chore: checkpoint before Python removal

vendor/psm/src/arch/x86_windows_gnu.s

@@ -0,0 +1,94 @@
+/* FIXME: this works locally but not on appveyor??!? */
+/* NOTE: fastcall calling convention used on all x86 targets */
+.text
+
+.def @rust_psm_stack_direction@0
+.scl 2
+.type 32
+.endef
+.globl @rust_psm_stack_direction@0
+.p2align 4
+@rust_psm_stack_direction@0:
+/* extern "fastcall" fn() -> u8 (%al) */
+.cfi_startproc
+    movb $2, %al # always descending on x86_64
+    retl
+.cfi_endproc
+
+
+.def @rust_psm_stack_pointer@0
+.scl 2
+.type 32
+.endef
+.globl @rust_psm_stack_pointer@0
+.p2align 4
+@rust_psm_stack_pointer@0:
+/* extern "fastcall" fn() -> *mut u8 (%rax) */
+.cfi_startproc
+    leal 4(%esp), %eax
+    retl
+.cfi_endproc
+
+
+.def @rust_psm_replace_stack@16
+.scl 2
+.type 32
+.endef
+.globl @rust_psm_replace_stack@16
+.p2align 4
+@rust_psm_replace_stack@16:
+/* extern "fastcall" fn(%ecx: usize, %edx: extern "fastcall" fn(usize), 4(%esp): *mut u8) */
+.cfi_startproc
+/*
+   All we gotta do is set the stack pointer to 4(%esp) & tail-call the callback in %edx
+
+   Note, that the callee expects the stack to be offset by 4 bytes (normally, a return address
+   would be store there) off the required stack alignment on entry. To offset the stack in such a
+   way we use the `calll` instruction, however it would also be possible to to use plain `jmpl` but
+   would require to adjust the stack manually, which cannot be easily done, because the stack
+   pointer argument is already stored in memory.
+ */
+    movl 8(%esp), %eax
+    mov %eax, %fs:0x08
+    movl 4(%esp), %esp
+    mov %esp, %fs:0x04
+    calll *%edx
+    ud2
+.cfi_endproc
+
+
+.def @rust_psm_on_stack@16
+.scl 2
+.type 32
+.endef
+.globl @rust_psm_on_stack@16
+.p2align 4
+@rust_psm_on_stack@16:
+/* extern "fastcall" fn(%ecx: usize, %edx: usize, 4(%esp): extern "fastcall" fn(usize, usize), 8(%esp): *mut u8) */
+.cfi_startproc
+    pushl %ebp
+    .cfi_def_cfa %esp, 8
+    .cfi_offset %ebp, -8
+    pushl %fs:0x04
+    .cfi_def_cfa %esp, 12
+    pushl %fs:0x08
+    .cfi_def_cfa %esp, 16
+    movl  %esp, %ebp
+    .cfi_def_cfa_register %ebp
+
+    movl 24(%ebp), %eax
+    movl %eax, %fs:0x08
+    movl  20(%ebp), %esp
+    movl %esp, %fs:0x04
+    calll *16(%ebp)
+
+    movl  %ebp, %esp
+    popl %fs:0x08
+    .cfi_def_cfa %esp, 12
+    popl %fs:0x04
+    .cfi_def_cfa %esp, 8
+    popl  %ebp
+    .cfi_def_cfa %esp, 4
+    retl  $12
+.cfi_endproc
+