Drop the extra full-screen copy at render scale < 1

The SDL_GPU pipeline renders into its own view-sized colour target and
already composites it over the whole render target, so routing it through
gpu_scene_texture cost one extra same-size copy per frame. The scaled
intermediate is now used only by the SDL geometry fallback; the hardware
composite upscales with linear filtering.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
shenlei
2026-10-12 17:19:51 +09:00
co-authored by Claude Opus 5.5
parent b57f2fba99
commit df87bc64eb
4 changed files with 21 additions and 2 deletions
+13
View File
@@ -80,6 +80,19 @@ int main() {
for(int i=0;i<100;++i)UI::DrawText(renderer,"NPC label",12,0,0,i%2 ? 0xffff0000 : 0xff00ff00);
assert(SDL_RenderPresent(renderer));
}
{ // Render scale < 1: a smaller colour target is upscaled over the whole render target in the composite.
ava2::HardwareScene scaled(device);
std::vector<ava2::HardwareVertex> quad;
for(auto p: {std::pair{-1.f,-1.f},std::pair{1.f,-1.f},std::pair{1.f,1.f},std::pair{-1.f,1.f}})
quad.push_back({{p.first,p.second,.5f,1},{255,0,0,255},{0,0}});
SDL_SetRenderDrawColor(renderer,0,255,0,255);SDL_RenderClear(renderer);
assert(scaled.Render(renderer,32,32,quad,{0,1,2,0,2,3},{{0,6,nullptr,1,true,true,false,false,-1}}));
auto* pixels=SDL_RenderReadPixels(renderer,nullptr);assert(pixels);
auto redAt=[&](int x,int y){Uint32 v=0;std::memcpy(&v,static_cast<Uint8*>(pixels->pixels)+y*pixels->pitch+x*SDL_BYTESPERPIXEL(pixels->format),SDL_BYTESPERPIXEL(pixels->format));
Uint8 r,g,b,a;SDL_GetRGBA(v,SDL_GetPixelFormatDetails(pixels->format),nullptr,&r,&g,&b,&a);return r>245&&g<5;};
assert(redAt(1,1)&&redAt(62,62)&&redAt(62,1)&&redAt(1,62)&&redAt(32,32));
SDL_DestroySurface(pixels);SDL_RenderPresent(renderer);
}
{
ava2::Gr2SceneRenderer raster(64,64);
raster.SetGPUDevice(device);