5.0.2 mul(matrix, vector) function throws an error

More
9 months 2 weeks ago #1 by Fu-Bama 5.0.2 mul(matrix, vector) function throws an error was created by Fu-Bama
In 5.0.2 ReShade (dxgi), function mul(matrix3x2, vector3) throws an error, whereas it worked as expected in 4.9.1, resulting in vector2.
error X3004: undeclared identifier or no matching intrinsic overload for 'mul'

Please Log in or Create an account to join the conversation.

More
9 months 2 weeks ago - 9 months 2 weeks ago #2 by crosire Replied by crosire on topic 5.0.2 mul(matrix, vector) function throws an error
What you are trying to do is mathematically impossible and ReShade 5.0 now detects this (this didn't work in 4.9.1 either, the compiler just didn't always issue an error and instead did funky stuff like trying to reinterpret the 3x2 matrix as 2x3 which doesn't do what you'd expect or want).
A matrix multiplication of 3x2 * 3x1 is not defined, but 1x3 * 3x2 is, so you can solve this by switching the arguments to mul(vector3, matrix3x2).
Last edit: 9 months 2 weeks ago by crosire.
The following user(s) said Thank You: Fu-Bama

Please Log in or Create an account to join the conversation.

More
9 months 2 weeks ago - 9 months 2 weeks ago #3 by Fu-Bama Replied by Fu-Bama on topic 5.0.2 mul(matrix, vector) function throws an error
Matrix multiplication would work if vector3 was interpreted as 1x3 matrix, not 3x1. Such that mul(matrix3x2, vector3) = vector2 results in 3x2 * 1x3 multiplication.
Code for that would look something like this:
float2 mul(float3x2 mat, float3 vec) { return float2(dot(mat[0], vec), dot(mat[1], vec)); }
Last edit: 9 months 2 weeks ago by Fu-Bama.

Please Log in or Create an account to join the conversation.

More
9 months 2 weeks ago - 9 months 2 weeks ago #4 by crosire Replied by crosire on topic 5.0.2 mul(matrix, vector) function throws an error
No?. For matrix multiplication to be valid the number of columns in the first matrix must be equal to the number of rows in the second matrix (see also en.wikipedia.org/wiki/Matrix_multiplication ). With 3x2 * 1x3, that is not the case either (2 != 1).

You are probably thinking of 2x3 * 3x1. That's valid, can call that via mul(matrix2x3, vector3). Or as said do mul(vector3, matrix3x2).
Last edit: 9 months 2 weeks ago by crosire.

Please Log in or Create an account to join the conversation.

More
9 months 2 weeks ago - 9 months 2 weeks ago #5 by Fu-Bama Replied by Fu-Bama on topic 5.0.2 mul(matrix, vector) function throws an error
Thanks for the response. Column*row or row*column is an arbitrary choice, depending on convention. It's called row-major or column-major order.

The following:
mul(matrix3x3, vector3).yz
gives same result as proposed:
mul(matrix3x2, vector3)
,
where
matrix3x3[1] = matrix3x2[0]
matrix3x3[2] = matrix3x2[1]

This is how I fixed the error, using matrix3x3 and swizzle.
Since it works, I propose convention, where matrix vector components (accessed by index) represent rows. And vector on the right side of mul function is transposed to colum-vector. Therefore mul operation performs row*column multiplication, resulting in an column vector. At first glace it should go along DirectX convention, but I have not confirmed that.[/code][/code][/code]
Last edit: 9 months 2 weeks ago by Fu-Bama.

Please Log in or Create an account to join the conversation.

More
9 months 2 weeks ago - 9 months 2 weeks ago #6 by crosire Replied by crosire on topic 5.0.2 mul(matrix, vector) function throws an error
I'm aware, but it's not a choice you have in ReShade: In ReShade FX all matrices are row-major by definition. That's it. Their storage format cannot be changed to be column-major, like in HLSL with "row_major" and "column_major" keywords (notice those keywords are intentionally missing in ReShade FX). This is done to simplify translation between the various APIs and uniform storage updates. Similarly all matrices are logically expected to be row-major in operations and as such only those intrinsics that make sense for row-major matrices/vectors are available.
Last edit: 9 months 2 weeks ago by crosire.

Please Log in or Create an account to join the conversation.

We use cookies

We use cookies on our website. Some of them are essential for the operation of the site, while others help us to improve this site and the user experience (tracking cookies). You can decide for yourself whether you want to allow cookies or not. Please note that if you reject them, you may not be able to use all the functionalities of the site.