Here are a few guidelines to keep in mind regarding C data types on the C6000?

秒速五厘米 2022-05-21 01:04 127阅读 0赞

![1350467577_5730.png][]

\------------------------------------------------------------------------------------------------------------------------------

1. Use short types for integer multiplication. As with most fixed-point DSPs, our ‘C62x devices use a 16x16 integer multiplier. If you specify an int  multiply, a software function in the runtime support library will be called. (Note, the ‘C67x devices do have a 32x32→64-bit multiply instruction, MPYID.)

2. Use  int types for counters and indexes. As we examine during the next chapter, all registers  and data paths are 32-bits wide.

3. Avoid accidentally mixinglong andintvariables. Many compilers allocate 32-bits for both types, thus some users interchange these types. The ‘C6000 allocates longs at 40-bits to take advantage of 40-bit hardware within the CPU. If you mix types, the compiler may be forced to manage this – which will most likely cost you some performance. Why 40-bits? The extra 8-bits are often used to provide headroom in integer operations. Also, they can act like an 8-bit “carry bit”.

4. On ‘C67x devices, 32-bitfloatoperations are performed in hardware. The ‘C6000 supports  IEEE 32-bit floating-point math.

5. The  double precision floating-point hardware supports IEEE 64-bit floating-point math.

6.  Pointers, at 32-bits, can reach across the entire‘C6000 memory-map.

====================================================================================================

[1350467577_5730.png]: https://img-my.csdn.net/uploads/201210/17/1350467577_5730.png