stack alignment with VAR on aarch64

[sgraham]

It seems that 16 byte stack alignment doesn't get maintained on aarch64 when using ir_VAR, ir_VSTORE, ir_VLOAD. I'm testing on Mac/aarch64 and the C below generates this code:

myfunc:
	sub sp, sp, #8
	mov w0, #0x4d2
	str w0, [sp]
	ldr w0, [sp]
	add sp, sp, #8
	ret

which makes sp unaligned so bus error when trying to store w0.

If I fiddle around and add a few more local VAR (and disable opts) then it seems to get and stay aligned with more local allocation.

Or am I just ir_VAR incorrectly?

#include <stdlib.h>
#include "ir.h"
#include "ir_builder.h"

/*
 * int32_t myfunc(void) {
 *	int x = 1234;
 *	return x;
 * }
 */
typedef int32_t (*myfunc_t)(void);

void gen_myfunc(ir_ctx* ctx) {
  ir_START();
  ir_ref x = ir_VAR(IR_I32, "x");
  ir_VSTORE(x, ir_CONST_I32(1234));
  ir_ref val = ir_VLOAD(IR_I32, x);
  ir_RETURN(val);
}

int main(int argc, char** argv) {
  ir_ctx ctx;

  ir_consistency_check();

  ir_init(&ctx, IR_FUNCTION, IR_CONSTS_LIMIT_MIN, IR_INSNS_LIMIT_MIN);
  ctx.ret_type = IR_I32;

  gen_myfunc(&ctx);

  size_t size;
  void* entry = ir_jit_compile(&ctx, 0, &size);

  ir_disasm_init();
  ir_disasm("myfunc", entry, size, false, &ctx, stderr);
  ir_disasm_free();

  if (entry) {
    printf("func: %d\n", ((myfunc_t)entry)());
  }

  ir_free(&ctx);

  return 0;
}