Table of Tables

2.1: Improvement in SPARC code density resulting from instruction set changes

2.2: Frequency of 2-address instructions in SPARC code

2.3: Frequency of Last Result Re-use

3.1: Video Decompression Circuits from [RABA94]

3.2: Programmer-visibility of parallelism

3.3: Effective parallelism in a 6 stage pipeline

3.4: Instruction Mix

3.5: Performance Advantage of Major Processor Features (from [JOHN91])

3.6: Branch Prediction and Power Efficiency

4.1: Performance Of Synchronous and Asynchronous Pipelines

4.2: Performance Of Synchronous and Asynchronous Superscalar Processors

4.3: Recent Asynchronous Microprocessors

4.4: Characteristics of the AMULET1 compared with ARM6; slow-slow silicon

5.1: Number of times each instruction result is subsequently used (from[ENDE93])

5.2: Instruction Set Summary

5.3: Functional Unit Use for Conventional Program (Example 1)

5.4: Functional Unit Use for SCALP Program (Example 1)

5.5: Functional Unit Use for Conventional Program (example 2)

5.6: Functional Unit Use for SCALP Program (Example 2)

6.1: Macromodules used in the SCALP implementation

6.2: Issuer Decode Block Function

6.3: Instruction Issuer Architectural Performance

6.4: Total Gate Counts

6.5: Gates per Block

6.6: SCALP, AMULET and ARM sizes

7.1: Illustrating the use of the Geometric Mean

7.2: Static Code Size / bytes

7.3: Dynamic Code Size / bytes

7.4: Register Bank and Move Unit Instructions per Useful Instruction

7.5: FIG fetch performance, no branches

7.6: Instruction issuer performance

7.7: Functional Unit Performance

7.8: Network Performance

7.9: Performance of Isolated Pipeline Stages

7.10: Pipeline Decoupling Queue Lengths

7.11: Ideal and Actual Queue Lengths

7.12: Branch Latency

7.13: Instruction Sequence Throughput

7.14: Example Program Throughput

7.15: Maximum Potential Functional Unit Parallelism for Example Programs

7.16: Average Functional Unit Parallelism

7.17: Example Program Performance

7.18: Power Saving due to Inactive Functional Units

8.1: Example Program Code Sizes for SCALP with Separate Result Routing

9.1: Benchmark Programs