Synchronous network traffic processor

[00002] The present application claims priority to U.S. Provisional Patent Application entitled “SYNCHRONOUS NETWORK TRAFFIC PROCESSOR”, filed Dec. 8, 2000, and bearing Ser. No. 60/254,436.

[00003] The present invention relates generally to data processors for high speed communication systems and networks. More particularly, the present invention relates to processors for real-time analysis and processing of network data.

[00004] Network communication devices are, in general, protocol dependent. Since devices which communicate within computer and storage Networks must strictly adhere to rapidly changing protocols associated with those networks, it has become clear that the use of protocol independent-network processors to analyze, generate and process traffic within these networks is of extreme practical and business importance.

[00005] As such, network communication devices typically include specially designed protocol-specific state machines and decoder logic. Protocol-specific hardware offers the advantages of high performance and cost-effectiveness. However, high-speed networking protocol standards are in a state of flux—new protocols are emerging and changing all the time. Since protocol-specific hardware designs are not reusable for different protocols, major redesigning efforts are expended in producing protocol-specific hardware for these emerging protocols. Furthermore, protocol-specific hardware designs cannot be easily upgraded to include new features and functionality. In most cases, modifications to the hardware itself must be made.

[00006] An embodiment of the present invention includes a network traffic processor. The processor itself is protocol independent; it does not have any hardwired logic for recognizing packets, frames, or any other protocol-specific entities. Framing-based tasks are performed inside the processor using user-defined software instructions. Thus, the same processor may be used to implement network data processing systems for virtually any protocol. Furthermore, new features and functionality can be easily added to the network traffic processor through software upgrades. As a result, the development cost of network data processing systems, as well as the cost of upgrading the system, can also be greatly reduced.

[00007] The network traffic processor of the present invention is capable of synchronously processing and generating data for high-speed protocols (serial or otherwise), on a wire-speed, word-by-word basis. Significantly, the processor is capable of operating data directly on its input/output busses without requiring the data to be moved in and out of registers or internal memory units. The low overhead of operating on data directly on its input/output busses, minimizes the total clock cycles required to process and generate each I/O data word. The network processor receives and transmits data on every clock, and executes instructions upon the same clock, eliminating the need for polling or interrupts to determine whether data is ready to be read or written.

[00008] According to an embodiment of the present invention, multiple synchronous network traffic processors may be implemented in a system, in a chain mode or otherwise, for providing a multitude of programmable functions. The synchronous network traffic processor may also be integrated with other hardware functions, such as other types of processors, memory controllers, FIFOs, etc.

[00009] The synchronous network traffic processor, in one embodiment, has a low gate count and can be easily implemented using programmable logic (e.g., FPGA). An appropriately programmed synchronous network traffic processor may replace modules traditionally implemented with hard-wired logic or ASIC.

[00020] The present invention provides a processor for synchronously processing and generating data for high speed serial protocols on a word-by-word basis. In contrast to conventional microprocessors, whose main focus is on register and memory operations, an emphasis of the present invention is I/O processing. The processor of the present invention is capable of operating directly on the data streams in its I/O busses without requiring the data to be moved in and out of registers or internal memory. In addition, the processor of the present invention has a wide instruction set. These factors reduce the total clock cycles required to process and optionally modify each I/O data word. Indeed, in one embodiment of the present invention, a data word may be processed and modified in a single instruction clock cycle.

[00021] Significantly, the processor of the present invention executes instructions synchronously with a master clock that drives the I/O busses. In one embodiment, the processor interfaces directly to the inbound serial-parallel and outbound parallel-serial converters of the receive and transmit serial interfaces. Words are received and transmitted on every clock cycle, eliminating the need for polling or interrupts to determine whether data is ready to be read or written. The processor does not have any hardwired logic for recognizing packets, frames, or any other asynchronously-arriving protocol-specific entities. The emphasis is on individual words, which arrive synchronously with instruction execution. Any framing functionality is performed by software. Thus, the processor may be programmed to handle any network protocol.

[00022] FIG. 1 is a block diagram illustrating the main functional units of a synchronous network data processor 100 in accordance with an embodiment of the present invention. As illustrated, the synchronous network data processor 100 includes a data compare unit 110, a data modify unit 120, an execution control unit 130, a peripheral unit 140, an input pipeline unit 150, an instruction memory 160, and a bank of general-purpose registers 170. The peripheral unit 140 of the illustrated embodiment includes control signal decoders 141, counters 142, control registers 144, an external memory interface 146, and a local interface 148. In the preferred embodiment, instruction memory 160 is a 128-word instruction memory, and register bank 170 includes sixteen banks of 40-bit registers. Data are communicated between the main functional units via 40-bit wide data paths, corresponding to four ten-bit undecoded input characters and four eight-bit decoded characters plus control or status bits. Forty-bit wide data paths illustrated in FIG. 1 include: PTPIPE_A, PTPIPE_B, INPIPE_A, INPIPE_B, IMMDATA_1, IMMDATA_2, REG_RD_DATA1, REG_RD_DATA2, PERIPH_WR, DM_PERIPH_RD, DC_PERIPH_RD, and REG_WR_DATA. Also illustrated are address busses and control signal paths such as PIPE_CTRL, CTRL_REG, DM_CTRL, DC_CTRL, INSTRUCTION, COMPARE_FLAGS, PERIPH_FLAG, START_STOP, IWR_ADDR, IWR_DATA, DM_PERIPH_CTRL, DM_REG_CTRL, DC_PERIPH_CTRL, and DC_REG_CTRL. For simplicity, some addresses busses and control signals are omitted in FIG. 1.

[00023] The input pipeline unit 150, in the present embodiment, includes four 40-bit wide by 16-stage pipeline registers for the input busses. Two of these pipelines (INPIPE_A, INPIPE_B) feed data from input bus IN0 and IN1 to the data compare unit 110 and data modify unit 120; the other two pipelines (PTPIPE_A, PTPIPE_B) are used for automatic pass-through of data from the input busses IN0 and IN1 to output busses OUT0 and OUT1 without program intervention. The input pipeline unit 150 is driven by an externally generated clock signal CLK. Particularly, each pipeline of the input pipeline unit 150 is operable for receiving/outputting one word during one cycle of the clock signal CLK. The pipeline stages from which the outputs are taken are selectable by control signals PIPE_CTRL and CTRL_REG. The signal PIPE_CTRL is generated by the execution control unit 130 based on a currently executed instruction. The control signal CTRL_REG is generated by the control registers 144 based on the values stored therein by the execution control unit 130 in previous execution cycles.

[00024] In the present embodiment, the execution control unit 130 executes one instruction at every instruction cycle. Instructions are fetched and executed from the internal instruction memory 160. Any results the instruction generates may be used in the following instruction. Instruction execution may be interrupted by a trap, which can be generated either internally or from the external interrupt pins. Traps transfer control either to a fixed address or a relative offset from the current program counter (PC); the trap address, absolute/relative mode, and condition are all software-programmable. Every instruction may execute conditionally. Further, every instruction may specify up to two different conditional relative branches, each with its own destination address. Conditional execution control fields are shared with the control fields for the second branch. Therefore, if conditional execution is used the second branch must be disabled or use the same condition.

[00025] The processor 100 can execute two types of instructions: data compare instructions and data modify instructions. Data compare instructions are for generating control signals that control the data compare unit 110; data modify instructions are for generating control signals that control the data modify unit 120.

[00026] Significantly, the execution control unit 130 is synchronous with the input pipeline unit 150. That is, both the execution control unit 130 and the input pipeline unit 150 are driven by the same externally generated clock signal CLK. During each cycle of the clock signal CLK, one data word is received by each pipeline of the input pipeline unit 150 and one instruction is executed by the execution control unit 130. This is significantly different from conventional microprocessors where data is required to be moved in and out of registers or internal memory and where the instruction clock is not synchronous with the I/O clock.

[00027] With reference still to FIG. 1, the data compare unit 110 is operable for selectively performing mask/match comparisons of two instruction-specified operands during each instruction cycle. In the present embodiment, the instruction-specified operands may come from the input pipeline unit 150 (via INPIPE_A, INPIPE_B), the register bank 170 (via REG_RD_DATA2), peripheral units 140 (via DM_PERIPH_RD), and the execution control unit 130 (via IMMDATA_1, IMMDATA_2). The mask/match and compare operations performed by the data compare unit 110 are instruction-specified. In particular, the mask/match and compare operations performed are specified by the control signal DC_CTRL, which is generated by the execution control unit 130 based on the currently executed instruction. The data compare unit 110 stores the results of the mask/match comparisons to a set of compare flags, which are provided to the execution control unit 130 and peripheral unit 140 (via COMPARE_FLAGS). The set of compare flags may be used by the execution control unit 130 and the peripheral unit 140 in the next instruction cycle to conditionally branch, execute, trap, increment a counter, etc. In the present embodiment, there is one compare flag for each 8-bit byte of the 40 bit input word, allowing multiple independent byte comparisons as well as whole 40-bit word comparisons in one instruction. Also illustrated in FIG. 1 are the DC_REG_CTRL and the DC_PERIPH_CTRL signal paths that communicate addresses and commands from the data compare unit 110 to the register bank 170 and the peripheral unit 140, respectively.

[00028] The data modify unit 120 of the present embodiment includes arithmetic logic units (ALUs) operable for performing arithmetic and logic operations using instruction-specified operands and operators. In the present embodiment, instruction-specified operands and operators may come from the input pipeline unit 150 (via INPIPE_A, INPIPE_B), the register bank 170 (via REG_RD_DATA1), peripheral units 140 (DM_PERIPH_RD), and the execution control unit 130 (via IMMDATA_1, IMMDATA_2). Using the instruction-specified operands and operators, the data modify unit 120 generates output data words that are provided to the output busses OUT0 and OUT1, the register bank 170 (via REG_WR_DATA), and/or the peripheral units 140 (via PERIPH_WR). The data modify unit 120 also allows instruction-specified data to pass through unaltered to the output busses OUT0 and OUT1. The modification operations performed by the data modify unit 120 are instruction-specified. In particular, the data modifications performed by the data modify unit 120 are specified by the control signal DM_CTRL, which is generated by the execution control unit 130 according to the currently executed instruction. Also illustrated are the DM_REG_CTRL and the DM_PERIPH_CTRL signal paths that communicate addresses and commands from the data modify unit 120 to the register bank 170 and peripheral unit 140, respectively.

[00029] With reference still to FIG. 1, the peripheral unit 140 includes four 20-bit counters 142, control registers 144, an external memory/peripheral interface 146, and a local interface 148. The local interface 148 allows a host computer to download instructions to the instruction memory 160 via IWR_ADDR and IWR_DATA busses, and to control the operations of the processor 100 via START_STOP signals and PERIPH_FLAGS. In addition, the control register 144 generates the CTRL_REG signal for controlling the operations of the pass-through pipes of the input pipeline unit 150. The local interface 148 also allows the host computer to communicate with the processor 100 via shared mailbox registers (not shown). Counters 142 that maybe cascaded to give two 40-bit counters or one 40-bit and two 20-bit counters. Each counter 142 has an independently programmable increment enable, allowing it to increment in different modes: synchronously at every clock cycle, selectively when a register is written, or based on a mask/match of the compare flags generated by the data compare unit 110. Additionally, one or two counters 142 may be used as an address generator for the external memory/peripheral interface 146. The data modify unit 120 may configure the counters 142 and the control registers 144 by communicating appropriate data via the PERIPH_WR bus.

[heading-00030] An Exemplary Implementation of the Input Pipeline Unit

[00031] An exemplary implementation of the input pipeline unit 150 according to one embodiment of invention is illustrated in FIGS. 2A and 2B. FIG. 2A illustrates two input pipelines 210 and 220, and FIG. 2B illustrates two pass-through pipelines 230 and 240. Pipelines 210, 220, 230 and 240 each includes sixteen 40-bit wide registers 214 (herein called 16-stage pipeline registers) that are driven by the clock signal CLK.

[00032] As illustrated in FIG. 2A, input pipeline 210 includes a multiplexer 212 that selectively provides data from either one of the input busses IN0 and IN1 to the 40-bit wide by 16-stage pipeline registers 214 according to a control signal PA_SRC provided by the control registers 144 of the peripheral unit 140. Likewise, input pipeline 220 includes a multiplexer 212 that selectively provides data from either one of the input busses IN0 and IN1 to the pipeline registers 214 according to a control signal PB_SRC, which is also provided by the control registers 144.

[00033] In the illustrated embodiment, each stage of the pipeline registers 214 includes an output for outputting one of the input data words after a delay of a number of clock cycles corresponding to a position of the respective stage in the pipeline. The outputs of the pipelines 210 and 220 are determined by the pipeline stage select multiplexers 216, which select the stages from which the outputs are taken. The particular stages of the pipelines 210 and 220 from which the outputs are selected are controlled by control signals PA_WORD_SEL and PB_WORD_SEL, which are generated by the execution control unit 130 in accordance with the currently executed instruction.

[00034] Pass-through pipelines 230 and 240 of FIG. 2B are used for automatic pass-through of unmodified data from the input busses IN0 and IN1 to the output busses OUT0 and OUT1 without program intervention. Similar to pipelines 210 and 220, each stage of the pipeline registers 214 includes an output for outputting one of the input data words after a delay of a number of instructions cycles corresponding to a position of the respective stage in the pipeline. The outputs of the pipelines 230 and 240 are determined by the pipeline stage select multiplexers 226, which select the stages from which the outputs are taken. The particular stages of the pipelines 230 and 240 from which the outputs are selected are controlled by control signals P0_WORD_SEL and P1_WORD_SEL, which are provided by the control registers 144 of the peripheral unit 140.

[heading-00035] An Exemplary Implementation of the Data Compare Unit

[00036] An exemplary implementation of the data compare unit 110 is illustrated in FIGS. 3A-3C. As shown in FIG. 3A, the data compare unit 110 includes source select and mask units 310, comparators 320 and flag update units 330. Each source select and mask unit 310 is configured for receiving data from the input pipeline unit 150 (via INPIPE_A, INPIPE_B), the register bank 170 (via REG_RD_DATA2), the peripheral unit 140 (via DC_PERIPH_RD) and the execution control unit 130 (via IMMDATA_1, IMMDATA_2). The source select and mask units 310 perform instruction-specified masking operations on the data to generate masked data and comparands to be provided to the comparators 320. The comparators 320 perform comparisons or “matching” operations between the masked data and the comparands to generate match outputs, which are provided to the flag update units 330. The flag update units 330 in turn generate a set of compare flags DC0, DC1, DC2, DC3 and DC4 based on instruction-specified flag update modes.

[00037] In the present embodiment, there is one compare flag for each 8-bit byte of the 40 bit input word, allowing multiple independent byte comparisons as well as whole 40-bit word comparisons in one instruction. It should be appreciated that the data to be masked and the comparands to be generated by the source select and mask units 310 are instruction-specified. Specifically, each of the select and mask units 310 receives the control signal DC_CTRL, which is generated by the execution control unit 130 according to a currently executed instruction.

[00038] FIG. 3B illustrates an exemplary implementation of a source select and mask unit 310 in accordance with an embodiment of the present invention. As illustrated, the source select and mask unit 310 includes 8-bit multiplexers 342a-342f. Although it is not illustrated in FIG. 3B, it is appreciated that the multiplexers 342a-342f are controlled by the signal DC_CTRL. Thus, the sources of the data, the mask and the comparand are specified by the currently executed instruction.

[00039] It should also be noted that the data paths within the illustrated source select and mask unit 310 are only eight bits wide. For example, the source select and mask unit 310 processes bit-0 to bit-7 of the 40-bit wide data. The remaining bits of the 40-bit data words are handled by the other source select and mask units 310 of the data modify unit 120.

[00040] As illustrated, multiplexes 342a-342c each includes inputs for receiving data from the input pipeline unit 150 (via INPIPE_A and INPIPE_B). The output of the multiplexer 342a is coupled to one of the inputs of multiplexer 342d, which also receives data from the register bank 170 (via REG_DATA2) and from the peripheral unit (via DC_PERIPH_RD). Thus, by applying the appropriate control signals, the output of the multiplexer 342d, which is the data to be masked, can be chosen from any one of these sources. Similarly, because multiplexer 342e is coupled to receive data from input pipeline unit 150 (via multiplexer 342b), the register bank 170, or the execution control unit 130 (via IMMDATA_1), the output of the multiplexer 342a, which is the mask data, may be chosen from any one of these data sources. The outputs of multiplexer 342e-342f are coupled to an AND-gate 344, which performs a masking operation on the data. In the present embodiment, the comparand may be selected from data within the input pipeline unit 150, the register bank 170, the peripheral unit 140 or the execution control unit 130 (via IMMDATA_2) when appropriate control signals are applied to multiplexers 342c and 342f.

[00041] FIG. 3C is a block diagram illustrating an exemplary flag update unit 330 in accordance with an embodiment of the present invention. The flag update unit 330 provides additional programmability and flexibility to the processor 100 by allowing the instruction to specify how the compare flags are updated. Particularly, as illustrated in FIG. 3C, the flag update unit 330 includes an AND-gate 332, an OR-gate 334, and XOR-gate 336, each having an input for receiving a comparison result from a comparator 320. The outputs of the logic gates are coupled to inputs of multiplexer 338. Responsive to a flag update mode control signal generated by the execution control unit 130, the multiplexer 338 selects one of the outputs of AND-date 332, OR-gate 334, XOR-gate 336, or the comparison results from the comparator 320, to be provided to a memory element 342 (e.g., a D-flip-flop). The output of the memory element 342 is fed back to the inputs of the logic gates 332, 334 and 336 to form feed-back loops. In this way, the flag update unit 330 updates the compare flags according to the instruction and according to the state of the compare flags in a previous instruction cycle. It should be noted that the memory element 342 is synchronous with the clock signal CLK that drives the input pipeline unit 150 and the execution control unit 130. Thus, the updated compare flags are provided to the execution control unit 130 for use in the next clock cycle.

[heading-00042] An Exemplary Implementation of the Data Modify Unit

[00043] FIG. 4 is a block diagram illustrating an exemplary implementation of the data modify unit 120 in accordance with an embodiment of the present invention. According to the present invention, the data modify unit 120 may access any instruction-specified data stored within the input pipeline unit 150, and modify the instruction specified data using an instruction-specified operator during one instruction cycle. The data modify unit 120 may also allow data to pass-through without any modification.

[00044] Particularly, as illustrated in FIG. 4, the data modify unit 120 includes two multiplexers 410a-410b, which are operable to receive data from input pipeline unit 150 (via INPIPE_A, INPIPE_B), the register bank 170 (via REG_RD_DATA1), or the peripheral unit 140 (via DM_PERIPH_RD). The outputs of the multiplexers 410a-410b are coupled to ALUs 420a-420b, which also receive data from the execution control unit 130 as operands (via IMMDATA_1, IMMDATA_2). The outputs of the ALUs 420a-420b are provided as inputs to another ALU 420c. The outputs of the ALUs 420a-420c are also provided to multiplexers 430a-430b. The multiplexers 430a-430b are also coupled to receive data directly from the pass-through pipelines PTPIPE_A and PTPIPE_B of the input pipeline unit 150. The control signals out0_src and out1_src, received from the control registers, are for selecting the inputs to the output multiplexers 430a and 430b, respectively. The output of the multiplexers 430a-430b are coupled to output registers 440a-440b, which provide data to the output busses OUT0 and OUT1 of the processor 100.

[00045] According the present embodiment, the sources of the data to be modified, as well as the operators, are instruction-specified. Particularly, the data modify unit 120 receives the control signals SRC1_SEL, SRC2_SEL, op1, op2, op3 (via control signal bus DM_CTRL), which are generated by the execution control unit 130 according to the current instruction. The control signals SRC1_SEL and SRC2_SEL are for selecting the inputs of multiplexers 410a-410b. The control signals “op1”, “op2”, and “op3” are for controlling the logic operations of ALUs 420a-420c. Thus, by using appropriate instructions, the data modify unit 120 may be configured for performing a variety of instruction-specified data modification operations during each clock cycle to generate the desired data for output.

[heading-00046] Exemplary Applications of the Processor of the Present Invention

[00047] FIG. 5 is a block diagram illustrating a high-speed data modification system 520 coupled between network devices 510 and 512. As illustrated, network devices 510 and 512 communicate with one another via high speed communication paths 514 and 516. Inserted into the high speed communication paths 514 and 516, the data modification system 520 enables real-time system-level testing of the devices 510 and 512 by injecting errors into the communication paths 514 and 516, and monitoring the responses of the devices 510 and 512.

[00048] As illustrated, data modification system 520 includes two trace memories 522 for capturing the data that are communicated between the devices 510 and 512 for output to an analyzer. Additionally, data modification system 520 includes a trigger subsystem 526 and two data jammers 524. The trigger subsystem 526 monitors the data paths 514 and 516, waiting for a datum in the streams to match a predefined pattern. When the trigger subsystem 526 detects an input datum matching the predefined pattern, the trigger subsystem 526 generates a trigger signal to the data jammers 524. The data jammers 524 respond to the trigger signal by “jamming”—altering selected portions of the input datum in a predefined manner in real time.

[00049] The trigger subsystem 526 and the data jammers 524 may be implemented with the high-speed synchronous network data processor of the present invention. Particularly, one synchronous network data processor 100 may be used to implement the trigger subsystem 526 by loading appropriate data compare instructions and data modify instructions into the processor. Each of the data jammers 524 may also be implemented with a synchronous network data processor 100 by loading appropriate instructions therein. A significant advantage of using the synchronous network data processor of the present invention in the data modification system 520 is that the system may be re-programmed for different types of protocols as well as to perform different tasks.

[00050] Application of synchronous network data processor of the present invention is not limited to data modification systems. FIG. 6 is a block diagram illustrating a general network data processing system 600 implemented with synchronous network data processors of the present invention. As shown, the general network data processing system 600 includes four synchronous network data processor 100 interconnected by an interconnect fabric 670. Also interconnected by the interconnect fabric 670 are a FIFO module 610, a RAM module 620, a CAM module 630, I/O modules 640, a RX data path 650, and a TX data path 660. According to the present invention, the RX data path 650 is a inbound serial-to-parallel interface, and the TX data path module 660 is an outbound parallel-to-serial interface. The I/O modules 640 are for coupling the network data processing system 600 to data analyzers and other network data processing systems.

[heading-00051] Branch Control and Conditional Execution of Instructions by the Processor

[00052] According to the present invention, the processor 100 may execute every instruction conditionally. Further, every instruction may specify up to two different conditional relative branches, each with its own destination address. In the present embodiment, conditional execution control fields are shared with the control files for the second branch. If conditional execution is used, the second branch is disabled or use the same condition.

[00053] The bits that are examined when determining whether to conditionally branch, execute, or trap are referred to as the “flags,” and are held in the flags register of the execution control unit 130. There are six flags in total, which include the five flags generated by data compare instructions (DC4-DC0) and one programmable “P” flag generated by the peripheral unit 140. The “P” flag is selectable from one of several sources including counter wrap flags, the external memory interface ready signal, and the carry output of the data modify unit 120. The format of the flags register is shown below in Table 1.

[00002]

TABLE 1
Bit	39-6	5	4	3	2	1	0
Name	Reserved	P	DC4	DC3	DC2	DC1	DC0

[00054] A branch or execute condition is specified by three fields: Mask, Match, and True/False. Mask and Match are the same width as the flags register (40-bit), and True/False is a single bit. The execution control unit 130 evaluates the condition by logically ANDing the flags with Mask, and then comparing this result to Match. If the comparison result (True if equal, False if not equal) is the same as the True/False bit, the condition is considered satisfied and the branch or conditional execution takes place.

[00055] The branch conditions and the execution conditions of an instruction are defined by its common control fields. The syntax and operations of the common control fields are described below in Table 2.

[00002]

TABLE 2
Common Control Field	Function
br(mask1, match1, tf1,	Conditional branch control. The two
addr1,mask2, match2,	conditions are evaluated as described above.
tf2, addr2)	If condition 1 is satisfied, a branch is taken
	to addr1. Otherwise, if condition 2 is satisfied,
	a branch is taken to addr2. Otherwise, control
	transfers to the following instruction. Legal
	values are any 6-bit constant for the mask and
	match fields, T or F for the tf field, and a 12-bit
	constant or a label (string) for addr1 and addr2.
	The second branch condition and address may
	be omitted if not used. If no branch control
	field is given at all, control falls through to the
	next instruction.
	The second branch condition is shared with the
	execute condition; therefore if both conditional
	execution and the second branch are used, their
	conditions must be the same.
	When the second branch is not specified, the
	assembler encodes either an always-satisfied
	condition or the execute condition specified by
	the exec_on field. In each case, the second
	branch target is the next instruction. When
	neither branch is specified, the assembler
	encodes always-satisfied conditions for both
	branches, and the next instruction for both
	branch targets.
	Address 0xF80 has a special function when
	used as the branch 2 address. It causes a branch
	to the program counter (PC) saved by a
	previous subroutine call and is used to return
	from the subroutine. The branch 2
	mask/match/tf controls still function
	normally, allowing conditional returns.
exec_on(mask, match,	Conditional execution control. The condition is
tf)	evaluated as described above. If it is satisfied,
	the instruction executes; otherwise it does not
	execute (is treated as a no-op). All common
	control fields with the exception of bg_run are
	active regardless of whether the instruction
	executes or not.
	The execute condition is shared with the second
	branch condition (see above).
	If no conditional execution control field is
	specified, the instruction executes.
save_pc(ctrl)	Save the current program counter (PC). Used to
	implement subroutine calls. The ctrl field
	defines how the PC is saved:
	0: don't save PC
	1: store current address + 1 to saved_PC
	(subroutine returns to next instruction)
	2: store branch address 2 to saved_PC
	(subroutine returns to branch address 2. Branch
	2 still behaves normally).
	Others: reserved
bg_run	When present, causes the instruction to run in
	the background (i.e., execute continuously until
	interrupted by the execution of another
	instruction of the same type). If not present, the
	instruction executes for the present instruction
	cycle only. Once an instruction is running in the
	background, it is no longer subject to any
	execution condition it may have been issued with.
	An interruption of a background-running
	instruction occurs only if the interrupting
	instruction actually executes; i.e., its execution
	condition is satisfied.
	While background run mode is only supported for
	data compare instructions in one preferred
	embodiment, in an alternate embodiment
	background run mode is supported for both data
	compare and data modify instructions..

[00056] Some pseudo-control operations that can be implemented using the execution control fields are shown below in Table 3. Appropriate macros for these can be defined in a standard header file. Software written using the pseudo-control codes may be translated into the processor-specific common control fields using a pre-processor.

[00002]

TABLE 3
Pseudo-control	Operation	Implementation
jmp	Jump to address	br(0, 0, T, addr)
	(unconditionally)
jsr	Jump to subroutine	br(0, 0, T, subr) save_pc(1)
	(unconditionally)
jsrr	Jump to subroutine;	br(0, 0, T, subr, 0, 0, T,
	return to specified	retaddr) save_pc(2)
	address (unconditionally)
ret	Return from subroutine	br(0, 0, F, 0, 0, 0, T, 0xF80)
	(unconditionally)
bcs	Branch if carry	br(0x20, 0x20, T, addr)
	set
	(P = DM carry flag)
bcc	Branch if carry clear	br(0x20, 0x20, F, addr)
	(P = DM carry flag)
loop	Jump if still in loop	br(0x20, 0x20, F, addr)
	(P = counter wrap flag)
exec_loopend	Execute on end of loop	exec_on(0x20, 0x20, T)
	(P = counter wrap flag)
br_c8t/f	Branch on 1-5 byte	br(0x01, 0x01, T/F, addr)
br_c16t/f	comparison true/false	br(0x03, 0x03, T/F, addr)
br_c24t/f		br(0x07, 0x07, T/F, addr)
br_c32t/f		br(0x0f, 0x0f, T/F, addr)
br_c40t/f		br(0x1f, 0x1f, T/F, addr)

Data Compare Instructions Executable by the Processor

[00058] Data compare instructions perform a three operand (data, mask, and match) comparison operation of up to 40 bits at a time. The sources of the data to be compared can be the input pipeline unit 150, the register bank 170, the peripheral unit 140, and/or the execution control unit 130. According to the present embodiment, the input pipelines are fed from the processor's input busses IN0 and IN1, and the pipeline stage read by the compare instruction can be selected on the fly by the currently executed instruction.

[00059] Data compare instructions are carried out by the data compare unit 110 which includes five independent 8-bit comparators 330, each of which has selectable inputs for its data, mask, and match values. Each comparator 330 updates its own comparison result flag, which can be used as part of a conditional branch or execution condition. This flag can either be set to the comparison result, or to the logical AND, OR, or XOR of the comparison result and current flag value.

[00060] The syntax of a data compare instruction executable by the processor 100 is:

• • [00061] compare data, mask, match [data compare specific control fields] [Common control fields];

[00062] The C-equivalent logical operation performed by a data compare instruction is described below in Table 4.

[00002]

TABLE 4
for (comp = 0; comp < 5; comp++) // do all 5 comparators
{
	// perform 8-bit mask/match comparison
	if ((data[comp] & mask[comp]) == match[comp]) result[comp] = 1;
	else
	result[comp] = 0;
	// update comparison result flag (SET, AND, OR, or XOR)
	switch(update_mode)
	{
	case SET: flag[comp] = result[comp]; break;
	case AND: flag[comp] &= result[comp]; break;
	case OR: flag[comp] |= result[comp]; break;
	case XOR: flag[comp] ^= result[comp]; break;
	}
}

[00063] The compare flags are updated one clock after the instruction executes, and therefore may be used in the following instruction. Note that if a branch or execute condition is used in the same instruction as the compare, the flag values are those that existed BEFORE the compare instruction executes.

[00064] Although data for the data compare instructions may come from numerous sources and may be specified on the fly by the currently executed instruction, there are a few limitations. Table 5 below shows the legal values for the three comparator source fields.

[00002]

TABLE 5
	Input	Input	Register	Peripheral	Immediate
Source	Pipeline A	Pipeline B	Bank	Data	data
Mnemonic	ina[n]	inb[n]	r[n]	periph[n]	[value]
data	YES	YES	YES	YES	NO
mask	YES	YES	YES	YES	YES
match	YES	YES	YES	YES	YES

[00065] The comparator source fields are also subject to the following restrictions:

• • [00066] (A) If an input pipe is used for the mask source, it may not be the same as that used for the data.
  • [00067] (B) If the same input pipe is used in more than one source, the pipe word number (n) (i.e., the point at which the input pipe is tapped) must be the same in both uses.
  • [00068] (C) If a register or peripheral is used in more than one source, the number (n) must be the same in both uses. The parameters of r and periph are the register or internal peripheral number. Legal values for these parameters are 0-15.

[00069] The immediate data value is a 40-bit constant specified in the instruction. Two different values may be specified for the mask and match fields.

[00070] The parameters of the input pipelines specify the stage in the input pipelines from which data are accessed. For example, an instruction including the field “ina[4]” indicates using the word in the fourth stage of input pipeline INPIPE_A. Legal values for these parameters are 0-15. The input bus feeding each pipeline and the pipeline enables are set by fields in the control registers 144.

[00071] Table 6 shows the type-specific control fields that are supported by data compare instructions.

[00002]

TABLE 6
Control Field	Function
byte_sel(c4, c3, c2, c1, c0)	Selects the byte number of the 40-bit source
	word to apply to each comparator's data
	input. This field is only valid when using an
	input pipe as the data source, and has no
	effect otherwise. Legal values for c4-c0 are
	4-0 (byte 4 is the msb of the 40 bit input
	word, and byte 0 is the lsb). For the mask
	and match fields, or for non input pipe data
	sources, the byte number of the input word is
	the same as the comparator number; e.g., the
	third comparator uses byte 3 of the mask
	word.
	If this field is not given, the byte selects
	default to the previous values given, or
	4,3,2,1,0 if no previous values were given.
update_mode()	Used in conjunction with the
	FLAG_UPD_CFG field of the control
	registers to set the flag update mode for all
	comparators. The truth table for
	FLAG_UPD_CFG can be found in
	Appendix-A. Legal values for mode are 0
	and 1. If this field is not given, the mode
	defaults to the previous value given, or “0”if
	no previous value was given.

[00072] Data compare instructions may be run in background mode by applying the bg_run common control field to the instruction. In background run mode, a data compare instruction runs continuously, updating the compare flags, until the next compare instruction executes. Normal conditional branching and execution may be performed based on the flags generated by the background-running instruction.

[00073] Instruction examples illustrating both legal and illegal uses of the data compare instructions are illustrated below in Table 7.

[00002]

TABLE 7
Code Examples	Description
compare ina[0], 0xffffffffff,	40-bit straight comparison of the
0x123456789a byte_sel(4, 3, 2, 1, 0)	word in the first stage of input
update_mode(SET);	pipe A to a constant. The word
	was equal to 0x123456789a if
	all five comparator flags are true
	after the instruction executes.
compare ina[0], 0xfffffffff0,	Same as above but with the
0x1234567890;	lower 4 bits masked off (ignored
	in the comparison). The control
	fields default to the previous
	values used if not specified.
compare ina[0], r[2], inb[8];	Compare the first stage of input
	pipe A with the ninth stage of
	input pipe B, after masking the
	data in pipe B with data in
	r[2].
compare inb[12], r[8], periph[4];	Compare Pipe B stage 12 with
	peripheral 4, using mask in r[8].
compare ina[1], r[2], inb[0];	Compare a word in the input
	pipeline to the word received
	one clock ago. Assumes Pipes A
	and B both have the same source
	bus (in0 or in1). (The pipe
	source busses are set by bits in
	CTRL_REG).
compare inb[4], ina[0], ina[0];	See if all the bits set in the
	first stage of input pipe A are
	also set in the fifth stage
	of input pipe B.
compare inb[4], r[13], r[13];	Same as above, but using
	registers.
compare ina[0], 0x0fffffffff, SOFi3	Background run example: start
bg_run;	up the compare unit looking for
	SOFi3 in the input data stream,
	and then let other instructions
	execute. “SOFi3” is a C-style
	definition of the numeric value
	of a “start of frame” ordered
	set.
compare ina[3], 0xffffffffff,	Byte_sel example: Compare
0x123456789a byte_sel(2, 2, 2, 2, 2);	input pipe A stage 3 byte 2 with
	five different values (0x12,
	0x34, 0x56, 0x78, and 0x9a).
	The five flags hold the results of
	the five comparisons.
compare ina[3], 0x73ff3f7ff8,	Same as above, but with five
0x123456789a, byte_sel(2, 2, 2, 2, 2);	different 8-bit masks for the
	comparisons.
compare ina[3], 0xffffffffff,	Compare the 16-bit word in Pipe
0xaa12345678 byte_sel(4, 1, 0, 1, 0);	A stage 3 bytes 1-0 to two
	different values (0x1234
	and 0x5678), and byte 4 to 0xaa.
compare ina[7], 0xffffffffff; WORD_A	Update_mode example: if
update_mode(SET);	WORD_A, WORD_B, and
compare ina[7], 0xffffffffff, WORD_B	WORD_C are received in
update_mode(AND);	succession. The comparison
compare ina[7], 0xffffffffff, WORD_C	flags are set on the first
update_mode(AND);	comparison, then ANDed with
	the current flags. The pipes
	advance 1 stage per instruction,
	so reading the same pipe word
	on successive instructions has
	the effect of reading successive
	input words. This could
	alternatively be done with
	conditional branching. If
	the five flags are true after
	execution of the third compare
	instruction, the three specified
	words have been received in
	succession.
compare ina[1], 0xff, ina[2];	Examples of illegal usages.
compare r[2], 0xff, r[4];
compare ina[3], periph[2], periph[3];
compare inb[0], inb[0], 0xff;
compare 0xff, ina[1], r[2];

Data Modify Instructions Executable by the Processor

[00075] A description of the data modify instructions executable by the processor 100 of the preferred embodiment follows. Data modify instructions perform arithmetic and logic operations using up to four operands and three operation codes (opcodes), and store the results to one or more write destinations. The instructions use the same sources as data compare instructions: the input pipeline unit 150, the register bank 170, the peripheral unit 140, or immediate data from the execution control unit 130 as defined in the currently executed instruction.

[00076] Data modify instructions are performed by the data modify unit 120, which includes three two-operand arithmetic logic units ALU1-ALU3. ALU1 and ALU2 have their first operand (X) selectable from among the input pipeline unit 150, the register bank 170, or the peripheral unit 140. Their second operand (Y) is an immediate data value provided by the execution control unit 130 and specified in the currently executed instruction. The operands of ALU3 are the outputs of ALU1 and ALU2. ALU3 also generates a carry flag, which can be selected as a source flag for conditional branching or execution.

[00077] An optional ALU-bypass mode is available to the instructions. In the ALU-bypass mode, the results from ALU1 and ALU2 are provided to the output busses (OUT0 and OUT1), bypassing the ALU3. This mode allows both busses to be updated with one instruction.

[00078] The data modify unit 120 also supports an internal pass-through mode where data from the input pipeline unit 150 are provided directly to the output busses OUT0 and OUT1. In this pass-through mode, “default” data can be supplied to the output busses whenever data modify instructions are not executing. The pass-through operation is configured by fields in the control registers 144 of the peripheral unit 140. The opcodes supported by data modify instructions are shown below in Table 8. Operations are shown as C equivalents.

[00002]

TABLE 8
			Supported by
Opcode	Operation	Description	ALU's
and	X & Y	Bitwise logical AND of	1, 2, 3
		X and Y
or	X | Y	Bitwise logical OR of	1, 2, 3
		X and Y
xor	X {circumflex over ( )} Y	Bitwise logical XOR of	1, 2, 3
		X and Y
nor	˜(X | Y)	Bitwise logical NOR of	1, 2
		X and Y
ror8a	ror(X, 8) & Y	Rotate X right 8 bits,	1
		AND with Y
ror1a	ror(X, 1) & Y	Rotate X right 1 bit,	1
		AND with Y
rol8a	rol(X, 8) & Y	Rotate X left 8 bits,	2
		AND with Y
rol2a	rol(X, 2) & Y	Rotate X left 2 bits,	2
		AND with Y
add	X + Y	Sum of X and Y	3
addp1	X + Y + 1	Sum of X and Y, plus 1	3
pass_imm	Y	Pass Y (immediate data)	1, 2
		to result
tbd12	tbd	tbd	1, 2
tbd3_a	tbd	tbd	3
tbd3_b	tbd	tbd	3
tbd3_c	tbd	tbd	3

[00079] Table 9 below shows pseudo-opcodes that may be implemented using the native opcodes. Appropriate macros for these can be defined in a standard header file.

[00002]

TABLE 9
Pseudo-
op	Operation	Description	Implementation	Note
nop	(none)	No operation	null = or(0, 0)
not	˜A	Bitwise	xor(A, 0xffffffffff)
		inverse of A
inc	A + 1	Increment A	add(A, 1) or addp1(A, 0)
dec	A − 1	Decrement A	add(A, 0xffffffffff)
sub	A − B	Difference of	addp1(A, not(B))
		A and B
subi	A − B	Difference of	addp1(A, ˜B)
		A and B, B
		constant
neg	−A	Negate A	addp1(0, not(A))
adc	A + C	Sum of A	add(A, 1)	1
		and carry	exec_on(0x20,0x20,T)
sec	C = 1	Carry = 1	add(1, 0xffffffffff)
clc	C = 0	Carry = 0	add(0, 0)
testge	A >= B	Carry = 1 if	null = sub(A, B)
		A >= B, 0 if
		A < B
testnz	A != 0	Carry = 1 if	null = add(A,
		A != 0, 0 if	0xffffffffff)
		A >= 0
testneg	A < 0	Carry = 1 if	null = add(A,
		A < 0, 0 if	0x8000000000)
		A == 0
ror8	ror(A, 8)	Rotate A	ror8a(A, 0xffffffffff)
		right 8
		bits
ro18	ro1(A, 8)	Rotate A	ro18a(A, 0xffffffffff)
		left 8
		bits
shr	A >> 1	Shift A right	ror1a(A, 0xefffffffff)
		1 bit
sh1	A << 1	Shift A left	add(A, A)
		1 bit
shr8	A >> 8	Shift A right	ror8a(A, 0x00ffffffff)
		8 bits
sh18	A << 8	Shift A left	ro18a(A, 0xffffffff00)
		8 bits
shrn	A >> N	Shift A right	(Various)	2
		N bits (N =
		1..39)
shln	A << N	Shift A left	(Various)	2
		N bits (N =
		1..39)
bset	bset(A, N)	Set bit N in	or(A, 1 << N)
		A
bclr	bclr(A, N)	Clear bit N	and(A, ˜(1 << N))
		in A
bswap01	bswap(0,1)	Swap bytes 0	or(ror8a(A,
		and 1 in A,	0x00000000ff),
		zero others	rol8a(A, 0x000000ff00))
bswap12	bswap(1,2)	Swap bytes 1	or(ror8a(A,
		and 2 in A,	0x000000ff00),
		zero others	rol8a(A, 0x0000ff0000))
bswap23	bswap(2,3)	Swap bytes 2	or(ror8a(A,
		and 3 in A,	0x0000ff0000),
		zero others	rol8a(A, 0x00ff000000))
bswap34	bswap(3,4)	Swap bytes 3	or(ror8a(A,
		and 4 in A,	0x00ff000000),
		zero others	rol8a(A, 0xff00000000))
Notes:
(1) Assumes P flag is programmed to be the ALU3 carry flag. See the PERIPH_CTRL register.
(2) Can be implemented with multi-instruction macros using ror1a, ror8a, ro12a, and ro18a opcodes. Worst case N requires 5 instructions.

[00080] Data modify instructions write their results to one or more of the following write destinatins: either of the two output bussed OUT0 and OUT1, the register bank 170, or the peripheral unit 140.

[00081] The syntax of the data modify instructions in normal mode is:

• • [00082] dest1 [,dest2 . . . ]=op3(op1(src1, imm1), op2(src2, imm2)) [Common control fields];

[00083] ALU3 bypass mode is specified by assigning one or more of the output busses to the ALU1 or ALU2 results, using the following syntax.

• • [00084] dest1 [,dst2 . . . ]=op3(out0=op1(src1, imm1), op2(src2, imm2)) [Common control fields];
  • [00085] dest1 [,dest2 . . . ]=op3(op1(src1, imm1), out1=op2(src2, imm2)) [Common control fields];
  • [00086] dest1 [,dest2 . . . ]=op3(out0=op1(src1, imm1), out1=op2(src2, imm2)) [Common control fields];

[00087] The first syntax places out0 in bypass mode. The second syntax places out1 in bypass mode, and the third places both outputs in bypass mode. When an output is in bypass mode, it is illegal to also use it as an ALU3 destination.

[00088] The operation codes op1-op3 are for ALUs 420a-420c, respectively; src1 and src2 are the selectable source fields for ALU 420a and ALU 420b, and imm1 and imm2 are the two 40-bit immediate data values. The C-equivalent logic operation performed by a data modify instruction is illustrated below in Table 10.

[00002]

	TABLE 10
	result1 = alu12_operation(op1, src1, imm1);
	result2 = alu12_operation(op2, src2, imm2);
	if (out0_bypass)
	out0 = result1;
	if (out1_bypass)
	out1 = result2;
	dest(s) = alu3_operation(op3, result1, result2);

[00089] Additionally, the ALU3 carry flag is updated if the ALU3 opcode is “add” or “addp1” (other opcodes and DC instructions do not change the carry flag value). The carry is set if the addition overflowed, and cleared otherwise. In addition to arithmetic operations, the carry flag (not shown) can be used as a general-purpose branch and execute control flag.

[00090] Table 11 below shows the legal sources for the source (src1 and src2) and destination (dest) fields of a data modify instruction. Note that null can be specified for dest, in which case the ALU3 result is ignored. The immediate data operands (imm1 and imm2) are 40-bit constants specified in the instruction.

[00002]

TABLE 11
Source/	Input 0	Input 1	Register	Peripheral	Output	Output
Dest	Pipeline	Pipeline	Bank	Data	Bus	Bus	None
Mnemonic	in0[n]	in1[n]	r[n]	periph[n]	out0	out1	null
src1	YES	YES	YES	YES	NO	NO	NO
src2	NO	YES	YES	NO	NO	NO	NO
dest	NO	NO	YES	YES	YES	YES	YES

[00091] The parameters of r and periph are the register or internal peripheral number. Legal values for these parameters are 0-15.

[00092] The parameters of in0 and in1 are the word in the input pipeline register to operate on. For example, in0[4] means use the word in stage 4 of the input 0 pipeline. Legal values for these parameters are 0-15.

[00093] In the present embodiment, the source and destination fields are subject to the following additional restrictions:

[00094] (A) If the same input pipe is used in more than one source, the pipe word number (n) must be the same in both uses.

• • [00095] (B) If two registers are used as sources and a register is also used as a destination, the register number (n) of one of the source registers must be the same as that of the destination register.
  • [00096] (C) If a peripheral is used in more than one source, the number (n) must be the same in both uses.
  • [00097] (D) If both a register and peripheral are used as destinations, the number (n) must be the same in both uses.
  • [00098] (E) No more than one register may be used as a destination.
  • [00099] (F) No more than one peripheral may be used as a destination.

[00100] Table 12 below illustrates some exemplary usages of the data modify instructions.

[00002]

TABLE 12
Code Examples	Description
out0 = in0[0];	Pass-through data.
out1 = r[4];	Output data from register.
out0 = 0x08BCB51717;	Send an SOF (Start of Frame).
r[0] = 0x12345678;	Initialize register to constant.
r[1] = r[0 ];	Move register to register.
r[2] = periph[3];	Move peripheral value to
	register (save DC flags).
periph[3] = r[2];	Move register to peripheral.
r[3] = in0[1];	Move input value to register.
periph[11] = 0xaa;	Store constant to peripheral.
r[0] = r[0];	No operation.
r[0] = add(r[0], r[1]);	Add register to register.
out1, r[6] = 0x0123456789;	set output and register to 40
	bit constant
out0, out1, r[12] = periph[3];	set both outputs and register
	to peripheral value
out0, out1, r[5], periph[5] = in1[3];	Multiple destinations.
r[0] = or(out0 = 1, out1 = 2)	ALU-3 bypass mode.
null = or(out0 = 1, out1 = 2)	ALU-3 results ignored.
out0 = or(r[2], periph[3]);	Logical OR of register and
	peripheral value
out0 = xor(in0[0], 1);	Toggle bit 0 of input, send
	to output bus 0
r[3] =and(in1[6], 0xffff);	Store lower 16 bits of input
	to r[3]
r[7] =add(r[7], 1);	increment r[7]
out0 = or(and(in1[4], 0xffffffO0), 0x8b);	output = input with byte 0
	changed to 0x8b
out0, out1, r[3], periph[3] =	Example of complex data
addp1(xor(in0[8], 0x123456789a),	modify instruction.
or(periph[2], 0xfedcba9876));
r[3], periph[3] =addp1(out0 = xor(in0[8],	With ALU3 bypass mode on
0x123456789a), out1 = or(periph[2],	both outputs
0xfedcba9876));
r[3], periph[3], out1 = addp1(out0 =	With ALU3 bypass mode on
xor(in0[8], 0x123456789a), or(periph[2],	OUT0 only
0xfedcba9876));
r[3], periph[3], out0 = addp1(xor(in0[8],	With ALU3 bypass mode on
0x123456789a), out1 = or(periph[2],	OUT1 only
0xfedcba9876));
out0 = or(in0[1], in0[2]);	Examples of illegal usage
r[0]= and(r[1], r[2]);
r[0] = add(periph[0], periph[1]);
r[0], periph[1]= 2;
r[0], r[1]= 0;
periph[0], periph[1]= r[6];

Peripheral Unit and Control Registers

[00102] The peripheral unit 140 is accessed via a set of registers referenced by the instructions as periph[n]. The peripheral unit 140 is divided into a number of subunits, which are described in more detail below. Table 13 below shows the address map of the subunits and registers in the peripheral unit.

[00002]

TABLE 13
				Read/
Register Name	Address	Description	Subunit	Write
EXT_WR_DATA	periph[0]	External Memory	External	W
		Interface write	Memory
		data with normal	Interface
		addressing	Unit
EXT_RD_DATA	periph[0]	External Memory	External	R
		Interface read	Memory
		data with normal	Interface
		addressing	Unit
MAILBOX_W	periph[1]	Mailbox Register	Local	W
		to host	Interface
			Unit
MAILBOX_R	periph[1]	Mailbox Register	Local	R
		from host	Interface
			Unit
CTR_32	periph[3]	Counter 3 (upper	Counter	R
		20) and Counter 2	Unit
		(lower 20 bits)
CTR_INC	periph[3]	Counter	Counter	W
		Increment	Unit
		register
ENG_CTRL	periph[4]	Control	[Global]	W
		Register
TRAP_CTRL	periph[5]	Trap Control	Trap Unit	W
		Register
CTR_DATA	periph[6]	Counter Data	Counter	W
		register	Unit
PERIPH_CTRL	periph[7]	Peripheral	[Global]	W
		Control
		register
EXT_WR_DATA_I	periph[8]	External Memory	External	W
		Interface	Memory
		write data	Interface
		with ALU2	Unit
		indexed
		addressing
EXT_RD_DATA_I	periph[8]	External Memory	External	R
		Interface	Memory
		read data	Interface
		with ALU2	Unit
		indexed
		addressing
RESERVED	others	Reserved

[00103] The format of the peripheral subunits are described in Appendix-A.

[heading-00104] Alternate Embodiments

[00105] While the present invention has been described with reference to a few specific embodiments, the description is illustrative of the invention and is not to be construed as limiting the invention. Various modifications may occur to those skilled in the art without departing from the true spirit and scope of the invention as defined by the claims below.

[00002]

APPENDIX A
Peripheral Register Formats
Field Name	Bits	Function
EXT_WR_DATA - External Memory Interface Write Data - Write Only
data	39-0	This value is written to the external
	memory interface write data bus. Writing
	this value also causes the interface chip
	select and write strobe to be asserted. The
	address presented to the external memory
	interface during the write is the
	concatenated value of Counter 3 (upper 20
	bits) and Counter2 (lower 20 bits)).
	The instruction writing the memory
	interface does not stall due to a
	deasserted interface RDY signal; instead,
	this signal can be used as part of a
	branch/execute/trap condition to provide
	software-based wait states (during which
	other useful instructions may execute). The
	write value has not necessarily been
	accepted by the external memory until it
	asserts RDY.
EXT_WR_DATA_I - External Memory Interface Write Data with
ALU2 Indexed Addressing - Write Only
data	39-0	This register functions equivalently to the
	EXT_WR_DATA register, except that the
	address presented to the external memory
	interface is Counter32 + the ALU2 result.
EXT_RD_DATA - External Memory Interface Read Data - Read Only
data	39-0	This value is read from the external memory
	interface read data bus. Reading this value
	also causes the interface chip select and
	read strobe to be asserted. The address
	presented to the external memory interface
	during the read is the concatenated value
	of Counter 3 (upper 20 bits) and Counter 2
	(lower 20 bits)
	The instruction reading the memory
	interface does not stall due to a
	deasserted interface RDY signal; instead,
	this signal can be used as part of a
	branch/execute/trap condition to provide
	software-based wait states (during which
	other useful instructions may execute). The
	read value is not necessarily valid until
	the external memory asserts RDY.
EXT_RD_DATA_I - External Memory Interface Read Data with
ALU2 Indexed Addressing - Read Only
data	39-0	This register functions equivalently to the
	EXT_RD_DATA register, except that the
	address presented to the external memory
	interface is Counter32 + the ALU2 result.
MAILBOX_W - Mailbox Register to Host - Write Only (Processor),
Read Only (Host)
res	39-32	Reserved, write 0
data	31-0	Mailbox register value. This value is
	writeable by the PicoEngine and readable by
	the host CPU for communication between the
	PicoEngine and host. The data contained in
	this register is application-dependent.
MAILBOX_R - Mailbox Register from Host - Read Only (Processor),
Write Only (Host)
res	39-32	Reserved, write 0
data	31-0	Mailbox register value. This value is
	readable by the PicoEngine and writeable by
	the host CPU for communication between the
	PicoEngine and host. The data contained in
	this register is application-dependent.
CTR_32 - Counter 32 Register - Read Only
counter3	39-20	Value of counter 3, also used for external
		memory address high bits.
counter2	19-0	Value of counter 2, also used for external
		memory address low bits.
CTR_INC - Counter Increment Register - Write Only
x	39-0	Writing this register increments any
	counter programmed to increment on a write
	to CTR_INC (as determined by the
	ctr*_inc_on_wr bits in the PERIPH_CTRL
	register). The value written is
	irrelevant.
CTR_DATA - Counter Data Register - Write Only
ctr_31	39-20	This data is written to counters 3 and 1
		when those counters are enabled by the
		corresponding ctr_wren bits in the
		PERIPH_CTRL register.
ctr_20	19-0	This data is written to counters 2 and 0
	when those counters are enabled by the
	corresponding ctr_wren bits in the
	PERIPH_CTRL register.
ENG_CTRL - Control Register - Write Only
res	39-38	Reserved, write 0
reg_bank—	37-36	Register bank read enable. Selects which
ren		register bank will be read when a register
	(r[0] through r[15]) is used as a source in
	Data Compare or Data Modify instructions.
	Each bank includes 16 independent
	registers. Background-running instructions
	read from the bank that was active at the
	time the background-running instruction was
	issued. [Note: Engines currently only
	support Bank 0 unless specially configured
	during hardware synthesis. Ask PG if in
	doubt].
	11: Bank 3
	10: Bank 2
	01: Bank 1
	00: Bank 0
reg_bank—	35-32	Write enable bits for the four register
wen		banks. Selects which banks will be written
	when the Data Modify unit writes a register
	(r[0] through r[15]). Each bank includes 16
	independent registers. More than one bank
	may be written simultaneously. [Note:
	Engines currently only support Bank 0
	unless specially configured during hardware
	synthesis. Ask PG if in doubt].
	1xxx: Enable bank 3 for write; 0xxx:
	disable
	x1xx: Enable bank 2 for write; x0xx:
	disable
	xx1x: Enable bank 1 for write; xx0x:
	disable
	xxx1: Enable bank 0 for write; xxx0:
	disable
out1_en	31	Output bus 1 update enable. When this bit
		is 1, the output bus is in passthrough mode
		and passes data from its default source
		whenever the bus is not being written by a
		Data Modify instruction. When 0, the bus
		holds its previous value.
out0_en	30	Same as above, for output bus 0.
out1_src	29	Selects the default source for output bus
		1. The data from this source is passed to
		the output bus whenever a Data Modify
		instruction isn't updating the bus, and the
		bus update enable (out1_en) is 1. The
		values for src are:
		0: input bus 0 passthrough pipeline
		1: input bus 1 passthrough pipeline
		The number of clocks of input to output
		delay is set by the p1_word_sel field.
out0_src	28	Same as above, for output bus 0.
p1_word—	27-24	Word select for the in1 to output bus
sel		passthrough pipeline. This gives the number
		of clocks (equal to p1_word_sel + 2) of
		delay between input bus 1 and the output
		bus in passthrough mode. An output bus is
		in passthrough mode whenever it isn't being
		updated by a DM instruction, and its out_en
		field is 1.
p0_word—	23-20	Same functionality as above, for the in0 to
sel		output bus passthrough pipeline.
flag_upd—	19	DC instruction compare flag update control.
cfg		Used in conjunction with the DC control
		field flag_update() to set the compare flag
		update mode as follows:
	flag_upd_cfg	update	Update mode
	0	0	SET
	0	1	AND
	1	0	OR
	1	1	XOR
comp_mode	18-14	Selects the comparator mode (0 = equality,
		1 = magnitude) for each DC comparator. In
		equality mode, the comparator result is 1
		if (data & mask) == match, otherwise 0. In
		magnitude mode, the result is 1 if (data &
		mask) >= match, otherwise 0.
		[Magnitude mode issues and description]
pb_en	13	Enable for Data Compare input pipeline B.
		0: disable pipeline (does not advance)
		1: enable pipeline (advances 1 word per
		instruction)
pb_src	12-8	Source bus for Data Compare input pipeline
		B (one bit per input bus byte).
		0: input bus 0
		1: input bus 1
res	7-6	Reserved, write 0
pa_en	5	Enable for Data Compare input pipeline A.
		0: disable pipeline (does not advance)
		1: enable pipeline (advances 1 word per
		instruction)
pa_src	4-0	Source bus for Data Compare input pipeline
		A (one bit per input bus byte).
		0: input bus 0
		1: input bus 1
TRAP_CTRL—Trap Control Register—Write Only
res	39-32	Reserved, write 0
trap_relative	31	Trap relative address enable. When 1,
		trap_addr is treated as a sign-extended
		relative address from the current PC; a
		trap causes control to transfer to the PC +
		trap_addr. When 0, trap_addr is treated as
		an absolute address; a trap causes control
		to transfer to trap_addr.
trap_restore	30	Trap restore. When 1, enables restoring the
		state of the trap_en bit after a return
		from the trap routine. Otherwise, trap_en
		remains disabled after the return from the
		trap routine.
trap_en	29	Trap enable. enables traps when 1, disables
		them when 0. When the trap is enabled and
		its match/mask/tf condition is satisfied,
		control transfers to the target address
		specified by the trap_addr and
		trap_relative fields.
		Trap_en is cleared upon entry to the trap
		routine, thus disabling further traps. If
		trap_restore is set, the bit will be
		restored to its value before the trap upon
		return from the trap routine (which occurs
		via a branch to the saved PC) . However, if
		software writes this bit before the trap
		routine returns, the bit written will be
		preserved upon the return.
trap_f	28	Trap on match/mask true/false. Determines
		whether trap should be taken if its
		match/mask condition is true (trap_f = 0)
		or false
		(trap_f = 1).
trap_match	27-20	Trap condition match bits. These bits
		specify the trap condition in the same
		manner as the branch/execute condition
		bits.
		bits 27-26: match bits for external
		interrupts 1-0 respectively
		bit 25: match bit for the Peripheral
		flag
		bits 24-20: match bits for Data Compare
		flags 4-0 respectively
trap_mask	19-12	Trap condition mask bits. These bits
		specify the trap condition in the same
		manner as the branch/execute condition
		bits.
		bits 19-18: mask bits for external
		interrupts 1-0 respectively
		bit 17: mask bit for the Peripheral
		flag
		bits 16-12: mask bits for Data Compare
		flags 4-0 respectively
res	11-10	Reserved, write 0
trap_addr	9-0	Trap destination address.
		Holds the target address for traps. Control
		is transferred to trap_addr (if
		trap_relative = 0) or the current PC +
		trap_addr (if trap relative = 1) when traps
		are enabled and the trap match/mask/tf
		condition is satisfied. Indirect branching
		may be implemented by writing the target
		address to this field and trapping on an
		always-satisfied condition.
PERIPH_CTRL—Peripheral Control Register—Write Only
res	39	Reserved, write 0
ct_f	38	Count on match/mask true/false. Determines
		whether counting should occur if the
		match/mask condition is true (ct_f = 0) or
		false
		(ct_f = 1).
ct_mask	37-32	Count enable condition mask bits. These
		bits specify the count condition (when
		count enable on match/mask/tf is configured
		by ctr*_ie_sel) in the same manner as the
		branch/execute condition bits.
		bit 37: mask bit for the Peripheral flag
		bits 36-32: mask bits for Data Compare
		flags 4-0 respectively
pf_en_hi	31-30	(See pf_en)
ct_match	29-24	Count enable condition match bits. These
		bits specify the count condition (when
		count enable on match/mask/tf is configured
		by ctr*_ie_sel) in the same manner as the
		branch/execute condition bits.
		bit 29: match bit for the Peripheral
		flag
		bits 28-24: match bits for Data Compare
		flags 4-0 respectively
ctr_wren	23-20	Counter write enables. These bits enable
		one or more of the counters for writing
		when the CTR_DATA register is written.
		bit 23: 1 = enable write to counter 3, 0 =
		disable
		bit 22: 1 = enable write to counter 2, 0 =
		disable
		bit 21: 1 = enable write to counter 1, 0 =
		disable
		bit 20: 1 = enable write to counter 0, 0 =
		disable
pf_en	19-16	Peripheral flag enable bits, used in
		combination with pf_en_hi. Selects the
		source(s) of the Peripheral flag (the P bit
		of the Flags register) used in branch,
		execute, trap, and count conditions. All
		sources with an enable bit of 1 are
		logically ANDed to generate the P bit;
		sources with an enable bit of 0 are
		ignored.
		pf_en_hi, pf_en, source:
		1x xxxx: Data Modify unit ALU3 carry flag
		x1 xxxx: EXT_RDY (ready flag) signal from
		External Memory Interface
		xx 1xxx: Counter 3 wrap flag; 1 when
		counter 3 wraps from 0xfffff to 0
		xx x1xx: Counter 2 wrap flag; 1 when
		counter 2 wraps from 0xfffff to 0
		xx xx1x: Counter 1 wrap flag; 1 when
		counter 1 wraps from 0xfffff to 0
		xx xxx1: Counter 0 wrap flag; 1 when
		counter 0 wraps from 0xfffff to 0
		Note: each counter wrap flag maintains its
		state until the counter is next updated,
		either by an increment or software write.
		Software writes to the CTR_DATA register
		reset the wrap flags of any counters
		written to.
ctr3_inc	15	Counter 3 increment enable on peripheral
_on_wr		register write. If this bit is 1, counter 3
		will be incremented on any write to the
		CTR_INC register as well as any conditions
		generated due to the ctr3_ie_sel bits. If
		this bit is 0 or whenever CTR_INC is not
		written, counting is controlled by the
		ctr3_ie_sel bits.
ctr3_ie_sel	14-12	Counter 3 default increment enable bits.
		Selects the condition for incrementing
		counter 3.
		111: increment when previous counter wraps
		(cascade with previous)
		110: increment always
		100: increment when counter mask/match/tf
		condition is satisfied
		000: increment on external memory interface
		read or write (memory address
		autoincrement)
		others: reserved
ctr2_inc	11	Same functionality as ctr3_inc_on_wr, for
_on_wr		counter 2.
ctr2_ie_	10-8	Same functionality as ctr3_ie_sel, for
sel		counter 2, with the following exception:
		0111: don't increment
ctrl_inc	7	Same functionality as ctr3_inc_on_wr, for
_on_wr		counter 1.
ctrl_ie—	6-4	Same functionality as ctr3_ie_sel, for
sel		counter 1.
ctr0_inc	3	Same functionality as ctr3_inc_on_wr, for
_on_wr		counter 0.
ctr0_ie—	2-0	Same functionality as ctr3_ie_sel, for
sel		counter 0, with the following exception:
		111: don't increment