How it Works
NSC
Loading...
NS32FX164V-20
Datasheet
102 pgs1.17 Mb0

NSC NS32FX164V-20, NS32FX164V-25, NS32FX164AV-25, N32FX164AVLJ-25 Datasheet

TL/EE11267
NS32FX161-15/NS32FX161-20/NS32FX164-20/NS32FX164-25/NS32FV16-20/NS32FV16-25
Advanced Imaging/Communication Signal Processors
February 1992
NS32FX161-15/NS32FX161-20/NS32FX164-20/ NS32FX164-25/NS32FV16-20/NS32FV16-25 Advanced Imaging/Communication Signal Processors
General Description
É
/
EP
TM
family of National’s Embedded System Processors
TM
specifically optimized for CCITT Group 2 and Group 3 Fac­simile Applications, Data Modems, Voice Mail Systems, La­ser Printers, or any combination of the above.
Unless specified otherwise any reference to the NS32FX164 in this document applies to the NS32FV16 and the NS32FX161 as well.
The NS32FX164 can perform all the computations and con­trol functions required for a stand-alone Fax system, a PC add-in Fax/Voice/Data Modem card or a Laser/Fax sys­tem.
It also meets the performance requirements to implement 14400, 9600 and 7200 bps modems complying with CCITT V.17, V.29 and V.27 standards. The NS32FV16 supports V.29 and V.27 standards as well as voice. The NS32FX161 supports V.29 and V.27 standards.
The NS32FX164 provides a 16 Mbyte Linear external ad­dress space and a 16-bit external data bus.
The CPU core, which is the same as that of the NS32CG16, incorporates a 32-bit ALU and instruction pipeline, and an 8-byte prefetch queue.
Also integrated on-chip with the CPU are a DSP Module (DSPM) and a 4K-byte RAM Array (2K in the NS32FV16 and NS32FX161). The DSPM is a complete processing unit, ca­pable of autonomous operation parallel to the CPU core operation. The DSPM executes programs stored in an inter­nal on-chip Random Access Memory (RAM), and manipu­lates data stored either in the internal RAM or in an external off-chip memory. To maximize utilization of hardware re­sources, the DSPM contains a pipelined DSP-oriented data­path, and a control logic that implements a set of DSP vec­tor commands.
The NS32FX164 capabilities can be expanded by using an external floating point unit (FPU) which directly interfaces to the NS32FX164 using the slave protocol. The CPU-FPU cluster features high speed execution of the floating-point instructions.
The NS32FX164 highly-efficient architecture combined with the NS32CG16 graphics instructions and the high-perform­ance vector operation capability, makes the device the ideal choice for Postscript
TM
and Fax applications.
Features
Y
Software compatible with the Series 32000/EP processors
Y
Designed around the CPU core of the NS32CG16
Y
Pin compatible with the NS32FX16
Y
32-bit architecture and implementation
Y
On-chip DSP Module for high-speed DSP operations
Y
Special support for graphics applications Ð 18 graphics instructions Ð Binary compression/expansion capability for font
storage using RLL encoding Ð Pattern magnification Ð Interface to an external BITBLT processing units for
fast color BITBLT operations
Y
4K-byte on-chip RAM array (2K in NS32FV16 and NS32FX161)
Y
On-chip clock generator
Y
Floating-point support via the NS32081 or NS32181
Y
Optimal interface to large memory arrays via the NS32CG821 and the DP84xx family of DRAM controllers
Y
Power save mode
Y
High-speed CMOS technology
Y
68-pin PLCC package
Block Diagram
TL/EE/11267– 1
FIGURE 1-1. CPU Block Diagram
Series 32000Éis a registered trademark of National Semiconductor Corporation. EP
TM
and Embedded System ProcessorsTMare trademarks of National Semiconductor Corporation.
Postscript
TM
is a trademark of Adobe Systems, Inc.
C
1995 National Semiconductor Corporation RRD-B30M115/Printed in U. S. A.
Table of Contents
1.0 PRODUCT INTRODUCTION АААААААААААААААААААААА6
1.1 NS32FX164 Special Features АААААААААААААААААААА6
2.0 ARCHITECTURAL DESCRIPTION ААААААААААААААААА7
2.1 Register Set ААААААААААААААААААААААААААААААААААА7
2.1.1 General Purpose Registers ААААААААААААААААА7
2.1.2 Address Registers ААААААААААААААААААААААААА8
2.1.3 Processor Status Register АААААААААААААААААА8
2.1.4 Configuration Register ААААААААААААААААААААА9
2.1.5 DSP Module Registers ААААААААААААААААААААА9
2.2 Memory Organization АААААААААААААААААААААААААА11
2.2.1 Address MappingААААААААААААААААААААААААА12
2.3 Modular Software Support АААААААААААААААААААААА12
2.4 Instruction Set АААААААААААААААААААААААААААААААА12
2.4.1 General Instruction Format АААААААААААААААА12
2.4.2 Addressing ModesАААААААААААААААААААААААА14
2.4.3 Instruction Set Summary АААААААААААААААААА16
2.5 Graphics SupportАААААААААААААААААААААААААААААА20
2.5.1 Frame Buffer Addressing АААААААААААААААААА20
2.5.2 BITBLT Fundamentals АААААААААААААААААААА20
2.5.2.1 Frame Buffer ArchitectureААААААААААА21
2.5.2.2 Bit Alignment АААААААААААААААААААААА21
2.5.2.3 Block Boundaries and Destination MasksАААААААААААААААААААААААААААА21
2.5.2.4 BITBLT Directions ААААААААААААААААА22
2.5.2.5 BITBLT Variations ААААААААААААААААА23
2.5.3 Graphics Support Instructions АААААААААААААА23
2.5.3.1 BITBLT (BIT-aligned BLock Transfer)À23
2.5.3.2 Pattern Fill АААААААААААААААААААААААА24
2.5.3.3 Data Compression, Expansion and MagnifyААААААААААААААААААААААААААА24
2.5.3.3.1 Magnifying Compressed Data ААААААААААААААААААААА26
3.0 FUNCTIONAL DESCRIPTION АААААААААААААААААААА26
3.1 Instruction Execution АААААААААААААААААААААААААА26
3.1.1 Operating States ААААААААААААААААААААААААА26
3.1.2 Instruction Endings ААААААААААААААААААААААА26
3.1.2.1 Completed Instructions ААААААААААААА27
3.1.2.2 Suspended InstructionsААААААААААААА27
3.1.2.3 Terminated InstructionsААААААААААААА27
3.1.2.4 Partially Completed Instructions ААААА27
3.1.3 Slave Processor Instructions ААААААААААААААА27
3.1.3.1 Slave Processor Protocol ААААААААААА27
3.1.3.2 Floating-Point Instructions АААААААААА28
3.2 Exception Processing АААААААААААААААААААААААААА29
3.2.1 Exception Acknowledge Sequence ААААААААА29
3.2.2 Returning from an Exception Service Procedure ААААААААААААААААААААААААААААААА30
3.2.3 Maskable InterruptsААААААААААААААААААААААА34
3.2.3.1 Non-Vectored Mode ААААААААААААААА34
3.2.3.2 Vectored Mode: Non-Cascaded
Case ААААААААААААААААААААААААААААА35
3.2.3.3 Vectored Mode: Cascaded Case ААААА35
3.2.4 Non-Maskable Interrupt ААААААААААААААААААА37
3.2.5 Traps ААААААААААААААААААААААААААААААААААА37
3.2.6 Priority among Exceptions ААААААААААААААААА37
3.2.7 Exception Acknowledge Sequences: Detailed Flow АААААААААААААААААААААААААААААААААААА39
3.2.7.1 Maskable/Non-Maskable Interrupt
Sequence АААААААААААААААААААААААА39
3.2.7.2 SLAVE/ILL/SVC/DVZ/FLG/BPT/UND
Trap Sequence АААААААААААААААААААА39
3.2.7.3 Trace Trap Sequence АААААААААААААА39
3.3 Debugging Support АААААААААААААААААААААААААААА40
3.3.1 Instruction TracingАААААААААААААААААААААААА40
3.4 DSP Module АААААААААААААААААААААААААААААААААА40
3.4.1 Programming Model АААААААААААААААААААААА40
3.4.2 RAM Organization and Data Types ААААААААА41
3.4.2.1 Integer ValuesААААААААААААААААААААА41
3.4.2.2 Aligned-Integer Values ААААААААААААА41
3.4.2.3 Real Values ААААААААААААААААААААААА41
3.4.3.4 Aligned-Real Values ААААААААААААААА41
3.4.2.5 Extended Precision Real Values ААААА41
3.4.2.6 Complex Values ААААААААААААААААААА42
3.4.3 Command List Format АААААААААААААААААААА42
3.4.4 CPU Core Interface ААААААААААААААААААААААА42
3.4.4.1 Synchronization of Parallel OperationÀ42
3.4.4.2 DSPM RAM Organization ААААААААААА43
3.4.5 DSPM Instruction Set ААААААААААААААААААААА43
3.4.5.1 Conventions АААААААААААААААААААААА43
3.4.5.2 Type Casting АААААААААААААААААААААА43
3.4.5.3 General NotesААААААААААААААААААААА44
3.4.5.4 Load Register Instructions АААААААААА44
3.4.5.5 Store Register Instructions АААААААААА45
3.4.5.6 Adjust Register Instructions ААААААААА46
3.4.5.7 Flow Control Instructions ААААААААААА47
3.4.5.8 Internal Memory Move Instructions ÀÀÀ48
3.4.5.9 External Memory Move Instructions ÀÀ48
3.4.5.10 Arithmetic/Logical Instructions ААААА49
3.4.5.11 Multiply-and-Accumulate
Instructions АААААААААААААААААААААА49
3.4.5.12 Multiply-and-Add InstructionsААААААА50
3.4.5.13 Clipping and Min/Max Instructions ÀÀ52
3.4.5.14 Special Instructions ААААААААААААААА53
2
Table of Contents (Continued)
3.5 System Interface АААААААААААААААААААААААААААААА55
3.5.1 Power and Grounding ААААААААААААААААААААА55
3.5.2 Clocking АААААААААААААААААААААААААААААААА56
3.5.3 Power Save Mode АААААААААААААААААААААААА57
3.5.4 ResettingАААААААААААААААААААААААААААААААА57
3.5.5 Bus Cycles АААААААААААААААААААААААААААААА58
3.5.5.1 Bus Status АААААААААААААААААААААААА58
3.5.5.2 Basic Read and Write Cycles АААААААА58
3.5.5.3 Cycle Extension ААААААААААААААААААА62
3.5.5.4 Instruction Fetch Cycles АААААААААААА63
3.5.5.5 Interrupt Control CyclesААААААААААААА64
3.5.5.6 Special Bus CyclesААААААААААААААААА65
3.5.5.7 Slave Processor Bus CyclesААААААААА65
3.5.5.8 Data Access Sequences АААААААААААА67
3.5.5.9 Bus Access Control АААААААААААААААА68
3.5.5.10 Instruction Status ААААААААААААААААА71
4.0 DEVICE SPECIFICATIONS АААААААААААААААААААААА71
4.1 NS32FX164 Pin Descriptions ААААААААААААААААААА71
4.1.1 Supplies АААААААААААААААААААААААААААААААА71
4.1.2 Input SignalsААААААААААААААААААААААААААААА71
4.1.3 Output Signals ААААААААААААААААААААААААААА71
4.1.4 Input-Output Signals АААААААААААААААААААААА72
4.2 Absolute Maximum Ratings ААААААААААААААААААААА74
4.3 Electrical Characteristics ААААААААААААААААААААААА74
4.4 Switching Characteristics ААААААААААААААААААААААА74
4.4.1 Definitions ААААААААААААААААААААААААААААААА74
4.4.2 Timing TablesАААААААААААААААААААААААААААА75
4.4.2.1 Output Signals: Internal Propagation Delays ААААААААААААААААААААААААААА75
4.4.2.2 Input Signal Requirements АААААААААА77
4.4.3 Timing Diagrams ААААААААААААААААААААААААА79
APPENDIX A: INSTRUCTION FORMATS ААААААААААААА89
APPENDIX B: INSTRUCTION EXECUTION TIMESААААА92
B.1 Basic and Floating-Point Instructions АААААААААААА92
B.1.1 Equations ААААААААААААААААААААААААААААААА92
B.1.2 Notes on Table Use АААААААААААААААААААААА93
B.1.3 Calculation of the Execution Time TEX for Basic
Instructions ААААААААААААААААААААААААААААА93
B.1.4 Calculation of the Execution Time TEX for
Floating-Point InstructionsААААААААААААААААА93
B.2 Special Graphics Instructions ААААААААААААААААААА99
B.2.1 Execution Time Calculation for Special
Graphics Instructions ААААААААААААААААААААА99
B.3 DSPM Instructions ААААААААААААААААААААААААААА100
List of Figures
FIGURE 1-1. CPU Block Diagram АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА1
FIGURE 2-1. NS32FX164 Internal Registers ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА7
FIGURE 2-2. Processor Status Register (PSR) ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА8
FIGURE 2-3. Configuration Register (CFG) АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА9
FIGURE 2-4. DSP Module Registers Address MapАААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА9
FIGURE 2-5. Accumulator Format АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА9
FIGURE 2-6. X, Y, Z Registers Format АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА9
FIGURE 2-7. EABR Register FormatААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА10
FIGURE 2-8. OVF Register Format АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА10
FIGURE 2-9. PARAM Register Format ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА10
FIGURE 2-10. REPEAT Register Format АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА10
FIGURE 2-11. EXT Register Format АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА11
FIGURE 2-12. CLSTAT Register Format АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА11
FIGURE 2-13. DSPINT and DSPMASK Register Format ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА11
FIGURE 2-14. NMISTAT Register Format ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА11
FIGURE 2-15. NS32FX164 Address Mapping АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА12
FIGURE 2-16. NS32FX164 Run-Time Environment ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА13
FIGURE 2-17. General Instruction Format ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА13
FIGURE 2-18. Index Byte FormatААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА13
FIGURE 2-19. Displacement Encodings ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА14
FIGURE 2-20. Correspondence between Linear and Cartesian Addressing АААААААААААААААААААААААААААААААААААААААААААА20
FIGURE 2-21. 32-Pixel by 32-Scan Line Frame Buffer ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА21
FIGURE 2-22. Overlapping BITBLT Blocks АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА22
FIGURE 2-23. BB Instructions Format АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА23
FIGURE 2-24. BITWT Instruction Format АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА24
FIGURE 2-25. EXTBLT Instruction Format ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА24
FIGURE 2-26. MOVMPi Instruction Format АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА24
3
List of Figures (Continued)
FIGURE 2-27. TBITS Instruction FormatААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА24
FIGURE 2-28. SBITS Instruction Format АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА25
FIGURE 2-29. SBITPS Instruction Format ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА25
FIGURE 2-30. Bus Activity for a Simple BITBLT Operation ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА25
FIGURE 3-1. Operating States АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА26
FIGURE 3-2. Slave Processor Protocol АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА28
FIGURE 3-3. Slave Processor Status Word ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА29
FIGURE 3-4. Interrupt Dispatch and Cascade Tables АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА30
FIGURE 3-5. Exception Acknowledge Sequence: Direct-Exception Mode Disabled АААААААААААААААААААААААААААААААААААА31
FIGURE 3-6. Exception Acknowledge Sequence: Direct-Exception Mode Enabled ААААААААААААААААААААААААААААААААААААА32
FIGURE 3-7. Return from Trap (RETTn) Instruction Flow: Direct-Exception Mode Disabled ААААААААААААААААААААААААААААА33
FIGURE 3-8. Return from Interrupt (RETI) Instruction Flow: Direct-Exception Mode Disabled АААААААААААААААААААААААААААА34
FIGURE 3-9. Interrupt Control Unit Connections (16 Levels) АААААААААААААААААААААААААААААААААААААААААААААААААААААААА35
FIGURE 3-10. Cascaded Interrupt Control Unit Connections ААААААААААААААААААААААААААААААААААААААААААААААААААААААААА36
FIGURE 3-11. Exception Processing Flowchart АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА38
FIGURE 3-12. Service Sequence ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА39
FIGURE 3-13. DSP Module Block Diagram АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА55
FIGURE 3-14. Power and Ground ConnectionsААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА56
FIGURE 3-15. Crystal InterconnectionsР30 MHz АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА56
FIGURE 3-16. Crystal InterconnectionsР40 MHz, 50 MHz ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА56
FIGURE 3-17. Recommended Reset Connections АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА56
FIGURE 3-18. Power-On Reset Requirements ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА57
FIGURE 3-19. General Reset TimingАААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА57
FIGURE 3-20. Bus Connections АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА59
FIGURE 3-21. Read Cycle Timing АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА60
FIGURE 3-22. Write Cycle Timing АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА61
FIGURE 3-23. Cycle Extension of a Read Cycle АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА63
FIGURE 3-24. Special Bus Cycle Timing АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА65
FIGURE 3-25. Slave Processor Read Cycle АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА66
FIGURE 3-26. Slave Processor Write Cycle АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА67
FIGURE 3-27. NS32FX164 and FPU Interconnections АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА67
FIGURE 3-28. Memory Interface ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА67
FIGURE 3-29. HOLD
Timing (Bus Initially Idle) ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА69
FIGURE 3-30. HOLD Timing (Bus Initially Not Idle) АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА70
FIGURE 4-1. Connection DiagramААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА73
FIGURE 4-2. Output Signals Specification Standard ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА74
FIGURE 4-3a. Input Signals Specification Standard ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА74
FIGURE 4-3b. RSTI
, INT, NMI HysteresisАААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА74
FIGURE 4-4. Read CycleААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА79
FIGURE 4-5. Write Cycle ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА80
FIGURE 4-6. Special Bus Cycle ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА81
FIGURE 4-7. HOLD
Acknowledge Timing (Bus Initially Not Idle) ААААААААААААААААААААААААААААААААААААААААААААААААААААА82
FIGURE 4-8. HOLD Timing (Bus Initially Idle) ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА83
FIGURE 4-9. External DMA Controller Bus Cycle АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА84
FIGURE 4-10. Slave Processor Write Timing ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА85
FIGURE 4-11. Slave Processor Read Timing ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА85
FIGURE 4-12. SPC
Timing АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА85
FIGURE 4-13. PFS Signal Timing ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА86
FIGURE 4-14. ILO Signal Timing ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА86
FIGURE 4-15. Clock Waveforms ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА86
FIGURE 4-16. INT
Signal Timing ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА87
4
List of Figures (Continued)
FIGURE 4-17. NMI
Signal TimingААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА87
FIGURE 4-18. Power-On Reset АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА87
FIGURE 4-19. Non-Power-On Reset АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА88
FIGURE 4-20. Interrupt Out АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА88
List of Tables
TABLE 2-1. NS32FX164 Addressing Modes ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА15
TABLE 2-2. NS32FX164 Instruction Set Summary АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА16
TABLE 2-3. ‘op’ and ‘i’ Field Encodings ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА23
TABLE 3-1. Floating-Point Instruction Protocols АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА28
TABLE 3-2. Summary of Exception ProcessingААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА40
TABLE 3-3. External Oscillator Specifications Crystal Characteristics ААААААААААААААААААААААААААААААААААААААААААААААААА57
TABLE 3-4. Interrupt Sequences ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА64
TABLE 3-5. Bus Cycle Categories АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА67
TABLE 3-6. Data Access Sequences ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА68
TABLE B-1. Basic Instructions ААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА94
TABLE B-2. Floating-Point Instructions: CPU Portion АААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААААА98
TABLE B-3. Average Instruction Execution Times with No Wait-States АААААААААААААААААААААААААААААААААААААААААААААААА99
TABLE B-4. Average Instruction Execution Times with Wait-States АААААААААААААААААААААААААААААААААААААААААААААААААА100
5
1.0 Product Introduction
The NS32FX164 is a high speed CMOS microprocessor in the Series 32000/EP family.
It includes two main execution units: the NS32CG16 com­patible CPU core and the DSP Module. The CPU core is designed for general purpose computations and system control functions. The DSP Module is tuned to perform the DSP primitives needed in Voice Band Modems. The NS32FX164 also incorporates a 4K-byte RAM Array as a shared resource for both the CPU core and the DSP Module (2K-byte in the NS32FV16 and the NS32FX161).
The NS32FX164 is software-compatible with all other CPUs in the family.
The device incorporates all of the Series 32000 advanced architectural features, with the exception of the virtual mem­ory capability.
Brief descriptions of the NS32FX164 features that are shared with other members of the family are provided be­low:
Powerful Addressing Modes. Nine addressing modes available to all instructions are included to access data structures efficiently.
Data Types. The architecture provides for numerous data types, such as byte, word, doubleword, and BCD, which may be arranged into a wide variety of data structures.
Symmetric Instruction Set. While avoiding special case instructions that compilers can’t use, the Series 32000 fami­ly incorporates powerful instructions for control operations, such as array indexing and external procedure calls, which save considerable space and time for compiled code.
Memory-to-Memory Operations. The Series 32000 CPUs represent two-address machines. This means that each op­erand can be referenced by any one of the addressing modes provided.
This powerful memory-to-memory architecture permits memory locations to be treated as registers for all useful operations. This is important for temporary operands as well as for context switching.
Large, Uniform Addressing. The NS32FX164 has 24-bit address pointers that can address up to 16 megabytes with­out any segmentation; this addressing scheme provides flexible memory management without add-on expense.
Modular Software Support. Any software package for the Series 32000 architecture can be developed independent of all other packages, without regard to individual addressing. In addition, ROM code is totally relocatable and easy to access, which allows a significant reduction in hardware and software cost.
Software Processor Concept. The Series 32000 architec­ture allows future expansions of the instruction set that can be executed by special slave processors, acting as exten-
sions to the CPU. This concept of slave processors is unique to the Series 32000 architecture. It allows software compatibility even for future components because the slave hardware is transparent to the software. With future ad­vances in semiconductor technology, the slaves can be physically integrated on the CPU chip itself.
To summarize, the architectural features cited above pro­vide three primary performance advantages and character­istics:
#
High-Level Language Support
#
Easy Future Growth Path
#
Application Flexibility
1.1 NS32FX164 SPECIAL FEATURES
In addition to the above Series 32000 features, the NS32FX164 provides features that make the device ex­tremely attractive for a wide range of applications where graphics support, low chip count, and low power consump­tion are required.
Graphics support is provided by eighteen instructions that allow operations such as BITBLT, data compression/expan­sion, fills, and line drawing, to be performed very efficiently. In addition, the device can be easily interfaced to an exter­nal BITBLT Processing Unit (BPU) for high BITBLT perform­ance.
The NS32FX164 allows systems to be built with a relatively small amount of random logic. The bus is highly optimized to allow simple interfacing to a large variety of DRAMs and peripheral devices. All the relevant bus access signals and clock signals are generated on-chip. The cycle extension logic is also incorporated on-chip.
The device is fabricated in a low-power, high speed CMOS technology. It also includes a power-save feature that al­lows the clock to be slowed down under software control, thus minimizing the power consumption. This feature can be used in those applications where power saving during peri­ods of low performance demand is highly desirable.
The power save feature, the DSP Module and the Bus Char­acteristics are described in the ‘‘Functional Description’’ section. A general overview of BITBLT operations and a description of the graphics support instructions is provided in Section 2.5. Details on all the NS32FX164 graphics in­structions can be found in the NS32CG16 Printer/Display Processor Programmer’s Reference Supplement.
6
1.0 Product Introduction (Continued)
Below is a summary of the instructions that are directly ap­plicable to graphics along with their intended use.
Instruction Application
BBAND The BITBLT group of instructions provide a BBOR method of quickly imaging characters, BBFOR creating patterns, windowing and other BBXOR block oriented effects. BBSTOD BITWT EXTBLT
MOVMP Move Multiple Pattern is a very fast
instruction for clearing memory and drawing patterns and lines.
TBITS Test Bit String will measure the length of 1’s
or 0’s in an image, supporting many data compression methods (RLL), TBITS may also be used to test for boundaries of images.
SBITS Set Bit String is a very fast instruction for
filling objects, outline characters and drawing horizontal lines. The TBITS and SBITS instructions support Group 3 and Group 4 CCITT standards for compression and decompression algorithms.
SBITPS Set Bit Perpendicular String is a very fast
instruction for drawing vertical, horizontal and 45
§
lines. In printing applications SBITS and SBITPS may be used to express portrait and landscape respectively from the same compressed font data. The size of the character may be scaled as it is drawn.
SBIT The Bit group of instructions enable single CBIT pixels anywhere in memory to be set, TBIT cleared, tested or inverted. IBIT
INDEX The INDEX instruction combines a multiply-
add sequence into a single instruction. This provides a fast translation of an X-Y address to a pixel relative address.
2.0 Architectural Description
2.1 REGISTER SET
The NS32FX164 has 32 internal registers. 17 of these regis­ters belong to the CPU portion of the device and are ad­dressed either implicitly by specific instructions or through the register addressing mode. The other 15 control the op­eration of the DSP Module, and are memory mapped.
Figure
2-1
shows the NS32FX164 internal registers.
CPU Registers
General Purpose
w
32 Bits
x
R0–R7
Address
PC
SP0, SP1
FP
SB
INTBASE
MOD
Processor Status
PSR
Configuration
CFG
Peripherals Registers
DSP Module
A
X
Y
Z
EABR
CLPTR
OVF
PARAM
REPEAT
ABORT
EXT
CLSTAT
DSPINT
DSPMASK
NMISTAT
FIGURE 2-1. NS32FX164 Internal Registers
2.1.1 General Purpose Registers
There are eight registers (R0 – R7) used for satisfying the high speed general storage requirements, such as holding temporary variables and addresses. The general purpose registers are free for any use by the programmer. They are 32 bits in length. If a general purpose register is specified for an operand that is 8 or 16 bits long, only the low part of the register is used; the high part is not referenced or modified.
7
2.0 Architectural Description (Continued)
2.1.2 Address Registers
The seven address registers are used by the processor to implement specific address functions. Except for the MOD register that is 16 bits wide, all the others are 32 bits. A description of the address registers follows.
PCÐProgram Counter. The PC register is a pointer to the first byte of the instruction currently being executed. The PC is used to reference memory in the program section.
When a reference is made to the selected Stack Pointer (see PSR S-bit), the terms ‘‘SP Register’’ or ‘‘SP’’ are used. SP refers to either SP0 or SP1, depending on the setting of the S bit in the PSR register. If the S bit in the PSR is 0, SP refers to SP0. If the S bit in the PSR is 1 then SP refers to SP1.
Stacks in the Series 32000 architecture grow downward in memory. A Push operation pre-decrements the Stack Point­er by the operand length. A Pop operation post-increments the Stack Pointer by the operand length.
FPÐFrame Pointer. The FP register is used by a procedure to access parameters and local variables on the stack. The FP register is set up on procedure entry with the ENTER instruction and restored on procedure termination with the EXIT instruction.
The frame pointer holds the address in memory occupied by the old contents of the frame pointer.
SBÐStatic Base. The SB register points to the global vari­ables of a software module. This register is used to support relocatable global variables for software modules. The SB register holds the lowest address in memory occupied by the global variables of a module.
INTBASEÐInterrupt Base. The INTBASE register holds the address of the dispatch table for interrupts and traps (Section 3.2.1).
MODÐModule. The MOD register holds the address of the module descriptor of the currently executing software mod­ule. The MOD register is 16 bits long, therefore the module table must be contained within the first 64 kbytes of memo­ry.
2.1.3 Processor Status Register
The Processor Status Register (PSR) holds status informa­tion for the microprocessor.
The PSR is sixteen bits long, divided into two eight-bit halves. The low order eight bits are accessible to all pro­grams, but the high order eight bits are accessible only to programs executing in Supervisor Mode.
15 8 7 0
BIPSUNZFJKLTC
FIGURE 2-2. Processor Status Register (PSR)
C The C bit indicates that a carry or borrow occurred after
an addition or subtraction instruction. It can be used with the ADDC and SUBC instructions to perform multiple­precision integer arithmetic calculations. It may have a setting of 0 (no carry or borrow) or 1 (carry or borrow).
T The T bit causes program tracing. If this bit is set to 1, a
TRC trap is executed after every instruction (Section
3.3.1).
L The L bit is altered by comparison instructions. In a com-
parison instruction the L bit is set to ‘‘1’’ if the second operand is less than the first operand, when both oper­ands are interpreted as unsigned integers. Otherwise, it is set to ‘‘0’’. In Floating-Point comparisons, this bit is always cleared.
K Reserved for use by the CPU.
J Reserved for use by the CPU.
F The F bit is a general condition flag, which is altered by
many instructions (e.g., integer arithmetic instructions use it to indicate overflow).
Z The Z bit is altered by comparison instructions. In a com-
parison instruction the Z bit is set to ‘‘1’’ if the second operand is equal to the first operand; otherwise it is set to ‘‘0’’.
N The N bit is altered by comparison instructions. In a
comparison instruction the N bit is set to ‘‘1’’ if the sec­ond operand is less than the first operand, when both operands are interpreted as signed integers. Otherwise, it is set to ‘‘0’’.
U If the U bit is ‘‘1’’ no privileged instructions may be exe-
cuted. If the U bit is ‘‘0’’ then all instructions may be executed. When U
e
0 the processor is said to be in Su-
pervisor Mode; when U
e
1 the processor is said to be in User Mode. A User Mode program is restricted from exe­cuting certain instructions and accessing certain regis­ters which could interfere with the operating system. For example, a User Mode program is prevented from changing the setting of the flag used to indicate its own privilege mode. A Supervisor Mode program is assumed to be a trusted part of the operating system, hence it has no such restrictions.
S The S bit specifies whether the SP0 register or SP1 reg-
ister is used as the Stack Pointer. The bit is automatical­ly cleared on interrupts and traps. It may have a setting of 0 (use the SP0 register) or 1 (use the SP1 register).
P The P bit prevents a TRC trap from occurring more than
once for an instruction (Section 3.3.1). It may have a setting of 0 (no trace pending) or 1 (trace pending).
I If I
e
1, then all interrupts will be accepted. If Ie0, only the NMI interrupt is accepted. Trap enables are not af­fected by this bit.
8
2.0 Architectural Description (Continued)
B Reserved for use by the CPU. This bit is set to 1 during
the execution of the EXTBLT instruction and causes the BPU
signal to become active. Upon reset, B is set to
zero and the BPU
signal is set high.
Note 1: When an interrupt is acknowledged, the B, I, P, S and U bits are set
to zero and the BPU
signal is set high. A return from interrupt will restore the original values from the copy of the PSR register saved in the interrupt stack.
Note 2: If BITBLT (BB) or EXTBLT instructions are executed in an interrupt
routine, the PSR bits J and K must be cleared first.
2.1.4 Configuration Register
The Configuration Register (CFG) is 32 bits wide, of which 5 bits are implemented. The implemented bits enable various operating modes for the CPU, including vectoring of inter­rupts, execution of floating-point instructions, processing of exceptions and selection of clock scaling factor. The CFG is programmed by the SETCFG instruction. The format of CFG is shown in
Figure 2-3
. The various control bits are de-
scribed below.
31 8 7 0
Reserved DE Res C M F I
FIGURE 2-3. Configuration Register (CFG)
I Interrupt vectoring. This bit controls whether maskable
interrupts are handled in nonvectored (I
e
0) or vec-
tored (I
e
1) mode. Refer to Section 3.2.3 for more in-
formation.
F Floating-point instruction set. This bit indicates wheth-
er a floating-point unit (FPU) is present to execute floating-point instructions. If this bit is 0 when the CPU executes a floating-point instruction, a Trap (UND) oc­curs. If this bit is 1, then the CPU transfers the instruc­tion and any necessary operands to the FPU using the slave-processor protocol described in Section 3.1.3.1.
M Clock scaling. This bit is used in conjunction with the
C-bit to select the clock scaling factor.
C Clock scaling. Same as the M-bit above. Refer to Sec-
tion 3.5.3 on ‘‘Power Save Mode’’ for details.
DE Direct-Exception mode enable. This bit enables the Di-
rect-Exception mode for processing exceptions. When this mode is selected, the CPU response time to inter­rupts and other exceptions is significantly improved. Refer to Section 3.2 for more information.
2.1.5 DSP Module Registers
The DSP Module (DSPM) contains 15 memory-mapped reg­isters. All the registers, except OVF, CLSTAT, ABORT, DSPINT and NMISTAT, are readable and writable. OVF, CLSTAT, DSPINT and NMISTAT are read-only. ABORT is write-only.
The DSPM registers are divided into two groups, according to their function. PARAM, OVF, X, Y, Z, A, REPEAT, CLPTR and EABR are called DSPM dedicated registers. CLSTAT, ABORT, DSPINT, DSPMASK, EXT and NMISTAT are called CPU core interface registers.
Accesses to these registers must be aligned; word and dou­ble-word accesses must occur on word and double-word address boundaries respectively. Failing to do so will cause unpredictable results.
Figure 2-4
shows the address map of
the DSP Module registers.
Register Register
Name Address
PARAM FFFF8000
OVF FFFF8004
X FFFF8008
Y FFFF800C
Z FFFF8010
A FFFF8014
REPEAT FFFF8018
CLPTR FFFF8020
EABR FFFF8024
CLSTAT FFFF9000
ABORT FFFF9004
DSPINT FFFF9008
DSPMASK FFFF900C
EXT FFFF9010
NMISTAT FFFF9014
FIGURE 2-4. DSP Module Registers Address Map
AÐAccumulator
The format of the accumulator is shown in
Figure 2-5
.
33 0 33 0
Imaginary Real
FIGURE 2-5. Accumulator Format
The A register is a complex accumulator. It has two 34-bit fields: a real part, and an imaginary part. Bits 15 through 30 of the real and the imaginary parts of the accumulator can be read or written by the core in one double-word access. Bits 15 through 30 of the real part are mapped to the oper­and’s bits 0 through 15, and bits 15 through 30 of the imagi­nary part are mapped to the operand’s bits 16 through 31. The accumulator can also be read and written by the com­mand-list execution unit using the SA, SEA, LA and LEA instructions (See Section 3.4 for more information).
Note that when a value is stored in the accumulator by the core, the value of PARAM.RND bit is copied into bit position 14 of both real and imaginary parts of the accumulator. This technique allows rounding of the accumulator’s value in the following DSPM instructions (See Section 3.4.5.3 for more information on rounding).
When the Accumulator is loaded either by the core or by the LA or LEA instructions, bits 31– 33 of the real and the imagi­nary accumulators are loaded with the values of bit 30 of the real and the imaginary parts respectively.
When the Accumulator is loaded either by the core or by the LA instruction, bits 0 – 13 of the real and the imaginary accu­mulators are loaded with zeros.
X, Y, ZÐVector Pointers
The format of X, Y, and Z registers is shown in
Figure 2-6
.
31 16 15 8 7 4 3 0
ADDRESS Reserved WRAP-AROUND INCREMENT
FIGURE 2-6. X, Y, Z Registers Format
9
2.0 Architectural Description (Continued)
The X, Y, and Z registers are used for addressing up to three vector operands. They are 32-bit registers, with three fields: ADDRESS, INCREMENT, and WRAP-AROUND. The value in the ADDRESS field specifies the address of a word in the on-chip memory. This field has 16 bits, and can ad­dress up to 64 Kwords of internal memory. The ADDRESS fields are initialized with the vector operands’ start-address­es by commands in the command list. At the beginning of each vector operation, the contents of the ADDRESS field are copied to incrementors. Increments can be used by vec­tor instructions to step through the corresponding vector operands while executing the appropriate calculations. There is an address wrap-around for those vector instruc­tions that require some of their operands to be located in cyclic buffers. The allowed values for the increment field are 0 through 15. The actual increment will be 2
increment
words. The allowed values for the WRAP-AROUND field are 0 through 15. The actual wrap-around will be 2
WRAP-AROUND
words. The WRAP-AROUND must be greater or equal to the INCREMENT.
The X, Y, and Z registers can be read and written by the core. These registers can be read and written by the com­mand-list execution unit, as well as by the core, when using SX, SXL, SXH, SY, SZ, LX, LY and LZ instructions.
EABRÐExternal Address Base Register
The format of the external address base register is shown in
Figure 2-7
.
31 17 16 0
ADDRESS 0
FIGURE 2-7. EABR Register Format
The EABR register is used together with a 16-bit address field to form a 32-bit external address. External addresses are specified as the sum of the value in EABR and two times the value of the 16-bit address pointed by registers X, Y or Z. The only value allowed to be written into bits 0 through 16 of EABR is ‘‘0’’. The EABR register can be read and written by the core. It can also be written by the command-list exe­cution unit by using the LEABR instruction.
EABR can hold any value except for FFFE0000. Accessing external memory with an FFFE0000 in the EABR will cause unpredictable results.
CLPTRÐCommand List Pointer
The CLPTR is a 16-bit register that holds the address of the current command in the internal RAM. Writing into the CLPTR causes the DSPM command-list execution unit to begin executing commands, starting from the address in CLPTR. The CLPTR can be read and written by the core while the command-list execution is idle.
Whenever the DSPM command-list execution unit reads a command from the DSPM RAM, the value of CLPTR is up­dated to contain the address of the next command to be executed. This implies, for example, that if the last com­mand in a list is in address N, the CLPTR will hold a value of N
a
1 following the end of command list execution.
OVFÐOverflow Register
The format of the overflow register is shown in
Figure 2-8
.
15 2 1 0
Reserved OVF SAT
FIGURE 2-8. OVF Register Format
The OVF register holds the current status of the DSPM arithmetic unit. It has two fields: OVF and SAT. The OVF bit is set to ‘‘1’’ whenever an overflow is detected in the DSPM 34-bit ALU (e.g., bits 32 and 33 of the ALU are not equal). No overflow detection is provided for integers. The SAT bit is set to ‘‘1’’ whenever a value read from the accumulator cannot be represented within the limits of its data type (e.g., 16 bits for real and integer, and 31 bits for extended real). In this case the value read from the accumulator will either be the maximum allowed value or the minimal allowed value for this data type depending on the sign of the accumulator value. Note that in some cases when the OVF is set, the SAT will not be set. The reason is that if an OVF occurred, the value in the accumulator can no longer be used for proper SAT detection. Upon reset, and whenever the ABORT register is written, the non reserved bits of the OVF register is cleared to ‘‘0’’.
The OVF is a read only register. It can be read by the core. It can also be read by the command-list execution unit using the SOVF instruction. Reading the OVF by either the core or the command-list execution unit clears it to ‘‘0’’.
PARAMÐVector Parameter Register
The format of the PARAM register is shown in Figure 2-9.
31 26 25 24 19 18 17 16 15 0
Reserved RND OP SUB CLR COJ LENGTH
FIGURE 2-9. PARAM Register Format
The PARAM register is used to specify the number of itera­tions and special options for the various instructions. The options are: RND, OP, SUB, CLR, and COJ. The effect of each of the bits of the PARAM register is specified in Sec­tion 3.4.
The PARAM register can be read and written by the core. It can also be written by the command-list execution unit, by using the LPARAM instruction. The value written into PAR­AM.LENGTH must be greater then 0.
The value of PARAM.LENGTH is not changed during com­mand-list execution, unless it is written into using the LPARAM instruction.
REPEATÐCommand-List Repeat Register
The format of the repeat register is shown in
Figure 2-10
.
31 16 15 0
COUNT TARGET
FIGURE 2-10. REPEAT Register Format
The REPEAT register is used, together with appropriate commands, to implement loops and branches in the com­mand list. The count is used to specify the number of times a loop in the command list is to be repeated. The target is used to specify a jump address within the command list.
The REPEAT register can be read and written by the core. It can also be read and written by the command-list execution unit by using SREPEAT and LREPEAT instructions respec­tively.
The value of REPEAT.COUNT changes during the execu­tion of the DJNZ command.
ABORTÐAbort Register
The ABORT register is used to force execution of the com­mand list to halt. Writing any value into this register stops execution, and clears the contents of OVF, EXT, DSPINT and DSPMASK. The ABORT register can only be written and only by the core.
10
2.0 Architectural Description (Continued)
EXTÐExternal Memory Reference Control Register
The format of the external memory reference control regis­ter is shown in
Figure 2-11
.
15 1 0
Reserved HOLD
FIGURE 2-11. EXT Register Format
The EXT register controls external references. The com­mand-list execution unit checks the value of EXT.HOLD be­fore each external memory reference. When EXT.HOLD is ‘‘0’’, external memory references are allowed. When EXT.HOLD is ‘‘1’’, and external memory references are re­quested, the execution of the command list will stop until EXT.HOLD is ‘‘0’’. Upon reset, and whenever the ABORT register is written, EXT.HOLD is cleared to ‘‘0’’. The EXT register can be read or written by the core.
CLSTATÐCommand-List Execution Status Register
The format of the command-list execution status register is shown in
Figure 2-12
.
15 1 0
Reserved RUN
FIGURE 2-12. CLSTAT Register Format
The CLSTAT register displays the current status of the exe­cution of the command list. When the command-list execu­tion is idle, CLSTAT.RUN is ‘‘0’’, and when it is active, CLSTAT.RUN is ‘‘1’’. Upon reset, the CLSTAT register is cleared to ‘‘0’’. It can only be read, and only by the core.
DSPINT, DSPMASK, NMISTATÐInterrupt Control Registers
The format of DSPINT and DSPMASK is shown in
Figure
2-13
.
15 1 0
Reserved HALT
FIGURE 2-13. DSPINT and DSPMASK Register Format
The DSPINT register holds the current status of interrupt requests. Whenever execution of the command list is stopped, the DSPINT.HALT bit is set to ‘‘1’’. The DSPINT is a read only register. It is cleared to ‘‘0’’ whenever it is read, whenever the ABORT register is written, and upon reset.
The DSPMASK register is used to mask the DSPINT. HALT flag. An interrupt request is transferred to the interrupt logic of the IOUT
output pin whenever the DSPINT.HALT bit is set to ‘‘1’’, and the DSPMASK.HALT bit is unmasked (set to ‘‘1’’). See Section 4.0 for the functionality of IOUT
. DSPMASK can be read and written by the core. Upon reset, and whenever the ABORT register is written, all the bits in DSPMASK are cleared to ‘‘0’’.
The format of the NMISTAT register is shown in
Figure 2-14
.
15 3 2 1 0
Reserved ERR UND EXT
FIGURE 2-14. NMISTAT Register Format
The NMISTAT holds the status of the current pending Non­Maskable Interrupt (NMI) requests.
Whenever the core attempts to access the DSPM address space while the CLSTAT.RUN bit is ‘‘1’’ (except for access­es to the CLSTAT, EXT, DSPINT, NMISTAT, DSPMASK, and ABORT registers) NMISTAT.ERR is set to ‘‘1’’.
Whenever there is an attempt to execute a DBPT instruc­tion, or a reserved DSPM instruction (Section 3.4), the NMISTAT.UND bit is set to ‘‘1’’.
When a high to low transition is detected on the NMI
input
pin, NMISTAT.EXT bit is set to ‘‘1’’.
When one of the bits in NMISTAT is set to ‘‘1’’, an NMI request to the core is issued.
The NMISTAT register is cleared to 0 upon reset, and each time its contents are read.
When one of the bits in NMISTAT is set to 1, an NMI occurs. The NMI handler can read the NMISTAT register to deter­mine the source of the interrupt. Note that since NMIs may be nested, it is possible that a second NMI handler (invoked while the previous handler has not yet exited) will read and handle more than one set bit in NMISTAT. Since the read operation clears the register, the interrupted handler may find that no bits are set.
2.2 MEMORY ORGANIZATION
The main memory of the NS32FX164 is a uniform linear address space. Memory locations are numbered sequential­ly starting at zero and ending at 2
24
b
1. The number speci­fying a memory location is called an address. The contents of each memory location is a byte consisting of eight bits. Unless otherwise noted, diagrams in this document show data stored in memory with the lowest address on the right and the highest address on the left. Also, when data is shown vertically, the lowest address is at the top of a dia­gram and the highest address at the bottom of the diagram. When bits are numbered in a diagram, the least significant bit is given the number zero, and is shown at the right of the diagram. Bits are numbered in increasing significance and toward the left.
70
A
Byte at Address A
15 8 7 0
Aa1A
MSB LSB
Word at Address A
Two contiguous words are called a double-word. Except where noted, the least significant word of a double-word is stored at the lowest address and the most significant word of the double-word is stored at the address two higher. In memory, the address of a double-word is the address of its least significant byte, and a double-word may start at any address.
31 24 23 16 15 8 7 0
Aa3A
a
2A
a
1A
MSB LSB
Double Word at Address A
11
2.0 Architectural Description (Continued)
2.2.1 Address Mapping
The NS32FX164 supports the use of memory-mapped pe­ripheral devices and coprocessors. Such memory-mapped devices can be located at arbitrary locations within the 16-Mbyte address range available externally.
Addresses marked as Reserved in
Figure 2-15
are not avail­able in the present implementation of the NS32FX164, and should not be used. The top 8-Mbyte block is reserved by National Semiconductor Corporation, and only a few loca­tions within this block are presently used to access the on­chip RAM array and DSP Module registers.
Figure 2-15
shows the NS32FX164 address mapping.
Start Address
(HEX)
00000000 Memory and I/O
00FFFE00 Interrupt Control
01000000 Reserved
FFFE0000 DSPM Internal RAM
FFFE1000 Reserved
FFFF8000 DSPM Dedicated Registers
FFFF8028 Reserved
FFFF9000 DSPM Control/Status Registers
FFFF9014 Reserved
FIGURE 2-15. NS32FX164 Address Mapping
2.3 MODULAR SOFTWARE SUPPORT
The NS32FX164 provides special support for software mod­ules and modular programs.
Each module in a NS32FX164 software environment con­sists of three components:
1. Program Code Segment.
This segment contains the module’s code and constant data.
2. Static Data Segment.
Used to store variables and data that may be accessed by all procedures within the module.
3. Link Table.
This component contains two types of entries: Absolute Addresses and Procedure Descriptors.
An Absolute Address is used in the external addressing mode, in conjunction with a displacement and the current MOD Register contents to compute the effective address of an external variable belonging to another module.
The Procedure Descriptor is used in the call external pro­cedure (CXP) instruction to compute the address of an external procedure.
Normally, the linker program specifies the locations of the three components. The Static Data and Link Table typically reside in RAM; the code component can be either in RAM or in ROM. The three components can be mapped into non­contiguous locations in memory, and each can be indepen­dently relocated. Since the Link Table contains the absolute addresses of external variables, the linker need not assign absolute memory addresses for these in the module itself; they may be assigned at load time.
To handle the transfer of control from one module to anoth­er, the NS32FX164 uses a module table in memory and two registers in the CPU.
The Module Table is located within the first 64 kbytes of memory. This table contains a Module Descriptor (also called a Module Table Entry) for each module in the ad­dress space of the program. A Module Descriptor has four 32-bit entries corresponding to each component of a mod­ule:
#
The Static Base entry contains the address of the begin­ning of the module’s static data segment.
#
The Link Table Base points to the beginning of the mod­ule’s Link Table.
#
The Program Base is the address of the beginning of the code and constant data for the module.
#
A fourth entry is currently unused but reserved.
The MOD Register in the CPU contains the address of the Module Descriptor for the currently executing module.
The Static Base Register (SB) contains a copy of the Static Base entry in the Module Descriptor of the currently execut­ing module, i.e., it points to the beginning of the current module’s static data area.
In an NS32FX164 software environment modules need not be linked together prior to loading. As modules are loaded, a linking loader simply updates the Module Table and fills the Link Table entries with the appropriate values. No modi­fication of a module’s code is required. Thus, modules may be stored in read-only memory and may be added to a sys­tem independently of each other, without regard to their in­dividual addressing.
Figure 2-16
shows a typical
NS32FX164 run-time environment.
2.4 INSTRUCTION SET
2.4.1 General Instruction Format
Figure 2-17
shows the general format of a Series 32000 instruction. The Basic Instruction is one to three bytes long and contains the Opcode and up to two 5-bit General Ad­dressing Mode (‘‘Gen’’) fields. Following the Basic Instruc­tion field is a set of optional extensions, which may appear depending on the instruction and the addressing modes se­lected.
Index Bytes appear when either or both Gen fields specify Scaled Index. In this case, the Gen field specifies only the Scale Factor (1, 2, 4 or 8), and the Index Byte specifies which General Purpose Register to use as the index, and which addressing mode calculation to perform before index­ing.
12
2.0 Architectural Description (Continued)
Following Index Bytes come any displacements (addressing constants) or immediate values associated with the select­ed addressing modes. Each Disp/lmm field may contain one of two displacements, or one immediate value. The size of a Displacement field is encoded within the top bits of that field, as shown in
Figure 2-19
, with the remaining bits inter­preted as a signed (two’s complement) value. The size of an immediate value is determined from the Opcode field. Both Displacement and Immediate fields are stored most-signifi­cant byte first. Note that this is different from the memory representation of data (Section 2.2).
Some instructions require additional ‘‘implied’’ immediates and/or displacements, apart from those associated with ad­dressing modes. Any such extensions appear at the end of the instruction, in the order that they appear within the list of operands in the instruction definition (Section 2.4.3).
TL/EE/11267– 3
FIGURE 2-18. Index Byte Format
TL/EE/11267– 2
Note: Dashed lines indicate information copied to register during transfer of control between modules.
FIGURE 2-16. NS32FX164 Run-Time Environment
TL/EE/11267– 4
FIGURE 2-17. General Instruction Format
13
2.0 Architectural Description (Continued)
2.4.2 Addressing Modes
The NS32FX164 CPU generally accesses an operand by calculating its Effective Address based on information avail­able when the operand is to be accessed. The method to be used in performing this calculation is specified by the pro­grammer as an ‘‘addressing mode’’.
Addressing modes in the NS32FX164 are designed to opti­mally support high-level language accesses to variables. In nearly all cases, a variable access requires only one ad­dressing mode, within the instruction that acts upon that variable. Extraneous data movement is therefore minimized.
NS32FX164 Addressing Modes fall into nine basic types:
Register: The operand is available in one of the eight Gen­eral Purpose Registers. In certain Slave Processor instruc­tions, an auxiliary set of eight registers may be referenced instead.
Register Relative: A General Purpose Register contains an address to which is added a displacement value from the instruction, yielding the Effective Address of the operand in memory.
Memory Space: Identical to Register Relative above, ex­cept that the register used is one of the dedicated registers PC, SP, SB or FP. These registers point to data areas gen­erally needed by high-level languages.
Memory Relative: A pointer variable is found within the memory space pointed to by the SP, SB or FP register. A displacement is added to that pointer to generate the Effec­tive Address of the operand.
Immediate: The operand is encoded within the instruction. This addressing mode is not allowed if the operand is to be written.
Absolute: The address of the operand is specified by a displacement field in the instruction.
External: A pointer value is read from a specified entry of the current Link Table. To this pointer value is added a dis­placement, yielding the Effective Address of the operand.
Top of Stack: The currently-selected Stack Pointer (SP0 or SP1) specifies the location of the operand. The operand is pushed or popped, depending on whether it is written or read.
Scaled Index: Although encoded as an addressing mode, Scaled Indexing is an option on any addressing mode ex­cept Immediate or another Scaled Index. It has the effect of calculating an Effective Address, then multiplying any Gen­eral Purpose Register by 1, 2, 4 or 8 and adding into the total, yielding the final Effective Address of the operand.
Table 2-1 is a brief summary of the addressing modes. For a complete description of their actions, see the Series 32000 Instruction Set Reference Manual.
In addition to the general modes, Register-Indirect with auto-increment/decrement and warps or pitch are available on several of the graphics instructions.
Byte Displacement: Range
b
64 toa63
Word Displacement: Range
b
8192 toa8191
Double Word Displacement:
Range (Entire Addressing Space)
TL/EE/11267– 5
FIGURE 2-19. Displacement Encodings
14
2.0 Architectural Description (Continued)
TABLE 2-1. NS32FX164 Addressing Modes
ENCODING MODE ASSEMBLER SYNTAX EFFECTIVE ADDRESS Register
00000 Register 0 R0 or F0 None: Operand is in the specified 00001 Register 1 R1 or F1 register. 00010 Register 2 R2 or F2 00011 Register 3 R3 or F3 00100 Register 4 R4 or F4 00101 Register 5 R5 or F5 00110 Register 6 R6 or F6 00111 Register 7 R6 or F7
Register Relative
01000 Register 0 relative disp(R0) Disp
a
Register. 01001 Register 1 relative disp(R1) 01010 Register 2 relative disp(R2) 01011 Register 3 relative disp(R3) 01100 Register 4 relative disp(R4) 01101 Register 5 relative disp(R5) 01110 Register 6 relative disp(R6) 01111 Register 7 relative disp(R7)
Memory Relative
10000 Frame memory relative disp2(disp1 (FP)) Disp2
a
Pointer; Pointer found at
10001 Stack memory relative disp2(disp1 (SP)) address Disp 1
a
Register. ‘‘SP’’
10010 Static memory relative disp2(disp1 (SB)) is either SP0 or SP1, as selected
in PSR.
Reserved
10011 (Reserved for Future Use)
Immediate
10100 Immediate value None: Operand is input from
instruction queue.
Absolute
10101 Absolute
@
disp Disp.
External
10110 External EXT (disp1)
a
disp2 Disp2aPointer; Pointer is found
at Link Table Entry number Disp1.
Top Of Stack
10111 Top of stack TOS Top of current stack, using either
User or Interrupt Stack Pointer, as selected in PSR. Automatic Push/Pop included.
Memory Space
11000 Frame memory disp(FP) Disp
a
Register; ‘‘SP’’ is either 11001 Stack memory disp(SP) SP0 or SP1, as selected in PSR. 11010 Static memory disp(SB) 11011 Program memory *
a
disp
Scaled Index
11100 Index, bytes mode[Rn:B
]
EA (mode)
a
Rn.
11101 Index, words mode[Rn:W
]
EA (mode)
a
2cRn.
11110 Index, double words mode[Rn:D
]
EA (mode)
a
4cRn.
11111 Index, quad words mode[Rn:Q
]
EA (mode)
a
8cRn. ‘‘Mode’’ and ‘‘n’’ are contained within the Index Byte. EA (mode) denotes the effective address generated using mode.
15
2.0 Architectural Description (Continued)
2.4.3 Instruction Set Summary
Table 2-2 presents a brief description of the NS32FX164 instruction set. The Format column refers to the Instruction Format tables (Appendix A). The Instruction column gives the instruction as coded in assembly language, and the De­scription column provides a short description of the function provided by that instruction. Further details of the exact op­erations performed by each instruction may be found in the Series 32000 Instruction Set Reference Manual and the NS32CG16 Printer/Display Processor Programmer’s Refer­ence.
Notations:
i
e
Integer length suffix: BeByte
W
e
Word
D
e
Double Word
f
e
Floating Point length suffix: FeStandard Floating
L
e
Long Floating
gen
e
General operand. Any addressing mode can be speci-
fied.
short
e
A 4-bit value encoded within the Basic Instruction
(see Appendix A for encodings).
imm
e
Implied immediate operand. An 8-bit value appended
after any addressing extensions.
disp
e
Displacement (addressing constant): 8, 16 or 32 bits.
All three lengths legal.
reg
e
Any General Purpose Register: R0– R7.
areg
e
Any Processor Register: SP, SB, FP, INTBASE,
MOD, PSR, US (bottom 8 PSR bits).
cond
e
Any condition code, encoded as a 4-bit field within
the Basic Instruction (see Appendix A for encodings).
TABLE 2-2. NS32FX164 Instruction Set Summary
MOVES
Format Operation Operands Description
4 MOVi gen,gen Move a value. 2 MOVQi short,gen Extend and move a signed 4-bit constant. 7 MOVMi gen,gen,disp Move multiple: disp bytes (1 to 16). 7 MOVZBW gen,gen Move with zero extension. 7 MOVZiD gen,gen Move with zero extension. 7 MOVXBW gen,gen Move with sign extension. 7 MOVXiD gen,gen Move with sign extension. 4 ADDR gen,gen Move effective address.
INTEGER ARITHMETIC
Format Operation Operands Description
4 ADDi gen,gen Add. 2 ADDQi short,gen Add signed 4-bit constant. 4 ADDCi gen,gen Add with carry. 4 SUBi gen,gen Subtract. 4 SUBCi gen,gen Subtract with carry (borrow). 6 NEGi gen,gen Negate (2’s complement). 6 ABSi gen,gen Take absolute value. 7 MULi gen,gen Multiply. 7 QUOi gen,gen Divide, rounding toward zero. 7 REMi gen,gen Remainder from QUO. 7 DIVi gen,gen Divide, rounding down. 7 MODi gen,gen Remainder from DIV (Modulus). 7 MEIi gen,gen Multiply to extended integer. 7 DEIi gen,gen Divide extended integer.
PACKED DECIMAL (BCD) ARITHMETIC
Format Operation Operands Description
6 ADDPi gen,gen Add packed. 6 SUBPi gen,gen Subtract packed.
16
2.0 Architectural Description (Continued)
TABLE 2-2. NS32FX164 Instruction Set Summary (Continued)
INTEGER COMPARISON
Format Operation Operands Description
4 CMPi gen,gen Compare. 2 CMPQi short,gen Compare to signed 4-bit constant. 7 CMPMi gen,gen,disp Compare multiple: disp bytes (1 to 16).
LOGICAL AND BOOLEAN
Format Operation Operands Description
4 ANDi gen,gen Logical AND. 4 ORi gen,gen Logical OR. 4 BICi gen,gen Clear selected bits. 4 XORi gen,gen Logical exclusive OR. 6 COMi gen,gen Complement all bits. 6 NOTi gen,gen Boolean complement: LSB only. 2 Scondi gen Save condition code (cond) as a Boolean variable of size i.
SHIFTS
Format Operation Operands Description
6 LSHi gen,gen Logical shift, left or right. 6 ASHi gen,gen Arithmetic shift, left or right. 6 ROTi gen,gen Rotate, left or right.
BIT FIELDS
Bit fields are values in memory that are not aligned to byte boundaries. Examples are PACKED arrays and records used in Pascal. ‘‘Extract’’ instructions read and align a bit field. ‘‘Insert’’ instructions write a bit field from an aligned source.
Format Operation Operands Description
8 EXTi reg,gen,gen,disp Extract bit field (array oriented). 8 INSi reg,gen,gen,disp Insert bit field (array oriented). 7 EXTSi gen,gen,imm,imm Extract bit field (short form). 7 INSSi gen,gen,imm,imm Insert bit field (short form). 8 CVTP reg,gen,gen Convert to bit field pointer.
ARRAYS
Format Operation Operands Description
8 CHECKi reg,gen,gen Index bounds check. 8 INDEXi reg,gen,gen Recursive indexing step for multiple-dimensional arrays.
STRINGS
String instructions assign specific functions to the General Purpose Registers:
R4 Ð Comparison Value
R3 Ð Translation Table Pointer
R2 Ð String 2 Pointer
R1 Ð String 1 Pointer
R0 Ð Limit Count
Options on all string instructions are:
B (Backward): Decrement string pointers after each
step rather than incrementing.
R4.
W (While match): End instruction if String 1 entry does not
match R4.
All string instructions end when R0 decrements to zero.
17
2.0 Architectural Description (Continued)
TABLE 2-2. NS32FX164 Instruction Set Summary (Continued)
Format Operation Operands Description
5 MOVSi options Move string 1 to string 2.
MOVST options Move string, translating bytes.
5 CMPSi options Compare string 1 to string 2.
CMPST options Compare, translating string 1 bytes.
5 SKPSi options Skip over string 1 entries.
SKPST options Skip, translating bytes for until/while.
JUMPS AND LINKAGE
Format Operation Operands Description
3 JUMP gen Jump. 0 BR disp Branch (PC Relative). 0 Bcond disp Conditional branch. 3 CASEi gen Multiway branch. 2 ACBi short,gen,disp Add 4-bit constant and branch if non-zero. 3 JSR gen Jump to subroutine. 1 BSR disp Branch to subroutine. 1 CXP disp Call external procedure 3 CXPD gen Call external procedure using descriptor. 1 SVC Supervisor call. 1 FLAG Flag trap. 1 BPT Breakpoint trap. 1 ENTER
[
reg list], disp Save registers and allocate stack frame (Enter Procedure).
1 EXIT
[
reg list
]
Restore registers and reclaim stack frame (Exit Procedure). 1 RET disp Return from subroutine. 1 RXP disp Return from external procedure call. 1 RETT disp Return from trap. (Privileged) 1 RETI Return from interrupt. (Privileged)
CPU REGISTER MANIPULATION
Format Operation Operands Description
1 SAVE
[
reg list
]
Save general purpose registers. 1 RESTORE
[
reg list
]
Restore general purpose registers. 2 LPRi areg,gen Load dedicated register. (Privileged if PSR or INTBASE) 2 SPRi areg,gen Store dedicated register. (Privileged if PSR or INTBASE) 3 ADJSPi gen Adjust stack pointer. 3 BISPSRi gen Set selected bits in PSR. (Privileged if not Byte length) 3 BICPSRi gen Clear selected bits in PSR. (Privileged if not Byte length) 5 SETCFG
[
option list
]
Set configuration register. (Privileged)
18
2.0 Architectural Description (Continued)
TABLE 2-2. NS32FX164 Instruction Set Summary (Continued)
FLOATING POINT
Format Operation Operands Description
11 MOVf gen,gen Move a floating point value.
9 MOVLF gen,gen Move and shorten a long value to standard. 9 MOVFL gen,gen Move and lengthen a standard value to long. 9 MOVif gen,gen Convert any integer to standard or long floating. 9 ROUNDfi gen,gen Convert to integer by rounding. 9 TRUNCfi gen,gen Convert to integer by truncating, toward zero.
9 FLOORfi gen,gen Convert to largest integer less than or equal to value. 11 ADDf gen,gen Add. 11 SUBf gen,gen Subtract. 11 MULf gen,gen Multiply. 11 DIVf gen,gen Divide. 11 CMPf gen,gen Compare. 11 NEGf gen,gen Negate. 11 ABSf gen,gen Take absolute value.
9 LFSR gen Load FSR.
9 SFSR gen Store FSR.
12 POLYf gen,gen Polynomial Step. 12 DOTf gen,gen Dot Product. 12 SCALBf gen,gen Binary Scale. 12 LOGBf gen,gen Binary Log.
MISCELLANEOUS
Format Operation Operands Description
1 NOP No operation. 1 WAIT Wait for interrupt. 1 DIA Diagnose. Single-byte ‘‘Branch to Self’’ for hardware
breakpointing. Not for use in programming.
GRAPHICS
Format Operation Operands Description
5 BBOR options* Bit-aligned block transfer ‘OR’. 5 BBAND options Bit-aligned block transfer ‘AND’. 5 BBFOR Bit-aligned block transfer fast ‘OR’. 5 BBXOR options Bit-aligned block transfer ‘XOR’. 5 BBSTOD options Bit-aligned block source to destination. 5 BITWT Bit-aligned word transfer. 5 EXTBLT options External bit-aligned block transfer. 5 MOVMPi Move multiple pattern. 5 TBITS options Test bit string. 5 SBITS Set bit string. 5 SBITPS Set bit perpendicular string.
BITS
Format Operation Operands Description
4 TBITi gen,gen Test bit. 6 SBITi gen,gen Test and set bit. 6 SBITIi gen,gen Test and set bit, interlocked. 6 CBITi gen,gen Test and clear bit. 6 CBITIi gen,gen Test and clear bit, interlocked. 6 IBITi gen,gen Test and invert bit. 8 FFSi gen,gen Find first set bit.
*Note: Options are controlled by fields of the instruction, PSR status bits, or dedicated register values.
19
2.0 Architectural Description (Continued)
2.5 GRAPHICS SUPPORT
The following sections provide a brief description of the NS32FX164 graphics support capabilities. Basic discus­sions on frame buffer addressing and BITBLT operations are also provided. More detailed information on the NS32FX164 graphics support instructions can be found in the NS32CG16 Printer/Display Processor Programmer’s Reference.
2.5.1 Frame Buffer Addressing
There are two basic addressing schemes for referencing pixels within the frame buffer: Linear and Cartesian (or x-y). Linear addressing associates a single number to each pixel representing the physical address of the corresponding bit in memory. Cartesian addressing associates two numbers to each pixel representing the x and y coordinates of the pixel relative to a point in the Cartesian space taken as the origin. The Cartesian space is generally defined as having the origin in the upper left. A movement to the right increas­es the x coordinate; a movement downward increases the y coordinate.
The correspondence between the location of a pixel in the Cartesian space and the physical (BIT) address in memory is shown in
Figure 2-20
. The origin of the Cartesian space
(x
e
0, ye0) corresponds to the bit address ‘ORG’. Incre­menting the x coordinate increments the bit address by one. Incrementing the y coordinate increments the bit address by an amount representing the warp (or pitch) of the Cartesian space. Thus, the linear address of a pixel at location (x, y) in the Cartesian space can be found by the following expres­sion.
ADDR
e
ORGay * WARPax
Warp is the distance (in bits) in the physical memory space between two vertically adjacent bits in the Cartesian space.
Example 1 below shows two NS32FX164 instruction se­quences to set a single pixel given the x and y coordinates. Example 2 shows how to create a fat pixel by setting four adjacent bits in the Cartesian space.
Example 1: Set pixel at location (x, y)
Setup: R0 x coordinate
R1 y coordinate
Instruction Sequence 1:
MULD WARP, R1 ; Y*WARP
ADDD R0, R1 ; 0 X 4 BIT OFFSET
SBITD R1, ORG ; SET PIXEL
Instruction Sequence 2:
INDEXD R1, (WARP-1), R0 ; Y*WARP 0 X
SBITD R1, ORG ; SET PIXEL
Example 2: Create fat pixel by setting bits at locations
(x, y), (x
a
1, y), (x, ya1) and (xa1, ya1).
Setup: R0 x coordinate
R1 y coordinate
Instruction Sequence:
INDEXD R1, (WARP-1), R0 ; BIT ADDRESS
SBITD 41, ORG ; SET FIRST PIXEL
ADDQD 1, R1 ; (X01, Y)
SBITD R1, ORG ; SECOND PIXEL
ADDD (WARP-1), R1 ; (X, Y01)
SBITD R1, ORG ; THIRD PIXEL
ADDQD 1, R1 ; (X01, Y01)
SBITD R1, ORG ; LAST PIXEL
TL/EE/11267– 6
FIGURE 2-20. Correspondence between
Linear and Cartesian Addressing
2.5.2 BITBLT Fundamentals
BITBLT, BIT-aligned BLock Transfer, is a general operator that provides a mechanism to move an arbitrary size rectan­gle of an image from one part of the frame buffer to another. During the data transfer process a bitwise logical operation can be performed between the source and the destination data. BITBLT is also called RasterOp: operations on rasters. It defines two rectangular areas, source and destination, and performs a logical operation (e.g., AND, OR, XOR) be­tween these two areas and stores the result back to the destination. It can be expressed in simple notation as:
Source op Destination
x
Destination
op: AND, OR, XOR, etc.
20
2.0 Architectural Description (Continued)
2.5.2.1 Frame Buffer Architecture
There are two basic types of frame buffer architectures: plane-oriented or pixel-oriented. BITBLT takes advantage of the plane-oriented frame buffer architecture’s attribute of multiple, adjacent pixels-per-word, facilitating the movement of large blocks of data. The source and destination starting addresses are expressed as pixel addresses. The width and height of the block to be moved are expressed in terms of pixels and scan lines. The source block may start and end at any bit position of any word, and the same applies for the destination block.
2.5.2.2 Bit Alignment
Before a logical operation can be performed between the source and the destination data, the source data must first be bit aligned to the destination data. In
Figure 2-21
, the source data needs to be shifted three bits to the right in order to align the first pixel (i.e., the pixel at the top left corner) in the source data block to the first pixel in the desti­nation data block.
2.5.2.3 Block Boundaries and Destination Masks
Each BITBLT destination scan line may start and end at any bit position in any data word. The neighboring bits (bits shar­ing the same word address with any words in the destination data block, but not a part of the BITBLT rectangle) of the BITBLT destination scan line must remain unchanged after the BITBLT operation.
Due to the plane-oriented frame buffer architecture, all memory operations must be word-aligned. In order to pre­serve the neighboring bits surrounding the BITBLT destina­tion block, both a left mask and a right mask are needed for all the leftmost and all the rightmost data words of the desti­nation block. The left mask and the right mask both remain the same during a BITBLT operation.
The following example illustrates the bit alignment require­ments. In this example, the memory data path is 16 bits wide.
Figure 2-21
shows a 32 pixel by 32 scan line frame buffer which is organized as a long bit stream which wraps around every two words (32 bits). The origin (top left corner) of the frame buffer starts from the lowest word in memory (word address 00 (hex)).
Each word in the memory contains 16 bits, D0 – D15. The least significant bit of a memory word, D0, is defined as the first displayed pixel in a word. In this example, BITBLT ad­dresses are expressed as pixel addresses relative to the origin of the frame buffer. The source block starting address is 021 (hex) (the second pixel in the third word). The desti­nation block starting address is 204 (hex) (the fifth pixel in the 33rd word). The block width is 13 (hex), and the height is 06 (hex) (corresponding to 6 scan lines). The shift value is 3.
TL/EE/11267– 7
FIGURE 2-21. 32-Pixel by 32-Scan Line Frame Buffer
21
2.0 Architectural Description (Continued)
2.5.2.4 BITBLT Directions
A BITBLT operation moves a rectangular block of data in a frame buffer. The operation itself can be considered as a subroutine with two nested loops. The loops are preceded by setup operations. In the outer loop the source and desti­nation starting addresses are calculated, and the test for completion is performed. In the inner loop the actual data movement for a single scan line takes place. The length of the inner loop is the number of (aligned) words spanned by each scan line. The length of the outer loop is equal to the height (number of scan lines) of the block to be moved. A skeleton of the subroutine representing the BITBLT opera­tion follows.
BITBLT: calculate BITBLT setup parameters;
(once per BITBLT operation).
such as
width, height
bit misalignment (shift number)
left, right masks
horizontal, vertical directions
etc
# #
OUTERLOOP: calculate source, dest addresses;
(once per scanline).
INNERLOOP: move data, (logical operation) and incre-
ment addresses; (once per word).
UNTIL done horizontally
UNTIL done vertically
RETURN (from BITBLT).
Note: In the NS32FX164 only the setup operations must be done by the
programmer. The inner and outer loops are automatically executed by the BITBLT instructions.
Each loop can be executed in one of two directions: the inner loop from left to right or right to left, the outer loop from top to bottom (down) or bottom to top (up).
The ability to move data starting from any corner of the BITBLT rectangle is necessary to avoid destroying the BITBLT source data as a result of destination writes when the source and destination are overlapped (i.e., when they share pixels). This situation is routinely encountered while panning or scrolling.
A determination of the correct execution directions of the BITBLT must be performed whenever the source and desti­nation rectangles overlap. Any overlap will result in the de­struction of source data (from a destination write) if the cor­rect vertical direction is not used. Horizontal BITBLT direc­tion is of concern only in certain cases of overlap, as will be explained below.
Figures 2-22(a)
and
(b)
illustrate two cases of overlap. Here, the BITBLT rectangles are three pixels wide by five scan lines high; they overlap by a single pixel in
(a)
and a single
column of pixels in
(b)
. For purposes of illustration, the BITBLT is assumed to be carried out pixel-by-pixel. This convention does not affect the conclusions.
In
Figure 2-22(a)
, if the BITBLT is performed in the UP direc­tion (bottom-to-top) one of the transfers of the bottom scan line of the source will write to the circled pixel of the destina­tion. Due to the overlap, this pixel is also part of the upper­most scan line of the source rectangle. Thus, data needed later is destroyed. Therefore, this BITBLT must be per­formed in the DOWN direction. Another example of this oc-
TL/EE/11267– 8
(a)
TL/EE/11267– 9
(b)
FIGURE 2-22. Overlapping BITBLT Blocks
The left mask and the right mask are 0000,1111,1111,1111 and 1111,1111,0000,0000 respectively.
Note 1: Zeros in either the left mask or the right mask indicate the destination bits which will not be modified.
Note 2: The BB(function) and EXTBLT instructions use different set up parameters, and techniques.
22
2.0 Architectural Description (Continued)
curs any time the screen is moved in a purely vertical direc­tion, as in scrolling text. It should be noted that, in both of these cases, the choice of horizontal BITBLT direction may be made arbitrarily.
Figure 2-22(b)
demonstrates a case in which the horizontal BITBLT direction may not be chosen arbitrarily. This is an instance of purely horizontal movement of data (panning). Because the movement from source to destination involves data within the same scan line, the incorrect direction of movement will overwrite data which will be needed later. In this example, the correct direction is from right to left.
2.5.2.5 BITBLT Variations
The ‘‘classical’’ definition of BITBLT, as described in ‘‘Smalltalk-80 The Language and its Implementation’’, by Adele Goldberg and David Robson, provides for three oper­ands: source, destination and mask/texture. This third oper­and is commonly used in monochrome systems to incorpo­rate a stipple pattern into an area. These stipple patterns provide the appearance of multiple shades of gray in single­bit-per-pixel systems, in a manner similar to the ‘‘halftone’’ process used in printing.
Texture op1 Source op2 Destination
x
Destination
While the NS32FX164 and the external BPU (if used) are essentially two-operand devices, three-operand BITBLT op­erations can be implemented quite flexibly and efficiently by performing the two operations serially.
2.5.3 GRAPHICS SUPPORT INSTRUCTIONS
The NS32FX164 provides eleven instructions for supporting graphics oriented applications. These instructions are divid­ed into three groups according to the operations they per­form. General descriptions for each of them and the related formats are provided in the following sections.
2.5.3.1 BITBLT (BIT-aligned BLock Transfer)
This group includes seven instructions. They are used to move characters and objects into the frame buffer which will be printed or displayed. One of the instructions works in conjunction with an external BITBLT Processing Unit (BPU) to maximize performance. The other six are executed by the NS32FX164.
BIT-aligned BLock Transfer
Syntax: BB(function) Options
Setup: R0 base address, source data
R1 base address, destination data R2 shift value R3 height (in lines) R4 first mask R5 second mask R6 source warp (adjusted) R7 destination warp (adjusted) 0(SP) width (in words)
Function: AND, OR, XOR, FOR, STOD
Options: IA Increasing Address (default option).
When IA is selected, scan lines are transferred in the increasing BIT/BYTE order.
DA Decreasing Address.
S True Source (default option).
b
S Inverted Source.
These five instructions perform standard BITBLT operations between source and destination blocks. The operations available include the following:
BBAND: src AND dst
b
src AND dst
BBOR: src OR dst
b
src OR dst
BBXOR: src XOR dst
b
src XOR dst
BBFOR: src OR dst BBSTOD: src TO dst
b
src TO dst
‘src’ and ‘
b
src’ stand for ‘True Source’ and ‘Inverted
Source’ respectively; ‘dst’ stands for ‘Destination’.
Note 1: For speed reasons, the BB instructions require the masks to be
specified with respect to the source block. In
Figure 2-21
masking
was defined relative to the destination block.
Note 2: The options
b
S and DA are not available for the BBFOR instruc-
tion.
Note 3: BBFOR performs the same operation as BBOR with IA and S op-
tions.
Note 4: IA and DA are mutually exclusive and so are S and
b
S.
Note 5: The width is defined as the number of words of source data to read.
Note 6: An odd number of bytes can be specified for the source warp.
However, word alignment of source scan lines will result in faster execution.
The horizontal and vertical directions of the BITBLT opera­tions performed by the above instructions, with the excep­tion of BBFOR, are both programmable. The horizontal di­rection is controlled by the IA and DA options. The vertical direction is controlled by the sign of the source and destina­tion warps.
Figure 2-23
and Table 2-3 show the format of the BB instructions and the encodings for the ‘op’ and ‘i’ fields.
23 16 15 8 7 0
000 000 D X S 0 op i 00001110
#
D is set when the DA option is selected
#
S is set when thebS option is selected
#
X is set for BBAND, and it is clear for all other BB instructions
FIGURE 2-23. BB Instructions Format
TABLE 2-3. ‘op’ and ‘i’ Field Encodings
Instruction Options ‘op’ Field ‘i’ Field
BBAND Yes 1010 11
BBOR Yes 0110 01
BBXOR Yes 1110 01
BBFOR No 1100 01
BBSTOD Yes 0100 01
BIT-aligned Word Transfer
Syntax: BITWT
Setup: R0 Base address, source word
R1 Base address, destination double word R2 Shift value
The BITWT instruction performs a fast logical OR operation between a source word and a destination double word, stores the result into the destination double word and incre­ments registers R0 and R1 by two. Before performing the OR operation, the source word is shifted left (i.e., in the direction of increasing bit numbers) by the value in register R2.
23
2.0 Architectural Description (Continued)
This instruction can be used within the inner loop of a block OR operation. Its use assumes that the source data is ‘clean’ and does not need masking. The BITWT format is shown in
Figure 2-24
.
23 16 15 8 7 0
0000000000100001 0 000 1110
FIGURE 2-24. BITWT Instruction Format
External BITBLT
Syntax: EXTBLT
Setup: R0 base addresses, source data
R1 base address, destination data R2 width (in bytes) R3 height (in lines) R4 horizontal increment/decrement R5 temporary register (current width) R6 source warp (adjusted) R7 destination warp (adjusted)
Note 1: R0 and R1 are updated after execution to point to the last source
and destination addresses plus related warps. R2, R3 and R5 will be modified. R4, R6, and R7 are returned unchanged.
Note 2: Source and destination pointers should point to word-aligned oper-
ands to maximize speed and minimize external interface logic.
This instruction performs an entire BITBLT operation in con­junction with an external BITBLT Processing Unit (BPU). The external BPU Control Register should be loaded by the software before the instruction is executed (refer to the DP8510 or DP8511 data sheets for more information on the BPU). The NS32FX164 generates a series of source read, destination read and destination write bus cycles until the entire data block has been transferred. The BITBLT opera­tion can be performed in either horizontal direction. As con­trolled by the sign of the contents of register R4.
Depending on the relative alignment of the source and des­tination blocks, an extra source read may be required at the beginning of each scan line, to load the pipeline register in the external BPU. The L bit in the PSR register determines whether the extra source read is performed. If L is 1, no extra read is performed. The instructions CMPQB 2,1 or CMPQB 1,2 could be executed to provide the right setting for the L bit just before executing EXTBLT.
Figure 2-25
shows the EXTBLT format. The bus activity for a simple BITBLT operation is shown in
Figure 2-30.
23 15 8 7 0
000000000001011100001110
FIGURE 2-25. EXTBLT Instruction Format
2.5.3.2 Pattern Fill
Only one instruction is in this group. It is usually used for clearing RAM and drawing patterns and lines.
Move Multiple Pattern
Syntax: MOVMPi
Setup: R0 base address of the destination
R1 pointer increment (in bytes) R2 number of pattern moves R3 source pattern
Note: R1 and R3 are not modified by the instruction. R2 will always be
returned as zero. R0 is modified to reflect the last address into which a pattern was written.
This instruction stores the pattern in register R3 into the destination area whose address is in register R0. The pat­tern count is specified in register R2. After each store oper­ation the destination address is changed by the contents of register R1. This allows the pattern to be stored in rows, in columns, and in any direction, depending on the value and sign of R1. The MOVMPi instruction format is shown in
Fig-
ure 2-26
.
23 15 8 7 0
00000000000111 i 00001110
FIGURE 2-26. MOVMPi Instruction Format
2.5.3.3 Data Compression, Expansion and Magnify
The three instructions in this group can be used to com­press data and restore data from compression. A com­pressed character set may require from 30% to 50% less memory space for its storage.
The compression ratio possible can be 50:1 or higher de­pending on the data and algorithm used. TBITS can also be used to find boundaries of an object. As a character is need­ed, the data is expanded and stored in a RAM buffer. The expand instructions (SBITS, SBITPS) can also function as line drawing instructions.
Test Bit String
Syntax: TBITS option
Setup: R0 base address, source (byte address)
R1 starting source bit offset R2 destination run length limited code R3 maximum value run length limit R4 maximum source bit offset
Option: 1 count set bits until a clear bit is found
0 count clear bits until a set bit is found
Note: R0, R3 and R4 are not modified by the instruction execution. R1
reflects the new bit offset. R2 holds the result.
This instruction starts at the base address, adds a bit offset, and tests the bit for clear if ‘‘option’’
e
0 (and for set if
‘‘option’’
e
1). If clear (or set), the instruction increments to the next higher bit and tests for clear (or set). This testing for clear proceeds through memory until a set bit is found or until the maximum source bit offset or maximum run length value is reached. The total number of clear bits is stored in the destination as a run length value.
When TBITS finds a set bit and terminates, the bit offset is adjusted to reflect the current bit address. Offset is then ready for the next TBITS instruction with ‘‘option’’
e
0. After the instruction is executed, the F flag is set to the value of the bit previous to the bit currently being pointed to (i.e., the value of the bit on which the instruction completed execu­tion). In the case of a starting bit offset exceeding the maxi­mum bit offset (R1
t
R4), the F flag is set if the option was 1 and clear if the option was 0. The L flag is set when the desired bit is found, or if the run length equalled the maxi­mum run length value and the bit was not found. It is cleared otherwise.
Figure 2-27
shows the TBITS instruction format.
23 15 8 7 0
00000000S010011100001110
#
S is set for ‘TBITS 1’ and clear for ‘TBITS 0’.
FIGURE 2-27. TBITS Instruction Format
24
2.0 Architectural Description (Continued)
Set Bit String
Syntax: SBITS
Setup: R0 base address of the destination
R1 starting bit offset (signed) R2 number of bits to set (unsigned) R3 address of string look-up table
Note: When the instruction terminates, the registers are returned un-
changed.
SBITS sets a number of contiguous bits in memory to 1, and is typically used for data expansion operations. The instruc­tion draws the number of ones specified by the value in R2, starting at the bit address provided by registers R0 and R1. In order to maximize speed and allow drawing of patterned lines, an external 1k byte lookup table is used. The lookup table is specified in the NS32CG16 Printer/Display Proces­sor Programmer’s Reference Supplement.
When SBITS begins executing, it compares the value in R2 with 25. If the value in R2 is less than or equal to 25, the F flag is cleared and the appropriate number of bits are set in memory. If R2 is greater than 25, the F flag is set and no other action is performed. This allows the software to use a faster algorithm to set longer strings of bits.
Figure 2-28
shows the SBITS instruction format.
23 15 8 7 0
000000000011011100001110
FIGURE 2-28. SBITS Instruction Format
Set BIT Perpendicular String
Syntax: SBITPS
Setup: R0 base address, destination (byte address)
R1 starting bit offset R2 number of bits to set R3 destination warp (signed value, in bits)
Note: When the instruction terminates, the R0 and R3 registers are re-
turned unchanged. R1 becomes the final bit offset. R2 is zero.
The SBITPS can be used to set a string of bits in any direc­tion. This allows a font to be expanded with a 90 or 270 degree rotation, as may be required in a printer application. SBITPS sets a string of bits starting at the bit address speci­fied in registers R0 and R1. The number of bits in the string is specified in R2. After the first bit is set, the destination warp is added to the bit address and the next bit is set. The process is repeated until all the bits have been set. A nega­tive raster warp offset value leads to a 90 degree rotation. A positive raster warp value leads to a 270 degree rotation. If the R3 value is
e
(space warpa1orb1), then the result is
a 45 degree line. If the R3 value is
a
1orb1, a horizontal
line results.
SBITS and SBITPS allow expansion on any 90 degree an­gle, giving portrait, landscape and mirror images from one font.
Figure 2-29
shows the SBITPS instruction format.
23 15 8 7 0
000000000010111100001110
FIGURE 2-29. SBITPS Instruction Format
TL/EE/11267– 10
FIGURE 2-30. Bus Activity for a Simple BITBLT Operation
Note 1: This example is for a block 4 words wide and 1 line high. Note 2: The sequence is common with all logical operations of the DP8510/DP8511 BPU. Note 3: Mask values, shift values and number of bit planes do not affect the performance. Note 4: Zero wait states are assumed throughout the BITBLT operation. Note 5: The extra read is performed when the BPU pipeline register needs to be preloaded.
25
2.0 Architectural Description (Continued)
2.5.3.3.1 Magnifying Compressed Data
As the pattern of data is expanded, it can be magnified by 2x, 3x, 4x, . . . , 10x and so on. This creates several sizes of the same style of character, or changes the size of a logo. A magnify in both dimensions X and Y can be accomplished by drawing a single line, then using the MOVS (Move String) or the BB instructions to duplicate the line, maintaining an equal aspect ratio.
More information on this subject is provided in the NS32CG16 Printer/Display Processor Programmer’s Refer­ence Supplement.
3.0 Functional Description
The chapter is divided into five main sections:
Instruction Execution, Exception Processing, Debugging, DSP Module and System Interface.
3.1 INSTRUCTION EXECUTION
To execute an instruction, the NS32FX164 performs the fol­lowing operations:
#
Fetch the Instruction
#
Read Source Operands, if Any (1)
#
Calculate Results
#
Write Result Operands, if Any
#
Modify Flags, if Necessary
#
Update the Program Counter
Under most circumstances, the CPU can be conceived to execute instructions by completing the operations above in strict sequence for one instruction and then beginning the sequence of operations for the next instruction. However, due to the internal instruction pipelining, as well as the oc­currence of exceptions, the sequence of operations per­formed during the execution of an instruction may be al­tered. Furthermore, exceptions also break the sequentiality of the instructions executed by the CPU.
Note 1: In this and following sections, memory locations read by the CPU to
calculate effective addresses for Memory-Relative and External ad­dressing modes are considered like source operands, even if the effective address is being calculated for an operand with access class of write.
3.1.1 Operating States
The CPU has four operating states regarding the execution of instructions and the processing of exceptions: Reset, Ex­ecuting Instructions, Processing An Exception and Waiting­For-An-Interrupt. The various states and transitions be­tween them are shown in
Figure 3-1
.
Whenever the RSTI signal is asserted, the CPU enters the reset state. The CPU remains in the reset state until the RSTI
signal is driven inactive, at which time it enters the Executing-Instructions state. In the Reset state the contents of certain registers are initialized. Refer to Section 3.5.4 for details.
TL/EE/11267– 11
FIGURE 3-1. Operating States
In the Executing-Instructions state, the CPU executes in­structions. It will exit this state when an exception is recog­nized or a WAIT instruction is encountered. At which time it enters the Processing-An-Exception state or the Waiting­For-An-Interrupt state respectively.
While in the Processing-An-Exception state, the CPU saves the PC, PSR and MOD register contents on the stack and reads the new PC and module linkage information to begin execution of the exception service procedure.
Following the completion of all data references required to process an exception, the CPU enters the Executing-In­structions state.
In the Waiting-For-An-Interrupt state, the CPU is idle. A spe­cial status identifying this state is presented on the system interface (Section 3.5). When an interrupt is detected, the CPU enters the Processing-An-Exception State.
3.1.2 Instruction Endings
The NS32FX164 checks for exceptions at various points while executing instructions. Certain exceptions, like inter­rupts, are in most cases recognized between instructions. Other exceptions, like Divide-By-Zero Trap, are recognized during execution of an instruction. When an exception is recognized during execution of an instruction, the instruction ends in one of four possible ways: completed, suspended, terminated, or partially completed. Each type of exception causes a particular ending, as specified in Section 3.2.
26
3.0 Functional Description (Continued)
3.1.2.1 Completed Instructions
When an exception is recognized after an instruction is completed, the CPU has performed all of the operations for that instruction and for all other instructions executed since the last exception occurred. Result operands have been written, flags have been modified, and the PC saved on the Interrupt Stack contains the address of the next instruction to execute. The exception service procedure can, at its con­clusion, execute the RETT instruction (or the RETI instruc­tion for maskable interrupts), and the CPU will begin execut­ing the instruction following the completed instruction.
3.1.2.2 Suspended Instructions
An instruction is suspended when one of several trap condi­tions is detected during execution of the instruction. A sus­pended instruction has not been completed, but all other instructions executed since the last exception occurred have been completed. Result operands and flags due to be affected by the instruction may have been modified, but only modifications that allow the instruction to be executed again and completed can occur. For certain exceptions (Trap (UND) the CPU clears the P-flag in the PSR before saving the copy that is pushed on the Interrupt Stack. The PC saved on the Interrupt Stack contains the address of the suspended instruction.
To complete a suspended instruction, the exception service procedure takes either of two actions:
1. The service procedure can simulate the suspended in­struction’s execution. After calculating and writing the in­struction’s results, the flags in the PSR copy saved on the Interrupt Stack should be modified, and the PC saved on the Interrupt Stack should be updated to point to the next instruction to execute. The service procedure can then execute the RETT instruction, and the CPU begins exe­cuting the instruction following the suspended instruction. This is the action taken when floating-point instructions are simulated by software in systems without a hardware floating-point unit.
2. The suspended instruction can be executed again after the service procedure has eliminated the trap condition that caused the instruction to be suspended. The service procedure should execute the RETT instruction at its con­clusion; then the CPU begins executing the suspended instruction again. This is the action taken by a debugger when it encounters a BPT instruction that was temporarily placed in another instruction’s location in order to set a breakpoint.
Note 1: It may be necessary for the exception service procedure to alter the
P-flag in the PSR copy saved on the Interrupt Stack: If the exception service procedure simulates the suspended instruction and the P­flag was cleared by the CPU before saving the PSR copy, then the saved T-flag must be copied to the saved P-flag (like the floating­point instruction simulation described above). Or if the exception service procedure executes the suspended instruction again and the P-flag was not cleared by the CPU before saving the PSR copy, then the saved P-flag must be cleared (like the breakpoint trap de­scribed above). Otherwise, no alteration to the saved P-flag is nec­essary.
3.1.2.3 Terminated Instructions
An instruction being executed is terminated when reset oc­curs. Any result operands and flags due to be affected by the instruction are undefined, as is the contents of the PC.
3.1.2.4 Partially Completed Instructions
When an interrupt condition is recognized during execution of a string instruction, the instruction is said to be partially completed. A partially completed instruction has not com­pleted, but all other instructions executed since the last ex­ception occurred have been completed. Result operands and flags due to be affected by the instruction may have been modified, but the values stored in the string pointers and other general-purpose registers used during the instruc­tion’s execution allow the instruction to be executed again and completed.
The CPU clears the P-flag in the PSR before saving the copy that is pushed on the Interrupt Stack. The PC saved on the Interrupt Stack contains the address of the partially completed instruction. The exception service procedure can, at its conclusion, simply execute the RETT instruction (or the RETI instruction for maskable interrupts), and the CPU will resume executing the partially completed instruc­tion.
3.1.3 Slave Processor Instructions
The NS32FX164 supports only one group of instructions, the floating-point instruction set, as being executable by a slave processor. The floating-point instruction set is validat­ed by the F-bit in the CFG register.
If a floating-point instruction is encountered and the F-bit in the CFG register is not set, a Trap (UND) will result, without any slave processor communication attempted by the CPU. This allows software emulation in case an external floating­point unit (FPU) is not used.
3.1.3.1 Slave Processor Protocol
Slave Processor instructions have a three-byte Basic In­struction field, consisting of an ID Byte followed by an Oper­ation Word. The ID Byte has three functions:
1. It identifies the instruction as being a Slave Processor instruction.
2. It specifies which Slave Processor will execute it.
3. It determines the format of the following Operation Word of the instruction.
Upon receiving a Slave Processor instruction, the CPU initi­ates the sequence outlined in
Figure 3-2
. While applying Status Code 1111 (Broadcast ID, Section 3.5.5.1), the CPU transfers the ID Byte on the least-significant half of the Data Bus (AD0 – AD7). All Slave Processors input this byte and decode it. The Slave Processor selected by the ID Byte is activated, and from this point the CPU is communicating only with it. If any other slave protocol was in progress (e.g., an aborted Slave instruction), this transfer cancels it.
27
3.0 Functional Description (Continued)
The CPU next sends the Operation Word while applying Status Code 1101 (Transfer Slave Operand, Section
3.5.5.1). Upon receiving it, the Slave Processor decodes it, and at this point both the CPU and the Slave Processor are aware of the number of operands to be transferred and their sizes. The Operation Word is swapped on the Data Bus; that is, bits 0–7 appear on pins AD8 – AD15 and bits 8 – 15 appear on pins AD0 –AD7.
Using the Address Mode fields within the Operation Word, the CPU starts fetching operands and issuing them to the Slave Processor. To do so, it references any Addressing Mode extensions which may be appended to the Slave Processor instruction. Since the CPU is solely responsible for memory accesses, these extensions are not sent to the Slave Processor. The Status Code applied is 1101 (Transfer Slave Processor Operand, Section 3.5.5.1).
After the CPU has issued the last operand, the Slave Proc­essor starts the actual execution of the instruction. Upon completion, it will signal the CPU by pulsing SPC
low.
While the Slave Processor is executing the instruction, the CPU is free to prefetch instructions into its queue. If it fills the queue before the Slave Processor finishes, the CPU will wait, applying Status Code 0011 (Waiting for Slave).
Upon receiving the pulse on SPC
, the CPU uses SPC to read a Status Word from the Slave Processor, applying Status Code 1110 (Read Slave Status). This word has the format shown in
Figure 3-3
. If the Q-bit (‘‘Quit’’, Bit 0) is set, this indicates that an error was detected by the Slave Proc­essor. The CPU will not continue the protocol, but will imme-
Status Combinations: Send ID (ID): Code 1111 Xfer Operand (OP): Code 1101 Read Status (ST): Code 1110
Step Status Action
1 ID CPU Sends ID Byte 2 OP CPU Sends Operation Word 3 OP CPU Sends Required Operands 4 Ð Slave Starts Execution.
CPU Pre-Fetches.
5 Ð Slave Pulses SPC
Low
6 ST CPU Reads Status Word.
(Trap? Alter Flags?)
7 OP CPU Reads Results (If Any).
FIGURE 3-2. Slave Processor Protocol
The last step in the protocol is for the CPU to read a result, if any, and transfer it to the destination. The Read cycles from the Slave Processor are performed by the CPU while applying Status Code 1101 (Transfer Slave Operand).
3.1.3.2 Floating-Point Instructions
Table 3-1 gives the protocols followed for each Floating­Point instruction. The instructions are referenced by their mnemonics. For the bit encodings of each instruction, see Appendix A.
TABLE 3-1. Floating-Point Instruction Protocols
Mnemonic
Operand 1 Operand 2 Operand 1 Operand 2 Returned Value PSR Bits
Class Class Issued Issued Type and Dest. Affected
ADDf read.f rmw.f f f f to Op.2 none SUBf read.f rmw.f f f f to Op.2 none MULf read.f rmw.f f f f to Op.2 none DIVf read.f rmw.f f f f to Op.2 none
MOVf read.f write.f f N/A f to Op.2 none ABSf read.f write.f f N/A f to Op.2 none NEGf read.f write.f f N/A f to Op.2 none
CMPf read.f read.f f f N/A N,Z,L
FLOORfi read.f write.i f N/A i to Op.2 none TRUNCfi read.f write.i f N/A i to Op.2 none ROUNDfi read.f write.i f N/A i to Op.2 none
MOVFL read.F write.L F N/A L to Op.2 none MOVLF read.L write.F L N/A F to Op.2 none
MOVif read.i write.f i N/A f to Op.2 none
LFSR read.D N/A D N/A N/A none SFSR N/A write.D N/A N/A D to Op. 2 none
POLYf read.f read.f f f f to F0 none DOTf read.f read.f f f f to F0 none SCALBf read.f rmw.f f f f to Op. 2 none LOGBf read.f write.f f N/A f to Op. 2 none
Notes:
D
e
Double Word
i
e
Integer size (B, W, D) specified in mnemonic.
f
e
Floating-Point type (F, L) specified in mnemonic.
N/A
e
Not Applicable to this instruction.
28
3.0 Functional Description (Continued)
The Operand class columns give the Access Class for each general operand, defining how the addressing modes are interpreted (see Series 32000 Instruction Set Reference Manual).
The Operand Issued columns show the sizes of the oper­ands issued to the Floating-Point Unit by the CPU. ‘‘D’’ indi­cates a 32-bit Double Word. ‘‘i’’ indicates that the instruction specifies an integer size for the operand (B
e
Byte,
W
e
Word, DeDouble Word). ‘‘f’’ indicates that the in­struction specifies a Floating-Point size for the operand (F
e
32-bit Standard Floating, Le64-bit Long Floating).
The Returned Value Type and Destination column gives the size of any returned value and where the CPU places it. The PSR Bits Affected column indicates which PSR bits, if any, are updated from the Slave Processor Status Word
(Figure
3-3)
.
TL/EE/11267– 12
FIGURE 3-3. Slave Processor Status Word
3.2 EXCEPTION PROCESSING
Exceptions are special events that alter the sequence of instruction execution. The CPU recognizes two basic types of exceptions: interrupts and traps.
An interrupt occurs in response to an event generated either internally, by the on-chip DSP Module, or externally, by acti­vating NMI
or INT. External interrupts are typically request-
ed by peripheral devices that require the CPU’s attention.
Traps occur as a result either of exceptional conditions (e.g., attempted division by zero) or of specific instructions whose purpose is to cause a trap to occur (e.g., supervisor call instruction).
When an exception is recognized, the CPU saves the PC, PSR and optionally the MOD register contents on the inter­rupt stack and then it transfers control to an exception serv­ice procedure.
It is to be noted that the reset operation is not treated here as an exception. Even though, like any exception, it alters the instruction execution sequence.
The reason being that the CPU handles reset in a signifi­cantly different way than it does for exceptions.
Refer to Section 3.5.4 for details on the reset operation.
3.2.1 Exception Acknowledge Sequence
When an exception is recognized, the CPU goes through three major steps:
1. Adjustment of Registers. Depending on the source of the exception, the CPU may restore and/or adjust the con­tents of the Program Counter (PC), the Processor Status Register (PSR) and the currently-selected Stack Pointer (SP). A copy of the PSR is made, and the PSR is then set to reflect Supervisor Mode and selection of the Interrupt Stack. Trap (TRC) always disabled. Maskable interrupts are also disabled if the exception is caused by an inter­rupt.
2. Vector Acquisition. A vector is either obtained from an external interrupt control unit or is supplied internally by default.
3. Service Call. The CPU performs one of two sequences common to all exceptions to complete the acknowledge process and enter the appropriate service procedure. The selection between the two sequences depends on whether the Direct-Exception mode is disabled or en­abled.
Direct-Exception Mode Disabled
The Direct-Exception mode is disabled while the DE bit in the CFG register is 0 (Section 2.1.4). In this case the CPU first pushes the saved PSR copy along with the contents of the MOD and PC registers on the interrupt stack. Then it reads the double-word entry from the Interrupt Dispatch ta­ble at address ‘‘INTBASE’’
a
vectorc4’’. See
Figures 3-4
and
3-5
. The CPU uses this entry to call the exception serv­ice procedure, interpreting the entry as an external proce­dure descriptor.
A new module number is loaded into the MOD register from the least-significant word of the descriptor, and the static­base pointer for the new module is read from memory and loaded into the SB register. Then the program-base pointer for the new module is read from memory and added to the most-significant word of the module descriptor, which is in­terpreted as an unsigned value. Finally, the result is loaded into the PC register.
Direct-Exception Mode Enabled
a
vectorc4’’. The CPU uses this entry to call the exception service procedure, interpreting the entry as an absolute address that is simply loaded into the PC register.
Figure 3-6
provides a pictorial of the acknowledge sequence. It is to be noted that while the direct-exception mode is enabled, the CPU can respond more quickly to interrupts and other exceptions because fewer memory references are required to process an excep­tion. The MOD and SB registers, however, are not initialized before the CPU transfers control to the service procedure. Consequently, the service procedure is restricted from exe­cuting any instructions, such as CXP, that use the contents of the MOD or SB registers in effective address calcula­tions.
29
3.0 Functional Description (Continued)
TL/EE/11267– 13
FIGURE 3-4. Interrupt Dispatch and Cascade Tables
3.2.2 Returning from an Exception Service Procedure
To return control to an interrupted program, one of two in­structions can be used: RETT (Return from Trap) and RETI (Return from Interrupt).
RETT is used to return from any trap or non-maskable inter­rupt service procedure. Since some traps are often used deliberately as a call mechanism for supervisor mode proce­dures, RETT can also adjust the Stack Pointer (SP) to dis­card a specified number of bytes from the original stack as surplus parameter space.
RETI is used to return from a maskable interrupt service procedure. A difference of RETT, RETI also informs the on­chip ICU as well as any external interrupt control logic that interrupt service has completed. Since interrupts are gener­ally asynchronous external events, RETI does not discard parameters from the stack.
Both of the above instructions always restore the Program Counter (PC) and the Processor Status Register from the interrupt stack. If the Direct-Exception mode is disabled, they also restore the MOD and SB register contents.
Fig-
ures 3-7
and
3-8
show the RETT and RETI instruction flows
when the Direct-Exception mode is disabled.
30
Loading...
+ 72 hidden pages
Buy Points How it Works FAQ Contact Us Questions and Suggestions Privacy Policy Cookie Policy Terms of Use