Extended assemblt and memory clobbering

[John V. Shahid]

Hi there,

I'm trying to write in-line assembly that clobbers an array in memory, but unfortunately I have no idea how to do that. Here is a piece of code from my program:

/*
There is OUTPUT_NODES arrays of 2 doubles. This code computes the distance between each of those OUTPUT_NODES arrays found in the network_t structure and the input array provided as an argument, too. It then stores the results in the distance and do some other stuff with it (this isn't shown in the code below).
*/

#define OUTPUT_NODES 10
#define INPUT_NODES 2
typedef struct
{
>     double weight[OUTPUT_NODES][INPUT_NODES];
} network_t;
void fillDistance(double input[INPUT_NODES], network_t *network)
{
     double distance[OUTPUT_NODES];
     asm(
          "xorq %%rax, %%rax\n\t"                                                 /*Clear rax*/
          "movupd %[input], %%xmm0\n"                                      /*load the input vector in xmm0*/
          "0:\t movupd (%[weights],%%rax,8), %%xmm1\n\t"       /* load the first weight vector into xmm1*/
          "subpd %%xmm0, %%xmm1\n\t"                                    /*subtract the two vectors */
          "mulpd %%xmm1,  %%xmm1\n\t"                                   /*square the result*/
          "unpckhpd %%xmm1, %%xmm2\n\t"                              /*bring the upper quadword of xmm1 in the low quadword of xmm2*/
          "addpd %%xmm2, %%xmm1\n\t"                                    /*add low quadword of xmm1 with low quadword of xmm2(former high quadword of xmm1)*/
          "movupd %%xmm1, (%[distance],%%rax,4)\n\t"            /*store the low quadword of xmm1 in the first elementof distance array*/
          "addq $2, %%rax\n\t"                                                        /*since scaling in x86 has a mximum of 8 and we are dealing with arrays of 2 doubles, we add 2 to rax and multiply the scale y 2*/
          "loop 0b" ::                                                                        /*start with the next weight array and store its results in the next element of distance*/
     "c"(OUTPUT_NODES), [input]"m"(input), [weights]"r"(network->weights), [distance]"r"(distance) :
          "rax", "xmm0", "xmm1", "xmm2"
     );
     /* Do some other stuff with distance and then return */
}

My question is: how can I tell the compiler that I'm clobbering the distance array. I couldn't just add "m" to the output list since the compiler requires an lvalue to be specified.