XC8从不使用POSTINC

这是它处理间接寻址的方法，包括PoST或Pro，而不是在指令集中。没有什么了不起的，只有1个寄存器是低效的，这就是为什么我不再使用8位芯片。不知道强制编译器这样做，也许一个付费编译器会这么做。就像你的SAI一样。使用汇编程序。



      以下为原文

    It's the way it handles indirect addressing including post/pre, just not in the instruction set.
Nothing really amazing, it's inefficient only having 1 register, that's why I don't use 8bit chips anymore.
 
Don't know about forcing the compiler to do that, maybe a paid compiler would do it.
 
Like you said, use the assembler.

您好，XC8编译器使用的是哪种许可证？自由模式或Pro模式。在自由模式下有很多优化，如果你还没有做过的话，有可能尝试2个月的最大优化。你不必在程序集中编写所有代码，只需要那些真正重要的部分。你可以完全编写一个函数。程序集，你可以在C中写一个函数条目和完成，并用一大块汇编代码填充它，或者你可以在C代码中加入汇编指令。注意，在C代码中加入汇编指令，如果没有C编译器知道发生了什么，可能会导致错误。例如改变FSR寄存器。R内容，而不是恢复原始值，可能会导致以下代码中的错误。这里是一个在C函数中封装的汇编代码的例子，它已经用于连接的交流测量和PIC16和PIC18上的RMS平均计算。



      以下为原文

    Hi,
Which kind of license for XC8 compiler are you using? Free  or Pro mode.
There are a lot of optimizations not done in Free mode,
 
There is a possibility to try out maximum optimizations for 2 months, if you haven't done so already.
You don't have to write all the code in assembly, only those parts where it really matter.
You may write a function entirely in assembly,
You may write a function entry and completion in C and fill it with a big block of assembly code,
or you may put in assembly instructions in C code.
 
Be careful, putting in assembly instructions in C code, without C compiler knowing what is going on, 
may cause errors.
E.g.  changing  FSR register contents, and not restoring the original value, may cause error in following code.
 
Here is an example of assembly code wrapped in a C function, it have been used in connection AC measurements and RMS average calculations on PIC16 and PIC18.:
/*
 *  PIC16, PIC18
 *    Calculate average from a running accumulator  Avg = Avg >> 10;
 *    Modified from code contributed on Microchip forum by "co_worker"
 *  containing suggestions from "1and0" and "andig"
 *  Microchip Forum http://www.microchip.com/forums/FindPost/848067
 */
#include
#include

#define IN_ASM
#ifndef     Avgshift
    #define Avgshift    10
#endif
/*
 * Unsigned integer shift
 *
 *   Result = Arg >> 10;
 *
 * Parameters:
 *   a = 24-bit unsigned integer accumulating samples.
 *
 * Return Value:
 *   The function returns a 16-bit unsigned integer Result.
 *
 * Program Memory:
 *   20 instructions including call and return
 *
 * Execution Time:
 *   14 cycles excluding the call (prolog) and return (epilog).
 *
 * Remarks:
 *
 *
 * History:
 *      2016-05-04                            - by Arne Bergseth "Mysil"
 */
uint16_t uShift_24_10(uint24_t Arg)
{
    uint16_t Result = 0;    /* Zero to make compiler happy, */
                             /* zeroing may be omitted if compiler warning is acceptable. */
#if defined(IN_C)
    result  = a >> avgshift;
#endif

#if defined(IN_ASM) /* ? instructions, ? cycles, ? avg ? to be tested. */
/*
 * Ma.
 *
 * rh:rl = (au:ah:al >> 10)
 *
 */
#asm
#ifdef _PIC18
    #define rh (uShift_24_10@Result+1)
    #define rl (uShift_24_10@Result)
    #define au (uShift_24_10@Arg+2)
    #define ah (uShift_24_10@Arg+1)
    #define al (uShift_24_10@Arg)

#elif  _PIC14 || _PIC14E
    #define rh (uShift_24_10@Result+1)&0x7F
    #define rl (uShift_24_10@Result)&0x7F
    #define au (uShift_24_10@Arg+2)&0x7F
    #define ah (uShift_24_10@Arg+1)&0x7F
    #define al (uShift_24_10@Arg)&0x7F
#endif
    BANKSEL(uShift_24_10@Arg)
                    /*    rh:rl = au:ah >> 1    Do 8 bit shift by picking bytes
                     *                            then 1 bit shift by instruction */
    bcf        STATUS,0    /* clear the carry bit. */
#if (Avgshift == 8)        /* Do 8 bit shift by picking bytes . */
    movf    au,W
    movwf    rh
    movf    ah,W
    movwf    rl
#elif (Avgshift > 8)    /* then 1 bit shift by instruction. */
    rrcf     au,W        /* Move from Au into W */
    movwf    rh            /* store in Rh */
    rrcf    ah,W
    movwf    rl
#endif
#if (Avgshift == 10)    /* ru:rh:rl = ru:rh:rl >> 1 */
    clrc                /* one more rotation for 10 bit shift */
    rrcf    rh,F
    rrcf    rl,F        /* Shift into rl from the carry bit */
//    rrcf    rl,W        ;/* may use W for this. */
#else
    movf    rl,W        /* start with low byte */
#endif  

#endasm
#endif
    return Result;
} 
 
Regards,
   Mysil

得到一个16位芯片。添加x，[y++]，[-Z]



      以下为原文

    Get a 16bit chip.
 
add x,[y++],[--z]

我不认为XC8将编译一个16位的ChIPWink：我最后写了C和汇编的混合。效率提高了很多，但是代码可读性却呈指数下降，在编译器中重用FSR对的内容（在PRO模式下），在所有程序中找不到一个点。但是我知道手册警告这一点。下面是我做的一个例子。原始C代码：混合1：MIX2：用模拟器运行代码给出了以下结果（我必须注释“For（FLAG）”，因为SPI模块的这一部分在模拟器中没有实现）：IFTXLEN＝1（1循环）：C：20 Cy；Mix1：18 Cy；Mix2：16 CyftxLeN＝10（10个循环）：C:146 Cy；Mix1:81 Cy；MIX2:70 Cy*Cy指令Cyrimix2是我能够做到的最优化的版本。最小化最后一次传输到结束的等待和下一次传输之间的指令数目。



      以下为原文

   
I don't think XC8 will compile for a 16-bit chip wink:
 
Well I ended up writing a mix of C and assembly. The efficiency increases a lot, but code readability decreases exponentially.

I could not find a single point in all program where the compiler reuses the contents of the FSR pair (in PRO mode). But I know the manual warns about this.
 
Here is an example of what I did.
Original C code:
for (i = 0; i < txLen; i++)
{
 PIR1bits.SSP1IF = 0;
 SSP1BUF = txBuffer;
 while (PIR1bits.SSP1IF == 0);
}
 
Mix 1:
 i = txLen + 1;
#asm
 MOVLW low(_txBuffer)
 MOVWF FSR2, C
 MOVLW high(_txBuffer)
 MOVWF FSR2H, C;
#endasm
 while (--i)
 {
  PIR1bits.SSP1IF = 0;
#asm
  MOVFF POSTINC2, SSP1BUF;
#endasm
  while (PIR1bits.SSP1IF == 0);
 }
 
Mix2:
 if (txLen)
 {
  i = txLen;
#asm
  MOVLW low(_txBuffer)
  MOVWF FSR2, C
  MOVLW high(_txBuffer)
  MOVWF FSR2H, C;
#endasm
  PHY_CS = 0;
#asm
  goto asm_send;
#endasm
  do
  {
   while (SSP1STATbits.BF == 0);
   SSP1BUF; // Clear BF bit
#asm
asm_send:
   MOVFF POSTINC2, SSP1BUF;
#endasm
  } while (--i);
  while (SSP1STATbits.BF == 0);
 }
 
Running the code with the simulator gives the following results (I had to comment out the "while(flag);", because this part of the SPI module is not implement in the simulator):
If txLen = 1 (1 loop):
  C: 20 cy; Mix1: 18 cy; Mix2: 16 cy
If txLen = 10 (10 loops):
  C: 146 cy; Mix1: 81 cy; Mix2: 70 cy
*cy - instruction cycle
 
Mix 2 is the most optimized version that I was able to do. The number of instructions between the wait for the last transfer to finish and the next transfer are minimized.

这可能不会节省很多时间，因为你在等待STAT标志。我会写整个FUNC。在ASM中使其更易于阅读。C和内联ASM的混合看起来不整洁。



      以下为原文

    That may not save that much time since you are waiting for the STAT flag.
 
I'd write the whole func. in asm to make it easier to read.
A mix of C and inline asm does not look tidy.

如果TXLEN是8位，DO-（）循环可能会更好，所以试试这个：



      以下为原文

   
If txLen is 8-bit a do() loop might be better, so try this:
asm("lfsr 2,_txBuffer");
i = txLen;
do {
   PIR1bits.SSP1IF = 0;
   SSP1BUF = POSTINC2;
   while (PIR1bits.SSP1IF == 0);
} while (--i);

FIY，这可以用C语言编写，但是使用LFSR指令更有效。



      以下为原文

    FYI, this
#asm
MOVLW low(_txBuffer)
MOVWF FSR2, C
MOVLW high(_txBuffer)
MOVWF FSR2H, C;
#endasm

can be written in C as
FSR2 = txBuffer;
but use of the LFSR instruction is more efficient.

这就解决了问题，感谢用户1和0！你回答了主要的问题。“强制”XC8使用PosiCin。我已经在MIX2版本中使用了DOE（）循环，现在更可读（只有一个汇编指令）：



      以下为原文

     
This solves the problem.
Thanks user 1and0! You answered the main topic question. It is possible to "force" XC8 to use POSTINC.
I'm already using the do() loop in the Mix2 version, and now is more readable (only one assembly instruction):
if (txLen)
{
  i = txLen;
  asm("LFSR 2,_txBuffer"); // FSR2 = txBuffer;
  goto sendPayload;
  do
  {
    while (SSP1STATbits.BF == 0);
    SSP1BUF; // Clear BF bit
sendPayload:
    SSP1BUF = POSTINC2; // SSP1BUF = *txBuffer++;
  } while (--i);
  while (SSP1STATbits.BF == 0);
}

过去它是这样做的：这是从PICC V8列表文件中得到的。



      以下为原文

    It used to do it back in the old days:   222                           ; _buff loaded to fsr1
   223  0078C0  C0D2  FFE1          movff   ?_spi_write_sector,fsr1l
   224  0078C4  C0D3  FFE2          movff   ?_spi_write_sector+1,fsr1h
   225                           ;mmc.c: 84: u8 len;
   234                           ;mmc.c: 89: do {
   243                           ;mmc.c: 94: SSPBUF = *buff++;
   244  0078D8  CFE6 FFC9           movff   postinc1,4041   ;volatile
   245                           ;mmc.c: 95: } while (--len);
That's from a PICC v8 list file.

不要认为GOTO是必要的。试试这个：



      以下为原文

   
Don't think the GOTO is necessary. Try this:
if (txLen) {
  i = txLen;
  asm("LFSR 2,_txBuffer"); // FSR2 = txBuffer;
  do {
    SSP1BUF = POSTINC2; // SSP1BUF = *txBuffer++;
    while (SSP1STATbits.BF == 0);
    SSP1BUF; // Clear BF bit
  } while (--i);
}

那样的话，我在最后一次传输之后会失去几个周期。不要介意Goto指令。我并没有真正使用它，因为SSP1BUF是在代码到达这个部分之前编写的。在这个例子中我用它来显示速度优化。



      以下为原文

   
That way I’m losing a couple cycles after the last transmission.
Don’t mind the goto instruction. I’m not really using it, because SSP1BUF is written before the code reaches this part. I used it in this example just to show speed optimization.