[Compiler-rt][MSan] Fix msan_print_shadow3.cc test for Big Endian archs

[eugenis]

maybe just define x_s as an array of char?

[mohit.bhakkad]

In D17264#353734, @eugenis wrote:

maybe just define x_s as an array of char?

Even after that we will have differences, as shadow is printed in hex format(i.e. each nibble)
so "12" will be printed as "21", "78" as "87" etc.

And it is because we are reversing it on the basis of endianness in DescribeMemoryRange()

[eugenis]

Ah, right. LGTM.
Btw, is that consistent with how other tools print memory on big endian platforms? Like gdb and objdump. If not, this can be changed.

[mohit.bhakkad]

In D17264#355720, @eugenis wrote:

Btw, is that consistent with how other tools print memory on big endian platforms? Like gdb and objdump. If not, this can be changed.

No, in gdb, endians are differentiated from byte level, so nibbles are not reversed while displaying.

[mohit.bhakkad]

@eugenis, I think we should remove changes in rL210335 and use char array in this test.

[eugenis]

Hm, r210335 looks weird. So, the way you suggest is to print the hex digits of each byte in the most significant first order, and print different bytes in the memory order. Sounds reasonable, let's do it.

[mohit.bhakkad]

I have removed endian based conditions for shadow printing in D17563.
This revision uses a char array to make this test compatible for both endians.

[eugenis]

LGTM together with D17563.

[mohit.bhakkad]

Commited in rL261982.

test/msan/msan_print_shadow3.cc

 // RUN: %clangxx_msan -O0 -g %s -o %t && %run %t >%t.out 2>&1
 // RUN: FileCheck %s < %t.out
 
 #include <stdint.h>
 #include <sanitizer/msan_interface.h>
 
 int main(void) {
   unsigned long long x = 0; // For 8-byte alignment.
-  uint32_t x_s = 0x12345678U;
+  char x_s[4] = {0x87, 0x65, 0x43, 0x21};
   __msan_partial_poison(&x, &x_s, sizeof(x_s));
   __msan_print_shadow(&x, sizeof(x_s));
   return 0;
 }
 
 // CHECK: Shadow map of [{{.*}}), 4 bytes:
 // CHECK: 0x{{.*}}: 87654321 ........ ........ ........