Rework of the shape operations. Shape operations are now done in shape specific
operators rather than being based on type.

shape_t is reworked to a list of size_t values.

Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: I2fca0728f9caa6a6fc34a8ce9e389bb581eea959

Kevin Petit authored Jan 25, 2024 and Eric Kunze committed Jan 28, 2024

Signed-off-by: Kevin Petit <kevin.petit@arm.com>
Change-Id: Icb824233e5be84ac0eb693459b65441e9ae6c545

Be explicit that the value is converted to int32 and then clamped to the
output type range.

Also remove use of apply_clip, replacing with either apply_clip_s or apply_clip_u

Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: I0efc7f4d653c1bef13c9989cfe2647208dad9617

Signed-off-by: Kevin Petit <kevin.petit@arm.com>
Change-Id: Ieb895e834d1f3b512a666b062b963b08f82c8a24

Signed-off-by: Kevin Petit <kevin.petit@arm.com>
Change-Id: Ifc17c8ab508a0d27f5f832fa76f7dc9c1ff57dbd

Signed-off-by: Kevin Petit <kevin.petit@arm.com>
Change-Id: I7cad7d2bc6212e1765fbe3027abfb1fe37111445

Change either the name in the pseudocode or XML to what feels more
"natural" or in line with other related names.

Change-Id: Ic12a0418a80642c548c288b6ba26a4603b94fedc
Signed-off-by: Kevin Petit <kevin.petit@arm.com>

Change-Id: Ic362e5cbc607d1b7560c09326fcebcba606454ce
Signed-off-by: Kevin Petit <kevin.petit@arm.com>

- Some operators were using True and False.
- Document the use of true and false in the definition of
  the boolean number format.

Signed-off-by: Kevin Petit <kevin.petit@arm.com>
Change-Id: Ic4c16c8527d46ca01d22e3c70c5371821a726df8

Thomas Preudhomme authored Jan 12, 2024 and Eric Kunze committed Jan 19, 2024

Axis is a dimension index of the input tensor, but is used to index the
output tensor.

Signed-off-by: Thomas Preud'homme <thomas.preudhomme@arm.com>
Change-Id: Ie50f3921153e01437fe763b5d2b6cfaa06049f25

The semicolons had no semantic meaning.

Signed-off-by: Kevin Petit <kevin.petit@arm.com>
Change-Id: I9447ea24c82a32c8bfe07fcc560f34a696e59831

- Add missing parenthesis in ARITHMETIC_SHIFT_RIGHT
- Add missing semicolons
- Remove stray dot in VARIABLE_WRITE

Change-Id: I28ab7c80c735995437588fffeffb1f12d112f3ec
Signed-off-by: Kevin Petit <kevin.petit@arm.com>

This makes it easier to process the pseudocode automatically.

Change-Id: I84394192598e589de07d43a7af60b96788e14f86
Signed-off-by: Kevin Petit <kevin.petit@arm.com>

TANH can have cancellation between exp(x) and exp(-x)
for small abs(x). To allow for this, a lower bound of
0.5*exp(-normal_fraction<in_out_t>) is set for the
error bound. The TANH formula also can contain 4
exp() values, so the error bound is increased to 4
multiplied by the exp() error bound.

SCATTER output size K must be as least as great
as the number of indices W, as each index is distinct.

Change-Id: I8d1f1c866bd59d8140cf7db76404f5006a04a3f3
Signed-off-by: Dominic Symes <dominic.symes@arm.com>

i48_t is 48-bit, not 32-bit.

Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: Icd602931ffb5d74842359fae2a8885e96b85bf59

Update the REDUCE_PRODUCT conformance description
to be consistent with EXP, POW, TANH, SIGMOID.

Signed-off-by: Dominic Symes <dominic.symes@arm.com>
Change-Id: I083c41ce09e70b51d07cfea30a4bbc4dff4a8657

Cannot raise a negative base to a non-integral power.

Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: I851006abc30bccb612b2a41e27f5c259bb221e41

Add i4_t and i48_t to the list of types
supported by the IDENTITY operator.

Signed-off-by: Dominic Symes <dominic.symes@arm.com>
Change-Id: Id0eeb5f665345e6fdb975b98af4926efc6412afc

Change SIGMOID, TANH precision to allow an
implementation using EXP. The bound is based
on the EXP bound.

Change-Id: I19582b31753ac614a1064a1b81961dbc54acf823
Signed-off-by: Dominic Symes <dominic.symes@arm.com>

For bf16_t data type, REDUCE_SUM is changed to use fp32_t
as the accumulator type to be consistent with CONV2D.
For other data types the accumulator type is the same
as in_out_t (and so no change).
Also correct the rank limit text.

Update reduction pseudo-code to be consistent with
REDUCE_SUM.

Change-Id: I9923066be7d1b7edb0efd9bcf3365b4af9501beb
Signed-off-by: Dominic Symes <dominic.symes@arm.com>

Change RSQRT precision to 2 ulp to allow
unfused square root and reciprocal.

Also fixes:
- a typo in EXP and POW conformance
- exp2() handling of large negative values
- symmetry about 0 of test set S=4 data generation
- err_bnd cannot be negative
- ulp not taken for reference value of 0

Change-Id: Idaeeb7b615f1634e8e09dea5f82827039780b462
Signed-off-by: Dominic Symes <dominic.symes@arm.com>

Previous description referred to nonexistant input tensor

Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: I162ab05ed1db7491ed29ddeabd3490e9ff7ee39b

- CustomOp's domain attribute name has been updated to domain_name
- Forget to update that for the description text part.

Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: Ifa731125dc2f224ecc1cd661e238c7b27945db68

Jerry-Ge authored Oct 31, 2023 and Eric Kunze committed Nov 01, 2023

- "operator" is a reserved keyword in C++
- Update the attribute names to avoid those conflicts

Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: Iecb1edc50eb1a1232e05250b8baa79cfceb8cd61

EXP and POW are complex to implement in a precise number of ulp
and a more flexible precision is given.

Additionally, the test generater S=5 has the bias values
changed to zero. If the bias values are non-zero then a
bias for all output tensor elements of the channel can affect
the dot product bias test.

The numeric accuracy helpers are updated for bf16.

Signed-off-by: Dominic Symes <dominic.symes@arm.com>
Change-Id: Ia46a2ef4d577244b6983a08ce850de3db9573a42

Bias broadcast was introduced earlier, this adds text to the bias argument
to call out the broadcast when only a single data element is given for
broadcast.

Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: I684afbe642e9b9cfb9f7c7e9b5bb4944019f206a

Clarify how to compare with an error range of a given
number of ulp in this specification. Denormals may be
flushed to zero.

Signed-off-by: Dominic Symes <dominic.symes@arm.com>
Change-Id: I1158e777030c4770d74f1acff84ab98e4a3420ac

REDUCE_MIN, REDUCE_MAX, SELECT did not have an accuracy bound defined.

Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: I0fea091752dfcdc2bb55eae7becc68b9c1860e9e

The output shape can be calculated from the
input shape and attributes so the out_shape
argument is redundant.

Change-Id: I7e6b0b7f03dc75ccdf967261fad4185c513ed187
Signed-off-by: Dominic Symes <dominic.symes@arm.com>

Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: I323bf261ac4a5219ee881a31e2840356ba32ab9a

Also express the fp32_t acc_min_normal calculation
without a double negation.

Signed-off-by: Dominic Symes <dominic.symes@arm.com>
Change-Id: I34ba274264bc511e60f0e01601d5d0a3f983e594

Define idiv_floor to give equivalent behavior to the floating-point
floor function for image coordinate calculation.

Change-Id: Id6268794b1e3ce5cc1114bda74dd06b892457a8e
Signed-off-by: Eric Kunze <eric.kunze@arm.com>

Update the error bound for dot product operations to
allow floating point accumulators where denormal
values are flushed to zero.

Change-Id: I1d5805f0ad613ca9d78c6a6da12e31d7df5f6cc4
Signed-off-by: Dominic Symes <dominic.symes@arm.com>

CUSTOM operators are still implementation specific, but now have
attributes added to identify and namespace the underyling operation.

Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: I8db2fffd0b34958bd8c718633a130941f32f962b

Integer inputs and outputs to TOSA operators are now defined as signless
values. In most instances the operator will used signed arithmetic as
indicated in previous versions of the specification resulting in little
functional change to the specification.

New attributes have been added to the RESCALE operator to indicate
whether the input and output values should be treated as signed or unsigned.

Explicit use of static_cast, sign_extend, zero_extend and truncate are added
to the pseudocode to avoid ambiguity.

Change-Id: I71c67d3e5aeaabc418c768f821fce6ee3eebb65b

Adds a local_bound flag to convolution operations.

If the local_bound flag is true then the output accuracy
for floating-point is measured relative to the
input tensor data local to the particular convolution output.

If the local_bound flag is false then the output accuracy
for floating-point is measured relative to the whole input tensor.
This accuracy measure is the default and more approapriate for
implementations with non local optimizations (such as
transform based fast convolutions).

Signed-off-by: Dominic Symes <dominic.symes@arm.com>
Change-Id: I64e3e4981a63e26e6391149e28d5d71e7ef5560a

Shape inference derives the shape of tensors in
the graph from input shapes. Operations such as RESHAPE
may need calculations to derive the new tensor shape.
This patch:

- Adds a DIM operator to get the size of a tensor in
  a given axis as a rank 0 tensor of type shape_t

- Allows RESHAPE to take a 1D shape tensor as input for
  the new shape

- Allows RESIZE, TILE, PAD to take input sizes based
  on shape tensors.

- Allows ADD, SUB, MUL, INTDIV to operate on rank 0
  shape_t tensors

- Allows CONCAT to concatenate 0D shape_t tensors to
  a 1D shape_t tensor

- Adds CONST support for shape_t tensors

In this version of the specification shape tensors must
be resolvable to constants at backend compile time.

Signed-off-by: Dominic Symes <dominic.symes@arm.com>
Change-Id: I484bd44452453b5e05d0d8a82689564587b224e4

We've seen networks with >64 scale ratios that we would expect
to run on systems implementing the 8K level. Raise the ratio to 256.

Change-Id: I5a72a533b2125fe56ffedc8b8d5c31bb96f60f10
Signed-off-by: Eric Kunze <eric.kunze@arm.com>

Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: I63a4c1202a1eddcdedb222e64cac34557647ff21

Bias tensors often have the same value for
each channel. To allow for this the number of
bias channels, BC, is permitted to be 1 or the
output channel size. If BC == 1 then the
bias is broadcast.

Change-Id: I3262a6c699bd045f44201e2a123f9b792b9f67c9
Signed-off-by: Dominic Symes <dominic.symes@arm.com>