Re: [PATCH 3/3] d2d1: Reuse PS constant buffer object.

[Henri Verbeet]

On Tue, 8 Jun 2021 at 10:57, Nikolay Sivov  wrote:
> -HRESULT d2d_brush_get_ps_cb(struct d2d_brush *brush, struct d2d_brush *opacity_brush,
> -
>    BOOL outline, BOOL is_arc, struct d2d_device_context *render_target, ID3D10Buffer **ps_cb)
>
> +HRESULT d2d_brush_init_ps_cb_data(struct d2d_brush *brush, struct d2d_brush *opacity_brush,
> -
>    BOOL outline, BOOL is_arc, struct d2d_ps_cb *cb_data)
>
> {
> - D3D10_SUBRESOURCE_DATA buffer_data;
> - struct d2d_ps_cb cb_data = {0};
> - D3D10_BUFFER_DESC buffer_desc;
> - HRESULT hr;
> -
> - cb_data.outline = outline;
> - cb_data.is_arc = is_arc;
> - if (!d2d_brush_fill_cb(brush, &cb_data.colour_brush))
> - cb_data->outline = outline;
> - cb_data->is_arc = is_arc;
> - if (!d2d_brush_fill_cb(brush, &cb_data->colour_brush))
>     return E_NOTIMPL;
> - if (!d2d_brush_fill_cb(opacity_brush, &cb_data.opacity_brush))
> - if (!d2d_brush_fill_cb(opacity_brush, &cb_data->opacity_brush))
>     return E_NOTIMPL;
> - buffer_desc.ByteWidth = sizeof(cb_data);
> - buffer_desc.Usage = D3D10_USAGE_DEFAULT;
> - buffer_desc.BindFlags = D3D10_BIND_CONSTANT_BUFFER;
> - buffer_desc.CPUAccessFlags = 0;
> - buffer_desc.MiscFlags = 0;
> -
> - buffer_data.pSysMem = &cb_data;
> - buffer_data.SysMemPitch = 0;
> - buffer_data.SysMemSlicePitch = 0;
> -
> - if (FAILED(hr = ID3D10Device_CreateBuffer(render_target->d3d_device, &buffer_desc, &buffer_data, ps_cb)))
> -
>    ERR("Failed to create constant buffer, hr %#x.\n", hr);
> -
> - return hr;
> - return S_OK;
>
> }

Arguably, what's left of the function could just be inlined in
d2d_device_context_update_ps_cb() now.
> @@ -266,6 +266,8 @@ static ULONG STDMETHODCALLTYPE d2d_device_context_inner_Release(IUnknown *iface)
>          ID3D10Buffer_Release(context->vb);
>          ID3D10Buffer_Release(context->ib);
>          ID3D10PixelShader_Release(context->ps);
> -
>    if (context->ps_cb)
> -
>        ID3D10Buffer_Release(context->ps_cb);
>    for (i = 0; i < D2D_SHAPE_TYPE_COUNT; ++i)
>    {
>        ID3D10VertexShader_Release(context->shape_resources[i].vs);

If we created "context->ps_cb" in d2d_device_context_init(), this
would become an unconditional release.
> +static HRESULT d2d_device_context_update_ps_cb(struct d2d_device_context *context,
> -
>    struct d2d_brush *brush, struct d2d_brush *opacity_brush, BOOL outline, BOOL is_arc)
>
> +{
> - D3D10_SUBRESOURCE_DATA buffer_data;
> - struct d2d_ps_cb cb_data = {0};
> - D3D10_BUFFER_DESC buffer_desc;
> - void *data;
> - HRESULT hr;
> -
> - if (FAILED(hr = d2d_brush_init_ps_cb_data(brush, opacity_brush, outline, is_arc, &cb_data)))
> - {
> -
>    WARN("Failed to initialize brush constant buffer data, hr %#x.\n", hr);
> -
>    return hr;
> - }
> -
> - if (context->ps_cb)
> - {
> -
>    if (SUCCEEDED(hr = ID3D10Buffer_Map(context->ps_cb, D3D10_MAP_WRITE_DISCARD, 0, &data)))
> -
>    {
> -
>        memcpy(data, &cb_data, sizeof(cb_data));
> -
>        ID3D10Buffer_Unmap(context->ps_cb);
> -
>    }
> -
>    else
> -
>        ERR("Failed to map constant buffer, hr %#x.\n", hr);
> - }

We don't need the memcpy(). I.e., we can just do the following:

struct d2d_ps_cb *cb_data;
    [...]
    ID3D10Buffer_Map(..., &cb_data);
    d2d_brush_init_ps_cb_data(..., cb_data);
    ID3D10Buffer_Unmap(...);
> - else
> - {
> -
>    buffer_desc.ByteWidth = sizeof(cb_data);
> -
>    buffer_desc.Usage = D3D10_USAGE_DYNAMIC;
> -
>    buffer_desc.BindFlags = D3D10_BIND_CONSTANT_BUFFER;
> -
>    buffer_desc.CPUAccessFlags = D3D10_CPU_ACCESS_WRITE;
> -
>    buffer_desc.MiscFlags = 0;
> -
> -
>    buffer_data.pSysMem = &cb_data;
> -
>    buffer_data.SysMemPitch = 0;
> -
>    buffer_data.SysMemSlicePitch = 0;
> -
> -
>    if (FAILED(hr = ID3D10Device_CreateBuffer(context->d3d_device, &buffer_desc, &buffer_data,
> -
>            &context->ps_cb)))
> -
>    {
> -
>        ERR("Failed to create constant buffer, hr %#x.\n", hr);
> -
>    }
> - }
> -
> - return hr;
>
> +}
> -

And this could just happen during d2d_device_context_init(). (The
"buffer_data" argument is optional; it can be convenient for
initialising resources without explicitly mapping them, but I don't
think it provides much of an advantage here.)
> @@ -1676,34 +1707,19 @@ static void STDMETHODCALLTYPE d2d_device_context_Clear(ID2D1DeviceContext *iface
>          return;
>      }
> - ps_cb_data.outline = FALSE;
> - ps_cb_data.colour_brush.type = D2D_BRUSH_TYPE_SOLID;
> - ps_cb_data.colour_brush.opacity = 1.0f;
> - c = &ps_cb_data.colour_brush.u.solid.colour;
> - memset(&brush, 0, sizeof(brush));
> - brush.type = D2D_BRUSH_TYPE_SOLID;
> - brush.opacity = 1.0f;
> - c = &brush.u.solid.color;
> if (colour)
>     *c = *colour;
> if (render_target->desc.pixelFormat.alphaMode == D2D1_ALPHA_MODE_IGNORE)
>     c->a = 1.0f;
> - c->r *= c->a;
> - c->g *= c->a;
> - c->b *= c->a;
> -

Constructing a brush like that is a little awkward, I think. To some
extent that's also true of constructing the d2d_ps_cb structure in the
original code, but brush objects are a bit more complex than the
d2d_ps_cb structure. It would seem preferable to either directly map
"ps_cb" here, and generate its contents in the existing way, or to
create a proper brush object using d2d_solid_color_brush_create().