[glsl-out] Invalid GLSL emission when using firstLeadingBit and others

[hanawatson]

Apologies for the lack of clarity in #1946, but I didn't mention specifically that the GLSL in that issue was emitted by naga after being provided valid WGSL. I assumed it was a glsl-in problem, but it seems I was incorrect - sorry for the confusion! I'm not sure that the GLSL mentioned in the linked issue should be being output by naga, as it's invalid (as answered in the issue).

WGSL:

@stage(compute) @workgroup_size(1)
fn main() {
    var var0 = firstLeadingBit(abs(0u));
}

GLSL output by naga (included in linked issue, also applies for 450 core):

#version 440 core
#extension GL_ARB_compute_shader : require
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;

void main() {
    uint var0_ = findMSB(abs(0u));
}

Error when validating above GLSL with naga (included in linked issue):

error: Unknown function 'findMSB'
  ┌─ temptest.comp:6:18
  │
6 │     uint var0_ = findMSB(abs(0u));
  │                  ^^^^^^^^^^^^^^^^

[JCapucho]

This actually a problem with wgsl-in since abs is defines as

@const fn abs(e: T) -> T

where T is a scalar or a vector of scalars, but firstLeadingBit is defined as

@const fn firstLeadingBit(e: T) -> T

where T is an i32 or vecN<i32>
Nevermind it's also defined for u32

[JCapucho]

I've looked at the other issue referenced and there are mentions of other functions causing the same problems but those definitely only support floats, could you provide some reproductions cases with the other functions like exp?

[hanawatson]

I've looked at the other issue referenced and there are mentions of other functions causing the same problems but those definitely only support floats, could you provide some reproductions cases with the other functions like exp?

Apologies for that - those other functions did cause the same problems in GLSL, but you pointed out that was expected behavior as it was incorrect GLSL. The same behavior doesn't occur in the GLSL output by naga when using e.g. exp, since it is invalid WGSL typing as you say (e.g. exp outputting a float, findLeadingBit expecting an integer) and therefore no GLSL is actually output.

If exp(0.0) is used instead of abs(0u) in the WGSL above, the error message is thrown by naga when validating the WGSL and is:

error: Incompatible operands: FindMsb(Scalar { kind: Float, width: 4 })

Could not parse WGSL