NEON Cortex-A8的装配示例

Question

我试图编译和例子唱霓虹灯装配代码为皮质A8使用这个二进制的BeagleBone黑板(BBB)。我使用eclipse工具GCC编译器和汇编程序，

GCC : arm-linux-gnueabi-gcc 汇编程序: arm-linux-gnueabi-as

下面的错误发生在一个示例中，对于我使用的每个示例，我都会发现类似的错误

Description Path    Resource    Location    Type
SP not allowed in register list -- `ldmia r12,{r4-r11,r13,lr}'      EXAMPLE_NEON    line 61, external location: /tmp/ccTXrczs.s C/C++ Problem

我用的代码

/************************

  \* neon.c       \*      \*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*/

#include <stdio.h>


__attribute__((aligned (16)))
unsigned short int data1[8];
unsigned short int data2[8];
unsigned short int out[8];

void* neontest_save_buffer[16];


void
neontest(unsigned short int *a, unsigned short int *b,
                unsigned short int* q)
{
  __asm__(
"   movw        r12, #:lower16:neontest_save_buffer\n\t"
"   movt        r12, #:upper16:neontest_save_buffer\n\t"
"   stmia       r12, {r4-r11, r13, lr}        @ save registers\n\t"
"   vld1.16     {q1}, [r0:128]\n\t"
"   vld1.16     {q2}, [r1:128]\n\t"
"   vadd.i16    q0, q1, q2\n\t"
"   vst1.32     {q0}, [r2:128]\n\t"
"   movw        r12, #:lower16:neontest_save_buffer\n\t"
"   movt        r12, #:upper16:neontest_save_buffer\n\t"
"   ldmia       r12, {r4-r11, r13, lr}        @ reload all registers and return\n\t"
"finish:\n\t"
    );

}

int
main(void)
{
    int i;

    for (i=0; i<8; i++)
    {
        data1[i]=i*10;
        data2[i]=5;
        out[i]=0;
    }

    neontest(data1, data2, out);

    printf("output is: ");
    for (i=0; i<7; i++)
    {
        printf("%d, ", out[i]);
    }
    printf("%d\n", out[i]);

  return(0);
}

Answer 1

看起来您使用的是Thumb32模式，在这种模式中，sp cannot be in the list of registers (来自[1])。

如果没有设置新堆栈，为什么需要保存堆栈？只需尝试从r13块和ldm块中删除ldm。

Answer 2

默认情况下，编译器使用拇指模式，在命令行中添加"-marm“以ARM模式编译代码：

arm-linux-gnueabihf-gcc -mcpu=cortex a8 -mfpu=neon -marm nonn.c

您还可以调整代码，让编译器进行寄存器保存/还原，这样就可以为ARM和thumb2指令集编译代码：

#include <stdio.h>


__attribute__((aligned (16)))
unsigned short int data1[8];
unsigned short int data2[8];
unsigned short int out[8];

void
neontest(unsigned short int *a, unsigned short int *b,
                unsigned short int* q)
{
    __asm volatile (
"   vld1.16     {q1}, [%[a]:128]\n\t"
"   vld1.16     {q2}, [%[b]:128]\n\t"
"   vadd.i16    q0, q1, q2\n\t"
"   vst1.32     {q0}, [%[q]:128]\n\t"

     : [q] "+r" (q)
     : [a] "r" (a), [b] "r" (b)
     : "q0", "q1", "q2"
    );
}

int
main(void)
{
    int i;

    for (i=0; i<8; i++)
    {
        data1[i]=i*10;
        data2[i]=5;
        out[i]=0;
    }

    neontest(data1, data2, out);

    printf("output is: ");
    for (i=0; i<7; i++)
    {
        printf("%d, ", out[i]);
    }
    printf("%d\n", out[i]);

  return(0);
}

臂-linux-gnueabihf-gcc-mcpu=皮层-a8 -mfpu=neon -marm nonn2.c

臂-linux-gnueabihf-gcc-mcpu=皮层-a8 -mfpu=neon -mthumb nonn2.c