[mingw-w64-clang-i686-gtk4] Segmentation fault in libgraphene-1.0-0.dll

[aidenfoxx]

Description / Steps to reproduce the issue

The "mingw-w64-clang-i686-gtk4" current causes a segmentation fault when used. This can be seen by executing the "gtk4-demo.exe" included with the package.

Thread 1 received signal SIGSEGV, Segmentation fault.
0x58e1429c in graphene_matrix_init_from_matrix () from C:\Development\MSYS2\clang32\bin\libgraphene-1.0-0.dll
(gdb) bt 10
#0  0x58e1429c in graphene_matrix_init_from_matrix () from C:\Development\MSYS2\clang32\bin\libgraphene-1.0-0.dll
#1  0x59561b68 in gtk_snapshot_push_color_matrix () from C:\Development\MSYS2\clang32\bin\libgtk-4-1.dll
#2  0x0e8db450 in ?? ()
#3  0x00000000 in ?? ()

Expected behavior

The library can be used without causing an immediate crash.

Actual behavior

An immediate segmentation fault.

Verification

• [X] I have verified that my MSYS2 is up-to-date before submitting the report (see https://www.msys2.org/docs/updating/)

Windows Version

MSYS_NT-10.0-19044

MINGW environments affected

• [ ] MINGW64
• [ ] MINGW32
• [ ] UCRT64
• [ ] CLANG64
• [X] CLANG32
• [ ] CLANGARM64

Are you willing to submit a PR?

No response

[Biswa96]

Thank you for reporting the issue. I can reproduce the issue in clang32 but not in ucrt64. Unfortunately, I am limited with my knowledge because gdb leads to the SIMD intrinsic used in graphene. Please wait for others to comment.

[lb90]

Here's a stacktrace:

(lldb) target create "./gtk4-demo.exe"
(rrent executable set to 'C:\Users\roberta\gtk4-clang32\gtk4-demo.exe' (i386).
(lldb) r
(lldb) Process 11004 launched: 'C:\Users\roberta\gtk4-clang32\gtk4-demo.exe' (i386)
Process 11004 stopped
* thread #1, stop reason = Exception 0xc0000005 encountered at address 0x667b429c: Access violation reading location 0xffffffff
    frame #0: 0x667b429c libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix + 12
libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix:
->  0x667b429c <+12>: movaps %xmm0, 0x30(%eax)
    0x667b42a0 <+16>: movaps 0x20(%ecx), %xmm0
    0x667b42a4 <+20>: movaps %xmm0, 0x20(%eax)
    0x667b42a8 <+24>: movaps (%ecx), %xmm0
(lldb) bt
* thread #1, stop reason = Exception 0xc0000005 encountered at address 0x667b429c: Access violation reading location 0xffffffff
  * frame #0: 0x667b429c libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix + 12
    frame #1: 0x674210ec libgtk-4-1.dll`gsk_color_matrix_node_new(child=0x085eafb0, color_matrix=0x0e3ee180, color_offset=0x0e3ee1c0) at gskrendernodeimpl.c:3892:3
    frame #2: 0x67050471 libgtk-4-1.dll`gtk_snapshot_collect_color_matrix(snapshot=0x0e77e8c8, state=0x0e3ee160, nodes=0x0e3b79e8, n_nodes=1) at gtksnapshot.c:644:16
    frame #3: 0x67052ab1 libgtk-4-1.dll`gtk_snapshot_pop_one(snapshot=0x0e77e8c8) at gtksnapshot.c:1691:14
    frame #4: 0x670525a2 libgtk-4-1.dll`gtk_snapshot_pop_internal(snapshot=0x0e77e8c8, is_texture_pop=0) at gtksnapshot.c:1776:10
    frame #5: 0x670516ae libgtk-4-1.dll`gtk_snapshot_pop(snapshot=0x0e77e8c8) at gtksnapshot.c:1902:10
    frame #6: 0x66f51ab7 libgtk-4-1.dll`gtk_icon_paintable_snapshot_symbolic(paintable=0x0e66d570, snapshot=0x0e77e8c8, width=16, height=16, colors=0x085948a0, n_colors=4) at gtkicontheme.c:3938:5
    frame #7: 0x67074857 libgtk-4-1.dll`gtk_symbolic_paintable_snapshot_symbolic(paintable=0x0e66d570, snapshot=0x0e77e8c8, width=16, height=16, colors=0x085948a0, n_colors=4) at gtksymbolicpaintable.c:101:7
    frame #8: 0x671886db libgtk-4-1.dll`gtk_css_image_icon_theme_snapshot(image=0x08594890, snapshot=0x0e77e8c8, width=16, height=36) at gtkcssimageicontheme.c:92:3
    frame #9: 0x67181a78 libgtk-4-1.dll`gtk_css_image_snapshot(image=0x08594890, snapshot=0x0e77e8c8, width=16, height=36) at gtkcssimage.c:290:3
    frame #10: 0x670076aa libgtk-4-1.dll`gtk_css_style_snapshot_icon(style=0x08594830, snapshot=0x0e77e8c8, width=16, height=36) at gtkrendericon.c:66:7
    frame #11: 0x6714db3a libgtk-4-1.dll`gtk_builtin_icon_snapshot(widget=0x033b2d40, snapshot=0x0e77e8c8) at gtkbuiltinicon.c:53:5
    frame #12: 0x6710ccb6 libgtk-4-1.dll`gtk_widget_create_render_node(widget=0x033b2d40, snapshot=0x0e77e8c8) at gtkwidget.c:11864:7
    frame #13: 0x67102202 libgtk-4-1.dll`gtk_widget_do_snapshot(widget=0x033b2d40, snapshot=0x0e77e8c8) at gtkwidget.c:11899:17
    frame #14: 0x67103148 libgtk-4-1.dll`gtk_widget_snapshot_child(widget=0x0623bcc0, child=0x033b2d40, snapshot=0x0e77e8c8) at gtkwidget.c:12320:3
    frame #15: 0x671071fd libgtk-4-1.dll`gtk_widget_real_snapshot(widget=0x0623bcc0, snapshot=0x0e77e8c8) at gtkwidget.c:762:5
    frame #16: 0x6710ccb6 libgtk-4-1.dll`gtk_widget_create_render_node(widget=0x0623bcc0, snapshot=0x0e77e8c8) at gtkwidget.c:11864:7
    frame #17: 0x67102202 libgtk-4-1.dll`gtk_widget_do_snapshot(widget=0x0623bcc0, snapshot=0x0e77e8c8) at gtkwidget.c:11899:17
    frame #18: 0x67103148 libgtk-4-1.dll`gtk_widget_snapshot_child(widget=0x062f15b0, child=0x0623bcc0, snapshot=0x0e77e8c8) at gtkwidget.c:12320:3
    frame #19: 0x671071fd libgtk-4-1.dll`gtk_widget_real_snapshot(widget=0x062f15b0, snapshot=0x0e77e8c8) at gtkwidget.c:762:5
    frame #20: 0x6710ccb6 libgtk-4-1.dll`gtk_widget_create_render_node(widget=0x062f15b0, snapshot=0x0e77e8c8) at gtkwidget.c:11864:7
    frame #21: 0x67102202 libgtk-4-1.dll`gtk_widget_do_snapshot(widget=0x062f15b0, snapshot=0x0e77e8c8) at gtkwidget.c:11899:17
    frame #22: 0x67103148 libgtk-4-1.dll`gtk_widget_snapshot_child(widget=0x06ae4d88, child=0x062f15b0, snapshot=0x0e77e8c8) at gtkwidget.c:12320:3
    frame #23: 0x671071fd libgtk-4-1.dll`gtk_widget_real_snapshot(widget=0x06ae4d88, snapshot=0x0e77e8c8) at gtkwidget.c:762:5
    frame #24: 0x6710cc8a libgtk-4-1.dll`gtk_widget_create_render_node(widget=0x06ae4d88, snapshot=0x0e77e8c8) at gtkwidget.c:11859:7
    frame #25: 0x67102202 libgtk-4-1.dll`gtk_widget_do_snapshot(widget=0x06ae4d88, snapshot=0x0e77e8c8) at gtkwidget.c:11899:17
    frame #26: 0x67103148 libgtk-4-1.dll`gtk_widget_snapshot_child(widget=0x06ac5d50, child=0x06ae4d88, snapshot=0x0e77e8c8) at gtkwidget.c:12320:3
    frame #27: 0x671071fd libgtk-4-1.dll`gtk_widget_real_snapshot(widget=0x06ac5d50, snapshot=0x0e77e8c8) at gtkwidget.c:762:5
    frame #28: 0x67020304 libgtk-4-1.dll`gtk_scrolled_window_snapshot(widget=0x06ac5d50, snapshot=0x0e77e8c8) at gtkscrolledwindow.c:2832:3
    frame #29: 0x6710ccb6 libgtk-4-1.dll`gtk_widget_create_render_node(widget=0x06ac5d50, snapshot=0x0e77e8c8) at gtkwidget.c:11864:7
    frame #30: 0x67102202 libgtk-4-1.dll`gtk_widget_do_snapshot(widget=0x06ac5d50, snapshot=0x0e77e8c8) at gtkwidget.c:11899:17
    frame #31: 0x67103148 libgtk-4-1.dll`gtk_widget_snapshot_child(widget=0x06a80a98, child=0x06ac5d50, snapshot=0x0e77e8c8) at gtkwidget.c:12320:3
    frame #32: 0x671071fd libgtk-4-1.dll`gtk_widget_real_snapshot(widget=0x06a80a98, snapshot=0x0e77e8c8) at gtkwidget.c:762:5
    frame #33: 0x6710ccb6 libgtk-4-1.dll`gtk_widget_create_render_node(widget=0x06a80a98, snapshot=0x0e77e8c8) at gtkwidget.c:11864:7
    frame #34: 0x67102202 libgtk-4-1.dll`gtk_widget_do_snapshot(widget=0x06a80a98, snapshot=0x0e77e8c8) at gtkwidget.c:11899:17
    frame #35: 0x67103148 libgtk-4-1.dll`gtk_widget_snapshot_child(widget=0x06a80540, child=0x06a80a98, snapshot=0x0e77e8c8) at gtkwidget.c:12320:3
    frame #36: 0x671071fd libgtk-4-1.dll`gtk_widget_real_snapshot(widget=0x06a80540, snapshot=0x0e77e8c8) at gtkwidget.c:762:5
    frame #37: 0x6710ccb6 libgtk-4-1.dll`gtk_widget_create_render_node(widget=0x06a80540, snapshot=0x0e77e8c8) at gtkwidget.c:11864:7
    frame #38: 0x67102202 libgtk-4-1.dll`gtk_widget_do_snapshot(widget=0x06a80540, snapshot=0x0e77e8c8) at gtkwidget.c:11899:17
    frame #39: 0x67103148 libgtk-4-1.dll`gtk_widget_snapshot_child(widget=0x0691b430, child=0x06a80540, snapshot=0x0e77e8c8) at gtkwidget.c:12320:3
    frame #40: 0x671071fd libgtk-4-1.dll`gtk_widget_real_snapshot(widget=0x0691b430, snapshot=0x0e77e8c8) at gtkwidget.c:762:5
    frame #41: 0x6710cc8a libgtk-4-1.dll`gtk_widget_create_render_node(widget=0x0691b430, snapshot=0x0e77e8c8) at gtkwidget.c:11859:7
    frame #42: 0x67102202 libgtk-4-1.dll`gtk_widget_do_snapshot(widget=0x0691b430, snapshot=0x0e77e8c8) at gtkwidget.c:11899:17
    frame #43: 0x671020a5 libgtk-4-1.dll`gtk_widget_snapshot(widget=0x0691b430, snapshot=0x0e77e8c8) at gtkwidget.c:11921:3
    frame #44: 0x6710237a libgtk-4-1.dll`gtk_widget_render(widget=0x0691b430, surface=0x085efd20, region=0x0625cb48) at gtkwidget.c:11953:3
    frame #45: 0x6711d108 libgtk-4-1.dll`surface_render(surface=0x085efd20, region=0x0625cb48, widget=0x0691b430) at gtkwindow.c:4745:3
    frame #46: 0x673621de libgtk-4-1.dll`_gdk_marshal_BOOLEAN__BOXEDv(closure=0x0e7b4a70, return_value=0x009ef780, instance=0x085efd20, args="H\xcb%\U00000006(\xf8\x9e", marshal_data=0x00000000, n_params=1, param_types=0x085e8900) at gdkmarshalers.c:130:14
    frame #47: 0x66c47969 libgobject-2.0-0.dll`__g_closure_invoke_va + 249
    frame #48: 0x66c5ae20 libgobject-2.0-0.dll`g_signal_emit_valist + 1120
    frame #49: 0x66c5b580 libgobject-2.0-0.dll`g_signal_emit + 64
    frame #50: 0x67400281 libgtk-4-1.dll`gdk_surface_paint_on_clock(clock=0x085ed600, data=0x085efd20) at gdksurface.c:1393:7
    frame #51: 0x66c4960c libgobject-2.0-0.dll`g_cclosure_marshal_VOID__VOIDv + 44
    frame #52: 0x66c47969 libgobject-2.0-0.dll`__g_closure_invoke_va + 249
    frame #53: 0x66c5b2ab libgobject-2.0-0.dll`g_signal_emit_valist + 2283
    frame #54: 0x66c5b580 libgobject-2.0-0.dll`g_signal_emit + 64
    frame #55: 0x673d924b libgtk-4-1.dll`_gdk_frame_clock_emit_paint(frame_clock=0x085ed600) at gdkframeclock.c:710:3
    frame #56: 0x673db22c libgtk-4-1.dll`gdk_frame_clock_paint_idle(data=0x085ed600) at gdkframeclockidle.c:634:19
    frame #57: 0x673daabf libgtk-4-1.dll`gdk_frame_clock_flush_idle(data=0x085ed600) at gdkframeclockidle.c:400:3
    frame #58: 0x66cc4180 libglib-2.0-0.dll`g_clear_list + 144
(lldb) f 1
frame #1: 0x674210ec libgtk-4-1.dll`gsk_color_matrix_node_new(child=0x085eafb0, color_matrix=0x0e3ee180, color_offset=0x0e3ee1c0) at gskrendernodeimpl.c:3892:3
   3889   node->offscreen_for_opacity = child->offscreen_for_opacity;
   3890
   3891   self->child = gsk_render_node_ref (child);
-> 3892   graphene_matrix_init_from_matrix (&self->color_matrix, color_matrix);
   3893   graphene_vec4_init_from_vec4 (&self->color_offset, color_offset);
   3894
   3895   gsk_rect_init_from_rect (&node->bounds, &child->bounds);
(lldb) p *self
(GskColorMatrixNode) {
  render_node = {
    parent_instance = {
      g_class = 0x0334cad8
    }
    ref_count = 1
    bounds = {
      origin = (x = 0, y = 0)
      size = (width = 0, height = 0)
    }
    preferred_depth = 0
    offscreen_for_opacity = 0
  }
  child = 0x085eafb0
  color_matrix = {
    __graphene_private_value = {
      x = (x = 0, y = 0, z = 0, w = 0)
      y = (x = 0, y = 0, z = 0, w = 0)
      z = (x = 0, y = 0, z = 0, w = 0)
      w = (x = 0, y = 0, z = 0, w = 0)
    }
  }
  color_offset = {
    __graphene_private_value = (x = 0, y = 0, z = 0, w = 0)
  }
}
(lldb) p color_matrix
(const graphene_matrix_t *) 0x0e3ee180
(lldb) p *color_matrix
(const graphene_matrix_t) {
  __graphene_private_value = {
    x = (x = 0.200000003, y = 0.819607853, z = 0.478431374, w = 0)
    y = (x = 0.960784316, y = 0.474509805, z = 0, w = 0)
    z = (x = 0.800000011, y = 0, z = 0, w = 0)
    w = (x = 0, y = 0, z = 0, w = 1)
  }
}
(lldb) f 0
frame #0: 0x667b429c libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix + 12
libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix:
->  0x667b429c <+12>: movaps %xmm0, 0x30(%eax)
    0x667b42a0 <+16>: movaps 0x20(%ecx), %xmm0
    0x667b42a4 <+20>: movaps %xmm0, 0x20(%eax)
    0x667b42a8 <+24>: movaps (%ecx), %xmm0
(lldb) registers read
error: 'registers' is not a valid command.
(lldb) register read
General Purpose Registers:
       eax = 0x0E66D688
       ebx = 0x009EE300
       ecx = 0x0E3EE180
       edx = 0x00000000
       edi = 0x0E77E8C8
       esi = 0x0E77E8C8
       ebp = 0x009EE1E0
       esp = 0x009EE1B4
       eip = 0x667B429C  libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix + 12
    eflags = 0b00000000000000010000001000000110

(lldb)

[lb90]

Doesn't happen if libgraphene is built with --buildtype=debug 🤷‍♂️

[lb90]

And here's with debugoptimized builds:

$ lldb ./gtk4-demo.exe
(lldb) target create "./gtk4-demo.exe"
(rrent executable set to 'C:\Users\roberta\gtk4-clang32\gtk4-demo.exe' (i386).
(lldb) r
(lldb) Process 13304 launched: 'C:\Users\roberta\gtk4-clang32\gtk4-demo.exe' (i386)
Process 13304 stopped
* thread #1, stop reason = Exception 0xc0000005 encountered at address 0x667542cc: Access violation reading location 0xffffffff
    frame #0: 0x667542cc libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix(m=0x0deea508, src=0x009ee310) at graphene-matrix.c:248:19
   245  graphene_matrix_init_from_matrix (graphene_matrix_t       *m,
   246                                    const graphene_matrix_t *src)
   247  {
-> 248    m->value = src->value;
   249
   250    return m;
   251  }
(lldb) p m
(graphene_matrix_t *) 0x0deea508
(lldb) p src
(const graphene_matrix_t *) 0x009ee310
(lldb) p *src
(const graphene_matrix_t) {
  value = {
    x = (x = 0.200000003, y = 0.819607853, z = 0.478431374, w = 0)
    y = (x = 0.960784316, y = 0.474509805, z = 0, w = 0)
    z = (x = 0.800000011, y = 0, z = 0, w = 0)
    w = (x = 0, y = 0, z = 0, w = 1)
  }
}
(lldb) p *m
(graphene_matrix_t) {
  value = {
    x = (x = -0.00132703932, y = -0.00132703932, z = -0.00132703932, w = -0.00132703932)
    y = (x = -0.00132703932, y = -0.00132703932, z = -0.00132703932, w = -0.00132703932)
    z = (x = -0.00132703932, y = -0.00132703932, z = -0.00132703932, w = -0.00132703932)
    w = (x = -0.00132703932, y = -0.00132703932, z = -0.00132703932, w = -0.00132703932)
  }
}
(lldb)

I can't understand where 0xffffffff comes out of..guess we have to check the function with a disassembler

[lb90]

Here's graphene_matrix_init_from_matrix compiled with debugoptimized (crashes):

(lldb) disassemble
libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix:
    0x668742c0 <+0>:  movl   0x4(%esp), %eax
    0x668742c4 <+4>:  movl   0x8(%esp), %ecx
    0x668742c8 <+8>:  movaps 0x30(%ecx), %xmm0
->  0x668742cc <+12>: movaps %xmm0, 0x30(%eax)
    0x668742d0 <+16>: movaps 0x20(%ecx), %xmm0
    0x668742d4 <+20>: movaps %xmm0, 0x20(%eax)
    0x668742d8 <+24>: movaps (%ecx), %xmm0
    0x668742db <+27>: movaps 0x10(%ecx), %xmm1
    0x668742df <+31>: movaps %xmm1, 0x10(%eax)
    0x668742e3 <+35>: movaps %xmm0, (%eax)
    0x668742e6 <+38>: retl

Versus debug (which runs fine):

(lldb) disassemble
libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix:
    0x668b5a30 <+0>:  pushl  %ebp
    0x668b5a31 <+1>:  movl   %esp, %ebp
    0x668b5a33 <+3>:  subl   $0xc, %esp
    0x668b5a36 <+6>:  movl   0xc(%ebp), %eax
    0x668b5a39 <+9>:  movl   0x8(%ebp), %eax
->  0x668b5a3c <+12>: movl   0x8(%ebp), %ecx
    0x668b5a3f <+15>: movl   0xc(%ebp), %eax
    0x668b5a42 <+18>: movl   %ecx, (%esp)
    0x668b5a45 <+21>: movl   %eax, 0x4(%esp)
    0x668b5a49 <+25>: movl   $0x40, 0x8(%esp)
    0x668b5a51 <+33>: calll  0x668d9ac2                ; _memcpy
    0x668b5a56 <+38>: movl   0x8(%ebp), %eax
    0x668b5a59 <+41>: addl   $0xc, %esp
    0x668b5a5c <+44>: popl   %ebp
    0x668b5a5d <+45>: retl

[lb90]

Here's debugoptimized with GCC from MSYS2/MINGW32 (runs fine):

(lldb) disassemble
libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix:
->  0x667366b0 <+0>:  pushl  %ebp
    0x667366b1 <+1>:  movl   %esp, %ebp
    0x667366b3 <+3>:  movl   0xc(%ebp), %edx
    0x667366b6 <+6>:  movl   0x8(%ebp), %eax
    0x667366b9 <+9>:  movdqu (%edx), %xmm1
    0x667366bd <+13>: movups %xmm1, (%eax)
    0x667366c0 <+16>: movdqu 0x10(%edx), %xmm2
    0x667366c5 <+21>: movups %xmm2, 0x10(%eax)
    0x667366c9 <+25>: movdqu 0x20(%edx), %xmm3
    0x667366ce <+30>: movups %xmm3, 0x20(%eax)
    0x667366d2 <+34>: movdqu 0x30(%edx), %xmm0
    0x667366d7 <+39>: movups %xmm0, 0x30(%eax)
    0x667366db <+43>: popl   %ebp
    0x667366dc <+44>: retl
(lldb) register read
General Purpose Registers:
       eax = 0x009EE3F0
       ebx = 0x009EE76C
       ecx = 0x009EE5B0
       edx = 0x667542B0  _graphene_identity_matrix_floats + 48
       edi = 0x00000000
       esi = 0x00000000
       ebp = 0x009EE458
       esp = 0x009EE32C
       eip = 0x667366B0  libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix at 247:1
    eflags = 0b00000000000000000000001001000110

[lb90]

I believe it's a dupe of https://github.com/llvm/llvm-project/issues/55844

movaps %xmm0, 0x30(%eax)

With EAX=0x0E66D688 is not correct since movaps can only operate on addresses that are 16 bytes aligned.

Edit: however graphene_matrix_t is meant to be 16 byte aligned...

[lb90]

graphene_matrix_t is aligned to 16 bytes (even without SSE support). That can also be verified with alignof. Anyway in https://github.com/msys2/MINGW-packages/issues/18799#issuecomment-1837537008 we have m=0x0deea508, which isn't a multiple of 16.

    frame #0: 0x667542cc libgraphene-1.0-0.dll`graphene_matrix_init_from_matrix(m=0x0deea508, src=0x009ee310) at graphene-matrix.c:248:19
   245  graphene_matrix_init_from_matrix (graphene_matrix_t       *m,
   246                                    const graphene_matrix_t *src)
   247  {
-> 248    m->value = src->value;
   249
   250    return m;
   251  }
(lldb) p m
(graphene_matrix_t *) 0x0deea508

The matrix is heap allocated, perhaps the required alignment is not respected by g_type_create_instance

[lb90]

This is going to be fixed within GTK, see https://gitlab.gnome.org/GNOME/gtk/-/issues/6256

[ebassi]

How did you build graphene? Which build flags did you use?

[lb90]

Hi @ebassi! :) Build recipe is here https://github.com/msys2/MINGW-packages/blob/master/mingw-w64-graphene/PKGBUILD#L24

[ebassi]

Thanks, @lb90. Care to check what is the generated graphene-config.h? Because this looks like something is enabling the SSE or GCC autovectorisation implementations of graphene_simd4f_t on i686, which should just not happen.

You can try and rebuild Graphene with -Dsse2=false -Dgcc_vector=false, and see if it reproduces the issue.

[aidenfoxx]

Really amazing work @lb90. ❤️

[lb90]

Here's the configure output:

  Directories
    prefix          : D:/msys64/clang32
    libdir          : lib
    includedir      : include

  SIMD
    SSE             : NO
    GCC/Clang vector: NO
    ARM NEON        : NO

  Features
    GObject types   : YES
    Introspection   : NO

  Build
    Documentation   : NO
    Tests           : YES
    Installed tests : NO

  Subprojects
    mutest          : YES

  User defined options
    buildtype       : release
    gtk_doc         : false
    installed_tests : false

And graphene-config.h.txt

Setting options explicitly with -Dsse2=false -Dgcc_vector=false doesn't seem to help, still crashes.

[lb90]

My understanding is that the compiler emits movaps instructions because it sees that graphene_matrix_t is aligned to 16 bytes and takes advantage of that. That happens regardless if the graphene's SSE2 code path is included for compilation or not.

We may work around that by setting a low -march or something like -mno-sse