Segfault in GpuArray_reshape_inplace()

[obilaniu]

A person in the lab had their compute job die on a signal while using the new Theano backend, libgpuarray, Git master branch tip as of today (SHA-1 23c8f004968f2d7703612f2ebb0e2da1b112bc7d). I recognized the signal code as SIGSEGV asked and was provided a truncated-at-3MB core file and libgpuarray.so.2.0 file.

After some reverse engineering, I found the problem to originate in GpuArray_reshape_inplace(), src/gpuarray_array.c:665

657  while (ni < nd && oi < a->nd) {
658    np = newdims[ni];
659    op = a->dimensions[oi];
660
661    while (np != op) {
662      if (np < op) {
663        np *= newdims[nj++];
664      } else {
665        op *= a->dimensions[oj++];
666      }
667    }
668
669    for (ok = oi; ok < oj - 1; ok++) {
670      if (ord == GA_F_ORDER) {

Through some more reverse engineering I've been able to establish the following:

• Segfaulting instruction: imul (%rcx,%r10,8),%rdi
• Effective Address of access that caused the segfault: 0x9f7dc000. This is the beginning of a new page, likely in heap.
  It was probably not mapped. Due to ASLR and the truncation of the core file I can't pursue this further.
• a->dimensions == 0x997cc9c0, in rcx.
• newdims == 0x7ffdc3e10310, in rbx
• oj == 12590793 in r8d. It had been incremented by the instruction preceding the load that segfaulted. The segfaulting access was made with a copy of oj in r10d, made when oj held the value 12590792.
• op == 0, in rdi.
• nj == 1, in esi
• np == 1, in rdx
• The value of esi, and thus nj, on entry to the while (np != op) loop was probably the same as that in r11d, which currently holds 1.
• The value of rdx, and thus np, on entry to the while (np != op) loop was probably the same as that in newdims[r13], and r13 held at the time of the crash the value 0. The register r13 is not accessed within the loop.
• r13d's last value was itself copied from ebp. ebp is also not accessed within the while (np != op) loop.
• The value of rdi, and thus np, on entry to the while (np != op) loop was probably the same as that in a->dimensions[r12], and r12 held at the time of the crash the value 0. The register r12 is not accessed within the while (np != op) loop.
• r12d's last value was itself copied from eax. eax is also not accessed within the while (np != op) loop.
• r9 probably holds the value a->strides. This points 80 bytes behind a->dimensions, so we might be able to compute an upper bound on the dimensions of a by taking into account internal GNU ptmalloc data structure sizes.
• r14 was probably set to 8x the value of r12 before entry to the loop. Since r12 is 0, so is r14.
• Within GpuArray_reshape_inplace(), r15 is used most commonly to hold malloc()'ed pointers. At the moment of the crash it held 0x99d65de0, which does look like a heap address.

Opinion

I think that somehow, op turned zero, whether by corrupt input, integer overflow or logic error. From that moment on the function was doomed and kept on multiplying all memory contents together until it hit the end of the heap and died.

The loop that segfaulted in GpuArray_reshape_inplace() does not defend itself against this possibility, and more generally GpuArray_reshape_inplace() is spaghetti code. It should be refactored.

Appendix

The contents of the registers at the moment of the crash were:

rax            0x0      0
rbx            0x7ffdc3e10310   140727889756944
rcx            0x997cc9c0       2575092160
rdx            0x1      1
rsi            0x1      1
rdi            0x0      0
rbp            0x0      0x0
rsp            0x7ffdc3e101b0   0x7ffdc3e101b0
r8             0xc01ec9 12590793
r9             0x997cc970       2575092080
r10            0xc01ec8 12590792
r11            0x1      1
r12            0x0      0
r13            0x0      0
r14            0x0      0
r15            0x99d65de0       2580962784
rip            0x7f6fbdb11187   0x7f6fbdb11187
eflags         0x10206  [ PF IF RF ]
cs             0x33     51
ss             0x2b     43
ds             0x0      0
es             0x0      0
fs             0x0      0
gs             0x0      0
st0            0        (raw 0x00000000000000000000)
st1            0        (raw 0x00000000000000000000)
st2            0        (raw 0x00000000000000000000)
st3            0        (raw 0x00000000000000000000)
st4            0        (raw 0x00000000000000000000)
st5            0.010000000000000000208166817117216851   (raw 0x3ff8a3d70a3d70a3d800)
st6            0.010000000000000000208166817117216851   (raw 0x3ff8a3d70a3d70a3d800)
st7            0.010000000000000000208166817117216851   (raw 0x3ff8a3d70a3d70a3d800)
fctrl          0x37f    895
fstat          0x0      0
ftag           0xffff   65535
fiseg          0x7f6f   32623
fioff          0xd02674d8       -802786088
foseg          0x7ffd   32765
fooff          0xc3e4e388       -1008409720
fop            0x0      0
mxcsr          0x1fa2   [ DE PE IM DM ZM OM UM PM ]
ymm0           {v8_float = {0x4, 0x0, 0x0, 0x2, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x2, 0x0, 0x0}, v32_int8 = {0x0, 0x0, 0x80, 0x40, 0x0, 0x0, 0x0, 0x8, 0x80, 0x7, 0x0, 0x5b, 0x0, 0x20, 0x0, 0x40, 0x0 <repeats 16 times>}, v16_int16 = {0x0, 0x4080, 0x0, 0x800, 0x780, 0x5b00, 0x2000, 0x4000, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0x40800000, 0x8000000, 0x5b000780, 0x40002000, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x800000040800000, 0x400020005b000780, 0x0, 0x0}, v2_int128 = {0x400020005b0007800800000040800000, 0x00000000000000000000000000000000}}
ymm1           {v8_float = {0x0, 0x2, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x4, 0x0, 0x0, 0x0}, v32_int8 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x10, 0x40, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xc, 0x0 <repeats 17 times>}, v16_int16 = {0x0, 0x0, 0x0, 0x4010, 0x0, 0x0, 0x0, 0xc, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0x0, 0x40100000, 0x0, 0xc0000, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x4010000000000000, 0xc000000000000, 0x0, 0x0}, v2_int128 = {0x000c0000000000004010000000000000, 0x00000000000000000000000000000000}}
ymm2           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}}
ymm3           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x13, 0x0, 0x80, 0x4, 0x0 <repeats 28 times>}, v16_int16 = {0x13, 0x480, 0x0 <repeats 14 times>}, v8_int32 = {0x4800013, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x4800013, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000004800013, 0x00000000000000000000000000000000}}
ymm4           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x90, 0x2, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x30, 0x0 <repeats 23 times>}, v16_int16 = {0x290, 0x0, 0x0, 0x0, 0x30, 0x0 <repeats 11 times>}, v8_int32 = {0x290, 0x0, 0x30, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x290, 0x30, 0x0, 0x0}, v2_int128 = {0x00000000000000300000000000000290, 0x00000000000000000000000000000000}}
ymm5           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0, 0x0, 0x5c, 0xb, 0x13, 0x0, 0x0, 0x0, 0x40, 0xe, 0x4c, 0xb, 0x13, 0x0 <repeats 19 times>}, v16_int16 = {0x0, 0xb5c, 0x13, 0x0, 0xe40, 0xb4c, 0x13, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0xb5c0000, 0x13, 0xb4c0e40, 0x13, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x130b5c0000, 0x130b4c0e40, 0x0, 0x0}, v2_int128 = {0x000000130b4c0e40000000130b5c0000, 0x00000000000000000000000000000000}}
ymm6           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}}
ymm7           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}}
ymm8           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}}
ymm9           {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}}
ymm10          {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}}
ymm11          {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}}
ymm12          {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0, 0x0, 0x0, 0x0, 0x0, 0xff, 0x0, 0x0, 0x0, 0xff, 0x0 <repeats 22 times>}, v16_int16 = {0x0, 0x0, 0xff00, 0x0, 0xff00, 0x0 <repeats 11 times>}, v8_int32 = {0x0, 0xff00, 0xff00, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xff0000000000, 0xff00, 0x0, 0x0}, v2_int128 = {0x000000000000ff000000ff0000000000, 0x00000000000000000000000000000000}}
ymm13          {v8_float = {0x0, 0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0xfb, 0xb4, 0x35, 0x3b, 0x86, 0x10, 0xae, 0xbf, 0x0 <repeats 24 times>}, v16_int16 = {0xb4fb, 0x3b35, 0x1086, 0xbfae, 0x0 <repeats 12 times>}, v8_int32 = {0x3b35b4fb, 0xbfae1086, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xbfae10863b35b4fb, 0x0, 0x0, 0x0}, v2_int128 = {0x0000000000000000bfae10863b35b4fb, 0x00000000000000000000000000000000}}
ymm14          {v8_float = {0x0, 0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0xfb, 0xb4, 0x35, 0x3b, 0x86, 0x10, 0xae, 0xbf, 0x0 <repeats 24 times>}, v16_int16 = {0xb4fb, 0x3b35, 0x1086, 0xbfae, 0x0 <repeats 12 times>}, v8_int32 = {0x3b35b4fb, 0xbfae1086, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xbfae10863b35b4fb, 0x0, 0x0, 0x0}, v2_int128 = {0x0000000000000000bfae10863b35b4fb, 0x00000000000000000000000000000000}}
ymm15          {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x54, 0xdc, 0x1c, 0x1a, 0xe1, 0x6b, 0x5f, 0xbc, 0x0 <repeats 24 times>}, v16_int16 = {0xdc54, 0x1a1c, 0x6be1, 0xbc5f, 0x0 <repeats 12 times>}, v8_int32 = {0x1a1cdc54, 0xbc5f6be1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xbc5f6be11a1cdc54, 0x0, 0x0, 0x0}, v2_int128 = {0x0000000000000000bc5f6be11a1cdc54, 0x00000000000000000000000000000000}}

[nouiz]

Are you able to reproduce this or it was a once time error?

…

On Sat, Mar 11, 2017 at 4:35 PM Olexa Bilaniuk ***@***.***> wrote: A person in the lab had their compute job die on a signal while using the new Theano backend, libgpuarray, Git master branch tip as of today (SHA-1 23c8f00). I recognized the signal code as SIGSEGV asked and was provided a truncated-at-3MB core file and libgpuarray.so.2.0 file. After some reverse engineering, I found the problem to originate in GpuArray_reshape_inplace(), src/gpuarray_array.c:665 <https://github.com/Theano/libgpuarray/blob/23c8f004968f2d7703612f2ebb0e2da1b112bc7d/src/gpuarray_array.c#L665> 657 while (ni < nd && oi < a->nd) { 658 np = newdims[ni]; 659 op = a->dimensions[oi]; 660 661 while (np != op) { 662 if (np < op) { 663 np *= newdims[nj++]; 664 } else { 665 op *= a->dimensions[oj++]; 666 } 667 } 668 669 for (ok = oi; ok < oj - 1; ok++) { 670 if (ord == GA_F_ORDER) { Through some more reverse engineering I've been able to establish the following: - Segfaulting instruction: imul (%rcx,%r10,8),%rdi - Effective Address of access that caused the segfault: 0x9f7dc000. This is the beginning of a new page, likely in heap. It was probably not mapped. Due to ASLR and the truncation of the core file I can't pursue this further. - a->dimensions == 0x997cc9c0, in rcx. - newdims == 0x7ffdc3e10310, in rbx - oj == 12590793 in r8d. It had been incremented by the instruction preceding the load that segfaulted. The segfaulting access was made with a copy of oj in r10d, made when oj held the value 12590792. - op == 0, in rdi. - nj == 1, in esi - np == 1, in rdx - The value of esi, and thus nj, on entry to the while (np != op) loop was probably the same as that in r11d, which currently holds 1. - The value of rdx, and thus np, on entry to the while (np != op) loop was probably the same as that in newdims[r13], and r13 held at the time of the crash the value 0. The register r13 is not accessed within the loop. - r13d's last value was itself copied from ebp. ebp is also not accessed within the while (np != op) loop. - The value of rdi, and thus np, on entry to the while (np != op) loop was probably the same as that in a->dimensions[r12], and r12 held at the time of the crash the value 0. The register r12 is not accessed within the while (np != op) loop. - r12d's last value was itself copied from eax. eax is also not accessed within the while (np != op) loop. - r9 probably holds the value a->strides. This points 80 bytes behind a->dimensions, so we might be able to compute an upper bound on the dimensions of a by taking into account internal GNU ptmalloc data structure sizes. - r14 was probably set to 8x the value of r12 before entry to the loop. Since r12 is 0, so is r14. - Within GpuArray_reshape_inplace(), r15 is used most commonly to hold malloc()'ed pointers. At the moment of the crash it held 0x99d65de0, which does look like a heap address. Opinion I think that somehow, op turned zero, whether by corrupt input or integer overflow. From that moment on the function was doomed and kept on multiplying all memory contents together until it hit the end of the heap and died. The loop that segfaulted in GpuArray_reshape_inplace() does not defend itself against this possibility, and more generally GpuArray_reshape_inplace() is spaghetti code. It should be refactored. Appendix The contents of the registers at the moment of the crash were: rax 0x0 0 rbx 0x7ffdc3e10310 140727889756944 rcx 0x997cc9c0 2575092160 rdx 0x1 1 rsi 0x1 1 rdi 0x0 0 rbp 0x0 0x0 rsp 0x7ffdc3e101b0 0x7ffdc3e101b0 r8 0xc01ec9 12590793 r9 0x997cc970 2575092080 r10 0xc01ec8 12590792 r11 0x1 1 r12 0x0 0 r13 0x0 0 r14 0x0 0 r15 0x99d65de0 2580962784 rip 0x7f6fbdb11187 0x7f6fbdb11187 eflags 0x10206 [ PF IF RF ] cs 0x33 51 ss 0x2b 43 ds 0x0 0 es 0x0 0 fs 0x0 0 gs 0x0 0 st0 0 (raw 0x00000000000000000000) st1 0 (raw 0x00000000000000000000) st2 0 (raw 0x00000000000000000000) st3 0 (raw 0x00000000000000000000) st4 0 (raw 0x00000000000000000000) st5 0.010000000000000000208166817117216851 (raw 0x3ff8a3d70a3d70a3d800) st6 0.010000000000000000208166817117216851 (raw 0x3ff8a3d70a3d70a3d800) st7 0.010000000000000000208166817117216851 (raw 0x3ff8a3d70a3d70a3d800) fctrl 0x37f 895 fstat 0x0 0 ftag 0xffff 65535 fiseg 0x7f6f 32623 fioff 0xd02674d8 -802786088 foseg 0x7ffd 32765 fooff 0xc3e4e388 -1008409720 fop 0x0 0 mxcsr 0x1fa2 [ DE PE IM DM ZM OM UM PM ] ymm0 {v8_float = {0x4, 0x0, 0x0, 0x2, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x2, 0x0, 0x0}, v32_int8 = {0x0, 0x0, 0x80, 0x40, 0x0, 0x0, 0x0, 0x8, 0x80, 0x7, 0x0, 0x5b, 0x0, 0x20, 0x0, 0x40, 0x0 <repeats 16 times>}, v16_int16 = {0x0, 0x4080, 0x0, 0x800, 0x780, 0x5b00, 0x2000, 0x4000, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0x40800000, 0x8000000, 0x5b000780, 0x40002000, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x800000040800000, 0x400020005b000780, 0x0, 0x0}, v2_int128 = {0x400020005b0007800800000040800000, 0x00000000000000000000000000000000}} ymm1 {v8_float = {0x0, 0x2, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x4, 0x0, 0x0, 0x0}, v32_int8 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x10, 0x40, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xc, 0x0 <repeats 17 times>}, v16_int16 = {0x0, 0x0, 0x0, 0x4010, 0x0, 0x0, 0x0, 0xc, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0x0, 0x40100000, 0x0, 0xc0000, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x4010000000000000, 0xc000000000000, 0x0, 0x0}, v2_int128 = {0x000c0000000000004010000000000000, 0x00000000000000000000000000000000}} ymm2 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}} ymm3 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x13, 0x0, 0x80, 0x4, 0x0 <repeats 28 times>}, v16_int16 = {0x13, 0x480, 0x0 <repeats 14 times>}, v8_int32 = {0x4800013, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x4800013, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000004800013, 0x00000000000000000000000000000000}} ymm4 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x90, 0x2, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x30, 0x0 <repeats 23 times>}, v16_int16 = {0x290, 0x0, 0x0, 0x0, 0x30, 0x0 <repeats 11 times>}, v8_int32 = {0x290, 0x0, 0x30, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x290, 0x30, 0x0, 0x0}, v2_int128 = {0x00000000000000300000000000000290, 0x00000000000000000000000000000000}} ymm5 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0, 0x0, 0x5c, 0xb, 0x13, 0x0, 0x0, 0x0, 0x40, 0xe, 0x4c, 0xb, 0x13, 0x0 <repeats 19 times>}, v16_int16 = {0x0, 0xb5c, 0x13, 0x0, 0xe40, 0xb4c, 0x13, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v8_int32 = {0xb5c0000, 0x13, 0xb4c0e40, 0x13, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x130b5c0000, 0x130b4c0e40, 0x0, 0x0}, v2_int128 = {0x000000130b4c0e40000000130b5c0000, 0x00000000000000000000000000000000}} ymm6 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}} ymm7 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}} ymm8 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}} ymm9 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}} ymm10 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}} ymm11 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0 <repeats 32 times>}, v16_int16 = {0x0 <repeats 16 times>}, v8_int32 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0x0, 0x0, 0x0, 0x0}, v2_int128 = {0x00000000000000000000000000000000, 0x00000000000000000000000000000000}} ymm12 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x0, 0x0, 0x0, 0x0, 0x0, 0xff, 0x0, 0x0, 0x0, 0xff, 0x0 <repeats 22 times>}, v16_int16 = {0x0, 0x0, 0xff00, 0x0, 0xff00, 0x0 <repeats 11 times>}, v8_int32 = {0x0, 0xff00, 0xff00, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xff0000000000, 0xff00, 0x0, 0x0}, v2_int128 = {0x000000000000ff000000ff0000000000, 0x00000000000000000000000000000000}} ymm13 {v8_float = {0x0, 0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0xfb, 0xb4, 0x35, 0x3b, 0x86, 0x10, 0xae, 0xbf, 0x0 <repeats 24 times>}, v16_int16 = {0xb4fb, 0x3b35, 0x1086, 0xbfae, 0x0 <repeats 12 times>}, v8_int32 = {0x3b35b4fb, 0xbfae1086, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xbfae10863b35b4fb, 0x0, 0x0, 0x0}, v2_int128 = {0x0000000000000000bfae10863b35b4fb, 0x00000000000000000000000000000000}} ymm14 {v8_float = {0x0, 0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0xfb, 0xb4, 0x35, 0x3b, 0x86, 0x10, 0xae, 0xbf, 0x0 <repeats 24 times>}, v16_int16 = {0xb4fb, 0x3b35, 0x1086, 0xbfae, 0x0 <repeats 12 times>}, v8_int32 = {0x3b35b4fb, 0xbfae1086, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xbfae10863b35b4fb, 0x0, 0x0, 0x0}, v2_int128 = {0x0000000000000000bfae10863b35b4fb, 0x00000000000000000000000000000000}} ymm15 {v8_float = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_double = {0x0, 0x0, 0x0, 0x0}, v32_int8 = {0x54, 0xdc, 0x1c, 0x1a, 0xe1, 0x6b, 0x5f, 0xbc, 0x0 <repeats 24 times>}, v16_int16 = {0xdc54, 0x1a1c, 0x6be1, 0xbc5f, 0x0 <repeats 12 times>}, v8_int32 = {0x1a1cdc54, 0xbc5f6be1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int64 = {0xbc5f6be11a1cdc54, 0x0, 0x0, 0x0}, v2_int128 = {0x0000000000000000bc5f6be11a1cdc54, 0x00000000000000000000000000000000}} — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#374>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AALC-5O9BDV3ju0BipdIYbhQCJUEWS66ks5rkxO7gaJpZM4MaV9Q> .