I am working on a point light shadow mapping system which is working well but could probably be improved.

In particular, to upload the 'light view projection' matrices I'm using Queue.WriteBuffer, mainly because it's what I'm most familiar with.

However, this means I have to stop/restart rendering (ie: Create a new command encoder) for every shadow pass (ie: lightCount * 6 cube faces) and my understanding is that I should avoid creating 'too many' command encoders.

I can think of 2 alternatives to this:

1. Upload ALL LVP matrices in one hit into an intermmediate buffer, and then use CommandEncoder.CopyBuffer to copy each LVP matrix to uniform data per shadow pass. I'm guessing this is the right way to go as it means I only ever need to create 1 command encoder for everything, and there is only the CPU->GPU copy and the buffer 'chunk' copying may be more efficient on the GPU?

2. Create a unique bind group for each shadow pass and use SetBindGroup for each shadow pass (I'm actually sort of doing this already, I actually use 6 per-face boid groups so I only have to create lightCount command encoders). The downside to this approach is that it could potentially complicate the use of render bundles, something I'm not doing yet but would like to try at some point.

Am I missing another way? Another approach could involve using an array of LVP matrices, but then I've got the same problem of having to upload an 'index' uniform.

What's the best approach here?

Hey all,

I'm trying to do something like the following in WGSL but it doesn't seem to be allowed:

var myVec4 : vec4f;
myVec4.xy = vec2f(1.0, 0.0);

Am I missing some convenient way to do this ? ( For example there is not ternary operator but I can use select instead ). If not, it's not the end of the world but it would be a nice to have addition to the language.

Hey all,

I'm trying to do something like the following in WGSL but it doesn't seem to be allowed:

var myVec4 : vec4f;
myVec4.xy = vec2f(1.0, 0.0);

Am I missing some convenient way to do this ? ( For example there is no ternary operator but I can use select instead ). If not, it's not the end of the world but it would be a nice to have addition to the language.

In reply to @marksibly:matrix.org

I am working on a point light shadow mapping system which is working well but could probably be improved.

In particular, to upload the 'light view projection' matrices I'm using Queue.WriteBuffer, mainly because it's what I'm most familiar with.

However, this means I have to stop/restart rendering (ie: Create a new command encoder) for every shadow pass (ie: lightCount * 6 cube faces) and my understanding is that I should avoid creating 'too many' command encoders.

I can think of 2 alternatives to this:

1. Upload ALL LVP matrices in one hit into an intermmediate buffer, and then use CommandEncoder.CopyBuffer to copy each LVP matrix to uniform data per shadow pass. I'm guessing this is the right way to go as it means I only ever need to create 1 command encoder for everything, and there is only the CPU->GPU copy and the buffer 'chunk' copying may be more efficient on the GPU?

2. Create a unique bind group for each shadow pass and use SetBindGroup for each shadow pass (I'm actually sort of doing this already, I actually use 6 per-face boid groups so I only have to create lightCount command encoders). The downside to this approach is that it could potentially complicate the use of render bundles, something I'm not doing yet but would like to try at some point.

Am I missing another way? Another approach could involve using an array of LVP matrices, but then I've got the same problem of having to upload an 'index' uniform.

What's the best approach here?

In reply to @marksibly:matrix.org

I am working on a point light shadow mapping system which is working well but could probably be improved.

In particular, to upload the 'light view projection' matrices I'm using Queue.WriteBuffer, mainly because it's what I'm most familiar with.

However, this means I have to stop/restart rendering (ie: Create a new command encoder) for every shadow pass (ie: lightCount * 6 cube faces) and my understanding is that I should avoid creating 'too many' command encoders.

I can think of 2 alternatives to this:

1. Upload ALL LVP matrices in one hit into an intermmediate buffer, and then use CommandEncoder.CopyBuffer to copy each LVP matrix to uniform data per shadow pass. I'm guessing this is the right way to go as it means I only ever need to create 1 command encoder for everything, and there is only the CPU->GPU copy and the buffer 'chunk' copying may be more efficient on the GPU?

2. Create a unique bind group for each shadow pass and use SetBindGroup for each shadow pass (I'm actually sort of doing this already, I actually use 6 per-face boid groups so I only have to create lightCount command encoders). The downside to this approach is that it could potentially complicate the use of render bundles, something I'm not doing yet but would like to try at some point.

Am I missing another way? Another approach could involve using an array of LVP matrices, but then I've got the same problem of having to upload an 'index' uniform.

What's the best approach here?