xref: /freebsd/contrib/llvm-project/llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp (revision 7a6dacaca14b62ca4b74406814becb87a3fefac0)
1 //===-- AMDGPULowerModuleLDSPass.cpp ------------------------------*- C++ -*-=//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This pass eliminates local data store, LDS, uses from non-kernel functions.
10 // LDS is contiguous memory allocated per kernel execution.
11 //
12 // Background.
13 //
14 // The programming model is global variables, or equivalently function local
15 // static variables, accessible from kernels or other functions. For uses from
16 // kernels this is straightforward - assign an integer to the kernel for the
17 // memory required by all the variables combined, allocate them within that.
18 // For uses from functions there are performance tradeoffs to choose between.
19 //
20 // This model means the GPU runtime can specify the amount of memory allocated.
21 // If this is more than the kernel assumed, the excess can be made available
22 // using a language specific feature, which IR represents as a variable with
23 // no initializer. This feature is referred to here as "Dynamic LDS" and is
24 // lowered slightly differently to the normal case.
25 //
26 // Consequences of this GPU feature:
27 // - memory is limited and exceeding it halts compilation
28 // - a global accessed by one kernel exists independent of other kernels
29 // - a global exists independent of simultaneous execution of the same kernel
30 // - the address of the global may be different from different kernels as they
31 //   do not alias, which permits only allocating variables they use
32 // - if the address is allowed to differ, functions need help to find it
33 //
34 // Uses from kernels are implemented here by grouping them in a per-kernel
35 // struct instance. This duplicates the variables, accurately modelling their
36 // aliasing properties relative to a single global representation. It also
37 // permits control over alignment via padding.
38 //
39 // Uses from functions are more complicated and the primary purpose of this
40 // IR pass. Several different lowering are chosen between to meet requirements
41 // to avoid allocating any LDS where it is not necessary, as that impacts
42 // occupancy and may fail the compilation, while not imposing overhead on a
43 // feature whose primary advantage over global memory is performance. The basic
44 // design goal is to avoid one kernel imposing overhead on another.
45 //
46 // Implementation.
47 //
48 // LDS variables with constant annotation or non-undef initializer are passed
49 // through unchanged for simplification or error diagnostics in later passes.
50 // Non-undef initializers are not yet implemented for LDS.
51 //
52 // LDS variables that are always allocated at the same address can be found
53 // by lookup at that address. Otherwise runtime information/cost is required.
54 //
55 // The simplest strategy possible is to group all LDS variables in a single
56 // struct and allocate that struct in every kernel such that the original
57 // variables are always at the same address. LDS is however a limited resource
58 // so this strategy is unusable in practice. It is not implemented here.
59 //
60 // Strategy | Precise allocation | Zero runtime cost | General purpose |
61 //  --------+--------------------+-------------------+-----------------+
62 //   Module |                 No |               Yes |             Yes |
63 //    Table |                Yes |                No |             Yes |
64 //   Kernel |                Yes |               Yes |              No |
65 //   Hybrid |                Yes |           Partial |             Yes |
66 //
67 // "Module" spends LDS memory to save cycles. "Table" spends cycles and global
68 // memory to save LDS. "Kernel" is as fast as kernel allocation but only works
69 // for variables that are known reachable from a single kernel. "Hybrid" picks
70 // between all three. When forced to choose between LDS and cycles we minimise
71 // LDS use.
72 
73 // The "module" lowering implemented here finds LDS variables which are used by
74 // non-kernel functions and creates a new struct with a field for each of those
75 // LDS variables. Variables that are only used from kernels are excluded.
76 //
77 // The "table" lowering implemented here has three components.
78 // First kernels are assigned a unique integer identifier which is available in
79 // functions it calls through the intrinsic amdgcn_lds_kernel_id. The integer
80 // is passed through a specific SGPR, thus works with indirect calls.
81 // Second, each kernel allocates LDS variables independent of other kernels and
82 // writes the addresses it chose for each variable into an array in consistent
83 // order. If the kernel does not allocate a given variable, it writes undef to
84 // the corresponding array location. These arrays are written to a constant
85 // table in the order matching the kernel unique integer identifier.
86 // Third, uses from non-kernel functions are replaced with a table lookup using
87 // the intrinsic function to find the address of the variable.
88 //
89 // "Kernel" lowering is only applicable for variables that are unambiguously
90 // reachable from exactly one kernel. For those cases, accesses to the variable
91 // can be lowered to ConstantExpr address of a struct instance specific to that
92 // one kernel. This is zero cost in space and in compute. It will raise a fatal
93 // error on any variable that might be reachable from multiple kernels and is
94 // thus most easily used as part of the hybrid lowering strategy.
95 //
96 // Hybrid lowering is a mixture of the above. It uses the zero cost kernel
97 // lowering where it can. It lowers the variable accessed by the greatest
98 // number of kernels using the module strategy as that is free for the first
99 // variable. Any futher variables that can be lowered with the module strategy
100 // without incurring LDS memory overhead are. The remaining ones are lowered
101 // via table.
102 //
103 // Consequences
104 // - No heuristics or user controlled magic numbers, hybrid is the right choice
105 // - Kernels that don't use functions (or have had them all inlined) are not
106 //   affected by any lowering for kernels that do.
107 // - Kernels that don't make indirect function calls are not affected by those
108 //   that do.
109 // - Variables which are used by lots of kernels, e.g. those injected by a
110 //   language runtime in most kernels, are expected to have no overhead
111 // - Implementations that instantiate templates per-kernel where those templates
112 //   use LDS are expected to hit the "Kernel" lowering strategy
113 // - The runtime properties impose a cost in compiler implementation complexity
114 //
115 // Dynamic LDS implementation
116 // Dynamic LDS is lowered similarly to the "table" strategy above and uses the
117 // same intrinsic to identify which kernel is at the root of the dynamic call
118 // graph. This relies on the specified behaviour that all dynamic LDS variables
119 // alias one another, i.e. are at the same address, with respect to a given
120 // kernel. Therefore this pass creates new dynamic LDS variables for each kernel
121 // that allocates any dynamic LDS and builds a table of addresses out of those.
122 // The AMDGPUPromoteAlloca pass skips kernels that use dynamic LDS.
123 // The corresponding optimisation for "kernel" lowering where the table lookup
124 // is elided is not implemented.
125 //
126 //
127 // Implementation notes / limitations
128 // A single LDS global variable represents an instance per kernel that can reach
129 // said variables. This pass essentially specialises said variables per kernel.
130 // Handling ConstantExpr during the pass complicated this significantly so now
131 // all ConstantExpr uses of LDS variables are expanded to instructions. This
132 // may need amending when implementing non-undef initialisers.
133 //
134 // Lowering is split between this IR pass and the back end. This pass chooses
135 // where given variables should be allocated and marks them with metadata,
136 // MD_absolute_symbol. The backend places the variables in coincidentally the
137 // same location and raises a fatal error if something has gone awry. This works
138 // in practice because the only pass between this one and the backend that
139 // changes LDS is PromoteAlloca and the changes it makes do not conflict.
140 //
141 // Addresses are written to constant global arrays based on the same metadata.
142 //
143 // The backend lowers LDS variables in the order of traversal of the function.
144 // This is at odds with the deterministic layout required. The workaround is to
145 // allocate the fixed-address variables immediately upon starting the function
146 // where they can be placed as intended. This requires a means of mapping from
147 // the function to the variables that it allocates. For the module scope lds,
148 // this is via metadata indicating whether the variable is not required. If a
149 // pass deletes that metadata, a fatal error on disagreement with the absolute
150 // symbol metadata will occur. For kernel scope and dynamic, this is by _name_
151 // correspondence between the function and the variable. It requires the
152 // kernel to have a name (which is only a limitation for tests in practice) and
153 // for nothing to rename the corresponding symbols. This is a hazard if the pass
154 // is run multiple times during debugging. Alternative schemes considered all
155 // involve bespoke metadata.
156 //
157 // If the name correspondence can be replaced, multiple distinct kernels that
158 // have the same memory layout can map to the same kernel id (as the address
159 // itself is handled by the absolute symbol metadata) and that will allow more
160 // uses of the "kernel" style faster lowering and reduce the size of the lookup
161 // tables.
162 //
163 // There is a test that checks this does not fire for a graphics shader. This
164 // lowering is expected to work for graphics if the isKernel test is changed.
165 //
166 // The current markUsedByKernel is sufficient for PromoteAlloca but is elided
167 // before codegen. Replacing this with an equivalent intrinsic which lasts until
168 // shortly after the machine function lowering of LDS would help break the name
169 // mapping. The other part needed is probably to amend PromoteAlloca to embed
170 // the LDS variables it creates in the same struct created here. That avoids the
171 // current hazard where a PromoteAlloca LDS variable might be allocated before
172 // the kernel scope (and thus error on the address check). Given a new invariant
173 // that no LDS variables exist outside of the structs managed here, and an
174 // intrinsic that lasts until after the LDS frame lowering, it should be
175 // possible to drop the name mapping and fold equivalent memory layouts.
176 //
177 //===----------------------------------------------------------------------===//
178 
179 #include "AMDGPU.h"
180 #include "AMDGPUTargetMachine.h"
181 #include "Utils/AMDGPUBaseInfo.h"
182 #include "Utils/AMDGPUMemoryUtils.h"
183 #include "llvm/ADT/BitVector.h"
184 #include "llvm/ADT/DenseMap.h"
185 #include "llvm/ADT/DenseSet.h"
186 #include "llvm/ADT/STLExtras.h"
187 #include "llvm/ADT/SetOperations.h"
188 #include "llvm/Analysis/CallGraph.h"
189 #include "llvm/CodeGen/TargetPassConfig.h"
190 #include "llvm/IR/Constants.h"
191 #include "llvm/IR/DerivedTypes.h"
192 #include "llvm/IR/IRBuilder.h"
193 #include "llvm/IR/InlineAsm.h"
194 #include "llvm/IR/Instructions.h"
195 #include "llvm/IR/IntrinsicsAMDGPU.h"
196 #include "llvm/IR/MDBuilder.h"
197 #include "llvm/IR/ReplaceConstant.h"
198 #include "llvm/InitializePasses.h"
199 #include "llvm/Pass.h"
200 #include "llvm/Support/CommandLine.h"
201 #include "llvm/Support/Debug.h"
202 #include "llvm/Support/Format.h"
203 #include "llvm/Support/OptimizedStructLayout.h"
204 #include "llvm/Support/raw_ostream.h"
205 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
206 #include "llvm/Transforms/Utils/ModuleUtils.h"
207 
208 #include <vector>
209 
210 #include <cstdio>
211 
212 #define DEBUG_TYPE "amdgpu-lower-module-lds"
213 
214 using namespace llvm;
215 
216 namespace {
217 
218 cl::opt<bool> SuperAlignLDSGlobals(
219     "amdgpu-super-align-lds-globals",
220     cl::desc("Increase alignment of LDS if it is not on align boundary"),
221     cl::init(true), cl::Hidden);
222 
223 enum class LoweringKind { module, table, kernel, hybrid };
224 cl::opt<LoweringKind> LoweringKindLoc(
225     "amdgpu-lower-module-lds-strategy",
226     cl::desc("Specify lowering strategy for function LDS access:"), cl::Hidden,
227     cl::init(LoweringKind::hybrid),
228     cl::values(
229         clEnumValN(LoweringKind::table, "table", "Lower via table lookup"),
230         clEnumValN(LoweringKind::module, "module", "Lower via module struct"),
231         clEnumValN(
232             LoweringKind::kernel, "kernel",
233             "Lower variables reachable from one kernel, otherwise abort"),
234         clEnumValN(LoweringKind::hybrid, "hybrid",
235                    "Lower via mixture of above strategies")));
236 
237 bool isKernelLDS(const Function *F) {
238   // Some weirdness here. AMDGPU::isKernelCC does not call into
239   // AMDGPU::isKernel with the calling conv, it instead calls into
240   // isModuleEntryFunction which returns true for more calling conventions
241   // than AMDGPU::isKernel does. There's a FIXME on AMDGPU::isKernel.
242   // There's also a test that checks that the LDS lowering does not hit on
243   // a graphics shader, denoted amdgpu_ps, so stay with the limited case.
244   // Putting LDS in the name of the function to draw attention to this.
245   return AMDGPU::isKernel(F->getCallingConv());
246 }
247 
248 template <typename T> std::vector<T> sortByName(std::vector<T> &&V) {
249   llvm::sort(V.begin(), V.end(), [](const auto *L, const auto *R) {
250     return L->getName() < R->getName();
251   });
252   return {std::move(V)};
253 }
254 
255 class AMDGPULowerModuleLDS {
256   const AMDGPUTargetMachine &TM;
257 
258   static void
259   removeLocalVarsFromUsedLists(Module &M,
260                                const DenseSet<GlobalVariable *> &LocalVars) {
261     // The verifier rejects used lists containing an inttoptr of a constant
262     // so remove the variables from these lists before replaceAllUsesWith
263     SmallPtrSet<Constant *, 8> LocalVarsSet;
264     for (GlobalVariable *LocalVar : LocalVars)
265       LocalVarsSet.insert(cast<Constant>(LocalVar->stripPointerCasts()));
266 
267     removeFromUsedLists(
268         M, [&LocalVarsSet](Constant *C) { return LocalVarsSet.count(C); });
269 
270     for (GlobalVariable *LocalVar : LocalVars)
271       LocalVar->removeDeadConstantUsers();
272   }
273 
274   static void markUsedByKernel(Function *Func, GlobalVariable *SGV) {
275     // The llvm.amdgcn.module.lds instance is implicitly used by all kernels
276     // that might call a function which accesses a field within it. This is
277     // presently approximated to 'all kernels' if there are any such functions
278     // in the module. This implicit use is redefined as an explicit use here so
279     // that later passes, specifically PromoteAlloca, account for the required
280     // memory without any knowledge of this transform.
281 
282     // An operand bundle on llvm.donothing works because the call instruction
283     // survives until after the last pass that needs to account for LDS. It is
284     // better than inline asm as the latter survives until the end of codegen. A
285     // totally robust solution would be a function with the same semantics as
286     // llvm.donothing that takes a pointer to the instance and is lowered to a
287     // no-op after LDS is allocated, but that is not presently necessary.
288 
289     // This intrinsic is eliminated shortly before instruction selection. It
290     // does not suffice to indicate to ISel that a given global which is not
291     // immediately used by the kernel must still be allocated by it. An
292     // equivalent target specific intrinsic which lasts until immediately after
293     // codegen would suffice for that, but one would still need to ensure that
294     // the variables are allocated in the anticpated order.
295     BasicBlock *Entry = &Func->getEntryBlock();
296     IRBuilder<> Builder(Entry, Entry->getFirstNonPHIIt());
297 
298     Function *Decl =
299         Intrinsic::getDeclaration(Func->getParent(), Intrinsic::donothing, {});
300 
301     Value *UseInstance[1] = {
302         Builder.CreateConstInBoundsGEP1_32(SGV->getValueType(), SGV, 0)};
303 
304     Builder.CreateCall(
305         Decl, {}, {OperandBundleDefT<Value *>("ExplicitUse", UseInstance)});
306   }
307 
308   static bool eliminateConstantExprUsesOfLDSFromAllInstructions(Module &M) {
309     // Constants are uniqued within LLVM. A ConstantExpr referring to a LDS
310     // global may have uses from multiple different functions as a result.
311     // This pass specialises LDS variables with respect to the kernel that
312     // allocates them.
313 
314     // This is semantically equivalent to (the unimplemented as slow):
315     // for (auto &F : M.functions())
316     //   for (auto &BB : F)
317     //     for (auto &I : BB)
318     //       for (Use &Op : I.operands())
319     //         if (constantExprUsesLDS(Op))
320     //           replaceConstantExprInFunction(I, Op);
321 
322     SmallVector<Constant *> LDSGlobals;
323     for (auto &GV : M.globals())
324       if (AMDGPU::isLDSVariableToLower(GV))
325         LDSGlobals.push_back(&GV);
326 
327     return convertUsersOfConstantsToInstructions(LDSGlobals);
328   }
329 
330 public:
331   AMDGPULowerModuleLDS(const AMDGPUTargetMachine &TM_) : TM(TM_) {}
332 
333   using FunctionVariableMap = DenseMap<Function *, DenseSet<GlobalVariable *>>;
334 
335   using VariableFunctionMap = DenseMap<GlobalVariable *, DenseSet<Function *>>;
336 
337   static void getUsesOfLDSByFunction(CallGraph const &CG, Module &M,
338                                      FunctionVariableMap &kernels,
339                                      FunctionVariableMap &functions) {
340 
341     // Get uses from the current function, excluding uses by called functions
342     // Two output variables to avoid walking the globals list twice
343     for (auto &GV : M.globals()) {
344       if (!AMDGPU::isLDSVariableToLower(GV)) {
345         continue;
346       }
347 
348       if (GV.isAbsoluteSymbolRef()) {
349         report_fatal_error(
350             "LDS variables with absolute addresses are unimplemented.");
351       }
352 
353       for (User *V : GV.users()) {
354         if (auto *I = dyn_cast<Instruction>(V)) {
355           Function *F = I->getFunction();
356           if (isKernelLDS(F)) {
357             kernels[F].insert(&GV);
358           } else {
359             functions[F].insert(&GV);
360           }
361         }
362       }
363     }
364   }
365 
366   struct LDSUsesInfoTy {
367     FunctionVariableMap direct_access;
368     FunctionVariableMap indirect_access;
369   };
370 
371   static LDSUsesInfoTy getTransitiveUsesOfLDS(CallGraph const &CG, Module &M) {
372 
373     FunctionVariableMap direct_map_kernel;
374     FunctionVariableMap direct_map_function;
375     getUsesOfLDSByFunction(CG, M, direct_map_kernel, direct_map_function);
376 
377     // Collect variables that are used by functions whose address has escaped
378     DenseSet<GlobalVariable *> VariablesReachableThroughFunctionPointer;
379     for (Function &F : M.functions()) {
380       if (!isKernelLDS(&F))
381         if (F.hasAddressTaken(nullptr,
382                               /* IgnoreCallbackUses */ false,
383                               /* IgnoreAssumeLikeCalls */ false,
384                               /* IgnoreLLVMUsed */ true,
385                               /* IgnoreArcAttachedCall */ false)) {
386           set_union(VariablesReachableThroughFunctionPointer,
387                     direct_map_function[&F]);
388         }
389     }
390 
391     auto functionMakesUnknownCall = [&](const Function *F) -> bool {
392       assert(!F->isDeclaration());
393       for (const CallGraphNode::CallRecord &R : *CG[F]) {
394         if (!R.second->getFunction()) {
395           return true;
396         }
397       }
398       return false;
399     };
400 
401     // Work out which variables are reachable through function calls
402     FunctionVariableMap transitive_map_function = direct_map_function;
403 
404     // If the function makes any unknown call, assume the worst case that it can
405     // access all variables accessed by functions whose address escaped
406     for (Function &F : M.functions()) {
407       if (!F.isDeclaration() && functionMakesUnknownCall(&F)) {
408         if (!isKernelLDS(&F)) {
409           set_union(transitive_map_function[&F],
410                     VariablesReachableThroughFunctionPointer);
411         }
412       }
413     }
414 
415     // Direct implementation of collecting all variables reachable from each
416     // function
417     for (Function &Func : M.functions()) {
418       if (Func.isDeclaration() || isKernelLDS(&Func))
419         continue;
420 
421       DenseSet<Function *> seen; // catches cycles
422       SmallVector<Function *, 4> wip{&Func};
423 
424       while (!wip.empty()) {
425         Function *F = wip.pop_back_val();
426 
427         // Can accelerate this by referring to transitive map for functions that
428         // have already been computed, with more care than this
429         set_union(transitive_map_function[&Func], direct_map_function[F]);
430 
431         for (const CallGraphNode::CallRecord &R : *CG[F]) {
432           Function *ith = R.second->getFunction();
433           if (ith) {
434             if (!seen.contains(ith)) {
435               seen.insert(ith);
436               wip.push_back(ith);
437             }
438           }
439         }
440       }
441     }
442 
443     // direct_map_kernel lists which variables are used by the kernel
444     // find the variables which are used through a function call
445     FunctionVariableMap indirect_map_kernel;
446 
447     for (Function &Func : M.functions()) {
448       if (Func.isDeclaration() || !isKernelLDS(&Func))
449         continue;
450 
451       for (const CallGraphNode::CallRecord &R : *CG[&Func]) {
452         Function *ith = R.second->getFunction();
453         if (ith) {
454           set_union(indirect_map_kernel[&Func], transitive_map_function[ith]);
455         } else {
456           set_union(indirect_map_kernel[&Func],
457                     VariablesReachableThroughFunctionPointer);
458         }
459       }
460     }
461 
462     return {std::move(direct_map_kernel), std::move(indirect_map_kernel)};
463   }
464 
465   struct LDSVariableReplacement {
466     GlobalVariable *SGV = nullptr;
467     DenseMap<GlobalVariable *, Constant *> LDSVarsToConstantGEP;
468   };
469 
470   // remap from lds global to a constantexpr gep to where it has been moved to
471   // for each kernel
472   // an array with an element for each kernel containing where the corresponding
473   // variable was remapped to
474 
475   static Constant *getAddressesOfVariablesInKernel(
476       LLVMContext &Ctx, ArrayRef<GlobalVariable *> Variables,
477       const DenseMap<GlobalVariable *, Constant *> &LDSVarsToConstantGEP) {
478     // Create a ConstantArray containing the address of each Variable within the
479     // kernel corresponding to LDSVarsToConstantGEP, or poison if that kernel
480     // does not allocate it
481     // TODO: Drop the ptrtoint conversion
482 
483     Type *I32 = Type::getInt32Ty(Ctx);
484 
485     ArrayType *KernelOffsetsType = ArrayType::get(I32, Variables.size());
486 
487     SmallVector<Constant *> Elements;
488     for (size_t i = 0; i < Variables.size(); i++) {
489       GlobalVariable *GV = Variables[i];
490       auto ConstantGepIt = LDSVarsToConstantGEP.find(GV);
491       if (ConstantGepIt != LDSVarsToConstantGEP.end()) {
492         auto elt = ConstantExpr::getPtrToInt(ConstantGepIt->second, I32);
493         Elements.push_back(elt);
494       } else {
495         Elements.push_back(PoisonValue::get(I32));
496       }
497     }
498     return ConstantArray::get(KernelOffsetsType, Elements);
499   }
500 
501   static GlobalVariable *buildLookupTable(
502       Module &M, ArrayRef<GlobalVariable *> Variables,
503       ArrayRef<Function *> kernels,
504       DenseMap<Function *, LDSVariableReplacement> &KernelToReplacement) {
505     if (Variables.empty()) {
506       return nullptr;
507     }
508     LLVMContext &Ctx = M.getContext();
509 
510     const size_t NumberVariables = Variables.size();
511     const size_t NumberKernels = kernels.size();
512 
513     ArrayType *KernelOffsetsType =
514         ArrayType::get(Type::getInt32Ty(Ctx), NumberVariables);
515 
516     ArrayType *AllKernelsOffsetsType =
517         ArrayType::get(KernelOffsetsType, NumberKernels);
518 
519     Constant *Missing = PoisonValue::get(KernelOffsetsType);
520     std::vector<Constant *> overallConstantExprElts(NumberKernels);
521     for (size_t i = 0; i < NumberKernels; i++) {
522       auto Replacement = KernelToReplacement.find(kernels[i]);
523       overallConstantExprElts[i] =
524           (Replacement == KernelToReplacement.end())
525               ? Missing
526               : getAddressesOfVariablesInKernel(
527                     Ctx, Variables, Replacement->second.LDSVarsToConstantGEP);
528     }
529 
530     Constant *init =
531         ConstantArray::get(AllKernelsOffsetsType, overallConstantExprElts);
532 
533     return new GlobalVariable(
534         M, AllKernelsOffsetsType, true, GlobalValue::InternalLinkage, init,
535         "llvm.amdgcn.lds.offset.table", nullptr, GlobalValue::NotThreadLocal,
536         AMDGPUAS::CONSTANT_ADDRESS);
537   }
538 
539   void replaceUseWithTableLookup(Module &M, IRBuilder<> &Builder,
540                                  GlobalVariable *LookupTable,
541                                  GlobalVariable *GV, Use &U,
542                                  Value *OptionalIndex) {
543     // Table is a constant array of the same length as OrderedKernels
544     LLVMContext &Ctx = M.getContext();
545     Type *I32 = Type::getInt32Ty(Ctx);
546     auto *I = cast<Instruction>(U.getUser());
547 
548     Value *tableKernelIndex = getTableLookupKernelIndex(M, I->getFunction());
549 
550     if (auto *Phi = dyn_cast<PHINode>(I)) {
551       BasicBlock *BB = Phi->getIncomingBlock(U);
552       Builder.SetInsertPoint(&(*(BB->getFirstInsertionPt())));
553     } else {
554       Builder.SetInsertPoint(I);
555     }
556 
557     SmallVector<Value *, 3> GEPIdx = {
558         ConstantInt::get(I32, 0),
559         tableKernelIndex,
560     };
561     if (OptionalIndex)
562       GEPIdx.push_back(OptionalIndex);
563 
564     Value *Address = Builder.CreateInBoundsGEP(
565         LookupTable->getValueType(), LookupTable, GEPIdx, GV->getName());
566 
567     Value *loaded = Builder.CreateLoad(I32, Address);
568 
569     Value *replacement =
570         Builder.CreateIntToPtr(loaded, GV->getType(), GV->getName());
571 
572     U.set(replacement);
573   }
574 
575   void replaceUsesInInstructionsWithTableLookup(
576       Module &M, ArrayRef<GlobalVariable *> ModuleScopeVariables,
577       GlobalVariable *LookupTable) {
578 
579     LLVMContext &Ctx = M.getContext();
580     IRBuilder<> Builder(Ctx);
581     Type *I32 = Type::getInt32Ty(Ctx);
582 
583     for (size_t Index = 0; Index < ModuleScopeVariables.size(); Index++) {
584       auto *GV = ModuleScopeVariables[Index];
585 
586       for (Use &U : make_early_inc_range(GV->uses())) {
587         auto *I = dyn_cast<Instruction>(U.getUser());
588         if (!I)
589           continue;
590 
591         replaceUseWithTableLookup(M, Builder, LookupTable, GV, U,
592                                   ConstantInt::get(I32, Index));
593       }
594     }
595   }
596 
597   static DenseSet<Function *> kernelsThatIndirectlyAccessAnyOfPassedVariables(
598       Module &M, LDSUsesInfoTy &LDSUsesInfo,
599       DenseSet<GlobalVariable *> const &VariableSet) {
600 
601     DenseSet<Function *> KernelSet;
602 
603     if (VariableSet.empty())
604       return KernelSet;
605 
606     for (Function &Func : M.functions()) {
607       if (Func.isDeclaration() || !isKernelLDS(&Func))
608         continue;
609       for (GlobalVariable *GV : LDSUsesInfo.indirect_access[&Func]) {
610         if (VariableSet.contains(GV)) {
611           KernelSet.insert(&Func);
612           break;
613         }
614       }
615     }
616 
617     return KernelSet;
618   }
619 
620   static GlobalVariable *
621   chooseBestVariableForModuleStrategy(const DataLayout &DL,
622                                       VariableFunctionMap &LDSVars) {
623     // Find the global variable with the most indirect uses from kernels
624 
625     struct CandidateTy {
626       GlobalVariable *GV = nullptr;
627       size_t UserCount = 0;
628       size_t Size = 0;
629 
630       CandidateTy() = default;
631 
632       CandidateTy(GlobalVariable *GV, uint64_t UserCount, uint64_t AllocSize)
633           : GV(GV), UserCount(UserCount), Size(AllocSize) {}
634 
635       bool operator<(const CandidateTy &Other) const {
636         // Fewer users makes module scope variable less attractive
637         if (UserCount < Other.UserCount) {
638           return true;
639         }
640         if (UserCount > Other.UserCount) {
641           return false;
642         }
643 
644         // Bigger makes module scope variable less attractive
645         if (Size < Other.Size) {
646           return false;
647         }
648 
649         if (Size > Other.Size) {
650           return true;
651         }
652 
653         // Arbitrary but consistent
654         return GV->getName() < Other.GV->getName();
655       }
656     };
657 
658     CandidateTy MostUsed;
659 
660     for (auto &K : LDSVars) {
661       GlobalVariable *GV = K.first;
662       if (K.second.size() <= 1) {
663         // A variable reachable by only one kernel is best lowered with kernel
664         // strategy
665         continue;
666       }
667       CandidateTy Candidate(
668           GV, K.second.size(),
669           DL.getTypeAllocSize(GV->getValueType()).getFixedValue());
670       if (MostUsed < Candidate)
671         MostUsed = Candidate;
672     }
673 
674     return MostUsed.GV;
675   }
676 
677   static void recordLDSAbsoluteAddress(Module *M, GlobalVariable *GV,
678                                        uint32_t Address) {
679     // Write the specified address into metadata where it can be retrieved by
680     // the assembler. Format is a half open range, [Address Address+1)
681     LLVMContext &Ctx = M->getContext();
682     auto *IntTy =
683         M->getDataLayout().getIntPtrType(Ctx, AMDGPUAS::LOCAL_ADDRESS);
684     auto *MinC = ConstantAsMetadata::get(ConstantInt::get(IntTy, Address));
685     auto *MaxC = ConstantAsMetadata::get(ConstantInt::get(IntTy, Address + 1));
686     GV->setMetadata(LLVMContext::MD_absolute_symbol,
687                     MDNode::get(Ctx, {MinC, MaxC}));
688   }
689 
690   DenseMap<Function *, Value *> tableKernelIndexCache;
691   Value *getTableLookupKernelIndex(Module &M, Function *F) {
692     // Accesses from a function use the amdgcn_lds_kernel_id intrinsic which
693     // lowers to a read from a live in register. Emit it once in the entry
694     // block to spare deduplicating it later.
695     auto [It, Inserted] = tableKernelIndexCache.try_emplace(F);
696     if (Inserted) {
697       Function *Decl =
698           Intrinsic::getDeclaration(&M, Intrinsic::amdgcn_lds_kernel_id, {});
699 
700       auto InsertAt = F->getEntryBlock().getFirstNonPHIOrDbgOrAlloca();
701       IRBuilder<> Builder(&*InsertAt);
702 
703       It->second = Builder.CreateCall(Decl, {});
704     }
705 
706     return It->second;
707   }
708 
709   static std::vector<Function *> assignLDSKernelIDToEachKernel(
710       Module *M, DenseSet<Function *> const &KernelsThatAllocateTableLDS,
711       DenseSet<Function *> const &KernelsThatIndirectlyAllocateDynamicLDS) {
712     // Associate kernels in the set with an arbirary but reproducible order and
713     // annotate them with that order in metadata. This metadata is recognised by
714     // the backend and lowered to a SGPR which can be read from using
715     // amdgcn_lds_kernel_id.
716 
717     std::vector<Function *> OrderedKernels;
718     if (!KernelsThatAllocateTableLDS.empty() ||
719         !KernelsThatIndirectlyAllocateDynamicLDS.empty()) {
720 
721       for (Function &Func : M->functions()) {
722         if (Func.isDeclaration())
723           continue;
724         if (!isKernelLDS(&Func))
725           continue;
726 
727         if (KernelsThatAllocateTableLDS.contains(&Func) ||
728             KernelsThatIndirectlyAllocateDynamicLDS.contains(&Func)) {
729           assert(Func.hasName()); // else fatal error earlier
730           OrderedKernels.push_back(&Func);
731         }
732       }
733 
734       // Put them in an arbitrary but reproducible order
735       OrderedKernels = sortByName(std::move(OrderedKernels));
736 
737       // Annotate the kernels with their order in this vector
738       LLVMContext &Ctx = M->getContext();
739       IRBuilder<> Builder(Ctx);
740 
741       if (OrderedKernels.size() > UINT32_MAX) {
742         // 32 bit keeps it in one SGPR. > 2**32 kernels won't fit on the GPU
743         report_fatal_error("Unimplemented LDS lowering for > 2**32 kernels");
744       }
745 
746       for (size_t i = 0; i < OrderedKernels.size(); i++) {
747         Metadata *AttrMDArgs[1] = {
748             ConstantAsMetadata::get(Builder.getInt32(i)),
749         };
750         OrderedKernels[i]->setMetadata("llvm.amdgcn.lds.kernel.id",
751                                        MDNode::get(Ctx, AttrMDArgs));
752       }
753     }
754     return OrderedKernels;
755   }
756 
757   static void partitionVariablesIntoIndirectStrategies(
758       Module &M, LDSUsesInfoTy const &LDSUsesInfo,
759       VariableFunctionMap &LDSToKernelsThatNeedToAccessItIndirectly,
760       DenseSet<GlobalVariable *> &ModuleScopeVariables,
761       DenseSet<GlobalVariable *> &TableLookupVariables,
762       DenseSet<GlobalVariable *> &KernelAccessVariables,
763       DenseSet<GlobalVariable *> &DynamicVariables) {
764 
765     GlobalVariable *HybridModuleRoot =
766         LoweringKindLoc != LoweringKind::hybrid
767             ? nullptr
768             : chooseBestVariableForModuleStrategy(
769                   M.getDataLayout(), LDSToKernelsThatNeedToAccessItIndirectly);
770 
771     DenseSet<Function *> const EmptySet;
772     DenseSet<Function *> const &HybridModuleRootKernels =
773         HybridModuleRoot
774             ? LDSToKernelsThatNeedToAccessItIndirectly[HybridModuleRoot]
775             : EmptySet;
776 
777     for (auto &K : LDSToKernelsThatNeedToAccessItIndirectly) {
778       // Each iteration of this loop assigns exactly one global variable to
779       // exactly one of the implementation strategies.
780 
781       GlobalVariable *GV = K.first;
782       assert(AMDGPU::isLDSVariableToLower(*GV));
783       assert(K.second.size() != 0);
784 
785       if (AMDGPU::isDynamicLDS(*GV)) {
786         DynamicVariables.insert(GV);
787         continue;
788       }
789 
790       switch (LoweringKindLoc) {
791       case LoweringKind::module:
792         ModuleScopeVariables.insert(GV);
793         break;
794 
795       case LoweringKind::table:
796         TableLookupVariables.insert(GV);
797         break;
798 
799       case LoweringKind::kernel:
800         if (K.second.size() == 1) {
801           KernelAccessVariables.insert(GV);
802         } else {
803           report_fatal_error(
804               "cannot lower LDS '" + GV->getName() +
805               "' to kernel access as it is reachable from multiple kernels");
806         }
807         break;
808 
809       case LoweringKind::hybrid: {
810         if (GV == HybridModuleRoot) {
811           assert(K.second.size() != 1);
812           ModuleScopeVariables.insert(GV);
813         } else if (K.second.size() == 1) {
814           KernelAccessVariables.insert(GV);
815         } else if (set_is_subset(K.second, HybridModuleRootKernels)) {
816           ModuleScopeVariables.insert(GV);
817         } else {
818           TableLookupVariables.insert(GV);
819         }
820         break;
821       }
822       }
823     }
824 
825     // All LDS variables accessed indirectly have now been partitioned into
826     // the distinct lowering strategies.
827     assert(ModuleScopeVariables.size() + TableLookupVariables.size() +
828                KernelAccessVariables.size() + DynamicVariables.size() ==
829            LDSToKernelsThatNeedToAccessItIndirectly.size());
830   }
831 
832   static GlobalVariable *lowerModuleScopeStructVariables(
833       Module &M, DenseSet<GlobalVariable *> const &ModuleScopeVariables,
834       DenseSet<Function *> const &KernelsThatAllocateModuleLDS) {
835     // Create a struct to hold the ModuleScopeVariables
836     // Replace all uses of those variables from non-kernel functions with the
837     // new struct instance Replace only the uses from kernel functions that will
838     // allocate this instance. That is a space optimisation - kernels that use a
839     // subset of the module scope struct and do not need to allocate it for
840     // indirect calls will only allocate the subset they use (they do so as part
841     // of the per-kernel lowering).
842     if (ModuleScopeVariables.empty()) {
843       return nullptr;
844     }
845 
846     LLVMContext &Ctx = M.getContext();
847 
848     LDSVariableReplacement ModuleScopeReplacement =
849         createLDSVariableReplacement(M, "llvm.amdgcn.module.lds",
850                                      ModuleScopeVariables);
851 
852     appendToCompilerUsed(M, {static_cast<GlobalValue *>(
853                                 ConstantExpr::getPointerBitCastOrAddrSpaceCast(
854                                     cast<Constant>(ModuleScopeReplacement.SGV),
855                                     PointerType::getUnqual(Ctx)))});
856 
857     // module.lds will be allocated at zero in any kernel that allocates it
858     recordLDSAbsoluteAddress(&M, ModuleScopeReplacement.SGV, 0);
859 
860     // historic
861     removeLocalVarsFromUsedLists(M, ModuleScopeVariables);
862 
863     // Replace all uses of module scope variable from non-kernel functions
864     replaceLDSVariablesWithStruct(
865         M, ModuleScopeVariables, ModuleScopeReplacement, [&](Use &U) {
866           Instruction *I = dyn_cast<Instruction>(U.getUser());
867           if (!I) {
868             return false;
869           }
870           Function *F = I->getFunction();
871           return !isKernelLDS(F);
872         });
873 
874     // Replace uses of module scope variable from kernel functions that
875     // allocate the module scope variable, otherwise leave them unchanged
876     // Record on each kernel whether the module scope global is used by it
877 
878     for (Function &Func : M.functions()) {
879       if (Func.isDeclaration() || !isKernelLDS(&Func))
880         continue;
881 
882       if (KernelsThatAllocateModuleLDS.contains(&Func)) {
883         replaceLDSVariablesWithStruct(
884             M, ModuleScopeVariables, ModuleScopeReplacement, [&](Use &U) {
885               Instruction *I = dyn_cast<Instruction>(U.getUser());
886               if (!I) {
887                 return false;
888               }
889               Function *F = I->getFunction();
890               return F == &Func;
891             });
892 
893         markUsedByKernel(&Func, ModuleScopeReplacement.SGV);
894       }
895     }
896 
897     return ModuleScopeReplacement.SGV;
898   }
899 
900   static DenseMap<Function *, LDSVariableReplacement>
901   lowerKernelScopeStructVariables(
902       Module &M, LDSUsesInfoTy &LDSUsesInfo,
903       DenseSet<GlobalVariable *> const &ModuleScopeVariables,
904       DenseSet<Function *> const &KernelsThatAllocateModuleLDS,
905       GlobalVariable *MaybeModuleScopeStruct) {
906 
907     // Create a struct for each kernel for the non-module-scope variables.
908 
909     DenseMap<Function *, LDSVariableReplacement> KernelToReplacement;
910     for (Function &Func : M.functions()) {
911       if (Func.isDeclaration() || !isKernelLDS(&Func))
912         continue;
913 
914       DenseSet<GlobalVariable *> KernelUsedVariables;
915       // Allocating variables that are used directly in this struct to get
916       // alignment aware allocation and predictable frame size.
917       for (auto &v : LDSUsesInfo.direct_access[&Func]) {
918         if (!AMDGPU::isDynamicLDS(*v)) {
919           KernelUsedVariables.insert(v);
920         }
921       }
922 
923       // Allocating variables that are accessed indirectly so that a lookup of
924       // this struct instance can find them from nested functions.
925       for (auto &v : LDSUsesInfo.indirect_access[&Func]) {
926         if (!AMDGPU::isDynamicLDS(*v)) {
927           KernelUsedVariables.insert(v);
928         }
929       }
930 
931       // Variables allocated in module lds must all resolve to that struct,
932       // not to the per-kernel instance.
933       if (KernelsThatAllocateModuleLDS.contains(&Func)) {
934         for (GlobalVariable *v : ModuleScopeVariables) {
935           KernelUsedVariables.erase(v);
936         }
937       }
938 
939       if (KernelUsedVariables.empty()) {
940         // Either used no LDS, or the LDS it used was all in the module struct
941         // or dynamically sized
942         continue;
943       }
944 
945       // The association between kernel function and LDS struct is done by
946       // symbol name, which only works if the function in question has a
947       // name This is not expected to be a problem in practice as kernels
948       // are called by name making anonymous ones (which are named by the
949       // backend) difficult to use. This does mean that llvm test cases need
950       // to name the kernels.
951       if (!Func.hasName()) {
952         report_fatal_error("Anonymous kernels cannot use LDS variables");
953       }
954 
955       std::string VarName =
956           (Twine("llvm.amdgcn.kernel.") + Func.getName() + ".lds").str();
957 
958       auto Replacement =
959           createLDSVariableReplacement(M, VarName, KernelUsedVariables);
960 
961       // If any indirect uses, create a direct use to ensure allocation
962       // TODO: Simpler to unconditionally mark used but that regresses
963       // codegen in test/CodeGen/AMDGPU/noclobber-barrier.ll
964       auto Accesses = LDSUsesInfo.indirect_access.find(&Func);
965       if ((Accesses != LDSUsesInfo.indirect_access.end()) &&
966           !Accesses->second.empty())
967         markUsedByKernel(&Func, Replacement.SGV);
968 
969       // remove preserves existing codegen
970       removeLocalVarsFromUsedLists(M, KernelUsedVariables);
971       KernelToReplacement[&Func] = Replacement;
972 
973       // Rewrite uses within kernel to the new struct
974       replaceLDSVariablesWithStruct(
975           M, KernelUsedVariables, Replacement, [&Func](Use &U) {
976             Instruction *I = dyn_cast<Instruction>(U.getUser());
977             return I && I->getFunction() == &Func;
978           });
979     }
980     return KernelToReplacement;
981   }
982 
983   static GlobalVariable *
984   buildRepresentativeDynamicLDSInstance(Module &M, LDSUsesInfoTy &LDSUsesInfo,
985                                         Function *func) {
986     // Create a dynamic lds variable with a name associated with the passed
987     // function that has the maximum alignment of any dynamic lds variable
988     // reachable from this kernel. Dynamic LDS is allocated after the static LDS
989     // allocation, possibly after alignment padding. The representative variable
990     // created here has the maximum alignment of any other dynamic variable
991     // reachable by that kernel. All dynamic LDS variables are allocated at the
992     // same address in each kernel in order to provide the documented aliasing
993     // semantics. Setting the alignment here allows this IR pass to accurately
994     // predict the exact constant at which it will be allocated.
995 
996     assert(isKernelLDS(func));
997 
998     LLVMContext &Ctx = M.getContext();
999     const DataLayout &DL = M.getDataLayout();
1000     Align MaxDynamicAlignment(1);
1001 
1002     auto UpdateMaxAlignment = [&MaxDynamicAlignment, &DL](GlobalVariable *GV) {
1003       if (AMDGPU::isDynamicLDS(*GV)) {
1004         MaxDynamicAlignment =
1005             std::max(MaxDynamicAlignment, AMDGPU::getAlign(DL, GV));
1006       }
1007     };
1008 
1009     for (GlobalVariable *GV : LDSUsesInfo.indirect_access[func]) {
1010       UpdateMaxAlignment(GV);
1011     }
1012 
1013     for (GlobalVariable *GV : LDSUsesInfo.direct_access[func]) {
1014       UpdateMaxAlignment(GV);
1015     }
1016 
1017     assert(func->hasName()); // Checked by caller
1018     auto emptyCharArray = ArrayType::get(Type::getInt8Ty(Ctx), 0);
1019     GlobalVariable *N = new GlobalVariable(
1020         M, emptyCharArray, false, GlobalValue::ExternalLinkage, nullptr,
1021         Twine("llvm.amdgcn." + func->getName() + ".dynlds"), nullptr, GlobalValue::NotThreadLocal, AMDGPUAS::LOCAL_ADDRESS,
1022         false);
1023     N->setAlignment(MaxDynamicAlignment);
1024 
1025     assert(AMDGPU::isDynamicLDS(*N));
1026     return N;
1027   }
1028 
1029   /// Strip "amdgpu-no-lds-kernel-id" from any functions where we may have
1030   /// introduced its use. If AMDGPUAttributor ran prior to the pass, we inferred
1031   /// the lack of llvm.amdgcn.lds.kernel.id calls.
1032   void removeNoLdsKernelIdFromReachable(CallGraph &CG, Function *KernelRoot) {
1033     KernelRoot->removeFnAttr("amdgpu-no-lds-kernel-id");
1034 
1035     SmallVector<Function *> Tmp({CG[KernelRoot]->getFunction()});
1036     if (!Tmp.back())
1037       return;
1038 
1039     SmallPtrSet<Function *, 8> Visited;
1040     bool SeenUnknownCall = false;
1041 
1042     do {
1043       Function *F = Tmp.pop_back_val();
1044 
1045       for (auto &N : *CG[F]) {
1046         if (!N.second)
1047           continue;
1048 
1049         Function *Callee = N.second->getFunction();
1050         if (!Callee) {
1051           if (!SeenUnknownCall) {
1052             SeenUnknownCall = true;
1053 
1054             // If we see any indirect calls, assume nothing about potential
1055             // targets.
1056             // TODO: This could be refined to possible LDS global users.
1057             for (auto &N : *CG.getExternalCallingNode()) {
1058               Function *PotentialCallee = N.second->getFunction();
1059               if (!isKernelLDS(PotentialCallee))
1060                 PotentialCallee->removeFnAttr("amdgpu-no-lds-kernel-id");
1061             }
1062 
1063             continue;
1064           }
1065         }
1066 
1067         Callee->removeFnAttr("amdgpu-no-lds-kernel-id");
1068         if (Visited.insert(Callee).second)
1069           Tmp.push_back(Callee);
1070       }
1071     } while (!Tmp.empty());
1072   }
1073 
1074   DenseMap<Function *, GlobalVariable *> lowerDynamicLDSVariables(
1075       Module &M, LDSUsesInfoTy &LDSUsesInfo,
1076       DenseSet<Function *> const &KernelsThatIndirectlyAllocateDynamicLDS,
1077       DenseSet<GlobalVariable *> const &DynamicVariables,
1078       std::vector<Function *> const &OrderedKernels) {
1079     DenseMap<Function *, GlobalVariable *> KernelToCreatedDynamicLDS;
1080     if (!KernelsThatIndirectlyAllocateDynamicLDS.empty()) {
1081       LLVMContext &Ctx = M.getContext();
1082       IRBuilder<> Builder(Ctx);
1083       Type *I32 = Type::getInt32Ty(Ctx);
1084 
1085       std::vector<Constant *> newDynamicLDS;
1086 
1087       // Table is built in the same order as OrderedKernels
1088       for (auto &func : OrderedKernels) {
1089 
1090         if (KernelsThatIndirectlyAllocateDynamicLDS.contains(func)) {
1091           assert(isKernelLDS(func));
1092           if (!func->hasName()) {
1093             report_fatal_error("Anonymous kernels cannot use LDS variables");
1094           }
1095 
1096           GlobalVariable *N =
1097               buildRepresentativeDynamicLDSInstance(M, LDSUsesInfo, func);
1098 
1099           KernelToCreatedDynamicLDS[func] = N;
1100 
1101           markUsedByKernel(func, N);
1102 
1103           auto emptyCharArray = ArrayType::get(Type::getInt8Ty(Ctx), 0);
1104           auto GEP = ConstantExpr::getGetElementPtr(
1105               emptyCharArray, N, ConstantInt::get(I32, 0), true);
1106           newDynamicLDS.push_back(ConstantExpr::getPtrToInt(GEP, I32));
1107         } else {
1108           newDynamicLDS.push_back(PoisonValue::get(I32));
1109         }
1110       }
1111       assert(OrderedKernels.size() == newDynamicLDS.size());
1112 
1113       ArrayType *t = ArrayType::get(I32, newDynamicLDS.size());
1114       Constant *init = ConstantArray::get(t, newDynamicLDS);
1115       GlobalVariable *table = new GlobalVariable(
1116           M, t, true, GlobalValue::InternalLinkage, init,
1117           "llvm.amdgcn.dynlds.offset.table", nullptr,
1118           GlobalValue::NotThreadLocal, AMDGPUAS::CONSTANT_ADDRESS);
1119 
1120       for (GlobalVariable *GV : DynamicVariables) {
1121         for (Use &U : make_early_inc_range(GV->uses())) {
1122           auto *I = dyn_cast<Instruction>(U.getUser());
1123           if (!I)
1124             continue;
1125           if (isKernelLDS(I->getFunction()))
1126             continue;
1127 
1128           replaceUseWithTableLookup(M, Builder, table, GV, U, nullptr);
1129         }
1130       }
1131     }
1132     return KernelToCreatedDynamicLDS;
1133   }
1134 
1135   bool runOnModule(Module &M) {
1136     CallGraph CG = CallGraph(M);
1137     bool Changed = superAlignLDSGlobals(M);
1138 
1139     Changed |= eliminateConstantExprUsesOfLDSFromAllInstructions(M);
1140 
1141     Changed = true; // todo: narrow this down
1142 
1143     // For each kernel, what variables does it access directly or through
1144     // callees
1145     LDSUsesInfoTy LDSUsesInfo = getTransitiveUsesOfLDS(CG, M);
1146 
1147     // For each variable accessed through callees, which kernels access it
1148     VariableFunctionMap LDSToKernelsThatNeedToAccessItIndirectly;
1149     for (auto &K : LDSUsesInfo.indirect_access) {
1150       Function *F = K.first;
1151       assert(isKernelLDS(F));
1152       for (GlobalVariable *GV : K.second) {
1153         LDSToKernelsThatNeedToAccessItIndirectly[GV].insert(F);
1154       }
1155     }
1156 
1157     // Partition variables accessed indirectly into the different strategies
1158     DenseSet<GlobalVariable *> ModuleScopeVariables;
1159     DenseSet<GlobalVariable *> TableLookupVariables;
1160     DenseSet<GlobalVariable *> KernelAccessVariables;
1161     DenseSet<GlobalVariable *> DynamicVariables;
1162     partitionVariablesIntoIndirectStrategies(
1163         M, LDSUsesInfo, LDSToKernelsThatNeedToAccessItIndirectly,
1164         ModuleScopeVariables, TableLookupVariables, KernelAccessVariables,
1165         DynamicVariables);
1166 
1167     // If the kernel accesses a variable that is going to be stored in the
1168     // module instance through a call then that kernel needs to allocate the
1169     // module instance
1170     const DenseSet<Function *> KernelsThatAllocateModuleLDS =
1171         kernelsThatIndirectlyAccessAnyOfPassedVariables(M, LDSUsesInfo,
1172                                                         ModuleScopeVariables);
1173     const DenseSet<Function *> KernelsThatAllocateTableLDS =
1174         kernelsThatIndirectlyAccessAnyOfPassedVariables(M, LDSUsesInfo,
1175                                                         TableLookupVariables);
1176 
1177     const DenseSet<Function *> KernelsThatIndirectlyAllocateDynamicLDS =
1178         kernelsThatIndirectlyAccessAnyOfPassedVariables(M, LDSUsesInfo,
1179                                                         DynamicVariables);
1180 
1181     GlobalVariable *MaybeModuleScopeStruct = lowerModuleScopeStructVariables(
1182         M, ModuleScopeVariables, KernelsThatAllocateModuleLDS);
1183 
1184     DenseMap<Function *, LDSVariableReplacement> KernelToReplacement =
1185         lowerKernelScopeStructVariables(M, LDSUsesInfo, ModuleScopeVariables,
1186                                         KernelsThatAllocateModuleLDS,
1187                                         MaybeModuleScopeStruct);
1188 
1189     // Lower zero cost accesses to the kernel instances just created
1190     for (auto &GV : KernelAccessVariables) {
1191       auto &funcs = LDSToKernelsThatNeedToAccessItIndirectly[GV];
1192       assert(funcs.size() == 1); // Only one kernel can access it
1193       LDSVariableReplacement Replacement =
1194           KernelToReplacement[*(funcs.begin())];
1195 
1196       DenseSet<GlobalVariable *> Vec;
1197       Vec.insert(GV);
1198 
1199       replaceLDSVariablesWithStruct(M, Vec, Replacement, [](Use &U) {
1200         return isa<Instruction>(U.getUser());
1201       });
1202     }
1203 
1204     // The ith element of this vector is kernel id i
1205     std::vector<Function *> OrderedKernels =
1206         assignLDSKernelIDToEachKernel(&M, KernelsThatAllocateTableLDS,
1207                                       KernelsThatIndirectlyAllocateDynamicLDS);
1208 
1209     if (!KernelsThatAllocateTableLDS.empty()) {
1210       LLVMContext &Ctx = M.getContext();
1211       IRBuilder<> Builder(Ctx);
1212 
1213       // The order must be consistent between lookup table and accesses to
1214       // lookup table
1215       auto TableLookupVariablesOrdered =
1216           sortByName(std::vector<GlobalVariable *>(TableLookupVariables.begin(),
1217                                                    TableLookupVariables.end()));
1218 
1219       GlobalVariable *LookupTable = buildLookupTable(
1220           M, TableLookupVariablesOrdered, OrderedKernels, KernelToReplacement);
1221       replaceUsesInInstructionsWithTableLookup(M, TableLookupVariablesOrdered,
1222                                                LookupTable);
1223 
1224       // Strip amdgpu-no-lds-kernel-id from all functions reachable from the
1225       // kernel. We may have inferred this wasn't used prior to the pass.
1226       //
1227       // TODO: We could filter out subgraphs that do not access LDS globals.
1228       for (Function *F : KernelsThatAllocateTableLDS)
1229         removeNoLdsKernelIdFromReachable(CG, F);
1230     }
1231 
1232     DenseMap<Function *, GlobalVariable *> KernelToCreatedDynamicLDS =
1233         lowerDynamicLDSVariables(M, LDSUsesInfo,
1234                                  KernelsThatIndirectlyAllocateDynamicLDS,
1235                                  DynamicVariables, OrderedKernels);
1236 
1237     // All kernel frames have been allocated. Calculate and record the
1238     // addresses.
1239     {
1240       const DataLayout &DL = M.getDataLayout();
1241 
1242       for (Function &Func : M.functions()) {
1243         if (Func.isDeclaration() || !isKernelLDS(&Func))
1244           continue;
1245 
1246         // All three of these are optional. The first variable is allocated at
1247         // zero. They are allocated by AMDGPUMachineFunction as one block.
1248         // Layout:
1249         //{
1250         //  module.lds
1251         //  alignment padding
1252         //  kernel instance
1253         //  alignment padding
1254         //  dynamic lds variables
1255         //}
1256 
1257         const bool AllocateModuleScopeStruct =
1258             MaybeModuleScopeStruct &&
1259             KernelsThatAllocateModuleLDS.contains(&Func);
1260 
1261         auto Replacement = KernelToReplacement.find(&Func);
1262         const bool AllocateKernelScopeStruct =
1263             Replacement != KernelToReplacement.end();
1264 
1265         const bool AllocateDynamicVariable =
1266             KernelToCreatedDynamicLDS.contains(&Func);
1267 
1268         uint32_t Offset = 0;
1269 
1270         if (AllocateModuleScopeStruct) {
1271           // Allocated at zero, recorded once on construction, not once per
1272           // kernel
1273           Offset += DL.getTypeAllocSize(MaybeModuleScopeStruct->getValueType());
1274         }
1275 
1276         if (AllocateKernelScopeStruct) {
1277           GlobalVariable *KernelStruct = Replacement->second.SGV;
1278           Offset = alignTo(Offset, AMDGPU::getAlign(DL, KernelStruct));
1279           recordLDSAbsoluteAddress(&M, KernelStruct, Offset);
1280           Offset += DL.getTypeAllocSize(KernelStruct->getValueType());
1281         }
1282 
1283         // If there is dynamic allocation, the alignment needed is included in
1284         // the static frame size. There may be no reference to the dynamic
1285         // variable in the kernel itself, so without including it here, that
1286         // alignment padding could be missed.
1287         if (AllocateDynamicVariable) {
1288           GlobalVariable *DynamicVariable = KernelToCreatedDynamicLDS[&Func];
1289           Offset = alignTo(Offset, AMDGPU::getAlign(DL, DynamicVariable));
1290           recordLDSAbsoluteAddress(&M, DynamicVariable, Offset);
1291         }
1292 
1293         if (Offset != 0) {
1294           (void)TM; // TODO: Account for target maximum LDS
1295           std::string Buffer;
1296           raw_string_ostream SS{Buffer};
1297           SS << format("%u", Offset);
1298 
1299           // Instead of explictly marking kernels that access dynamic variables
1300           // using special case metadata, annotate with min-lds == max-lds, i.e.
1301           // that there is no more space available for allocating more static
1302           // LDS variables. That is the right condition to prevent allocating
1303           // more variables which would collide with the addresses assigned to
1304           // dynamic variables.
1305           if (AllocateDynamicVariable)
1306             SS << format(",%u", Offset);
1307 
1308           Func.addFnAttr("amdgpu-lds-size", Buffer);
1309         }
1310       }
1311     }
1312 
1313     for (auto &GV : make_early_inc_range(M.globals()))
1314       if (AMDGPU::isLDSVariableToLower(GV)) {
1315         // probably want to remove from used lists
1316         GV.removeDeadConstantUsers();
1317         if (GV.use_empty())
1318           GV.eraseFromParent();
1319       }
1320 
1321     return Changed;
1322   }
1323 
1324 private:
1325   // Increase the alignment of LDS globals if necessary to maximise the chance
1326   // that we can use aligned LDS instructions to access them.
1327   static bool superAlignLDSGlobals(Module &M) {
1328     const DataLayout &DL = M.getDataLayout();
1329     bool Changed = false;
1330     if (!SuperAlignLDSGlobals) {
1331       return Changed;
1332     }
1333 
1334     for (auto &GV : M.globals()) {
1335       if (GV.getType()->getPointerAddressSpace() != AMDGPUAS::LOCAL_ADDRESS) {
1336         // Only changing alignment of LDS variables
1337         continue;
1338       }
1339       if (!GV.hasInitializer()) {
1340         // cuda/hip extern __shared__ variable, leave alignment alone
1341         continue;
1342       }
1343 
1344       Align Alignment = AMDGPU::getAlign(DL, &GV);
1345       TypeSize GVSize = DL.getTypeAllocSize(GV.getValueType());
1346 
1347       if (GVSize > 8) {
1348         // We might want to use a b96 or b128 load/store
1349         Alignment = std::max(Alignment, Align(16));
1350       } else if (GVSize > 4) {
1351         // We might want to use a b64 load/store
1352         Alignment = std::max(Alignment, Align(8));
1353       } else if (GVSize > 2) {
1354         // We might want to use a b32 load/store
1355         Alignment = std::max(Alignment, Align(4));
1356       } else if (GVSize > 1) {
1357         // We might want to use a b16 load/store
1358         Alignment = std::max(Alignment, Align(2));
1359       }
1360 
1361       if (Alignment != AMDGPU::getAlign(DL, &GV)) {
1362         Changed = true;
1363         GV.setAlignment(Alignment);
1364       }
1365     }
1366     return Changed;
1367   }
1368 
1369   static LDSVariableReplacement createLDSVariableReplacement(
1370       Module &M, std::string VarName,
1371       DenseSet<GlobalVariable *> const &LDSVarsToTransform) {
1372     // Create a struct instance containing LDSVarsToTransform and map from those
1373     // variables to ConstantExprGEP
1374     // Variables may be introduced to meet alignment requirements. No aliasing
1375     // metadata is useful for these as they have no uses. Erased before return.
1376 
1377     LLVMContext &Ctx = M.getContext();
1378     const DataLayout &DL = M.getDataLayout();
1379     assert(!LDSVarsToTransform.empty());
1380 
1381     SmallVector<OptimizedStructLayoutField, 8> LayoutFields;
1382     LayoutFields.reserve(LDSVarsToTransform.size());
1383     {
1384       // The order of fields in this struct depends on the order of
1385       // varables in the argument which varies when changing how they
1386       // are identified, leading to spurious test breakage.
1387       auto Sorted = sortByName(std::vector<GlobalVariable *>(
1388           LDSVarsToTransform.begin(), LDSVarsToTransform.end()));
1389 
1390       for (GlobalVariable *GV : Sorted) {
1391         OptimizedStructLayoutField F(GV,
1392                                      DL.getTypeAllocSize(GV->getValueType()),
1393                                      AMDGPU::getAlign(DL, GV));
1394         LayoutFields.emplace_back(F);
1395       }
1396     }
1397 
1398     performOptimizedStructLayout(LayoutFields);
1399 
1400     std::vector<GlobalVariable *> LocalVars;
1401     BitVector IsPaddingField;
1402     LocalVars.reserve(LDSVarsToTransform.size()); // will be at least this large
1403     IsPaddingField.reserve(LDSVarsToTransform.size());
1404     {
1405       uint64_t CurrentOffset = 0;
1406       for (size_t I = 0; I < LayoutFields.size(); I++) {
1407         GlobalVariable *FGV = static_cast<GlobalVariable *>(
1408             const_cast<void *>(LayoutFields[I].Id));
1409         Align DataAlign = LayoutFields[I].Alignment;
1410 
1411         uint64_t DataAlignV = DataAlign.value();
1412         if (uint64_t Rem = CurrentOffset % DataAlignV) {
1413           uint64_t Padding = DataAlignV - Rem;
1414 
1415           // Append an array of padding bytes to meet alignment requested
1416           // Note (o +      (a - (o % a)) ) % a == 0
1417           //      (offset + Padding       ) % align == 0
1418 
1419           Type *ATy = ArrayType::get(Type::getInt8Ty(Ctx), Padding);
1420           LocalVars.push_back(new GlobalVariable(
1421               M, ATy, false, GlobalValue::InternalLinkage,
1422               PoisonValue::get(ATy), "", nullptr, GlobalValue::NotThreadLocal,
1423               AMDGPUAS::LOCAL_ADDRESS, false));
1424           IsPaddingField.push_back(true);
1425           CurrentOffset += Padding;
1426         }
1427 
1428         LocalVars.push_back(FGV);
1429         IsPaddingField.push_back(false);
1430         CurrentOffset += LayoutFields[I].Size;
1431       }
1432     }
1433 
1434     std::vector<Type *> LocalVarTypes;
1435     LocalVarTypes.reserve(LocalVars.size());
1436     std::transform(
1437         LocalVars.cbegin(), LocalVars.cend(), std::back_inserter(LocalVarTypes),
1438         [](const GlobalVariable *V) -> Type * { return V->getValueType(); });
1439 
1440     StructType *LDSTy = StructType::create(Ctx, LocalVarTypes, VarName + ".t");
1441 
1442     Align StructAlign = AMDGPU::getAlign(DL, LocalVars[0]);
1443 
1444     GlobalVariable *SGV = new GlobalVariable(
1445         M, LDSTy, false, GlobalValue::InternalLinkage, PoisonValue::get(LDSTy),
1446         VarName, nullptr, GlobalValue::NotThreadLocal, AMDGPUAS::LOCAL_ADDRESS,
1447         false);
1448     SGV->setAlignment(StructAlign);
1449 
1450     DenseMap<GlobalVariable *, Constant *> Map;
1451     Type *I32 = Type::getInt32Ty(Ctx);
1452     for (size_t I = 0; I < LocalVars.size(); I++) {
1453       GlobalVariable *GV = LocalVars[I];
1454       Constant *GEPIdx[] = {ConstantInt::get(I32, 0), ConstantInt::get(I32, I)};
1455       Constant *GEP = ConstantExpr::getGetElementPtr(LDSTy, SGV, GEPIdx, true);
1456       if (IsPaddingField[I]) {
1457         assert(GV->use_empty());
1458         GV->eraseFromParent();
1459       } else {
1460         Map[GV] = GEP;
1461       }
1462     }
1463     assert(Map.size() == LDSVarsToTransform.size());
1464     return {SGV, std::move(Map)};
1465   }
1466 
1467   template <typename PredicateTy>
1468   static void replaceLDSVariablesWithStruct(
1469       Module &M, DenseSet<GlobalVariable *> const &LDSVarsToTransformArg,
1470       const LDSVariableReplacement &Replacement, PredicateTy Predicate) {
1471     LLVMContext &Ctx = M.getContext();
1472     const DataLayout &DL = M.getDataLayout();
1473 
1474     // A hack... we need to insert the aliasing info in a predictable order for
1475     // lit tests. Would like to have them in a stable order already, ideally the
1476     // same order they get allocated, which might mean an ordered set container
1477     auto LDSVarsToTransform = sortByName(std::vector<GlobalVariable *>(
1478         LDSVarsToTransformArg.begin(), LDSVarsToTransformArg.end()));
1479 
1480     // Create alias.scope and their lists. Each field in the new structure
1481     // does not alias with all other fields.
1482     SmallVector<MDNode *> AliasScopes;
1483     SmallVector<Metadata *> NoAliasList;
1484     const size_t NumberVars = LDSVarsToTransform.size();
1485     if (NumberVars > 1) {
1486       MDBuilder MDB(Ctx);
1487       AliasScopes.reserve(NumberVars);
1488       MDNode *Domain = MDB.createAnonymousAliasScopeDomain();
1489       for (size_t I = 0; I < NumberVars; I++) {
1490         MDNode *Scope = MDB.createAnonymousAliasScope(Domain);
1491         AliasScopes.push_back(Scope);
1492       }
1493       NoAliasList.append(&AliasScopes[1], AliasScopes.end());
1494     }
1495 
1496     // Replace uses of ith variable with a constantexpr to the corresponding
1497     // field of the instance that will be allocated by AMDGPUMachineFunction
1498     for (size_t I = 0; I < NumberVars; I++) {
1499       GlobalVariable *GV = LDSVarsToTransform[I];
1500       Constant *GEP = Replacement.LDSVarsToConstantGEP.at(GV);
1501 
1502       GV->replaceUsesWithIf(GEP, Predicate);
1503 
1504       APInt APOff(DL.getIndexTypeSizeInBits(GEP->getType()), 0);
1505       GEP->stripAndAccumulateInBoundsConstantOffsets(DL, APOff);
1506       uint64_t Offset = APOff.getZExtValue();
1507 
1508       Align A =
1509           commonAlignment(Replacement.SGV->getAlign().valueOrOne(), Offset);
1510 
1511       if (I)
1512         NoAliasList[I - 1] = AliasScopes[I - 1];
1513       MDNode *NoAlias =
1514           NoAliasList.empty() ? nullptr : MDNode::get(Ctx, NoAliasList);
1515       MDNode *AliasScope =
1516           AliasScopes.empty() ? nullptr : MDNode::get(Ctx, {AliasScopes[I]});
1517 
1518       refineUsesAlignmentAndAA(GEP, A, DL, AliasScope, NoAlias);
1519     }
1520   }
1521 
1522   static void refineUsesAlignmentAndAA(Value *Ptr, Align A,
1523                                        const DataLayout &DL, MDNode *AliasScope,
1524                                        MDNode *NoAlias, unsigned MaxDepth = 5) {
1525     if (!MaxDepth || (A == 1 && !AliasScope))
1526       return;
1527 
1528     for (User *U : Ptr->users()) {
1529       if (auto *I = dyn_cast<Instruction>(U)) {
1530         if (AliasScope && I->mayReadOrWriteMemory()) {
1531           MDNode *AS = I->getMetadata(LLVMContext::MD_alias_scope);
1532           AS = (AS ? MDNode::getMostGenericAliasScope(AS, AliasScope)
1533                    : AliasScope);
1534           I->setMetadata(LLVMContext::MD_alias_scope, AS);
1535 
1536           MDNode *NA = I->getMetadata(LLVMContext::MD_noalias);
1537           NA = (NA ? MDNode::intersect(NA, NoAlias) : NoAlias);
1538           I->setMetadata(LLVMContext::MD_noalias, NA);
1539         }
1540       }
1541 
1542       if (auto *LI = dyn_cast<LoadInst>(U)) {
1543         LI->setAlignment(std::max(A, LI->getAlign()));
1544         continue;
1545       }
1546       if (auto *SI = dyn_cast<StoreInst>(U)) {
1547         if (SI->getPointerOperand() == Ptr)
1548           SI->setAlignment(std::max(A, SI->getAlign()));
1549         continue;
1550       }
1551       if (auto *AI = dyn_cast<AtomicRMWInst>(U)) {
1552         // None of atomicrmw operations can work on pointers, but let's
1553         // check it anyway in case it will or we will process ConstantExpr.
1554         if (AI->getPointerOperand() == Ptr)
1555           AI->setAlignment(std::max(A, AI->getAlign()));
1556         continue;
1557       }
1558       if (auto *AI = dyn_cast<AtomicCmpXchgInst>(U)) {
1559         if (AI->getPointerOperand() == Ptr)
1560           AI->setAlignment(std::max(A, AI->getAlign()));
1561         continue;
1562       }
1563       if (auto *GEP = dyn_cast<GetElementPtrInst>(U)) {
1564         unsigned BitWidth = DL.getIndexTypeSizeInBits(GEP->getType());
1565         APInt Off(BitWidth, 0);
1566         if (GEP->getPointerOperand() == Ptr) {
1567           Align GA;
1568           if (GEP->accumulateConstantOffset(DL, Off))
1569             GA = commonAlignment(A, Off.getLimitedValue());
1570           refineUsesAlignmentAndAA(GEP, GA, DL, AliasScope, NoAlias,
1571                                    MaxDepth - 1);
1572         }
1573         continue;
1574       }
1575       if (auto *I = dyn_cast<Instruction>(U)) {
1576         if (I->getOpcode() == Instruction::BitCast ||
1577             I->getOpcode() == Instruction::AddrSpaceCast)
1578           refineUsesAlignmentAndAA(I, A, DL, AliasScope, NoAlias, MaxDepth - 1);
1579       }
1580     }
1581   }
1582 };
1583 
1584 class AMDGPULowerModuleLDSLegacy : public ModulePass {
1585 public:
1586   const AMDGPUTargetMachine *TM;
1587   static char ID;
1588 
1589   AMDGPULowerModuleLDSLegacy(const AMDGPUTargetMachine *TM_ = nullptr)
1590       : ModulePass(ID), TM(TM_) {
1591     initializeAMDGPULowerModuleLDSLegacyPass(*PassRegistry::getPassRegistry());
1592   }
1593 
1594   void getAnalysisUsage(AnalysisUsage &AU) const override {
1595     if (!TM)
1596       AU.addRequired<TargetPassConfig>();
1597   }
1598 
1599   bool runOnModule(Module &M) override {
1600     if (!TM) {
1601       auto &TPC = getAnalysis<TargetPassConfig>();
1602       TM = &TPC.getTM<AMDGPUTargetMachine>();
1603     }
1604 
1605     return AMDGPULowerModuleLDS(*TM).runOnModule(M);
1606   }
1607 };
1608 
1609 } // namespace
1610 char AMDGPULowerModuleLDSLegacy::ID = 0;
1611 
1612 char &llvm::AMDGPULowerModuleLDSLegacyPassID = AMDGPULowerModuleLDSLegacy::ID;
1613 
1614 INITIALIZE_PASS_BEGIN(AMDGPULowerModuleLDSLegacy, DEBUG_TYPE,
1615                       "Lower uses of LDS variables from non-kernel functions",
1616                       false, false)
1617 INITIALIZE_PASS_DEPENDENCY(TargetPassConfig)
1618 INITIALIZE_PASS_END(AMDGPULowerModuleLDSLegacy, DEBUG_TYPE,
1619                     "Lower uses of LDS variables from non-kernel functions",
1620                     false, false)
1621 
1622 ModulePass *
1623 llvm::createAMDGPULowerModuleLDSLegacyPass(const AMDGPUTargetMachine *TM) {
1624   return new AMDGPULowerModuleLDSLegacy(TM);
1625 }
1626 
1627 PreservedAnalyses AMDGPULowerModuleLDSPass::run(Module &M,
1628                                                 ModuleAnalysisManager &) {
1629   return AMDGPULowerModuleLDS(TM).runOnModule(M) ? PreservedAnalyses::none()
1630                                                  : PreservedAnalyses::all();
1631 }
1632