ryujinx/SDL
1
0
Fork 0
mirror of https://github.com/Ryujinx/SDL.git synced 2024-12-23 15:15:39 +00:00
Code Issues Packages Projects Releases Wiki Activity

optimize Emscripten_UpdateWindowFramebuffer

Browse source 
- avoid creating contexts and images all the time
 - use set and then fix alpha directly
This commit is contained in:
Alon Zakai 2016-09-13 00:03:43 -07:00
parent 63200ea395
commit 2b367cb6b0
1 changed files with 59 additions and 17 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

76
src/video/emscripten/SDL_emscriptenframebuffer.c
View file

@ -69,15 +69,25 @@ int Emscripten_UpdateWindowFramebuffer(_THIS, SDL_Window * window, const SDL_Rec
/* Send the data to the display */ /* Send the data to the display */
EM_ASM_INT({ EM_ASM_INT({
//TODO: don't create context every update var w = $0;
var ctx = Module['canvas'].getContext('2d'); var h = $1;
var pixels = $2;
//library_sdl.js SDL_UnlockSurface if (!Module['SDL2']) Module['SDL2'] = {};
var image = ctx.createImageData($0, $1); var SDL2 = Module['SDL2'];
var data = image.data; if (SDL2.ctxCanvas !== Module['canvas']) {
var src = $2 >> 2; SDL2.ctx = Module['canvas'].getContext('2d');
SDL2.ctxCanvas = Module['canvas'];
}
if (SDL2.w !== w || SDL2.h !== h || SDL2.imageCtx !== SDL2.ctx) {
SDL2.image = SDL2.ctx.createImageData(w, h);
SDL2.w = w;
SDL2.h = h;
SDL2.imageCtx = SDL2.ctx;
}
var data = SDL2.image.data;
var src = pixels >> 2;
var dst = 0; var dst = 0;
var isScreen = true;
var num; var num;
if (typeof CanvasPixelArray !== 'undefined' && data instanceof CanvasPixelArray) { if (typeof CanvasPixelArray !== 'undefined' && data instanceof CanvasPixelArray) {
// IE10/IE11: ImageData objects are backed by the deprecated CanvasPixelArray, // IE10/IE11: ImageData objects are backed by the deprecated CanvasPixelArray,
@ -90,26 +100,58 @@ int Emscripten_UpdateWindowFramebuffer(_THIS, SDL_Window * window, const SDL_Rec
data[dst ] = val & 0xff; data[dst ] = val & 0xff;
data[dst+1] = (val >> 8) & 0xff; data[dst+1] = (val >> 8) & 0xff;
data[dst+2] = (val >> 16) & 0xff; data[dst+2] = (val >> 16) & 0xff;
data[dst+3] = isScreen ? 0xff : ((val >> 24) & 0xff); data[dst+3] = 0xff;
src++; src++;
dst += 4; dst += 4;
} }
} else { } else {
var data32 = new Uint32Array(data.buffer); if (SDL2.data32Data !== data) {
SDL2.data32 = new Int32Array(data.buffer);
SDL2.data8 = new Uint8Array(data.buffer);
}
var data32 = SDL2.data32;
num = data32.length; num = data32.length;
if (isScreen) { // logically we need to do
while (dst < num) { // while (dst < num) {
// HEAP32[src++] is an optimization. Instead, we could do {{{ makeGetValue('buffer', 'dst', 'i32') }}}; // data32[dst++] = HEAP32[src++] | 0xff000000
data32[dst++] = HEAP32[src++] | 0xff000000; // }
// the following code is faster though, because
// .set() is almost free - easily 10x faster due to
// native memcpy efficiencies, and the remaining loop
// just stores, not load + store, so it is faster
data32.set(HEAP32.subarray(src, src + num));
var data8 = SDL2.data8;
var i = 3;
var j = i + 4*num;
if (num % 8 == 0) {
// unrolling gives big speedups
while (i < j) {
data8[i] = 0xff;
i = i + 4 | 0;
data8[i] = 0xff;
i = i + 4 | 0;
data8[i] = 0xff;
i = i + 4 | 0;
data8[i] = 0xff;
i = i + 4 | 0;
data8[i] = 0xff;
i = i + 4 | 0;
data8[i] = 0xff;
i = i + 4 | 0;
data8[i] = 0xff;
i = i + 4 | 0;
data8[i] = 0xff;
i = i + 4 | 0;
} }
} else { } else {
while (dst < num) { while (i < j) {
data32[dst++] = HEAP32[src++]; data8[i] = 0xff;
i = i + 4 | 0;
} }
} }
} }
ctx.putImageData(image, 0, 0); SDL2.ctx.putImageData(SDL2.image, 0, 0);
return 0; return 0;
}, surface->w, surface->h, surface->pixels); }, surface->w, surface->h, surface->pixels);