perf: defer unused GPU work and streamline preview rendering

这个提交包含在:
mac 于 2026-11-08 16:10:13 +08:00 提交
1 父节点 c9e4d3f330
当前提交 ccc34734e3
18 个文件已更改
+362 -115

没有与搜索条件相匹配的文件

+1 -1
查看文件
@@ -9,7 +9,7 @@
auto dispatch=[&](NSString*name,const void*params,NSUInteger length,const void*extra=nullptr,NSUInteger extraLength=0){id<MTLTexture>out=next();if(!out)return fail(error,@"无法分配一级 CST 缓冲区。");BOOL ok=[self managedDispatch:name textures:@[current,out] data:params length:length extra:extra extraLength:extraLength entry:entry error:error];if(ok)current=out;return ok;};
auto transform=[&](CSTParams p){if(p.inputTransfer==p.outputTransfer&&p.inputGamut==p.outputGamut&&!p.toneMapping)return YES;return dispatch(@"dyPrimaryCST",&p,sizeof(p));};
auto tile=[&](id<MTLTexture>t){DYExportTile*v=[[[DYExportTile alloc]initWithTexture:t registry:d.registryID]autorelease];[v setBounds:src.tilePixelBounds origin:src.imageOrigin];[v setPixelAspect:src.pixelAspect];return v;};
auto blend=[&](id<MTLTexture>before,NSArray*masks,float amount,NSString*type){if(!masks.count&&amount>=1)return YES;id<MTLTexture>out=next();if(!out)return fail(error,@"无法分配一级混合缓冲区。");DYToolMaskHeader h{};id<MTLBuffer>mb=[self maskBuffer:@{type:masks?:@[]} tile:src device:d header:&h];h.section=(int)[DYToolTypes()indexOfObject:type];struct {CSTParams toLinear,fromLinear,originalTo709;float amount;}p={{wg,wt,2,20,0},{2,20,wg,wt,0},{initialGamut,initialTransfer,22,20,0},amount};id<MTLComputePipelineState>pipeline=[self managedKernel:@"dyUnifiedBlend" entry:entry error:error];if(!pipeline||!mb)return NO;id<MTLCommandBuffer>cb=[entry[@"queue"]commandBuffer];id<MTLComputeCommandEncoder>ce=[cb computeCommandEncoder];[ce setComputePipelineState:pipeline];[ce setTexture:before atIndex:0];[ce setTexture:current atIndex:1];[ce setTexture:out atIndex:2];[ce setTexture:originalWorking?:before atIndex:3];[ce setBuffer:mb offset:0 atIndex:0];[ce setBytes:&h length:sizeof(h) atIndex:1];[ce setBytes:&p length:sizeof(p) atIndex:2];[ce dispatchThreads:MTLSizeMake(source.width,source.height,1) threadsPerThreadgroup:MTLSizeMake(16,16,1)];[ce endEncoding];[cb commit];[cb waitUntilCompleted];if(cb.status!=MTLCommandBufferStatusCompleted)return fail(error,cb.error.localizedDescription);current=out;return YES;};
auto blend=[&](id<MTLTexture>before,NSArray*masks,float amount,NSString*type){if(!masks.count&&amount>=1)return YES;id<MTLTexture>out=next();if(!out)return fail(error,@"无法分配一级混合缓冲区。");DYToolMaskHeader h{};id<MTLBuffer>mb=[self maskBuffer:@{type:masks?:@[]} tile:src device:d header:&h];h.section=(int)[DYToolTypes()indexOfObject:type];struct {CSTParams toLinear,fromLinear,originalTo709;float amount;}p={{wg,wt,2,20,0},{2,20,wg,wt,0},{initialGamut,initialTransfer,22,20,0},amount};id<MTLComputePipelineState>pipeline=[self computePipeline:@"dyUnifiedBlend" entry:entry error:error];if(!pipeline||!mb)return NO;id<MTLCommandBuffer>cb=[entry[@"queue"]commandBuffer];id<MTLComputeCommandEncoder>ce=[cb computeCommandEncoder];[ce setComputePipelineState:pipeline];[ce setTexture:before atIndex:0];[ce setTexture:current atIndex:1];[ce setTexture:out atIndex:2];[ce setTexture:originalWorking?:before atIndex:3];[ce setBuffer:mb offset:0 atIndex:0];[ce setBytes:&h length:sizeof(h) atIndex:1];[ce setBytes:&p length:sizeof(p) atIndex:2];[ce dispatchThreads:MTLSizeMake(source.width,source.height,1) threadsPerThreadgroup:MTLSizeMake(16,16,1)];[ce endEncoding];[cb commit];[cb waitUntilCompleted];if(cb.status!=MTLCommandBufferStatusCompleted)return fail(error,cb.error.localizedDescription);current=out;return YES;};
auto grade=[&](DYUniforms values,NSArray*masks,BOOL local){if(values.primarySectionEnabled<.5||values.cm.mix<=0)return YES;DYUniforms neutral=values;neutral.inputSectionEnabled=neutral.cm.outputEnabled=0;BOOL gradeNeutral=identityCM(neutral);if(!local&&gradeNeutral)return YES;if(local&&identityCM(values))return YES;
DayuRenderParams p=DYCMRender(values.cm);p.input.toneMapping=0;if((int)lroundf(values.cm.workingTransfer)==20){p.reserved|=8;p.basis.colorspace=10;p.basisInput={(int)lroundf(values.cm.workingGamut),20,(int)lroundf(values.cm.workingGamut),10,0};p.basisOutput={(int)lroundf(values.cm.workingGamut),10,(int)lroundf(values.cm.workingGamut),20,0};}CSTParams back={wg,wt,wg,wt,0};
if(local){p.stages=(values.inputSectionEnabled>.5?1:0)|(gradeNeutral?0:2)|(values.cm.outputEnabled>.5?4:0);back={values.cm.outputEnabled>.5?(int)lroundf(values.cm.outputGamut):(int)lroundf(values.cm.workingGamut),values.cm.outputEnabled>.5?(int)lroundf(values.cm.outputTransfer):(int)lroundf(values.cm.workingTransfer),wg,wt,0};}