Kenton’s Score-O-Meter: 5 of 5.  CEVA’s new Application Optimizer closes the performance gap between C code and assembly, dramatically reducing—and sometimes eliminating—the need for assembly-level coding.  The toolchain radically cuts development time, and it puts CEVA on a much stronger competitive footing.

Review: CEVA’s Application Optimizer toolchain contains major upgrades that (according to CEVA) bring out-of-the box C code to within 1.5X the performance of hand-optimized assembly code—a commendable achievement.  Optimized C code can reach 1.2X the performance of assembly code—good enough to eliminate assembly coding in many cases.  The improved C performance means that developers can look forward to a much shorter DSP software development cycle.

The biggest upgrade in the toolchain is the new project build optimizer.  The project build optimizer generates near-optimal configurations based on system-level profiling.  The following video explains this tool:

Here’s how it works:  First, the user specifies a set of configurations to test.  Each configuration specifies build options such as compiler settings, memory sizes, libraries, data sets, and use cases (for example, control vs. DSP functionality).  The build optimizer builds, simulates, and profiles the code for each configuration.  The profiling results in a function-level database of performance and code size.

The build optimizer inspects this database, and for each function selects the configurations that returned the best balance of speed and performance.   These function-level configurations are mixed together to create a new set of project-level configurations.  The user can then select the configuration that best meets the project goals.

The project build optimizer is a remarkable tool.  Its ability to mix and match configurations at the function level is unusual—as far as I know, no other licensable processor company offers this capability.  The build optimizer is also notable for its system-level profiling capabilities.  The tools allow the user to specify the behavior of the cache system, the memory bus, and so on.  This behavioral data allows the build optimizer to return highly realistic performance data.

The toolchain includes a number of other performance-boosting advancements. These include:

• Upgrades to the C compiler that greatly improve the C vs. assembly performance ratio
• An enhanced post-linker optimizer that reduces code size
• Expanded DSP libraries
• All-new communication libraries

The other major new feature in the toolchain is test automation using built-in Perl support.  In addition to allowing users to write Perl scripts, the automation feature can capture mouse and keyboard inputs.  Among other things, this feature allows a user to record their actions during a session and send the recording to a colleague.  This ensures that developers in physically separate locations see the exact same scenarios.  The recording feature also allows users to leverage the automation features without learning Perl.

All of the new features give CEVA a DSP toolchain that is in many ways better than those offered by other IP vendors.  The upgrades also put CEVA on a more equal footing with industry leader Texas Instruments.  Just as importantly, the upgrades show that CEVA is paying attention to its customer’s needs and delivering solutions.  CEVA clearly understands that software development has become the bottleneck in SoC design, and its new tools go a long way towards relieving that bottleneck.