CN101620546B - Realizing method of register window switching in binary translation - Google Patents
Realizing method of register window switching in binary translation Download PDFInfo
- Publication number
- CN101620546B CN101620546B CN 200910056334 CN200910056334A CN101620546B CN 101620546 B CN101620546 B CN 101620546B CN 200910056334 CN200910056334 CN 200910056334 CN 200910056334 A CN200910056334 A CN 200910056334A CN 101620546 B CN101620546 B CN 101620546B
- Authority
- CN
- China
- Prior art keywords
- register
- window
- source platform
- register window
- virtual
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Landscapes
- Executing Machine-Instructions (AREA)
Abstract
The invention relates to a realizing method of register window switching technology in dynamic binary translation, belonging to the technical field of computers. The realizing method comprises the following steps: corresponding to all register windows on a source platform by using a section of continuous memory space; storing the address of the current register window in the corresponding simulated memory space on the source platform using a virtual register and simulating a control/state register and a global register by using the other fixed memory space, wherein the control/state register and the global register are irrelevant to the register windows; determining the positions of the registers of the source platform in the simulated memory in a second-level addressing mode; and dynamically adjusting the value of the virtual register when each register window switching instruction is translated at the front end of a translator. The invention uses the second-level addressing mode to enable the value of the virtual register to be adjusted only when each register window switching instruction is translated into an intermediate instruction, thereby effectively reducing memory access frequency and further improving the execution efficiency of object codes after translation.
Description
Technical field
The present invention relates to a kind of register window changing method of field of computer technology, particularly a kind of register window switches in the implementation method in the binary translation.
Background technology
Binary translation is most widely used a kind of method in the virtual machine technique, to solve legacy code and improve the adaptive a kind of effective means of software platform, it is in the situation of the source code that does not need executable program, the binary program on the conversion source gallery dynamically, and make it to run on other target machine platforms.For dynamic binary translator self, execution performance is a very important measurement index.So-called execution performance refers to, the efficient of moving under former framework take source program is as reference, and its loss in efficiency degree under the running environment that dynamic binary translator provides is lower, and the execution performance of translater is better.The dynamic binary translator that comprises metainstruction generally comprises front-end and back-end.Optimization to interpreter code just concentrates on intermediate language level and target machine code level, and the translation quality of front end has directly affected the quality of the metainstruction that generates and the quality of target machine code.
Some system architecture, such as SPARC, AMD 29000 and Intel i960 have a large number of register, and work register forms several windows, forms the structure of ring-type, utilizes the overlapping register window technology to come the running of faster procedure.The register window technology is exactly the register group that register is divided into a lot of fixed qties, for each function call process is distributed a register group, when calling, automatically CPU being transformed into different register groups uses, no longer need to do the operation of Save and restore, this register group is register window.The register that fixed qty is arranged between adjacent register window is overlapping, is used for carrying out between two register windows data sharing.The parameter transmission utilizes this mechanism to realize.When function call being arranged or returns from function, there is respectively an instruction (such as on the SPARC being SAVE and RESTORE) to finish the switching of register window, but owing to be not that every kind of system architecture has corresponding mechanism, so just cause that efficient seriously is affected when the program that the framework with register window technology compiles is translated on other platform architecture, such as, the binary translation take SPARC as front end just has significantly performance reduction.The solution register window switches in the method that realizes in the binary translation at present has:
Method one: the content that all registers are preserved in every SAVE instruction, and the register read of lap got in the new register window, every RESTORE instruction is read the content of the register of preserving.
Method two: the data flow diagram of routine analyzer, the content that draws accordingly which register need to be used in the back of this instruction, to the content that these registers that need to use are only preserved in every SAVE instruction, every RESTORE instruction is read the content of the register of preserving.
Can see that first method all needs to carry out to copy between register many times operation with internal storage access for every register window switching command, efficient is very low; Second method has improved some in efficient, but need to analyze data flow diagram, increased the complexity of translater front end, and just eliminated the operation of a part of access memory, not from raising the efficiency in essence, because the frequency that the register window switching command is performed is very high, the value of every less important some registers of preservation will certainly increase the read-write operation of many internal memories to internal memory.
Therefore, for above-mentioned technical barrier, in the prior art field, need to propose a kind of novel method that is applicable to based on the register window switching command in the binary translation of metainstruction, be used for improving the execution performance on heterogeneous platform after the program that the platform with register window technology compiles is translated.
Summary of the invention
The object of the invention is to for the deficiencies in the prior art, provide a kind of register window to switch in implementation method in the binary translation.The present invention reduces the expense when moving by reducing the operation to internal memory, and the program execution performance of making has larger lifting.
It is as follows to the present invention includes step:
1. when the front end of translater is set up mapping relations with all register windows on the corresponding source platform of one section continuous analog memory headroom;
2. preserve the address of current register window in the emulated memory space of correspondence on the source platform with a virtual register, the pointer of mark register window position is set up mapping relations in this virtual register and the emulated memory space;
3. with other fixing memory headroom simulation and register window irrelevant control/status register and global register;
4. determine the position of source platform register in the emulated memory space for the metainstruction of map source platform registers and translater virtual register with the mode of second-level addressing;
5. when every register window switching command of translater front end translation dynamic adjustment step 2. in the value of virtual register.
Described binary translation comprises two processes: the one, the reflection that moves under source platform is carried out destructing, and translate by fundamental block one by one, convert the intermediate code piece that is formed by intermediate code to; The 2nd, the intermediate code block translation is become the target code block on target platform, can move.
When the front end at translater of step described in 1. set up mapping relations, by the whole register group of dummy source platform, then realized corresponding one by one with the register window of source platform by the two-stage addressing mode.
The register window of described and source platform corresponding one by one, all source-register groups and a memory headroom are set up mapping relations, the memory headroom of application window number size is set up vertical one by one mapping relations with register window in the memory headroom of dummy source platform registers.
Step 5. in, described dynamic adjustment refers to adjust the window's position by virtual register of described operation, has replaced the operation that a upper register window is switched to internal memory.
The metainstruction of described map source platform registers and translater virtual register is the GET/PUT instruction.
Binary translation of the present invention comprises two processes: the one, the reflection that moves under source platform is carried out destructing, and translate by fundamental block one by one, convert the intermediate code piece that is formed by intermediate code to; The 2nd, the intermediate code block translation is become the target code block on target platform, can move.Dynamic binary translator comprises the intermediate code layer, because if will reach the translation characteristic of multi-source multi-target, many these one decks will reduce the translation path of translating target platform from source platform.Be that dynamic binary translator is divided into front-end and back-end.The quality of front end translation has directly affected the quality of the intermediate code that generates.The present invention is used in the front end just, finishes a register window switching command and succinctly is translated as efficiently metainstruction.
The invention enables the virtual register of value only need to adjust to(for) every register window switching command, i.e. the value of the pointer of the mark current register window position of step described in 2..This has greatly reduced the operation to internal memory, has effectively improved the quality of the intermediate code that generates, thereby has improved the execution performance of object code.Which compare with the method that determines preservation register by analyzing data flow diagram, the present invention is effectively succinct, and replace traditional mode that copies a large amount of memory contents by the value of adjusting a virtual register and fundamentally reduced operation to internal memory, thereby the execution efficient of the object code that translates is higher.
Description of drawings
Fig. 1 is the memory headroom distribution plan of the dummy source platform registers that adopts of the present invention.
Embodiment
For understanding better technical scheme of the present invention, be further described below in conjunction with accompanying drawing and by specific embodiment.Following examples do not consist of limitation of the invention.
SPARC is a risc microcontroller of SUN and the cooperative development of TI company.The processor of SPARC comprises two types register: general-purpose register and control/status register, general-purpose register are divided into global register and register window.At any one time, an instruction can be accessed 8 overall situation (globals) registers and a window that comprises 24 registers in the integer register cell.A register window is comprised of 8 in registers, 8 local registers and 8 out registers, and wherein 8 out registers are with next identical in abutting connection with 8 in register addresss of window.SAVE and RESTORE instruction cause the switching of window.
CrossBit is a dynamic binary translation system, and it can be for carrying out the source program in multiple different architecture, and the method by translation and optimization provides the execution environment under the isomery framework.In order to realize multi-source multi-target, CrossBit has used the intermediate code layer, can reduce like this translation path.Only need to translate into the intermediate code piece to front end binary image program (front end), and then the intermediate code block translation is become corresponding target platform code (rear end), just can form a kind of binary translator.If target platform changes, only need to change a rear end.At present, CrossBit supports SimpleScalar, MIPS, X86, SPARC front end, X86 and SPARC rear end.CrossBit adopts VINST as intermediate command set, it is the self-designed cover intermediate command set of CrossBit project team, be a kind of virtual machine instruction set of simplifying of low level, have infinite a plurality of 32 virtual registers, Load-Store style architecture, single inclined to one side location addressing mode.The execution framework of CrossBit is, at first load the source binary image, find out the entry address of source image, in TCache, search afterwards target code block that should entrance, so-called TCache is a region of memory, be used for depositing translated target code block, if search successfully, then switch contexts is carried out this target code block, if search unsuccessfully, then start the spot decoder of CrossBit, to in the source image fundamental block that should the entry address being decoded into metainstruction, and be packaged into intermediate code piece (VBlock), the back-end code device that then calls CrossBit converts the metainstruction piece to target code block (TBlock).Then be submitted in the TCache, and then search TCache, TCache will occur and hit, at this moment carry out again contextual switching, carry out the execution of target code block.After target code block is finished, will turn back to the context of CrossBit, carry out the operation of next samsara.And so forth, until whole program be finished.The front end binary image program (front end) of using the register window handoff technique is being translated in the process of intermediate code piece, will used method of the present invention and carry out translation to the register window switching command.
A task of front end is that the register of source platform is simulated.The mode of dummy source platform registers is among the CrossBit, at first in memory headroom the application a continuous space, then the source platform register mappings in this piece memory headroom.
The present invention mainly comprises following several process, the mode of the register of the interpretative system of the initialization of the buffer status of particular source platform, register window switching command, back-end realization map source platform and the virtual register of CrossBit.Be without loss of generality, present embodiment relies on the execution framework of SPARC register window switching characteristic, CrossBit and the generation technique of target code block.
1. the buffer status of initialization source platform
I. the initialization in the emulated memory space of register window mapping
All source-register groups and a memory headroom are set up mapping relations, rather than only shine upon a register window at every turn.The application size is the memory headroom regs of REG_WINDOWS_SIZE in the SPARCProcessorState class, and this space content is initialized as 0:
volatile?XTInt32?regs[REG_WINODW_SIZE]；
UTIL::XTMemset((void*)®s，0，sizeof(regs))；
The initialization in the emulated memory space of ii. global register and control/status register mapping
Two memory headroom global[8 that initialization is other] and asr[EOASR] respectively in order to set up mapping relations with global register and the control/status register of above-mentioned SPARC, the value initialization of two memory headrooms is 0:
volatile?XTInt32?global[8]；
volatile?XTInt32?asr[EOASR]；
UTIL::XTMemset((void*)&global，0，sizeof(global))；
UTIL::XTMemset((void*)&asr，0，sizeof(asr))；
Iii. the initialization of register window pointer
The pointer of the current register window of mark position and the virtual register CWP of CrossBit set up mapping relations, and this pointer content is carried out initialization:
asr[CWP]＝REG_WINODWS_SIZE-24；
24 is numbers of register in each register window herein.
Iv. the address of source platform register in the emulated memory space
XTMemAddr get (XTRegNum reg) is the method that this class provides, and is used for obtaining the single level address of the corresponding simulation address space of certain register of source platform.The performing step of XTMemAddr get (XTRegNumregNum) method is as follows:
If it is global register that register number regNum, illustrates this register so less than 8, directly return global+regNum; If register number regNum is less than 32 and be not less than 8, illustrate that then this register is a register in the register window, returns its single level address regs+regNum-8 here; If register number regNum is not less than 32, then this register is control/status register, returns its address asr+regNum-32.
Owing to can't know the position of register window in internal memory in the translating phase, namely can't determine the value of flag register the window's position pointer, so the position of certain register only could be determined when operation in the register window.So, can be by the value of the pointer of the current register window of mark position being added the value of single level address, the address of source platform register in memory headroom that obtains simulateding when operation.
2. translation register windows exchange instruction
For the SAVE instruction, the pointer of the register window position of the virtual register CWP of CrossBit and mark source platform is set up mapping relations, then the value with virtual register CWP deducts 16, namely window is slided into adjacent next position, at last the value of window registers is shone upon back in the memory headroom of dummy source platform registers.Code is as follows:
sbb->add(LI(16，TEMPREG2))；
sbb->add(GET(CWP，CWP))；
sbb->add(SUB(CWP，TEMPREG2，CWP))；
sbb->add(PUT(CWP，CWP))；
The interpretative system of RESTORE instruction is substantially identical with the interpretative system of SAVE instruction, and unique difference is in the process of translation RESTORE instruction, need to add 16 to the value of virtual register CWP, rather than deduct 16.Code is as follows:
sbb->add(LI(16，TEMPREG3))；
sbb->add(GET(CWP，CWP))；
sbb->add(ADD(CWP，TEMPREG3，CWP))；
sbb->add(PUT(CWP，CWP))；
In above-mentioned code, metainstruction GET and PUT are the instructions of centre of the virtual register of the register of map source platform and CrossBit.Can see under this interpretative system, copy a large amount of memory contents to no longer needing in the translation of windows exchange instruction, but only need to adjust the value of virtual register CWP.
3. realize the register of map source platform and the virtual register of CrossBit
The important step that the instruction of source platform machine is decoded is the mapping relations of setting up between the virtual register of the register of source platform machine and CrossBit.Comprise the register mappings of the source platform virtual register (GET) to CrossBit, and the register (PUT instruction) that the virtual register of CrossBit is shone upon back the source platform machine.
In the rear end, the register of map source platform and the virtual register of CrossBit are realized by decoding metainstruction GET and metainstruction PUT.The below describes the decoding process of GET in detail, and the PUT instruction class seemingly.
Metainstruction GET (srcReg, dstReg) is the virtual register dstReg that the register srcReg on the source platform is mapped to CrossBit.Its realization flow is as follows:
A) obtain the single level address of source platform register in the emulated memory space.
Since front end to source platform instruction resolve, can directly obtain the register number of the corresponding dummy source platform registers of a certain source platform instruction here.Pass the instruction inst that comes for front end, obtain the register number of source platform instruction, then call above-mentioned XTMemAddr get (XTRegNum regNum) method and obtain the single level address of this register:
XTMemAddr?addr＝src_state->get(inst->getOperand(1).reg)；
B) obtain the CrossBit virtual register that will be mapped to
Described in a), front end to source platform instruction resolve, can directly obtain the virtual register that will be mapped to here:
XTRegNum?reg＝inst->getOperand(2).reg；
C) register in the mapping register window
According to the initialized agreement of front end, the register in the register window is register number greater than 7 and less than 32 register.Step has calculated the single level address of register in the emulated memory space in a), here obtain the pointer src_state-of the current register window of mark position〉get (CWP), and the value of this pointer moved among the register Tsrc (implementation is X86::movwMemToReg (src_state-〉get (CWP), Tsrc)); The value of Tsrc is added single level address addr, namely obtain the source platform register inst-that will shine upon the real address of getOperand (1) .reg in the emulated memory space.Then applying for that a register and this address are shone upon gets final product that (implementation is X86::movMemToReg (Tsrc, addr, WORD, X86RegAlloc (reg, DEF)); ).
D) shine upon other register
Other register comprises global register and control/status register.These two kinds of register addresses in the emulated memory space are fixed, therefore very simple to their processing, directly then register of application gets final product address and this register mappings that obtains in a) that (implementation is X86::movwMemToReg (addr, X86RegAlloc (reg, DEF)); ).
Claims (6)
1. a register window switches in the implementation method in the binary translation, it is characterized in that, comprises the steps:
1. when the front end of translater is set up mapping relations with all register windows on the corresponding source platform of one section continuous analog memory headroom;
2. preserve the address of current register window in the emulated memory space of correspondence on the source platform with a virtual register, the pointer of mark register window position is set up mapping relations in this virtual register and the emulated memory space;
3. with other fixing memory headroom simulation and register window irrelevant control/status register and global register;
4. determine the position of source platform register in the emulated memory space for the metainstruction of map source platform registers and translater virtual register with the mode of second-level addressing;
5. when every register window switching command of translater front end translation dynamic adjustment step 2. in the value of virtual register.
2. register window according to claim 1 switches in the implementation method in the binary translation, it is characterized in that, described binary translation, comprise two processes: the one, the reflection that moves under source platform is carried out destructing, translate by fundamental block one by one, convert the intermediate code piece that is formed by intermediate code to; The 2nd, the intermediate code block translation is become the target code block on target platform, can move.
3. register window according to claim 1 switches in the implementation method in the binary translation, it is characterized in that, when the front end at translater of step described in 1. set up mapping relations, by the whole register group of dummy source platform, then realized corresponding one by one with the register window of source platform by the two-stage addressing mode.
4. register window according to claim 3 switches in the implementation method in the binary translation, it is characterized in that, the register window of described and source platform corresponding one by one, all source-register groups and a memory headroom are set up mapping relations, the memory headroom of application window number size is set up vertical one by one mapping relations with register window in the memory headroom of dummy source platform registers.
5. register window according to claim 1 switches in the implementation method in the binary translation,
It is characterized in that, step 5. in, described dynamic adjustment refers to adjust the window's position by operating a described virtual register, has replaced the operation that a upper register window is switched to internal memory.
6. register window according to claim 1 switches in the implementation method in the binary translation, it is characterized in that, the metainstruction of described map source platform registers and translater virtual register is the GET/PUT instruction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910056334 CN101620546B (en) | 2009-08-13 | 2009-08-13 | Realizing method of register window switching in binary translation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910056334 CN101620546B (en) | 2009-08-13 | 2009-08-13 | Realizing method of register window switching in binary translation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101620546A CN101620546A (en) | 2010-01-06 |
| CN101620546B true CN101620546B (en) | 2013-03-27 |
Family
ID=41513794
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200910056334 Expired - Fee Related CN101620546B (en) | 2009-08-13 | 2009-08-13 | Realizing method of register window switching in binary translation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101620546B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102073533B (en) * | 2011-01-14 | 2013-06-19 | 中国人民解放军国防科学技术大学 | Multicore architecture supporting dynamic binary translation |
| CN103077011B (en) * | 2012-10-09 | 2015-12-09 | 华为技术有限公司 | A kind of method and apparatus of pre-generatmg machine code instruction in JavaScript script |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101145109A (en) * | 2007-11-01 | 2008-03-19 | 上海交通大学 | Binary translation method using intermediate command set |
| CN101387969A (en) * | 2008-10-16 | 2009-03-18 | 上海交通大学 | Dynamic binary translation method for cooperation design of software and hardware |
-
2009
- 2009-08-13 CN CN 200910056334 patent/CN101620546B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101145109A (en) * | 2007-11-01 | 2008-03-19 | 上海交通大学 | Binary translation method using intermediate command set |
| CN101387969A (en) * | 2008-10-16 | 2009-03-18 | 上海交通大学 | Dynamic binary translation method for cooperation design of software and hardware |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101620546A (en) | 2010-01-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US7536682B2 (en) | Method and apparatus for performing interpreter optimizations during program code conversion | |
| US7543284B2 (en) | Partial dead code elimination optimizations for program code conversion | |
| US5742802A (en) | Method and system for efficiently mapping guest instruction in an emulation assist unit | |
| US8615643B2 (en) | Operational efficiency of virtual TLBs | |
| US8145470B2 (en) | Accelerated access device emulator access scheme in a hypervisor environment with child and root partitions | |
| US5815686A (en) | Method and apparatus for address space translation using a TLB | |
| CN101763316B (en) | Method for dynamically distributing isomerism storage resources on instruction parcel based on virtual memory mechanism | |
| US20060206687A1 (en) | Method and system for a second level address translation in a virtual machine environment | |
| CN101299192B (en) | Non-aligning access and storage processing method | |
| US20030149963A1 (en) | Condition code flag emulation for program code conversion | |
| US8589143B2 (en) | Virtualization apparatus and processing method thereof | |
| RU2008103216A (en) | PREVENTING MULTIPLE ACCESSES TO THE QUICK CONVERTER OF THE ADDRESS FOR ONE AND SAME PAGES IN MEMORY | |
| CN101620546B (en) | Realizing method of register window switching in binary translation | |
| JP5261503B2 (en) | RISC processor device and instruction address conversion search method | |
| US7200841B2 (en) | Method and apparatus for performing lazy byteswapping optimizations during program code conversion | |
| CN104871144A (en) | Speculative addressing using a virtual address-to-physical address page crossing buffer | |
| US9465595B2 (en) | Computing apparatus, computing method, and computing program | |
| CN101882113A (en) | Memory virtualization method based on guest operatiion system kernel code replacement | |
| JP2990195B2 (en) | Emulation device and method for legacy command | |
| US5732235A (en) | Method and system for minimizing the number of cycles required to execute semantic routines | |
| US20160011889A1 (en) | Simulation method and storage medium | |
| US9223600B1 (en) | In-processor dynamic address redirection table for substituting instruction strings | |
| TWI660307B (en) | Binary translation device and method | |
| CN107341002B (en) | Memory access method | |
| KR100930831B1 (en) | Apparatus for accelerating bytecode translation and the method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20130327 Termination date: 20200813 |