Interpreting GDB registers (SSE registers)

Question

I've been using GDB for 1 day and I've accumulated a decent understanding of it. However when I set a breakpoint at the final semicolon using GDB and print registers I can't fully interpret the meaning of the data stored into the XMM register.

I don't know if the data is in (MSB > LSB) format or vice versa.

    __m128i S = _mm_load_si128((__m128i*)Array16Bytes);

}

So this is the result that I'm getting.

(gdb) print $xmm0
$1 = {
  v4_float = {1.2593182e-07, -4.1251766e-18, -5.43431603e-31, -2.73406277e-14}, 
  v2_double = {4.6236050467459811e-58, -3.7422963639201271e-245}, 
  v16_int8 = {52, 7, 55, -32, -94, -104, 49, 49, -115, 48, 90, -120, -88, -10, 67, 50}, 
  v8_int16 = {13319, 14304, -23912, 12593, -29392, 23176, -22282, 17202}, 
  v4_int32 = {872888288, -1567084239, -1926210936, -1460255950}, 
  v2_int64 = {3749026652749312305, -8273012972482837710}, 
  uint128 = 0x340737e0a29831318d305a88a8f64332
}

So would someone kindly guide me how to interpret the data.

Answer 1

SSE (XMM) registers can be interpreted in various different ways. The register itself has no knowledge of the implicit data representation, it just holds 128 bits of data. An XMM register can represent:

4 x 32 bit floats        __m128
2 x 64 bit doubles       __m128d
16 x 8 bit ints          __m128i
8 x 16 bit ints          __m128i
4 x 32 bit ints          __m128i
2 x 64 bit ints          __m128i
128 individual bits      __m128i

So when gdb displays an XMM register it gives you all possible interpretations, as seen in your example above.

If you want to display a register using a specific interpretation (e.g. 16 x 8 bit ints) then you can do it like this:

(gdb) p $xmm0.v16_int8
$1 = {0, 0, 0, 0, 0, 0, 0, 0, -113, -32, 32, -50, 0, 0, 0, 2}

As for endianness, gdb displays the register contents in natural order, i.e. left-to-right, from MS to LS.

So if you have the following code:

#include <stdio.h>
#include <stdint.h>

#include <xmmintrin.h>
int main(int argc, char *argv[])
{
    int8_t buff[16] __attribute__ ((aligned(16))) = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 };

    __m128i v = _mm_load_si128((__m128i *)buff);

    printf("v = %vd\n", v);

    return 0;
}

If you compile and run this you will see:

v = 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

However if you step through the code in gdb and examine v you will see:

v16_int8 = {15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0}