## sum(x, axis=1)，f32[8,128]：与 CPU 结果一致 True；HLO 段：%reduce_sum.7 = reduce(%copy-done, %constant.1)
  cld.8x128×1、vadd.8x128.f32×24、vmov.8x128×1、vpop.8x128×17、vrot.slane.down.8x128.u32×7、vxpose.0.start.end.8x128×1

## sum(x, axis=0)，f32[8,128]：与 CPU 结果一致 True；HLO 段：%reduce_sum.7 = reduce(%copy-done, %constant.1)
  cld.8x128×1、vadd.8x128.f32×8、vmov.8x128×1、vpop.8x128×1、vrot.slane.down.8x128.u32×7

## max(x, axis=0)，f32[8,128]：与 CPU 结果一致 True；HLO 段：%reduce_max.7 = reduce(%copy-done, %constant.1)
  cld.8x128×1、vmax.8x128.f32×8、vmov.8x128×1、vpop.8x128×1、vrot.slane.down.8x128.u32×7

## sum(x)，f32[8,128]：与 CPU 结果一致 True；HLO 段：%reduce_sum.7 = reduce(%copy-done, %constant.1)
  cld.8x128×1、vadd.8x128.f32×8、vadd.xlane.0.8x128.f32×1、vmov.8x128×1、vpop.8x128×2、vrot.slane.down.8x128.u32×7

## sum(x, axis=0)，f32[64,128]：与 CPU 结果一致 True；HLO 段：%reduce_sum.7 = reduce(%copy-done, %constant.1)
  cld.8x128×8、vadd.8x128.f32×15、vmov.8x128×1、vpop.8x128×8、vrot.slane.down.8x128.u32×7