Merge webkit.org at r54731 : Initial merge by git

Change-Id: Ia79977b6cf3b0b00c06ef39419989b28e57e4f4a

5 files changed, 174 insertions, 24 deletions

diff --git a/JavaScriptCore/jit/JIT.h b/JavaScriptCore/jit/JIT.h
index 8e0c9ac..bfbb1ee 100644
--- a/JavaScriptCore/jit/JIT.h
+++ b/JavaScriptCore/jit/JIT.h
@@ -317,10 +317,10 @@ namespace JSC {
             jit.privateCompilePutByIdTransition(stubInfo, oldStructure, newStructure, cachedOffset, chain, returnAddress);
         }
 
-        static void compileCTIMachineTrampolines(JSGlobalData* globalData, RefPtr<ExecutablePool>* executablePool, CodePtr* ctiStringLengthTrampoline, CodePtr* ctiVirtualCallLink, CodePtr* ctiVirtualCall, CodePtr* ctiNativeCallThunk)
+        static void compileCTIMachineTrampolines(JSGlobalData* globalData, RefPtr<ExecutablePool>* executablePool, TrampolineStructure *trampolines)
         {
             JIT jit(globalData);
-            jit.privateCompileCTIMachineTrampolines(executablePool, globalData, ctiStringLengthTrampoline, ctiVirtualCallLink, ctiVirtualCall, ctiNativeCallThunk);
+            jit.privateCompileCTIMachineTrampolines(executablePool, globalData, trampolines);
         }
 
         static void patchGetByIdSelf(CodeBlock* codeblock, StructureStubInfo*, Structure*, size_t cachedOffset, ReturnAddressPtr returnAddress);
@@ -361,7 +361,7 @@ namespace JSC {
         void privateCompileGetByIdChain(StructureStubInfo*, Structure*, StructureChain*, size_t count, size_t cachedOffset, ReturnAddressPtr returnAddress, CallFrame* callFrame);
         void privateCompilePutByIdTransition(StructureStubInfo*, Structure*, Structure*, size_t cachedOffset, StructureChain*, ReturnAddressPtr returnAddress);
 
-        void privateCompileCTIMachineTrampolines(RefPtr<ExecutablePool>* executablePool, JSGlobalData* data, CodePtr* ctiStringLengthTrampoline, CodePtr* ctiVirtualCallLink, CodePtr* ctiVirtualCall, CodePtr* ctiNativeCallThunk);
+        void privateCompileCTIMachineTrampolines(RefPtr<ExecutablePool>* executablePool, JSGlobalData* data, TrampolineStructure *trampolines);
         void privateCompilePatchGetArrayLength(ReturnAddressPtr returnAddress);
 
         void addSlowCase(Jump);
@@ -800,6 +800,9 @@ namespace JSC {
         void emit_op_to_jsnumber(Instruction*);
         void emit_op_to_primitive(Instruction*);
         void emit_op_unexpected_load(Instruction*);
+#if ENABLE(JIT_OPTIMIZE_MOD)
+        void softModulo();
+#endif
 
         void emitSlow_op_add(Instruction*, Vector<SlowCaseEntry>::iterator&);
         void emitSlow_op_bitand(Instruction*, Vector<SlowCaseEntry>::iterator&);
diff --git a/JavaScriptCore/jit/JITArithmetic.cpp b/JavaScriptCore/jit/JITArithmetic.cpp
index feee8d2..2f2ffe3 100644
--- a/JavaScriptCore/jit/JITArithmetic.cpp
+++ b/JavaScriptCore/jit/JITArithmetic.cpp
@@ -31,6 +31,7 @@
 #include "CodeBlock.h"
 #include "JITInlineMethods.h"
 #include "JITStubCall.h"
+#include "JITStubs.h"
 #include "JSArray.h"
 #include "JSFunction.h"
 #include "Interpreter.h"
@@ -1186,14 +1187,40 @@ void JIT::emit_op_mod(Instruction* currentInstruction)
     unsigned op1 = currentInstruction[2].u.operand;
     unsigned op2 = currentInstruction[3].u.operand;
 
+#if ENABLE(JIT_OPTIMIZE_MOD)    
+    emitLoad2(op1, regT1, regT0, op2, regT3, regT2);
+    addSlowCase(branch32(NotEqual, regT1, Imm32(JSValue::Int32Tag)));
+    addSlowCase(branch32(NotEqual, regT3, Imm32(JSValue::Int32Tag)));
+    
+    addSlowCase(branch32(Equal, regT2, Imm32(0)));
+    
+    emitNakedCall(m_globalData->jitStubs.ctiSoftModulo());
+
+    emitStoreInt32(dst, regT0, (op1 == dst || op2 == dst));
+#else
     JITStubCall stubCall(this, cti_op_mod);
     stubCall.addArgument(op1);
     stubCall.addArgument(op2);
     stubCall.call(dst);
+#endif
 }
 
-void JIT::emitSlow_op_mod(Instruction*, Vector<SlowCaseEntry>::iterator&)
+void JIT::emitSlow_op_mod(Instruction* currentInstruction, Vector<SlowCaseEntry>::iterator& iter)
 {
+#if ENABLE(JIT_OPTIMIZE_MOD)
+    unsigned result = currentInstruction[1].u.operand;
+    unsigned op1 = currentInstruction[2].u.operand;
+    unsigned op2 = currentInstruction[3].u.operand;
+    linkSlowCase(iter);
+    linkSlowCase(iter);
+    linkSlowCase(iter);
+    JITStubCall stubCall(this, cti_op_mod);
+    stubCall.addArgument(op1);
+    stubCall.addArgument(op2);
+    stubCall.call(result);
+#else
+    ASSERT_NOT_REACHED();
+#endif
 }
 
 #endif // CPU(X86) || CPU(X86_64)
@@ -2138,15 +2165,40 @@ void JIT::emit_op_mod(Instruction* currentInstruction)
     unsigned op1 = currentInstruction[2].u.operand;
     unsigned op2 = currentInstruction[3].u.operand;
 
+#if ENABLE(JIT_OPTIMIZE_MOD)    
+    emitGetVirtualRegisters(op1, regT0, op2, regT2);
+    emitJumpSlowCaseIfNotImmediateInteger(regT0);
+    emitJumpSlowCaseIfNotImmediateInteger(regT2);
+    
+    addSlowCase(branch32(Equal, regT2, Imm32(1)));
+
+    emitNakedCall(m_globalData->jitStubs.ctiSoftModulo());
+
+    emitPutVirtualRegister(result, regT0);
+#else
     JITStubCall stubCall(this, cti_op_mod);
     stubCall.addArgument(op1, regT2);
     stubCall.addArgument(op2, regT2);
     stubCall.call(result);
+#endif
 }
 
-void JIT::emitSlow_op_mod(Instruction*, Vector<SlowCaseEntry>::iterator&)
+void JIT::emitSlow_op_mod(Instruction* currentInstruction, Vector<SlowCaseEntry>::iterator& iter)
 {
+#if ENABLE(JIT_OPTIMIZE_MOD)
+    unsigned result = currentInstruction[1].u.operand;
+    unsigned op1 = currentInstruction[2].u.operand;
+    unsigned op2 = currentInstruction[3].u.operand;
+    linkSlowCase(iter);
+    linkSlowCase(iter);
+    linkSlowCase(iter);
+    JITStubCall stubCall(this, cti_op_mod);
+    stubCall.addArgument(op1, regT2);
+    stubCall.addArgument(op2, regT2);
+    stubCall.call(result);
+#else
     ASSERT_NOT_REACHED();
+#endif
 }
 
 #endif // CPU(X86) || CPU(X86_64)
diff --git a/JavaScriptCore/jit/JITOpcodes.cpp b/JavaScriptCore/jit/JITOpcodes.cpp
index d9a32d9..c3f20f1 100644
--- a/JavaScriptCore/jit/JITOpcodes.cpp
+++ b/JavaScriptCore/jit/JITOpcodes.cpp
@@ -40,8 +40,12 @@ namespace JSC {
 
 #if USE(JSVALUE32_64)
 
-void JIT::privateCompileCTIMachineTrampolines(RefPtr<ExecutablePool>* executablePool, JSGlobalData* globalData, CodePtr* ctiStringLengthTrampoline, CodePtr* ctiVirtualCallLink, CodePtr* ctiVirtualCall, CodePtr* ctiNativeCallThunk)
+void JIT::privateCompileCTIMachineTrampolines(RefPtr<ExecutablePool>* executablePool, JSGlobalData* globalData, TrampolineStructure *trampolines)
 {
+#if ENABLE(JIT_OPTIMIZE_MOD)
+    Label softModBegin = align();
+    softModulo();
+#endif
 #if ENABLE(JIT_OPTIMIZE_PROPERTY_ACCESS)
     // (1) This function provides fast property access for string length
     Label stringLengthBegin = align();
@@ -365,18 +369,21 @@ void JIT::privateCompileCTIMachineTrampolines(RefPtr<ExecutablePool>* executable
     CodeRef finalCode = patchBuffer.finalizeCode();
     *executablePool = finalCode.m_executablePool;
 
-    *ctiVirtualCall = trampolineAt(finalCode, virtualCallBegin);
-    *ctiNativeCallThunk = trampolineAt(finalCode, nativeCallThunk);
+    trampolines->ctiVirtualCall = trampolineAt(finalCode, virtualCallBegin);
+    trampolines->ctiNativeCallThunk = trampolineAt(finalCode, nativeCallThunk);
 #if ENABLE(JIT_OPTIMIZE_PROPERTY_ACCESS)
-    *ctiStringLengthTrampoline = trampolineAt(finalCode, stringLengthBegin);
+    trampolines->ctiStringLengthTrampoline = trampolineAt(finalCode, stringLengthBegin);
 #else
     UNUSED_PARAM(ctiStringLengthTrampoline);
 #endif
 #if ENABLE(JIT_OPTIMIZE_CALL)
-    *ctiVirtualCallLink = trampolineAt(finalCode, virtualCallLinkBegin);
+    trampolines->ctiVirtualCallLink = trampolineAt(finalCode, virtualCallLinkBegin);
 #else
     UNUSED_PARAM(ctiVirtualCallLink);
 #endif
+#if ENABLE(JIT_OPTIMIZE_MOD)
+    trampolines->ctiSoftModulo = trampolineAt(finalCode, softModBegin);
+#endif
 }
 
 void JIT::emit_op_mov(Instruction* currentInstruction)
@@ -1495,8 +1502,12 @@ void JIT::emit_op_profile_did_call(Instruction* currentInstruction)
 #define RECORD_JUMP_TARGET(targetOffset) \
    do { m_labels[m_bytecodeIndex + (targetOffset)].used(); } while (false)
 
-void JIT::privateCompileCTIMachineTrampolines(RefPtr<ExecutablePool>* executablePool, JSGlobalData* globalData, CodePtr* ctiStringLengthTrampoline, CodePtr* ctiVirtualCallLink, CodePtr* ctiVirtualCall, CodePtr* ctiNativeCallThunk)
+void JIT::privateCompileCTIMachineTrampolines(RefPtr<ExecutablePool>* executablePool, JSGlobalData* globalData, TrampolineStructure *trampolines)
 {
+#if ENABLE(JIT_OPTIMIZE_MOD)
+    Label softModBegin = align();
+    softModulo();
+#endif
 #if ENABLE(JIT_OPTIMIZE_PROPERTY_ACCESS)
     // (2) The second function provides fast property access for string length
     Label stringLengthBegin = align();
@@ -1827,11 +1838,14 @@ void JIT::privateCompileCTIMachineTrampolines(RefPtr<ExecutablePool>* executable
     CodeRef finalCode = patchBuffer.finalizeCode();
     *executablePool = finalCode.m_executablePool;
 
-    *ctiVirtualCallLink = trampolineAt(finalCode, virtualCallLinkBegin);
-    *ctiVirtualCall = trampolineAt(finalCode, virtualCallBegin);
-    *ctiNativeCallThunk = trampolineAt(finalCode, nativeCallThunk);
+    trampolines->ctiVirtualCallLink = trampolineAt(finalCode, virtualCallLinkBegin);
+    trampolines->ctiVirtualCall = trampolineAt(finalCode, virtualCallBegin);
+    trampolines->ctiNativeCallThunk = trampolineAt(finalCode, nativeCallThunk);
+#if ENABLE(JIT_OPTIMIZE_MOD)
+    trampolines->ctiSoftModulo = trampolineAt(finalCode, softModBegin);
+#endif
 #if ENABLE(JIT_OPTIMIZE_PROPERTY_ACCESS)
-    *ctiStringLengthTrampoline = trampolineAt(finalCode, stringLengthBegin);
+    trampolines->ctiStringLengthTrampoline = trampolineAt(finalCode, stringLengthBegin);
 #else
     UNUSED_PARAM(ctiStringLengthTrampoline);
 #endif
@@ -2978,6 +2992,81 @@ void JIT::emitSlow_op_to_jsnumber(Instruction* currentInstruction, Vector<SlowCa
 
 #endif // USE(JSVALUE32_64)
 
+// For both JSValue32_64 and JSValue32
+#if ENABLE(JIT_OPTIMIZE_MOD)
+#if CPU(ARM_TRADITIONAL)
+void JIT::softModulo()
+{
+    push(regS0);
+    push(regS1);
+    push(regT1);
+    push(regT3);
+#if USE(JSVALUE32_64)
+    m_assembler.mov_r(regT3, regT2);
+    m_assembler.mov_r(regT2, regT0);
+#else
+    m_assembler.mov_r(regT3, m_assembler.asr(regT2, 1));
+    m_assembler.mov_r(regT2, m_assembler.asr(regT0, 1));
+#endif
+    m_assembler.mov_r(regT1, ARMAssembler::getOp2(0));
+    
+    m_assembler.teq_r(regT3, ARMAssembler::getOp2(0));
+    m_assembler.rsb_r(regT3, regT3, ARMAssembler::getOp2(0), ARMAssembler::MI);
+    m_assembler.eor_r(regT1, regT1, ARMAssembler::getOp2(1), ARMAssembler::MI);
+    
+    m_assembler.teq_r(regT2, ARMAssembler::getOp2(0));
+    m_assembler.rsb_r(regT2, regT2, ARMAssembler::getOp2(0), ARMAssembler::MI);
+    m_assembler.eor_r(regT1, regT1, ARMAssembler::getOp2(2), ARMAssembler::MI);
+    
+    Jump exitBranch = branch32(LessThan, regT2, regT3);
+
+    m_assembler.sub_r(regS1, regT3, ARMAssembler::getOp2(1));
+    m_assembler.tst_r(regS1, regT3);
+    m_assembler.and_r(regT2, regT2, regS1, ARMAssembler::EQ);
+    m_assembler.and_r(regT0, regS1, regT3);
+    Jump exitBranch2 = branchTest32(Zero, regT0);
+    
+    m_assembler.clz_r(regS1, regT2);
+    m_assembler.clz_r(regS0, regT3);
+    m_assembler.sub_r(regS0, regS0, regS1);
+
+    m_assembler.rsbs_r(regS0, regS0, ARMAssembler::getOp2(31));
+
+    m_assembler.mov_r(regS0, m_assembler.lsl(regS0, 1), ARMAssembler::NE);
+
+    m_assembler.add_r(ARMRegisters::pc, ARMRegisters::pc, m_assembler.lsl(regS0, 2), ARMAssembler::NE);
+    m_assembler.mov_r(regT0, regT0);
+    
+    for (int i = 31; i > 0; --i) {
+        m_assembler.cmp_r(regT2, m_assembler.lsl(regT3, i));
+        m_assembler.sub_r(regT2, regT2, m_assembler.lsl(regT3, i), ARMAssembler::CS);
+    }
+
+    m_assembler.cmp_r(regT2, regT3);
+    m_assembler.sub_r(regT2, regT2, regT3, ARMAssembler::CS);
+    
+    exitBranch.link(this);
+    exitBranch2.link(this);
+    
+    m_assembler.teq_r(regT1, ARMAssembler::getOp2(0));
+    m_assembler.rsb_r(regT2, regT2, ARMAssembler::getOp2(0), ARMAssembler::GT);
+    
+#if USE(JSVALUE32_64)
+    m_assembler.mov_r(regT0, regT2);
+#else
+    m_assembler.mov_r(regT0, m_assembler.lsl(regT2, 1));
+    m_assembler.eor_r(regT0, regT0, ARMAssembler::getOp2(1));
+#endif
+    pop(regT3);
+    pop(regT1);
+    pop(regS1);
+    pop(regS0);
+    ret();
+}
+#else
+#error "JIT_OPTIMIZE_MOD not yet supported on this platform."
+#endif // CPU(ARM_TRADITIONAL)
+#endif
 } // namespace JSC
 
 #endif // ENABLE(JIT)
diff --git a/JavaScriptCore/jit/JITStubs.cpp b/JavaScriptCore/jit/JITStubs.cpp
index bf5168b..85471de 100644
--- a/JavaScriptCore/jit/JITStubs.cpp
+++ b/JavaScriptCore/jit/JITStubs.cpp
@@ -757,7 +757,7 @@ extern "C" {
 
 JITThunks::JITThunks(JSGlobalData* globalData)
 {
-    JIT::compileCTIMachineTrampolines(globalData, &m_executablePool, &m_ctiStringLengthTrampoline, &m_ctiVirtualCallLink, &m_ctiVirtualCall, &m_ctiNativeCallThunk);
+    JIT::compileCTIMachineTrampolines(globalData, &m_executablePool, &m_trampolineStructure);
 
 #if CPU(ARM_THUMB2)
     // Unfortunate the arm compiler does not like the use of offsetof on JITStackFrame (since it contains non POD types),
diff --git a/JavaScriptCore/jit/JITStubs.h b/JavaScriptCore/jit/JITStubs.h
index 99c2dd2..17fd0d9 100644
--- a/JavaScriptCore/jit/JITStubs.h
+++ b/JavaScriptCore/jit/JITStubs.h
@@ -74,6 +74,14 @@ namespace JSC {
         JSString* jsString() { return static_cast<JSString*>(asPointer); }
         ReturnAddressPtr returnAddress() { return ReturnAddressPtr(asPointer); }
     };
+    
+    struct TrampolineStructure {
+        MacroAssemblerCodePtr ctiStringLengthTrampoline;
+        MacroAssemblerCodePtr ctiVirtualCallLink;
+        MacroAssemblerCodePtr ctiVirtualCall;
+        MacroAssemblerCodePtr ctiNativeCallThunk;
+        MacroAssemblerCodePtr ctiSoftModulo;
+    };
 
 #if CPU(X86_64)
     struct JITStackFrame {
@@ -239,18 +247,16 @@ namespace JSC {
         static void tryCacheGetByID(CallFrame*, CodeBlock*, ReturnAddressPtr returnAddress, JSValue baseValue, const Identifier& propertyName, const PropertySlot&, StructureStubInfo* stubInfo);
         static void tryCachePutByID(CallFrame*, CodeBlock*, ReturnAddressPtr returnAddress, JSValue baseValue, const PutPropertySlot&, StructureStubInfo* stubInfo);
 
-        MacroAssemblerCodePtr ctiStringLengthTrampoline() { return m_ctiStringLengthTrampoline; }
-        MacroAssemblerCodePtr ctiVirtualCallLink() { return m_ctiVirtualCallLink; }
-        MacroAssemblerCodePtr ctiVirtualCall() { return m_ctiVirtualCall; }
-        MacroAssemblerCodePtr ctiNativeCallThunk() { return m_ctiNativeCallThunk; }
+        MacroAssemblerCodePtr ctiStringLengthTrampoline() { return m_trampolineStructure.ctiStringLengthTrampoline; }
+        MacroAssemblerCodePtr ctiVirtualCallLink() { return m_trampolineStructure.ctiVirtualCallLink; }
+        MacroAssemblerCodePtr ctiVirtualCall() { return m_trampolineStructure.ctiVirtualCall; }
+        MacroAssemblerCodePtr ctiNativeCallThunk() { return m_trampolineStructure.ctiNativeCallThunk; }
+        MacroAssemblerCodePtr ctiSoftModulo() { return m_trampolineStructure.ctiSoftModulo; }
 
     private:
         RefPtr<ExecutablePool> m_executablePool;
 
-        MacroAssemblerCodePtr m_ctiStringLengthTrampoline;
-        MacroAssemblerCodePtr m_ctiVirtualCallLink;
-        MacroAssemblerCodePtr m_ctiVirtualCall;
-        MacroAssemblerCodePtr m_ctiNativeCallThunk;
+        TrampolineStructure m_trampolineStructure;
     };
 
 extern "C" {