Running the encoder with akiyo, no specific flags needed otherwise, will cause the encoder to crash with a segfault. git bisect shows https://gitlab.com/AOMediaCodec/SVT-AV1/-/commit/ecd6bda2f4cfc84b994869beb054af8d5c58f550 as the culprit. ```c Thread 48 "svt-md7" received signal SIGSEGV, Segmentation fault. [Switching to Thread 0x7fffdd7d76c0 (LWP 399113)] 0x00005555558b48b5 in svt_aom_sub_pixel_variance16xh_ssse3.x_zero_y_zero_loop () at /home/ccom1/SVT-AV1/Source/Lib/ASM_SSE2/x86inc.asm:1352 1352 __instr %6, %7 (gdb) bt #0 0x00005555558b48b5 in svt_aom_sub_pixel_variance16xh_ssse3.x_zero_y_zero_loop () at /home/ccom1/SVT-AV1/Source/Lib/ASM_SSE2/x86inc.asm:1352 #1 0x00005555558b033e in svt_aom_sub_pixel_variance64x64_ssse3 ( src=0x7ffff6c65560 "#\037\037\037\037\035\037\034\027\035 \036 $)-''$$+28K[Z]_[bgdflrojinrtuxywx}\200\177ztka\\bn|\204\205\204\201~yttv}\204\205\203\201\177\177\202\204\205\205\202|sjit}~\177\177{ukZKKQPOPQRPOQRQOQRPQQOOOOPPPONMOMIIJF@@@=93/.-,--***++*,+,-,+,,--)*+.+", '\023' <repeats 41 times>..., src_stride=544, x_offset=0, y_offset=0, dst=0x7ffff5c688c8 "$! \036\036\036 \030\023\034 \036!#*-(& %.28M\\[]^^`eecmsqkiqttvwxwx}\177\177|tlaY_mz\203\205\204\203\200{vtt}\203\204\202\200\177\200\202\204\205\204\201|tjit}\177\177\177|tiZLKQQOQRQOPPPQOOQOQPOOPOPOPQNNNMJGIEA@?=81/.-++,,**++*+**,**,+,,+**,+", '\024' <repeats 41 times>..., dst_stride=496, sse_ptr=0x7fffdd7d6ac4) at /home/ccom1/SVT-AV1/Source/Lib/ASM_SSSE3/variance_impl_ssse3.c:72 #2 0x000055555570f003 in md_subpel_search_fixed_stage (ctx=0x5555620ec490, ref_pic=0x55555f9ad4e0, src=0x7ffff5c688c8 "$! \036\036\036 \030\023\034 \036!#*-(& %.28M\\[]^^`eecmsqkiqttvwxwx}\177\177|tlaY_mz\203\205\204\203\200{vtt}\203\204\202\200\177\200\202\204\205\204\201|tjit}\177\177\177|tiZLKQQOQRQOPPPQOOQOQPOOPOPOPQNNNMJGIEA@?=81/.-++,,**++*+**,**,+,,+**,+", '\024' <repeats 41 times>..., src_stride=496, me_mv=0x7fffdd7d6ba4) at /home/ccom1/SVT-AV1/Source/Lib/Codec/product_coding_loop.c:2724 #3 0x000055555570f4d4 in read_refine_me_mvs_light_pd1 (pcs=0x55555f4b7fd0, input_pic=0x5555600d0bf0, ctx=0x5555620ec490) at /home/ccom1/SVT-AV1/Source/Lib/Codec/product_coding_loop.c:2812 #4 0x0000555555729663 in md_encode_block_light_pd1 (pcs=0x55555f4b7fd0, ctx=0x5555620ec490, input_pic=0x5555600d0bf0) at /home/ccom1/SVT-AV1/Source/Lib/Codec/product_coding_loop.c:9090 #5 0x000055555572eebb in svt_aom_pick_partition_lpd1 (scs=0x55555e07e140, pcs=0x55555f4b7fd0, ctx=0x5555620ec490, mds=0x555562150220, pc_tree=0x555562151770, mi_row=16, mi_col=48) at /home/ccom1/SVT-AV1/Source/Lib/Codec/product_coding_loop.c:10644 #6 0x000055555560527b in svt_aom_mode_decision_kernel_iter (context=0x5555620ec3d0) at /home/ccom1/SVT-AV1/Source/Lib/Codec/enc_dec_process.c:3109 #7 0x0000555555606040 in svt_aom_mode_decision_kernel (input_ptr=0x55556093f550) at /home/ccom1/SVT-AV1/Source/Lib/Codec/enc_dec_process.c:3311 #8 0x0000555555771c23 in svt_thread_trampoline (p=0x555560964890) at /home/ccom1/SVT-AV1/Source/Lib/Codec/svt_threads.c:121 #9 0x00007ffff7c9caa4 in start_thread (arg=<optimized out>) at ./nptl/pthread_create.c:447 #10 0x00007ffff7d29c6c in clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:78 disassemble Dump of assembler code for function svt_aom_sub_pixel_variance16xh_ssse3.x_zero_y_zero_loop: 0x00005555558b48b1 <+0>: movdqu (%rdi),%xmm0 => 0x00005555558b48b5 <+4>: movdqa (%r8),%xmm1 0x00005555558b48ba <+9>: movdqa %xmm0,%xmm2 0x00005555558b48be <+13>: punpckhbw %xmm5,%xmm2 0x00005555558b48c2 <+17>: punpcklbw %xmm5,%xmm0 0x00005555558b48c6 <+21>: movdqa %xmm1,%xmm3 0x00005555558b48ca <+25>: punpckhbw %xmm5,%xmm3 0x00005555558b48ce <+29>: punpcklbw %xmm5,%xmm1 0x00005555558b48d2 <+33>: psubw %xmm3,%xmm2 0x00005555558b48d6 <+37>: psubw %xmm1,%xmm0 0x00005555558b48da <+41>: paddw %xmm2,%xmm6 0x00005555558b48de <+45>: pmaddwd %xmm2,%xmm2 0x00005555558b48e2 <+49>: paddw %xmm0,%xmm6 0x00005555558b48e6 <+53>: pmaddwd %xmm0,%xmm0 0x00005555558b48ea <+57>: paddd %xmm2,%xmm7 0x00005555558b48ee <+61>: paddd %xmm0,%xmm7 0x00005555558b48f2 <+65>: add %rsi,%rdi 0x00005555558b48f5 <+68>: add %r9,%r8 0x00005555558b48f8 <+71>: dec %eax 0x00005555558b48fa <+73>: jg 0x5555558b48b1 <svt_aom_sub_pixel_variance16xh_ssse3.x_zero_y_zero_loop> 0x00005555558b48fc <+75>: pcmpgtw %xmm6,%xmm5 0x00005555558b4900 <+79>: movhlps %xmm7,%xmm3 0x00005555558b4903 <+82>: movdqa %xmm6,%xmm4 0x00005555558b4907 <+86>: punpcklwd %xmm5,%xmm4 0x00005555558b490b <+90>: punpckhwd %xmm5,%xmm6 0x00005555558b490f <+94>: paddd %xmm3,%xmm7 0x00005555558b4913 <+98>: paddd %xmm4,%xmm6 0x00005555558b4917 <+102>: pshufd $0x1,%xmm7,%xmm3 0x00005555558b491c <+107>: movhlps %xmm6,%xmm4 0x00005555558b491f <+110>: paddd %xmm3,%xmm7 0x00005555558b4923 <+114>: paddd %xmm4,%xmm6 0x00005555558b4927 <+118>: mov 0x10(%rsp),%rsi 0x00005555558b492c <+123>: pshufd $0x1,%xmm6,%xmm4 0x00005555558b4931 <+128>: movd %xmm7,(%rsi) 0x00005555558b4935 <+132>: paddd %xmm4,%xmm6 0x00005555558b4939 <+136>: movd %xmm6,%eax 0x00005555558b493d <+140>: ret ``` The movdqu unaligned worked, but the aligned movdqa didn't, which is a bit suspicious. I did confirm that I could dereference `$r8`, and confirm that it shows `140737316817096` as `$r8`, which is not aligned to 16-bytes. Valgrind doesn't seem to show anything new. Originally reported via email.