数值检查通过

# 完整清单的结构概览
bundle    0: prelude-prefix-start
bundle    9: routine:wait-for-active-dma
bundle   16: routine:program-exit
bundle   19: routine:allocation-table-find
bundle   28: routine:allocation-table-get-or-add
bundle   54: routine:allocation-table-cut-and-add
bundle   71: routine:branch-impl
bundle  122: routine:branch
bundle  128: trampoline:program-start
bundle  484: entry bundle: %broadcast_multiply_fusion = fusion(%x.1)
bundle  517: exit bundle: %broadcast_multiply_fusion = fusion(%x.1)
bundle  620: 清单结束（共 620 个 bundle）

# fusion 段的指令统计
dma.simple: 2
por: 8
sadd.s32: 8
sand.u32: 4
shalt: 4
simm.s32: 3
slt.s32: 8
sne.s32: 4
vld.8x128: 1
vmul.8x128.f32: 1
vst.8x128: 1
vsyncadd.s32: 2
vsyncpa.u1: 3
vwait.ge: 2

# fusion 段
# entry bundle: %broadcast_multiply_fusion = fusion(%x.1)
{ misc: vsyncpa.u1 [sflag:52], 0 }
{ s0: sand.u32 s8, 0x7fffffff, s0 ;
  s1: sadd.s32 s12, 8, s0 ;
  misc: vsyncpa.u1 [sflag:53], 0 }
{ s0: sadd.s32 s9, 8, s8 ;
  s1: slt.s32 p1, s8, s0 }
{ s0: sne.s32 p0, s8, s9 ;
  s1: slt.s32 p2, s12, s9 }
{ s0: por p3, p2, p1 }
{ s0: por p4, !p3, !p0 }
{ s0: @!p4 shalt }  # BoundsCheck 0 [deref of opnd#0] for dma.hbm_to_vmem\nhlo: broadcast_multiply_fusion
{ s0: simm.s32 s15, 2147483647 ;  # materialized constant
  s1: simm.s32 s18, 0 ;  # materialized constant
  .encoding { s1.y = lo(imm2) } }
{ s0: sand.u32 s16, 0x0, s15 ;
  s1: sadd.s32 s19, 8, s18 ;
  .encoding { s0.y = lo(imm0) } }
{ s0: sadd.s32 s17, 8, s16 ;
  s1: slt.s32 p6, s16, 0 ;
  .encoding { s1.y = lo(imm0) } }
{ s0: sne.s32 p5, s16, s17 ;
  s1: slt.s32 p7, s19, s17 }
{ s0: por p8, p7, p6 }
{ s0: por p9, !p8, !p5 }
{ s0: @!p9 shalt }  # BoundsCheck 0 [deref of opnd#2] for dma.hbm_to_vmem\nhlo: broadcast_multiply_fusion
{ s0: dma.simple [vmem:s18], [hbm:s0], length=8, dst_flag=[sflag:52] }
{ misc: vwait.ge [sflag:52], 8 ;  # pipeline-emitter-dma-wait
  .encoding { misc.value = imm2 } }
{ s0: sand.u32 s0, 0x8, s15 ;
  s1: simm.s32 s23, 8 ;  # materialized constant
  vld: vld.8x128 v0, [vmem:0x0] ;
  misc: vsyncadd.s32 [sflag:52], -8 ;  # pipeline-emitter-dma-wait
  .encoding { misc.value = imm2 ; s1.y = lo(imm0) } }
{ s0: sadd.s32 s22, 8, s0 ;
  s1: sadd.s32 s24, 8, s23 ;
  va0: vmul.8x128.f32 v1, 2.0, v0 }
{ s0: sne.s32 p10, s0, s22 ;
  s1: slt.s32 p11, s0, 8 ;
  .encoding { s1.y = lo(imm0) } }
{ s0: slt.s32 p12, s24, s22 ;
  vst: vst.8x128 [vmem:0x8], v1 }
{ s0: por p13, p12, p11 }
{ s0: por p14, !p13, !p10 }
{ s0: @!p14 shalt }  # BoundsCheck 3 [deref of opnd#0] for dma.vmem_to_hbm\nhlo: broadcast_multiply_fusion
{ s0: sand.u32 s27, 0x7fffffff, s1 ;
  s1: sadd.s32 s2, 8, s1 }
{ s0: sadd.s32 s28, 8, s27 ;
  s1: slt.s32 p1, s27, s1 }
{ s0: sne.s32 p0, s27, s28 ;
  s1: slt.s32 p2, s2, s28 }
{ s0: por p3, p2, p1 }
{ s0: por p4, !p3, !p0 }
{ s0: @!p4 shalt }  # BoundsCheck 3 [deref of opnd#2] for dma.vmem_to_hbm\nhlo: broadcast_multiply_fusion
{ s0: dma.simple [hbm:s1], [vmem:s23], length=8, dst_flag=[sflag:53] }
{ misc: vwait.ge [sflag:53], 8 ;  # pipeline-emitter-dma-wait
  .encoding { misc.value = imm2 } }
{ misc: vsyncadd.s32 [sflag:53], -8 ;  # pipeline-emitter-dma-wait
  .encoding { misc.value = imm2 } }
{ misc: vsyncpa.u1 [sflag:52], 1 ;
  .encoding { misc.value = imm2 } }
# exit bundle: %broadcast_multiply_fusion = fusion(%x.1)