Bad codegen for inline functions, it just doesn't work

[lassade]

Zig Version

0.12.0-dev.3282+da5b16f9e

Steps to Reproduce and Observed Behavior

Please give a the code in action here: https://godbolt.org/z/6T1edKa94

In the link above you will see these 4 functions. Each pair of function should do the exact same thing. First function contains the manually inlined code that I copied and pasted just to make sure it's really inlined and in the other I called the function.

The problem is that the assembly of the versions where the function is used are just crazy, and do way more stuff that it really should.

I'm using -Doptimize=ReleaseFast

export fn inlinePtrTest(item: *Ptr) void {
    @as(*Id, @alignCast(@ptrCast(item.inner + @offsetOf(Block.Inner, "id")))).* = .{ 0 };
    @as(*Pos, @alignCast(@ptrCast(item.inner + @offsetOf(Block.Inner, "pos")))).* = .{ @splat(0.0) };
}

export fn fnPtrTest(item: *Ptr) void {
    item.field(.id).* = .{ 0 };
    item.field(.pos).* = .{ @splat(0.0) };
}

export fn directBlock(block: *Block) void {
    _ = Ptr{ .inner = @alignCast(@ptrCast(&block.inner.pos[0])) };
}

export fn indirectBlock(block: *Block) void {
    _ = block.index(0);
}

Expected Behavior

Both assemblies of each pair should be identical

[mlugg]

-Doptimize=ReleaseFast is an option you pass to zig build when a build script is using std.Built.standardOptimizeOption. When directly building a binary with build-exe/build-obj/build-lib, which is what Godbolt does, you need to use -OReleaseFast instead. The option you're passing isn't doing anything helpful (it's just defining a C macro!).

When you pass the correct optimization option, the codegen is identical for the two functions: https://godbolt.org/z/cKWTdTYWf

[lassade]

Sorry I got skilled issued