i need to do this because vulkan for some reason wants shader code as a [*:0]const u8
#how can i safely cast a `*const [N:0]u8` to `[]const u32`?
pine vector
actually, it might not need to be null terminated since you pass in the size of the code too
worn ice
This is not really possible, since the alignment of u8 is smaller than the alignment of u32. How do you get the u8 pointer variable
pine vector
from @embedFile
the question should probably be []const u8 to [*:0]const u32, the shader code doesnt need to be null terminated (and cannot be)
how can i safely cast a []const u8 to [*:0]const u32?
worn ice
you can do
const aligned_data align(@alignOf(u32)) = @embedFile("file").*;
const u32_ptr: *const [@divExact(aligned_data.len, 4)]u32 = @ptrCast(&aligned_data);
Other way around: @embedFile always returns a null terminated ptr, but for vulkan the shader does not need to be null terminated.
pine vector
you can do
```
const aligned_data align(@alignOf(u32)) = @embedFile("file").*;
...
what would be the type of aligned_data?
worn ice
[N]u8, where N is the comptime known len of your shader file
pine vector
oh so then could i just pass in a slice for a function?
worn ice
afaik your function needs a u32 slice and not a aligned u8 slice
I don't know the complete vulkan function interface, but maybe @ptrCast(&aligned_data); is possible
worn ice
Here is an example: https://github.com/Snektron/vulkan-zig/blob/master/examples/triangle.zig. See the vert_spv and frag_spv vars and how they are used.
pine vector
src\main.zig:524:62: error: TODO: implement @ptrCast between slices changing the length
const vertex_shader_module = self.createShaderModule(@ptrCast(&vertex_shader_code));
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
interesting error message
that vulkan binding might be doing other stuff
worn ice
then do @as(*const [@divExact(vertex_shader_code.len, 4)]u32, @ptrCast(&vertex_shader_code))
pine vector
did that, results in seg fault
worn ice
which os?
pine vector
windows
pine vector
how can i safely cast a *const [N:0]u8 to []const u32?
pine vector
bump
shrewd ivy
bump
does this work?
const bytes: []const u8 = @embedFile("my_file");
const data: [@divExact(bytes.len, 4)]u32 = @bitCast(bytes[0..bytes.len].*);
_ = &data; // <- your `[]const u32`
src\main.zig:524:83: error: @bitCast size mismatch: destination type '[376]u32' has 12032 bits but source type '[1504:0]u8' has 12040 bits
const vertex_shader_bytecode: [@divExact(vertex_shader_code.len, 4)]u32 = @bitCast(vertex_shader_code[0..vertex_shader_code.len].*);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
oh thats 8 bits off
so kinda close?
pine vector
oh i did see that
shrewd ivy
we need to rid of the sentinel terminator
pine vector
compiles, but is UB
shrewd ivy
UB how?
pine vector
seg fault
shrewd ivy
that's in a differnet part of your program then.
the whole calculation is done at compile time, no UB comes from there.
pine vector
If pCode is a pointer to SPIR-V code, pCode must adhere to the validation rules described by the Validation Rules within a Module section of the SPIR-V Environment appendix
hmm yeah this is what the validation layer says