[PATCH 3/5] wined3d: Optimize scanning changed shader constants in wined3d_device_apply_stateblock().

[Matteo Bruni]

On Tue, Feb 11, 2020 at 5:33 PM Henri Verbeet <hverbeet at gmail.com> wrote:
>
> On Mon, 10 Feb 2020 at 23:06, Matteo Bruni <mbruni at codeweavers.com> wrote:
> > +typedef HRESULT (CDECL *wined3d_device_shader_constant_setter)(struct wined3d_device *device,
> > +        unsigned int start_idx, unsigned int count, const void *constants);
> > +
> > +static void device_apply_shader_constants(struct wined3d_device *device,
> > +        const struct wined3d_stateblock_state *state,
> > +        DWORD *bitmap, unsigned int bit_count, const void *data, unsigned int stride,
> > +        wined3d_device_shader_constant_setter shader_constant_setter)
> > +{
> > +    const unsigned int word_bit_count = sizeof(DWORD) * CHAR_BIT;
> > +    unsigned int i, j, idx, start, last;
> > +    const BYTE *byte_data = data;
> > +    DWORD map;
> > +
> > +    start = last = ~0u;
> > +    for (i = 0; i < (bit_count + word_bit_count - 1) / word_bit_count; ++i)
> > +    {
> > +        map = bitmap[i];
> > +
> > +        if (map == ~0u)
> > +        {
> > +            if (last != ~0u && last != i * word_bit_count - 1)
> > +            {
> > +                shader_constant_setter(device, start, last - start + 1, &byte_data[start * stride]);
> > +                start = i * word_bit_count;
> > +            }
> > +            if (start == ~0u)
> > +                start = i * word_bit_count;
> > +            last = i * word_bit_count + word_bit_count - 1;
> > +            continue;
> > +        }
> > +        while (map)
> > +        {
> > +            j = wined3d_bit_scan(&map);
> > +            idx = i * word_bit_count + j;
> > +
> > +            if (start == ~0u)
> > +            {
> > +                start = last = idx;
> > +            }
> > +            else if (last != idx - 1)
> > +            {
> > +                shader_constant_setter(device, start, last - start + 1, &byte_data[start * stride]);
> > +                start = last = idx;
> > +            }
> > +            else
> > +            {
> > +                last = idx;
> > +            }
> > +        }
> > +    }
> > +    if (start != ~0u)
> > +        shader_constant_setter(device, start, last - start + 1, &byte_data[start * stride]);
> > +}
> > +
> This looks like it's a fair bit more complicated than it needs to be.
> I think the primitives you want are something like the following:
>
>     unsigned int wined3d_bitmap_ffs(uint32_t *bitmap, unsigned int
> start, unsigned int count);
>     unsigned int wined3d_bitmap_ffz(uint32_t *bitmap, unsigned int
> start, unsigned int count);
>
> and then you can extract the ranges relatively trivially.

Oh, interesting. I'll update the patch along those lines and see what
the end result looks like.