From 554fd8c5195424bdbcabf5de30fdc183aba391bd Mon Sep 17 00:00:00 2001 From: upstream source tree Date: Sun, 15 Mar 2015 20:14:05 -0400 Subject: obtained gcc-4.6.4.tar.bz2 from upstream website; verified gcc-4.6.4.tar.bz2.sig; imported gcc-4.6.4 source tree from verified upstream tarball. downloading a git-generated archive based on the 'upstream' tag should provide you with a source tree that is binary identical to the one extracted from the above tarball. if you have obtained the source via the command 'git clone', however, do note that line-endings of files in your working directory might differ from line-endings of the respective files in the upstream repository. --- gcc/testsuite/gcc.dg/vmx/varargs-4.c | 291 +++++++++++++++++++++++++++++++++++ 1 file changed, 291 insertions(+) create mode 100644 gcc/testsuite/gcc.dg/vmx/varargs-4.c (limited to 'gcc/testsuite/gcc.dg/vmx/varargs-4.c') diff --git a/gcc/testsuite/gcc.dg/vmx/varargs-4.c b/gcc/testsuite/gcc.dg/vmx/varargs-4.c new file mode 100644 index 000000000..a17551953 --- /dev/null +++ b/gcc/testsuite/gcc.dg/vmx/varargs-4.c @@ -0,0 +1,291 @@ +#include "harness.h" +#include +#include +#include + +typedef struct n_a +{ + signed char m1; + short m2; + int m3; + double m4; + vector float m5; +} +n_a; + +static n_a gn_a; + +static int +lay(char *p, int start, int end, int n) +{ + int b; + unsigned char ch; + unsigned int mask; + + start *= 8; + end *= 8; + n *= 8; + + for (b = 0; b + 8 <= start; b += 8) + { + ch = *p++; + if (ch != 0xff) + for (mask = 0x80; mask; b++, mask >>= 1) + if ((ch & mask) != mask) + return b; + } + + if (b < start) + { + ch = *p++; + for (mask = 0x80; b < start; b++, mask >>= 1) + if ((ch & mask) != mask) + return b; + for (; mask && b < end; b++, mask >>= 1) + if ((ch & mask) != 0) + return b; + } + + for (; b + 8 <= end; b += 8) + { + ch = *p++; + if (ch != 0) + for (mask = 0x80; mask; b++, mask >>= 1) + if ((ch & mask) != 0) + return b; + } + + if (b < end) + { + ch = *p++; + for (mask = 0x80; b < end; b++, mask >>= 1) + if ((ch & mask) != 0) + return b; + for (; mask && b < n; b++, mask >>= 1) + if ((ch & mask) != mask) + return b; + } + + for (; b + 8 <= n; b += 8) + { + ch = *p++; + if (ch != 0xff) + for (mask = 0x80; mask; b++, mask >>= 1) + if ((ch & mask) != mask) + return b; + } + + return n; +} + +static void +initn_a(signed char p1, short p2, int p3, double p4, vector float p5) +{ + n_a i; + + i.m1 = p1; + i.m2 = p2; + i.m3 = p3; + i.m4 = p4; + i.m5 = p5; + + check(i.m1 == 77, "i.m1"); + check(i.m2 == 1924, "i.m2"); + check(i.m3 == -1471601920, "i.m3"); + check(vec_all_eq(i.m5, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})), + "i.m5"); + + check(sizeof(n_a) == 32, "sizeof(n_a)"); + + check(offsetof(n_a, m1) == 0, "offsetof(m1)"); + check(offsetof(n_a, m2) == 2, "offsetof(m2)"); + check(offsetof(n_a, m3) == 4, "offsetof(m3)"); + check(offsetof(n_a, m4) == 8, "offsetof(m4)"); + check(offsetof(n_a, m5) == 16, "offsetof(m5)"); + + check(sizeof(i.m1) == 1, "sizeof(m1)"); + check(sizeof(i.m2) == 2, "sizeof(m2)"); + check(sizeof(i.m3) == 4, "sizeof(m3)"); + check(sizeof(i.m4) == 8, "sizeof(m4)"); + check(sizeof(i.m5) == 16, "sizeof(m5)"); + +#define lay_check(field) do { \ + memset((char *)&i, 0xFF, sizeof(i)); \ + lay_reset(field); \ + check(lay((char *)&i, \ + offsetof(n_a, field), \ + offsetof(n_a, field) + sizeof(i.field), \ + sizeof(i)) == sizeof(i)*8, \ + "lay(" #field ")"); \ + } while (0) +#define lay_reset(field) i.field = 0 + + lay_check(m1); + lay_check(m2); + lay_check(m3); + lay_check(m4); +#undef lay_reset +#define lay_reset(field) i.field = ((vector float){0,0,0,0}) + lay_check(m5); + +#undef lay_check +#undef lay_reset +} + +n_a +valuen_a(void) +{ + return gn_a; +} + +n_a * +addrn_a(void) +{ + return &gn_a; +} + +static void +eqn_a(n_a * a) +{ + check(a->m1 == 77, "a->m1"); + check(a->m2 == 1924, "a->m2"); + check(a->m3 == -1471601920, "a->m3"); + check(vec_all_eq(a->m5, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})), + "a->m5"); +} + +static void +getsn_a(n_a * a) +{ + a->m1 = 77; + a->m2 = 1924; + a->m3 = -1471601920; + a->m4 = 3.65e+18; + a->m5 = ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08}); +} + +static void +varlistn_a(signed char p1, va_list ap) +{ + n_a q; + q.m1 = p1; + q.m2 = va_arg(ap, int); + q.m3 = va_arg(ap, int); + q.m4 = va_arg(ap, double); + q.m5 = va_arg(ap, vector float); + + check(q.m1 == 77, "q.m1"); + check(q.m2 == 1924, "q.m2"); + check(q.m3 == -1471601920, "q.m3"); + check(vec_all_eq(q.m5, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})), + "q.m5"); +} + +static void +varargsn_a(signed char p1, ...) +{ + n_a q, r; + va_list ap; + + va_start(ap, p1); + q.m1 = p1; + q.m2 = va_arg(ap, int); + q.m3 = va_arg(ap, int); + q.m4 = va_arg(ap, double); + q.m5 = va_arg(ap, vector float); + va_end(ap); + + check(q.m1 == 77, "q.m1"); + check(q.m2 == 1924, "q.m2"); + check(q.m3 == -1471601920, "q.m3"); + check(vec_all_eq(q.m5, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})), + "q.m5"); + + va_start(ap, p1); + r.m1 = p1; + r.m2 = va_arg(ap, int); + r.m3 = va_arg(ap, int); + r.m4 = va_arg(ap, double); + r.m5 = va_arg(ap, vector float); + va_end(ap); + + check(r.m1 == 77, "r.m1"); + check(r.m2 == 1924, "r.m2"); + check(r.m3 == -1471601920, "r.m3"); + check(vec_all_eq(r.m5, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})), + "r.m5"); + + va_start(ap, p1); + varlistn_a(p1, ap); + va_end(ap); +} + +static void +test() +{ + static struct + { + char a; + n_a b; + } + s; + n_a v[3], a, *p; + + static n_a i = { 77, 1924, -1471601920, 3.65e+18, {-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08} }; + + memset((char *)&(v), -1, sizeof(v)); + v[1] = s.b; + check(lay((char *)&v, sizeof(n_a), sizeof(n_a)*2, sizeof(n_a)*3) == sizeof(n_a)*3*8, + "structure assignment"); + + check(i.m1 == 77, "i.m1"); + check(i.m2 == 1924, "i.m2"); + check(i.m3 == -1471601920, "i.m3"); + check(vec_all_eq(i.m5, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})), + "i.m5"); + + initn_a(77, 1924, -1471601920, 3.65e+18, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})); + varargsn_a(77, 1924, -1471601920, 3.65e+18, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})); + + gn_a.m1 = 77; + gn_a.m2 = 1924; + gn_a.m3 = -1471601920; + gn_a.m4 = 3.65e+18; + gn_a.m5 = ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08}); + a = valuen_a(); + + check(a.m1 == 77, "a.m1"); + check(a.m2 == 1924, "a.m2"); + check(a.m3 == -1471601920, "a.m3"); + check(vec_all_eq(a.m5, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})), + "a.m5"); + + p = addrn_a(); + + check(p->m1 == 77, "p->m1"); + check(p->m2 == 1924, "p->m2"); + check(p->m3 == -1471601920, "p->m3"); + check(vec_all_eq(p->m5, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})), + "p->m5"); + + eqn_a(&a); + + check(gn_a.m1 == 77, "gn_a.m1"); + check(gn_a.m2 == 1924, "gn_a.m2"); + check(gn_a.m3 == -1471601920, "gn_a.m3"); + check(vec_all_eq(gn_a.m5, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})), + "gn_a.m5"); + + getsn_a(&v[0]); + v[2].m1 = v[0].m1; + v[2].m2 = v[0].m2; + v[2].m3 = v[0].m3; + v[2].m4 = v[0].m4; + v[2].m5 = v[0].m5; + + check(v[2].m1 == 77, "v[2].m1"); + check(v[2].m2 == 1924, "v[2].m2"); + check(v[2].m3 == -1471601920, "v[2].m3"); + check(vec_all_eq(v[2].m5, ((vector float){-1.38e+09, 5.96e+08, 6.88e+08, -3.2e+08})), + "v[2].m5"); +} -- cgit v1.2.3