I'm trying to track down an endianess issue when running on a PowerPC with Power8. Big endian is OK, little endian is having some troubles.
Below uint8x16_p8 is a typedef for __vector unsigned char. On a big endian machine I see:
1110 uint8x16_p8 r5 = (uint8x16_p8)VectorLoadKey(s_mask);
(gdb)
1112 for (unsigned int i=0; i<rounds-2; ++i)
(gdb) p r5
$1 = {0xd, 0xe, 0xf, 0xc, 0xd, 0xe, 0xf, 0xc, 0xd, 0xe, 0xf, 0xc, 0xd, 0xe,
0xf, 0xc}
On a little endian machine I see:
1110 uint8x16_p8 r5 = (uint8x16_p8)VectorLoadKey(s_mask);
(gdb)
1112 for (unsigned int i=0; i<rounds-2; ++i)
(gdb) p r5
$1 = {0xc, 0xf, 0xe, 0xd, 0xc, 0xf, 0xe, 0xd, 0xc, 0xf, 0xe, 0xd, 0xc, 0xf,
0xe, 0xd}
When gdb prints the value it is using the memory layout:
(gdb) ptype r5
type = unsigned char __attribute__ ((vector_size(16)))
(gdb)
I want to see the 128-bit integer value as it is loaded in a vsx register. The vsx register value is the important one, and it is always big endian. If there's a difference in the vsx value, then I know I need to permute a vector during the load from memory.
Also, GDB does not appear to support uint128_t:
(gdb) p *(uint128_t)r5
No symbol "uint128_t" in current context.
How do I have GDB print the vsx register value (and not the memory layout value)?
Another problem with GDB is, it cannot disassemble form "here", so I can't find where I am to print registers. For example, disass . does not "disassemble from here" (it results in a syntax error), and using $pc does not disassemble where I am (it looks like the start of the function):
(gdb) disass $pc
Dump of assembler code for function Rijndael_UncheckedSetKey_POWER8(...):
0x00000000104b82c8 <+0>: lis r2,4213
0x00000000104b82cc <+4>: addi r2,r2,-29952
0x00000000104b82d0 <+8>: mflr r0
0x00000000104b82d4 <+12>: std r0,16(r1)
0x00000000104b82d8 <+16>: std r31,-8(r1)
0x00000000104b82dc <+20>: stdu r1,-272(r1)
0x00000000104b82e0 <+24>: mr r31,r1
0x00000000104b82e4 <+28>: std r3,208(r31)
0x00000000104b82e8 <+32>: std r4,216(r31)
0x00000000104b82ec <+36>: std r5,224(r31)
0x00000000104b82f0 <+40>: std r6,232(r31)
0x00000000104b82f4 <+44>: mr r9,r7
0x00000000104b82f8 <+48>: stw r9,240(r31)
0x00000000104b82fc <+52>: ld r9,216(r31)
0x00000000104b8300 <+56>: cmpdi cr7,r9,16
0x00000000104b8304 <+60>: bne cr7,0x104b8548 <Rijndael_UncheckedSetKey_POWER8(...)+640>
0x00000000104b8308 <+64>: ld r9,208(r31)
0x00000000104b830c <+68>: std r9,32(r31)
---Type <return> to continue, or q <return> to quit---
...
Try __int128_t, which is supported by gcc & gdb:
#include <stdint.h>
int main(void)
{
__uint128_t s;
vector int v = { 0, 1, 2, 3 };
s = (__uint128_t)v;
(void)s;
return 0;
}
Then, you should be able to print your __uint128_t scalar as a normal value:
[jk@p8 ~]$ gdb -quiet ./test
Reading symbols from ./test...done.
(gdb) break test.c:12
Breakpoint 1 at 0x100005ec: file test.c, line 12.
(gdb) run
Starting program: /home/jk/test
Breakpoint 1, main () at test.c:12
12 return 0;
(gdb) print s
$1 = 0x00000003000000020000000100000000
(gdb)
Or, if you know which register is used, just print the register value directly, using $<reg> notation:
(gdb) p $vs0
$5 = {uint128 = 0x00000003000000020000000100000000, v2_double = {
2.1219957909652723e-314, 6.3659873738839482e-314}, v4_float = {0,
1.40129846e-45, 2.80259693e-45, 4.20389539e-45}, v4_int32 = {0, 1, 2, 3},
v8_int16 = {0, 0, 1, 0, 2, 0, 3, 0}, v16_int8 = {0, 0, 0, 0, 1, 0, 0, 0, 2,
0, 0, 0, 3, 0, 0, 0}}
However, keep in mind that the cast to a scalar may affect the register value; check the disassembly if you're unsure.