Assembly code performance evaluation apparatus and method

• US 6,598,221 B1
• Filed: 04/13/2000
• Issued: 07/22/2003
• Est. Priority Date: 04/13/2000
• Status: Expired due to Fees

• Current Assignee: NXP B.V. (NXP Semiconductors NV)
• Original Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Inventors: Sohier, Olivier, Auguin, Michel, Pegatoquet, Alain
• Primary Examiner(s): Dam, Tuan Q.
• Assistant Examiner(s): VU, TUAN A
• Application Number:
  US09/549,234
• Time in Patent Office: 1,195 Days
• Field of Search: 717/140, 717/162, 717/124-135, 703/20, 703/22
• US Class Current: 717/152
• CPC Class Codes: G06F 11/3466 Performance evaluation by tracing or monitoring
• CPC Class Codes: G06F 2201/865 Monitoring of software
• CPC Class Codes: G06F 8/447 Target code generation

Abstract

An assembly code performance evaluation apparatus is provided which includes a host computer, a target digital signal processor (DSP) compiler, and a performance estimation program. The host computer includes processing circuitry, memory and a host compiler to use test sequences and generate dynamic information. The target digital signal processor compiler communicates with the processing circuitry. The performance estimation program is implemented on the host processing circuitry and is operative to annotate application source code and to generate an estimation of an optimized assembly code. A method is also provided.

Claims

1. An assembly code performance evaluation apparatus, comprising:

a host computer having processing circuitry, memory and a host compiler, the host compiler operative to execute the program using test sequences and generate dynamic information;

a target digital signal processor (DSP) compiler communicating with the processing circuitry; and

a performance estimation program implemented on the host processing circuitry and operative to annotate application source code, and generate an estimation of an optimized assembly code;

wherein the performance estimation program includes a plurality of rewriting rules that are applied to an RTL intermediate representation and wherein one of the plurality of rewriting rules is operative to remove a shift operation by 0, 1, or −

1 following a multiplication operation.

2. The assembly code performance evaluation apparatus of claim 1 wherein the rewriting rules comprise rules operative to remove useless operations.

3. The assembly code performance evaluation apparatus of claim 1 wherein the rewriting rules comprise rules operative to synthesize DSP specific instructions.

4. The assembly code performance evaluation apparatus of claim 1 wherein the rewriting rules are applied on a Directed Acyclic Graph (DAG) that represents basic blocks of the application program.

5. The assembly code performance evaluation apparatus of claim 1 wherein one of the rewriting rules is operative to remove all arithmetic operations that perform computations on address registers within nested loops.

6. A method of evaluating performance of assembly code, comprising:

providing a target compiler;

providing a C code annotated with dynamic information based upon execution of the C code with test sequences on a host computer;

generating an RTL intermediate representation of the application;

defining a set of rewriting rules;

applying the set of rewriting rules to the RTL intermediate representation; and

generating an estimation of an optimized assembly code;

wherein one of the rewriting rules is operative to remove a shift operation by 0, 1, or −

1 following a multiplication operation.

7. The method of claim 6 wherein the set of rewriting rules is applied to the RTL representation so as to match as closely as possible a DSP execution scheme.

8. The method of claim 6 wherein the set of rewriting rules comprises rules operative to remove useless operations.

9. The method of claim 6 wherein the set of rewriting rules comprises rules operative to synthesize DSP specific instructions.

10. The method of claim 6 wherein the rewriting rules are applied on a Directed Acyclic Graph (DAG) that represents basic blocks of the application program.

11. The method of claim 6 wherein the rewriting rules are derived experientially using a programmer'"'"'s knowledge.

12. The method of claim 6 wherein the rewriting rules are derived experimentally.

13. The method of claim 6 wherein one of the rewriting rules is operative to remove all arithmetic operations that perform computations on address registers within nested loops.

14. The method of claim 6 wherein one of the rewriting rules is operative to replace an addition operation with a negative operand by a subtraction operation with a positive operand.

15. The method of claim 6 wherein one of the rewriting rules is operative to synthesize a multiplication operation with two identical input data in a square root node.

16. A method of evaluating performance of assembly code, comprising:

providing a target compiler;

providing a C code annotated with dynamic information based upon execution of the C code with test sequences on a host computer;

generating an RTL intermediate representation of the application;

defining a set of rewriting rules;

applying the set of rewriting rules to the RTL intermediate representation; and

generating an estimation of an optimized assembly code;

wherein one of the rewriting rules is operative to delete a shift operation by +/−

16 after a memory read or write operation using an indirect addressing mode.

17. A method of evaluating performance of assembly code, comprising:

providing a target compiler;

providing a C code annotated with dynamic information based upon execution of the C code with test sequences on a host computer;

generating an RTL intermediate representation of the application;

defining a set of rewriting rules;

applying the set of rewriting rules to the RTL intermediate representation; and

generating an estimation of an optimized assembly code;

wherein one of the rewriting rules is operative to remove a comparison operation with 0 before a branch operation.