C2000's Special Memory Layout - 土星环的基地

The C2000 DSP is the one-and-only modern chip with 16-bit bytes, so its memory addressing differs from the usual CHAR_BIT == 8.

This needs extra attention when doing zero-copy casts on protocol-layer data. I've dealt with it before, but today I ran into it again and had forgotten all about it, so I had to manually test it again. Let me write it down.

	U32	Actual memory layout
C2000	0x87654321	4321 8765
Normal 8-bit little-endian	0x87654321	21 43 65 87

	Float value	Actual memory layout	Memory layout as U32
C2000	123.456	E979 42F6	0x42F6E979
Normal 8-bit little-endian	123.456	79 E9 F6 42	0x42F6E979

In addition, it has strict access-alignment requirements: 32-bit data also requires addresses to be aligned to 2 bytes. Example:

uint16_t buffer[8];

uint32_t t32;

/* In fact, the following two both read memory as [0][1] and convert to u32 */
// On a normal CPU, the following would read memory as [1][2] and convert to u32
t32 = *(uint32_t *)(& buffer[1]);
// On a normal CPU, the following is the correct zero-copy way to read [0][1]
t32 = *(uint32_t *)(& buffer[0]);

/* The following two will instead read memory as [2][3] and convert to u32 */
t32 = *(uint32_t *)(& buffer[3]);
t32 = *(uint32_t *)(& buffer[2]);