Maintaining Synchronism Between a Processor Pipeline and Subsystem Pipelines During Debugging of a Data Processing System

(C) Copyright 1997 Texas Instruments Incorporated. A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent disclosure, as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.

This invention relates to microprocessors, and particularly relates to architectures of very long instruction word processors.

As the technology for manufacturing integrated circuits advances, more and more logic functions may be included in a single integrated circuit device. Modern integrated circuit (IC) devices include large numbers of gates on a single semiconductor chip, with these gates interconnected so as to perform multiple and complex functions, such as, for example, those in a general-purpose microprocessor. The manufacture of such circuits incorporating such Very Large Scale Integration (VLSI) requires that the fabrication of the circuit be error free, as some manufacturing defects may prevent it from performing all of the functions that it is designed to perform. This requires verification of the design of the circuit and also various types of electrical testing after manufacture.

However, as the complexity of the circuit increases, so does the cost and difficulty of verifying and electrically testing each of the devices in the circuit. From an electrical test standpoint, in order to totally verify that each gate in a VLSI circuit functions properly, one must ideally be able to exercise each of the gates not only individually (in the digital sense, determining that it is neither stuck-open nor stuck-closed), but also in conjunction with the other gates in the circuit in all possible combinations of operations. This is normally accomplished by automated testing equipment (ATE) that employs test vectors to perform the desired tests. A test vector describes the desired test input (or signals), associated clock pulse (or pulses), and expected test output (or signals) for every package pin during a period of time, often in an attempt to "test" a particular gate (or macro). For complex circuitry, this may involve a large number of test vectors and accordingly a long test time. Macro and cell are used herein to mean the same thing and may be used interchangeably.

Circuit designers have used stuck-fault modeling techniques in improving the efficiency of the testing of such VLSI circuits. Stuck-fault modeling is directed not to stuck-open or stuck-closed defects in individual gates, but to the effect of such defective gates (and defective interconnections) resulting in stuck-high and stuck-low nodes of the logic circuit. Minimum patterns of test vectors are then derived for the exercising of the logic circuit. Applying such test vectors to the circuit detects stuck-high and stuck-low nodes if defects are present. Such techniques have been successful in improving the test efficiency of current generation VLSI circuits.

In addition, specific circuit configurations in the VLSI circuit may have some of its gates inaccessible for all but a special combination of signals, thereby hiding a fault unless a very specific pattern of signals is presented. However, the cost of performing such testing on 100% of the manufactured circuits is staggering, considering the high cost of the test equipment required to exercise each circuit in conjunction with the long time required to present each possible combination to each gate. This has in the past forced integrated circuit manufacturers to test less than all of the active devices in a chip, with the attendant quality levels of the product being less than optimal. Thus, one of the major problems in integrated circuit design is the ability to adequately test the final IC design, and this problem increases with increasing complexity of the integrated circuit.

One way to address this problem is through design for test (DFT). The key concepts in DFT are controllability and observability. Controllability is the ability to set and reset the state of every node in the circuit, while observability is the ability to observe either directly or indirectly the state of any node in the circuit. The purpose of DFT is to increase the ability to control and observe internal and external nodes from external inputs/outputs. That is, DFT techniques may be employed for logic verification and DC parametric tests.

Designing testability into any circuit will affect the circuitry to some degree. Additional logic will probably have to be added. This additional logic will increase the amount of silicon required to implement the design. The savings from enhanced testability do not usually show up until the development time and testing costs of the circuit and its end system are analyzed.

In conjunction with the stuck-fault modeling and associated test generation, other circuitry may be included in the VLSI circuit specifically designed to improving its testability. One type of test circuitry is a scan path in the logic circuit. A scan path consists of a chain of synchronously clocked master/slave latches (or registers), each of which is connected to a particular node in the logic circuit. These latches can be loaded with a serial data stream ("scan in") presetting the logic circuit nodes to a predetermined state. The logic circuit then can be exercised in normal fashion, with the result of the operation (at each of the nodes having a scan latch) stored in its respective latch. By serially unloading the contents of the latches ("scan out"), the result of the particular test operation at the associated nodes is read out and may be analyzed for improper node operation. Repetition of this operation with a number of different data patterns effectively tests all necessary combinations of the logic circuit, but with a reduced test time and cost compared to separately testing each active component or cell and all their possible interactions. Scan paths permit circuit initialization by directly writing to the latches (or registers) and directly observing the contents of the latches (or registers). Using scan paths helps to reduce the quantity of test vectors compared to traditional "functional mode" approaches. Techniques for scanning such data are discussed by E. J. McCluskey in A Survey of Design for Testability Scan Techniques, VLSI Design (Vol. 5, No. 12, pp. 38-61, December 1984).

Also as VLSI technology is advancing, users of integrated circuits are desiring specially designed and constructed integrated circuits, for performing functions customized for the user's application. Such integrated circuits have been called Application-Specific Integrated Circuits (ASICs). For an ASIC device to be cost-competitive with general purpose microcomputers which may have special functions implemented in programmable firmware, and cost-competitive with a board design made up of smaller scale integrated circuits, the design time of the ASIC circuit must be short and the ASIC circuit must be manufacturable and testable at low cost. Accordingly, it is useful for such circuits to be modular in design, with each of the modules performing a certain function, so that a new ASIC circuit may be constructed by combining previously-designed circuit modules. Such an approach can also be used for non-ASIC microcomputers and microprocessors. Regardless of the end product, the use of a modular approach allows the designer to use logic which has previously been verified, and proven manufacturable. However, if logic modules containing existing scan paths are placed into a new circuit application, new test patterns will generally be required for the new device, thereby lengthening the design/manufacture cycle time.

A modular approach to utilizing scan paths and other testability circuits has been used to provide thorough coverage of all possible faults in an efficient manner. However, this approach utilizes system buses to set up and operate the scan test, so that even though each module is tested independently, the test pattern designed for a given module depends upon the operation of other modules in the logic circuit for purposes of bus control and module selection. This results in the testability of a particular module depending upon the fault-free operation of other modules. In addition, the automatic test program generator (ATPG) program which sets the conditions for test of a given module depends upon the position of the module relative to other modules, and upon the operating features of such other modules. While reduced test times and costs are thus achieved by such modularity, the use of system buses to load and unload the scan paths in the individual modules may not only affect the operation of the particular module, but is likely to also preclude "porting" of the test program for a given module from one logic circuit to another.

Recently, MegaModules have been used in the design of ASICs. (MegaModule is a trademark of Texas Instruments Incorporated.) Types of MegaModules include SRAMs, FIFOs, register files, RAMs, ROMs, universal asynchronous receiver-transmitters (UARTs), programmable logic arrays and other such logic circuits. MegaModules are usually defined as integrated circuit modules of at least 500 gates in complexity and having a complex ASIC macro function. These MegaModules may be predesigned and stored in an ASIC design library. The MegaModules can then be selected by the designer and placed within a certain area on the desired IC chip. This allows ASIC designers to integrate MegaModules into their logic as easily as simple macros.

Another solution to this testing problem of an ASIC is the use of a so-called Parallel Module Test (PMT), which is often referred to as a "direct connect" scheme. (Parallel Module Test is a trademark of Texas Instruments Incorporated.) PMT is a direct connect scheme, because it connects external pins to a MegaModule bypassing all other logic, buffers, etc. It is primarily intended as a logic verification testability scheme and has recently been enhanced to address limited VIH/VIL and ICCQ testability schemes. However, even PMT may have problems since the logic states of the ASIC's circuitry may be disturbed as part of the test process during test selection and enabling.

Another solution is the test access port and boundary-scan architecture defined by the IEEE 1149.1 standard, a so-called JTAG test port. IEEE 1149.1 is primarily intended as a system test solution. The IEEE 1149.1 standard requires a minimum of four package pins to be dedicated to the test function. The IEEE 1149.1 standard requires boundary scan cells for each I/O buffer, which adds data delay to all normal operation function pins as well as silicon overhead. Although it has "hooks" for controlling some internal testability schemes, it is not optimized for chip-level testing. IEEE 1149.1 does not explicitly support testing of internal DC parametrics.

Software breakpoints (SWBP) provide another mechanism to allow the debug of microprocessor code and to evaluate performance. A SWBP is typically accomplished through opcode replacement, provided the program resides in a writable memory module which allows the opcode at the stop point to be replaced in memory with the software breakpoint opcode. In most machines, when a SWBP opcode reaches the first execute stage of an instruction execution pipeline, it causes the pipeline to stop advancing or trap to an interrupt service routine, and set a debug status bit indicating the pipeline has stopped or trapped. In processors classified as protected pipelines, instructions fetched into the pipeline after the SWBP are not executed. Instructions that are already in the pipeline are allowed to complete. To restart execution the pipeline can be cleared and then restarted by simply prefetching the opcode at the SWBP memory address after the opcode is replaced in memory with the original opcode.

Microprocessor designers have increasingly endeavored to exploit parallelism to improve performance. One parallel architecture that has found application in some modern microprocessors is the very long instruction word, or VLIW, architecture. VLIW architecture microprocessors are called that because they handle VLIW format instructions.

A VLIW format instruction is a long fixed-width instruction that encodes multiple concurrent operations. VLIW systems use multiple independent functional units. Instead of issuing multiple independent instructions to the units, a VLIW system combines the multiple operations into one very long instruction. In a VLIW system, computer instructions for multiple integer operations, floating point operations, and memory references may be combined in a single, wide, VLIW instruction.

Testing and debugging such a complex pipeline is difficult, even when the techniques described in the preceding paragraphs are used. These and other disadvantages of the prior art are overcome by the present invention, however, and improved methods and apparatus for chip-level testing, as well as system-level debugging, are provided.

According to the present invention, synchronization between an unprotected instruction execution pipeline internal to a microprocessor and a memory subsystem data pipeline external to the microprocessor can advantageously be maintained by utilizing a "ready" output signal or signals from the microprocessor to warn of a pending halt condition and a second "ready" output signal or signals to indicate the microprocessor has halted. In an unprotected pipeline load and store operations are completed independently of a load or store instruction which initiated the operation. An instruction which follows the initiating load or store instruction can take advantage of memory access latency to retrieve data from the target location of the initiating load or store instruction prior to completion of the load or store operation.

A method for maintaining synchronism between a processor instruction execution pipeline and a subsystem data pipeline in such a data processing system during debugging includes the following steps:

executing system code in the processor instruction execution pipeline in a normal operational manner to initiate a plurality of operations in the instruction execution pipeline and in the data pipeline;

sending a first signal from the processor to the subsystem to indicate a pending halt;

conditioning the subsystem for halting in response to receipt of the first signal;

halting the normal operation of the processor pipeline such that at least one of the plurality of operations is still pending;

sending a second signal to the subsystem to indicate the processor pipeline is halted;

halting the subsystem in response to receipt of the second signal such that any of the operations in the subsystem pipeline which correspond to the at least one of the plurality of operations still pending in the instruction execution pipeline is maintained; and

continuing execution of the system code in the processor instruction execution pipeline in a manner that no extraneous operations occur within the data processing system.

Other features and advantages of the present invention will become apparent by reference to the following detailed description when considered in conjunction with the accompanying drawings, in which:

FIG. 1 is a block diagram of a digital signal processor (DSP), showing components thereof pertinent to an embodiment of the present invention;

FIG. 2 is a block diagram of the functional units, data paths and register files of FIG. 1;

FIG. 3 shows the addressing mode register (AMR) of the DSP of FIG. 1;

FIG. 4 shows the control status register (CSR) which contains control and status bits of the DSP of FIG. 1;

FIG. 5 depicts a general-purpose input register (IN) which supports 32 general-purpose input signals of the DSP of FIG. 1;

FIG. 6 depicts a general-purpose output register (OUT) which supports 32 general-purpose output signals of the DSP of FIG. 1;

FIG. 7 illustrates the register storage scheme for 40-bit data of the DSP of FIG. 1;

FIGS. 8A-8J show an opcode map for the DSP of FIG. 1;

FIG. 9 shows the basic format of a fetch packet of the DSP of FIG. 1;

FIG. 10A depicts a fetch packet of FIG. 9 with fully serial p-bits;

FIG. 10B depicts a fetch packet of FIG. 9 with fully parallel p-bits;

FIG. 10C depicts a fetch packet of FIG. 9 with partially serial p-bits;

FIG. 11 shows the phases of the pipeline of the DSP of FIG. 1;

FIG. 12 shows the branch instruction phases;

FIG. 13 shows the operation of the pipeline of the DSP of FIG. 1 based on clock cycles and fetch packets;

FIG. 14 depicts fetch packet n, which contains three execute packets, shown followed by six fetch packets (n+1 through n+6), each with one execution packet (containing 8 parallel instructions);

FIG. 15 is a block diagram of an MTAP to Test Port Interface for the processor of FIG. 1;

FIG. 16 is a timing diagram of a MegaModule Reset Sequence for the processor of FIG. 1;

FIG. 17A shows the interrupt flag register (IFR) which contains the status of INT4-INT15 and NMI;

FIG. 17B show the interrupt enable register (IER) of the DSP of FIG. 1;

FIG. 17C shows the interrupt set register (ISR), which allows to setting or clearing interrupts manually in the IFR;

FIG. 17D shows the interrupt clear register (ICR), which allows to setting or clearing interrupts manually in the IFR;

FIG. 18 is a timing diagram of detection of Analysis interrupts for the processor of FIG. 1;

FIGS. 19A and 19B illustrate two analysis interrupt related instructions, SWI and B ARP;

FIG. 20 is a block diagram describing MTAP to CPU Interface Signals for the MTAP of FIG. 15;

FIG. 21 is a block diagram of an MTAP to MegaModule Domain Interface for the processor of FIG. 1;

FIG. 22 is a state diagram of the test port states for the processor of FIG. 1;

FIG. 23A is a timing diagram of a clock switch from Functional Run to Scan on UCLK for the processor of FIG. 1;

FIG. 23B is a timing diagram of a clock switch from Functional Run on TCLK for the processor of FIG. 1;

FIG. 23C is a timing diagram of a clock switch from Functional Run on UCLK to Functional Run on TCLK for the processor of FIG. 1;

FIG. 24 is a table of a scan chain for a Data Scan based on the MSEND bits for the processor of FIG. 1;

FIG. 25 is a timing diagram showing various cases of halting for the processor of FIG. 1;

FIG. 26 is a circuit diagram of circuitry to form signal ERDY;

FIG. 27A is a timing diagram of a CPU test port requested halt during interrupt processing for the processor of FIG. 1;

FIG. 27B is a timing diagram illustrating a Test Port Requested Test Halt;

FIG. 28 is a timing diagram of a pipeline halt showing a pipeline management process for emulation for the processor of FIG. 1;

FIG. 29 is a timing diagram showing a pipeline restoration process after emulation for the processor of FIG. 1;

FIG. 30A illustrates an analysis control register for the processor of FIG. 1;

FIG. 30B illustrates an analysis data register for the processor of FIG. 1;

FIG. 30C illustrates an analysis data interrupt return pointer register for the processor of FIG. 1;

FIG. 30D illustrates a data streaming register for the processor of FIG. 1;

FIG. 31 is a timing diagram of the instruction execution pipeline for the processor of FIG. 1 showing various pipeline phases;

FIG. 32 is a block diagram illustrating pin connections to a megamodule in the processor of FIG. 1;

FIG. 33 is a block diagram illustrating JTAG instruction and data register paths for the processor of FIG. 1;

FIG. 34A illustrates JTAG instruction register contents when Strap status is selected in the registers of FIG. 33;

FIG. 34B illustrates JTAG instruction register contents when Stop Emulation status is selected in the registers of FIG. 33;

FIG. 34C illustrates JTAG instruction register contents when Real Time Emulation status is selected in the registers of FIG. 33;

FIG. 34D illustrates JTAG instruction register contents when Emulation Error status is selected in the registers of FIG. 33;

FIG. 35 is a block diagram of a JTAG to MPSD Interface for the processor of FIG. 1;

FIG. 36 illustrates the emulation control register of FIG. 33;

FIG. 37 is a block diagram of a code state machine (CSM) for the MTAP of the processor of FIG. 1;

FIG. 38 is a schematic of a clock source switch for the CSM of FIG. 37;

FIG. 39 is a schematic of circuitry to generate an EVTA interrupt for the processor of FIG. 1;

FIG. 40 illustrates the counter register of FIG. 33;

FIG. 41 is a block diagram of domain interconnections for the processor of FIG. 1;

FIG. 42 is a block diagram illustrating a stream scan register within the MTAP of FIG. 41;

FIG. 43 is a schematic of EMU pin connection for the processor of FIG. 1; and

FIG. 44 is a block diagram of a JTAG TAP configuration for the processor of FIG. 1.

FIG. 1 is a block diagram of a microprocessor 1 which has an embodiment of the present invention. Microprocessor 1 is a VLIW digital signal processor ("DSP"). In the interest of clarity, FIG. 1 only shows those portions of microprocessor 1 that are relevant to an understanding of an embodiment of the present invention. Details of general construction for DSPs are well known, and may be found readily elsewhere. For example, U.S. Pat. No. 5,072,418 issued to Frederick Boutaud, et al, describes a DSP in detail and is incorporated herein by reference. U.S. Pat. No. 5,329,471 issued to Gary Swoboda, et al, describes in detail how to test and emulate a DSP and is incorporated herein by reference. Details of portions of microprocessor 1 relevant to an embodiment of the present invention are explained in sufficient detail hereinbelow, so as to enable one of ordinary skill in the microprocessor art to make and use the invention.

In microprocessor 1 there are shown a central processing unit (CPU) 10, data memory 22, program memory 23, peripherals 60 and an external memory interface (EMIF) with a direct memory access (DMA) 61. CPU 10 further has an instruction fetch/decode unit 10a-c, a plurality of execution units, including an arithmetic and load/store unit D1, a multiplier M1, an ALU/shifter unit S1, an arithmetic logic unit ("ALU") L1, a shared multiport register file 20a from which data are read and to which data are written. Decoded instructions are provided from the instruction fetch/decode unit 10a-c to the functional units D1, M1, S1, and L1 over various sets of control lines which are not shown. Data are provided to/from the register file 20a from/to to load/store units D1 over a first set of busses 32a, to multiplier M1 over a second set of busses 34a, to ALU/shifter unit S1 over a third set of busses 36a and to ALU L1 over a fourth set of busses 38a. Data are provided to/from the memory 22 from/to the load/store units D1 via a fifth set of busses 40a. Note that the entire data path described above is duplicated with register file 20b and execution units D2, M2, S2, and L2. Instructions are fetched by fetch unit 10a from instruction memory 23 over a set of busses 41. Emulation circuitry 50 provides access to the internal operation of integrated circuit 1 which can be controlled by an external test/development system (XDS) 51.

External test system 51 is representative of a variety of known test systems for debugging and emulating integrated circuits. One such system is described in U.S. Pat. No. 5,535,331 which is incorporated herein by reference. Test circuitry 52 contains control registers and parallel signature analysis circuitry for testing integrated circuit 1.

Note that the memory 22 and memory 23 are shown in FIG. 1 to be a part of a microprocessor 1 integrated circuit, the extent of which is represented by the box 42. The memories 22-23 could just as well be external to the microprocessor 1 integrated circuit 42, or part of it could reside on the integrated circuit 42 and part of it be external to the integrated circuit 42. These are matters of design choice. Also, the particular selection and number of execution units are a matter of design choice, and are not critical to the invention.

When microprocessor 1 is incorporated in a data processing system, additional memory or peripherals may be connected to microprocessor 1, as illustrated in FIG. 1. For example, Random Access Memory (RAM) 70, a Read Only Memory (ROM) 71 and a Disk 72 are shown connected via an external bus 73. Bus 73 is connected to the External Memory Interface (EMIF) which is part of functional block 61 within microprocessor 42. A Direct Memory Access (DMA) controller is also included within block 61. The DMA controller is generally used to move data between memory and peripherals within microprocessor 1 and memory and peripherals which are external to microprocessor 1.

FIG. 2 is a block diagram of the execution units and register files of the microprocessor of FIG. 1 and shows a more detailed view of the buses connecting the various functional blocks. In this figure, all data busses are 32 bits wide, unless otherwise noted. Bus 40a has an address bus DA1 which is driven by mux 200a. This allows an address generated by either load/store unit D1 or D2 to provide an address for loads or stores for register file 20a. Data Bus LD1 loads data from an address in memory 22 specified by address bus DA1 to a register in load unit D1. Unit D1 may manipulate the data provided prior to storing it in register file 20a. Likewise, data bus ST1 stores data from register file 20a to memory 22. Load/store unit D1 performs the following operations: 32-bit add, subtract, linear and circular address calculations. Load/store unit D2 operates similarly to unit D1, with the assistance of mux 200b for selecting an address.

ALU unit L1 performs the following types of operations: 32/40 bit arithmetic and compare operations; left most 1, 0, bit counting for 32 bits; normalization count for 32 and 40 bits; and logical operations. ALU L1 has input src1 for a 32 bit source operand and input src2 for a second 32 bit source operand. Input msb.sub.-- src is an 8 bit value used to form 40 bit source operands. ALU L1 has an output dst for a 32 bit destination operands. Output msb.sub.-- dst is an 8 bit value used to form 40 bit destination operands. Two 32 bit registers in register file 20a are concatenated to hold a 40 bit operand. Mux 211 is connected to input src1 and allows a 32 bit operand to be obtained from register file 20a via bus 38a or from register file 20b via bus 210. Mux 212 is connected to input src2 and allows a 32 bit operand to be obtained from register file 20a via bus 38a or from register file 20b via bus 210. ALU unit L2 operates similarly to unit L1.

ALU/shifter unit S1 performs the following types of operations: 32 bit arithmetic operations; 32/40 bit shifts and 32 bit bit-field operations; 32 bit logical operations; branching; and constant generation. ALU S1 has input src1 for a 32 bit source operand and input src2 for a second 32 bit source operand. Input msb.sub.-- src is an 8 bit value used to form 40 bit source operands. ALU S1 has an output dst for a 32 bit destination operands. Output msb.sub.-- dst is an 8 bit value used to form 40 bit destination operands. Mux 213 is connected to input src2 and allows a 32 bit operand to be obtained from register file 20a via bus 36a or from register file 20b via bus 210. ALU unit S2 operates similarly to unit S1, but can additionally perform register transfers to/from the control register file 102.

Multiplier M1 performs 16.times.16 multiplies. Multiplier M1 has input src1 for a 32 bit source operand and input src2 for a 32 bit source operand. ALU S1 has an output dst for a 32 bit destination operands. Mux 214 is connected to input src2 and allows a 32 bit operand to be obtained from register file 20a via bus 34a or from register file 20b via bus 210. Multiplier M2 operates similarly to multiplier M1.

As depicted in FIG. 2, one unit (.S2) can read from and write to the control register file 102 using buses 220 and 221. Table 2 lists the control registers contained in the control register file, and briefly describes each. The control registers are described more fully later herein. Each control register is accessed by the MVC instruction; see the MVC instruction description later herein.

FIG. 3 shows the addressing mode register, (AMR). Eight registers (A4-A7, B4-B7) can perform circular addressing. For each of these registers, the AMR specifies the addressing mode. A 2-bit field for each register is used to select the address modification mode: linear (the default) or circular mode. With circular addressing, the field also specifies which BK (block size) field to use for a circular buffer. In addition, the buffer must be aligned on a byte boundary equal to the block size. The mode select field encoding is shown in Table 3.

The block size fields, BK0 and BK1, specify block sizes for circular addressing. The five bits in BK0 and BK1 specify the width. The formula for calculating the block size width is:

where N is the value in BK1 or BK0

Table 4 block size calculations for all 32 possibilities.

The control status register (CSR), shown in FIG. 4, contains control and status bits. The function of the bit fields in the CSR are shown in Table 5.

A general-purpose input register (IN) depicted in FIG. 5 supports 32 general-purpose input signals, while a general-purpose output register (OUT) depicted in FIG. 6 supports 32 general-purpose output signals. The function of these signals is described later herein.

Table 6 below explains various symbols used herein.

Table 7 and Table 8 define the mapping between instructions and functional units.

The general-purpose register file supports 32- and 40-bit data. 32-bit data is contained in single registers. 40-bit data is contained across two registers; the 32 LSBs of the data are stored in an even register and the 8 MSBs are stored in the 8 LSBs of the next register (which is always an odd register). There are 16 valid register pairs for 40-bit data as shown in Table 9. In assembly language syntax, the register pairs are denoted by a colon in between the register names. The odd register is specified first.

FIG. 7 illustrates the register storage scheme for 40-bit data. Operations requiring a long input ignore the 24 MSBs of the odd register. Operations producing a long result zero-fill the 24 MSBs of the odd register. The even register is encoded in the opcode.

The DSP's opcode map is shown in FIGS. 8A-8J. Refer to Table 6 and the instruction descriptions later herein for explanations of the field syntaxes and values.

All instructions can be conditional. The condition is controlled by a 3-bit (creg) field specifying the register tested, and a 1-bit field (z) specifying a test for zero or nonzero. The four MSBs of every opcode are creg and z. The register is tested at the beginning of the E1 pipeline stage for all instructions. The pipeline is described later herein. If z=1, the test is for equality with zero. If z=0, the test is for nonzero. The case of condition register field (creg)=0 and z=0 is treated as always true to allow instructions to be executed unconditionally. The creg register field is encoded as shown in Table 10.

Conditional instructions are represented by "[ ]" surrounding the condition register. The following execute packet contains two ADD instructions in parallel. The first ADD is conditional on B0 being non-zero. The second ADD is conditional on B0 being zero. `!` indicates the `not` of the condition.

The above instructions are mutually exclusive. This means that only one will execute.

If they are scheduled in parallel, mutually exclusive instructions must still follow all resource constraints mentioned later herein.

If mutually exclusive instructions share any resources as described later herein, they cannot be scheduled in parallel (put in the same execute packet), even though only one will end up executing.

The execution of instructions can be defined in terms of delay slots. Table 11 shows the types of instructions, how many delay slots each type instruction has, and the execute phases it uses. Delay slots are the number of extra cycles it takes before a result is available to be read after the source operands are read. For a single-cycle type instruction (such as ADD), if source operands are read in cycle i, the result can be read in cycle i+1. For a multiply instruction (MPY), if source operands are read in cycle i, the result can be read in cycle I+2.

Instructions are always fetched eight at a time. This constitutes a fetch packet. The basic format of a fetch packet is shown in FIG. 9. The execution grouping of the fetch packet is specified by the p-bit, bit zero, of each instruction. Fetch packets are 8-word aligned.

The p bit controls the parallel execution of instructions. The p bits are scanned from left to right (lower to higher address). If the p bit of instruction i is 1, then instruction i+1 is to be executed in parallel with (in the same cycle as) instruction i. If the p-bit of instruction i is 0, then instruction i+1 is executed in the cycle after instruction i. All instructions executing in parallel constitute an execute packet. An execute packet can contain up to eight instructions. All instructions in an execute packet must use a unique functional unit.

An execute packet cannot cross an 8-word boundary. Therefore, the last p bit in a fetch packet is always set to 0, and each fetch packet starts a new execute packet. The following examples illustrate the conversion of a p-bit sequence into a cycle-by-cycle execution stream of instructions. There are three types of p-bit patterns for fetch packets. These three p-bit patterns result in the following execution sequences for the eight instructions: fully serial; fully parallel; or partially serial. These three sequences of execution are explained more fully below.

The fully serial p-bit pattern depicted in FIG. 10A results in this execution sequence:

The eight instructions are executed sequentially.

The fully parallel p-bit pattern depicted in FIG. 10B results in this execution sequence:

All eight instructions are executed in parallel.

The partially serial p-bit pattern depicted in FIG. 10C results in this execution sequence:

The .parallel. characters signify that an instruction is to execute in parallel with the previous instruction. In the previous partially serial example, the code would be represented as this:

If a branch into the middle of an execution packet occurs, all instructions at lower addresses are ignored. In the partially serial example, if a branch to the address containing instruction D occurs, then only D and E will execute. Even though instruction C is in the same execute packet, it is ignored. Instructions A and B are also ignored because they are in earlier execute packets.

No two instructions within the same execute packet can use the same resources. Also, no two instructions can write to the same register during the same cycle. The following describes each of the resources an instruction can use.

Two instructions using the same functional unit cannot be issued in the same execute packet.

The following execute packet is invalid:

The following execute packet is valid:

Cross Paths (1X and 2X): one unit (either a .S, .L, or .M) per data path, per execute packet, can read a source operand from its opposite register file via the cross paths (1X and 2X). For example, .S1 can read both operands from the A register file, or one operand from the B register file using the 1X cross path. This is denoted by an X following the unit name.

Two instructions using the same X cross path between register files cannot be issued in the same execute packet since there is only one path from A to B and one path from B to A.

The following execute packet is invalid:

The following execute packet is valid:

The operand will come from a register file opposite of the destination if the x bit in the instruction field is set (as shown in the opcode map).

Loads and stores can use an address pointer from one register file while loading to or storing from the other register file. Two loads and/or stores using an address pointer from the same register file cannot be issued in the same execute packet.

The following execute packet is invalid:

The following execute packet is valid:

Two loads and/or stores loading to and/or storing from the same register file cannot be issued in the same execute packet.

The following execute packet is invalid:

The following execute packet is valid:

Only one long result may be written per cycle on each side of the register file. Because the .S and .L units share a read register port for long source operands and a write register port for long results, only one may be issued per side in an execute packet.

The following execute packet is invalid:

The following execute packet is valid:

Because the .L and .S units share their long read port with the store port, operations that read a long value cannot be issued on the .L and/or .S units in the same execute packet as a store.

The following execute packet is invalid:

The following execute packet is valid:

More than four reads of the same register cannot occur on the same cycle. Conditional registers are not included in this count.

The following code sequence is invalid:

whereas this code sequence is valid:

Multiple writes to the same register on the same cycle can occur if instructions with different latencies writing to the same register are issued on different cycles. For example, an MPY issued on cycle i followed by an ADD on cycle i+1 cannot write to the same register since both instructions will write a result on cycle i+1. Therefore, the following code sequence is invalid:

MPY .M1 A0,A1,A2

ADD .L1 A4,A5,A2

Table 12 shows different multiple write conflicts. For example, the ADD and SUB in execute packet L1 write to the same register. This conflict is easily detectable.

The MPY in packet L2 and the ADD in packet L3 might both write to B2 simultaneously; however, if a branch instruction causes the execute packet after L2 to be something other than L3, this would not be a conflict. Thus, the potential conflict in L2 and L3 might not be detected by the assembler. The instructions in L4 do not constitute a write conflict because they are mutually exclusive. In contrast, because it is not obvious that the instructions in L5 are mutually exclusive, the assembler cannot determine a conflict. If the pipeline does receive commands to perform multiple writes to the same register, the result is undefined.

The addressing modes are linear, circular using BK0, and circular using BK1. The mode is specified by the addressing mode register (AMR).

Eight registers can perform circular addressing. A4-A7 are used by the .D1 unit and B4-B7 are used by the .D2 unit. No other units can perform circular addressing modes. For each of these registers, the AMR specifies the addressing mode.

The following instructions all use the AMR to determine what type of address calculations are performed for these registers: LD(B)(H)(W), ST(B)(H)(W), ADDA(B)(H)(W), and SUBA(B)(H)(W). All registers can perform linear mode addressing.

Linear mode addressing operates as follows with LD/ST Instructions: linear mode simply shifts the offsetR/cst operand to the left by 2, 1, or 0 for word, half-word, or byte access respectively and then performs an add or subtract to baseR, (depending on the operation specified).

Linear mode addressing operates as follows with ADDA/SUBA Instructions: linear mode simply shifts src1/cst operand to the left by 2, 1, or 0 for word, halfword, or byte access respectively and then performs an add or subtract, (depending on the operation specified).

Circular mode addressing uses the BK0 and BK1 fields in the AMR to specify block sizes for circular addressing. Circular mode addressing operates as follows with LD/ST Instructions: after shifting offsetR/cst to the left by 2, 1, or 0 for LDW, LDH, or LDB respectively, an add or subtract is performed with the carry/borrow inhibited between bits N and N+1. Bits N+1 to 31 of baseR remain unchanged. All other carries/borrows propagate as usual. Thus, if an offsetR/cst is greater than the circular buffer size, 2.sup.(N+1), is specified, the address will be outside of the circular buffer. The circular buffer size in the AMR is not scaled; for example: a size of 4 is 4 bytes, not 4.times.size of (type). So, to perform circular addressing on an array of 8 words, a size of 32 should be specified, or N=4. Table 12 shows an LDW performed with register A4 in circular mode, with BK0=4, so the buffer size is 32 bytes, 16 halfwords, or 8 words. The value put in the AMR for this example is 0004 0001h.

Circular mode addressing operates as follows with ADDA/SUBA Instructions: after shifting src1/cst to the left by 2, 1, or 0 for ADDAW, ADDAH, or ADDAB respectively, an add or subtract is performed with the carry/borrow inhibited between bits N and N+1. Bits N+1 to 31, inclusive, of src2 remain unchanged. All other carries/borrows propagate as usual. Thus, if src1 is greater than the circular buffer size, 2.sup.(N+1), is specified, the address will be outside of the circular buffer. The circular buffer size in the AMR is not scaled, for example: a size of 4 is 4 bytes, not 4.times.size of (type). So, to perform circular addressing on an array of 8 words, a size of 32 should be specified, or N=4. Table 14 shows an ADDAH performed with register A4 in circular mode, with BK0=4, so the buffer size is 32 bytes, 16 halfwords, or 8 words. The value put in the AMR for this example is 0004 0001h.

An instruction syntax is used to describe each instruction. The opcode map breaks down the various bit fields that make up each instruction. There are certain instructions that can be executed on more than one functional unit, as was shown in Table 8. The syntax specifies the functional unit and various resources used by an instruction, typically as follows:

EXAMPLE (.unit) src, dst

This is what the syntax looks like for the ADD instruction:

ADD (.unit) src1, src2, dst

OR

ADDU (.unit) src1, src2, dst

OR

ADD (.unit) src2, src1, dst

unit=.L1, .L2, S1, .S2, .D1, .D2

src and dst indicate source and destination respectively. The (.unit) dictates which functional unit the instruction is mapped to (.L1, .L2, .S1, .S2, .M1, .M2, .D1, or .D2). This instruction has three opcode map fields: src1, src2, and dst.

Pipeline Operation

The DSP pipeline has several key features which improve performance, decrease cost, and simplify programming. They are: increased pipelining eliminates traditional architectural bottlenecks in program fetch, data access, and multiply operations; control of the pipeline is simplified by eliminating pipeline interlocks; the pipeline can dispatch eight parallel instructions every cycle; parallel instructions proceed simultaneously through the same pipeline phases; sequential instructions proceed with the same relative pipeline phase difference; and load and store addresses appear on the CPU boundary during the same pipeline phase, eliminating read-after-write memory conflicts.

A multi-stage memory pipeline is present for both data accesses and program fetches. This allows use of high-speed synchronous memories both on-chip and off-chip, and allows infinitely nestable zero-overhead looping with branches in parallel with other instructions.

There are no internal interlocks in the execution cycles of the pipeline, so a new execute packet enters execution every CPU cycle. Therefore, the number of CPU cycles for a particular algorithm with particular input data is fixed. If during program execution, there are no memory stalls, the number of CPU cycles equals the number of clock cycles for a program to execute.

Performance can be inhibited only by stalls from the memory subsystems or interrupts. The reasons for memory stalls are determined by the memory architecture. To fully understand how to optimize a program for speed, the sequence of program fetch, data store, and data load requests the program makes, and how they might stall the CPU should be understood.

The pipeline operation, from a functional point of view, is based on CPU cycles. A CPU cycle is the period during which a particular execute packet is in a particular pipeline stage. CPU cycle boundaries always occur at clock cycle boundaries; however, memory stalls can cause CPU cycles to extend over multiple clock cycles. To understand the machine state at CPU cycle boundaries, one must be concerned only with the execution phases (E1-E5) of the pipeline. The phases of the pipeline are shown in FIG. 11 and described in Table 15.

The pipeline operation of the instructions can be categorized into seven types shown in Table 16. The delay slots for each instruction type are listed in the second column.

The execution of instructions can be defined in terms of delay slots (Table 16). A delay slot is a CPU cycle that occurs after the first execution phase (E1) of an instruction in which results from the instruction are not available. For example, a multiply instruction has 1 delay slot, this means that there is 1 CPU cycle before another instruction can use the results from the multiply instruction.

Single cycle instructions execute during the E1 phase of the pipeline. The operand is read, operation is performed and the results are written to a register all during E1. These instructions have no delay slots.

Multiply instructions complete their operations during the E2 phase of the pipeline. In the E1 phase, the operand is read and the multiply begins. In the E2 phase, the multiply finishes, and the result is written to the destination (dst) register. Multiply instructions have 1 delay slot.

Load instructions have two results: data loaded from memory and address pointer modification.

Data loads complete their operations during the E5 phase of the pipeline. In the E1 phase, the address of the data is computed. In the E2 phase, the data address is sent to data memory. In the E3 phase, a memory read is performed. In the E4 stage, the data is received at the CPU core boundary. Finally, in the E5 phase, the data is loaded into a register. Because data is not written to the register until E5, these instructions have 4 delay slots. Because pointer results are written to the register in E1, there are no delay slots associated with the address modification.

Store instructions complete their operations during the E3 phase of the pipeline. In the E1 phase, the address of the data is computed. In the E2 phase, the data address is sent to data memory. In the E3 phase, a memory write is performed. The address modification is performed in the E1 stage of the pipeline. Even though stores finish their execution in the E3 phase of the pipeline, they have no delay slots and follow the following rules (i=cycle):

1) When a load is executed before a store, the old value is loaded and the new value is stored.

2) When a store is executed before a load, the new value is stored and the new value is loaded.

3) When the instructions are in are in parallel, the old value is loaded and the new value is stored.

Branch instructions execute during the E1 phase of the pipeline five delay slots/CPU cycles after the branch instruction enters an initial E1 phase of the pipeline. FIG. 12 shows the branch instruction phases. FIG. 13 shows the operation of the pipeline based on clock cycles and fetch packets. In FIG. 13, if a branch is in fetch packet n, then the E1 phase of the branch is the PG phase of n+6. In cycle 7 n is in the E1 phase and n+6 is in the PG phase. Because the branch target is in PG on cycle 7, it will not reach E1 until cycle 13. Thus, it appears as if the branch takes six cycles to execute, or has five delay slots.

In FIG. 14, fetch packet n, which contains three execute packets, is shown followed by six fetch packets (n+1 through n+6), each with one execution packet (containing 8 parallel instructions). The first fetch packet (n) goes through the program fetch phases during cycles 1-4. During these cycles a program fetch phase is started for each of the following fetch packets.

In cycle 5, the program dispatch (DP) phase, the CPU scans the p bits and detects that there are three execute packets (k thru k+2) in fetch packet n. This forces the pipeline to stall, which allows the DP phase to start execute packets k+1 and k+2 in cycles 6 and 7. Once execute packet k+2 is ready to move on to the DC phase (cycle 8) the pipeline stall is released.

The fetch packets n+1 through n+4 were all stalled so the CPU would have time to perform the DP phase for each of the three execute packets (k thru k+2) in fetch packet n. Fetch packet n+5 was also stalled in cycles 6 and 7; it was not allowed to enter the PG phase until after the pipeline stall was released in cycle 8. The pipeline will continue as shown with fetch packets n+5 and n+6 until another fetch packet containing multiple execution packets enters the DP phase, or an interrupt occurs.

Pipeline discontinuities are caused by memory stalls, multi-cycle NOPs, and the STP instruction. During a memory stall, the CPU cycle (which normally occurs during one clock cycle) occurs in two or more cycles. All of the pipeline phases stall during these extra clock cycles. The results of the program execution are identical with or without the stall. With a memory stall execution will take more clock cycles to finish.

The NOP count instruction provides count cycles of NOPs. If count .gtoreq.2, the NOP is a multi-cycle NOP. A NOP 2 for example, fills in extra delay slots for the instructors in the execute packet it is contained in and for all previous execute packets. Thus, if a NOP 2 is in parallel with an MPY instruction, the MPY's results will be available for use by instructions in the next execute packet. If the delay slots of a branch complete while a multi-cycle NOP is still dispatching NOPs into the pipeline, the branch overrides the multi-cycle NOP and the branch target begins execution after 5 delay slots.

STP is an advanced instruction that can be used only when both these conditions are met: 1) it cannot contain a parallel branch instructor that would force a program fetch, and 2) no program fetch is happening because either its second delay slot contains a multi-cycle NOP, or its third and fourth delay slot execute packets are in the same fetch packet.

Memory System

DSP program memory system 23 includes 64K bytes of memory and a memory/cache controller. The program memory can operate as either a 64K byte internal program memory or as a direct mapped program cache. There are four modes under which the program memory system operates: program memory mode; cache enable mode; cache freeze mode; and cache bypass mode. The mode under which the program memory operates is determined by the program cache control (PCC) field (bits 5-7) in the CSR (FIG. 4). Table 17 shows various PCC values for configuring program memory system 23.

When the PCC field of the CSR contains the value 000b, the program memory is mapped as valid program memory space. The addresses that make up the program memory map depend on the value on the MP.sub.-- BOOT pin on the device.

Emulation Features

An aspect of the present invention involves novel and improved techniques for emulating the operation of DSP 1 in order to develop software programs or to test DSP 1 for correct operation. Portions of DSP 1 which relate to emulation will now be described in more detail.

Referring again to FIG. 1, CPU 10 has emulation circuitry 50 and interrupt circuitry 90 to support the following emulation functions which will be described in more detail: execution and scan control through the test ports; analysis support; and real-time emulation support.

Execution and scan control through the test ports involves halting CPU 10. CPU Halt Support is provided in the following ways: A RDY-based CPU halt based on a software breakpoint (SWBP) or an analysis event.

Analysis support includes the following: a single, exact-match hardware program address breakpoint (PABP); analysis events which can be triggered by the EMUOIN or EMU1IN inputs from the Megamodule Test Access Port (MTAP), or by a program address breakpoint; and a Special Emulation Event input (SEE) which can trigger an analysis event.

Real-time emulation support includes message passing and a CPU Analysis Interrupt (AINT) based on a software interrupt, an analysis event, or the next cycle boundary.

Referring now to FIG. 15, emulation circuitry 50 is illustrated in more detail. Megamodule Test Access Port (MTAP) 305 is connected to CPU test port (CPUTP) 310, analysis test port (ATP) 320, and Megamodule test port (APT) 330. Three domains, CPU domain 10, analysis domain 321, and Megamodule domain 331 are interconnected with the test ports. MTAP 305 provides scan and execution control to the various domains within the megamodule. The test ports provide an interface for each domain to the MTAP. Test ports also generate, distribute, and perform clock switching functions for the functional and scan clocks on the megamodule. MTAP 305 provides an interface between XDS 51 and the CPU real-time analysis and message passing features. According to an aspect of the present invention, MTAP 305 provides data streaming for fast memory download/upload. MTAP 305 also supports performance analysis through an event counter and test port control of execution and clocking for both emulation and test. The operation and design of emulation circuitry 50, including MTAP 305 and test ports 310, 320 and 330 will be described in detail in the following pages.

A Special Emulation (SE) device interfaces to MTAP 305 and Megamodule 330 as a whole to provide increased debug, trace, and breakpoint capability. A special emulation device (SE) will have a fourth domain for the SE Analysis (SEA) domain circuitry, which resides outside the megamodule. The SEA domain's interface to the Megamodule through the MTAP is also described in detailed in the following pages. The SEA domain includes: hardware data breakpoints on data and address; hardware program breakpoints on addresses of execute packets dispatched for execution; tracing of executed program addresses, program and data memory accesses; taken interrupts, functional unit usage, and branches; and event counters and sequencers.

FIG. 16 is a timing diagram that shows a Megamodule reset operation and related signals. Note that throughout this process an inactive RDY can still stall the CPU, in other words extend CPU cycles over multiple clock cycles. The sequence of events that occurs is as follows:

1. Clock n+2: This is the first clock cycle after NRESET goes low. The following actions occur if the CPU test port is in the FUNC state:

A) All internal tri-state busses tri-state. They remain tri-stated until cycle 6.

B) LOGXOFFD is asserted indicating all non-program memory system devices to ignore megamodule strobes (DBS, PWRDN, JACK).

C) LOGXOFFP is asserted indicating the program memory should ignore program memory strobes: PAS, PDS, PWS.

2. Cycle n+2: If the CPU test port is in the FUNC state, all instructions in the DP and E1 pipeline phases are annulled.

3. Cycle 1: An indefinite number of clock cycles later, reset interrupt processing occurs with the rising edge of NRESET.

4. Cycle 6: When JACK goes active all registers and boundary signals are set to their reset values except for PWS, PAS, PDS, LOGXOFFD is de-asserted.

5. Cycle 7: As the first PAS is active LOGXOFFP is de-asserted.

Interrupt Operation

The CPU has 14 interrupts available for normal DSP operation. These are reset, the non-maskable interrupt (NMI) and interrupts 4-15. These interrupts corresponds to the RESET, NMI, and INT4-INT15 signals on the CPU boundary. For some embodiments, these signals may be tied directly to pins on the device, may be connected to on-chip peripherals, or may be disabled by being permanently tied inactive on chip. Generally, RESET and NMI are directly connected to pins on the device.

The priorities of these interrupts are listed in Table 18. A low-to-high transition on an interrupt pin sets the pending status of the interrupt within the interrupt flag register (IFR). If the interrupt is properly enabled, the CPU begins processing the interrupt and redirecting program flow to the interrupt service routine.

The CPU cannot be prevented from processing a reset. A reset begins to be processed when RESET experiences a low-to-high transition. Unlike the other interrupts, the RESET signal is labeled active-low. A low value on RESET has the effect of stopping all CPU processing and returning all registers to their reset values.

The non-maskable interrupt (NMI) is the interrupt of second-highest priority. Two conditions prevent NMI from causing interrupt processing: the CPU is in the delay slots of a branch, whether the branch is taken or not; and the NMI-enable bit (NMIE) in the interrupt enable register (IER) is 0. NMIE is cleared upon reset to prevent interruption of processor initialization and upon NMI processing to prevent re-interruption of an NMI by another NMI. NMI is re-enabled by setting NMIE or by completing the execution of a B NRP instruction.

If NMIE is 0, INT4-INT15 are disabled. During NMI processing the return pointer, which continues previous program execution, is stored in the NMI return pointer register (NRP). Thus, the B NRP instruction returns to the previous program flow after servicing the NMI. Table 19 shows how to return for an NMI.

The following conditions can prevent INT4-INT15 from causing interrupt processing: the CPU is processing code which resides in the delay slots of a branch and this includes conditional branches that do not complete execution due to a false conditions; the NMIE bit in the interrupt enable register (IER) is 0; the corresponding interrupt enable (IE) bit in the IER is 0; or the global interrupt enable bit (GIE) bit in the control status register (CSR) is 0.

During interrupt processing the return pointer, which continues previous program execution, is stored in the interrupt return pointer register (IRP). Thus, the B IRP instruction returns to the program flow after servicing the interrupt. Table 20 shows how to return from a maskable interrupt.

The IACK and INUM signals alert hardware, external to the device 11, when interrupts have taken place. The IACK signal indicates that the CPU has begun processing an interrupt. The INUMx signals (INUM0-INUM3) indicate the number of the interrupt (bit position in the IFR) that is being processed.

Table 21 lists the seven interrupt control registers on the device.

The IFR and the ISR share a register address. The IFR can be read from, and the ISR can be written to. The other registers have unique addresses.

An interrupt can trigger interrupt processing only if the corresponding bit in the Interrupt Enable Register (IER) is set. Bit 0, corresponding to reset, is not writable and is always read as 1. The RESET interrupt is always enabled. RESET cannot be disabled. Bits IE4-IE15 may be written as 1 or 0, enabling or disabling the associated interrupt, respectively. The IER is shown in FIG. 17B.

The NMIE, if clear, disables all nonreset interrupts, preventing interruption of NMI. The NMI enable (NMIE) is unaffected by a write of 0 but is set by a write of 1. NMIE is initialized to 0 upon reset to prevent any interruption of processor initialization until enabled by you. After reset, you must set NMIE to enable NMI and to allow INT15-INT4 to be enabled by GIE and the appropriate IE bit. You cannot manually clear the NMIE. The NMIE is cleared by the occurrence of an NMI. If cleared, NMIE is set only by completing a B NRP instruction or by a write of 1 to NMIE.

The interrupt flag register (IFR) (see FIG. 17A) contains the status of INT4-INT15 and NMI. Table 22 lists the interrupt flags and the interrupts they correspond to. If you want to check the status of interrupts, use the MVC instruction to read the IFR.

The interrupt set register (ISR) and the interrupt clear register (ICR) (see FIG. 17C and FIG. 17D) allow interrupts to be set or cleared manually in the IFR. Writing a 1 to IS4-IS15 of the ISR causes the corresponding interrupt flag to be set. Similarly, writing a 1 to a bit of the ICR causes the corresponding interrupt flag to be cleared. Writing a 0 to any bit of either the ISR or the ICR has no effect. Incoming interrupts have priority and override any write to the ICR. Reset or NMI cannot be set or cleared. Any write to the ISR or ICR (by the MVC instruction) effectively has one delay slot because the results cannot be read (by the MVC instruction) in the IFR until 2 cycles after the write to the ISR or ICR.

Although they do no form a CPU control register, the interrupt pending bits hold the pending status of all CPU interrupts. RSTP, NMIP, AIP, MSGIP, and IP4-IP15 correspond to RESET, NMI, AINT, MSGINT, and INT4-INT15, respectively. The IP bits are set upon recognition of an interrupt. These bits, which are not directly visible to the user, reside in megamodule domain 331 and are updated every clock (i.e. not stalled by an in-active RDY). The user may observe the status of the IP bits, through the IFR which is updated to the value of the IP bits every cycle. The user may affect the status of the IP bits through writes of the Interrupt Set Register (ISR) and Interrupt Clear Register (ICR). These changes occur on the next update of the IP bits to the IFR. The IP-bits are all cleared at RESET. The CPU registers described in this section all reside within the CPU domain and are updated every cycle (i.e. are stalled by an inactive RDY).

The following bits are "reserved" in the sense that they are not available for use by a software program during normal operation of CPU 10, but they are available for use during emulation and testing.

IFR: Bits 2 and 3 of the IFR are reserved for the analysis interrupt (AINT) and message interrupt (MSGINT), respectively.

IER: Bits 2 and 3 of the IER are reserved for the analysis interrupt enable (AIE) and message interrupt enable (MSGIE), respectively.

ISR: Bits 2 and 3 of the ISR are reserved for the analysis interrupt set (AIS) and message interrupt set (MSGIS), respectively. These bits can only be used to set their associated IP bits if the EMU.sub.-- MVCEN is set via scan.

ICR: Bits 2 and 3 of the ICR are reserved for the analysis interrupt clear (AIC) and message interrupt clear (MSGIC), respectively. These bits can only be used to clear their associated IP bits if the EMU.sub.-- MVCEN is set via scan.

An analysis interrupt can be triggers by certain types of events. Certain events trigger a halt. For a halt to occur the CPU test port must be in the CNTL state. An exception is that an external halt request from the CPU test port will happen regardless of the execute state.

Certain events can trigger either a halt or an analysis interrupt, as follows:

1) Special Emulation Event (SEE) input in an active state.

2) On-chip program address breakpoint (PABP).

3) EMU0IN and EMU1IN inputs transitioning from inactive to active.

4) A floating-point resource conflict (FPX event).

Certain other events can trigger only an analysis interrupt, as follows:

5) A software interrupt SWI instruction.

6) The next cycle boundary (CYC event).

7) The XAINT signal from the MTAP.

Certain other events can trigger only a halt, as follows:

8) SWBP decoded. This indicates the creg field of the first instruction in the execute packet contains the SWBP code (0001).

9) External halt request from the CPU test port.

Events must be active long enough to be recognized if enabled, regardless of the stalled state of the processor. However, they cannot be active so long as to cause multiple events. Edge sensitive circuitry is used to eliminate time constraints on event length.

Many of the events described above are classified as analysis events and will be ignored when the SUSPEND signal (Table 29) within the CPU is active. However, the following events are no classified as analysis events and will not be affected by SUSPEND: SWBP, SWI, and an external halt requested from the test port.

The SUSPEND signal is driven by the OR of four terms:

1. CPU.sub.-- DONE active

2. The ECTL bit active (Table 28)

3. The CPU test port being in the PAUS, SDAT, or SCTL states (Table 33)

4. The AINTSUSP signal in the Analysis Control Register (Table 34).

Referring now to FIG. 18, which is a timing diagram illustrating detection of analysis interrupts. An analysis interrupt (AINT) may be caused by one of 7 sources:

1. EMUOIN MTAP to CPU signal.

2. EMU1IN MTAP to CPU signal.

3. XAINT signal from the MTAP. An interrupt is only generated if the CPU test port is in CTRL.

4. Special Emulation Event (SEE) megamodule input.

5. On-megamodule Program Address Breakpoint (PABP).

6. Software Interrupt Instruction (SWI). Unlike SWBP, SWI does not required the CPU test port to be in the CNTL state. The STP instruction is available to write to program memory to set SWI.

7. Crossing the next cycle boundary in execution (CYC). If enabled the first cycle boundary after a B ARP's target execute packet completes E1 will trigger a CYC event, setting AIP. This will be used for interrupt-based single-stepping with a monitor.

Some of these events require enabling through the Analysis Control Register (FIG. 30A). AIP (and thus the AIF bit in the IFR (FIG. 17A) is only cleared by processing the interrupt or by a write of 1 to bit 2 of the ICR (FIG. 17D). These interrupts like all others cannot occur while in the delay slots of a branch. Recognition is postponed until branch processing is completed. FIG. 18 shows the detection of these events. AIP is only set if the indicated interrupt is enabled by AIE and PRI (and GIE, IE, NMIE) as necessary.

A bit in the Analysis Control Register (FIG. 30A) can change the maskability and priority of interrupts. By setting PRI, AINT can be treated either a second non-maskable interrupt (PRI=1) which is of highest priority or the second lowest priority maskable interrupt (PRI=O). In some systems, analysis must react to events as quickly as possible. In these, the non-maskable mode should be used. In other systems, must disrupt the program flow as little as possible. In this case the maskable, low priority mode should be used.

If PRI=1, AIE disable all non-RESET interrupts. Also, AINT is not affected by GIE or NMIE. NMIE disables all non-RESET/non-AINT interrupts, preventing interruption of an NMI except by RESET or AINT. Note, for an interrupt to be reflected in HPIENT it does not have to be enabled by AIE when PRI=1 (unless it is AINT),

When PRI=1 the following different constraints on detection of an interrupt needing processing are imposed:

1. Neither GIE nor NMIE have to be set for an AINT to be taken.

2. For all interrupts except RESET, AIE must be set (AIE=1).

When PRI=1, the following differing actions occur during interrupt processing:

1. During an AINT, PGIE is not set to GIE and GIE is not cleared

2. AIE is cleared.

Regardless of the value of PRI, the return address for an AINT is always saved in the Analysis Interrupt Return Pointer Register (ARP), illustrated in FIG. 30C. A distinct ARP from NRP and IRP (see Table 20) is necessary as if PRI=1 in the ACR, an AINT may interrupt an NMI or other interrupt, and overwrite the value in the IRP.

FIG. 19A and FIG. 19B illustrate two analysis interrupt related instructions, SWI and BARP, respectively. Tables 23 and 24 describe these two instructions. Table 25 shows a return from AINT code sequence.

When an analysis interrupt begins processing, the AINTSUSP bit in the ACR signal will be set to disable recognition of any future analysis events interrupts or halts (i.e. AIP will not be set). AINTSUSP can only be cleared by a user write.

Referring again to FIG. 18, the operation of analysis interrupts in the presence of other interrupts will now be discussed. If the previous execute packet, n+4, had an interrupt (enabled in all necessary ways) occur in its DP phase, that interrupt will be postponed and the AINT taken. Thus, n+4 and n+5 will be annulled. Upon returning to n+5 any conditions which caused a program address based AIP will reoccur.

If the next execute packet, n+6, had an interrupt (enabled in all necessary ways) occur in its DP phase the AINT will take precedence because it occurred prior to the next interrupt. Thus, there does not need to be any special handling for program address based analysis interrupts: SWI, PABP, SEE, and CYC based AINTs.

Test Ports

Referring again to FIG. 15, each domain 10, 321 and 331 interfaces to MTAP 305 through test ports 310, 320 and 330, respectively. The MTAP supports up to four test ports, one for each domain. However, the test port for the SEA domain is not included in the megamodule. Test ports are generic test/emulation interfaces that provide domain clock generation as well as execution and scan control. Execution control allows the emulation or test software direct clock-by-clock control of the domain. The MTAP to test port interface is shown in described in more detail in Table 26 and Table 27. The interface between the SEA domain and the MTAP is similar and is described in detail with reference to FIG. 2 and FIG. 41.

Tables 28, 29, 30, 31 and 32 along with FIGS. 20 and 21 describe and define various signals and control bits relating to emulation and testing of Megamodule 300 which will be described in more detail later. Table 28 defines CUP domain bits that are used for emulation control and to provide status information that are transferred using scan chain 301a-d. Table 29 defines signals which are sent across Megamodule boundary 301. FIG. 20 illustrates interconnections between MTAP 305 and CPU domain 10. Table 30 defines signals illustrated by FIG. 20 which also drive Megamodule boundary 301. Table 31 defines the remaining signals illustrated by FIG. 20. FIG. 21 illustrates interconnections between MTAP 305 and Megamodule domain 331. Table 32 defines the signals illustrated in FIG. 21.

Various terms used in Tables 28-32 are as defined as follows:

Annul: If an execute packet is annulled during a particular pipeline stage, it does not set active any megamodule boundary strobes or status, write any values to user-visible or emulation-visible state. Address and data may change state as they are qualified by strobes as may an status signals which must be qualified by a strobe. Unqualified status must be set inactive. Annulling also forces an instruction to be annulled in future pipeline stages.

Analysis Control Register (ACR): Illustrated in FIG. 30A

Analysis Data Register (ADR): Illustrated in FIG. 30B

PRI Interrupt Priority control bit in the ACR: described in Table 34.

FIG. 22 is a state diagram of various states in which a test port can reside. A set of MTAP signals C0, C1, and Ce (Table 27) and the LOCK signal (Table 26) determine the state of the test port, as described in Table 33 and illustrated in FIG. 22. Each test port 310, 320 and 330 has an independent lock signal from the MTAP (FIG. 15). An active lock signal causes the current state to be latched within the test port until the lock signal is inactive. When a test port is not locked, the test port bus determines the state of the test port. When the test port is locked, signal changes on the test port bus and clock switching are ignored. If a domain is locked its scan modules are not added to the chain and are bypassed through the test port's Shared Shift Register (SSR) bits. The LOCK signal also locks the current clock generation mode of the device.

Software operating on XDS 51 guarantees that a test port transitions properly when the test port is unlocked with a MTAP code applied to the test port that is different from the code it was locked in. When switching from FUNC, CNTL, or HALT to a scan state (SCTL or SDAT), PAUS is transitioned through, as shown in FIG. 22. The MTAP does not allow the PAUS state to be exited until a clock switch completes, as will be described later.

If the test port is locked in PAUS, the emulation software must unlock the test port in PAUS. If it was unlocked in any other states, the resulting clock switch would contain glitches.

Referring again to FIG. 15, each test port supplies a Domain Bus Enable (DBENB) signal to its domain scan paths. If inactive, DBENB forces all domain boundary signals inactive and sets all domain tri-states to the HI-Z state. In emulation mode, DBENB is driven active if the test port is in FUNC, CNTL, or HALT. In test mode (ATPG.sub.-- DEVM=1), the test port's first SSR bit acts as the DBENB. This bit is canned as part of a data scan (SDAT) operation. In all modes, the DBEND signal is inactive when the test port is in SCTL, SDAT, or PAUS. When switching from PAUS to FUNC or CNTL, the MTAP forces the test port to enter HALT for at least one clock to assure DBENB is active prior to restarting execution.

Referring now to FIGS. 23A, 23B and 23C, which are timing diagrams, the operation of a clock switch will be described. The PAUS state is used for clock switching as controlled by UCLK.sub.-- SEL, TCLK.sub.-- SEL, MTAP codes, LOCK, and SHIFT.sub.-- DR. LM is turned off when the test port is in PAUS, SDAT, or SCTL. If ATPG.sub.-- DEVM=1, LM is also shut off when in HALT. Three types of clock switching are possible:

1. Switching from functional run on UCLK to scan (on TCLK) (FIG. 23A) for emulation control of the device.

2. Switching from functional run on TCLK to scan (on TCLK) (FIG. 23B) for test. Running on TCLK is used during test because TCLK can be guaranteed to come directly from device pins whereas UCLK may be controlled by cDSP design logic.

3. Switching from functional run on UCLK to functional run on TCLK (FIG. 23C) for test for the same reasons as above.

The sequences of changes shown in the inputs to the test port are consistent with the transitioning of the MTAP outputs. One exception is that SHIFT.sub.-- DR could transition simultaneously with the transition in and out of the SDAT or SCTL states.

The following rules are followed:

1. Clocking changes are based on the latched codes inside the test port.

2. Clocking and test port state does not change if the LOCK bit is active.

3. LS is always 180 degrees out of phase and non-overlapping with which every clock is being driven (SCLK or LM).

4. (T/U)CLK drives LM/LS, only if (T/U)CLK.sub.-- SEL is enabled and the test port is either in HALT if ATPG.sub.-- DEVM=0, or in FUNC, or in CNTL.

5. TCLK drives SCLK/LS, only if TCLK.sub.-- SEL, SHIFT.sub.-- DR, and SDAT.sub.-- SEL are enabled and LOCK is disabled. Note that SCTL.sub.-- SEL does not enable SCLK, because only SSR SRLs (i.e. no domain SRLs) are scanned.

6. Finally, if the device is in ATPG mode (ATPG.sub.-- DEVM=1), TCLK always drives LM/LS when the test port is in FUNC or CNTL. This assumes UCLK is driven low.

Scan operation is provided by the test port's internal Shared Shift Register (SSR) path. This path consists of one SRL for each distinct scan chain within that domain. The SSR bits are used to update the Scan Control (SCTL) path or as bypass bits for domain data (SDAT) scan paths. The SSR path is clocked with the scan clock since the test port must remain scannable even while the domain is running on functional clock. The clocks within the domain must switch from the functional clock to the scan clock for SDAT and SCTL (FIG. 22). Details of the scan paths will be described later with respect to FIG. 33.

A control scan operation (SCTL) accesses the Module Scan Enable (MSENB) bits through the SSR scan path. MSENBs select the domain scan paths that are enabled to form the data scan (SDAT) path, as illustrated in FIG. 24. Each test port contains at least one MSENB bit. An MSENB bit is needed for each scan path within the domain. All the test ports with in the megamodule are single slice: they have a single scan path and a single MSENB bit. Alternate single slice embodiments could replace the SCTL scan and the MSENB bits by the LOCK signal, or equivalent, for simplicity. The MSENB data is scanned into the SSR path while the MTAP asserts SHIFT.sub.-- DR qualified by an active SCTL.sub.-- SEL signal, the SSR is loaded into the MSENBs. When the MTAP asserts CAPTURE.sub.-- DR qualified by an active SCTL.sub.-- SEL, the MSENBs are loaded into the SSR. A bypassed control scan during functional run can be accomplished by locking the test port in FUNC or CNTL and then performing the scan. In that case, the SSR bits are the only ones added to the scan chain. Also, the test port does not respond to the UPDATE.sub.-- DR and CAPTURE.sub.-- DR signals.

A data scan operation (SDAT) provides access to the domain scan paths through the SSR can path. During a data scan, the domain data is scanned through the SSR paths. The SSR bits act as bypass bits for domain scan paths that are bypassed as controlled by the MSENB bits. The SSR path is selected by the SDAT state. Data is scanned into the SSR path while the MTAP asserts SHIFT.sub.-- DR and the test port is in the SDAT state. If the MSENB associated with the SSR is enabled, data is scanned from the SSR bit through that domain scan path. If the MSENB bit is disabled, data is scanned to the next SSR bit. Each domain contains at least one SDAT path. The SDAT path provides access to SRLs required for test and emulation. The UPDATE.sub.-- DR and CAPTURE.sub.-- DR signals are not used by the scan data operation.

Power-down

Three methods of power down are available on integrated circuit 1, as follows:

1. The IDLE instruction. IDLE continuously propagates NOPs into the pipeline until annulled by an interrupt. Clocking and RDY operation continues as normal. Thus, the CPU can still respond to a test port halt.

2. Completing the PDREQ/PDACK handshake to turn off LM/LS in the analysis and CPU domains. Microprocessor 1 design may use the EIP.sub.-- SET and IP.sub.-- SET signals from the magamodule to re-enable clocks based on changed interrupt conditions.

3. Modifying or shutting off UCLK to the CPU. However, if UCLK is shut off, the XDS may provide only limited visibility by running the Megamodule on TCLK.

FIG. 15 and Table 34 detail the power down interface between the test ports, power-down logic, and megamodule boundary. If a test port receives an inactive FCLK.sub.-- REQ and an active clocks off request (OFFREQ), it will shut off clocks (LM low and LS high) and acknowledge the request by asserting OFFACK. If OFFREQ is inactivated or FCLK.sub.-- REQ is activated, clocks are re-enabled and OFFACK is de-asserted. The power-down logic gives microprocessor 1 the ability to shut off the analysis and CPU domains. In the second case, the EiP.sub.-- SET or IP.sub.-- SET inputs can be used to signal interrupt conditions that microprocessor 1 may want to use to reassert clocks. Microprocessor 1 design is responsible to stall all other components as necessary while the CPU is powered down.

All SRLs in the megamodule are scannable and belong to one of the three test port domains or to the MTAP. A fourth off-megamodule domain, the SE analysis domain is supported by the MTAP but is not included as part of the megamodule. Each domain has a test port associated with it. Test ports are used in situations where clocks in a particular domain may have to be switched. The four domains are as follows:

Analysis Domain: The Analysis Data Register (ADR)

Megamodule Domain: This domain contains a single scan path including:

1. IP bits that fed the IFR and the associated detection latches

2. The SRLs that latch RDY at the boundary and control stalls

3. The Parallel Signature Analyzers (PSAs) for test

4. The circuitry that detects the EMU (0/1)IN signals from the MTAP

5. The latches for the power-down control bits

6. The circuitry that generates CPU done

7. The circuitry that generates the one-clock MSGSW pulse.

CPU Domain: The CPU domain consists of a single scan path containing all CPU SRLs that are stalled by RDY.

SE Analysis (SEA) Domain: A special emulation device (SE) will have a fourth domain for the SE logic, which resides outside the megamodule.

CPU Test Port Halt Control

This section describes a CPU domain halt for emulation. In emulation mode (EMU.sub.-- DEVM=1), the CPU stops execution on a cycle boundary. In test mode (ATPG.sub.-- DEVM=1), the CPU stops execution on a clock boundary. The Figures in this section depict a two-cycle off-megamodule RDY-based halt for illustrative purposes. Table 35 defines signals between the CPU Domain and the CPU Test Port which support halting.

FIG. 25 is a timing diagram that illustrates the different cases of a halt for emulation, as discussed earlier. Although the descriptions here indicate the test port goes to the halt state, for another embodiment the results would be the same if the test port went instead to the PAUS state. PAUS affects the ERDY.sub.-- DC, ERDY.sub.-- E1, and CPU.sub.-- DONE signals in the same way as HAL. Emulation halts occur in response to the following:

Case 1. The active transition of the EMU(0/1) inputs. Note that the EMU(0/1)IN signals generate one clock long pulses. Thus, these signals are detected in the megamodule domain (perhaps in the middle of a cycle) and their width is stretched to the end of the cycle as shown. Thus, because it must react during inactive RDY, logic in the megamodule domain will accomplish this.

Case 2. Program address breakpoint detected during the DP phase.

Case 3. A SWBP decoded during the DP phase. Note that once the pipeline is stopped in during the E1 phase of the instruction. The pipeline will not advance until the SWBP.sub.-- DEC field is cleared through scan.

Case 4. A special emulation event (SEE) based on a program address breakpoint match of the PCDP externally. This is seen internally during the DP phase. Note that other conditions could generate and SEE-based halt depending on the special emulation logic implemented. However, a program address breakpoint is shown because it has the strictest timing requirements.

Case 5. A floating-point resource conflict.

Case 6. The test port transitioning from CNTL to HALT during the DP phase. Like EMU(0/1)IN events, this can also occur during the middle of a DP cycle. Based on one of these events the following actions occur, as illustrated in FIG. 25:

The ERDY.sub.-- DC signal is set low externally during the entire DC phase.

The ERDY.sub.-- E1 signal is set low externally starting at the beginning of the E1 phase.

The EMURDY is set at the beginning of the associated E1. This feeds the CPU.sub.-- RDY signal, stalling the CPU.

Once all RDYs (the RDY1-RDY4 inputs) are active internally, CPU.sub.-- DONE is asserted.

After recognizing the CPU.sub.-- DONE, the MTAP moves the CPU test port from CNTL to HALT.

The XDS, through the MTAP, performs all necessary scans.

The MTAP moves the test port back to CNTL to allow the CPU to complete the pipeline phase. CPU.sub.-- DONE and ERDY.sub.-- E1 go inactive.

EMURDY goes inactive.

Notice that if the MTAP applies CNTL for only one clock, it can step the megamodule a single pipeline phase.

FIG. 26 is a schematic illustrating a circuit for forming an external memory ready signal ERDY. The external memory systems must track each other's RDYs to be aware of each other's stalls. Similarly, the ERDY.sub.-- DC and the ERDY.sub.-- E1 signals can be used by the external memory systems to be aware of an emulation halt. Because the CPU internally has two cycles warning (at least two clocks) before an pending emulation halt, it can generate the ERDY.sub.-- DC and ERDY.sub.-- E1 signals.

FIG. 27A is a timing diagram that illustrates the interactions of Halts, Interrupts and Reset. In FIG. 27, assume an execute packet n+5 has an enabled pending interrupt it its DC stage. If n+5 has a halt detected in its DP, the halt will take priority and the interrupt will not occur. In this way halts can block interrupts (including reset). Interrupt flags will continue to be set however. Interrupts and resets are also postponed during scan because EMURDY postpones interrupt response by stalling the pipeline.

The side-effects of a halt depend on the progress of the associated execute packet through the pipeline. Thus, if an execute packet following a multi-cycle NOP or IDLE encounters a halt condition in its DP, ERDY.sub.-- DC and EMURDY will not respond to the halt until that execute packet enters DC. Similarly, ERDY.sub.-- E1 and CPU.sub.-- DONE will not respond until the execute packet enters E1.

If the CPU's test port transitions from CNTL to HALT in response to a CPU test port-requested emulation halt during interrupt processing, CPU.sub.-- DONE does not go active until the interrupt service fetch packet (ISFP) reaches E1 (FIG. 27A).

FIG. 27B is a timing diagram that illustrates a test halt that is requested by a test port. When in test mode (ATPG.sub.-- DEVM=1), the CPU will halt immediately upon the test port transitioning from CNTL to HALT in response to a CPU Port-Requested test halt. This acts the same way regardless of the presence of interrupt processing.

Emulation Pipeline Control

Referring again to Table 28 which summarizes various emulation control and status bits, the following emulation control bits will now be described in more detail: EPDN, EPUP, ECTL, and EUPL. The Emulation Pipe-Down (EPDN) bit will perform annulling for an emulation stop which results in the following:

Further Fetching Disabled: PAS is disabled. Thus, no more program fetches are initiated. However, previously requested fetches are allowed to complete and be loaded into a program data input buffer PDATA.sub.-- I, which is located in program fetch circuitry 10a.

Execute Packets not yet in E1 annulled: Before an execute packet enters E1 it is annulled. Any execute packets entering E2 and afterward can be cycled to completion by the XDS. Note: The ANNUL and ANNUL.sub.-- ASU bits must both be scanned to 0 for annulling to work.

Interrupts Disabled: Taking of all interrupts including RESET is disabled. Pending bits continue to get set.

The Emulation Pipe-Up (EPUP) bit will perform annulling for an emulation restart. Interrupts are disabled so that taking of all interrupts including RESET is disabled; however, pending interrupt bits continue to get set.

The Emulation Control (ECTL) bit allows the emulator control of the CPU by performing the following steps:

1) Suspends Analysis Events: ECTL suspends any future recognition of analysis events.

2) Interrupts Disabled: ECTL disables all interrupts including RESET.IP bits still get set but interrupts are not taken. Due to suspension of analysis events AIP can only be set by an SWI.

3) Disables Fetches: ECTL (like EPDN) also disables program fetches by forcing PRS inactive. Thus, when the CPU is in halt, the XDS can scan instructions into PDATA.sub.-- I. If PDATA.sub.-- I contains multiple execute packets (a partially serial fetch packet), the CPU should process these as usual and wrap around to the first execute packet when complete. However, execute packets cannot cross the fetch packet boundary and wrap around. The XDS can trigger another emulation event by placing a SWBP as one of the 8 instructions.

The Emulation Program Upload Support (EUPL) bit supports program upload. EUPL re-enables program fetches, and sets the PFC to increment to the next fetch packet (i.e. no branching, PFC+=8). However, all stages DP and after are annulled. Fetches are re-enabled even if ECTL is set. However, if ECTL is set interrupts will still not be taken and analysis events will be suspended. This is used for program upload. On every cycle, PDATA.sub.-- I can be scanned to extract the fetch packet contents. Note: The ANNUL and ANNUL.sub.-- ASU bits must both be scanned to 0 for annulling to work.

FIG. 28 is a timing diagram that illustrates a sequence of halts in a procedure for stopping the pipeline and saving as a state the contents of various registers within the pipeline, referred to as a "pipe-down" procedure. Instructions in execute packets that have completed E1 (a-d) are allowed to proceed to completion. The following describes the sequence of operations XDS 51 performs in controlling a pipeline stop. Note that in this method of saving state, any user modifications of data sources of loads that have completed will not be reflected upon restart. After initiating a halt, as described with reference to FIG. 25, the following steps are performed:

1. Cycle 2 at time 350: Scan out and save all CPU domain state. Scan in with EPDN=1, EPUP=0, ECTL=0, EUPL=0, ANNUL=0, and ANNUL.sub.-- ASU=0. This stops new fetching, disables interrupts, and annuls execute packets that have not completed E1.

2. Apply CNTL 351 for one clock cycle.

3. Cycle 3 at time 352: Scan out and save DDATA.sub.-- I. Scan in state without any modifications.

4. Apply CNTL 353 for one clock cycle.

5. Cycle 4 at time 354: Scan out and save DDATA.sub.-- I. Scan in state without any modifications.

6. Apply CNTL 355 for one clock cycle.

7. Cycle 5 at time 356: Scan out and save DDATA.sub.-- I. Scan in state filling DP phase with NOPs and EPDN=1, EPUP=0, ECTL=l, EUPL=0, ANNUL=0, and ANNUL.sub.-- ASU=0. Setting ECTL sets SUSPEND to disable on-chip and off-chip analysis.

8. Apply CNTL 357 for one clock cycle.

9. Cycle 6 at time 358: Not scanning. Allow NOPs to propagate into pipeline.

10. Apply CNTL 359 for one clock cycle.

11. Cycle 7 at time 360: Scan out with no state to be saved. Scan in with EPDN=0, EPUP=0, ECTL=1, EUPL=0, ANNUL=1, and ANNUL.sub.-- ASU=1. Clearing EPDN allows the contents of the DP phase (IR) to feed the rest of the pipeline. Fetching and interrupts remain disabled via ECTL.

FIG. 29 is a timing diagram that shows how the entire pipeline state is restored in a "pipe-up" procedure by restoring each of the various states that were saved during the pipe-down procedure described above. This Figure assumes that the pipeline had been flushed of all execute packets inserted by the XDS for control and contains NOPs in all phases of the pipeline. This state is achieved manually by the XDS by filling the instruction register with NOPs and allowing other instructions in the pipeline to cycle to completion. After the re-scan, annulled phases are allowed to complete from saved state. All other state follows from the previous cycle. On cycle 2, because data memory strobes are disabled, data memory operations are advantageously not re-executed. Incoming load data latched at the beginning of E5 (end of E4) must be scanned in by the XDS. Thus, advantageously, the entire sequence of pipe-down, emulation, and pipe-up causes no extraneous memory or I/O cycles in data processing system of FIG. 1. The following describes the sequence of operations the XDS performs in controlling a pipeline restart:

1. Cycle 0 at time 370: Scan back in the address of the fetch packet fetched for fetch packet h. This information is available in the scan of cycle 2 during the emulation stop. Set EPDN=0, EPUP=0, ECTL=0, EUPL=1, ANNUL=1, and ANNUL.sub.-- ASU=1. This re-enables fetching while disabling interrupts.

2. Apply CNTL 371 for one clock cycle.

3. Cycle 1 at time 372: Scan back in the address of the fetch packet fetched for fetch packet i. This information is available in the scan of cycle 2 during the emulation stop.

4. Apply CNTL 373 for one clock cycle.

5. Cycle 2 at time 374: Restore the entire CPU domain state from cycle with the DBS disabled (set to 0). Set EPDN=0, EPUP=1, ECTL=0, EUPL=1, ANNUL=1, and ANNUL.sub.-- ASU=1. This re-enables fetches while disabling interrupts.

6. Apply CNTL 375 for one clock cycle.

7. Cycle 3 at time 376: Scan out and scan back in DDATA.sub.-- I from cycle 3 of pipeline stop.

8. Apply CNTL 377 for one clock cycle.

9. Cycle 4 at time 387: Scan out and scan back in DDATA.sub.-- I from cycle 4 of pipeline stop.

10. Apply CNTL 398 for one clock cycle.

11. Cycle 5 at time 380: Scan out and scan back in with DDATA.sub.-- I from cycle 5 of pipeline stop. Set EPDN=0, EPUP=0, ECTL=0, and EUPL=0, ANNUL=1, ANNUL.sub.-- ASU=1. This re-enables interrupts, analysis, and fetching.

A Software Breakpoint (SWBP) is set by replacing and saving the creg/z field in a target instruction with a code of "0001" to indicate a SWBP, as shown in Table 10. This field is decoded during the DC phase of the pipeline. Decoding a "0001" triggers a breakpoint operation and invokes XDS 51 to perform a debug or emulation function by performing pipe-down and pipe-up procedures, as described above. At the completion of the debug or emulation operation, the XDS replaces the SWBP instruction in the pipeline during E1 by returning the associated creg field of the field to its original value. In addition, a SWBP.sub.-- DEC SRL will be set by an SWBP being decoded. The XDS uses SWBP.sub.-- DEC (FIG. 20) as status. Upon replacing SWBP in the pipeline with the appropriate creg field, the XDS should clear SWBP.sub.-- DEC so another SWBP can be decoded. According to an aspect of the present invention, a debug function is performed in response to a software breakpoint instruction and normal pipeline operation is resumed without prefetching the instruction which was converted to a SWPB. Advantageously, no extraneous memory cycles are performed in the data processing system of FIG. 1.

Referring again to FIG. 29, assume the CPU is halted via an emulation stop sequence and that the XDS has extracted all necessary state to be displayed by the debugger. The XDS can single step microprocessor 1 by performing the first 6 steps of the restart sequence and then redoing the entire emulation stop sequence. Note that while single stepping for analysis events to occur the analysis test port must be locked in CNTL or FUNC. Similarly, for interrupts to be detected the megamodule test port must be locked in CNTL or FUNC. In other words for interrupts to be detected that portion of the megamodule must continue running. Also, the CPU will not single step into interrupt service fetch packets. The CPU must be allowed to run for a minimum number of cycles before an interrupt will be serviced.

Events can be suspended for various reason in response to the state of the various test ports 310, 320, and 330. Three conditions suspend analysis events during a halt:

1. During a halt, the PAUS/SDAT/SCTL states of the CPU test port and the CPU.sub.-- DONE signal force the SUSPEND signal active, disabling future analysis events. When the CPU test port returns to CNTL or FUNC the SUSPEND signal returns to inactive state.

2. In addition, analysis events are prevented from causing halts when the CPU test port is in the FUNC state.

3. The XDS can set the ECTL bit through scan to disable analysis events during emulation control.

Halt is unaffected by an IDLE or a multi-cycle NOP instruction in the pipeline. Upon a pipeline restart the previous state of the IDLE or multi-cycle NOP in the pipeline is restored.

In normal debug and emulation operations, neither the analysis domain nor megamodule domains are stalled; therefore a hand shake signal (such as through CPU.sub.-- DONE) is not provided for the MTAP to halt the test port 320 or 330. The MTAP simply performs the clock switch, moving through states on the appropriate clock boundaries. If for some reason these domains must be halted, they must be halted only after the CPU domain has been halted.

FIGS. 30A, 30B, 30C and 30D illustrate various registers that are used in the course of emulation and debugging, as described above. These include: the Analysis Control Register (ACR) 390, Analysis Data Register (ADR) 391, Analysis Interrupt Return Pointer (ARP) 392, and Data Streaming Register (STREAM) 393. These various registers have been discussed in the previous sections, and will now be described in more detail. In these Figures, the term "MVC readable" and "MVC writable" refers to whether particular bits in a CPU control register can be read or written by a MVC instruction. All reserved or write only bits are read as "0."

FIG. 30A describes ACR 390, which contains control and status bits for emulation and analysis. The status-bits do not cause a halt or analysis interrupt but simply reflect that the event occurred while events were not suspended. As soon as they occur, the actual events themselves (if enabled) trigger a halt or set AIP to show a pending interrupt. Also, reset only sets the ACR to its reset values if the CPU test port is in the FUNC state. The exception is RSTOCC which always is reset regardless of the CPU test port state. Table 36 describes the function of each ACR bit field.

FIG. 30B describes Analysis Data Register (ADR) 391. The top 30 bits of the ADR are used to compare with the program address in pipeline phase DC to generate a program address breakpoint. This register is in the analysis domain and will be used for message passing. A read of this register during an analysis scan will return O and set the MSGERR bit in the ACR. A write during analysis scan will have no effect and set the MSGERR bit. Any other read or write will clear the MSGERR bit.

FIG. 30C describes Analysis Interrupt Return Pointer (ARP) 392, which was described in detail earlier with reference to analysis interrupts.

Memory Access Support

The architecture of microprocessor 1 assumes that all memory locations whether in program memory or data memory are writable (if RAM or FLASH is used). The design of Microprocessor 1 obeys this convention to allow XDS-based data and program uploads and downloads as well as monitor replacement of an SWI. In addition, distinct program and data memories may not occupy the same address space. Caches are be writable for SWI and SWBP replacement.

For data downloads and uploads as well as program downloads, the XDS will do the following after an emulation stop sequence:

1. Set ECTL. Scan a fetch packet into the PDATA.sub.-- I 10a (FIG. 1). For data accesses these consists of 7 serial loads or stores. For program downloads, these consist of 3 MVC/STP pairs followed by a SWBP. The appropriate data and addresses will be scanned into the register files.

2. Let the CPU run free. Eventually, the SWBP will trigger an halt and a subsequent re-scan.

In this manner, the XDS can download/upload 224-bits/scan (7.times.32-bit stores) of data or download 96-bits/scan (3.times.32-bit stores) of program. According to an aspect of the present invention, these sequences of debug code instruction are loaded into the multi-word instruction register and executed to perform a debug operation on the processor 1. Advantageously, no extraneous memory operations occur with the data processing system of FIG. 1.

For program uploads the emulator performs the following emulation pipeline stop sequence:

1. Set ECTL and EUPL. Scan a value into the PFC and the PADDR.

2. Apply HALT/CNTL three times to move through the PS, PW, and PR phases.

3. Scan the PDATA.sub.-- I for the uploaded program.

4. Afterwards, the XDS may apply a single CNTL/HALT each cycle. Thus, the emulator may perform 256-bits/scan.

While the previous procedure is effective for uploading or downloading small amounts of data without disturbing the data processing system environment, there is a significant overhead time due to the amount of scanning that is required. A procedure referred to as "data streaming" provides the following data transfers at the full scan rate: program memory download (from the XDS), data memory download (from the XDS), and data memory upload (to the XDS).

FIG. 30D illustrates Data Streaming Register 393 which is used to support data streaming. This CPU control register connects the CPU register file to the data streaming bus from the MTAP (STRM.sub.-- I/O). This register cannot be used for storage as what is written will not be read back. The data streaming process will now be described in more detail with reference to Tables 38-42.

To prevent fetches, the XDS sets the ECTL bit to disable fetching. The XDS loads the appropriate fetch packet into the PDATA.sub.-- I. This fetch packet performs the necessary data movement between the data streaming busses (STRM.sub.-- ST and STRM.sub.-- LD) as accessed through the STREAM CPU control register. Upon the scanning of a new data word (every 32nd bit) the CPU is stepped one cycle. If CPU.sub.-- DONE is not received before the cycle next step, the MTAP latches this error condition. The XDS can then detect a streaming error occurred upon completion of the streaming process. Details of managing data movement and error detection within the MTAP will be discussed later. Advantageously, since instruction fetching is inhibited, the same fetch packet will be executed for the duration of the streaming process.

Table 38 and Table 39 show the fetch packet to control data streaming data downloads on data ports 2 and 1, respectively. Examples for both data ports are shown because, microprocessor 10 can make the two ports write to physically disconnected memory spaces. In both cases, an MVC instruction with the STREAM register as an operand is used to move the streaming data to the B and A register files. For data port 2, the XDS must scan the initial data value for the first STW into the register file (into B1). For data port 1 downloads, STREAM must first be moved to a B register and then to an A register because MVC cannot directly load an A register. In this case, the XDS must both scan in the initial two values (A1 and B1) into the register file. Note that microprocessor 10 advantageously can execute various combinations of instructions in parallel, such as MVC and STW in Table 38 and Table 39 to minimize overhead time.

Table 40 and Table 41 show fetch packets to control data streaming uploads on data ports 2 and 1, respectively. Examples for both data ports are shown because, microprocessor 1 can make the two ports write to physically disconnected memory spaces. In both cases an MVC instruction with the STREAM register as an operand is used to move the streaming data from the B and A register files. Also, the XDS must scan the final data value from the last LDW from the register file (from B1) after it allows manually clocks the last loads to complete after filling PDATA.sub.-- I with NOPs.

Table 42 shows the fetch packet to control data streaming program downloads. Like data downloads, the XDS must load the first value to be stored into the register file (into B0). However, unlike data streaming for program accesses, no instruction is available to move STREAM directly to the register to be stored, which is a program-data output register (PDATA.sub.-- O) located in program fetch circuitry 10a. Thus, the STRM.sub.-- SEL signal must be used to update PDATA.sub.-- O to the value of STREAM (via the STRM I/O bus) every cycle.

Caching Support

There are independent controls for data and program caches in the control status register (CSR) that allow:

1. Caches Freeze: Cached values are read. Reads do not update the cache.

2. Caches Flush: All cached data is invalidated. Updates do not occur.

3. Cache Bypass: Values are read from memory. Reads do not update the cache.

The following rules apply to XDS control of memory accesses with respect to caches.

1. On data or program COFF downloads, caches are flushed and then resumed to their previous control states.

2. For data or program COFF uploads, caches is frozen and returned to their previous control states.

3. For data or program memory reads in a CPU memory view, caches are frozen and the resumed to their previous control states.

4. For data or program memory reads in a physical memory view, caches are bypassed and returned to their previous control states.

5. For data writes, no change is made to the cache control bits.

6. For program writes, the cache is flushed and then returned to its previous control states.

Development Tools Notes

In the XDS and simulator, the programmer's model for debugging is as follows:

1. All information will be displayed on a cycle-by-cycle basis (as opposed to a clock by clock basis). Thus, memory stalls are not visible to the user.

2. The debugger will highlight the next execute packet to be executed in the disassembly.

3. All registers results written by the end of the last cycle will be displayed:

A) The results of the execute packet in E1 (single-cycle integer instructions, address modification) in the previous cycle.

B) Similarly, the results of the instruction in E2 from two cycles ago (integer multiplies).

C) The results of the instruction in E4 (floating-point instructions) from four cycles ago.

D) The results of the instruction in E5 (loads) from five cycles ago.

In monitor-based debuggers, the programmer must employ single assignment. Also, all instructions will appear to execute in order.

For example, in FIG. 31, if execute packet (f) is highlighted, all CPU state written at or before the double line 400 will be displayed. Thus, for example, if execute packet (e) executed a multiply, its results will not be displayed. Rather, the multiply's destination register will contain its prior contents from cycle 11.

Unlike CPU state, memory state will be allowed to advance to completion of stores. In FIG. 31, the stores in execute packets c-e will complete. Thus, the memory display (updated after all scans are complete) appears to be 3 cycles ahead of the CPU. Although this may appear inconsistent with loads having 4 delay slots, the programmer must view that the value of a load value is the one presented in the display immediately before stepping through a load instruction. As all accesses occur in order at the CPU boundary, the memory display will be self-consistent.

Real-Time Debug: Monitor Mode: Real-time debug provides user visibility--while microprocessor 1 is running user code. Variables can be displayed and modified both from the assembly and the high-level language (HLL) viewpoints. This real-time debug by itself is independent of emulation logic. Normally, the debug monitor is either the background task or a very low-level interrupt service routine running in the presence of real-time tasks driven by higher-priority interrupts. The monitor communicates to a host on which the debugger is resident. The debugger at the lowest level can request memory and register reads and modifies. Communication can occur through peripherals such as serial ports, through shared memory, or through dedicated message-passing hardware (such as the ADR). Considerations for developing monitor code are:

Codesize: The monitor codesize must be minimized.

Performance: The monitor code must run fast enough to provide adequate user accessibility.

Single Stepping: The debugger performs two types of stepping:

1. Stepping statements in high-level language.

2. Stepping execute packets on a cycle-by-cycle basis.

Because pipeline visibility is critical to programming the CPU, the debugger does not need to support instruction stepping (stepping an instruction all the way from E1 until E5). Instead, the step mechanism will only allow all state to proceed a single pipeline phase.

After an emulation stop, execute packet stepping will be performed by a stop pipeline scan sequence (FIG. 28) followed by a partial restart (FIG. 29). Specifically, the restart will be allowed to proceed one cycle. The emulation stop sequence will then reoccur in its entirety.

In general, execute packets or groups of execute packets will not correspond on a one-to-one basis with high-level language (HLL) statements. The following approaches are available to provide HLL statement-by-statement visibility:

1) The HLL compiler can generate code that generates side-effects that complete in order. However, this drastically inhibits achieving high-performance code. Also, this restriction masks problems which only occur when out-of-order execution is allowed.

2) The XDS may perform multiple stops and scans so that the results of a particular HLL statement may be isolated. New methods of embedding debug information in the object must be developed to facilitate this methodology. Additionally, in a mixed-mode display (both assembly and HLL) a method must be developed to show what assembly instructions have completed what phase of the pipeline as seen in the display. A disadvantage of this method is that the user may not be able to request new memory locations or symbolic values for display. Such information may have already been lost because the part may have run beyond when their values are valid. Both solutions are practicable given the current emulation description in this chapter.

Software replacement of a SWI instruction in program memory with the appropriate instruction is done by the user's monitor program. If interrupts are enabled and SWI is the highest priority pending interrupt it will be taken when the SWI is in DC. If there are higher priority interrupts, SWI be re-fetched upon return from interrupt and thus re-encountered. SWI-recognition will be postponed by being in the delay slots of a taken branch or if AINT is not enabled.

Megamodule Test Access Port (MTAP)

The MTPA will now be described in detail. Aspects of the present invention relate to improvements on the structure of the IEEE 1149.1-1990 Standard Test Access Port and Boundary Scan Architecture, which is incorporated herein by reference. Terms and concepts relating to IEEE 1149.1 which are used herein are explained fully in this IEEE standard. Likewise, aspects of the present invention also relate to improvements on the structure of Texas Instruments' Modular Port Scan Design (MPSD) as disclosed in U.S. Pat. No. 4,860,290 and incorporated herein by reference. In particular, the operation of Shift register latches (SRLs), which are distributed throughout microprocessor 1 like a string of beads on a serial scan path respective to each module provide access to all "important" registers, is described in incorporated reference U.S. Pat. No. 4,860,290.

Referring again to FIG. 15, Megamodule Test Access Port (MTAP) 305 supports a subset of the IEEE 1149.1 Standard Test Access Port features. Since Megamodule 300's MTAP does not drive the pins of microprocessor 1 directly there is no requirement to support boundary scan. MTAP 305 does provide an 1149.1 compliant JTAG State Machine and serial scan communications between a remote host 51's scan controller and the Megamodule's Domain Test Ports (DTPs) 310, 320 and 330. In addition to the JTAG interface, MTAP 305 provides test support, automated execution control of the DTPs, a data stream control mechanism that provides a significant performance advantage, multiprocessor emulation support, and performance analysis; as discussed in the preceding paragraphs. MTAP 305's test feature set supports the application of production Automatic Test Pattern Generation (ATPG) patterns. Functional test patterns can be loaded, executed and debugged using emulation capabilities.

FIG. 32 is a block diagram illustrating signal connections from megamodule 300 to pins 410-416 on microprocessor 1. Also referring to FIG. 15, as discussed earlier and summarized here, Megamodule 300's architecture is split between MTAP 305 and four domains; CPU Core 10, CPU Analysis 321, Megamodule 331 (contains all megamodule features outside the CPU core) and Special Emulation (SE) 400. SE module 400 is external to Megamodule 300. Each domain provides execution control and access to the domain scan paths through a Domain Test Port (DTP). The CPU's DTP 310 provides stop mode and real time emulation features that control program execution (start, stop, software breakpoint) and visibility to the programmer's model (registers and memory). CPU Analysis Domain 321 provides core analysis features that in the case of the microprocessor 1 includes a hardware breakpoint and real time emulation scan communication support. SE Analysis Domain 400 provides advance emulation analysis features. These features include magnitude comparators for hardware breakpoints, program bus address breakpoints, data bus address and data breakpoints, event counters, program discontinuity trace, and analysis state sequencer. Megamodule Domain 331 only provides execution control and scan access of the features within the Megamodule that are outside of the CPU, such as the test PSA registers in test circuitry 52.

Real time emulation support provides execution control and visibility of the programmers model while the processor continues to service interrupts and multiplex tasks. The Real Time support is also accessible through an applications-based debugger and includes the use of the embedded analysis capability without connecting to a remote test/emulation controller.

Referring still to FIG. 32, Megamodule Test Access Port (MTAP) 305 is IEEE 1149.1 compliant in that it supports the standard JTAG interface and JTAG State Machine feature set. MTAP 305 is not 1149.1 compliant in that it does not support boundary scan or any public STAG commands other than BYPASS. JTAG based emulation and test requirements for microprocessor 1 are provided by utilizing provisions within IEEE 1149.1 that allow for application specific JTAG commands and data path extensions. References herein to the required JTAG capability are referred to as "public capability," while extensions are referred to as "private capability."

The support of emulation and test features in the Megamodule (MTAP) 300 requires connection to microprocessor 1's standard JTAG five pin interface (TMS 410, TRST 411, TDI 412, TDO 413, TCLK 414) complemented by two additional bi-directional emulation interrupt pins (EMU1 415, EMUO 416). There are multiple JTAG/MTAP configurations that will be discussed in the following paragraphs. The EMUO and EMU1 pins can be configured to facilitate multiprocessor stop events and performance analysis features.

The EMU0/1 pins are functional pins and as such conform to all Megamodule boundary cell rules.

The following constructs have been added within the JTAG framework to support communication and control between the JTAG pins of the device, MTAP 305, and multiple Domain Test Ports.

Data Path Extensions--extended private JTAG IR commands provide MTAP data scan selection of Domain Status, EMU1 and EMUO configuration. Emulation Control. Domain scan path select, and Domain lock information.

Command Generation--extended private JTAG IR commands provide test and emulation commands that are initiated through the JTAG IDLE state.

Instruction Register Capture--JTAG instruction register capture of miscellaneous emulation state, domain status, and test information has been added to facilitate emulation software operation and MTAP testability.

MTAP 305 JTAG signals are as follows:

TMS: Test Mode Select. This signal controls the transition of the JTAG state diagram. The various states traversed cause the instruction and data paths to be scanned. and JTAG instructions to be executed.

TCK: Test Clock. This signal provides a clock for JTAG logic and state machines. The JTAG interface is clocked at a frequency supplied external to the megamodule and allows the megamodule to be compatible with other JTAG devices, controllers and test equipment designed for different clock rates. In this specification this clock input is referred to as TCLK. The normal system clock input is referred to as UCLK (functional clock).

TDI: Test Data Input. This signal provides input data for all JTAG instruction and data scans of the megamodule.

TDO: Test Data Output. This signal provides output data from all JTAG instruction and data scans of the megamodule.

TRST: Test Reset. This signal resets the JTAG module and is provided to assure the test access port (TAP) is rapidly initialized upon power up.

Relationships between these signals are defined in the IEEE 1149.1 specification, which has been incorporated herein.

MTAP 305 EMU signals are as follows:

EMUI [1:0]: EMU Input. The analysis domain allows the detection of a logic zero on EMUIn to be used as an event. Since EMUI can be asynchronous and any pulse width. EMUI must be presented to a pulse catcher and a synchronizer to assure a single clock synchronous event is sent to the CPU domain (EMUI signals) for event processing. The pulse catcher is self clearing by the synchronized event pulse.

EMUO[1:0]: EMU Output. This signal is utilized to output events selected through the EMUC bits of the ECR (in emulation and test modes) or the CPU.sub.-- EMUC bits of the ACR (in strap mode). The EMUO[1:0] signals from MTAP 305 must be active for at least 5 Nsec (Test Bus Controller requirement).

EMUOEN[1:0]: EMU Output Enable. This signal provides an enable for the EMUO signal's external tri-state buffer utilized to drive the EMU pin pad. The operation mode of this signal. shared or dedicated. is selected through the EMUC bits of the ECR (in emulation and test modes) or the CPU.sub.-- EMUC bits of the ACR (in strap mode). In shared mode this signal is only enabled when the event is active, leaving the output buffer in the tri-state mode when an event is not enabled. When disabled in shared mode the EMU pin pad state can be driven into the EMUI driver from an external event. In dedicated mode this signal is always active, which forces the EMU pin pad to the EMUO signal level. Examples of using this signal external to the Megamodule will be discussed later.

JTAG Opcode Utilization: JTAG opcode space 0x00-0x1F is reserved for JTAG specific use. The scan paths and commands associated with these opcodes are defined in the IEEE 1149.1 specification. This opcode space is currently only partially defined with all undefined opcodes in this group being reserved for future use. References herein refers to any opcode in this group as a public opcode.

According to an aspect of the present invention, a portion of the undefined JTAG opcode space from 0x20-0x2F is reserved for emulation specific use. These opcodes are referred to as private opcodes. All undefined opcodes within this group are reserved for future emulation use.

Opcodes within the 0x20-0x2F range are defined to select data scan paths and as emulation and test commands. An opcode is utilized as a data scan path select when the JTAG state machine is in the JTAG SHIFT.sub.-- DR state (shift the selected data register). The same opcode is utilized as a command when the JTAG state machine is in the JTAG IDLE state. The JTAG opcodes are also utilized to directly apply control codes to a DTP in a test mode that maps the JTAG states to DTP control states. Each of these three uses of the opcodes are decoupled from the others. This document discusses them independently.

FIG. 33 is a block diagram of the MTAP scan path organization. Emulation/test scan paths are provided in addition to the IR and Bypass paths mandated by the JTAG specification. Note that MTAP 305 does not support a boundary scan path (which makes MTAP 305 non-JTAG compliant). The emulation/test paths are scanned by traversing the JTAG state machine to the SHIFT-DA state, preceded by an SHIFT-IR scan to select the scan path. More than one opcode may address the same public or private scan paths.

The public and private scan paths are briefly described in Table 43.

JTAG Data Path Control

Still referring to FIG. 33, scan data is shifted through a selected data shift register during JTAG SHIFT.sub.-- DR state. The data shift register shifts from MSB to LSB with the LSB of the shift register being output on TDO during the first SHIFT.sub.-- DR state. Data shifted into the data shift register from TDI may be transferred into a parallel holding register at the end of the shift sequence if a static value of the transferred data bit is required. The parallel holding register is called a shadow register or a shadow bit for a holding register length of one. MTAP 305 implementation uses a mix of shadowed bits and shift register bits to interface to processor emulation functions. When shift register bits are used directly, the shifting of the data shift register has been made a don't care in the bit's end use.

The data register group of JTAG states is entered when the SELECT.sub.-- DR state moves to CAPTURE.sub.-- DR, and ends when the execution of the UPDATE.sub.-- DR state is complete. This group of states is assigned to a specific data path that is selected by the opcode held in the JTAG instruction register while scan data register state group is traversed. This state group contains three states with path specific importance. These states, CAPTURE.sub.-- DR SHIFT.sub.-- DR and UPDATE.sub.-- DR, may be combined with the opcode contained in the JTAG instruction register to give path specific directives to manage scan data.

The CAPTURE.sub.-- DR state may selectively load information into the data shift register at the beginning of the shift sequence. This information is then visible at the chip's TDO pin as a path specific data shift register is advanced by the SHIFT.sub.-- DR state. The UPDATE.sub.-- DR state is used to transfer data shifted into the data shift register to the appropriate parallel data register denoted by the JTAG IR opcode.

Most MTAP private data paths do not require the CAPTURE.sub.-- DR function to be implemented for every data path bit position. In some cases an entire data path does not require the CAPTURE.sub.-- DR function to be implemented. This is especially true of DTP data scan paths. Some MTAP private data paths share a data shift register. This implementation results in a single data shift register whose data is selectively transferred to different shadow registers or bits by a qualified UPDATE.sub.-- DR signal. This approach is utilized within the DTPs to utilize a single physical shift register that supports both the DTP.sub.-- SCTL and DTP.sub.-- SDAT scan paths. This shared approach also allows the CAPTURE.sub.-- DR state function for the paths to be merged. All physical implementation options discussed above DO NOT alter the data path configuration or operation as it is viewed by software controlling the paths.

When the JTAG TAP transitions through the SELECT.sub.-- DR state the data scan path selected and registers captured and updated are dependent on the command in the JTAG IR. IR commands that direct data to the DTPs, other than the DTP data scan commands, will require an update and in some cases a capture signal (based on JTAG TAP CAPTURE.sub.-- DR and UPDATE.sub.-- DR states). The DTP.sub.-- SDAT and DTP SCTL scan paths share a common data shift register through the DTPs. In the case of the DTP SCTL path the path select signal combined with the CAPTURE.sub.-- DR state causes the common shift register to be loaded from the MSENB shadow bits. After a scan is completed (SHIFT-DR), the UPDATE.sub.-- DR state causes the new value in the common shift register to be transferred to the unique MSENB shadow registers. In the case of the DTP.sub.-- SDAT path the DTP's common shift register is utilized as a bypass bit. All other scan paths are directly scanned (not scanned through the DTPs). Table 44 details the requirements for each scan path.

The select signals, SCTL.sub.-- SEL, SECR.sub.-- SEL, SCTR.sub.-- SEL, STRM.sub.-- SEL, and SECT.sub.-- SEL are mutually exclusive and are driven inactive by the UPDATE.sub.-- IR TAP state.

Still referring to FIG. 33, MTAP 305's JTAG instruction register (IR) 450 is eight bits in length with all bits used in instruction decode. The decoded instructions are utilized to select scan paths or execute commands. Access to the IR is provided by a group of JTAG states. These states are entered when the SELECT.sub.-- IR state transitions to CAPTURE.sub.-- IR, and exited when the UPDATE.sub.-- IR state is exited. When the JTAG TAP state machine transitions through the SELECT.sub.-- IR state, the IR register is connected to the JTAG data scan path. The CAPTURE.sub.-- IR state loads information into the instruction shift register at the beginning of the shift sequence. The data captured is dependent on the state of the Status Select (STSL) bits in the emulation control register (ECR) (see FIG. 36). This information is then visible at the chip's TDO pin as the instruction shift register is advanced through the SHIFT.sub.-- IR state. When the SHIFT.sub.-- IR state is entered, the DPC (Data Path Controller) enables the scan path to shift on every TCLK. The instruction shift register is shifted from MSB to LSB with the LSB capture information register being output on TDO on the first SHIFT.sub.-- IR state. This mechanism is used to load and export emulation and test status. The UPDATE.sub.-- IR state is used to transfer data shifted into the chip (contents of the instruction shift register) to the shadow bits of the IR.

FIG. 34A illustrates the JTAG IR with Strap Status selected. Strap status is the default status state which is selected by TRST- or the TLR JTAG state. The two LSBs are loaded with a fixed pattern (0,1) per section 6.1.1 of the IEEE 1149.1 Specification. This fixed pattern is derived from MTAP state flags that are forced to the correct pattern during Strap mode.

FIG. 34B illustrates the JTAG IR with Stop Status selected. The status bits illustrated here are generally utilized during stop mode emulation. All status bits are sourced from the MTAP except ABP.sub.-- DET and SWBP.sub.-- DEC (CPU domain). MTAP 305 status bits are defined in Table 45. The stop mode emulation status bits are defined in Table 68.

FIG. 34C illustrates the JTAG IR with Real Time Status selected. The status bits illustrated here are generally utilized during real time mode emulation. All status bits are sourced from MTAP 305 except for MSGFLG and MSGSW (CPU domain). MTAP 305 status bits are defined in Table 44. The real time mode CPU domain emulation status bits are defined in Table 69.

FIG. 34D illustrates the JTAG IR with Emulation Error Status selected. The status bits illustrated here are generally utilized during emulation error processing for both stop and real time modes. All status bits are sourced from MTAP 305 except MINT.sub.-- EN and AINT.sub.-- EN (CPU domain). MTAP 305 status bits are defined in Table 44. The MINT.sub.-- EN and AINT.sub.-- EN CPU domain emulation error status bits are defined in Table 69.

Table 45 defines status bits that are generated within MTAP 305 module. See Tables 68-71 for the definition of status bits sourced outside of MTAP 305.

Table 45 defines a set of JTAG Instructions that are supported by MTAP 305. JTAG instructions within MTAP 305 may be broken into the standard JTAG instructions required by the IEEE 1149.1 specification (public instructions) and private instructions which have been added for emulation and test. Private JTAG opcodes have three basic functions:

1) Selection of the scan path for the SHIFT.sub.-- DR JTAG state and the control logic used in conjunction with the CAPTURE.sub.-- DR and UPDATE.sub.-- DR JTAG states.

2) Determine the disposition of private commands generated when private opcodes are in the JTAG IR and the JTAG state machine transitions into the JTAG IDLE state.

3) Support ATPG testing with opcodes by a direct mapping of JTAG states to MPSD codes when the device mode is Test.

Private commands are actions which are initiated in the JTAG environment by entering the JTAG IDLE state from either UPDATE.sub.-- IR or UPDATE.sub.-- DR. The command initiation is synchronized to the functional clock (UCLK) and results in one functional clock wide commands to functional logic in MTAP 305 and DTPs. The initiation of a private command inhibits further updates of the JTAG IR until the command has been issued. Completion of a private command can be determined by capturing the IRBUSY (instruction register busy flag) in the JTAG CAPTURE.sub.-- IR state.

The JTAG instructions described in Table 46 are supported by MTAP 305. They are broken into Public and Private groups. Required opcodes (public) are noted as such in the discussion. Any opcode not denoted as required is a private opcode. The description of the private group includes the data register selected by the opcode in addition to a short description of the command that may be initiated with this opcode. Test mode use of the opcodes is discussed later; particularly with respect to Table 63 and Table 65.

Referring again to FIG. 33, bypass path 451 is a one bit data path that is selected when no other data path is specifically selected through the JTAG instruction register. The bypass bit always captures a logic one in the CAPTURE.sub.-- DR state.

MTAP Support of JTAG to MPSD DTPs

FIG. 35 is a block diagram of MTAP 305 which has a JTAG to MPSD interface. MTAP 305 provides external access to and control of domain test ports (DTPs). MTAP 305 is utilized as an interface between the JTAG and MPSD scan technologies. It capitalizes on the use of the JTAG protocol as a communication medium in addition to its use in boundary scan control. MTAP 305 allows access to test and emulation system resources. MTAP 305 block performs the following functions: test support, scan support and execution support.

Test support allows on board application of ATPG test patterns through direct conversion of JTAG TAP state to DTP MPSD codes and substitution of TCLK for UCLK. Also supported is chip tester application of ATPG test patterns through a direct MPSD mode. Scan support includes management of the switching and application of clocks sourced to each domain to assure test clocks are supplied in scan modes and generation of DTP MPSD Scan Codes from JTAG shift data state (SHIFT.sub.-- DR) and the path select in the JTAG instruction register. Execution Support includes management of the switching and application of clocks sourced to each domain to ensure that functional clocks are supplied in execution modes and generation of MPSD code sequences which control the execution modes of the domains through the DTPs.

Sequences of scanned information coupled with JTAG state diagram transitions direct the functionality provided by MTAP 305. A MTAP subsystem, called the Code State Machine (CSM), creates the MPSD execution code sequences used to control the individual domains. Since the JTAG logic is driven by Test Clock (TCLK) and functional logic is driven by Universal Clock (UCLK), MTAP 305 is designed to control the clocking of each domain.

Referring still to FIG. 35, Megamodule Test Access Port (MTAP) 305 consists of a JTAG TAP 500 (specified in the 1149.1 JTAG Specification), JTAG Data Path Control logic 510, Emulation Control Register (ECR) 520, the Code State Machine (CSM) 530, the Start Control 540, Command Decode 550, Command Control 560, and the MPSD Code Generator 570.

Still referring to FIG. 35, MTAP 305 can be divided between JTAG and DTP control blocks. The JTAG block includes the JTAG TAP 500, JTAG IR 580, and Data Path Control 510. This block decodes the JTAG activity at the device pins, receiving instructions and dispositioning them as scan path control qualifiers or providing them to the DTP block as command qualifiers. Command requests pass from the JTAG block to the DTP block, where they are processed and dispositioned to the appropriate functional logic block. The DTP Control block includes Start Control 540, Command Control 560, Emulation Control Register 520, MPSD Code State Machine 530, and MPSD Code Generator sections 570.

JTAG TAP 500 contains a state machine which tracks the JTAG state activity directed by the TCLK, TRST-, and TMS pins. This state machine generates all control signals used to capture, shift, and update both the instruction register and data register scan paths. Data Path Control 510 then uses the TAP and Instruction Register 580 information to generate scan control for the JTAG scan paths.

Still referring to FIG. 35, MPSD Code State Machine (CSM) 530 and Emulation Control Register (ECR) 520 are used together to generate MPSD execution code sequences from MPSD Code Generator 570 and control DTP clocks. Commands from Command Control block 560 initiate CSM 530's operations. CSM 530 selects and applies one of two programmable CO and Ce values from ECR 520 when directed by a command. The application sequence of the two preloaded code values is also specified by bits in ECR 520. This application sequence is programmable and can be made dependent on processor actions. CSM 530 operation is covered in greater detail in a following section. ECR 520 also supports device mode selection and miscellaneous test functions that are covered in the next section.

Code generator 570 takes inputs from the ECR mode bits, the TAP states, decode of the DTP JTAG IR control opcode, and the MPSD Code State Machine to form the MPSD codes supplied to the DTPs.

FIG. 36 illustrates MTAP Emulation Control Register 520. Emulation Control Register (ECR) 520 is a private scan path 452 (FIG. 33) within MTAP 305. It is specified by a SECR opcode placed in the JTAG instruction register. The ECR is implemented as a shift register and a shadow register, although not all bits are shadowed. The Emulation control register fields are described in Table 47.

Table 49 defines the clock select table based on the MCS bit and device mode field bits.

Referring to Table 54, if the EMUC code selects the HI-Z state for either or both of the EMUOE signals, the appropriate EMUOE signals are permanently driven low holding the EMUO driver in the HI-Z state. If the code selects an Open Collector state for one or both pins. the signal selected to drive EMUO also controls the state of EMUOE. If a false condition occurs on EMUO, EMUOE is switched low forcing the EMUO output driver to its HI-Z state. If a true condition occurs on EMUO, EMUOE is switch to a high causing the EMUn pin driver to change from a HI-Z output to the driving state. The execution of the IF.sub.-- CLRO or IF CLR1 JTAG commands inhibit the signals driving EMUO0 and EMUO1 until the signals return to their inactive state. TRST- causes the EMUC bits to be loaded with zeros.

The EMUC configuration code bits are implemented as shadow bits of the ECR. The configuration bits are updated when the JTAG TAP transitions through the IDLE state. This will allow the new event selection to be synchronized to UCLK thus eliminating the chance of glitching the EMUO signals.

If the Device Mode bits are set to STRAP, the EMU configuration codes are selected by the CPU.sub.-- EMUC bits of the ACR as defined in an earlier section. The EMUC field is initialized to 0000 by TRST- and the TLR JTAG state.

The Device Mode fields and MCS bit are implemented with shadow bits that are loaded during the UPDATE-DR JTAG state while the MPSD Code and Remote Event fields are not shadowed and are physically within the Emulation Control shift register bits. References to the ECR refer to the shadowed and unshadowed bits that supply functional information to other logic.

Since there is no synchronization when changing the device modes or the MCS bit, to avoid an invalid state caused by glitching the software must assure that when switching modes or the MCS bit all test ports are unlocked and have PAUS applied, or the currently selected clock is disabled (power down).

The mix of shadowed and unshadowed bits in ECR 520 result from the use of the MPSD Code and Remote Event fields with logic which runs on the functional clock (UCLK). These fields are applied directly to logic or registers which are being evaluated with UCLK. Shift register bits updated with respect to the test clock (TCLK) will change asynchronous with respect to UCLK. generating erroneous results in the UCLK controlled logic. The synchronization problem is addressed by inhibiting the use of ECR data by the UCLK controlled logic while the ECR shift register is being scanned. The inhibit has the effect of freezing or locking the state of the functional logic using the non shadowed shift register bits and is generated by emulation software through the use of a functional command. This inhibit process is an alternate form of synchronized shadowing, assuring that functional logic using information scanned by the test clock (TCLK) gives the appearance of being static at all times. The locking process is part of the MPSD Code State Machine and is discussed in a later section which describes the MTAP code state machine.

ECR 520 can not be scanned, even if the host software is just checking the ECR state, without the software first locking the CSM.

MTAP Start Control and Command Control

Referring again to FIG. 35, MTAP Start control logic 540 accepts command directives from the JTAG section of MTAP 305, synchronizes the directive and then, with the aid of MTAP Command Control 560, dispositions the command to MPSD Code State Machine 530 and logic blocks external to MTAP 305. A Domain command is initiated in the JTAG section only when JTAG IR 580 contains an opcode in the range of Ox20-Ox2F and the TAP moves to the IDLE state from either UPDATE.sub.-- IR or UPDATE.sub.-- DR.

The Domain command request is transferred to MTAP Start Control logic 540 where a state machine generates an IRBUSY signal synchronous to TCLK which inhibits further JTAG instruction register updates. Table 55 is a sequence of events which facilitate the generation of a start pulse for the command to be generated. START.sub.-- REQT and IRBUSY are reset by the TAP Test Logic Reset (TLR) state while START.sub.-- REQ1 and START.sub.-- REQ2 are reset by IRBUSY low.

Still referring to FIG. 36, signal SWINPROG inhibits the generation of START if the JTAG opcode is 0x20-0x23, 0x30, 0x32 or 0x24 (all commands targeted at the selection of a domain scan path or the ECR).

A truth table for the START Generation state machine is shown by Table 56.

The command generation process requires that the emulation software be responsible for not issuing a second command until the first one is complete. The IRBUSY SRL, visible through the JTAG IR capture information, allows the busy indicator to be examined by emulation software on the same instruction register scan that loads the next command. Since the capture state occurs before the update state, a logic zero capture of the IRBUSY flag on a command load assures the emulation software the instruction register update will positively occur. Commands scanned into the instruction register should end in the PAUSE.sub.-- IR state. allowing the emulation software to decide if progression to the JTAG IDLE state is warranted. Certain events which result from the generation of start pulses may be programmed to be seen at the emulation pins as interrupts. These interrupts may be used by the emulation software in lieu of polling the instruction register in some cases. The determination of interrupt applicability is left to the discretion of the programmer.

If the JTAG state machine is driven to the IDLE state with the SGABORT command loaded in JTAG IR 520, or the JTAG state machine is driven to the TEST-LOGIC.sub.-- RESET state, the Start Generator will be directed to its clear state. The SGABORT command will override IRBUSY and allow the command to be loaded into the JTAG IR.

Referring still to FIG. 35, Command Controller 560 generates all command strobes for CSM 530 and Domains. The state of the lock bit may affect which domains receive a command such as SDAT.sub.-- HALT. START is a one clock wide pulse generated by the JTAG START Control Logic when a command has been initiated from the JTAG interface. START is sent to the Command Control to be routed to its destination. The Command Control logic combines the START pulse with the JTAG IR value to form a specific command. These specific commands are sent to either MPSD Code State Machine 530 within MTAP 305 or a Domain outside MTAP 305. The commands are listed in Table 57.

MTAP Code State Machine

Still referring to FIG. 35, MPSD Code State Machine (CSM) 530 controls the application of MPSD codes from the EXE and TERM register fields of ECR 520 to MPSD code generator 570(which drives the MPSD bus). It also manages the clock switches required to scan (TCLK) and execute (UCLK) while in emulation mode.

CSM 530 is operational in all modes except ATPG mode. The CSM is used in emulation mode to generate programmable MPSD execution code sequences to the MPSD Code generator. The Execution code sequences are defined as FUNC, CNTL, HALT, and PAUS. The emulation software is required to direct the CSM to apply a PAUS code to the code generator prior to attempting a scan of the DIP Data or Control paths. The CSM uses the execution code to determine the domain clock selection. FUNC, CNTL, or HALT code application requires that domains have UCLK selected while the PAUS code may be applied with either UCLK or TCLK applied. Moving from a FUNC, CNTL. or HALT code to a PAUS causes a clock switch from UCLK to TCLK while a request to move from PAUS to FUNC, CNTL. or HALT causes a clock switch from TCLK to UCLK. All clock switches occur while a PAUS code is applied to unlocked domains.

FIG. 37 is a block diagram of CSM 530. The Code State Machine can be broken into an MPSD code controller 600, a code register 610, and clock switcher 620 sections. The code register section contains a two bit code register, a clock select bit, and a code multiplexer. The clock switcher contains a break before make synchronizing clock switcher with a switch in progress indicator. The code controller contains two state machine SRLs and all combinational logic to accommodate MTAP commands and REVT modes specified in the ECR. The state machine also generates multiplexer controls to select the EXE 611 and TERM 612 codes into the code register and load the clock select bit (TCLKON 613).

The code source for code register 610 is determined by code controller 600. The code register normally feeds back on itself in the absence of another directive. This allows the code controller to select EXE or TERM code fields from the ECR for a period of one clock. The one clock wide select window allows a new code to be entered into the code register after which the code recirculates. This scheme allows the code controller to pass commands directly from MTAP command control 560 to the multiplexer control within the code register block. The CSEL(3:0) 614 signals control the muxes on the code register.

If the next code to be applied by the CSM to the code generator is a run code (CNTL or FUNC), and the present code is PAUS then the CSM will force the HALT code for one clock cycle before the run code is applied. The reason for this is to allow the domains to enable their buses a clock early since buses are enabled (DBENB) for the HALT code.

Code controller 600 derives the multiplexer selects after resolving the priority of all requests. The priority is Clock Switch requests first, MTAP commands second, and Remote Event requests third. Clock switch requests take priority over MTAP commands by inhibiting MTAP START Control while a clock switch is in progress (SWINPROG is true) and a CSM command is specified in the JTAG IR. If a MTAP Command occurs simultaneously with a remote event command the event command is ignored. Any MTAP command or Remote Event can request a clock switch in either direction. The clock polarity required is embedded in the MPSD code with PAUS requesting the application of TCLK to all unlocked domains, while HALT, CNTL, and FUNC request the application of UCLK.

When a code load request is detected, the code controller determines which of three types of loads will occur. They are a code load with no clock switch, a switch from UCLK to TCLK, or a switch from TCLK to UCLK. The code controller treats each of these cases differently. No switch is detected if either of the following occur:

1) The current code is PAUS and the code to be loaded is PAUS, or

2) The current code is not PAUS and the requested code is not PAUS.

A UCLK to TCLK switch is detected if the current code is HALT, CNTL, or FUNC and the requested code is PAUS. A TCLK to UCLK clock switch is detected if the current code is PAUS and the requested code is HALT, CNTL, or FUNC. Processing the types differently assures that all clock switches occur while the MPSD PAUS code is applied to the Domains.

When no clock switch is required, code controller 600 selects the ECR field to be loaded and determines the next CSM state. Code Register 610 is loaded on the next clock along with the clock select bit and the CSM state.

When a code of PAUS is requested to be loaded into the code register while the current value is a HALT, CNTL, or FUNC, the code controller selects the ECR field to be loaded and determines the next CSM state. The code register is loaded to PAUS the next state and the TCLKON bit is loaded to a one. The clock switcher, comparing TCLKON to TCLK.sub.-- SEL (Test Clock Select) determines a clock switch is requested and SWINPROG (switch in progress) becomes active. After the clock switch is complete SWINPROG returns inactive and the code controller and MTAP START Control may proceed.

A request requiring a switch from TCLK to UCLK takes special processing by code controller 600 as it must hold the current PAUS code while the clocks are switched, then installing the code that caused the switch to occur. In this case the code controller loads the requested clock state in TCLKON, updates the CSM state, and inhibits the code register 610 load.

TCLKON 613 requests clock switcher 620 to generate the clock switch to functional. The updated CSM 530 state points to the code that must be loaded when the clock switch is completed. Since TCLKON represents a PAUS code, it can be compared to code register 610 contents to see if the clock selection and code value agree. When TCLKON is a zero, and the code register contains a PAUS, a delayed code register load is pending. TCLKON 613 is always sent directly to clock switcher 620 where a clock switch sequence is initiated if required. This assures that the clock is switched while a PAUS code is applied to the domains. Switch in progress (SWINPROG) 621 is generated by the clock switcher if TCLKON disagrees with the output of the clock switcher or PAUS.sub.-- NOW 615, the decode of the pause code in the code register. When a clock switch from TCLK to UCLK is complete, the switcher generates a LD.sub.-- CODE signal to the code controller. The code controller combines the LD.sub.-- CODE signal with the current CSM state to request either the EXE or TERM code field be loaded into the code register. SWINPROG inhibits START generation in MTAP 305 START Control when the JTAG IR contains 0x20-0x23, 0x30, 0x32 or 0x24.

FIG. 38 is a schematic of MTAP CSM clock switch circuit 620, which is a break before make switching circuit. A HALT to PAUS truth table is illustrated in Table 58, and a PAUS to HALT truth table is shown in Table 59.

Special CSM considerations with ECR scans are now noted. Since CSM 530 uses ECR 520 fields as static inputs, the ECR can only be scanned while the CSM is in the LOCKED state. Code controller 600 encodes the three code management states of EXECUTE, TERMINATE. and LOCKED indicated by signal 601, 602 and 603 respectively, into the two state register bits. The EXECUTE and TERMINATE states reflect the source of the current code on the code register. These two states may be directed through MTAP commands at any time. The Remote event field may direct either of these two states when the CSM state is not in the LOCKED state. The LOCKED state is directed by MTAP command and cannot be directed by the Remote Event field of the ECR. See Table 57 for command information. The state encoding is shown in Table 60.

MTAP MPSD Code Generator

Referring again to FIG. 35, the MPSD codes supplied to the domains are built in MPSD Code Generator 570. Code generator 570 drives the MPSD bus from either the DTP Data and Control Scan States, the MPSD Code State Machine (CSM) outputs, by mapping the JTAG TAP Controller state directly to MPSD codes, or by forcing the MPSD Strap state (normal operation mode) to the bus.

The generation of codes by code generator 570 has a hierarchy which assures that the device is functional in JTAG test logic reset state (TLR). This hierarchy also allows MTAP 305 to be initialized with emulation software while all Domains receive an MPSD functional run code (FUNC) along with a functional clock select. The code value developed by the MPSD Code Generator is the logical OR of inputs from several sources. It becomes the emulation software's responsibility to apply compatible code sources to achieve the desired code output.

MPSD Code Generator 570 performs a logical OR of code inputs from Code State Machine 530, Strap mode, Scan DTP Data, Scan DTP Control. and ATPG mode. When both device mode bits in ECR 520 are set, STRAP is active. STRAP is ORed with C1, C0, and Ce, forcing them to the logic one state and generating the MPSD code of FUNC. This allows other code sources to be initialized with emulation software prior to resetting the STRAP signal with software. STRAP masks all other inputs to the code generation logic. The device mode bits are set to logic ones by a logic zero on TRST-, the JTAG TAP transition to the Test Logic Reset (TLR) state or by programming the ECR mode bits by an ECR scan.

The DTP Path Scan, MPSD Code State Machine, and ATPG Mode Mapping code generation sources are made mutually exclusive by the emulation software and MPSD code generation hardware. MPSD execution codes of FUNC, CNTL, and HALT can only be generated by the MPSD Code State Machine or, ATPG Mode Mapping logic that converts JTAG TAP states directly to MPSD codes. Only one source for the execute codes is active at a time, with the inactive source supplying logic zeros to the code building function in MPSD Code Generator 570.

Both MPSD execution code sources are to be placed in their inactive state (supplying logic zeros to the MPSD Code Generator) before any DTP Data Control Scans are attempted. The state mapping for ATPG mode is chosen to assure compliance with this rule. The emulation software is responsible for assuring that the MPSD Code State Machine outputs are placed in the inactive state (PAUS, CSM.sub.-- C0 and CSM.sub.-- Ce logic zeros) prior to attempting MPSD scans. Since all non scan MPSD codes (FUNC, CNTL, HALT, and PAUS) have C1 a logic one, C1 is generated as NOT SHEFT.sub.-- DR OR NOT JTAG opcodes (Ox20-Ox23, Ox30 and Ox32) and C0 generated as SHIFT.sub.-- DR AND JTAG opcodes (Ox22-Ox23 and Ox32). This yields a zero on C1 only when actually scanning the MPSD Data or MPSD Control Paths and frees both execution code generators from developing C1.

Code generator 570 takes inputs from the ECR mode bits, the TAP states, decodes of the MPSD Data and MPSD Control opcodes, and the MPSD Code State Machine to form the MPSD codes supplied to the domains. When not in STRAP, C1 is assigned to the inverse of the SHIFT.sub.-- DR state ANDed with a JTAG opcode of Ox20-Ox23 and Ox30 and Ox32 (scanning MPSD Data or MPSD Control paths). This makes the generation of the MPSD scan codes of SDAT and SCTL impossible to generate unless the correct scan path is selected and the data register scan state occurs. C0 is set to a one in the JTAG SHIFT.sub.-- DR state if the MPSD Control path is selected (a JTAG opcode of Ox22-Ox23, 0x30 and Ox32).

An additional constraint is placed on scanning the DTP Data path. The CAPTURE.sub.-- DR state samples TCLK.sub.-- SEL and if it is false, the MPSD PAUS code is applied to the DTPs during SHIFT.sub.-- DR states. In addition the output of the DTP data scan path is forced to a zero. The captured value of TCLK SEL is to be output on the first bit of the DTP Data scan path when TCLK.sub.-- SEL is false. Scan codes can only be generated when emulation software assures that the CSM.sub.-- C0 and CSM.sub.-- Ce outputs of the CSM are a logic zero.

Tables 61, 62 and 63 show the contribution of the various code generator sources. Table 61 shows a truth table for STRAP code, Table 62 shows a truth table for Emulation mode, and Table 63 shows a truth table for ATPG mode.

In Emulation mode MPSD Code Generator (MCG) 570 controls the C1 bit of the MPSD code that is applied to the MPSD Test Port bus 306 (FIG. 15). The MCG forces C1 high except when a JTAG MPSD data scan is in progress (C1=0). When the CSM outputs an MPSD PAUS state the MCG maps the JTAG TAP states to MPSD scan codes. The JTAG data scan path is coupled directly to the MPSD bus data scan path. All unlocked MPSD Test Ports are then scanned on TCLK with data when the JTAG TAP state is transitioned to the SHIFT.sub.-- DR state. The MPSD data or control path are selected through the JTAG IR path selects. During the JTAG SHIFT DR state cycle C1 is driven low if either the SDAT.sub.-- HALT, SDAT.sub.-- CNTL, SCTL.sub.-- HALT, or SCTL.sub.-- CNTL opcode are present (Ox20-Ox23, Ox30 and Ox32). CO is driven to a one if SCTL.sub.-- HALT, or SCTL.sub.-- CNTL (Ox22-Ox23 and Ox32) paths are selected or a zero if SDAT.sub.-- HALT or SDAT.sub.-- CNTL (Ox20-Ox21 and Ox30) paths are selected.

In ATPG mode all DTP MPSD codes are mapped directly from the JTAG TAP state controller except the Run-Test/Idle state. In Run-Test/Idle the CSM is utilized to drive MPSD execute code states (CNTL or HALT). The CSM is unlocked and the EXE or TRM state (selected by JTAG IR command) drives the MPSD bus. In Run-Test/Idle, if the JTAG-IR contains a code other than Ox20 or Ox21, the previous mapped state (HALT or PAUS) continues to be applied to the MPSD bus. The previous mapped state also continues to be applied if the CSM is locked (or unlocked but the new MPSD code has not been applied to the MPSD bus). When the Run-Test1Idle state is exited the CSM will be locked and the MPSD code generator will utilize the JTAG state mapping to drive the MPSD bus. As in Emulation mode the DTP scan paths are selected through a DTP scan opcode loaded in the IR. In ATPG mode the mapping of the data scan register group of scan states forces MAP.sub.-- CO and MAP.sub.-- Ce to zeros except for the SHIFT.sub.-- DR state where the normal JTAG to MPSD shift code conversion occurs. If emulation mode is switched to from ATPG mode (from an ECR scan) the CSM will continue to drive PAUS (the SHIFT.sub.-- DR mapped state) until the CSM is unlocked.

Table 65 shows the state decode for both Emulation and ATPG modes.

In STRAP mode all ECR lock bits are driven inactive, forcing the domains to use current MPSD codes. The MPSD Strap mode also forces the MPSD FUNC code onto MPSD bus 306.

MTAP Domain Clock Management

The Domain clock management is controlled by the Device mode. The UCLK.sub.-- SEL and TCLK.sub.-- SEL signals are gated in MCG 570 by the mode bits. In MPSD Strap mode the MCG overrides the Clock Switcher to generate UCLK.sub.-- SEL. ATPG mode overrides the Clock Switcher to select TCLK. In Emulation mode the UCLK.sub.-- SEL and TCLK.sub.-- SEL are sourced at Clock Switcher 620 outputs.

When UCLK.sub.-- SEL and TCLK.sub.-- SEL are inactive they gate off the functional and scan clocks sourced from the DTP to the domain. When clock switcher 620 in the CSM detects a condition that requires the domain clock to switch from UCLK to TCLK, UCLK.sub.-- SEL is driven inactive switching off the master clock while the slave phase is high and the master phase is low. The DTP's clock mux assures that the slave phase remains high at the clock mux output while both clocks are deselected. TCLK.sub.-- SEL is driven active after a synchronization delay when UCLK.sub.-- SEL is inactive enabling TCLK (while slave is high and master is low) to the Test Port. This mechanism performs a break before make switch of the functional clock. Switching from TCLK back to UCLK works identically. The CSM applies the PAUS code on the MPSD bus until the clock switch is complete. Also, only Test Ports that are unlocked are clock switched.

When a domain is in the process of being unlocked, the state of the clock switching signals must be such that they match the state of the clocks in the domain being unlocked. If the states do not match then a clock switch without synchronization will occur. This situation will be avoided by the software.

FIG. 39 is a schematic for MTAP Event Generation circuit (EVTA) 590. EVTA is an event which is enabled by unlocked domains. Two types of events can generate EVTA, these are MSGSW and CPU.sub.-- DONE.

It can be seen from FIG. 39 that either an MSGSW or a CPU.sub.-- DONE can generate EVTA depending on the state of CPULOCK, i.e. if the CPU is unlocked then the only event which can generate EVTA is DONE. The DONE is also qualified by ANYSTOP such that if the DONE was generated by ANYSTOP then no EVTA is generated, i.e. only MPSD generated DONEs can generate an EVTA. The reason for this is that if ANYSTOP was active there is no reason to generate EVTA and hence alter the state outputted by the CSM and hence the state of the MPSD bus.

MTAP Power Down Support

In emulation mode MTAP commands and scan clock switching relays on an active functional clock. In power down mode this may not be true. Therefore a functional clock request (FCLK.sub.-- REQ) signal will be generated from MTAP 305 to the megamodule that requests functional clocks be enabled. FCLK.sub.-- REQ is generated as defined in Table 66.

Also note that CPU.sub.-- DONE must be driven inactive when the CSM applies CNTL or FUNC if the megamodule is in a power down state. CPU DONE must be driven active when the CSM applies HALT.

FIG. 40 illustrates MTAP Counter circuitry 630, which is located in CSM 530. Counter 631 is a 10 bit down counter that is configurable for different functions (performance analysis, execution control, or data streaming). Counter 631 and its control bits are shadowed for scan. The counter runs off of UCLK. while the shadow bits are run off of TCLK. When the counter's scan path is selected (LEVT.sub.-- CNTR command), the shadow bits are loaded with the counter's value during the CAPTURE.sub.-- DR JTAG state (not in Data Streaming Mode) and the counter is loaded from the shift bits during the JTAG IDLE state. Using the IDLE state to perform the counter load will allow a synchronous load to be performed. MTAP.sub.-- BORROW and XFER.sub.-- DATA signals are a single clock cycle wide.

The AUX bit is only in the counter's shift path and is utilized as the 16th bit for the ID bus value during a SMM.sub.-- ID scan. During a SMM.sub.-- ID scan operation the 16 bits of the counter's shift register is loaded by the CAPTURE DR state with the 16 bit ID bus value.

The counter mode field (CM[1:0]) is utilized to configure counter 631 for the selected function. Table 67 defines the CM bit field of counter 630, and Table 68 defines the CES bit field.

Special Emulation (SE) Device Support

Referring again to FIG. 15, megamodule boundary 301 contains signals to support a special emulation device. An SE device includes megamodule 300 and additional control registers in a fourth SE analysis (SEA) domain. The control registers for an SE device are memory mapped.

Referring back to FIG. 11, the program counter value of the execution packet in the DP phase is latched in a register referred to as PCDP. The PCDP can be compared to values held in the emulation logic for exact matches, range comparison, or bit-field masking comparison. If the PCDP matches a particular address set for hardware program breakpoint, the SE emulation logic will set SEE during that cycle.

DADDR, DDATA.sub.-- I, DDATA.sub.-- O, DRNW, and DBS are available on the CPU boundary to detect breakpoints on data accesses. Both address and data (qualified by the appropriate strobes) can be used for exact match, range, or mask compares of breakpoint values. Data breakpoints differ from program breakpoints in that data addresses are present at the CPU boundary during the E2 phase. The execute packet causing the breakpoint cannot be halted before entering execution. However, the emulation logic can reconstruct from the PCDP stream the instruction that caused the breakpoint. Some buffering of the PCDP is necessary for this feature.

Trace and Performance Analysis

Memory control and interrupt acknowledge signals are available on the bus for trace. A signal BRTK indicates the address represented by PCDP was a branch target. Eight signals (FU.sub.-- LL, FU.sub.-- M1, FU.sub.-- D1, FU.sub.-- S1, FU.sub.-- L2, FU.sub.-- M2, FU.sub.-- D2 FU.sub.-- S2) indicate whether the execute packet in E2 (in E1 during the previous cycle) executed an instruction on the L-,M-,D-, or S-units, on the A- or B- sides, respectively. The unit assignment refers to the decode block used for the decode. This can be used to evaluate parallelism and whether conditions for a particular instruction evaluate as true. Finally, a 3-bit EPSIZE output indicates the size of the execute packet (in words) in DC. This should be zero during interrupt processing and all extra cycles introduced by IDLEs and multi-cycle NOPs.

FIG. 41 is an interconnect diagram showing in more detail the interconnect between MTAP 305 and the domains and DTPs of megamodule 300. As indicated on FIG. 33, SE Analysis utilizes an independent scan path 457 and 458, although it shares MTAP MPSD control bus (C0, C1, Ce) with the megamodule DTPs. The signals described in Table 69 interface MTAP 305 to the SE Analysis module.

The SE utilizes a DTP identical to the megamodule DTPs. The SE Analysis C0, C1, and Ce bits are timed such that they are presented to the SE's DTP in parallel with the megamodule DTP's C0, C1 and Ce bits.

Still referring to FIG. 41, MTAP 305 module provides the interface fan out between the JTAG environment and each DTP. The fanout to each DTP consists of an MPSD code (CO, C1, Ce) bus, the signals that control domain clock switching between TCLK and UCLK, the update signals that control the loading of control SRLs in the Test Port, emulation status, and MTAP commands that are not broadcast on the MPSD BUS.

Still referring to FIG. 41, Data Stream Scan Path circuitry 700 includes circuitry to transfer stream data to streaming register 393 (FIG. 30D). A STRM.sub.-- I/O bus 701 transfers stream data to CPU domain 10. This circuitry is described in detail with respect to FIG. 42.

FIG. 42 is a schematic showing details of datascan path circuitry 700. Scan register 710 forms a portion of DATA.sub.-- STRM scan path 456 (FIG. 33). STRM.sub.-- I/O bus 701 transfers stream data to data streaming register STREAM 393 (described previously with reference to FIG. 30D) which is in CPU domain 10. Stream data can be transferred to and from various storage locations in CPU domain 10 as described previously with reference to Tables 37-41. Five bit counter 632 is a portion of counter 631 which was described with reference to FIG. 40. comparator 635 monitor JTAG IR 580 to determine when a SDAT.sub.-- STRM JTAG instruction is present in JTAG IR 580. Gate 636 decrements counter 632 in response to each JTAG shift state indicated by signal 637. comparator 712 detects when counter 632 has reached 00 and asserts signal XFER.sub.-- DATA 702.

As described with reference to Table 64, stream data is transferred from scan register 710 when a load instruction is in CPU 10s instruction register. In this case, STREAM register 393 is loaded in response to signal 702. When a store instruction is in CPU 10s instruction register, data is transferred from STREAM register 393 to scan register 710. In this case, write enable signal is asserted in response to the store instruction and gate 713 asserts signal 717 to load scan register 710 in response to XFER.sub.-- DATA signal 702.

Still referring to FIG. 42, status circuit 730 forms MTAP status bits STRY.sub.-- TGLE and STSW.sub.-- TGLE in response to XFER.sub.-- DATA signal 702, as described with reference to Table 44. Handshake signals 731 include CPU.sub.-- DONE and TERM (0, 1) from ECT 520.

FIG. 43 is a schematic showing EMU pin connections. EMU[1:0] are EMU Input/Output pins. These pins provide emulation event input for multiprocessor support and output for external event detection. All JTAG and emulation pins may be left as no connects in an end system. To facilitate, this the TMS, TCLK, TDI, EMU1, and EMU0 pins have small pull up resistors on them, as indicated by pull-up 900. TRST- has a small internal pull down resistor to assure that the JTAG TAP and boundary logic remains reset if left as a no connect.

FIG. 44 is a block diagram of an alternative embodiment of a device 1000 which uses aspect of the present invention. There are multiple configurations that can be supported by MTAP 305 JTAG Interface. FIG. 44 is one example which has a megamodule 1010 and a custom logic module 1020. It is important to note that all JTAG module TDO/TDI signals are connected serially and all TMS and TRST signals are connected in parallel. The order of the modules is not important. The module order must be provided for the target systems JTAG device configuration file. This file is utilized by the emulation software to manage multiple JTAG devices and JTAG modules on the same scan chain.

EMU Status

Referring again to FIG. 41, various emulation status bits are generated within the CPU or the on-chip memory system of microprocessor 1. These signals are brought into MTAP 305 as levels. They are latched into JTAG IR 580's shift register during the CAPTURE-IR state. Table 70 describes Stop Mode Emulation status signals. Table 71 describes real time mode emulation status signals. Table 72 describes CPU Emulation Status signals. Table 73 describes CPU Emulation Event signals.

As used herein, the terms "applied," "connected," and "connection" mean electrically connected, including where additional elements may be in the electrical connection path.

While the invention has been described with reference to illustrative embodiments, this description is not intended to be construed in a limiting sense. Various other embodiments of the invention will be apparent to persons skilled in the art upon reference to this description. It is therefore contemplated that the appended claims will cover any such modifications of the embodiments as fall within the true scope and spirit of the invention.