps(dst, Operand(src)); }
   void divps(XMMRegister dst, const Operand& src);
   void divps(XMMRegister dst, XMMRegister src) { divps(dst, Operand(src)); }
 
   // SSE2 instructions
+  void sse2_instr(XMMRegister dst, const Operand& src, byte prefix, byte escape,
+                  byte opcode);

[bbudge, 2017/02/15 16:00:12]

This should probably be either private, or even better, a static function
defined in assembler-ia32.cc. You don't want clients of assembler to call it.

[bbudge, 2017/02/15 16:07:31]

Looking at the implementation, this has to be a class method, so make it
protected or private.

[Jing, 2017/02/16 05:11:17]

Done.

+#define DECLARE_SSE2_INSTRUCTION(instruction, prefix, escape, opcode) \
+  void instruction(XMMRegister dst, XMMRegister src) {                \
+    instruction(dst, Operand(src));                                   \
+  }                                                                   \
+  void instruction(XMMRegister dst, const Operand& src) {             \
+    sse2_instr(dst, src, 0x##prefix, 0x##escape, 0x##opcode);         \
+  }
+
+  SSE2_INSTRUCTION_LIST(DECLARE_SSE2_INSTRUCTION)

[bbudge, 2017/02/15 16:00:12]

I think it would be more readable if SSE2_INSTRUCTION_LIST was defined in this
file, rather than in a separate header.

[Jing, 2017/02/16 05:11:17]

I ported the header file from x64 and planned to expand the instruction list to
at least as many instructions as x64.
https://chromium.googlesource.com/v8/v8.git/+/master/src/x64/sse-instr.h
I don't know which is preferred? To add the list in current file or add a new
header file?

[bbudge, 2017/02/17 00:44:40]

I think it's more consistent with V8 style to inline the list here.

[Jing, 2017/02/17 03:51:35]

Done.

[Jing, 2017/02/20 15:28:13]

Sorry, I recover the separate header file again, because it has to be included
in disasm-ia32.cc

[bbudge, 2017/02/22 21:51:09]

OK, I see the precedent in x64.

+#undef DECLARE_SSE2_INSTRUCTION
+
   void cvttss2si(Register dst, const Operand& src);
   void cvttss2si(Register dst, XMMRegister src) {
     cvttss2si(dst, Operand(src));
   }
   void cvttsd2si(Register dst, const Operand& src);
   void cvttsd2si(Register dst, XMMRegister src) {
     cvttsd2si(dst, Operand(src));
   }
   void cvtsd2si(Register dst, XMMRegister src);
 
@@ skipping 84 matching lines @@
   void pshufd(XMMRegister dst, XMMRegister src, uint8_t shuffle);
   void pextrd(Register dst, XMMRegister src, int8_t offset) {
     pextrd(Operand(dst), src, offset);
   }
   void pextrd(const Operand& dst, XMMRegister src, int8_t offset);
   void pinsrd(XMMRegister dst, Register src, int8_t offset) {
     pinsrd(dst, Operand(src), offset);
   }
   void pinsrd(XMMRegister dst, const Operand& src, int8_t offset);
 
   // AVX instructions

[bbudge, 2017/02/15 16:07:31]

Is there a plan to use AVX if it's available?

   void vfmadd132sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
     vfmadd132sd(dst, src1, Operand(src2));
   }
   void vfmadd213sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
     vfmadd213sd(dst, src1, Operand(src2));
   }
   void vfmadd231sd(XMMRegister dst, XMMRegister src1, XMMRegister src2) {
     vfmadd231sd(dst, src1, Operand(src2));
   }
   void vfmadd132sd(XMMRegister dst, XMMRegister src1, const Operand& src2) {
@@ skipping 494 matching lines @@
   Assembler* assembler_;
 #ifdef DEBUG
   int space_before_;
 #endif
 };
 
 }  // namespace internal
 }  // namespace v8
 
 #endif  // V8_IA32_ASSEMBLER_IA32_H_