[MLIR][Quant] Fix equations in Quantization.md

This patch fixes the equations on the Quantization page
(https://mlir.llvm.org/docs/Quantization/).

I don't know what caused the equations to be broken, it
might be https://github.com/llvm/mlir-www/pull/152, but
I'm not sure. Irregardless, let's just fix it and be
done with it.

I've fixed the equations by moving some subscripts to
the text. For some reason, the large number of subscripts
caused Mathjax to fail. I've also tried KaTeX, which
failed at exactly the same number of subscripts.

The workflow to inspect the fix is as follows:

```
$ git clone --depth=1 https://github.com/llvm/mlir-www.git /some/path/mlir-www

$ git clone --depth=1 https://github.com/llvm/llvm-project.git /some/path/llvm-project

$ cp /some/path/llvm-project/mlir/docs/Quantization.md \
    /some/path/mlir-www/website/content/Quantization.md

$ cd /some/path/mlir-www/website

$ hugo serve
[...]
Web Server is available at http://localhost:1313/ (bind address 127.0.0.1)
Press Ctrl+C to stop
```
and view the page at http://localhost:1313/Quantization/.

Reviewed By: stellaraccident

Differential Revision: https://reviews.llvm.org/D152651
This commit is contained in:
rikhuijzer 2023-06-12 13:28:39 +02:00
parent dff3ebc8de
commit fc560cdb46

View File

@ -44,8 +44,8 @@ previous example, when $ scale = \pi $, the maximum rounding error will be $
Multiplication can be performed on scaled values with different scales, using
the same algorithm as multiplication of real values (note that product scaled
value has $$ scale_{product} = scale_{left \mbox{ } operand} * scale_{right
\mbox{ } operand} $$). Addition can be performed on scaled values, so long as
value has $ scale_{product} = scale_{left \mbox{ } operand} * scale_{right
\mbox{ } operand} $). Addition can be performed on scaled values, so long as
they have the same scale, using the same algorithm for addition of real values.
This makes it convenient to represent scaled values on a computer as signed
integers, and perform arithmetic on those signed integers, because the results
@ -115,17 +115,23 @@ not required that all representable values of the integral type are used):
$$
\begin{align*}
af&fine\\_value_{uint8 \\, or \\, uint16} \\\\
&= clampToTargetSize(roundToNearestInteger( \frac{real\\_value_{Single}}{scale_{Single}})_{sint32} + zero\\_point_{uint8 \, or \, uint16})
af&fine\\\_value \\\\
&= clampToTargetSize(roundToNearestInteger( \frac{real\\\_value}{scale}) + zero\\\_point \\\\
\end{align*}
$$
In the above, we assume that $real\\_value$ is a Single, $scale$ is a Single,
$roundToNearestInteger$ returns a signed 32-bit integer, and $zero\\_point$
is an unsigned 8-bit or 16-bit integer. Note that bit depth and number of fixed
point values are indicative of common types on typical hardware but is not
constrained to particular bit depths or a requirement that the entire range of
an N-bit integer is used.
where we assume the following types:
- `real_value`: Single
- `scale`: Single
- `roundToNearestInteger`: returns a 32-bit integer
- `zero_point`: 8-bit or 16-bit integer
- `affine_value`: 8-bit or 16-bit integer
Note that bit depth and number of fixed point values are indicative
of common types on typical hardware but is not constrained to
particular bit depths or a requirement that the entire range of an
N-bit integer is used.
#### Affine to real
@ -136,13 +142,19 @@ can be performed:
$$
\begin{align*}
re&al\\_value_{Single} \\\\
&= roundToNearestFloat((affine\\_value_{uint8 \\, or \\, uint16} - zero\\_point_{uint8 \\, or \\, uint16})_{sint32})_{Single} * scale_{Single}
re&al\\\_value \\\\
&= roundToNearestFloat(affine\\\_value - zero\\\_point) * scale
\end{align*}
$$
In the above, we assume that the result of subtraction is in 32-bit signed
integer format, and that $roundToNearestFloat$ returns a Single.
where we assume the following types:
- `real_value`: Single
- `scale`: Single
- `affine_value`: 8-bit or 16-bit integer
- `zero_point`: 8-bit or 16-bit integer
- `roundToNearestFloat`: returns a Single
- `-` (subtraction): returns a 32-bit signed integer
#### Affine to fixed point
@ -151,7 +163,7 @@ from the affine value to get the equivalent fixed point value.
$$
\begin{align*}
scaled\\_value = affine\\_value_{non\mbox{-}negative} - zero\\_point_{non\mbox{-}negative}
scaled\\\_value = affine\\\_value_{non\mbox{-}negative} - zero\\\_point_{non\mbox{-}negative}
\end{align*}
$$
@ -162,7 +174,7 @@ fixed point value to get the equivalent affine value.
$$
\begin{align*}
affine\\_value_{non\mbox{-}negative} = scaled\\_value + zero\\_point_{non\mbox{-}negative}
affine\\\_value_{non\mbox{-}negative} = scaled\\\_value + zero\\\_point_{non\mbox{-}negative}
\end{align*}
$$