.data
  .globl _heapend
  .globl environ
_heapend:
  .word 0
environ:
  .word 0

  .text
  .globl _start

_start:
  li  x1, 0
  li  x2, 0
  li  x3, 0
  li  x4, 0
  li  x5, 0
  li  x6, 0
  li  x7, 0
  li  x8, 0
  li  x9, 0
  li  x10,0
  li  x11,0
  li  x12,0
  li  x13,0
  li  x14,0
  li  x15,0
  li  x16,0
  li  x17,0
  li  x18,0
  li  x19,0
  li  x20,0
  li  x21,0
  li  x22,0
  li  x23,0
  li  x24,0
  li  x25,0
  li  x26,0
  li  x27,0
  li  x28,0
  li  x29,0
  li  x30,0
  li  x31,0

  # enable fp
  mfpcr  x1,cr0
  ori  x1,x1,0x2
  mtpcr  x1,cr0

  # enable vec
  mfpcr  x1,cr0
  ori  x1,x1,0x4
  mtpcr  x1,cr0

  ## if that didn't stick, we don't have an FPU, so don't initialize it
  mfpcr x1,cr0
  andi x1,x1,0x2
  beqz x1,1f

  mtfsr   x0
  mxtf.s  f0, x0
  mxtf.s  f1, x0
  mxtf.s  f2, x0
  mxtf.s  f3, x0
  mxtf.s  f4, x0
  mxtf.s  f5, x0
  mxtf.s  f6, x0
  mxtf.s  f7, x0
  mxtf.s  f8, x0
  mxtf.s  f9, x0
  mxtf.s  f10,x0
  mxtf.s  f11,x0
  mxtf.s  f12,x0
  mxtf.s  f13,x0
  mxtf.s  f14,x0
  mxtf.s  f15,x0
  mxtf.s  f16,x0
  mxtf.s  f17,x0
  mxtf.s  f18,x0
  mxtf.s  f19,x0
  mxtf.s  f20,x0
  mxtf.s  f21,x0
  mxtf.s  f22,x0
  mxtf.s  f23,x0
  mxtf.s  f24,x0
  mxtf.s  f25,x0
  mxtf.s  f26,x0
  mxtf.s  f27,x0
  mxtf.s  f28,x0
  mxtf.s  f29,x0
  mxtf.s  f30,x0
  mxtf.s  f31,x0
1:

  # get core id and number of cores
  mfpcr a0,cr10
  lw a1, 4(zero)

  slli a2, a0, 13
  la sp, stacktop
  sub sp, sp, a2

  la tp, tlstop
  sub tp, tp, a2

  jal thread_entry

  .bss
  .globl stacktop
  .globl tlstop

  .align 4
  .skip 32768
stacktop:
  .skip 65536
tlstop: