ethosu/vela/weight_compressor.py

# Copyright (C) 2020 Arm Limited or its affiliates. All rights reserved.
#
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the License); you may
# not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an AS IS BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# Description:
# Compresses and pads the weigths. It also calculates the scales and packs with the biases.
import math
from collections import namedtuple

import numpy as np

from .architecture_features import Accelerator
from .architecture_features import ArchitectureFeatures
from .data_type import DataType
from .errors import typecheck
from .errors import UnsupportedFeatureError
from .nn_graph import SchedulingStrategy
from .numeric_util import round_up
from .numeric_util import round_up_divide
from .operation import NpuBlockType
from .scaling import quantise_scale
from .scaling import reduced_quantise_scale
from .tensor import TensorBlockTraversal
from .tensor import TensorFormat
from .tensor import TensorPurpose
from .tensor import TensorSubPurpose
from ethosu import mlw_codec


# Contains meta info for a weight compression. If two tensors have identical weight compression config,
# then they also will have identical compressed weights.
WeightCompressionConfig = namedtuple(
    "WeightCompressionConfig", ["npu_block_type", "ofm_block_depth", "ofm_depth_step", "dilation", "equivalence_id"]
)


@typecheck
def encode_weights(
    accelerator: Accelerator,
    weights_volume: np.ndarray,
    dilation_xy: tuple,
    ifm_bitdepth: int,
    ofm_block_depth: int,
    is_depthwise: bool,
    is_partkernel: bool,
):
    """
    Public facing API to use the ethosu weight encoding.

    :param accelerator: architecture_features.Accelerator enum to pick the correct ethosu accelerator
    :param weights_volume: numpy.ndarray in OHWI layout with a shape of four
    :param dilation_xy: a two element tuple of dilation attributes in x,y dimension
    :param ifm_bitdepth: the bitdepth of input feature map
    :param ofm_block_depth: the depth of blocks for ethosu processing
    :param is_depthwise: a boolean indicating these weights are used for a depthwise traversal
    :param is_partkernel: a boolean indicating these weights are traversed on sub-kernal basis
    :return: a bytearray of compressed weights
    """

    # Checks for weight layout
    assert len(weights_volume.shape) == 4, "weights ndarray should have a shape of 4"

    # It cannot be both partkernel and depthwise
    assert not (is_depthwise and is_partkernel), "encode_weights :: partkernel and depthwise are mutually exclusive"

    # Check valid values for dilation
    assert dilation_xy[0] in (1, 2), "encode_weights :: dilation x should be 1 or 2 not {}".format(dilation_xy[0])
    assert dilation_xy[1] in (1, 2), "encode_weights :: dilation y should be 1 or 2 not {}".format(dilation_xy[1])

    ifm_ublock = ArchitectureFeatures.accelerator_configs[accelerator].ifm_ublock
    ofm_ublock = ArchitectureFeatures.accelerator_configs[accelerator].ofm_ublock
    raw_stream = generate_brick(
        ifm_ublock=ifm_ublock,
        ofm_ublock=ofm_ublock,
        brick_weights=weights_volume,
        ofm_block_depth=ofm_block_depth,
        is_depthwise=is_depthwise,
        is_partkernel=is_partkernel,
        ifm_bitdepth=ifm_bitdepth,
        dilation=dilation_xy,
    )
    encoded_stream = encode(raw_stream)
    return encoded_stream


def create_weight_compression_config(tens, npu_block_type, ofm_block_depth, ofm_depth_step, dilation):
    # Note: for an ofm block only its depth is used in weight compression.
    # And block depth > ofm depth gives same result as block depth == ofm depth
    block_depth = min(ofm_block_depth, tens.quant_values.shape[-1])
    return WeightCompressionConfig(npu_block_type, block_depth, ofm_depth_step, dilation, tens.equivalence_id)


def set_storage_shape(tens):
    # Sets the storage shape depending on the tensor's sub purpose
    if tens.sub_purpose == TensorSubPurpose.DoubleBuffer and len(tens.compressed_values) > 2:
        offset = 2 * np.amax([len(x) for x in tens.compressed_values])
        assert offset % 16 == 0
    else:
        offset = tens.weight_compressed_offsets[-1]
    tens.storage_shape = [1, 1, 1, offset]


class CompressedWeightCache:
    # Contains weight compressions for all weight tensors in a graph
    def __init__(self):
        self.cache = {}  # maps from WeightCompressionConfig to a tensor clone containing compressed weights

    def get_tensor_with_same_compression(self, wcc):
        return self.cache.get(wcc)

    def add(self, tens):
        # Adds the compressed weights from the tensor to the cache
        wcc = tens.weight_compression_config
        # Clone the tensor to make sure that nothing related to the weight compression is modified
        tens_clone = tens.clone("_weights{}_{}".format(wcc.ofm_block_depth, wcc.ofm_depth_step))
        self.cache[wcc] = tens_clone


def encode(weight_stream):
    if len(weight_stream) == 0:
        return []
    assert np.amin(weight_stream) >= -255
    assert np.amax(weight_stream) <= 255

    # Encode flattened signed weight stream
    compressed = mlw_codec.encode(weight_stream)

    # pad with 0xFF as needed so the length of the weight stream
    # is a multiple of 16

    while (len(compressed) % 16) != 0:
        compressed.append(0xFF)

    return compressed


def generate_brick(
    ifm_ublock, ofm_ublock, brick_weights, ofm_block_depth, is_depthwise, is_partkernel, ifm_bitdepth, dilation
):

    decomp_h = ArchitectureFeatures.SubKernelMax.height // dilation[0]
    decomp_w = ArchitectureFeatures.SubKernelMax.width // dilation[1]
    # Expect weights formatted OHWI
    ofm_depth = brick_weights.shape[-4]
    ifm_depth = brick_weights.shape[-1]
    kernel_width = brick_weights.shape[-2]
    kernel_height = brick_weights.shape[-3]
    # IFM block depth
    if is_partkernel or (ifm_bitdepth == 16):
        # IFM block depth is always 16 for part-kernel-first
        ifm_block_depth = 16
    elif ifm_bitdepth == 8:
        ifm_block_depth = 32
    else:
        assert False

    stream = []

    # Top level striping - OFM blocks in the entire brick's depth
    for ofm_block_z in range(0, ofm_depth, ofm_block_depth):
        clipped_ofm_block_depth = min(ofm_block_depth, ofm_depth - ofm_block_z)
        # IFM blocks required for the brick
        for ifm_block_z in range(0, (1 if is_depthwise else ifm_depth), ifm_block_depth):
            if is_depthwise:
                clipped_ifm_block_depth = ifm_ublock.depth
            else:
                clipped_ifm_block_depth = (
                    min(ifm_block_depth, ifm_depth - ifm_block_z) if is_partkernel else ifm_block_depth
                )
            # Weight decomposition
            # Subkernel Splitting  (H)
            for subkernel_y in range(0, kernel_height, decomp_h):
                sub_height = min(kernel_height - subkernel_y, decomp_h)
                # Subkernel splitting (W)
                for subkernel_x in range(0, kernel_width, decomp_w):
                    sub_width = min(kernel_width - subkernel_x, decomp_w)
                    subkernel_elements = sub_width * sub_height
                    # Part kernel first works across the kernel H/W and needs padding
                    if is_partkernel:
                        if ifm_bitdepth == 16 and subkernel_elements % 2 != 0:
                            subkernel_elements = int(math.ceil(subkernel_elements / 2) * 2)
                        elif ifm_bitdepth == 8 and subkernel_elements % 4 != 0:
                            subkernel_elements = int(math.ceil(subkernel_elements / 4) * 4)

                    # Depthwise Conv requires multiple of 4 kernel elements in its weight block
                    # this is different from normal conv which is considered "weights depth-first"
                    elif is_depthwise:
                        subkernel_elements = int(math.ceil(subkernel_elements / 4.0) * 4)

                    ifm_block_depth_outer = clipped_ifm_block_depth if is_partkernel else 1
                    ifm_block_depth_inner = 1 if is_partkernel else clipped_ifm_block_depth
                    # IFM Ublocks in IFM-block over depth for part-kernel-first mode
                    # For depth-first IFM Ublocks are traversed after subkernel elements so this loop is ignored.
                    for ifm_ublk_outer in range(0, ifm_block_depth_outer, ifm_ublock.depth):
                        # OFM Ublocks in OFM-block over depth
                        for ofm_ublk in range(0, clipped_ofm_block_depth, ofm_ublock.depth):
                            # HW Kernel element traversal - cannot be a H/W loop due to element
                            # padding requirement on depthwise/part-kernel configurations
                            for element in range(subkernel_elements):
                                kx = element % sub_width
                                ky = element // sub_width
                                # IFM Ublocks in IFM-block over depth (only 1 ublock if depthwise)
                                # In case of part-kernel-first IFM Ublock traversal have already been handled
                                # and this loop is ignored.
                                for ifm_ublk_inner in range(0, ifm_block_depth_inner, ifm_ublock.depth):
                                    # Feed OFM ublock elements
                                    for ofm_ublock_z in range(ofm_ublock.depth):
                                        # Source IFM ublock elements (only 1 element deep if depthwise)
                                        for ifm_ublock_z in range(1 if is_depthwise else ifm_ublock.depth):
                                            # Source position within the current subkernel
                                            wx = subkernel_x + kx
                                            wy = subkernel_y + ky
                                            # Source IFM/OFM slices
                                            ifm_ublk = ifm_ublk_inner + ifm_ublk_outer
                                            ifm_z = ifm_block_z + ifm_ublk + ifm_ublock_z
                                            ofm_z = ofm_block_z + ofm_ublk + ofm_ublock_z
                                            if (ifm_z >= ifm_depth) or (ofm_z >= ofm_depth) or (ky >= sub_height):
                                                stream.append(0)
                                            else:
                                                stream.append(brick_weights[ofm_z][wy][wx][ifm_z])
    return stream


def core_deinterleave(hwio, core, ncores):
    # Put weights back into OHWI
    ohwi = np.transpose(hwio, (3, 0, 1, 2))
    return ohwi[core : ohwi.shape[0] : ncores]


# Compress the weights
def compress_weights(arch, nng, tens, npu_block_type, ofm_block_depth, ofm_depth_step, dilation):
    assert tens.purpose == TensorPurpose.Weights
    assert tens.format == TensorFormat.WeightsCompressed

    # Check the weight cache
    if nng.weight_cache is None:
        nng.weight_cache = CompressedWeightCache()
    wcc = create_weight_compression_config(tens, npu_block_type, ofm_block_depth, ofm_depth_step, dilation)
    tens.weight_compression_config = wcc
    tens_cached = nng.weight_cache.get_tensor_with_same_compression(wcc)
    if tens_cached is not None:
        # Cache hit, copy weights from the cache
        tens.copy_compressed_weight_info(tens_cached)
        set_storage_shape(tens)
        return

    # No cache hit, perform the compression
    assert tens.quantization is not None
    assert tens.quantization.scale_f32 is not None
    assert tens.quantization.zero_point is not None

    zero_point = tens.quantization.zero_point
    quant_buf = tens.quant_values.astype(np.int64)

    # Early zero-point correction
    weights = quant_buf - zero_point

    if len(weights.shape) == 2:
        weights = np.expand_dims(np.expand_dims(weights, axis=0), axis=0)
        weights_shape = (weights.shape[0], 1, 1, weights.shape[1])
    else:
        weights_shape = weights.shape

    compression_scales = []
    compressed_offsets = []
    encoded_streams = []
    encoded_streams_substream_offsets = []
    offset = 0
    max_single_buffer_len = 0

    ifm_bitdepth = tens.consumer_list[0].inputs[0].dtype.size_in_bits()
    ifm_depth = weights.shape[-2]
    if npu_block_type == NpuBlockType.ConvolutionDepthWise:
        tens.block_traversal = TensorBlockTraversal.DepthWise
    if npu_block_type == NpuBlockType.ConvolutionMxN:
        # Determine which block traversal strategy has better DPU utilization
        kernel_size = weights_shape[0] * weights_shape[1]
        depth_utilization = weights_shape[2] / round_up(weights_shape[2], 32 if ifm_bitdepth == 8 else 16)
        part_kernel_utilization = (weights_shape[2] / round_up(weights_shape[2], 8)) * (
            kernel_size / round_up(kernel_size, 4 if ifm_bitdepth == 8 else 2)
        )
        if part_kernel_utilization >= depth_utilization or ifm_depth <= 8:
            # Part-kernel first is always better for ifm depths <= 8
            tens.block_traversal = TensorBlockTraversal.PartKernelFirst
        else:
            tens.block_traversal = TensorBlockTraversal.DepthFirst

    is_depthwise = tens.block_traversal == TensorBlockTraversal.DepthWise
    is_partkernel = tens.block_traversal == TensorBlockTraversal.PartKernelFirst

    if tens.consumer_list[0].type == "Conv2DBackpropInputSwitchedBias":
        # Transpose Convoluion, reverse weights in H and W axes
        weights = np.flip(weights, axis=(0, 1))

    # Calculate brick size
    brick_size = (weights_shape[0], weights_shape[1], weights_shape[2], min(tens.shape[-1], ofm_depth_step))
    elements_in_brick = np.prod(brick_size)

    # Slice weight stream up depth-ways into bricks and compress
    full_ofm_depth = quant_buf.shape[-1]
    for idx in range(0, full_ofm_depth, ofm_depth_step):
        # Get the weights necessary for this brick
        count = min(full_ofm_depth - idx, ofm_depth_step)
        brick_weights = weights[:, :, :, idx : idx + count]

        substream_offsets = [0]
        encoded_stream = []

        # For each core, deinterleave weights from the larger volume
        # and generate separate compressed streams.
        for core in range(0, min(arch.ncores, full_ofm_depth)):
            core_weights = core_deinterleave(brick_weights, core, arch.ncores)

            block_depth = (ofm_block_depth + arch.ncores - 1 - core) // arch.ncores
            encoded_substream = []
            if block_depth != 0:
                encoded_substream = encode_weights(
                    accelerator=arch.accelerator_config,
                    weights_volume=core_weights,
                    dilation_xy=dilation,
                    ifm_bitdepth=ifm_bitdepth,
                    ofm_block_depth=block_depth,
                    is_depthwise=is_depthwise,
                    is_partkernel=is_partkernel,
                )
            encoded_stream.extend(encoded_substream)
            substream_offsets.append(len(encoded_stream))

        encoded_streams.append(encoded_stream)
        encoded_streams_substream_offsets.append(substream_offsets)

        # Remember maximum encoded length for DoubleBuffering
        max_single_buffer_len = max(max_single_buffer_len, len(encoded_stream))

        # Remember where we put it for linear addressing
        compressed_offsets.append(offset)
        offset += len(encoded_stream)
        assert offset % 16 == 0

        # Compression scale tracking
        compression_scales.append(len(encoded_stream) / elements_in_brick)

    # Track total length as last element of the offsets array
    compressed_offsets.append(offset)

    tens.weight_compression_scales = compression_scales
    tens.weight_compressed_offsets = compressed_offsets
    tens.compression_scale_for_worst_weight_stream = np.amax(compression_scales)
    tens.storage_compression_scale = tens.bandwidth_compression_scale = np.average(compression_scales)
    tens.compressed_values = encoded_streams
    tens.compressed_values_substream_offsets = encoded_streams_substream_offsets
    tens.brick_size = brick_size
    set_storage_shape(tens)
    nng.weight_cache.add(tens)


def calc_scales_and_pack_biases(tens, arch, ofm_depth_step, rescale_for_faf=False):
    assert tens.purpose == TensorPurpose.FeatureMap
    assert tens.format == TensorFormat.NHWC
    # the connected operator should expect a bias input unless it is a FullyConnected
    assert "Bias" in tens.consumer_list[0].type or tens.consumer_list[0].type.startswith("FullyConnected")
    # the input bias tensor is the same as that connected to the operator
    _, _, bias_tens, _ = tens.consumer_list[0].get_ifm_weights_biases_ofm()
    assert tens is bias_tens

    # the operator should only have a single output
    assert len(tens.consumer_list[0].outputs) == 1

    def pack_bias_and_scale(bias, scale, shift):
        bias = np.int64(bias)
        assert -(1 << (40 - 1)) <= bias < (1 << (40 - 1))  # signed 40-bit range
        assert 0 <= scale < (1 << 32)  # unsigned 32-bit range
        assert 0 <= shift < (1 << 6)  # unsigned 6-bit range

        # pack the 80 bit value = [0(2-bits),shift(6-bits),scale(32-bits),bias(40-bits)]
        data = bytearray(10)
        data[0] = (bias >> (0 * 8)) & 0xFF
        data[1] = (bias >> (1 * 8)) & 0xFF
        data[2] = (bias >> (2 * 8)) & 0xFF
        data[3] = (bias >> (3 * 8)) & 0xFF
        data[4] = (bias >> (4 * 8)) & 0xFF
        data[5] = (scale >> (0 * 8)) & 0xFF
        data[6] = (scale >> (1 * 8)) & 0xFF
        data[7] = (scale >> (2 * 8)) & 0xFF
        data[8] = (scale >> (3 * 8)) & 0xFF
        data[9] = shift & 0x3F
        return data

    biases = tens.quant_values

    first_consumer_op = tens.consumer_list[0]
    ifm_dtype = first_consumer_op.inputs[0].dtype
    ifm_scale = first_consumer_op.inputs[0].quantization.scale_f32
    ofm_scale = first_consumer_op.outputs[0].quantization.scale_f32
    weight_scales = first_consumer_op.inputs[1].quantization.scale_f32

    # biases can have multiple consumers for rnn cells. if so, then check that they are all the same
    for op in tens.consumer_list[1:]:
        assert ifm_scale == op.inputs[0].quantization.scale_f32
        assert ofm_scale == op.outputs[0].quantization.scale_f32
        assert weight_scales == op.inputs[1].quantization.scale_f32

    if not hasattr(weight_scales, "__iter__"):
        # If weight_scales is not already an iterable make it into a list
        weight_scales = [weight_scales]

    # Convert scales to np.double (from np.float32) to conform to TensorFlow Lite which
    # uses double during scaling calculations
    # TensorFlow Lite casts the scales slightly differently for uint8 and int8
    if not rescale_for_faf:
        if ifm_dtype == DataType.uint8:
            scales = [np.double(ifm_scale * weight_scale) / np.double(ofm_scale) for weight_scale in weight_scales]
        elif ifm_dtype == DataType.int8 or ifm_dtype == DataType.int16:
            scales = [
                (np.double(ifm_scale) * np.double(weight_scale)) / np.double(ofm_scale)
                for weight_scale in weight_scales
            ]
        else:
            raise UnsupportedFeatureError(
                "Compression of {} is not implemented; tensor: {}".format(ifm_dtype, tens.name)
            )
    else:
        if ifm_dtype == DataType.uint8:
            scales = [np.double(ifm_scale * weight_scale * 0x3000) for weight_scale in weight_scales]
        elif ifm_dtype == DataType.int8 or ifm_dtype == DataType.int16:
            scales = [(np.double(ifm_scale * 0x3000) * np.double(weight_scale)) for weight_scale in weight_scales]
        else:
            raise UnsupportedFeatureError(
                "Compression of {} is not implemented; tensor: {}".format(ifm_dtype, tens.name)
            )

    # quantise all of the weight scales into (scale_factor, shift)
    if ifm_dtype == DataType.int16:
        quantised_scales = [reduced_quantise_scale(scale) for scale in scales]
    else:
        quantised_scales = [quantise_scale(scale) for scale in scales]

    for _, shift in quantised_scales:
        assert shift >= 16

    # pack the biases and scales
    if len(quantised_scales) == 1:
        # If only 1 quantised scale is used, repeat that value for the length of the biases
        quantised_scales = [quantised_scales[0]] * len(biases)

    assert len(quantised_scales) == len(biases)
    tens.element_size_bytes = 10
    tens.compressed_values = []
    tens.compressed_values_substream_offsets = []

    total_elements = len(quantised_scales)
    alignment_bytes = 0
    for i in range(0, total_elements, ofm_depth_step):
        # Extract streams from brick to generate substreams for each core
        stream = bytearray()
        substream_offsets = [0]
        max_len = min(ofm_depth_step, total_elements - i)
        for core in range(0, min(arch.ncores, max_len)):
            core_scales = quantised_scales[i + core : i + core + max_len : arch.ncores]
            core_biases = biases[i + core : i + core + max_len : arch.ncores]
            for j, core_bias in enumerate(core_biases):
                stream.extend(pack_bias_and_scale(core_bias, *core_scales[j]))

            # Align to 16 for start for next substream
            remainder = (len(stream)) % 16
            if remainder > 0:
                stream.extend(bytearray(16 - remainder))
                alignment_bytes += 16 - remainder

            substream_offsets.append(len(stream))

        # Add to compressed values with their substream offset lists to the tensor
        tens.compressed_values.append(stream)
        tens.compressed_values_substream_offsets.append(substream_offsets)

    tens.storage_shape = [total_elements + round_up_divide(alignment_bytes, tens.element_size_bytes)]


def update_pass_weight_and_scale_tensors(nng, arch):
    for sg in nng.subgraphs:
        for ps in sg.passes:
            tens = ps.weight_tensor
            if tens is not None:
                op = tens.find_npu_op()
                npu_usage_of_tensor = op.attrs["npu_block_type"]
                needs_dma = tens.needs_dma()
                if ps.cascade.strategy == SchedulingStrategy.WeightStream and needs_dma:
                    ofm_depth_step = ps.block_config[-1]
                else:
                    ofm_depth_step = tens.shape[-1]
                compress_weights(
                    arch, nng, tens, npu_usage_of_tensor, ps.block_config[-1], ofm_depth_step, op.get_dilation_h_w()
                )
                # Update source tensor
                if needs_dma:
                    src_tens = tens.get_dma_src_tensor()
                    src_tens.shape = tens.shape
                    src_tens.quant_values = tens.quant_values
                    src_tens.copy_compressed_weight_info(tens)
                    set_storage_shape(src_tens)

            if ps.scale_tensor is not None:
                rescale_for_faf = False
                activation_ops = set(("Sigmoid", "Tanh"))
                if (ps.ops[-1].type in activation_ops) and (ps.npu_block_type != NpuBlockType.ElementWise):
                    rescale_for_faf = True
                calc_scales_and_pack_biases(ps.scale_tensor, arch, ofm_depth_step, rescale_for_faf)