@@ -337,6 +337,30 @@ void print_f (vector float vec_expected,
}
#endif
+typedef union
+{
+ vector __int128_t i1;
+ __int128_t i2;
+ vector __uint128_t u1;
+ __uint128_t u2;
+ struct
+ {
+ long long d1;
+ long long d2;
+ };
+} U128;
+
+/* For a given long long VALUE, ensure it's stored from the beginning
+ of {u,}int128 memory storage (the low address), it avoids to load
+ unexpected data without the whole vector length. */
+
+static inline void
+getU128 (U128 *pu, unsigned long long value)
+{
+ pu->d1 = value;
+ pu->d2 = 0;
+}
+
int main() {
int i, j;
size_t len;
@@ -835,21 +859,24 @@ int main() {
#endif
}
+ U128 u_temp;
/* vec_xl_len() tests */
for (i = 0; i < 100; i++)
{
- data_c[i] = i;
- data_uc[i] = i+1;
- data_ssi[i] = i+10;
- data_usi[i] = i+11;
- data_si[i] = i+100;
- data_ui[i] = i+101;
- data_sll[i] = i+1000;
- data_ull[i] = i+1001;
- data_f[i] = i+100000.0;
- data_d[i] = i+1000000.0;
- data_128[i] = i + 12800000;
- data_u128[i] = i + 12800001;
+ data_c[i] = i;
+ data_uc[i] = i + 1;
+ data_ssi[i] = i + 10;
+ data_usi[i] = i + 11;
+ data_si[i] = i + 100;
+ data_ui[i] = i + 101;
+ data_sll[i] = i + 1000;
+ data_ull[i] = i + 1001;
+ data_f[i] = i + 100000.0;
+ data_d[i] = i + 1000000.0;
+ getU128 (&u_temp, i + 12800000);
+ data_128[i] = u_temp.i2;
+ getU128 (&u_temp, i + 12800001);
+ data_u128[i] = u_temp.u2;
}
len = 16;
@@ -1160,34 +1187,38 @@ int main() {
#endif
}
- vec_s128_expected1 = (vector __int128_t){12800000};
+ getU128 (&u_temp, 12800000);
+ vec_s128_expected1 = u_temp.i1;
vec_s128_result1 = vec_xl_len (data_128, len);
if (vec_s128_expected1[0] != vec_s128_result1[0])
{
#ifdef DEBUG
- printf("Error: vec_xl_len(), len = %d, vec_s128_result1[0] = %lld %llu; ",
- len, vec_s128_result1[0] >> 64,
- vec_s128_result1[0] & (__int128_t)0xFFFFFFFFFFFFFFFF);
- printf("vec_s128_expected1[0] = %lld %llu\n",
- vec_s128_expected1[0] >> 64,
- vec_s128_expected1[0] & (__int128_t)0xFFFFFFFFFFFFFFFF);
+ U128 u1, u2;
+ u1.i1 = vec_s128_result1;
+ u2.i1 = vec_s128_expected1;
+ printf ("Error: vec_xl_len(), len = %d,"
+ "vec_s128_result1[0] = %llx %llx; ",
+ len, u1.d1, u1.d2);
+ printf ("vec_s128_expected1[0] = %llx %llx\n", u2.d1, u2.d2);
#else
abort ();
#endif
}
vec_u128_result1 = vec_xl_len (data_u128, len);
- vec_u128_expected1 = (vector __uint128_t){12800001};
+ getU128 (&u_temp, 12800001);
+ vec_u128_expected1 = u_temp.u1;
if (vec_u128_expected1[0] != vec_u128_result1[0])
#ifdef DEBUG
{
- printf("Error: vec_xl_len(), len = %d, vec_u128_result1[0] = %lld; ",
- len, vec_u128_result1[0] >> 64,
- vec_u128_result1[0] & (__int128_t)0xFFFFFFFFFFFFFFFF);
- printf("vec_u128_expected1[0] = %lld\n",
- vec_u128_expected1[0] >> 64,
- vec_u128_expected1[0] & (__int128_t)0xFFFFFFFFFFFFFFFF);
+ U128 u1, u2;
+ u1.u1 = vec_u128_result1;
+ u2.u1 = vec_u128_expected1;
+ printf ("Error: vec_xl_len(), len = %d,"
+ "vec_u128_result1[0] = %llx %llx; ",
+ len, u1.d1, u1.d2);
+ printf ("vec_u128_expected1[0] = %llx %llx\n", u2.d1, u2.d2);
}
#else
abort ();
@@ -1421,8 +1452,13 @@ int main() {
and post data initialization is done. */
len = 16;
- vec_uc_expected1 = (vector unsigned char){ 16,15, 14, 13, 12, 11, 10, 9,
- 8, 7, 6, 5, 4, 3, 2, 1 };
+#if __LITTLE_ENDIAN__
+ vec_uc_expected1 = (vector unsigned char){16, 15, 14, 13, 12, 11, 10, 9,
+ 8, 7, 6, 5, 4, 3, 2, 1};
+#else
+ vec_uc_expected1 = (vector unsigned char){1, 2, 3, 4, 5, 6, 7, 8,
+ 9, 10, 11, 12, 13, 14, 15, 16};
+#endif
input_uc = (vector unsigned char){ 1, 2, 3, 4, 5, 6, 7, 8, 9,
10, 11, 12, 13, 14, 15, 16 };
vec_uc_result1 = vec_xl_len_r (&input_uc[0], len);
@@ -1441,30 +1477,3 @@ int main() {
}
}
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Hi, Test case builtins-6-p9-runnable.c doesn't work well on BE due to two problems: - When applying vec_xl_len onto data_128 and data_u128 with length 8, it expects to load 1280000[01] from the memory, but unfortunately assigning 1280000[01] to a {vector} {u,}int128 type variable, the value isn't guaranteed to be at the beginning of storage (in the low part of memory), which means the loaded value can be unexpected (as shown on BE). So this patch is to introduce getU128 which can ensure the given value shows up as expected and also update some dumping code for debugging. - When applying vec_xl_len_r with length 16, on BE it's just like the normal vector load, so the expected data should not be reversed from the original. Tested well on both powerpc64{,le}-linux-gnu, I'm going to push this soon. BR, Kewen ----- PR testsuite/114744 gcc/testsuite/ChangeLog: * gcc.target/powerpc/builtins-6-p9-runnable.c: Adjust for BE by fixing data_{u,}128, their uses and vec_uc_expected1, also adjust some formats. --- .../powerpc/builtins-6-p9-runnable.c | 119 ++++++++++-------- 1 file changed, 64 insertions(+), 55 deletions(-) -- 2.43.0