mirror of
https://sourceware.org/git/binutils-gdb.git
synced 2024-12-15 04:31:49 +08:00
d6dafb7c8e
* ppc-sysv-tdep.c (ppc64_sysv_abi_push_dummy_call): Implement correct ABI for AltiVec vector arguments. testsuite/ChangeLog: * gdb.arch/altivec-abi.c (vec_func): Make use of intv_on_stack_f when computing result. * gdb.arch/altivec-abi.exp: Update expected results.
142 lines
5.3 KiB
C
142 lines
5.3 KiB
C
#include <altivec.h>
|
|
|
|
vector short vshort = {111, 222, 333, 444, 555, 666, 777, 888};
|
|
vector unsigned short vushort = {100, 200, 300, 400, 500, 600, 700, 800};
|
|
vector int vint = {-10, -20, -30, -40};
|
|
vector unsigned int vuint = {1111, 2222, 3333, 4444};
|
|
vector char vchar = {'a','b','c','d','e','f','g','h','i','l','m','n','o','p','q','r'};
|
|
vector unsigned char vuchar = {'A','B','C','D','E','F','G','H','I','L','M','N','O','P','Q','R'};
|
|
vector float vfloat = {1.25, 3.75, 5.5, 1.25};
|
|
|
|
vector short vshort_d = {0,0,0,0,0,0,0,0};
|
|
vector unsigned short vushort_d = {0,0,0,0,0,0,0,0};
|
|
vector int vint_d = {0,0,0,0};
|
|
vector unsigned int vuint_d = {0,0,0,0};
|
|
vector char vchar_d = {'z','z','z','z','z','z','z','z','z','z','z','z','z','z','z','z'};
|
|
vector unsigned char vuchar_d = {'Z','Z','Z','Z','Z','Z','Z','Z','Z','Z','Z','Z','Z','Z','Z','Z'};
|
|
vector float vfloat_d = {1.0, 1.0, 1.0, 1.0};
|
|
|
|
struct test_vec_struct
|
|
{
|
|
vector signed short vshort1;
|
|
vector signed short vshort2;
|
|
vector signed short vshort3;
|
|
vector signed short vshort4;
|
|
};
|
|
|
|
static vector signed short test4[4] =
|
|
{
|
|
(vector signed short) {1, 2, 3, 4, 5, 6, 7, 8},
|
|
(vector signed short) {11, 12, 13, 14, 15, 16, 17, 18},
|
|
(vector signed short) {21, 22, 23, 24, 25, 26, 27, 28},
|
|
(vector signed short) {31, 32, 33, 34, 35, 36, 37, 38}
|
|
};
|
|
|
|
void
|
|
struct_of_vector_func (struct test_vec_struct vector_struct)
|
|
{
|
|
vector_struct.vshort1 = vec_add (vector_struct.vshort1, vector_struct.vshort2);
|
|
vector_struct.vshort3 = vec_add (vector_struct.vshort3, vector_struct.vshort4);
|
|
}
|
|
|
|
void
|
|
array_of_vector_func (vector signed short *matrix)
|
|
{
|
|
matrix[0] = vec_add (matrix[0], matrix[1]);
|
|
matrix[2] = vec_add (matrix[2], matrix[3]);
|
|
}
|
|
|
|
vector int
|
|
vec_func (vector short vshort_f, /* goes in v2 */
|
|
vector unsigned short vushort_f, /* goes in v3 */
|
|
vector int vint_f, /* goes in v4 */
|
|
vector unsigned int vuint_f, /* goes in v5 */
|
|
vector char vchar_f, /* goes in v6 */
|
|
vector unsigned char vuchar_f, /* goes in v7 */
|
|
vector float vfloat_f, /* goes in v8 */
|
|
vector short x_f, /* goes in v9 */
|
|
vector int y_f, /* goes in v10 */
|
|
vector char a_f, /* goes in v11 */
|
|
vector float b_f, /* goes in v12 */
|
|
vector float c_f, /* goes in v13 */
|
|
vector int intv_on_stack_f)
|
|
{
|
|
|
|
vector int vint_res;
|
|
vector unsigned int vuint_res;
|
|
vector short vshort_res;
|
|
vector unsigned short vushort_res;
|
|
vector char vchar_res;
|
|
vector float vfloat_res;
|
|
vector unsigned char vuchar_res;
|
|
|
|
vint_res = vec_add (vint_f, intv_on_stack_f);
|
|
vint_res = vec_add (vint_res, y_f);
|
|
vuint_res = vec_add (vuint_f, ((vector unsigned int) {5,6,7,8}));
|
|
vshort_res = vec_add (vshort_f, x_f);
|
|
vushort_res = vec_add (vushort_f,
|
|
((vector unsigned short) {1,2,3,4,5,6,7,8}));
|
|
vchar_res = vec_add (vchar_f, a_f);
|
|
vfloat_res = vec_add (vfloat_f, b_f);
|
|
vfloat_res = vec_add (c_f, ((vector float) {1.1,1.1,1.1,1.1}));
|
|
vuchar_res = vec_add (vuchar_f,
|
|
((vector unsigned char) {'a','a','a','a','a','a','a','a','a','a','a','a','a','a','a','a'}));
|
|
|
|
return vint_res;
|
|
}
|
|
|
|
void marker(void) {};
|
|
|
|
int
|
|
main (void)
|
|
{
|
|
vector int result = {-1,-1,-1,-1};
|
|
vector short x = {1,2,3,4,5,6,7,8};
|
|
vector int y = {12, 22, 32, 42};
|
|
vector int intv_on_stack = {12, 34, 56, 78};
|
|
vector char a = {'v','e','c','t','o','r',' ','o','f',' ','c','h','a','r','s','.' };
|
|
vector float b = {5.5, 4.5, 3.75, 2.25};
|
|
vector float c = {1.25, 3.5, 5.5, 7.75};
|
|
|
|
vector short x_d = {0,0,0,0,0,0,0,0};
|
|
vector int y_d = {0,0,0,0};
|
|
vector int intv_on_stack_d = {0,0,0,0};
|
|
vector char a_d = {'q','q','q','q','q','q','q','q','q','q','q','q','q','q','q','q'};
|
|
vector float b_d = {5.0, 5.0, 5.0, 5.0};
|
|
vector float c_d = {3.0, 3.0, 3.0, 3.0};
|
|
|
|
int var_int = 44;
|
|
short var_short = 3;
|
|
struct test_vec_struct vect_struct;
|
|
|
|
vect_struct.vshort1 = (vector signed short){1, 2, 3, 4, 5, 6, 7, 8};
|
|
vect_struct.vshort2 = (vector signed short){11, 12, 13, 14, 15, 16, 17, 18};
|
|
vect_struct.vshort3 = (vector signed short){21, 22, 23, 24, 25, 26, 27, 28};
|
|
vect_struct.vshort4 = (vector signed short){31, 32, 33, 34, 35, 36, 37, 38};
|
|
|
|
marker ();
|
|
#if 0
|
|
/* This line is useful for cutting and pasting from the gdb command line. */
|
|
vec_func(vshort,vushort,vint,vuint,vchar,vuchar,vfloat,x,y,a,b,c,intv_on_stack)
|
|
#endif
|
|
result = vec_func (vshort, /* goes in v2 */
|
|
vushort, /* goes in v3 */
|
|
vint, /* goes in v4 */
|
|
vuint, /* goes in v5 */
|
|
vchar, /* goes in v6 */
|
|
vuchar, /* goes in v7 */
|
|
vfloat, /* goes in v8 */
|
|
x, /* goes in v9 */
|
|
y, /* goes in v10 */
|
|
a, /* goes in v11 */
|
|
b, /* goes in v12 */
|
|
c, /* goes in v13 */
|
|
intv_on_stack);
|
|
|
|
struct_of_vector_func (vect_struct);
|
|
array_of_vector_func (test4);
|
|
|
|
return 0;
|
|
}
|
|
|