MIPS-3D® ASE
The MIPS-3D graphics and multimedia extension gives developers of digital entertainment and multimedia products a new tool that simultaneously lowers the cost and improves the performance of their designs. The MIPS-3D extension is intended for 64-bit consumer applications that need three-dimensional graphics but require minimal implementation costs for low-power or System-on-Chip (SOC) solutions.- MIPS-3D technology achieves 37 million polygons/sec for transform, clip check and perspective divide operations and 15 million polygons/sec with complex lighting operations.
- MIPS-3D technology is both a cost-effective and an efficient way to achieve high-performance 3D geometry processing within the context of a MIPS64® architecture. It eliminates the need to deploy expensive, dedicated 3D geometry engines and provides excellent code reduction and execution times in common 3D geometry and lighting operations.
- MIPS-3D technology extends the MIPS64 architecture and leverages the established infrastructure of software, development systems, operating systems and other hardware and software development tools designed for the MIPS® architecture.
Instruction Usage
CVT.PS.PW and CVT.PW.PS
- Paired-single and paired-word instructions provide fast data format conversions
ADDR and MULR
- Reduction add and multiply instructions accelerate matrix multiplication, an important operation in vertex transformation
CABS and BC1ANYnx
- The absolute compare instruction and the multiple-condition code branch instruction are useful in image clipping, which consists of identifying and eliminating vertices that are outside the view volume
RECIP1 and RECIP2
- The perspective divide operation assists in transforming verticles to screen coordinates
RSQRT1 and RSQRT2
- Reciporcal square root instructions are used in the normalization functions found in infinite and local lighting
MIPS-3D floating-point instructions exploit the MIPS64 floating-point unit and the paired-single data type. The PS (paired-single) operations provide 2-way SIMD (single instruction multiple data) capability by operating on two 32-bit floating-point values packed into a single 64-bit register. The primary use for paired-single is to implement 3D geometry operations, such as transforms, clip checking and lighting, and shading calculations. It is also useful in boosting the performance of floating-point digital signal processing.
