gcc-4.4 optimization of a program using -O3 leads to incorrect assembly code

Issue

• gcc-4.4 optimization of a program using -O3 leads to incorrect assembly code
• The program is as follows:

#include <stdlib.h>

typedef struct bmRecord bmRecord;
struct bmRecord {
    bmRecord    *next;
};

typedef struct bmHeader bmhdr;
struct bmHeader {
    void        *blocklist;
    bmRecord    *freelist;

    int
        nrec,
        datasize,
        realsize,
        blocksize;
};

typedef  struct {
    int         x;
    int         y;
} vPoint;

typedef struct varPoint varPoint;
struct varPoint {
    varPoint    *next;
    vPoint       point;
};

typedef struct varShape varShape;
struct varShape {
    varShape    *next;
    varPoint    *points;
    int          nPoints;
};

static bmhdr    VarPointPool;

int getShape(void *inPtr, int *layer, int *nPoints, vPoint **points)
{
    static vPoint  tmpPoints[4096];
    varPoint      *p, *nextP;
    varShape      *s, **inShapes = inPtr;

    *layer = *nPoints = 0;
    s = *inShapes;
    if( !s ) return 1;

    *points = (vPoint*) tmpPoints;
    for( p = s->points; p; p = nextP ) {
        tmpPoints[(*nPoints)++] = p->point;
        nextP = p->next;
        {
            bmRecord *bm__y;
            bmhdr    *bm__z;

            bm__z = &(VarPointPool);
            bm__y = (bmRecord *) (void *)(p);
            bm__y->next = bm__z->freelist;
            bm__z->freelist = bm__y;
        }
    }
    *inShapes = s->next;
    return 1;
}

int call_getShape(void)
{
    varShape  inPtr1, *inPtr = &inPtr1;
    int       layer=0, nPoints=0;
    vPoint   *pouts[1024];
    varPoint *points = (varPoint *)calloc(1,sizeof(varPoint));

    inPtr->points = points;

    points->next = (varPoint *)calloc(1, sizeof(varPoint));
    points->next->next = 0;

    VarPointPool.freelist = (bmRecord *)0x100;
    getShape( &inPtr, &layer, &nPoints, &pouts[0]);

    return 0;
}

int main(int argc, char *argv[])
{
    call_getShape();
    return 0;
}