fixpipe

功能说明

矩阵计算完成后，对结果进行处理，例如对计算结果进行加偏置和对数据进行量化、并把数据从L1OUT 搬迁到GM Memory 中。

函数原型

fixpipe(dst, src, cburst_num, burst_len, dst_stride, src_stride, extend_params=None)

参数说明

支持的型号

Atlas 200/300/500 推理产品

Atlas 训练系列产品

Atlas推理系列产品（Ascend 310P处理器）AI Core

Atlas A2训练系列产品/Atlas 800I A2推理产品

Atlas 200/500 A2推理产品

注意事项

• 当src和dst的数据类型一致时，extend_params参数的key：quantize_params不生效。
• 单步调试时间比较长，不建议使用。
• 当extend_params中同时使能多个功能时，执行顺序如下所示，不使能的功能请忽略：bias->quantize->relu。

• 该接口不支持与vector相关指令一起使用。
• 操作数地址偏移对齐要求请见通用约束。
• 对于量化输入的单scale或Tensor float32数据类型说明如下：

  （1）标准的IEEE 754 float32格式为：1bit符号位，8bits指数位，23bits尾数位；当前float32格式为：1bit符号位，8bits指数位，10bits尾数位；

  （2）如果用户提供的是标准的IEEE 754 float32输入，API内部会处理成芯片支持的float32格式进行计算，此时如果golden数据生成过程中使用的是标准的IEEE 754 float32数据，则可能引入精度不匹配问题，需要修正golden数据的生成，将量化参数的23bits尾数位的低13bits数据位清零再参与量化计算。

返回值

无

调用示例

• 示例：src和dst的数据类型分别为int32和float16，无bias，量化mode_param为tensor

  from tbe import tik
  tik_instance = tik.Tik()
  dtype_size = {
      "int8": 1,
      "uint8": 1,
      "int16": 2,
      "uint16": 2,
      "float16": 2,
      "int32": 4,
      "uint32": 4,
      "float32": 4,
      "int64": 8,
  }
  fm_dtype = "int8"
  ker_dtype = "int8"
  deq_dtype="float16"
  dst_dtype = "int32"
  fm_shape = [1, 4, 4, 32]
  kernel_shape = [1, 2, 2, 32, 32]
  dst_shape = [2, 16, 16]
  dst_l1_shape = [2, 16, 16]
  deq_shape = [16]
  # 卷积步长, [stride_h, stride_w]
  stride = [1, 1]
  # padding行数/列数，格式是[pad_left, pad_right, pad_top, pad_bottom]
  pad = [0, 0, 0, 0]
  # 空洞卷积参数，格式[dilation_h, dilation_w];
  dilation = [1, 1]
  # padding填充值的数值
  pad_value = 0
  # 定义tensor
  feature_map_gm = tik_instance.Tensor(fm_dtype, fm_shape, name='feature_map_gm', scope=tik.scope_gm)
  weight_gm = tik_instance.Tensor(ker_dtype, kernel_shape, name='weight_gm', scope=tik.scope_gm)
  deqscale_gm = tik_instance.Tensor(deq_dtype, deq_shape, name='deqscale_gm', scope=tik.scope_gm)
  dst_gm = tik_instance.Tensor(deq_dtype, dst_shape, name='dst_gm', scope=tik.scope_gm)
  feature_map = tik_instance.Tensor(fm_dtype, fm_shape, name='feature_map', scope=tik.scope_cbuf)
  weight = tik_instance.Tensor(ker_dtype, kernel_shape, name='weight', scope=tik.scope_cbuf)
  deqscale = tik_instance.Tensor(deq_dtype, deq_shape, name='deqscale', scope=tik.scope_cbuf)
  dst_l1out = tik_instance.Tensor(dst_dtype, dst_l1_shape, name='dst_l1out', scope=tik.scope_cbuf_out)
  # 将数据从gm搬入源操作数tensor
  tik_instance.data_move(feature_map, feature_map_gm, 0, 1, 16, 0, 0)
  tik_instance.data_move(weight, weight_gm, 0, 1, 128, 0, 0)
  tik_instance.data_move(deqscale, deqscale_gm, 0, 1, 1, 0, 0)
  # 进行卷积操作
  tik_instance.conv2d(dst_l1out, feature_map, weight, fm_shape, kernel_shape, stride, pad, dilation, pad_value)
  # fixpipe实现量化
  # 传输数据片段数目, 处理conv2d数据时矩阵计算后的结果cburst_num设为cout_blocks, 处理matmul数据时矩阵计算后的结果cburst_num设为N1
  cburst_num = dst_l1_shape[0]
  # 连续传输数据片段长度,处理conv2d数据时应为howo*16*src_dtype_size/32, 处理matmul数据时应为m*N0*src_dtype_size/32
  burst_len = dst_l1_shape[1] *16 * dtype_size[dst_dtype] // 32
  # dst_stride/src_stride tensor相邻连续数据片段间隔,前burst尾与后burst头,此处暂为连续计算，以0 为例
  dst_stride, src_stride = 0, 0
  
  tik_instance.fixpipe(dst_gm, dst_l1out, cburst_num, burst_len, dst_stride, src_stride, extend_params={"bias": None, "quantize_params": {"mode": "int322fp16", "mode_param": deqscale}})
  tik_instance.BuildCCE(kernel_name="fixpipe", inputs=[feature_map_gm, weight_gm, deqscale_gm], outputs=[dst_gm])

  结果示例：

  输入数据：
  feature_map_gm：
  [[[[3, 2, 4, 2, ..., 4, 3]]]]
  weight_gm：
  [[[[[0, -5, -3, ..., -4, -2]]]]]
  deqscale_gm：
  [ 0.1214, -0.2238, ..., 0.4883, 0.2788]
  输出数据：
  dst_gm：
  [[[-13.48, 39.38, -114.8, 30.38, ..., 9.766, -24.81]]]

• 示例：src和dst的数据类型分别为int32和float16，无bias，量化mode_param为scalar

  from tbe import tik
  tik_instance = tik.Tik()
  dtype_size = {
      "int8": 1,
      "uint8": 1,
      "int16": 2,
      "uint16": 2,
      "float16": 2,
      "int32": 4,
      "uint32": 4,
      "float32": 4,
      "int64": 8,
  }
  fm_dtype = "int8"
  ker_dtype = "int8"
  deq_dtype="float16"
  dst_dtype = "int32"
  fm_shape = [1, 4, 4, 32]
  kernel_shape = [1, 2, 2, 32, 32]
  dst_shape = [2, 16, 16]
  dst_l1_shape = [2, 16, 16]
  # 卷积步长, [stride_h, stride_w]
  stride = [1, 1]
  # padding行数/列数，格式是[pad_left, pad_right, pad_top, pad_bottom]
  pad = [0, 0, 0, 0]
  # 空洞卷积参数，格式[dilation_h, dilation_w];
  dilation = [1, 1]
  # padding填充值的数值
  pad_value = 0
  # 定义tensor
  feature_map_gm = tik_instance.Tensor(fm_dtype, fm_shape, name='feature_map_gm', scope=tik.scope_gm)
  weight_gm = tik_instance.Tensor(ker_dtype, kernel_shape, name='weight_gm', scope=tik.scope_gm)
  dst_gm = tik_instance.Tensor(deq_dtype, dst_shape, name='dst_gm', scope=tik.scope_gm)
  feature_map = tik_instance.Tensor(fm_dtype, fm_shape, name='feature_map', scope=tik.scope_cbuf)
  weight = tik_instance.Tensor(ker_dtype, kernel_shape, name='weight', scope=tik.scope_cbuf)
  dst_l1out = tik_instance.Tensor(dst_dtype, dst_l1_shape, name='dst_l1out', scope=tik.scope_cbuf_out)
  # 将数据从gm搬入源操作数tensor
  tik_instance.data_move(feature_map, feature_map_gm, 0, 1, 16, 0, 0)
  tik_instance.data_move(weight, weight_gm, 0, 1, 128, 0, 0)
  scalar = tik_instance.Scalar(dtype="float32", init_value=2)
  # 进行卷积操作
  tik_instance.conv2d(dst_l1out, feature_map, weight, fm_shape, kernel_shape, stride, pad, dilation, pad_value)
  # fixpipe实现量化
  # 传输数据片段数目, 处理conv2d数据时矩阵计算后的结果cburst_num设为cout_blocks, 处理matmul数据时矩阵计算后的结果cburst_num设为N1
  cburst_num = dst_l1_shape[0]
  # 连续传输数据片段长度,处理conv2d数据时应为howo*16*src_dtype_size/32, 处理matmul数据时应为m*N0*src_dtype_size/32
  burst_len = dst_l1_shape[1] *16 * dtype_size[dst_dtype] // 32
  # dst_stride/src_stride tensor相邻连续数据片段间隔,前burst尾与后burst头,此处暂为连续计算，以0 为例
  dst_stride, src_stride = 0, 0
  tik_instance.fixpipe(dst_gm, dst_l1out, cburst_num, burst_len, dst_stride, src_stride, extend_params={"quantize_params": {"mode": "int322fp16", "mode_param": scalar}})
  tik_instance.BuildCCE(kernel_name="fixpipe", inputs=[feature_map_gm, weight_gm], outputs=[dst_gm])
  

  结果示例：

  输入数据：
  feature_map_gm：
  [[[[-3,  4, -3, -1， ...,4, -1]]]]
  weight_gm：
  [[[[[ 1,  4, -3，...,-1, -4, -2]]]]]
  输出数据：
  dst_gm：
  [[[ -78.  120.  -76.  144. -114. ...0.    0.    0.]]]

• 示例：src和dst的数据类型分别为float32和float16，有bias，量化mode_param为None

  from tbe import tik
  tik_instance = tik.Tik()
  # 定义tensor
  feature_map_gm = tik_instance.Tensor("float16", [2, 4, 4, 16], name='feature_map_gm', scope=tik.scope_gm)
  weight_gm = tik_instance.Tensor("float16", [2, 2, 2, 16, 16], name='weight_gm', scope=tik.scope_gm)
  bias_gm = tik_instance.Tensor("float32", (16,), name='bias_gm', scope=tik.scope_gm)
  dst_gm = tik_instance.Tensor("float16", [1, 4, 16], name='dst_gm', scope=tik.scope_gm)
  feature_map = tik_instance.Tensor("float16", [2, 4, 4, 16], name='feature_map', scope=tik.scope_cbuf)
  weight = tik_instance.Tensor("float16", [2, 2, 2, 16, 16], name='weight', scope=tik.scope_cbuf)
  bias = tik_instance.Tensor("float32", (16,), name='bias', scope=tik.scope_cbuf)
  dst_l1out = tik_instance.Tensor("float32", [1, 16, 16], name='dst_l1out', scope=tik.scope_cbuf_out)
  # 将数据从gm搬入源操作数tensor
  tik_instance.data_move(feature_map, feature_map_gm, 0, 1, 32, 0, 0)
  tik_instance.data_move(weight, weight_gm, 0, 1, 128, 0, 0)
  tik_instance.data_move(bias, bias_gm, 0, 1, 2, 0, 0)
  # 进行卷积操作
  tik_instance.conv2d(dst_l1out, feature_map, weight, [2, 4, 4, 16], [2, 2, 2, 16, 16], [1, 1], [0, 0, 0, 0], [2, 2], 0)
  # fixpipe实现bias功能和量化功能
  tik_instance.fixpipe(dst_gm, dst_l1out, 1, 8, 0, 0, extend_params={"bias": bias, "quantize_params": {"mode": "fp322fp16", "mode_param": None}})
  tik_instance.BuildCCE(kernel_name="conv2d", inputs=[feature_map_gm, weight_gm, bias_gm], outputs=[dst_gm])

  结果示例:

  输入数据
  feature_map_gm：
  [[[[0.0, 0.01, 0.02, 0.03, 0.04, ..., 5.09, 5.1, 5.11]]]]
  weight_gm:
  [[[[[0.0, 0.01, 0.02, 0.03, 0.04, ..., 20.46, 20.47]]]]]
  bias_gm:
  [0.0, 1.0, 2.0, 3.0, ..., 14.0, 15.0]
  
  输出数据
  dst_gm：
  [[[3568., 3614., 3660., 3704., 3750., 3794., 3840., 3884., 3930.,
     3976., 4020., 4066., 4110., 4156., 4200., 4250.],
    [3754., 3802., 3850., 3900., 3948., 3996., 4044., 4094., 4140.,
     4188., 4240., 4290., 4336., 4384., 4430., 4480.],
    [4308., 4370., 4424., 4484., 4544., 4600., 4660., 4716., 4776.,
     4830., 4892., 4950., 5010., 5068., 5124., 5184.],
    [4496., 4556., 4616., 4680., 4740., 4804., 4864., 4924., 4988.,
     5050., 5108., 5172., 5230., 5296., 5356., 5416.]]]

• 示例：src和dst的数据类型分别为float32和float16，有bias，量化mode_param为None

  from tbe import tik
  tik_instance = tik.Tik()
  # 定义tensor
  feature_map_gm = tik_instance.Tensor("float16", [2, 4, 4, 16], name='feature_map_gm', scope=tik.scope_gm)
  weight_gm = tik_instance.Tensor("float16", [2, 2, 2, 16, 16], name='weight_gm', scope=tik.scope_gm)
  bias_gm = tik_instance.Tensor("float32", (16,), name='bias_gm', scope=tik.scope_gm)
  dst_gm = tik_instance.Tensor("float16", [1, 4, 16], name='dst_gm', scope=tik.scope_gm)
  feature_map = tik_instance.Tensor("float16", [2, 4, 4, 16], name='feature_map', scope=tik.scope_cbuf)
  weight = tik_instance.Tensor("float16", [2, 2, 2, 16, 16], name='weight', scope=tik.scope_cbuf)
  bias = tik_instance.Tensor("float32", (16,), name='bias', scope=tik.scope_cbuf)
  dst_l1out = tik_instance.Tensor("float32", [1, 16, 16], name='dst_l1out', scope=tik.scope_cbuf_out)
  # 将数据从gm搬入源操作数tensor
  tik_instance.data_move(feature_map, feature_map_gm, 0, 1, 32, 0, 0)
  tik_instance.data_move(weight, weight_gm, 0, 1, 128, 0, 0)
  tik_instance.data_move(bias, bias_gm, 0, 1, 2, 0, 0)
  # 进行卷积操作
  tik_instance.conv2d(dst_l1out, feature_map, weight, [2, 4, 4, 16], [2, 2, 2, 16, 16], [1, 1], [0, 0, 0, 0], [2, 2], 0, bias=bias)
  # fixpipe实现bias功能和量化功能
  tik_instance.fixpipe(dst_gm, dst_l1out, 1, 8, 0, 0, extend_params={"quantize_params": {"mode": "fp322fp16", "mode_param": None}})
  tik_instance.BuildCCE(kernel_name="conv2d", inputs=[feature_map_gm, weight_gm, bias_gm], outputs=[dst_gm])

  结果示例:

  输入数据
  feature_map_gm：
  [[[[0.0, 0.01, 0.02, 0.03, 0.04, ..., 5.09, 5.1, 5.11]]]]
  weight_gm:
  [[[[[0.0, 0.01, 0.02, 0.03, 0.04, ..., 20.46, 20.47]]]]]
  bias_gm:
  [0.0, 1.0, 2.0, 3.0, ..., 14.0, 15.0]
  
  输出数据
  dst_gm：
  [[[3568., 3614., 3660., 3704., 3750., 3794., 3840., 3884., 3930.,
     3976., 4020., 4066., 4110., 4156., 4200., 4250.],
    [3754., 3802., 3850., 3900., 3948., 3996., 4044., 4094., 4140.,
     4188., 4240., 4290., 4336., 4384., 4430., 4480.],
    [4308., 4370., 4424., 4484., 4544., 4600., 4660., 4716., 4776.,
     4830., 4892., 4950., 5010., 5068., 5124., 5184.],
    [4496., 4556., 4616., 4680., 4740., 4804., 4864., 4924., 4988.,
     5050., 5108., 5172., 5230., 5296., 5356., 5416.]]]