Lines Matching defs:Kernel

152 static const omp::GV &getGridValue(const Triple &T, Function *Kernel) {
155         Kernel->getFnAttribute("target-features").getValueAsString();
6210   Function *Kernel = DebugKernelWrapper;
6213   StringRef KernelName = Kernel->getName();
6217     Kernel = M.getFunction(KernelName);
6218     assert(Kernel && "Expected the real kernel to exist");
6224     writeTeamsForKernel(T, *Kernel, Attrs.MinTeams, Attrs.MaxTeams.front());
6231         int32_t(getGridValue(T, Kernel).GV_Default_WG_Size), Attrs.MinThreads);
6234     writeThreadBoundsForKernel(T, *Kernel, Attrs.MinThreads, MaxThreadsVal);
6344   Function *Kernel = Builder.GetInsertBlock()->getParent();
6346   StringRef KernelName = Kernel->getName();
6363 static MDNode *getNVPTXMDNode(Function &Kernel, StringRef Name) {
6364   Module &M = *Kernel.getParent();
6370     if (!KernelOp || KernelOp->getValue() != &Kernel)
6380 static void updateNVPTXMetadata(Function &Kernel, StringRef Name, int32_t Value,
6383   MDNode *ExistingOp = getNVPTXMDNode(Kernel, Name);
6392     LLVMContext &Ctx = Kernel.getContext();
6393     Metadata *MDVals[] = {ConstantAsMetadata::get(&Kernel),
6398     Module &M = *Kernel.getParent();
6405 OpenMPIRBuilder::readThreadBoundsForKernel(const Triple &T, Function &Kernel) {
6407       Kernel.getFnAttributeAsParsedInteger("omp_target_thread_limit");
6410     const auto &Attr = Kernel.getFnAttribute("amdgpu-flat-work-group-size");
6423   if (MDNode *ExistingOp = getNVPTXMDNode(Kernel, "maxntidx")) {
6432                                                  Function &Kernel, int32_t LB,
6434   Kernel.addFnAttr("omp_target_thread_limit", std::to_string(UB));
6437     Kernel.addFnAttr("amdgpu-flat-work-group-size",
6442   updateNVPTXMetadata(Kernel, "maxntidx", UB, true);
6446 OpenMPIRBuilder::readTeamBoundsForKernel(const Triple &, Function &Kernel) {
6448   return {0, Kernel.getFnAttributeAsParsedInteger("omp_target_num_teams")};
6451 void OpenMPIRBuilder::writeTeamsForKernel(const Triple &T, Function &Kernel,
6455       updateNVPTXMetadata(Kernel, "maxclusterrank", UB, true);
6457     Kernel.addFnAttr("amdgpu-max-num-workgroups", llvm::utostr(LB) + ",1,1");
6459   Kernel.addFnAttr("omp_target_num_teams", std::to_string(LB));