Changeset 40963 in webkit for trunk/JavaScriptCore/assembler/MacroAssemblerX86Common.h

Timestamp:

Feb 12, 2009, 8:22:02 PM (16 years ago)

Author:

[email protected]

Message:

2009-02-12 Gavin Barraclough <​[email protected]>

Reviewed by Sam Weinig.

Remove (/reduce) use of hard-wired register names from the JIT.
Currently there is no abstraction of registers used in the JIT,
which has a number of negative consequences. Hard-wiring x86
register names makes the JIT less portable to other platforms,
and prevents us from performing dynamic register allocation to
attempt to maintain more temporary values in machine registers.
(The latter will be more important on x86-64, where we have more
registers to make use of).

Also, remove MacroAssembler::mod32. This was not providing a
useful abstraction, and was not in keeping with the rest of the
MacroAssembler interface, in having specific register requirements.

• assembler/MacroAssemblerX86Common.h:
• jit/JIT.cpp: (JSC::JIT::compileOpStrictEq): (JSC::JIT::emitSlowScriptCheck): (JSC::JIT::privateCompileMainPass): (JSC::JIT::privateCompileSlowCases): (JSC::JIT::privateCompile): (JSC::JIT::privateCompileCTIMachineTrampolines):
• jit/JIT.h:
• jit/JITArithmetic.cpp: (JSC::JIT::compileFastArith_op_lshift): (JSC::JIT::compileFastArithSlow_op_lshift): (JSC::JIT::compileFastArith_op_rshift): (JSC::JIT::compileFastArithSlow_op_rshift): (JSC::JIT::compileFastArith_op_bitand): (JSC::JIT::compileFastArithSlow_op_bitand): (JSC::JIT::compileFastArith_op_mod): (JSC::JIT::compileFastArithSlow_op_mod): (JSC::JIT::compileFastArith_op_post_inc): (JSC::JIT::compileFastArithSlow_op_post_inc): (JSC::JIT::compileFastArith_op_post_dec): (JSC::JIT::compileFastArithSlow_op_post_dec): (JSC::JIT::compileFastArith_op_pre_inc): (JSC::JIT::compileFastArithSlow_op_pre_inc): (JSC::JIT::compileFastArith_op_pre_dec): (JSC::JIT::compileFastArithSlow_op_pre_dec): (JSC::JIT::compileFastArith_op_add): (JSC::JIT::compileFastArith_op_mul): (JSC::JIT::compileFastArith_op_sub): (JSC::JIT::compileBinaryArithOp):
• jit/JITCall.cpp: (JSC::JIT::compileOpCallInitializeCallFrame): (JSC::JIT::compileOpCallSetupArgs): (JSC::JIT::compileOpCallEvalSetupArgs): (JSC::JIT::compileOpConstructSetupArgs): (JSC::JIT::compileOpCall): (JSC::JIT::compileOpCallSlowCase):
• jit/JITInlineMethods.h: (JSC::JIT::emitGetVirtualRegister): (JSC::JIT::emitPutVirtualRegister): (JSC::JIT::emitNakedCall): (JSC::JIT::restoreArgumentReference): (JSC::JIT::restoreArgumentReferenceForTrampoline):
• jit/JITPropertyAccess.cpp: (JSC::JIT::compileGetByIdHotPath): (JSC::JIT::compilePutByIdHotPath): (JSC::JIT::compileGetByIdSlowCase): (JSC::JIT::compilePutByIdSlowCase): (JSC::JIT::privateCompilePutByIdTransition): (JSC::JIT::privateCompilePatchGetArrayLength): (JSC::JIT::privateCompileGetByIdSelf): (JSC::JIT::privateCompileGetByIdProto): (JSC::JIT::privateCompileGetByIdSelfList): (JSC::JIT::privateCompileGetByIdProtoList): (JSC::JIT::privateCompileGetByIdChainList): (JSC::JIT::privateCompileGetByIdChain): (JSC::JIT::privateCompilePutByIdReplace):

File:

• trunk/JavaScriptCore/assembler/MacroAssemblerX86Common.h (modified) (1 diff)

trunk/JavaScriptCore/assembler/MacroAssemblerX86Common.h

@@ -120 +120 @@
     }
     
-    // Take the value from dividend, divide it by divisor, and put the remainder in remainder.
-    // For now, this operation has specific register requirements, and the three register must
-    // be unique.  It is unfortunate to expose this in the MacroAssembler interface, however
-    // given the complexity to fix, the fact that it is not uncommmon  for processors to have
-    // specific register requirements on this operation (e.g. Mips result in 'hi'), or to not
-    // support a hardware divide at all, it may not be 
-    void mod32(RegisterID divisor, RegisterID dividend, RegisterID remainder)
-    {
-#ifdef NDEBUG
-#pragma unused(dividend,remainder)
-#else
-        ASSERT((dividend == X86::eax) && (remainder == X86::edx));
-        ASSERT((dividend != divisor) && (remainder != divisor));
-#endif
-
-        m_assembler.cdq();
-        m_assembler.idivl_r(divisor);
-    }
-
     void mul32(RegisterID src, RegisterID dest)
     {