Method and System for Storing Information in a Processing System

This patent application relates in general to systems for processing information and in particular to a processing system and method of operation.

A superscalar processing system includes multiple execution units for simultaneously executing multiple instructions. In some processing systems, instructions are executable out-of-order relative to their programmed sequence within the instruction stream. Accordingly, instructions' results are not necessarily available in-order of the instructions' programmed sequence. Nevertheless, some of these processing systems are designed to write instructions' results back to architectural registers in-order of the instructions' programmed sequence.

For this reason, some processing systems include rename buffers for intermediately storing instructions' results until a suitable moment for writing the intermediately stored results back to architectural registers. Previous techniques have been developed for controlling a processing system's rename buffers to accommodate data dependencies between the executed instructions. Nevertheless, shortcomings of such previous techniques include relatively slow speed, high complexity, and large physical size of systems implementing the techniques. With slow speed, high complexity, and large physical size, such systems are poorly suited for use in advanced integrated circuit microprocessors.

Thus, a need has arisen for a processing system and method of operation, in which speed of a system for controlling rename buffers is increased relative to previous techniques. Also, a need has arisen for a processing system and method of operation, in which complexity of a system for controlling rename buffers is decreased relative to previous techniques. Further, a need has arisen for a processing system and method of operation, in which physical size of a system for controlling rename buffers is decreased relative to previous techniques. Moreover, a need has arisen for a processing system and method of operation, in which a system for controlling rename buffers is more suited for use in advanced integrated circuit microprocessors relative to previous techniques.

In a processing system and method of operation, at least one execution unit processes information of a register in response to an instruction specifying the register. Each of multiple control units selectively allocates a respective one of multiple buffers to store the information in response to the instruction.

It is a technical advantage of the present invention that speed of a system for controlling rename buffers is increased relative to previous techniques.

It is another technical advantage of the present invention that complexity of a system for controlling rename buffers is decreased relative to previous techniques.

It is a further technical advantage of the present invention that physical size of a system for controlling rename buffers is decreased relative to previous techniques.

It is yet another technical advantage of the present invention that a system for controlling rename buffers is more suited for use in advanced integrated circuit microprocessors relative to previous techniques.

An exemplary embodiment of the present invention and its advantages are better understood by referring to the following descriptions and accompanying drawings, in which:

FIG. 1 is a block diagram of a processing system according to the preferred embodiment;

FIG. 2 is a block diagram of rename buffers of the system of FIG. 1;

FIG. 3 is a block diagram of an instruction queue of the system of FIG. 1;

FIG. 4 is an illustration of an instruction format according to the system of FIG. 1;

FIG. 5 is a block diagram of a source determining, tag assigning, allocating and deallocating memory ("STAM") of the system of FIG. 1;

FIG. 6 is a block diagram of a resource allocation array of the STAM of FIG. 5;

FIG. 7 is a schematic electrical circuit diagram of allocation bit logic of the STAM of FIG. 5;

FIG. 8 is a schematic electrical circuit diagram of a hold-off generate unit of the STAM of FIG. 5;

FIG. 9 is a schematic electrical circuit diagram of most recently allocated ("MRA") set logic of the STAM of FIG. 5;

FIG. 10 is a schematic electrical circuit diagram of MRA reset logic of the STAM of FIG. 5; and

FIG. 11 is a schematic electrical circuit diagram of a priority assign unit of the STAM of FIG. 5.

An exemplary embodiment of the present invention and its advantages are better understood by referring to FIGS. 1-11 of the drawings, like numerals being used for like and corresponding parts of the accompanying drawings.

FIG. 1 is a block diagram of a processing system 10 according to the preferred embodiment. In the preferred embodiment, system 10 is an integrated circuit superscalar microprocessor. System 10 is connected to a system bus 12.

System 10 includes a bus interface unit ("BIU") 14, an instruction cache 16, and a data cache 18. Moreover, system 10 includes a fixed point unit 20, a load/store unit 22, and a floating point unit 24. Also, system 10 includes fixed point rename buffers 26, floating point rename buffers 28, general purpose architectural registers ("GPR's") 30, and floating point architectural registers ("FPR's") 32.

Further, system 10 includes a control section, indicated by dashed enclosure 34. Control section 34 includes an instruction queue 36, an instruction decoder 38, and completion logic 40. In a significant aspect of the preferred embodiment, control section 34 further includes a fixed point source determining, tag assigning, allocating and deallocating memory ("STAM") 42 and a floating point STAM 44.

As discussed further hereinbelow in connection with FIG. 5, STAM 42 advantageously includes multiple identical cascadable rename buffer control units BFR0, BFR1, BFR2, BFR3 and BFR4. Control unit BFR0 is associated with physical rename buffer 0 (FIG. 2) of fixed point rename buffers 26. Accordingly, in a significant aspect of the preferred embodiment, control unit BFR0 controls whether rename buffer 0 stores data for any of GPR's 30. Likewise, control units BFR1-BFR4 are associated with physical rename buffers 1-4, respectively.

Control units BFR0-BFR4 are connected in a cascaded manner as shown in FIG. 5. Notably, if rename buffers 26 includes only one rename buffer, then STAM 42 includes only control unit BFR0. Similarly, if rename buffers 26 include more than five rename buffers, then each additional rename buffer has a respectively associated additional control unit (in addition to control units BFR0-BFR4 and identical to control unit BFR0) within STAM 42. Each such additional control unit is added to STAM 42 by connecting in a cascaded manner in series beginning at control unit BFR4.

Significantly, control units BFR0-BFR4 are cascadably connectable and are respectively associated with physical rename buffers 0-4. Such a modularized cascadable structure facilitates formation of simulation source files, layout of semiconductor fabrication plans, scalability of designs, adaptability to various processing systems, and usefulness as part of a library of reusable component designs. Accordingly, circuitry of STAM 42 is more suited for use in advanced integrated circuit microprocessors (such as superscalar microprocessor 10 of FIG. 1) relative to previous techniques for controlling rename buffers. Moreover, circuitry of STAM 42 has decreased complexity and decreased physical size relative to previous techniques for controlling rename buffers. Further, circuitry of STAM 42 has increased speed relative to previous techniques for controlling rename buffers.

Referring to FIG. 1, system 10 outputs and inputs information to and from system bus 12 through bus interface unit 14. Such information includes instructions and data. Instruction cache 16 stores instructions input through bus interface unit 14 from system bus 12. Data cache 18 stores data input through bus interface unit 14 from system bus 12. Moreover, bus interface unit 14 inputs stored data from data cache 18 and outputs such data to system bus 12.

Load/store unit 22 transfers data from data cache 18 to GPR's 30 through fixed point rename buffers 26. Moreover, load/store unit 22 transfers data from data cache 18 to FPR's 32 through floating point rename buffers 28. Further, load/store unit 22 transfers data from GPR's 30 to data cache 18 and from FPR's 32 to data cache 18. Load/store unit 22 executes such transfers of data between GPR's 30, FPR's 32, and data cache 18 in response to instructions decoded by instruction decoder 38 of control section

Similarly, in response to instructions decoded by instruction decoder 38, fixed point unit 20 inputs data from GPR's 30, processes such data, and outputs the processed data to GPR's 30 through fixed point rename buffers 26. Also, in response to instructions decoded by instruction decoder 38, floating point unit 24 inputs data from FPR's 32, processes such data, and outputs the processed data to FPR's 32 through floating point rename buffers 28. Together, fixed point unit 20, load/store unit 22, and floating point unit 24 form execution units of system 10. For clarity, FIG. 1 does not show connections from instruction decoder 38 to the execution units.

As previously mentioned, floating point unit 24 outputs data to FPR's 32 through floating point rename buffers 28. For example, floating point unit 24 modifies data for a particular one of FPR's 32 and stores such modified data in floating point rename buffers 28. Later, in response to a write-back signal on line 85b from completion logic 40 and a write-back address on lines 87b from floating point STAM 44, such modified data are written back to the particular FPR from floating point rename buffers 28.

Floating point unit 24 is able to input data from floating point rename buffers 28 and accordingly is able to modify data for the particular FPR more than once before such modified data are written back to the particular FPR from floating point rename buffers 28. In such a situation, multiple ones of floating point rename buffers 28 store data for the particular FPR. As discussed further hereinbelow in connection with FIGS. 5-11, floating point STAM 44 monitors which of floating point rename buffers 28 stores the most recent version of data for the particular FPR. In response to a write-back signal from completion logic 40 and a write-back address from floating point STAM 44, data from only one of floating point rename buffers 28 are written back to the particular FPR at any one time.

In a comparable manner, data are transferred between load/store unit 22, FPR's 32, and floating point rename buffers 28, responsive to indications from floating point STAM 44. Moreover, in a comparable manner, data are transferred between load/store unit 22, GPR's 30, and fixed point rename buffers 26, responsive to indications from fixed point STAM 42. Also, in a comparable manner, data are transferred between fixed point unit 20, GPR's 30, and fixed point rename buffers 26, responsive to indications from fixed point STAM 42.

In a significant aspect of the preferred embodiment, fixed point STAM 42 controls whether a particular one of fixed point rename buffers 26 stores data for a particular one of GPR's 30. Similarly, floating point STAM 44 controls whether a particular one of floating point rename buffers 28 stores data for a particular one of FPR's 32. Fixed point STAM 42 is representative one of STAM'S 42 and 44. Fixed point STAM 42 is discussed further hereinbelow in connection with FIGS. 5-11.

FIG. 2 is a block diagram of fixed point rename buffers 26. As shown in FIG. 2, fixed point rename buffers 26 include a rename buffer 0, a rename buffer 1, a rename buffer 2, a rename buffer 3, and a rename buffer 4. In the preferred embodiment, each of rename buffers 0-4 is 32-bits wide.

FIG. 3 is a block diagram of instruction queue 36. Instruction queue 36 stores an I0 instruction and an I1 instruction in a buffer I0 and a buffer I1, respectively, of dispatch buffers 56. In the preferred embodiment, in response to a cycle of system 10, either the I0 instruction is dispatched by itself to an execution unit, both the I0 and I1 instructions are dispatched together to respective execution units, or the I1 instruction is dispatched by itself to an execution unit. The contents of buffers I0 and I1 are output to instruction decoder 38 (FIG. 1) through lines 49a-b, respectively.

If the I0 instruction is dispatched by itself, then the shift1 signal (FIG. 1) is asserted, and the I1 instruction is dispatched in response to a later cycle of system 10. By comparison, if both the I0 and I1 instructions are dispatched together, then the shift2 signal (FIG. 1) is asserted. If the I1 instruction is dispatched by itself, then neither the shift1 signal nor the shift2 signal is asserted; the I0 instruction is dispatched in response to a later cycle of system 10, and the shift2 signal is asserted at that later cycle.

In the preferred embodiment, instruction queue 36 is able to input two instructions in parallel from instruction cache 16 through buses 50a-b, respectively. In response to assertion of the shift2 signal, instruction queue 36 transfers previously stored instructions from instruction buffers 54a-b to buffers I0 and I1, respectively. Also, instruction queue 36 transfers previously stored instructions from instruction buffers 52a-b to instructions buffers 54a-b, respectively. Moreover, instruction queue 36 inputs two instructions from instruction cache 16 through buses 50a-b and stores such instructions in instruction buffers 52a-b, respectively.

In response to assertion of the shift1 signal, instruction queue 36 transfers the previously stored instruction from buffer I1 to buffer I0. Also, instruction queue 36 transfers previously stored instructions from instruction buffer 54a to buffer I1, from instruction buffer 54b to instruction buffer 54a, from instruction buffer 52a to instruction buffer 54b, and from instruction buffer 52b to instruction buffer 52a. Moreover, instruction queue 36 inputs an instruction from instruction cache 16 through bus 50b and stores this instruction in instruction buffer 52b.

In the preferred embodiment, the I1 instruction is dispatchable to an execution unit for execution prior to the I0 instruction, even though the I0 instruction precedes the I1 instruction in their programmed sequence within the instruction stream. In such a situation, neither shift1 nor shift2 is asserted; the I0 instruction is dispatched in response to a later cycle of system 10, and the shift2 signal is asserted at that later cycle. In a significant aspect of the preferred embodiment, fixed point STAM 42 accommodates such an "out-of-order" dispatch of instructions in buffers I0 and I1.

FIG. 4 is an illustration of an instruction format according to system 10. FIG. 4 shows a generic instruction 60 having an instruction code field 62, an R.sub.T register field 64, an R.sub.A register field 66, and an R.sub.B register field 68. Instruction code field 62 indicates whether instruction 60 is to be executed by either fixed point unit 20, load/store unit 22, or floating point unit 24.

For example, instruction code 62 can identify instruction 60 as an ADD instruction for execution by fixed point unit 20. In response to such an instruction, fixed point unit 20 adds the contents of GPR registers R.sub.A and R.sub.B and stores the result in a selected one of fixed point rename buffers 26 in response to a destination tag from fixed point STAM 42. Advantageously, fixed point STAM 42 associates register R.sub.T with data stored in the selected one of fixed point rename buffers 26.

Later, in response to a write-back signal on line 85a from completion logic 40 and a write-back address on lines 87a from fixed point STAM 42, the associated data are written back to GPR register R.sub.T from the selected one of fixed point rename buffers 26. The write-back address indicates which of GPR's 30 is to store data input from the selected one of fixed point rename buffers 26, according to the register associated by fixed point STAM 42 with the selected one of fixed point rename buffers 26.

In the preferred embodiment, a register designated by R.sub.T register field 64 of instruction 60 is always a destination register into which a result of the instruction is to be stored. By comparison, a register designated by R.sub.B register field 68 is always a source register from which data are to be input and processed according to the instruction code designated in field 62. Depending upon the instruction code designated in field 62, a register designated by R.sub.A register field 66 is either a source register or a destination register.

FIG. 5 is a block diagram of STAM 42. In a significant aspect of the preferred embodiment, STAM 42 includes multiple identical cascadable rename buffer control units BFR0, BFR1, BFR2, BFR3 and BFR4. Control unit BFR0 is a representative one of control units BFR0-BFR4. Accordingly, for clarity, FIG. 1 shows detail of control unit BFR0 only.

Control unit BFR0 is associated with physical rename buffer 0 (FIG. 2) of fixed point rename buffers 26. Accordingly, in a significant aspect of the preferred embodiment, control unit BFR0 controls whether rename buffer 0 stores data for any of GPR's 30. Likewise, control units BFR1-BFR4 are associated with physical rename buffers 1-4, respectively.

Control units BFR0-BFR4 are connected in a cascaded manner as shown in FIG. 5. Notably, if rename buffers 26 includes only one rename buffer, then STAM 42 includes only control unit BFR0. Similarly, if rename buffers 26 include more than five rename buffers, then each additional rename buffer has a respectively associated additional control unit (in addition to control units BFR0-BFR4 and identical to control unit BFR0) within STAM 42. Each such additional control unit is added to STAM 42 by connecting in a cascaded manner in series beginning at control unit BFR4.

Referring also to FIG. 1, completion logic 40 outputs five completion lines to fixed point STAM 42 through lines 80a, namely ccc.sub.-- gpr.sub.-- wr.sub.-- 0, ccc.sub.-- gpr.sub.-- wr.sub.-- 1, ccc.sub.-- gpr.sub.-- wr.sub.-- 2, ccc.sub.-- gpr.sub.-- wr.sub.-- 3 and ccc.sub.-- gpr.sub.-- wr.sub.-- 4. Similarly, completion logic 40 outputs five completion lines to floating point STAM 44 through lines 80b (FIG. 1). Referring to FIG. 5, completion line ccc.sub.-- gpr.sub.-- wr.sub.-- 0 is associated with rename buffer 0 of fixed point rename buffers 26. Accordingly, assertion by completion logic 40 of completion line ccc.sub.-- gpr.sub.-- wr.sub.-- 0 indicates to control unit BFR0 that rename buffer 0 of fixed point rename buffers 26 is reallocatable to a different one of GPR's 30. Similarly, control unit BFR1 inputs completion line ccc.sub.-- gpr.sub.-- wr.sub.-- 1, control unit BFR2 inputs completion line ccc.sub.-- gpr.sub.-- wr .sub.-- 2, control unit BFR3 inputs completion line ccc.sub.-- gpr.sub.-- wr.sub.-- 3, and control unit BFR4 inputs completion line ccc.sub.-- gpr.sub.-- wr.sub.-- 4.

Each of control units BFR0-BFR4 inputs the shift1 line and shift2. Moreover, each of control units BFR0-BFR4 inputs request lines I0Rt.sub.-- rq, I0Ra.sub.-- rq, I1Rt.sub.-- rq and I1Ra.sub.-- rq from instruction decoder 38 through lines 82a (FIG. 1). Similarly, floating point STAM 44 inputs request lines from instruction decoder 38 through lines 82b. Instruction decoder 38 indicates to fixed point STAM 42 that an instruction in buffer I0 (FIG. 3) is a valid fixed point or load/store instruction by asserting one or more of request lines I0Rt.sub.-- rq and I0Ra.sub.-- rq.

For example, instruction decoder 38 asserts request line I0Rt.sub.-- rq in response to a valid R.sub.T field 64 (FIG. 4) of an instruction in buffer I0 (FIG. 3) specifying one of GPR's 30 into which a result of the instruction is destined to be stored (as specified by the instruction code in field 62). Moreover, instruction decoder 38 asserts request line I0Ra.sub.-- rq in response to a valid R.sub.A field 66 of an instruction in buffer I0 specifying one of GPR's 30 into which a result of the instruction is destined to be stored. Also, instruction decoder 38 asserts request line I1Rt.sub.-- rq in response to a valid R.sub.T field 64 of an instruction in buffer I1 specifying one of GPR's 30 into which a result of the instruction is destined to be stored. Similarly, instruction decoder 38 asserts request line I1Ra.sub.-- rq in response to a valid R.sub.A field 66 of an instruction in buffer I1 specifying one of GPR's 30 into which a result of the instruction is destined to be stored.

Fixed point STAM 42 and floating point STAM 44 input 5-bit register fields I0Rt.sub.-- d, I0Ra.sub.-- d, I1Rt.sub.-- d, I1Ra.sub.-- d, I0Ra.sub.-- s, I0Rb.sub.-- s, I1Ra.sub.-- s and I1Rb.sub.-- s from instruction queue 36 through lines 84. Referring to FIG. 5, each of control units BFR0-BFR4 inputs these 5-bit register fields. I0Rt.sub.-- d is a 5-bit binary encoded register number from R.sub.T register field 64 of an instruction in buffer I0. I0Ra.sub.-- d is a 5-bit binary encoded register number from R.sub.A register field 66 of an instruction in buffer I0, where register R.sub.A is a destination register into which a result of the instruction is to be stored.

I1Rt.sub.-- d is a 5-bit binary encoded register number from R.sub.T register field 64 of an instruction in buffer I1. I1Ra.sub.-- d is a 5-bit binary encoded register number from R.sub.A register field 66 of an instruction in buffer I1, where register R.sub.A is a destination register into which a result of the instruction is to be stored.

I0Ra.sub.-- s is a 5-bit binary encoded register number from R.sub.A register field 66 of an instruction in buffer I0, where register R.sub.A is a source register from which data are to be input by the instruction. I0Rb.sub.-- s is a 5-bit binary encoded register number from R.sub.B register field 68 of an instruction in buffer I0.

I1Ra.sub.-- s is a 5-bit binary encoded register number from R.sub.A register field 66 of an instruction in buffer I1, where register R.sub.A is a source register from which data are to be input by the instruction. I1Rb.sub.-- s is a 5-bit binary encoded register number from R.sub.B register field 68 of an instruction in buffer I1.

FIG. 6 is a block diagram of a resource allocation array ("RAA") 100 of STAM 42. RAA 100 includes four cells 102a-d. Cell 102a inputs alloc0 through a line 103. If alloc0 is set (indicating rename buffer 0 is not presently allocated to a GPR) and I0Rt.sub.-- rq is set (indicating the R.sub.T register field of an instruction in buffer I0 is valid), then cell 102a sets assign.sub.-- I0Rt, clears avail.sub.-- I0Rt, and clears rq.sub.-- I0Rt.sub.-- after0. If alloc0 is cleared (indicating rename buffer 0 is presently allocated to a GPR), then cell 102a equates rq.sub.-- I0Rt.sub.-- after0 to I0Rt.sub.-- rq, clears avail.sub.-- I0Rt, and clears assign.sub.-- I0Rt. If alloc0 is set and I0Rt.sub.-- rq is cleared, then cell 102a clears rq.sub.-- I0Rt.sub.-- after0, clears assign.sub.-- I0Rt, and sets avail.sub.-- I0Rt.

Referring to cell 102b, if avail.sub.-- I0Rt is set (indicating rename buffer 0 is not presently allocated to a GPR) and I0Ra.sub.-- rq is set (indicating the R.sub.A register field of an instruction in buffer I0 is valid), then cell 102b sets assign.sub.-- I0Ra, clears avail.sub.-- I0Ra, and clears rq.sub.-- I0Ra.sub.-- after0. If avail.sub.-- I0Rt is cleared (indicating rename buffer 0 is presently allocated to a GPR), then cell 102b equates rq.sub.-- I0Ra.sub.-- after0 to I0Ra.sub.-- rq, clears avail.sub.-- I0Ra, and clears assign.sub.-- I0Ra. If avail.sub.-- I0Rt is set and I0Ra.sub.-- rq is cleared, then cell 102b clears rq.sub.-- I0Ra.sub.-- after0, clears assign.sub.-- I0Ra, end sets avail.sub.-- I0Ra.

Referring to cell 102c, if avail.sub.-- I0Ra is set (indicating rename buffer 0 is not presently allocated to a GPR) and I1Rt.sub.-- rq is set (indicating the R.sub.T register field of an instruction in buffer I1 is valid), then cell 102c sets assign.sub.-- I1Rt, clears avail.sub.-- I1Rt, and clears rq.sub.-- I1Rt.sub.-- after0. If avail.sub.-- I0Ra is cleared (indicating rename buffer 0 is presently allocated to a GPR), then cell 102c equates rq.sub.-- I1Rt.sub.-- after0 to I1Rt.sub.-- rq, clears avail.sub.-- I1Rt, and clears assign.sub.-- I1Rt. If avail.sub.-- I0Ra is set and I1Rt rq is cleared, then cell 102c clears rq.sub.-- I1Rt.sub.-- after0, clears assign.sub.-- I1Rt, and sets avail.sub.-- I1Rt.

Referring to cell 102d, if avail.sub.-- I1Rt is set (indicating rename buffer 0 is not presently allocated to a GPR) and I1Ra.sub.-- rq is set (indicating the R.sub.A register field of an instruction in buffer I1 is valid), then cell 102d sets assign.sub.-- I1Ra and clears rq.sub.-- I1Ra.sub.-- after0. If avail.sub.-- I1Rt is cleared (indicating rename buffer 0 is presently allocated to a GPR), then cell 102d equates rq.sub.-- I1Ra.sub.-- after0 to I1Ra.sub.-- rq and clears assign.sub.-- I1Ra. If avail.sub.-- I1Rt is set and I1Ra.sub.-- rq is cleared, then cell 102d clears rq.sub.-- I1Ra.sub.-- after0 and clears assign.sub.-- I1Ra.

Referring also to FIG. 5, through lines 109a-d, the resource allocation array of control unit BFR1 inputs rq.sub.-- I0Rt.sub.-- after0, rq.sub.-- I0Ra.sub.-- after0, rq.sub.-- I1Rt.sub.-- after0, and rq.sub.-- I1Ra.sub.-- after0 in place of I0Rt.sub.-- rq, I0Ra.sub.-- rq, I1Rt.sub.-- rq and I1Ra.sub.-- rq, respectively. In this manner, control unit BFR1 is connected in series with control unit BFR0 in a cascaded fashion.

Moreover, RAA 100 outputs assign.sub.-- I0Rt, assign.sub.-- I0Ra, assign.sub.-- I1Rt and assign.sub.-- I1Ra to Destination Compares and MRA Set Logic 104, multiplexer 106, and allocation bit logic 110 through lines 108. In response to lines 108, multiplexer 106 outputs one of destination fields I0Rt.sub.-- d, I0Ra.sub.-- d, I1Rt.sub.-- d and I1Ra.sub.-- d to a register address value latch 113. At any single moment, only one of lines assign.sub.-- I0Rt, assign.sub.-- I0Ra, assign.sub.-- I1Rt and assign.sub.-- I1Ra is true.

Lines assign.sub.-- I0Rt, assign.sub.-- I0Ra, assign.sub.-- I1Rt and assign.sub.-- I1Ra are associated with destination fields I0Rt.sub.-- d, I0Ra.sub.-- d, I1Rt.sub.-- d and I1Ra.sub.-- d, respectively. For example, if assign.sub.-- I0Rt=0, assign.sub.-- I0Ra=0, assign.sub.-- I1Rt=1, and assign.sub.-- I1Ra=0, then multiplexer 106 outputs destination field I1Rt.sub.-- d to register address value latch 113. In response to a positive edge transition of the clock signal on clock line CLK, register address value latch 113 latches the output destination field from multiplexer 106. Accordingly, if 5-bit destination field I1Rt.sub.-- d is binary encoded to represent general purpose register number-27 of GPR's 30, then register address value latch 113 latches the binary encoded representation of 27. In this manner, control unit BFR0 associates the register number of one of GPR's 30 with allocated rename buffer 0 (FIG. 2) of fixed point rename buffers 26.

FIG. 7 is a schematic electrical circuit diagram of allocation bit logic 110 of STAM 42. Allocation bit logic 110 includes latches 112a-d and OR-gates 114a-d. Together, the outputs of OR-gates 114a-d form I0Rt.sub.-- alloc.sub.-- 0, I0Ra.sub.-- alloc.sub.-- 0, I1Rt.sub.-- alloc.sub.-- 0 and I1Ra.sub.-- alloc.sub.-- 0, respectively.

If assign.sub.-- I0Rt is true, then I0Rt.sub.-- alloc.sub.-- 0 is true. Similarly, if assign.sub.-- I0Ra is true, then I0Ra.sub.-- alloc.sub.-- 0 is true. Also, if assign.sub.-- I1Rt is true, then I1Rt.sub.-- alloc.sub.-- 0 is true. Moreover, if assign.sub.-- I1Ra is true, then I1Ra.sub.-- alloc.sub.-- 0 is true.

If shift1 is false, then the outputs of multiplexers 116a-b are equal to assign.sub.-- I0Rt and assign.sub.-- I0Ra, respectively. Otherwise, if shift1 is true, then the outputs of multiplexers 116a-b are equal to I1Rt.sub.-- alloc.sub.-- 0 and I1Ra.sub.-- alloc.sub.-- 0, respectively. In this manner, if shift1 is true (such that the I1 instruction is moved into buffer I0 as discussed further hereinabove in connection with FIG. 3), then the states of I1Rt.sub.-- alloc.sub.-- 0 and I1Ra.sub.-- alloc.sub.-- 0 are preserved by copying I1Rt.sub.-- alloc.sub.-- 0 and I1Ra.sub.-- alloc.sub.-- 0 to the outputs of multiplexers 116a-b, respectively.

By preserving the states of I1Rt.sub.-- alloc.sub.-- 0 and I1Ra.sub.-- alloc.sub.-- 0, fixed point STAM 42 advantageously accommodates a situation where the I1 instruction is dispatched to an execution unit "out-of-order" prior to the I0 instruction, as discussed further hereinabove in connection with FIG. 3. Such preservation is important because rename buffer 0 (FIG. 2) is subject to being allocated by control unit BFR0 in response to the previously dispatched I1 instruction, and such allocation is indicated by the states of I1Rt.sub.-- alloc.sub.-- 0 and I1Ra.sub.-- alloc.sub.-- 0.

If the output of multiplexer 116a is true, then the output of latch 112a is set in response to a positive edge transition of a clock signal on clock line CLK, so that I0Rt.sub.-- alloc.sub.-- 0 is latched true. Similarly, if the output of multiplexer 116b is true, then the output of latch 112b is set in response to a positive edge transition of the clock signal on clock line CLK, so that I0Ra.sub.-- alloc.sub.-- 0 is latched true. Also, if assign.sub.-- I1Rt is true, then the output of latch 112c is set in response to a positive edge transition of the clock signal on clock line CLK, so that I1Rt.sub.-- alloc.sub.-- 0 is latched true. Moreover, if assign.sub.-- I1Ra is true, then the output of latch 112d is set in response to a positive edge transition of the clock signal on clock line CLK, so that I1Ra.sub.-- alloc.sub.-- 0 is latched true. As previously mentioned, only one of lines assign.sub.-- I0Rt, assign.sub.-- I0Ra, assign.sub.-- I1Rt and assign.sub.-- I1Ra is true at any single moment.

If assign.sub.-- I0Rt is false and the output of latch 112a is false, then I0Rt.sub.-- alloc.sub.-- 0 is false. Similarly, if assign.sub.-- I0Ra is false and the output of latch 112b is false, then I0Ra.sub.-- alloc.sub.-- 0 is false. Also, if assign.sub.-- I1Rt is false and the output of latch 112c is false, then I1Rt.sub.-- alloc.sub.-- 0 is false. Moreover, if assign.sub.-- I1Ra is false and the output of latch 112d is false, then I1Ra.sub.-- alloc.sub.-- 0 is false.

Referring to OR gate 117, if either shift1 or shift2 is true, then the outputs of latches 112c-d are cleared in response to a positive edge transition of a clock signal on clock line CLK. Referring to logic indicated by dashed enclosures 118a-b, if shift2 is true, then the outputs of latches 112a-b are cleared in response to a positive edge transition of the clock signal on clock line CLK.

Referring to logic 118a, if shift1 is true and I1Rt.sub.-- alloc.sub.-- 0 is false, then the output of latch 112a is cleared in response to a positive edge transition of the clock signal on clock line CLK. Referring to logic 118b, if shift1 is true and I1Ra.sub.-- alloc.sub.-- 0 is false, then the output of latch 112b is cleared in response to a positive edge transition of the clock signal on clock line CLK. In this manner, if shift1 is true (such that the I1 instruction is moved into buffer I0 as discussed further hereinabove in connection with FIG. 3), then the states of I1Rt.sub.-- alloc.sub.-- 0 and I1Ra.sub.-- alloc.sub.-- 0 are preserved by copying I1Rt.sub.-- alloc.sub.-- 0 and I1Ra.sub.-- alloc.sub.-- 0 to the outputs of latches 112a-b, respectively.

If any of I0Rt.sub.-- alloc.sub.-- 0, I0Ra.sub.-- alloc.sub.-- 0, I1Rt.sub.-- alloc.sub.-- 0, or I1Ra.sub.-- alloc.sub.-- 0 is true, then rename buffer 0 is allocated to one of GPR's 30 by control unit BFR0. At any single moment, only one of lines I0Rt.sub.-- alloc.sub.-- 0, I0Ra.sub.-- alloc.sub.-- 0, I1Rt.sub.-- alloc.sub.-- 0 and I1Ra.sub.-- alloc.sub.-- 0 is true. Referring to logic indicated by dashed enclosure 119, if shift2 is asserted, then latch 120 sets allocated.sub.-- 0 in response to a positive edge transition of the clock signal on clock line CLK if either I0Rt.sub.-- alloc.sub.-- 0, I0Ra.sub.-- alloc 0, I1Rt.sub.-- alloc.sub.-- 0, or I1Ra.sub.-- alloc.sub.-- 0 is true.

By comparison, if shift1 is asserted, then latch 120 sets allocated.sub.-- 0 in response to a positive edge transition of the clock signal on clock line CLK if either I0Rt.sub.-- alloc.sub.-- 0 or I0Ra.sub.-- alloc.sub.-- 0 is true. Moreover, if shift1 is asserted, then the states of I1Rt.sub.-- alloc.sub.-- 0 and I1Ra.sub.-- alloc.sub.-- 0 are preserved by copying I1Rt.sub.-- alloc.sub.-- 0 and I1Ra.sub.-- alloc.sub.-- 0 to the outputs of latches 112a-b, respectively.

Accordingly, allocated.sub.-- 0 indicates to MRA reset logic 130 (through a line 132 shown in FIG. 5) whether rename buffer 0 is allocated in response to a dispatched instruction. If allocated.sub.-- 0 is true, then rename buffer 0 (FIG. 2) is allocated to store data for a particular one of GPR's 30 as specified by a dispatched instruction. The particular GPR's number is latched by register address value latch 113 of control unit BFR0, as discussed further hereinabove in connection with register address value latch 113. If ccc.sub.-- gpr.sub.-- wr.sub.-- 0 is true, then latch 120 clears allocated.sub.-- 0 in response to a positive edge transition of the clock signal on clock line CLK, thereby indicating that rename buffer 0 is not allocated to any of GPR's 30.

Notably, if shift1 is asserted (indicating the instruction in buffer I1 is not being dispatched), then allocated.sub.-- 0 does not reflect an indication by I1Rt.sub.-- alloc.sub.-- 0 or I1Ra.sub.-- alloc.sub.-- 0 that rename buffer 0 is allocated. Accordingly, a NOR-gate 124 inputs allocated.sub.-- 0 and the outputs of latches 112a-d. If any of these inputs is true, then alloc0 is cleared; otherwise, alloc0 is set. In this manner, alloc0 indicates to RAA 100 whether rename buffer 0 is allocated in response to an instruction, irrespective of whether the instruction is dispatched or not dispatched.

The outputs of latches 112a-d are output to instruction decoder 38 (FIG. 1) through lines 122.sub.-- 0. Likewise, fixed point STAM 42 outputs lines 122.sub.-- 1-122.sub.-- 4 (FIG. 1) to instruction decoder 38 from control units BFR1-BFR4, respectively. For clarity, lines 122.sub.-- 1-122.sub.-- 4 are not shown in FIG. 5.

Lines 122.sub.-- 0-122.sub.-- 4 indicate to instruction decoder 38 whether fixed point STAM 42 has allocated a rename buffer in response to assertion by instruction decoder 38 of one or more of I0Rt.sub.-- rq, I0Ra.sub.-- rq, I1Rt.sub.-- rq and I1Ra.sub.-- rq. In response to such indications from fixed point STAM 32 through lines 122.sub.-- 0-122.sub.-- 4, instruction decoder 38 clears any previously asserted one of I0Rt.sub.-- rq, I0Ra.sub.-- rq, I1Rt.sub.-- rq and I1Ra.sub.-- rq for which fixed point STAM 42 has allocated a rename buffer. In this manner, fixed point STAM avoids redundant allocation of multiple rename buffers in response to a single request from instruction decoder 38.

Referring also to FIG. 5, as control unit BFR0 outputs lines I0Rt.sub.-- alloc.sub.-- 0, I0Ra.sub.-- alloc.sub.-- 0, I1Rt.sub.-- alloc.sub.-- 0, and I1Ra.sub.-- alloc.sub.-- 0, likewise each of control units BFR1-BFR4 outputs lines I0Rt.sub.-- alloc.sub.-- n, I0Ra.sub.-- alloc.sub.-- n, I1Rt.sub.-- alloc.sub.-- n, and I1Ra.sub.-- alloc.sub.-- n (where n=1 for BFR1, n=2 for BFR2, n=3 for BFR3, and n=4 for BFR4). Referring to FIG. 5, 5-bit destination tag I0Rt is formed by bundling lines I0Rt.sub.-- alloc.sub.-- 0, I0Rt.sub.-- alloc.sub.-- 1, I0Rt.sub.-- alloc.sub.-- 2, I0Rt.sub.-- alloc.sub.-- 3 and I0Rt.sub.-- alloc.sub.-- 4. Similarly, 5-bit destination tags I0Ra, I1Rt and I1Ra are formed by bundling lines I0Ra.sub.-- alloc.sub.-- n, I1Rt.sub.-- alloc.sub.-- n, and I1Ra.sub.-- alloc.sub.-- n, respectively (where n=[0 . . . 4]).

Referring also to FIG. 1, destination tags I0Rt, I0Ra, I1Rt and I1Ra are output from fixed point STAM 42 to fixed point unit 20 and to load/store unit 22 through lines 128. For example, if fixed point STAM 42 outputs a destination tag I0Rt=10000 while fixed point unit 20 executes an instruction from buffer I0 (FIG. 3), then fixed point unit 20 stores the instruction's R.sub.T result data in rename buffer 0 (FIG. 2) of fixed point rename buffers 26. Moreover, control unit BFR0 of fixed point STAM 42 associates allocated rename buffer 0 with the register number specified in the instruction's R.sub.T register field, as discussed further hereinabove in connection with register address value latch 113.

FIG. 8 is a schematic electrical circuit diagram of a hold-off generate unit 140 of fixed point STAM 42. Normally, control unit BFR0 sets its MRAbit when it allocates its associated rename buffer 0 (FIG. 2) to store data for a particular one of GPR's 30 in response to a dispatched instruction. Nevertheless, with hold-off generate unit 140, fixed point STAM 42 accommodates a situation where (1) the I1 instruction is dispatched before or together with the I0 instruction and (2) the I0 and I1 instructions each have a common destination register specified in either R.sub.T register field 64 (FIG. 4) or R.sub.A register field 66.

For example, if

I0 instruction is: instr. code, R3, R4, R5

I1 instruction is: instr. code, R3, R6, R7

then register number 3 (i.e. R3) is specified in R.sub.T register field 64 of both the I0 and I1 instructions. In such a situation, if control unit BFR0 allocates rename buffer 0 to store data for register R3 in response to dispatched instruction I0, then control unit BFR0 does not set its MRA bit. This is because a different rename buffer is allocated by fixed point STAM 42 to store data for register R3 in response to the previously or simultaneously dispatched I1 instruction. Since the I1 instruction follows the I0 instruction in sequence, this different rename buffer is designated by fixed point STAM 42 as the "most recently allocated" rename buffer storing data for register R3.

Accordingly, FIG. 5 and FIG. 8 show four comparators 150a-d and their respective outputs 152a-d. Referring to FIG. 8, comparator 150a asserts output 152a in response to a match between I0Rt.sub.-- d and I1Rt.sub.-- d. Similarly, comparator 150b asserts output 152b in response to a match between I0Ra.sub.-- d and I1Rt.sub.-- d. Also, comparator 150c asserts output 152c in response to a match between I0Rt.sub.-- d and I1Ra.sub.-- d. Moreover, comparator 150d asserts output 152d in response to a match between I0Ra.sub.-- d and I1Ra.sub.-- d.

If I1Rt.sub.-- rq is true (indicating a valid R.sub.T field 64 of an instruction in buffer I1) and output 152a is asserted, then hold-off generate unit 140 asserts I0Rt.sub.-- holdoff 154a. Similarly, if I1Rt.sub.-- rq is true and output 152b is asserted, then holdsoff generate unit 140 asserts I0Ra.sub.-- holdoff 154b. Also, if I1Ra.sub.-- rq is true (indicating a valid R.sub.A field 66 of an instruction in buffer I1) and output 152c is asserted, then hold-off generate unit 140 asserts I0Rt.sub.-- holdoff 154a. Moreover, if I1Ra.sub.-- rq is true and output 152d is asserted, then hold-off generate unit 140 asserts I0Ra.sub.-- holdoff 154b.

FIG. 9 is a schematic electrical circuit diagram of MRA set logic 160 of Destination Compares and MRA Set Logic 104. As shown in FIG. 5, Destination Compares and MRA Set Logic 104 inputs I0Rt.sub.-- holdoff 154a and I0Ra.sub.-- holdoff 154b from hold-off generate unit 140.

Referring to FIG. 9, MRA set logic 160 asserts mra.sub.-- set.sub.-- 0 if:

(1) assign.sub.-- I0Rt.sub.-- 0 is true and shift1 is true;

(2) assign.sub.-- I0Ra.sub.-- 0 is true and shift1 is true;

(3) assign.sub.-- I1Rt.sub.-- 0 is true and shift2 is true; or

(4) assign.sub.-- I1Ra.sub.-- 0 is true and shift2 is true.

Moreover, MRA set logic 160 asserts mra.sub.-- set.sub.-- 0 if:

(1) assign.sub.-- I0Rt.sub.-- 0 is true and shift2 is true (unless I0Rt.sub.-- holdoff is true); or

(2) assign.sub.-- I0Ra.sub.-- 0 is true and shift2 is true (unless I0Ra.sub.-- holdoff is true).

Accordingly, if the I0 instruction is dispatched after or together with the I1 instruction (indicated by assertion of shift2), and if the I0 and I1 instructions each have a common destination register specified in either the R.sub.T or R.sub.A register fields, and if control unit BFR0 allocates rename buffer 0 to store data for the common destination register in response to dispatched instruction I0, then control unit BFR0 does not set its MRAbit. Instead, a different control unit of fixed point STAM 42 allocates a different rename buffer to store data for the common destination register in response to dispatched instruction I1, and this different control unit sets its MRAbit.

Referring also to FIG. 5, MRA set logic 160 outputs mra.sub.-- set.sub.-- 0 as line 162 to an MRA latch 164. In response to assertion of mra.sub.-- set.sub.-- 0, MRA latch 164 sets MRAbit 166.

FIG. 10 is a schematic electrical circuit diagram of MRA reset logic 130 of fixed point STAM 42. Further, FIG. 10 shows destination compares logic 170 of Destination Compares and MRA Set Logic 104. As shown in FIGS. 5 and 10, destination compares logic 170 inputs the latched register number from register address value latch 113 as reg.sub.-- adr.sub.-- 0. Moreover, destination compares logic 170 inputs I0Rt.sub.-- d, I0Ra.sub.-- d, I1Rt.sub.-- d and I1Ra.sub.-- d.

Notably, the MRA reset logic of each of control units BFR0-BFR4 inputs request lines I0Rt.sub.-- rq, I0Ra.sub.-- rq, I1Rt.sub.-- rq and I1Ra.sub.-- rq from instruction decoder 38 (FIG. 1). Also, the MRA reset logic of each of control units BFR0-BFR4 inputs shift1, shift2, allocated.sub.-- n, and ccc.sub.-- gpr.sub.-- wr.sub.-- n (where n=1 for BFR1, n=2 for BFR2, n=3 for BFR3, and n=4 for BFR4).

Control unit BFR0 clears its MRA bit in response to any of three situations. In the first situation, ccc.sub.-- gpr.sub.-- wr.sub.-- 0 is asserted by completion logic 40 as discussed further hereinabove in connection with FIG. 1 and FIG. 5. In the second situation, the I0 instruction is dispatched, and the dispatched I0 instruction has a valid destination register equal to the GPR register number associated by control unit BFR0 with allocated rename buffer 0. In the third situation, the I0 instruction is dispatched after or together with the I1 instruction, and the dispatched I1 instruction has a valid destination register equal to the GPR register number associated by control unit BFR0 with allocated rename buffer 0.

Accordingly, destination compares logic 170 includes four comparators 172a-d. If reg.sub.-- adr.sub.-- 0 matches I0Rt.sub.-- d, then comparator 172a sets line 174a to MRA reset logic 130. Similarly, if reg.sub.-- adr.sub.-- 0 matches I0Ra.sub.-- d, then comparator 172b sets line 174b to MRA reset logic 130. Also, if reg.sub.-- adr.sub.-- 0 matches I1Rt.sub.-- d, then comparator 172c sets line 174c to MRA reset logic 130. Moreover, if reg.sub.-- adr0 matches I1Ra.sub.-- d, then comparator 172d sets line 174d to MRA reset logic 130.

MRA reset logic 130 asserts mra.sub.-- reset.sub.-- 0 if allocated.sub.-- 0 is true and:

(1) line 174a is true, shift1 is true, and I0Rt.sub.-- rq is true;

(2) line 174a is true, shift2 is true, and I0Rt.sub.-- rq is true;

(3) line 174b is true, shift1 is true, and I0Ra.sub.-- rq is true;

(4) line 174b is true, shift2 is true, and I0Ra.sub.-- rq is true;

(5) line 174c is true, shift2 is true, and I1Rt.sub.-- rq is true; or

(6) line 174d is true, shift2 is true, and I1Ra.sub.-- rq is true.

Moreover, MRA reset logic 130 asserts mra.sub.-- reset.sub.-- 0 if ccc.sub.-- gpr.sub.-- wr.sub.-- 0 is true. Referring also to FIG. 5, MRA reset logic 130 outputs mra.sub.-- reset.sub.-- 0 as line 176 to MRA latch 164. In response to assertion of mra.sub.-- reset.sub.-- 0, MRA latch 164 clears MRAbit 166.

Referring to FIG. 5, fixed point STAM 42 includes source compares logic 190. Source compares logic 190 inputs the latched register number from register address value latch 113 as reg.sub.-- adr.sub.-- 0. Moreover, source compares logic 190 inputs I0Ra.sub.-- s, I0Rb.sub.-- s, I1Ra.sub.-- s, and I1Rb.sub.-- s.

Source compares logic 190 compares reg.sub.-- adr.sub.-- 0 with I0Ra.sub.-- s, with I0Rb.sub.-- s, with I1Ra.sub.-- s, and with I1Rb.sub.-- s. In response to a match between reg.sub.-- adr.sub.-- 0 and I0Ra.sub.-- s, source compares logic 190 asserts a line to an AND gate 192a, so that AND gate 192a asserts a line I0Ra.sub.-- source.sub.-- 0 if MRA bit 166 is set. Similarly, in response to a match between reg.sub.-- adr.sub.-- 0 and I0Rb.sub.-- s, source compares logic 190 asserts a line to an AND gate 192b, so that AND gate 192b asserts a line I0Rb.sub.-- source.sub.-- 0 if MRA bit 166 is set.

In addition to lines I0Ra.sub.-- source.sub.-- 0 and I0Rb.sub.-- source.sub.-- 0, control unit BFR0 further outputs lines I1Ra.sub.-- source.sub.-- 0 and I1Rb.sub.-- source.sub.-- 0 from multiplexers 194a-b, respectively. Multiplexers 194a-b are discussed further hereinbelow in connection with FIG. 11.

As control unit BFR0 outputs lines I0Ra.sub.-- source.sub.-- 0, I0Rb.sub.-- source.sub.-- 0, I1Ra.sub.-- source.sub.-- 0 and I1Rb.sub.-- source.sub.-- 0, likewise each of control units BFR1-BFR4 outputs lines I0Ra.sub.-- source.sub.-- n, I0Rb.sub.-- source.sub.-- n, I1Ra.sub.-- source.sub.-- n and I1Rb.sub.-- source.sub.-- n (where n=1 for BFR1, n=2 for BFR2, n=3 for BFR3, and n=4 for BFR4). A 5-bit source tag I0Ra is formed by bundling lines I0Ra.sub.-- source.sub.-- 0, I0Ra.sub.-- source.sub.-- 1, I0Ra.sub.-- source.sub.-- 2, I0Ra.sub.-- source.sub.-- 3 and I0Ra.sub.-- source.sub.-- 4. Similarly, 5-bit source tags I0Rb, I1Ra and I1Rb are formed by bundling lines I0Rb.sub.-- source.sub.-- n, I1Ra.sub.-- source.sub.-- n, and I1Rb.sub.-- source.sub.-- n, respectively (where n=[0 . . . 4]).

Referring also to FIG. 1, source tags I0Ra, I0Rb, I1Ra and I1Rb are output from fixed point STAM 42 to fixed point unit 20 and to load/store unit 22 through lines 198. In this manner, fixed point STAM 42 indicates whether a most recently allocated one of fixed point rename buffers 26 stores data for a register number specified in an instruction's R.sub.A register field or R.sub.B register field. Accordingly, if multiple rename buffers store data for the specified register number, then fixed point STAM 42 indicates only the most recently allocated one of the rename buffers.

For example, if fixed point STAM 42 outputs a source tag I0Ra=00100 while fixed point unit 20 executes an instruction from buffer I0 (FIG. 3), then fixed point unit 20 inputs the instruction's Ra source data from rename buffer 2 (FIG. 2) of fixed point rename buffers 26. As another example, if fixed point STAM 42 outputs a source tag I0Ra=00000 while fixed point unit 20 executes an instruction from buffer I0 (FIG. 3), then fixed point unit 20 inputs the instruction's R.sub.A source data from the one of GPR's 30 specified in the instruction's R.sub.A register field.

FIG. 11 is a schematic electrical circuit diagram of a priority assign unit 200 of fixed point STAM 42. With priority-assign unit 200, fixed point STAM 42 accommodates a situation where (1) the I1 instruction is dispatched before or together with the I0 instruction and (2) the I0 instruction has a destination register which also is a source register for the I1 instruction.

For example, if

I0 instruction is: instr. code, R3, R4, R5

I1 instruction is: instr. code, R8, R6, R3

then register number 3 (i.e. R3) is specified in the I0 instruction's R.sub.T destination register field and also in the I1 instruction's R.sub.B source register field. In such a situation, fixed point STAM 42 does not allocate a rename buffer to store data for register R3 (in response to dispatched instruction I0) prior to the dispatch of instruction I1. This is because the I1 instruction is dispatched before or together with the I0 instruction.

Moreover, in response to the dispatched I0 instruction, fixed point STAM 42 allocates a rename buffer to store data for the I0 instruction's destination register. This allocated rename buffer is the most recently allocated one of fixed point rename buffers 26, and it stores the most recent version of data for the I1 instruction's source register.

Advantageously, fixed point STAM 42 indicates whether a most recently allocated one of fixed point rename buffers 26 stores data for a register number specified in the I1 instruction's R.sub.A register field or R.sub.B register field, even if the I1 instruction is dispatched before or together with the I0 instruction, and even if the I0 instruction's destination register also is a source register for the I1 instruction.

Accordingly, FIG. 5 and FIG. 11 show four comparators 202a-d and their respective outputs 204a-d. Referring to FIG. 11, comparator 202a asserts output 204a in response to a match between I1Ra.sub.-- s and I0Rt.sub.-- d. Similarly, comparator 202b asserts output 204b in response to a match between I1Ra.sub.-- s and I0Ra.sub.-- d. Also, comparator 202c asserts output 204c in response to a match between I1Rb.sub.-- s and I0Rt.sub.-- d. Moreover, comparator 202d asserts output 204d in response to a match between I1Rb.sub.-- s and I0Ra.sub.-- d.

If I0Rt.sub.-- alloc.sub.-- 0 is true (indicating control unit BFR0 is allocating rename buffer 0 to store data for a valid R.sub.T field 64 of an instruction in buffer I0) and output 204a is true (indicating a match between I1Ra.sub.-- s and I0Rt.sub.-- d), then priority assign Knit 200 asserts I1Ra.sub.-- first.sub.-- 0 206a. Similarly, if I0Ra.sub.-- alloc.sub.-- 0 is true (indicating control unit BFR0 is allocating rename buffer 0 to store data for a valid R.sub.A field 66 of an instruction in buffer I0) and output 204b is true (indicating a match between I1Ra.sub.-- s and I0Ra.sub.-- d), then priority assign unit 200 asserts I1Ra.sub.-- first.sub.-- 0 206a.

Also, if I0Rt.sub.-- alloc.sub.-- 0 is true and output 204c is true (indicating a match between I1Rb.sub.-- s and I0Rt.sub.-- d), then priority assign unit 200 asserts I1Rb.sub.-- first.sub.-- 0 206b. Moreover, if I0Ra.sub.-- alloc.sub.-- .sub.-- 0 is true and output 204d is true (indicating a match between I1Rb.sub.-- s and I0Ra.sub.-- d), then priority assign unit 200 asserts I1Rb.sub.-- first.sub.-- 0 206b.

Referring again to FIG. 5, in response to a match between reg.sub.-- adr.sub.-- 0 and I1Ra.sub.-- s, source compares logic 190 asserts a line to an AND gate 192c, so that AND gate 192c asserts I1Ra.sub.-- second.sub.-- 0 208a if MRA bit 166 is set. Moreover, in response to a match between reg.sub.-- adr.sub.-- 0 and I1Rb.sub.-- s, source compares logic 190 asserts a line to an AND gate 192d, so that AND gate 192d asserts I1Rb.sub.-- second.sub.-- 0 208b if MRA bit 166 is set.

As control unit BFR0 outputs lines I1Ra.sub.-- first.sub.-- 0 and I1Rb.sub.-- first.sub.-- 0, likewise each of control units BFR1-BFR4 outputs lines I1Ra.sub.-- first.sub.-- n and I1Rb.sub.-- first.sub.-- n (where n=1 for BFR1, n=2 for BFR2, n=3 for BFR3, and n=4 for BFR4). An OR gate 210a inputs lines I1Ra.sub.-- first.sub.-- 0, I1Ra.sub.-- first.sub.-- 1, I1Ra.sub.-- first.sub.-- 2, I1Ra.sub.-- first.sub.-- 3 and I1Ra.sub.-- first.sub.-- 4. Similarly, an OR gate 210b inputs lines I1Rb.sub.-- first.sub.-- 0, I1Rb.sub.-- first.sub.-- 1, I1Rb.sub.-- first.sub.-- 2, I1Rb.sub.-- first.sub.-- 3 and I1Rb.sub.-- first.sub.-- 4.

If any of lines I1Ra.sub.-- first.sub.-- n (where n=[0 . . . 4]) is true (indicating fixed point STAM 42 is allocating a rename buffer to store data for the I0 instruction's destination register which also is an Ra source register for the I1 instruction), then an output S1 of OR gate 210a is true. If S1 is true, then multiplexer 194a outputs I1Ra.sub.-- first.sub.-- 0 as I1Ra.sub.-- source.sub.-- 0. Otherwise, multiplexer 194a outputs I1Ra.sub.-- second.sub.-- 0 as I1Ra.sub.-- source.sub.-- 0.

Similarly, if any of lines I1Rb.sub.-- first.sub.-- n (where n=[0 . . . 4]) is true (indicating fixed point STAM 42 is allocating a rename buffer to store data for the I0 instruction's destination register which also is an Rb source register for the I1 instruction), then an output S2 of OR gate 210b is true. If S2 is true, then multiplexer 194b outputs I1Rb.sub.-- first.sub.-- 0 as I1Rb.sub.-- source.sub.-- 0. Otherwise, multiplexer 194a outputs I1Rb.sub.-- second.sub.-- 0 as I1Rb.sub.-- source.sub.-- 0.

In this manner, fixed point STAM 42 indicates whether a most recently allocated one of fixed point rename buffers 26 stores data for a register number specified in the I1 instruction's R.sub.A register field or R.sub.B register field, even if the I1 instruction is dispatched before or together with the I0 instruction, and even if the I0 instruction's destination register also is a source register for the I1 instruction.

Referring again to FIG. 1, in response to a writeback signal on line 85a from completion logic 40 and a Write-back address on lines 87a from fixed point STAM 42, data from a selected one of fixed point rename buffers 26 are written back to one of GPR's 30. The write-back address indicates which of GPR's 30 is to store data input from the selected one of fixed point rename buffers 26, according to the register associated by fixed point STAM 42 with the selected one of fixed point rename buffers 26.

Accordingly, referring also to FIG. 5, completion logic 40 outputs 5-bit write-back requests ccc.sub.-- wrt and ccc.sub.-- win through lines 220a and 220b, respectively. Write-back requests ccc.sub.-- wrt 220a and ccc.sub.-- win 220b are input by a write-back arbitration unit 222 of fixed point STAM 42. Further, arbitration unit 222 inputs reg.sub.-- adr.sub.-- 0 from control unit BFR0. Likewise, arbitration unit 222 inputs reg.sub.-- adr.sub.-- 1 from control unit BFR1, reg.sub.-- adr.sub.-- 2 from control unit BFR2, reg.sub.-- adr.sub.-- 3 from control unit BFR3, and reg.sub.-- adr.sub.-- 4 from control unit BFR4.

Advantageously, arbitration unit 222 is able to output two 5-bit write-back addresses, wb0 and wb1, on lines 87a. Through write-back request ccc.sub.-- wrt 220a, completion logic 40 selects two of fixed point rename buffers 26 to be written back to GPR's 30. Each of the 5-bits of ccc.sub.-- wrt 220a is associated with a respective one of rename buffers 0-4 of fixed point rename buffers 26. For example, if ccc.sub.-- wrt 220a=00101, then data from rename buffer 2 and rename buffer 4 are to be written back to GPR's 30. As another example, if ccc.sub.-- wrt 220a=11000, then data from rename buffer 0 and rename buffer 1 are to be written back to GPR's 30.

In response to ccc.sub.-- wrt 220a, arbitration logic 222 multiplexes two of the five reg.sub.-- adr.sub.-- n (where n=[0 . . . 4]) onto lines 87a as wb0 and wb1. GPR's 30 input wb0 and wb1. In this manner, wb0 and wb1 indicate two of GPR's 30 for storing data input from the two fixed point rename buffers selected by completion logic 40. The two indicated GPR's are the registers associated by fixed point STAM 42 with the two selected fixed point rename buffers.

It is possible for write-back request ccc.sub.-- wrt 220a to select two rename buffers storing data for the same GPR (although only one of the two rename buffers is most recently allocated). Through write-back request ccc.sub.-- win 220b, completion logic 40 specifies which of the two ccc.sub.-- wrt-selected rename buffers is to be written back to GPR's 30 in such a situation. Like ccc.sub.-- wrt 220a, each of the 5-bits of ccc.sub.-- win 220b is associated with a respective one of rename buffers 0-4 of fixed point rename buffers 26. For example, if ccc.sub.-- wrt 220a=01010 and ccc.sub.-- wtn 220b=00010, and if reg.sub.-- adr.sub.-- 1=reg.sub.-- adr.sub.-- 3, then arbitration logic 222 multiplexes only reg.sub.-- adr.sub.-- 3 (specified by ccc.sub.-- win 220b) onto lines 87a as wb0. In this example, write-back address wb1 is a null address of 00000.

Although an exemplary embodiment of the present invention and its advantages have been described in detail hereinabove, they have been described as example and not as limitation. Various changes, substitutions and alterations can be made in the exemplary embodiment without departing from the breadth, scope and spirit of the present invention. The breadth, scope and spirit of the present invention should not be limited by the exemplary embodiment, but should be defined only in accordance with the following claims and equivalents thereof.