1 //===- AMDGPUResourceUsageAnalysis.h ---- analysis of resources -----------===//
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //===----------------------------------------------------------------------===//
11 /// functions.
20 /// hardware-entrypoints. Therefore the register usage of functions with
21 /// indirect calls is estimated as the maximum of all non-entrypoint functions
24 //===----------------------------------------------------------------------===//
41 #define DEBUG_TYPE "amdgpu-resource-usage"
48 // only due to dynamic / non-entry block allocas.
50     "amdgpu-assume-external-call-stack-size",
55     "amdgpu-assume-dynamic-stack-object-size",
68   return cast<Function>(Op.getGlobal()->stripPointerCastsAndAliases());  in getCalleeFunction()
104   const TargetMachine &TM = TPC->getTM<TargetMachine>();  in runOnModule()
125     Function *F = IT->getFunction();  in runOnModule()
126     if (!F || F->isDeclaration())  in runOnModule()
134     SIFunctionResourceInfo &Info = CI.first->second;  in runOnModule()
141   // It's possible we have unreachable functions in the module which weren't  in runOnModule()
146     if (!F || F->isDeclaration())  in runOnModule()
151     if (!CI.second) // Skip already visited functions  in runOnModule()
154     SIFunctionResourceInfo &Info = CI.first->second;  in runOnModule()
180   const SIRegisterInfo &TRI = TII->getRegisterInfo();  in analyzeResourceUsage()
184                          MRI.isLiveIn(MFI->getPreloadedReg(  in analyzeResourceUsage()
193   if (Info.UsesFlatScratch && !MFI->getUserSGPRInfo().hasFlatScratchInit() &&  in analyzeResourceUsage()
207   if (MFI->isStackRealigned())  in analyzeResourceUsage()
258   int32_t MaxVGPR = -1;  in analyzeResourceUsage()
259   int32_t MaxAGPR = -1;  in analyzeResourceUsage()
260   int32_t MaxSGPR = -1;  in analyzeResourceUsage()
502         int MaxUsed = HWReg + Width - 1;  in analyzeResourceUsage()
517             TII->getNamedOperand(MI, AMDGPU::OpName::callee);  in analyzeResourceUsage()
527         if (Callee && AMDGPU::isEntryFunctionCC(Callee->getCallingConv()))  in analyzeResourceUsage()
530         bool IsIndirect = !Callee || Callee->isDeclaration();  in analyzeResourceUsage()
535         if (!Callee || !Callee->doesNotRecurse()) {  in analyzeResourceUsage()
547             CalleeFrameSize = std::max(  in analyzeResourceUsage()
555               std::max(CalleeFrameSize,  in analyzeResourceUsage()
566           MaxSGPR = std::max(I->second.NumExplicitSGPR - 1, MaxSGPR);  in analyzeResourceUsage()
567           MaxVGPR = std::max(I->second.NumVGPR - 1, MaxVGPR);  in analyzeResourceUsage()
568           MaxAGPR = std::max(I->second.NumAGPR - 1, MaxAGPR);  in analyzeResourceUsage()
570               std::max(I->second.PrivateSegmentSize, CalleeFrameSize);  in analyzeResourceUsage()
571           Info.UsesVCC |= I->second.UsesVCC;  in analyzeResourceUsage()
572           Info.UsesFlatScratch |= I->second.UsesFlatScratch;  in analyzeResourceUsage()
573           Info.HasDynamicallySizedStack |= I->second.HasDynamicallySizedStack;  in analyzeResourceUsage()
574           Info.HasRecursion |= I->second.HasRecursion;  in analyzeResourceUsage()
575           Info.HasIndirectCall |= I->second.HasIndirectCall;  in analyzeResourceUsage()
590   // Collect the maximum number of registers from non-hardware-entrypoints.  in propagateIndirectCallRegisterUsage()
591   // All these functions are potential targets for indirect calls.  in propagateIndirectCallRegisterUsage()
597     if (!AMDGPU::isEntryFunctionCC(I.getFirst()->getCallingConv())) {  in propagateIndirectCallRegisterUsage()
599       NonKernelMaxSGPRs = std::max(NonKernelMaxSGPRs, Info.NumExplicitSGPR);  in propagateIndirectCallRegisterUsage()
600       NonKernelMaxVGPRs = std::max(NonKernelMaxVGPRs, Info.NumVGPR);  in propagateIndirectCallRegisterUsage()
601       NonKernelMaxAGPRs = std::max(NonKernelMaxAGPRs, Info.NumAGPR);  in propagateIndirectCallRegisterUsage()
605   // Add register usage for functions with indirect calls.  in propagateIndirectCallRegisterUsage()
606   // For calls to unknown functions, we assume the maximum register usage of  in propagateIndirectCallRegisterUsage()
607   // all non-hardware-entrypoints in the current module.  in propagateIndirectCallRegisterUsage()
611       Info.NumExplicitSGPR = std::max(Info.NumExplicitSGPR, NonKernelMaxSGPRs);  in propagateIndirectCallRegisterUsage()
612       Info.NumVGPR = std::max(Info.NumVGPR, NonKernelMaxVGPRs);  in propagateIndirectCallRegisterUsage()
613       Info.NumAGPR = std::max(Info.NumAGPR, NonKernelMaxAGPRs);  in propagateIndirectCallRegisterUsage()