JP3727640B2 - Data processing device - Google Patents

Description

  The present invention relates to a logic LSI (Large Scale Integrated circuit) integrated into a semiconductor integrated circuit having a central processing unit and a digital signal processing unit, and is applied to a data processing device such as a microcomputer and a digital signal processor that require high-speed arithmetic processing. Related to effective technology.

  A general-purpose arithmetic processing unit (Central Processing Unit (CPU)) required to control the entire system and a digital signal processor (Digital Signal Processor) equipped with a product-sum operation function required to efficiently process digital signals As an example of a microcomputer in which (DSP) is mounted on the same chip, there is one described in Non-Patent Document 1. According to this, a digital signal processing unit having a product-sum operation function can efficiently execute typical arithmetic processing of digital signal processing such as digital filtering with a small number of steps.

“Kawasaki et al.,“ SH Series with Built-in DSP Function ”, Nikkei Electronics, November 23, 1992, no. 568, pp. 99-112 "

  However, although the digital signal processing unit described in the above prior art includes a product-sum operation unit, data to be operated is handled as integer data in the same manner as the general-purpose operation processing unit. Usually, data handled in the world of digital signal processing is fixed point data or floating point data. Floating-point data is a data format in which mantissa data and exponent data are set, and has a completely different numerical system, but fixed-point data is very similar to integer data only in the position of the decimal point. . In fact, in addition and subtraction, the integer data and processing are basically the same.

  However, as shown in FIG. 1 (a), in the case of integer data in multiplication, the source data uses the lower word of the specified register for the operation, whereas in the case of fixed-point data, it is specified. Use the upper side of the selected register. As shown in FIG. 1 (b), integer data is considered to have a decimal point located to the right of the least significant bit, whereas fixed point data is typically immediately to the right of the most significant bit. This is because the part close to is more important. Therefore, in order to execute fixed-point multiplication with an integer multiplier, the source data must be shifted from the upper side to the lower side in advance. Further, as shown in FIG. 1C, alignment (digit alignment) is performed based on the decimal point position, and the calculation result is shifted by 1 bit between the two. As a result, there is a problem that an actual processing program may require a shift process to correct the discrepancy between the two.

  Furthermore, in digital signal processing, the bit length of data read from a memory or data when outputting a calculation result to the memory or the outside is often smaller than the bit precision during the calculation. Therefore, in an actual digital signal processing unit, data transfer to or from a memory or the outside is generally performed by single-precision word data (for example, 16 bits), and operations are generally performed by double precision (for example, 32 bits) or more. When data with a bit length shorter than the calculation accuracy is transferred, the operation content differs greatly between integer data and fixed-point data.

  An arithmetic unit premised on handling integer data inputs / outputs the lower side of a register holding data when transferring data of word data or byte data (8 bits) having a shorter bit length. However, an arithmetic unit that is premised on handling fixed-point data inputs and outputs the upper side of the data. The cause of this difference comes from the difference in decimal point position described above. In other words, when the bit length of the data to be transferred is shorter than the bit length of the operand to be stored, the portion closer to the decimal point is more important in terms of data accuracy and range, but for integer data, the decimal point is to the right of the least significant bit. Such a discrepancy arises because fixed-point data is usually immediately to the right of the most significant bit. As a result, there is a problem in that when a data transfer with a bit length shorter than the calculation accuracy is performed in an arithmetic unit premised on handling integer data, it is necessary to perform shift processing one by one.

  If the bit length of the data at the time of transfer is the same as the bit length of the data at the time of operation, such a problem will not occur, but an extra bus width is required to transfer redundant bits, There is a problem that the memory for storing the memory also requires an extra capacity.

  An object of the present invention is to provide a data processing apparatus such as a microcomputer or a digital signal processor that incorporates a central processing unit and a digital signal processing unit for processing fixed-point data.

  A further object of the present invention is to provide a general-purpose arithmetic processing unit necessary for controlling the entire system and a digital signal processing unit having a product-sum arithmetic function necessary for efficiently processing digital signals on the same chip. In the microcomputer and digital signal processor installed in the PC, the increase in the number of processing steps due to the difference in the data format handled by the arithmetic unit is prevented, and the digital signal processing is made highly efficient.

  Another object of the present invention is to eliminate the unnecessary shift operation associated with the correction of the bit position of the multiplication result and the data transfer, thereby speeding up the digital signal processing.

  The above and other objects and novel features of the present invention will be apparent from the description of this specification and the accompanying drawings.

  The following is a brief description of an outline of typical inventions disclosed in the present application.

  (A) The data processing device (1) includes a CPU (100) and a digital signal processing unit (104) whose operation is controlled by the CPU (100) decoding an instruction on one semiconductor substrate, The digital signal processing unit (104) includes an addition / subtraction circuit (105) for processing fixed-point data and a multiplication circuit (106) for processing fixed-point data.

  (B) The data processing device (1) includes a first processing unit (100) having a first register (103) and a first computing unit (101, 102) for computing data in the first register (103). And a second processing unit (104) having a second register (108) and a second computing unit (105, 106) for computing data in the second register (108), the first processing The unit (100) processes integer data, and the second processing unit (104) processes fixed point data.

  (C) The data processing device (104) includes a register (108) and an arithmetic unit (105, 106) for calculating data in the register (108), and data having a bit length shorter than the bit length of the register. When the first instruction to transfer the data to the register (108) from the outside of the data processing device (104) is executed, the data is packed into the upper side of the register (108) and input, and the register (108 ) Is input to the extra low-order side, and a second instruction for transferring data having a bit length shorter than the bit length of the register (108) from the register (108) to the outside of the data processing device (104). When executing, data having a required bit length is output to the outside from the upper side of the register (108).

  (D) The data processing device (1) selectively transmits an address from the central processing unit (100) including an arithmetic circuit (101) that executes arithmetic operation or logical operation, and the central processing unit (100). Connected to the first, second and third address buses (109, 110, 111), the first address bus (109) and the second address bus (110), from the central processing unit (100) Connected to the first memory (115), the first address bus (109) and the third address bus (111), and accessed by the address from the central processing unit (100). Second memory (116) and the first and second memories ( 15, 116) and the central processing unit (100) to which data is transmitted, and a first data bus (112) to which data is transmitted, and a first data bus (112) to which data is transmitted. Two data buses (113), a third data bus (114) connected to the second memory (116) to transmit data, and the first, second and third data buses (112). , 113, 114) and a digital signal processing unit (104) that operates synchronously with the central processing unit (100). The digital signal processing unit (104) is an addition / subtraction circuit (105) that processes fixed-point data. And a multiplication circuit (106) for processing fixed-point data.

  (E) The data processing apparatus includes a multiplier (106) that inputs a multiplier and a multiplicand, outputs a multiplication result of the multiplier and the multiplicand, and a shifter (107) that shifts the output of the multiplier. When multiplying integer data, the shifter outputs the output of the multiplier without shifting, and when multiplying fixed-point data, the shifter shifts the output of the multiplier to the left by 1 bit, Enter zero in the bit.

  That is, in the data transfer operation between the digital signal processing unit and the memory or the outside, when transferring data having a bit length shorter than the calculation accuracy, the data on the upper side of the register for storing the data is input / output. This can be solved by providing a function and providing a data transfer instruction as fixed-point data separately from a transfer instruction based on conventional integer data.

  When a fixed-point data transfer instruction is issued and data with a bit length shorter than that of the destination register is transferred, the data is stored in the upper side of the destination register and the extra lower bits are cleared. . Conversely, when data is output from the source register, data of the required number of bits is also output from the higher order of the source register. As a result, it is not necessary to perform an extra shift operation.

  Further, a microcomputer in which a general-purpose arithmetic processing unit necessary for controlling the entire system and a digital signal processing unit having a product-sum arithmetic function necessary for efficiently processing a digital signal are mounted on the same chip. In a digital signal processor, the digital signal processing unit may be an arithmetic unit that handles fixed-point data, and an instruction for executing fixed-point data arithmetic may be provided separately from a conventional integer arithmetic instruction.

  That is, when a fixed-point multiplication instruction is issued in an arithmetic circuit that performs a multiplication operation, source data is output from the upper side of the register, and leftward is arithmetically left by one bit with respect to the output of the conventional integer data multiplication circuit. Control is made to store the data shifted to the designated destination register.

  The effects obtained by the representative ones of the inventions disclosed in the present application will be briefly described as follows.

  That is, a microcomputer in which a general-purpose arithmetic processing unit necessary for controlling the entire system and a digital signal processing unit having a product-sum arithmetic function necessary for efficiently processing digital signals are mounted on the same chip. In a digital signal processor, since the digital signal processing unit can handle fixed point data, more complicated digital signal processing can be performed.

  Further, in the data transfer operation between the digital signal processing unit and the memory or the outside, when transferring data having a bit length shorter than the calculation accuracy, data on the upper side of the register for storing the data is input / output. Provided with a function, a data transfer instruction as fixed-point data is provided separately from the transfer instruction based on the conventional integer data, so that unnecessary shift operations associated with data transfer can be omitted, and high-speed performance is improved. be able to.

  By providing the digital signal processing unit with an instruction for executing fixed-point data arithmetic separately from the conventional integer arithmetic instruction, the bit position of the multiplication result is automatically corrected, and high-speed performance can be improved.

"overall structure"
FIG. 2 is an overall block diagram of a microcomputer according to an embodiment of the present invention. The microcomputer 1 shown in the figure is formed on a single semiconductor substrate such as single crystal silicon by a semiconductor integrated circuit manufacturing technique. In the figure, 100 is a general-purpose arithmetic processing unit (central processing unit: CPU) having an integer arithmetic processing function, 101 is an arithmetic logic unit (ALU) in the general-purpose arithmetic processing unit, and 102 is a first arithmetic unit in the general-purpose arithmetic processing unit. 2 is an integer arithmetic unit (PAU) that performs an address operation of 2, a register file 103 is a source or destination operand of each arithmetic unit, 104 is a digital signal processing unit (DSP) having a fixed-point data arithmetic processing function, 105 Is an arithmetic and logic unit (ALU) in the digital signal processing unit, 106 is a multiplier in the digital signal processing unit, 107 is a shifter, 108 is a register file serving as a source or destination operand of each of the arithmetic units, 109 supports the entire address space 32-bit address bus (IAB [31: 0]) 110 and 111 are 16-bit address buses (XAB [16] dedicated to accessing word data (16-bit data) supporting only a part of the address space. 1], YAB [16: 1]), 112 is a 32-bit data path (IDB [31: 0]), 113 and 114 are 16-bit data buses (XDB [15: 0], YDB [15 : 0]), 115 and 116 are on-chip memories (X memory and Y memory), and 117 is a module (I / O) for interfacing with peripheral circuits and the outside. It should be noted that other element circuits naturally included in the original data processing apparatus, that is, peripheral circuits, instruction decoding circuits, flow control circuits, and the like are omitted here because they are not directly related to the present invention. The details of the microcomputer 1 are described in the previous application (Japanese Patent Application No. 7-132906) of the present inventors.

  First, basic operations and functions of the present embodiment will be described. The microcomputer 1 supports two types of instructions, a CPU instruction and a DSP instruction. The CPU instruction is an instruction executed exclusively by the general-purpose arithmetic processing unit (CPU) 100 without operating the digital signal processing unit (DSP) 104. The DSP instruction is an instruction executed by the DSP 104 while the CPU 100 bears some processing. The DSP instructions include integer arithmetic instructions and instructions that handle fixed-point data.

  The CPU 100 fetches an instruction from the on-chip memory 115, the on-chip memory 116, or an external memory (not shown), decodes the instruction, and determines whether the instruction is a CPU instruction or a DSP instruction. If the result of decoding is a DSP instruction, the CPU 100 supplies a DSP control signal to the DSP 104. The DSP 104 decodes the DSP control signal and generates a control signal inside the DSP 104. That is, different control signals are generated for integer arithmetic instructions and instructions that handle fixed-point data.

  The general-purpose arithmetic processing unit 100 has basic functions provided in a central processing unit (CPU) that is the core of a normal one-chip microcomputer LSI. An arithmetic logic unit (ALU) 101 executes data and address arithmetic processing. An integer arithmetic unit (PAU) 102 that performs an address operation generates an address when the digital signal processing unit 104 needs to read a plurality of source data from a memory for a product-sum operation process together with an arithmetic logic unit 101. It is a computing unit. Source operand data required for each of the arithmetic units 101 and 102 is selected from the register file 103 and supplied. The calculation result is stored in the selected destination register in the register file 103.

  The address generated by the general-purpose arithmetic processing unit 100 is output to the address bus 109, 110 or 111. An address bus (IAB) 109 supports the entire address space, and accesses each peripheral circuit and external address space through on-chip memories 115 and 116 and an interface module (I / O) 117. Data to be written / read in the address area accessed by the address bus 109 is performed by a data bus (IDB) 112. The address bus (XAB) 110 accesses only the on-chip memory (X memory) 115. Data to be written / read in the address area accessed by the address bus 110 is transmitted by the data bus 113. The address bus (YAB) 111 accesses only the on-chip memory (Y memory) 116. Data to be written / read in the address area accessed by the address bus 111 is performed by a data bus (YDB) 114.

  The digital signal processing unit 104 has a function of processing fixed point data. Having the function of processing integer data does not hinder the implementation of the present invention. The arithmetic logic unit 105 performs addition / subtraction and logic operation processing. The multiplier 106 multiplies two 16-bit word data and outputs a 32-bit result. In the case of integer multiplication, the lower word (0th to 15th bits) of the source register is input as source data, and in the case of fixed-point multiplication, the upper word (16th to 31st bits) of the source register is used as source data. input. Since it is clear that 106 is a product-sum operation unit and does not hinder the implementation of the present invention, description will be made here in the case of a multiplier. The shifter 107 has a function of shifting the output of the multiplier 106 one bit to the left. Source operand data required for each of the arithmetic units 105 and 106 is selected from the register file 108 and supplied. The operation result is stored in the selected destination register in the register file 108.

  Data to be processed by the digital signal processing unit 104 is supplied to each peripheral circuit and the register file 108 from the outside through the on-chip memories 115 and 116 or the interface module 117 via the data bus 112. The processed data is output from the register file 108 via the data buses 112, 113 and 114 to each peripheral circuit and the outside through the on-chip memories 115 and 116 or the interface module 117. Data processed by the digital signal processing unit 104 can be transferred via the data buses 113 and 114, but can be transferred only between the register file 108 and the on-chip memory 115 on the data bus 113. In addition, data can be transferred only between the register file 108 and the on-chip memory 116 on the data bus 114. Data transfer using the data buses 113 and 114 can be executed in parallel because the resources are all separate. When data transfer between the register file 108 and others is executed, necessary addresses are generated by the general-purpose arithmetic processing unit 100.

  The on-chip memories 115 and 116 are mapped to different addresses. The type of memory is not particularly limited, and a ROM such as a mask ROM or a flash memory that is programmed during an LSI manufacturing process even in a random access memory (RAM) such as SRAM (Static RAM) or DRAM (Dynamic RAM). (Read Only Memory) may be used. That is, it may be a volatile memory or a nonvolatile memory. The on-chip memory 115 receives an address supply from the address buses 109 and 110 and writes / reads data through the data buses 112 and 113. On-chip memory 116 receives address supply from address buses 109 and 111 and writes / reads data via data buses 112 and 114. As a result, it is possible to perform data write / read operations in parallel in the same operation cycle as described above.

《Shifter configuration》
A detailed embodiment of the shifter 107 is shown in FIG. In the figure, 200 is an inverter, 201 is a logical product circuit, 202 is a logical sum circuit, and 203 is a control signal for controlling whether or not to perform shift in the shifter 107. The logical sum circuit 202 and the two logical product circuits 201 constitute a selection circuit. The number attached to the output of the multiplier 106 represents the bit position. The 31st bit is the most significant bit and the 0th bit is the least significant bit. Other symbols are the same as those in FIG. This embodiment of the shifter is an embodiment when the data processing apparatus supports both integer multiplication and fixed-point multiplication. Here, the multiplier 106 always performs integer multiplication. As a result, when an integer multiplication instruction is executed, the control signal 203 is in a low state, and the output result of the multiplier 106 is passed as it is. When a fixed-point multiplication instruction is being executed, the control signal 203 is in a high state, and the output result of the multiplier 106 is left-shifted by 1 bit and output. Zero is output in the 0th bit. In this way, fixed-point multiplication is realized. If the integer multiplication instruction is not supported, the shifter 107 does not need a through function and always needs to output 1-bit shift, so the control signal 203 is also unnecessary, and the shift function itself is not actually required, and is simply stored at the result storage destination. The connection should be made by shifting the bit position to the left by one bit. Therefore, the present invention is not required to have a shift circuit such as 107, and it is an important point of the present invention that at least a fixed-point multiplication function is provided in the digital signal processing unit 104.

  By providing the multiplier that performs only integer multiplication with a shift circuit having different shift functions depending on instructions, both fixed-point multiplication and integer multiplication can be executed. Therefore, high functions can be realized with a small amount of hardware, and an increase in chip area can be suppressed. Further, there is no need to execute a CPU instruction such as a shift operation after execution of multiplication.

<< Connection between DSP and data bus >>
A more detailed block diagram of the register file 108 and an example of connection to the data bus are shown in FIG. However, in this figure, only the configuration related to the connection between the data bus 112 and the register file 108 is shown for the purpose of focusing on the main points, and the connection structure with other data buses and each arithmetic unit is omitted. ing.

  In the figure, 300a, 300b, 300c, and 300d are individual registers, 301 is a local bus that connects the upper word (16th bit to 31st bit) of each register and the buffer & driver 303, and 302 is a lower level of each register. A local bus connecting the word (0th to 15th bits) and the buffer & driver 304; 303 a buffer & driver for relaying data exchange between the upper word of each register and the data bus 112; 304 Is a buffer and driver that relays the transfer of data between the lower word of each register and the data bus 112, and 305 selects either the upper word or the lower word of the data bus 112 to transfer data. Direction control signal, 306 is connected to the lower word of data bus 112 It is a signal for controlling the data transfer direction. In FIG. 4, FIG. 5 and FIG. 7, the data bus 112 is described by being divided into a lower data bus 112a and an upper data bus 112b for explanation. FIG. 5 shows a circuit diagram of the buffers & drivers 303 and 304. FIG. 6 shows the relationship between the control signal 305 (305a, 305b, 305c, 305d, 306e) of the buffer & driver 303 and the control signal 306 (306a, 306b, 306c) of the buffer & driver 304 and the handling data.

  Here, for simplification of description, 16-bit data is referred to as word data, and 32-bit data is referred to as long word data.

(1) Input / output of long word data When long word data is input via the data bus 112 (indicated as long word load in FIG. 6), the nature of the data (integer data or fixed point data) Regardless of the operation, the operation is the same. That is, when the control signal 306a is set to “1” (High level), the input buffer 505 is enabled, and the lower data bus 112a and the local bus 302 are electrically connected. Therefore, the data on the lower data bus 112a is stored in the lower word of the destination register (one of 300a to 300d) designated via the buffer & driver 304 and the local bus 302. At the same time, when the control signal 305a is set to “1” (High level), the input buffer 501 is enabled, and the upper data bus 112b and the local bus 301 are electrically connected. Therefore, the data on the upper data bus 112b is stored in the upper word of the destination register (the same register as the lower word) designated via the buffer & driver 303 and the local bus 301.

  Even when longword data is output to the data bus 112 (shown as longword store in FIG. 6), the operation is the same regardless of the nature of the data (integer data or fixed-point data). . That is, when the control signal 306b is set to “1” (High level), the output buffer 506 is enabled, and the local data bus 302 and the lower data bus 112a are electrically connected. Accordingly, the word data output from the lower word of the designated source register (one of 300a to 300d) is output to the lower data bus 112a via the local bus 302 and the buffer & driver 304. At the same time, when the control signal 305b is set to “1” (High level), the output buffer 502 is enabled, and the local data bus 301 and the upper data bus 112b are electrically connected. Accordingly, the word data output from the upper word of the designated source register (the same register as the lower word) is output to the upper data bus 112b via the local bus 301 and the buffer & driver 303.

(2) Input / output of word data In the transfer of word data, data is always transferred using the lower data bus 112a.

(I) Integer data First, the case of integer data will be described. When word data is input via the data bus 112 (indicated as integer data word load in FIG. 6), when the control signal 306a is set to “1” (High level), the input buffer 505 is When enabled, the lower data bus 112a and the local bus 302 are electrically connected. Therefore, the data on the lower data bus 112a is stored in the lower word of the destination register (one of 300a to 300d) designated via the buffer & driver 304 and the local bus 302. At the same time, when the control signal 305e is set to “1” (High level), the input buffer 507 is enabled, and the 15th bit of the lower data bus 112a and the local bus 301 are electrically connected via the sign extension circuit 510. The Therefore, the buffer & driver 303 takes in only the 15th bit data of the lower data bus 112a and performs an extended copy to 16 bits, and the higher order of the destination register (the same register as the lower word) specified via the local bus 301. Store in word. As a result, the code data of the transferred word data is copied to the upper word of the destination register.

  When word data is output to the data bus 112 (denoted as integer data word store in FIG. 6), the output buffer 506 is enabled when the control signal 306b is set to “1” (High level). Then, the local data bus 302 and the lower data bus 112a are electrically connected. Accordingly, the word data output from the lower word of the designated source register (one of 300a to 300d) is output to the lower data bus 112a via the local bus 302 and the buffer & driver 304. At this time, the buffer & driver 303 side does not operate. That is, the control signals 305a, 305b, 305c, 305d, and 305e are all “0” (Low level), and the input buffers 501, 504, and 507 and the output buffers 502 and 503 are disabled.

(Ii) Fixed-point data Next, the case of fixed-point data will be described. When word data is input via the data bus 112 (indicated as fixed-point data word load in FIG. 6), the input buffer 503 is set when the control signal 305c is set to “1” (High level). Is enabled, and the lower data bus 112a and the local bus 301 are electrically connected. Accordingly, the data on the lower data bus 112a is stored in the upper word of the destination register (one of 300a to 300d) designated via the buffer & driver 303 and the local bus 301. At the same time, when the control signal 306c is set to “1” (High level), the all-zero circuit 512 of the buffer & driver 304 generates all-zero data for 16 bits, and the destination register (the upper word and the upper word) is specified via the local bus 301. Stored in the lower word of the same register). As a result, the lower word of the destination register is automatically cleared. In order to clear the lower word, a circuit that directly clears the destination register instead of generating all zeros by the buffer & driver 304 may be provided.

  When word data is output to the data bus 112 (indicated as fixed-point data word store in FIG. 6), the output buffer 504 is enabled when the control signal 306d is set to “1” (High level). The local data bus 301 and the lower data bus 112a are electrically connected. Accordingly, the word data output from the upper word of the designated source register (one of 300a to 300d) is output to the lower data bus 112a via the local bus 301 and the buffer & driver 303. At this time, the buffer & driver 304 side does not operate. That is, the control signals 306a, 306b, and 306c are all “0” (Low level), and the input buffer 505, the output buffer 506, and the all zero circuit 512 are disabled.

  The state of the control signals 305 (305a, 305b, 305c, 305d, 305e) and 306 (306a, 306b, 306c) differs depending on the difference between the integer data transfer instruction and the fixed-point data transfer instruction. By controlling the & drivers 303 and 304, respectively, transfer from the upper word to the upper word, from the upper word to the lower word, or from the lower word to the upper word becomes possible. As a result, it is possible to shorten the calculation time without having to execute an operation such as shifting the source data to the lower side before execution of the fixed-point multiplication by the CPU instruction.

<Connection between CPU and data bus>
FIG. 7 shows a more detailed block diagram of the register file 103 of the general-purpose arithmetic processing unit 100 and a connection example with the data bus. However, this figure also shows only the configuration related to the connection between the data bus 112 and the register file 103, and the connection structure with other data buses and each arithmetic unit is omitted. In FIG. 7, similarly to FIG. 4, the data bus 112 is described by being divided into a lower data bus 112a and an upper data bus 112b. In the figure, 400a, 400b, 400c, and 400d are individual registers, 401 is a local bus that connects the upper word of each register (from the 16th bit to the 31st bit) and the buffer & driver 403, and 402 is the lower level of each register. A local bus that connects the word (0th to 15th bits) and the buffer & driver 404; 403 is a buffer & driver that relays data transfer between the upper word of each register and the upper data bus 112b; Reference numeral 404 denotes a buffer and driver that relays data transfer between the lower word of each register and the lower data bus 112a. Reference numeral 405 denotes a control signal that is connected to the upper data bus 112b and controls the data transfer direction. A signal connected to the lower data bus 112a to control the data transfer direction. It is.

  In this register file 103, all data is handled as integer data. Accordingly, the data transfer operation is basically the same, although the operation and timing or pipeline operation for integer data in the register file 108 of the digital signal processing unit 104 may be different. That is, the buffer & driver 403 includes circuits corresponding to the input buffers 501 and 507, the output buffer 502, and the sign extension circuit 510 in the buffer & driver 303. The buffer & driver 404 includes circuits corresponding to the input buffer 505 and the output buffer 506 in the buffer & driver 304. Therefore, the control signal 405 includes control signals corresponding to the control signals 305a, 306b, and 306e. The control signal 406 includes a control signal corresponding to the control signals 306a and 306b.

  Although the invention made by the present inventor has been specifically described based on the embodiments, it is needless to say that the present invention is not limited thereto and can be variously modified without departing from the gist thereof. For example, the present invention is not limited to a microcomputer, and is also applied to a digital signal processor.

  In the present embodiment, the register file 108 also supports the case where both integer data and fixed-point data transfer instructions are supported. However, in the present invention, integer data is transferred to the digital signal processing unit 104 for data transfer. It is not always necessary to support a word data transfer instruction, and at least only a transfer instruction for fixed-point data may be supported. Needless to say, the bit length of data is not necessarily 16 bits or 32 bits. Furthermore, in the present embodiment, it has been described on the assumption that only the lower word of the data bus is used when transferring word data. However, when transferring word data of fixed-point data, the upper word of the data bus is also used. If the connected word is switched depending on the nature of the data, the same function can be realized. In this case, since the buffer & driver 303 may be connected to the upper word side of the data bus at any time, it is not necessary to connect to the lower data bus 112a. Further, in this embodiment, the fixed point is between the 30th and 31st bits, and the numerical range that can be expressed is assumed to be −1.0 or more and less than +1.0. The bit may be a register that is further expanded and supported. In this case, the word data from the 16th bit to the 31st bit is transferred when the word data transfer command is executed, the guard bit portion is sign-extended when data is input, and can be ignored when data is output.

Explanatory drawing which shows the relationship between integer data and fixed-point data. 1 is an overall block diagram of a microcomputer according to an embodiment of the present invention. The detailed example explanatory drawing of the shifter 107 in one Example of this invention. 2 is a block diagram showing a more detailed configuration of a register file 108 of the digital signal processing unit 104 and an example of connection with a data bus in an embodiment of the present invention. FIG. The circuit diagram of a buffer & driver circuit. Explanatory drawing showing the relationship between the control signal 305 and the control signal 306, and handling data. FIG. 2 is a block diagram showing a more detailed configuration of a register file 103 of a general-purpose arithmetic processing unit 101 and an example of connection with a data bus in an embodiment of the present invention.

Explanation of symbols

100: General-purpose arithmetic processing unit (CPU) having an integer arithmetic processing function
101: Arithmetic logic unit (ALU) in general-purpose arithmetic processing unit
102: An integer arithmetic unit (PAU) that performs the second address calculation in the general-purpose arithmetic processing unit
103: Register file in the general-purpose arithmetic processing unit 104: Digital signal processing unit (DSP) having a fixed-point data arithmetic processing function
105: arithmetic logic unit 106 in the digital signal processing unit 106: multiplier in the digital signal processing unit 107: shifter 108 in the digital signal processing unit: register file 109 in the digital signal processing unit 109: 32-bit length address bus 110-111: 16-bit address bus 112: 32-bit data bus 113-114: 16-bit data bus 115-116: On-chip memory (X memory, Y memory)
117: Module (I / O) for interfacing with peripheral circuits and the outside
200: inverter circuit 201: logical product circuit 202: logical sum circuit 203: control signals 300a, 300b, 300c, 300d for controlling whether or not to perform shift in the shifter 107: individual registers 301 to 302 in the register file: Local bus 303 to 304: Buffer & driver 305 to 306: Data transfer control signal 112a: Lower word of data bus 112 (lower data bus)
112b: upper word of data bus 112 (upper data bus)
400a, 400b, 400c, 400d: individual registers 401-402 in the register file: local buses 403-404: buffers & drivers 405-406: control signals for data transfer 501, 503, 505, 507: input buffer 502, 504, 506: Output buffer 510: Sign extension circuit 512: All zero circuits 305a, 305b, 305c, 305d, 305e: Control signals 306a, 306b, 306c: Control signals

Claims (8)

A first processing unit having an integer arithmetic function;
A data processing apparatus controlled by the first processing unit and having a second processing unit having a fixed-point arithmetic function,
The second processing unit includes a multiplication circuit capable of performing integer arithmetic and fixed point arithmetic,
The first processing unit is capable of executing a first type instruction;
The second processing unit is capable of executing a second type instruction;
The first type instruction is an instruction executed exclusively by the first processing unit without operating the second processing unit;
The second type instruction includes a first instruction that performs an integer operation and a second instruction that performs a fixed-point operation,
The first instruction includes an integer multiplication instruction, the second instruction includes a fixed-point multiplication instruction;
The first processing unit has a function of decoding a first type instruction and a second type instruction,
The first processing unit outputs a control signal to the second processing unit when the second type instruction is decoded.
The first processing unit causes the multiplication circuit to execute integer multiplication using the control signal when the integer multiplication instruction is decoded, and uses the control signal to decode the multiplication when the fixed-point multiplication instruction is decoded. a shall be executed a fixed point multiplication circuit,
The data processing apparatus further includes a first memory and a second memory,
A first data bus connected to the first processing unit, the second processing unit, the first memory, and the second memory;
A second data bus connected to the second processing unit and the first memory;
A third data bus connected to the second processing unit and the second memory;
A data processing apparatus , wherein when the second type instruction is executed by the second processing unit, operation data is transferred through the second data bus and the third data bus . The multiplication circuit includes a shift circuit,
The first processing unit outputs, without shifting, the operation result of the multiplication circuit using the shift circuit when the integer multiplication instruction is decoded, and outputs the shift circuit when the fixed-point multiplication instruction is decoded. The data processing apparatus according to claim 1, wherein the data processing apparatus shifts and outputs the operation result of the multiplication circuit. The multiplication circuit includes a shift circuit,
The first processing unit outputs, without shifting, the operation result of the multiplication circuit using the shift circuit when the integer multiplication instruction is decoded, and outputs the shift circuit when the fixed-point multiplication instruction is decoded. 2. The data processing apparatus according to claim 1, wherein the operation result of the multiplication circuit is arithmetically shifted to the left by one bit and output.   4. The data processing apparatus according to claim 1, wherein the first and second processing units are formed on the same semiconductor substrate. A data processing apparatus comprising a central processing unit, a digital signal processing unit, and a memory,
The digital signal processing unit has a register,
The data processing device includes:
A first function of transferring fixed-point data on the memory to the register;
A second function of transferring the fixed-point data on the register to the memory;
A third function of transferring integer data on the memory to the register;
A fourth function of transferring integer data on the register to the memory;
A fixed-point arithmetic function for the data stored in the register;
An integer arithmetic function on the data stored in the register, and
The data processing device includes:
A first instruction for executing in parallel the first function and another first function for data different from the first function;
A second instruction for executing any one of the first function and the second function;
A third instruction designating the integer arithmetic function;
A fourth instruction designating the third function;
A fifth instruction designating the fourth function can be executed;
The central processing unit decodes the first to fifth instructions or other instructions and realizes all the functions of the instructions by the central processing unit itself based on the decoding result, or Controlling the digital signal processing unit in parallel with the realization of a part of the functions of the instructions,
The memory includes a first memory and a second memory,
A first data bus connected to the digital signal processing unit and a first memory; a second data bus connected to the digital signal processing unit and a second memory;
When executing the first instruction, the central processing unit applies an address on the first memory in which data related to the first function is stored to data different from the first function. And an address on the second memory in which data relating to another first function is stored, and the digital signal processing unit transmits each of the data via the first bus and the second bus. Is to be transferred,
When executing the second instruction, the central processing unit relates to an address on the first or second memory in which data related to the first function is stored or to the second function. A data processing apparatus for generating an address on the first or second memory in which data is stored . In the case of executing the first function, when the data is fixed-point data having a bit length shorter than the bit length of the register, the fixed-point data is input to the upper side of the register. Yes,
When the second function is executed and the data is fixed-point data having a bit length shorter than the bit length of the register, the fixed-point data is necessary for the fixed-point data from the upper side of the register Output a bit length of data,
In the case of executing the third function, when the data is integer data having a bit length shorter than the bit length of the register, the integer data is input to the lower side of the register.
When the fourth function is executed and the data is integer data having a bit length shorter than the bit length of the register, the bit length required for the integer data from the lower order side of the register 6. The data processing apparatus according to claim 5, wherein the data is output . The fixed-point arithmetic function includes a fixed-point multiplication function;
The integer arithmetic function includes an integer multiplication function;
7. The fixed-point multiplication function fetches data from the upper side of the register, shifts the output of the integer arithmetic function leftward arithmetically by 1 bit, and outputs the result. The data processing apparatus described . A third data bus connected to the central processing unit, the digital signal processing unit, the first memory, and the second memory;
The said 1st thru | or 5th instruction | command or another instruction | command is transferred to the said central processing unit via the said 3rd bus | bath, The any one of Claim 5 thru | or 7 characterized by the above-mentioned. Data processing equipment.

Images