Arithmetic unit for structure arithmetic

• US 5,379,387 A
• Filed: 08/02/1991
• Issued: 01/03/1995
• Est. Priority Date: 08/02/1990
• Status: Expired due to Fees

• Current Assignee: Carlstedt Elektronik Ab
• Original Assignee: Carlstedt Elektronik Ab
• Inventors: Carlstedt, L. Gunnar
• Primary Examiner(s): Lall, Parshotam S.
• Assistant Examiner(s): MOHAMED, AYNI Y
• Application Number:
  US07/739,532
• Time in Patent Office: 1,250 Days
• Field of Search: 395/375, 395/425, 395/800
• US Class Current: 712/221
• CPC Class Codes: G06F 7/483 Computations with numbers represented by a non-linear combination of denominational numbers, e.g. rational numbers, logarithmic number system or floating-point numbers G06F7/4806, G06F7/4824, G06F7/49, G06F7/491, G06F7/544 take precedence
• CPC Class Codes: G06F 8/31 Programming languages or programming paradigms
• CPC Class Codes: G06F 8/311 Functional or applicative languages; Rewrite languages
• CPC Class Codes: G06F 9/30014 with variable precision
• CPC Class Codes: G11C 11/419 Read-write [R-W] circuits
• CPC Class Codes: G11C 15/04 using semiconductor elements

Abstract

A method and an arithmetic unit for structural arithmetic processing is described. Data words are stored in several registers, each data word having a mark part and an information part. The mark part includes a mark indicating if the register in question being in use or not. The data words are arranged in lists. Each of the lists is stored in a predetermined number of the registers. The mark part of each of the words in the lists stored in the registers is marked in use indicating that one of the lists has at least a part stored in the actual register. The list having a part stored in said actual register includes a list instruction denoting which kind of list it is and the relation between the lists is apparent from the arrangement of the lists in the registers. The registers are controlled by a control device making use of the list instructions belonging to lists stored in the registers to rearrange the lists among the registers and for input/output of register content in accordance with the list instructions.

Claims

1. A method for structural arithmetic processing, including:

a) storing data words in several registers,

each data word having a mark part and an information part,

said mark part including a mark indicating if the register in question being in use or not,

b) said data words being arranged in lists,

storing each of said lists in a predetermined number of said registers,

said mark part of each of said words in said lists stored in said registers being marked in use indicating that one of said lists has at least a part stored in a register, said list having a part stored in said register including a list instruction denoting which kind of list it is,

c) arranging said lists stored in said registers as a tree of lists, of which one of the lists is a root list, where a relation between said lists is determined from an arrangement of said lists in said registers,

d) controlling said registers and making use of said list instructions belonging to lists stored in said registers to rearrange said lists among said registers and to perform input/output operations of register content in accordance with said list instructions, wherein an information regarding what kind of structural arithmetic processing to perform can be derived from a type of said root list.

2. A method according to claim 1, wherein an identifier of the stored tree of lists is stored in a separate identifier register.

3. A method according to claim 1, wherein an environment of the stored tree of lists is stored in a separate environment register.

4. A method according to claim 1, wherein a placement of a root list of said tree of lists in different registers is dependent upon a level of said tree to be stored.

5. A method according to claim 4, wherein some of said registers are arranged in a matrix of base registers including a row of main registers.

6. A method according to claim 4, wherein a tree including only one level is stored in a main register.

7. A method according to claim 4, wherein a tree including two levels is stored so as to provide its root list in said main register and its branch lists in said base registers.

8. A method according to claim 4, wherein an extra set of registers, called auxiliary register, are arranged outside said matrix.

9. A method according to claim 8, wherein a tree including three levels is stored so as to have its root list stored in said auxiliary register and one of its element stored in said matrix of registers.

10. A method according to claim 1, wherein said information in said type of said root list includes an instruction code representing the instruction to execute.

11. A method according to claim 1, wherein if said type is a function application, a first element of said root list includes an instruction code or the root of a tree of lists representing a function definition.

12. A method according to claim 1, wherein a maximum number of words in a list is four.

13. A method according to claim 1, wherein a maximal depth of said tree of lists is three levels.

14. A method according to claim 13, wherein if said depth is three levels and if said list instruction of said root list stored in said registers indicates that said root list has more than one branch, only one of said branches is stored in said registers.

15. A method according to claim 5, wherein structure reduction is provided on data objects placed in said registers, such as in the base registers or in the base and auxiliary registers.

16. A method according to claim 15, wherein a mapping instruction is executed such that an instruction map having a two level structure

17. A method according to claim 15, wherein a transpose instruction is executed such that an instruction transpose having a three level structure ##EQU9## where the instruction transpose is written into the auxiliary register and the list of lists of instruction elements e_i,j, i and j being the denotations of the positions of the instruction elements in said matrix, in the matrix of base registers, such that the base register complex contains a square of elements, is rewritten by control from a control unit into the two level structure ##EQU10## where the auxiliary register values are made empty and the information elements in the matrix of base registers are transposed to be placed in a mirror position in relation to a diagonal through the matrix.

18. A method according to claim 15, wherein a swap instruction is executed such that an instruction swap having a three level structure ##EQU11## where the list of lists of instruction elements e_i,j, i and j being the denotations of the positions of the instruction elements in said matrix of base registers, such that the base register complex contain a square of elements, is rewritten by control from a control unit into a two level structure ##EQU12## such that the element (e_m,1, . . . ) changes place with the element (e_m+1,1, . . . ).

19. A method according to claim 15, wherein a skip instruction is executed such that an instruction skip having a three level structure ##EQU13## where the list of lists of instruction elements e_i,j, i and j being the denotations of the positions of the instruction elements in said matrix of base registers is rewritten by control from a control unit into ##EQU14## such that the element (e_m,1, . . . ) is deleted.

39. An arithmetic unit according to claim 6, wherein each register cell in a register comprises:

a) an internal one-bit register (r_R),

b) at least one internal one-wire bus (a_R, b_R) connectable to said one-bit register,

c) at least one internal, controllable connection, each including a switch (SW_N, SW_E, SW_Vi, SW_Vo, SW_Hi, SW_Ho, SW_Db, SW_a0, SW_a1, SW_b0, SW_b1) controllable by said control means (6p) making one of said at least one one-wire bus connectable to one of following elements;

a bus outside the cell, one of the cells belonging to another of said registers.

40. An arithmetic unit according to claim 39, wherein said at least one internal one-bit register (r_R) includes an input buffer means (Q1), such as an inverter, and an output buffer means (Q2), such as an inverter, and a controllable switch (SW_Q) connected between said buffer means.

41. An arithmetic unit according to claim 40, wherein said input buffer means and said output buffer means are separately connectable to said at least one internal one-wire bus (a_R, b_R) by controllable switches.

45. An arithmetic unit according to claim 40, wherein each inverter includes one of the following arrangements:

two complementary MOS-FETs (FIG. 9C), two MOS-FET of enhancement type (FIG. 9A), one MOS-FET of enhancement type and one MOS-FET of depletion type (FIG. 9B).

46. An arithmetic unit according to claim 40, wherein a comparator (G1, G2) is connected to compare the content in said part registers and provide the compared result to a wire in an external bus called access.

47. An arithmetic unit according to claim 46, wherein said comparator comprises:

a) a first NAND gate having a first input connected to the side of said first buffer means (Q1) turned from said internal register switch and a second input connected to the junction between said internal register switch and said second buffer means (Q2),

b) a second NAND gate having a first input connected to the side of said second buffer means (Q2) turned from said internal register switch and a second input connected to the junction between said internal register switch and said first buffer means (Q1);

the outputs of said NAND gates being interconnected and connected to said external bus wires in said bus called access.

48. An arithmetic unit according to claim 47, wherein each of said NAND gates includes two series coupled MOS-FET transistors having series coupled source/drain paths, their gates being the NAND gate inputs, and the drain of the topmost MOS-FET transistor being the output to said external bus wire.

20. An arithmetic unit for structural arithmetic processing, comprising:

a) at least one input/output means (v0, v1, v2, v3, id, env) for input and output of data lists,

b) several registers (S₀,0 to S₃,3, F0 to F3, ID, ENV) each being adapted to store a data word, said data words being arranged in said lists,

each data word having a mark part and an information part,

said mark part including a mark specifying if a register containing said data word is in use or not,

each said data lists being storable in a predetermined number of said registers,

said mark part of each said register among said registers being marked in use indicating that one of said lists has at least a part stored in said register, said list having a part stored in said register includes a list instruction denoting which kind of list it is, said list parts stored in said registers being arranged as a tree of list parts, of which one of the list parts is a root list part, a relation between said list parts being apparent from an arrangement of said list parts in said registers,

c) control means (6p) for controlling said registers and for making use of said list instructions belonging to list parts stored in said registers to rearrange said lists among said registers and for performing input/output operations of register content in accordance with said list instructions, forming an array of said registers (S₀,0 to S₃,3) having a predefined register (S₀,0 to S₃,0) for storing said root list part and a predefined set of registers for storing sublist parts, said array being automatically controlled by said control means (6p) to make a structure reduction of data objects placed in said registers in a rewrite operation performed in one register transfer cycle.

21. An arithmetic unit according to claim 20, wherein a maximum of said predetermined number of registers storing a list is four.

22. An arithmetic unit according to claim 20, wherein a maximal depth of said tree of lists is three levels.

23. An arithmetic unit according to claim 22, wherein if said depth is three levels and if said list instruction of said root list stored in said registers indicates that said root list has one or more than one branch, said control means is adapted to only store one of said branches in said registers.

24. An arithmetic unit according to claim 20, wherein at least one extra register (ID) is provided, in which an identifier of the stored tree of lists is storable.

25. An arithmetic unit according to claim 20, wherein at least one extra register (ENV) is provided, in which an environment of the stored tree of lists is storable.

26. An arithmetic unit according to claim 20, including a matrix of registers (S₀,0 to S₃,3), having a peripheral row (S₀,0, to S₃,0) providing main registers, the columns of said matrix providing base registers.

27. An arithmetic unit according to claim 26, wherein a number of auxiliary registers (F0 to F3) are provided outside said matrix.

28. An arithmetic unit according to claim 20, wherein said root list of said tree of lists is adapted to be placed in different registers dependent upon the level of said actual tree to be stored.

29. An arithmetic unit according to claim 26, wherein a tree including only one level is adapted to be stored in said main registers (S₀,0 to S₃,0).

30. An arithmetic unit according to claim 26, wherein a tree including two levels is adapted to have its root list stored in said main registers (S₀,0 to S₃,0) and its branch lists in base registers (S₀,0 to S₀,3, S₁,0 to S₁,3, S₂,0 to S₂,3, S₃,0 to S₃,3) of said base registers.

31. An arithmetic unit according to claim 27, wherein a tree including three levels is adapted to have its root list stored in said auxiliary registers (F0 to F3) and one of its element stored in said matrix of registers.

32. A method according to any of the claim 20, wherein the information regarding what kind of reduction to perform can be derived from the type of said root list by said control means (6p).

33. A method according to claim 32, wherein said information in said type of said root list includes an instruction code used by the control means (6p) to indicate what kind of instruction to execute.

34. A method according to claim 32, wherein if said type is a function application, the first element of said root list includes an instruction code or the root of a tree of lists representing a function definition, which is used by the control means (6p) to indicate the action to take in order to reduce said root list.

35. An arithmetic unit according to claim 20, wherein said registers are logically arranged in a sliced manner in planes in the core cell, each plane including at the most one register cell from each register, where each register cell is able to store one bit of information and where register cells within a plane are connectable to each other. (FIG. 7).

36. An arithmetic unit according to claim 35, wherein some of said registers are longer than other registers, such that some of said planes only have register cells belonging to said longer registers (FIG. 12).

37. An arithmetic unit according to claim 35, wherein at least some of said registers, named base registers, are arranged in rows and columns in a matrix of N times N registers, N being an integer;

said register cells of said base registers are interconnected in a bitwise manner and for each bit in said base registers in a column wire (v0, v1, v2, v3) is provided for each column and a row wire (h0, h1, h2, h3) is provided for each row, a controllable switch (SW_v0, SW_v1, SW_v2, SW_v3) being provided at each intersection point between a said column wire and a said row wire having the same row and column number, and each base register having a controllable register connection at least with its nearest row wire and column wire;

a connection being provided between the neighboring base registers both along said rows and along said columns.

38. An arithmetic unit according to claim 37, wherein said control means (6p) is adapted to control said controllable switches and said controllable register connections and to make one of at least three kinds of connections depending upon the kind of instruction to be executed:

a) a simple connection in one direction from one register to another,

b) two separate connections between registers, one in either direction,

c) as a time multiplexed connection between registers, in which transport of stored list elements are made in one direction and in the other direction in two sequent phases.

42. An arithmetic unit according to claim 36, wherein each register cell at least one of comprises:

a) at least one of a first (a_R) and a second (b_R) internal one-wire bus,

b) an internal one-bit register (r_R),

c) a first set of internal, controllable connections including switches controllable by said control means making said first bus connectable to a first input of said internal register and to a first set of several external buses coupled to other register cells,

d) a second set of internal, controllable connections including switches controllable by said control means making said second bus connectable to a second input of said internal register and to a second set of several external buses coupled to other register cells,

e) a third set of internal, controllable connections including switches controllable by said control means making the output of said internal one-bit register connectable to a third set of several external buses.

43. An arithmetic unit according to claim 42, wherein dependency upon the location of said registers, some of said register cells have fixed values fixedly connected to at least one of its internal connections not being one of said connections to said internal register.

44. An arithmetic unit according to claim 42, wherein each said controllable connection is at least one controllable switch, including one of the following arrangements:

a MOS-FET (FIG. 9E), two complementary MOS-FETs (FIG. 9F).

49. An arithmetic unit according to claim 27, wherein a mapping instruction is adapted to be executed such that an instruction map having a two level structure

50. An arithmetic unit according to claim 27, wherein a transpose instruction is adapted to be executed such that an instruction transpose having a three level structure ##EQU15## where the instruction transpose is written into the auxiliary register and the list of lists of instruction elements e_i,j, i and j being the denotations of the positions of the instruction elements in said matrix, in the matrix of base registers, such that the base register complex contains a square of elements, is adapted to be rewritten by control from said control means into ##EQU16## where the auxiliary register values are made empty and the information elements in the matrix of base registers are transposed to be placed in a mirror position in relation to a diagonal through the matrix.

51. An arithmetic unit according to claim 27, wherein a swap instruction is adapted to be executed such that an instruction swap having a three level structure ##EQU17## where the list of lists of instruction elements e_i,j, i and j being the denotations of the positions of the instruction elements in said matrix of base registers, such that the base register complex contain a square of elements, is adapted to be rewritten by control from said control means into ##EQU18## such that the element (e_m,1, . . . ) changes place with the element (e_m+1,1, . . . ).

52. An arithmetic unit according to claim 27, wherein a skip instruction is adapted to be executed such that an instruction skip having a three level structure ##EQU19## where the list of lists of instruction elements e_i,j, i and j being the denotations of the positions of the instruction elements in said matrix of base registers, such that the base register complex contain a square of elements, is adapted to be rewritten by control from said control means into ##EQU20## such that the element (e_m,1, . . . ) is deleted.

53. An arithmetic unit according to claim 20, wherein it is a part of a central processing unit.