profiler; rsf, makefile and image loading tweaks; pngcrush'd largest images; removed leftover debug text from asset loading

libstarlight/source/starlight/ThemeManager.cpp

@@ -25,6 +25,10 @@
 
 #include "starlight/util/JsonConversions.h"
 
+#include "starlight/util/Profiler.h"
+using starlight::util::Profiler;
+#include <sstream>
+
 using std::string;
 using std::shared_ptr;
 using std::make_shared;
@@ -75,40 +79,48 @@ namespace {
     }
     
     CTexture* LoadPNG(const std::string& path, bool isPremult = false) {
+        //Profiler::TaskStart();
         unsigned char* imgbuf;
         unsigned width, height;
+        lodepng::State state;
         lodepng_decode32_file(&imgbuf, &width, &height, path.c_str());
+        if (state.info_png.color.colortype != 6) isPremult = true; // expect no alpha if not rgba
+        /*{
+            std::stringstream ss; ss << "loaded png, ";
+            ss << width; ss << "x"; ss << height;
+            Profiler::TaskFinish(ss.str());
+        }*/
         
         unsigned bw = NextPow2(width), bh = NextPow2(height);
         
         u8* gpubuf = static_cast<u8*>(linearAlloc(bw*bh*4));
+        //Profiler::TaskStart();
+        //memset(gpubuf, 0, bw*bh*4);
+        //Profiler::TaskFinish("cleared canvas");
         
-        u8* src = static_cast<u8*>(imgbuf); u8* dst = static_cast<u8*>(gpubuf);
         
+        //Profiler::TaskStart();
         if (isPremult) {
+            u32* src = reinterpret_cast<u32*>(imgbuf); u32* dst = reinterpret_cast<u32*>(gpubuf);
             // just convert endianness
             for(unsigned iy = 0; iy < height; iy++) {
                 for (unsigned ix = 0; ix < width; ix++) {
-                    int r = *src++;
-                    int g = *src++;
-                    int b = *src++;
-                    int a = *src++;
+                    u32 clr = *src;
+                    *dst = __builtin_bswap32(clr);
                     
-                    *dst++ = a;
-                    *dst++ = b;
-                    *dst++ = g;
-                    *dst++ = r;
+                    src+=4; dst+=4;
                 }
                 dst += (bw - width) * 4; // skip the difference
             }
         } else {
+            u8* src = static_cast<u8*>(imgbuf); u8* dst = static_cast<u8*>(gpubuf);
             // convert and premultiply
             for(unsigned iy = 0; iy < height; iy++) {
                 for (unsigned ix = 0; ix < width; ix++) {
-                    int r = *src++;
-                    int g = *src++;
-                    int b = *src++;
-                    int a = *src++;
+                    u8 r = *src++;
+                    u8 g = *src++;
+                    u8 b = *src++;
+                    u8 a = *src++;
                     
                     float aa = (1.0f / 255.0f) * a;
                     
@@ -120,11 +132,14 @@ namespace {
                 dst += (bw - width) * 4; // skip the difference
             }
         }
+        //Profiler::TaskFinish("made into canvas, flipped and premult");
         // completely skipping over the difference instead of erasing might eventually lead to garbage outside of frame,
         // but meh; that'll only be visible if you intentionally push the UVs outside the image proper
         
+        //Profiler::TaskStart();
         CTexture* tx = RenderCore::LoadTexture(static_cast<void*>(gpubuf), bw, bh);
         tx->size = Vector2(width, height); // and for now just fix the size after the fact
+        //Profiler::TaskFinish("copied into linear");
         
         std::free(imgbuf);
         linearFree(gpubuf);
@@ -216,7 +231,7 @@ shared_ptr<Drawable> ThemeManager::LoadAsset(string& path, ThemeRefContainer<Dra
     static shared_ptr<Drawable> nulldrw = make_shared<starlight::gfx::DrawableTest>();
     
     string ext = FindExtension(path);
-    printf("load: %s (%s)\n", path.c_str(), ext.c_str());
+    //printf("load: %s (%s)\n", path.c_str(), ext.c_str());
     /**/ if (ext == "png") {
         return make_shared<DrawableImage>(LoadPNG(path));
     }
@@ -227,7 +242,7 @@ shared_ptr<Drawable> ThemeManager::LoadAsset(string& path, ThemeRefContainer<Dra
             fs >> j;
         }
         auto st = j.dump();
-        printf("file contents: %s\n", st.c_str());
+        //printf("file contents: %s\n", st.c_str());
         
         string type = j["assetType"];
         /**/ if (type == "ninepatch") {

libstarlight/source/starlight/gfx/RenderCore.cpp

@@ -264,7 +264,7 @@ CTexture* RenderCore::LoadTexture(void* src, int width, int height) {
     
     C3D_TexBind(0, tex->texture);
     
-    printf("loaded image w %i (%i) h %i (%i)\n", width, owidth, height, oheight);
+    //printf("loaded image w %i (%i) h %i (%i)\n", width, owidth, height, oheight);
     
     return tex;
 }

libstarlight/source/starlight/ui/DebugConsole.cpp

@@ -88,6 +88,7 @@ void DebugConsole::PreDraw() {
         if (!buffer) buffer = std::make_unique<gfx::DrawContextCanvas>(rect.size + Vector2(0, 8));
         buffer->Clear();
         GFXManager::PushContext(buffer.get());
+        GFXManager::PrepareForDrawing(); // force clear even if nothing to write
         textConfig.Print(buffer->rect, text);
         GFXManager::PopContext();
     }

libstarlight/source/starlight/ui/Label.cpp

@@ -53,6 +53,7 @@ void Label::PreDraw() {
         buffer = std::make_unique<gfx::DrawContextCanvas>(rect.size + Vector2(0, 8));
         buffer->Clear();
         GFXManager::PushContext(buffer.get());
+        GFXManager::PrepareForDrawing(); // force clear even if nothing to write
         textConfig.ROGet().Print(buffer->rect, text);
         GFXManager::PopContext();
     }

libstarlight/source/starlight/util/Profiler.cpp

@@ -0,0 +1,19 @@
+#include "Profiler.h"
+
+#include <cstdio>
+
+#include "3ds.h"
+
+using starlight::util::Profiler;
+
+Profiler::TickCount Profiler::tc = Profiler::TickCount();
+
+void Profiler::TaskStart() {
+    osTickCounterUpdate(reinterpret_cast<TickCounter*>(&tc));
+}
+
+void Profiler::TaskFinish(const std::string& msg) {
+    osTickCounterUpdate(reinterpret_cast<TickCounter*>(&tc));
+    double tm = osTickCounterRead(reinterpret_cast<TickCounter*>(&tc));
+    printf("T:%f - %s\n", tm, msg.c_str());
+}

libstarlight/source/starlight/util/Profiler.h

@@ -0,0 +1,22 @@
+#pragma once
+#include "starlight/_global.h"
+
+#include <string>
+
+namespace starlight {
+    namespace util {
+        class Profiler {
+        private:
+            struct TickCount {
+                unsigned long long int elapsed;
+                unsigned long long int ref;
+            };
+            static TickCount tc;
+            Profiler() {};
+            
+        public:
+            static void TaskStart();
+            static void TaskFinish(const std::string& msg);
+        };
+    }
+}

libstarlight/todo.txt

@@ -13,6 +13,9 @@ roadmap to v0.5.1 {
     
     - libctru console as ui element
     
+    pngcrush the biggest assets (default and osk backdrops etc.)
+    - profile image loading (for large images, loading the png and spanned copy/premult take about the same time; memset to preclear is only ~2ms with a 512x512)
+    
     - fix the hang on osk when pressing (L|R)+up+left
     figure out what (else) to put on the left side of the keyboard (opposite backspace and enter)
     temporary drawable loading, local themeref, discard etc.
@@ -28,6 +31,7 @@ roadmap to v0.5.1 {
         ...and some mechanism for allowing it to opt out of the rest of the cycle
         
         Trackable sideclass for threads; float progress 0..1, etc.
+        ^ make progress bar and use it for a progress/"please wait" dialog
         - MAKE THREADS END CLEANLY
         ^ observed a single instance of being stalled on redscreen, not really sure what that was about
         lambda task thread

testbed/Makefile

@@ -173,7 +173,7 @@ send: $(BUILD)
 #---------------------------------------------------------------------------------
 send-cia: $(TARGET)-strip.elf
 	@makerom -f cia -o $(TARGET).cia -elf $(TARGET)-strip.elf -rsf resources/$(TARGET).rsf -icon resources/icon.icn -banner resources/banner.bnr -exefslogo -target t
-	@sockme $(TARGET).cia $(3DSIP)
+	@sockme $(TARGET).cia $(3DSIP) || sockme $(TARGET).cia $(3DSIP) || sockme $(TARGET).cia $(3DSIP)
 #---------------------------------------------------------------------------------
 run: $(BUILD)
 	@citra $(TARGET).3dsx

testbed/resources/starlight-testbed.rsf

@@ -91,6 +91,12 @@ AccessControlInfo:
   Priority                      : 16
 
   MaxCpu                        : 0x9E # Default
+  
+  # New3DS Exclusive Process Settings
+  SystemModeExt                 : Legacy # Legacy(Default)/124MB/178MB  Legacy:Use Old3DS SystemMode
+  CpuSpeed                      : 804MHz # 268MHz(Default)/804MHz
+  EnableL2Cache                 : true # false(default)/true
+  CanAccessCore2                : true 
 
   DisableDebug                  : true
   EnableForceDebug              : false

testbed/source/Core.cpp

@@ -11,6 +11,7 @@
 #include "starlight/gfx/RenderCore.h"
 
 #include "starlight/util/Path.h"
+#include "starlight/util/Profiler.h"
 
 #include "starlight/ui/ParallaxLayer.h"
 #include "starlight/ui/ScrollField.h"
@@ -36,6 +37,7 @@ using starlight::GFXManager;
 using starlight::gfx::RenderCore;
 
 using starlight::util::Path;
+using starlight::util::Profiler;
 
 using starlight::Application;