数值检查通过

# fusion 段的指令统计
dma.strided: 2
por: 8
sadd.s32: 8
sand.u32: 4
shalt: 4
simm.s32: 4
slt.s32: 8
sne.s32: 4
vld.8x128: 8
vmul.8x128.f32: 8
vst.8x128: 8
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, 64, s0 ;
  misc: vsyncpa.u1 [sflag:53], 0 }
{ s0: sadd.s32 s9, 64, 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, 64, s18 ;
  .encoding { s0.y = lo(imm0) } }
{ s0: sadd.s32 s17, 64, 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: simm.s32 s20, 8 }  # materialized constant
{ s0: dma.strided [vmem:s18], [hbm:s0], length=64, dst_stride=s20, src_stride=s20, elements_per_stride=s20, dst_flag=[sflag:52] }  # base_bounds: (8, 1)\ndynamic_base_bounds: (8, 1)\nwindow_bounds: (8, 1)\niteration_bounds: (1, 1)\nstrides: (8, 1)\npad_low: (0, 0)\npad_high: (0, 0)\nelement_size_in_bytes: 4096\nsecond_minor_factor: 1
{ misc: vwait.ge [sflag:52], 64 ;  # pipeline-emitter-dma-wait
  .encoding { misc.value = imm2 } }
{ s0: sand.u32 s0, 0x40, s15 ;
  s1: simm.s32 s24, 64 ;  # materialized constant
  vld: vld.8x128 v0, [vmem:0x0] ;
  misc: vsyncadd.s32 [sflag:52], -64 ;  # pipeline-emitter-dma-wait
  .encoding { misc.value = imm2 ; s1.y = lo(imm0) } }
{ s0: sadd.s32 s23, 64, s0 ;
  s1: sadd.s32 s25, 64, s24 ;
  va0: vmul.8x128.f32 v1, 2.0, v0 ;
  vld: vld.8x128 v2, [vmem:0x8] }
{ s0: sne.s32 p10, s0, s23 ;
  s1: slt.s32 p11, s0, 64 ;
  va0: vmul.8x128.f32 v3, 2.0, v2 ;
  vld: vld.8x128 v4, [vmem:0x10] ;
  .encoding { s1.y = lo(imm0) } }
{ s0: slt.s32 p12, s25, s23 ;
  va0: vmul.8x128.f32 v5, 2.0, v4 ;
  vst: vst.8x128 [vmem:0x40], v1 ;
  vld: vld.8x128 v6, [vmem:0x18] ;
  .encoding { vst.offset = imm2 } }
{ s0: por p13, p12, p11 ;
  va0: vmul.8x128.f32 v7, 2.0, v6 ;
  vst: vst.8x128 [vmem:0x48], v3 ;
  vld: vld.8x128 v8, [vmem:0x20] ;
  .encoding { vst.offset = imm2 } }
{ s0: por p14, !p13, !p10 ;
  va0: vmul.8x128.f32 v9, 2.0, v8 ;
  vst: vst.8x128 [vmem:0x50], v5 ;
  vld: vld.8x128 v10, [vmem:0x28] ;
  .encoding { vst.offset = imm2 } }
{ va0: vmul.8x128.f32 v11, 2.0, v10 ;
  vst: vst.8x128 [vmem:0x58], v7 ;
  vld: vld.8x128 v12, [vmem:0x30] ;
  .encoding { vst.offset = imm2 } }
{ va0: vmul.8x128.f32 v13, 2.0, v12 ;
  vst: vst.8x128 [vmem:0x60], v9 ;
  vld: vld.8x128 v14, [vmem:0x38] ;
  .encoding { vst.offset = imm2 } }
{ va0: vmul.8x128.f32 v15, 2.0, v14 ;
  vst: vst.8x128 [vmem:0x68], v11 }
{ vst: vst.8x128 [vmem:0x70], v13 }
{ vst: vst.8x128 [vmem:0x78], v15 }
{ s0: @!p14 shalt }  # BoundsCheck 17 [deref of opnd#0] for dma.vmem_to_hbm\nhlo: broadcast_multiply_fusion
{ s0: sand.u32 s28, 0x7fffffff, s1 ;
  s1: sadd.s32 s3, 64, s1 }
{ s0: sadd.s32 s29, 64, s28 ;
  s1: slt.s32 p1, s28, s1 }
{ s0: sne.s32 p0, s28, s29 ;
  s1: slt.s32 p2, s3, s29 }
{ s0: por p3, p2, p1 }
{ s0: por p4, !p3, !p0 }
{ s0: @!p4 shalt }  # BoundsCheck 17 [deref of opnd#2] for dma.vmem_to_hbm\nhlo: broadcast_multiply_fusion
{ s0: dma.strided [hbm:s1], [vmem:s24], length=64, dst_stride=s20, src_stride=s20, elements_per_stride=s20, dst_flag=[sflag:53] }  # base_bounds: (8, 1)\ndynamic_base_bounds: (8, 1)\nwindow_bounds: (8, 1)\niteration_bounds: (1, 1)\nstrides: (8, 1)\npad_low: (0, 0)\npad_high: (0, 0)\nelement_size_in_bytes: 4096\nsecond_minor_factor: 1
{ misc: vwait.ge [sflag:53], 64 ;  # pipeline-emitter-dma-wait
  .encoding { misc.value = imm2 } }
{ misc: vsyncadd.s32 [sflag:53], -64 ;  # 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)