// SPDX-License-Identifier: MIT // Host checks for the diagnostic store, with the allocation/clock boundary replaced. #include #include #include #include #include #include #include "NativeGpuTrace.h" static void PE_parse_boot_argn(const char *,unsigned *out,size_t){*out=1;} static void *IOMallocZero(size_t n){return calloc(1,n);} static uint64_t mach_absolute_time(){return std::chrono::steady_clock::now().time_since_epoch().count();} static void absolutetime_to_nanoseconds(uint64_t t,uint64_t *out){*out=t;} #include "NativeGpuTrace.inc" int main() { NativeGpuThreadTargets targets;assert(targets.find(61,1)==0);targets.record(61,1,100); targets.record(61,1,200);targets.record(61,2,250);targets.record(0,1,300); assert(targets.find(61,1)==200);assert(targets.find(61,2)==250);assert(targets.find(0,1)==0); gpuTraceInit(); uint64_t thread[6]={3456,1,0,0,0,0};native_gpu_trace_record(63,61,thread,6); uint64_t encode[6]={3457,2,0,0,0,0};native_gpu_trace_record(63,61,encode,6); assert(native_gpu_role_thread(61,1)==3456 && native_gpu_role_thread(61,2)==3457); NativeGpuTraceSlot slots[8]{};NativeGpuTraceStore store;store.slots=slots;store.capacity=8; for(unsigned i=0;i<32;++i){NativeGpuTraceRecord r{};r.value[0]=i;assert(store.push(r));} NativeGpuTraceRecord row{};assert(!store.read(24,&row));assert(store.read(25,&row) && row.value[0]==24); std::vector writers; for(unsigned t=0;t<4;++t)writers.emplace_back([&,t]{for(unsigned i=0;i<10000;++i){NativeGpuTraceRecord r{};r.pid=t;r.value[0]=i;r.value[1]=~uint64_t(i);store.push(r);}}); for(auto &t:writers)t.join(); for(uint64_t i=store.issued-7;i<=store.issued;++i)if(store.read(i,&row))assert(row.value[1]==~row.value[0]); gpuTraceInit();gpuTraceEvent(GpuTraceBuild,0,1234); for(unsigned i=0;i output(32+(NativeGpuTraceCapacity+NativeGpuTraceMetadata)*120); unsigned bytes=native_gpu_trace_snapshot(output.data(),output.size());assert(bytes>32); auto *head=(NativeGpuTraceHeader *)output.data();auto *rows=(NativeGpuTraceRecord *)(head+1); bool build=false;for(unsigned i=0;icount;++i)build|=rows[i].event==GpuTraceBuild && rows[i].value[0]==1234; assert(build); auto issued=gpuTrace.issued; uint64_t noise[3]={123,53,0}; for(uint32_t event:{12u,13u,14u,15u,38u,39u,43u,61u,62u,64u,68u})native_gpu_trace_record(event,61,noise,3); assert(gpuTrace.issued==issued); native_gpu_trace_record(GpuTraceRetire,61,noise,3);assert(gpuTrace.issued==issued+1); /* The render server's GMEM swap trace: Turnip passes, transfers and attachments. */ for(uint32_t event:{36u,37u,44u})native_gpu_trace_record(event,61,noise,3); assert(gpuTrace.issued==issued+4); NativeGpuLatencyWaits waits;uint64_t omitted=0; assert(waits.keep(1,0,55,3,99,omitted)); for(unsigned i=2;i<100001;++i)assert(!waits.keep(i,0,55,3,99,omitted)); assert(waits.keep(100002,0,55,3,99,omitted) && omitted==99999); assert(waits.keep(100003,0,55,4,99,omitted)); assert(waits.keep(100004,0,55,1,99,omitted)); assert(waits.keep(100005,0,55,3,100,omitted)); return 0; }