[Polly] [PPCGCodeGeneration] Teach PPCGCodeGen to use live range reordering.

[bollu]

Summary

• Currently, code is a big pile of hack. Need to generate correct

kill information. Probably can be done in DependenceInfo.

• Contains changes to /lib/External/ppcg for debugging purposes. All

this will be removed before merging.

[bollu]

• [WIP] [2 failures] change algorithm to pick phi nodes. privatization passes, other tests fail

[bollu]

• [NFC] code is readable and sane now. Can be reviewed.

[bollu]

ping, @grosser, @singam-sanjay, @Meinersbur (post-commit, probably since you are busy)

[bollu]

• Fix forgotten LHS assignment in isl_schedule_set(Info.KillsSchedule, KillSchedule);

[bollu]

• Fix forgotten LHS assignment in isl_schedule_set(Info.KillsSchedule, KillSchedule)
• Push from correct branch (clean diff branch, not dirty dev branch with debug code)

[bollu]

• [NFC wrt patch] Updated clang, hence fixed diff in unittests/ScopPassManager/PassManagerTest.cpp

[grosser]

Hi Siddharth,

the patch looks already very good. I have a couple of stylistic comments below, but nothing fundamental.

lib/CodeGen/PPCGCodeGeneration.cpp

124

What does "grafted" mean.

You name your class according to the context it is used in, rather than what it does. As there may be different uses in the future, I would prefer to name it according to what it does -- aka compute kills.

129

Any reason you do not use the new C++ interface?

134

Say why we only derive kill information for PHI nodes. Something like:

We currently only derive kill information for phi nodes, as phi nodes allow us to easily derive kill information. PHI nodes are not alive outside the scop and can consequently all be "killed".

147

Just say you compute the kill information -- and define what a kill is.

163

Would it make sense to use all 80 columns, rather than to break early?

174

Why do you create multiple kill statements? Is a single kill statement not enough>

185

Why two white spaces?

193

Unnecessary newline?

195

The C++ bindings would allow this to be written in a single line. ;)

198

For some reason it feels more canonical to just append std::strings rather then using a stringstream. Not sure if you agree.

216

Use all 80 columns for the comments.

[bollu]

the comments I haven't replied to, I agree with and will fix.

lib/CodeGen/PPCGCodeGeneration.cpp

129

I did not wish to mix the old C interface and the new C++ interface in the same file. I was considering a cleanup patch where I convert everything to C++ once this is done,

163

yes, I agree here.

174

I'm mimicking what PPCG does when it hands in the schedule to compute_dependences. I haven't tested with just one statement.

216

I wanted to emphasize the "[param] -> { Stmt_phantom[] }" so I broke it into a separate line.

I should probably write it as:

// 3. Create the kill schedule of the form:
// "[param] -> { Stmt_phantom[] }"
// Then add this to Info.KillsSchedule.

to show that the newline is deliberate.

[bollu]

• [NFC] style fixes.

[bollu]

• [NFC wrt patch] update to C++ API. Code reads much nicer now

[bollu]

• [NFC] remove sstream that is no longer used

[grosser]

LGTM, just some minor comments!

lib/CodeGen/PPCGCodeGeneration.cpp

117

Talk about kill, not live-range reordering.

163

Still does not use 80 cols.

lib/CodeGen/PPCGCodeGeneration.cpp

 #include "ppcg/gpu.h"
 #include "ppcg/gpu_print.h"
 #include "ppcg/ppcg.h"
 #include "ppcg/schedule.h"
 }
 
 #include "llvm/Support/Debug.h"
 
+#include <sstream>
+
 using namespace polly;
 using namespace llvm;
 
 #define DEBUG_TYPE "polly-codegen-ppcg"
 
 static cl::opt<bool> DumpSchedule("polly-acc-dump-schedule",
                                   cl::desc("Dump the computed GPU Schedule"),
                                   cl::Hidden, cl::init(false), cl::ZeroOrMore,
CudaVersion("polly-acc-cuda-version",
                 cl::desc("The CUDA version to compile for"), cl::Hidden,
                 cl::init("sm_30"), cl::ZeroOrMore, cl::cat(PollyCategory));
 
 static cl::opt<int>
     MinCompute("polly-acc-mincompute",
                cl::desc("Minimal number of compute statements to run on GPU."),
                cl::Hidden, cl::init(10 * 512 * 512));
 
+/// Used to store information PPCG wants for live range reordering.
+///
+/// @see computeLiveRangeReordering
+/// @see GPUNodeBuilder::createPPCGScop
+/// @see GPUNodeBuilder::createPPCGProg
+struct LiveRangeReorderingInfo {
+  /// Collection of all kill statements that will be grafted to the end of
+  /// PPCGScop->schedule.
+  ///
+  /// The nodes in `KillsSchedule` will be merged using `isl_schedule_set`
+  /// which merges schedules in *arbitrary* order.
+  /// (we don't care about the order of the kills anyway).
+  isl_schedule *KillsSchedule;
+  /// Map from kill statement instances to scalars that need to be
+  /// killed.
+  ///
+  /// We choose to only only kill phi nodes in a scop.
+  /// [params] -> { [Stmt_phantom[] -> ref_phantom[]] -> phi_ref[] }
+  isl_union_map *TaggedMustKills;
+
+  LiveRangeReorderingInfo()
+      : KillsSchedule(nullptr), TaggedMustKills(nullptr){};
+};
+
+/// Compute information needed to enable live range reordering information
+/// with PPCG.
+///
+/// @params S The Scop to compute live range reordering information
+/// @returns live range reordering information that can be used to setup
+/// PPCG.
+static LiveRangeReorderingInfo computeLiveRangeReorderingInfo(const Scop &S) {
+  LiveRangeReorderingInfo Info;
+
+  // 1. Collect phi nodes in scop.
+  SmallVector<isl_id *, 4> KillMemIds;
+  for (ScopArrayInfo *SAI : S.arrays()) {
+    if (!SAI->isPHIKind())
+      continue;
+
+    KillMemIds.push_back(SAI->getBasePtrId());
+  }
+
+  Info.TaggedMustKills =
+      isl_union_map_from_map(isl_map_empty(S.getParamSpace()));
+
+  // Initialising KillsSchedule to isl_set_empty
+  // creates an empty node in the schedule node:
+  // - filter: "[control] -> { }"
+  // So, we choose to not create this to keep the output a little nicer,
+  // at the cost of some code complexity.
+  Info.KillsSchedule = nullptr;
+
+  for (isl_id *phiId : KillMemIds) {
+
+    std::stringstream KillStmtSS;
+    KillStmtSS << "Skill_phantom_" << isl_id_get_name(phiId);
+    isl_id *KillStmtId =
+        isl_id_alloc(S.getIslCtx(), KillStmtSS.str().c_str(), NULL);
+
+    // NOTE: construction of tagged_must_kill
+    // 2. We need to construct a map
+    // [param] -> { [Stmt_phantom[] -> ref_phantom[]] -> phi_ref }
+    // To construct this, we use `isl_map_domain_product` on 2 maps`
+    // 2a. StmtToPhi:
+    //         [param] -> { Stmt_phantom[] -> phi_ref[] }
+    // 2b. PhantomRefToPhi:
+    //         [param] -> { ref_phantom[] -> phi_ref[] }
+    //
+    // Combining these with  `isl_map_domain_product` gives us
+    // TaggedMustKill:
+    //     [param] -> { [Stmt[] -> phantom_ref[]] -> memref[] }
+
+    // 2a. [param] -> { S_2[] -> phi_ref[] }
+    isl_map *StmtToPhi = isl_map_universe(S.getParamSpace());
+    StmtToPhi =
+
+        isl_map_set_tuple_id(StmtToPhi, isl_dim_in, isl_id_copy(KillStmtId));
+    StmtToPhi =
+        isl_map_set_tuple_id(StmtToPhi, isl_dim_out, isl_id_copy(phiId));
+
+    std::stringstream PhantomRefSS;
+    PhantomRefSS << "ref_phantom_" << isl_id_get_name(phiId);
+    isl_id *PhantomRefId =
+        isl_id_alloc(S.getIslCtx(), PhantomRefSS.str().c_str(), NULL);
+
+    // 2b. [param] -> { phantom_ref[] -> memref[] }
+    isl_map *PhantomRefToPhi = isl_map_universe(S.getParamSpace());
+    PhantomRefToPhi =
+        isl_map_set_tuple_id(PhantomRefToPhi, isl_dim_in, PhantomRefId);
+    PhantomRefToPhi = isl_map_set_tuple_id(PhantomRefToPhi, isl_dim_out, phiId);
+
+    // 2. [param] -> { [Stmt[] -> phantom_ref[]] -> memref[] }
+    isl_map *TaggedMustKill =
+        isl_map_domain_product(StmtToPhi, PhantomRefToPhi);
+    Info.TaggedMustKills = isl_union_map_union(
+        Info.TaggedMustKills, isl_union_map_from_map(TaggedMustKill));
+
+    // 3. Create the kill schedule of the form
+    // "[param] -> { Stmt_phantom[] }"
+    // and add to Info.KillsSchedule
+    isl_space *KillStmtSpace = S.getParamSpace();
+    KillStmtSpace =
+        isl_space_set_tuple_id(KillStmtSpace, isl_dim_set, KillStmtId);
+    isl_union_set *KillStmtDomain =
+        isl_union_set_from_set(isl_set_universe(KillStmtSpace));
+
+    isl_schedule *KillSchedule = isl_schedule_from_domain(KillStmtDomain);
+    if (Info.KillsSchedule)
+      Info.KillsSchedule = isl_schedule_set(Info.KillsSchedule, KillSchedule);
+    else
+      Info.KillsSchedule = KillSchedule;
+  }
+
+  return Info;
+}
+
 /// Create the ast expressions for a ScopStmt.
 ///
 /// This function is a callback for to generate the ast expressions for each
 /// of the scheduled ScopStmts.
 static __isl_give isl_id_to_ast_expr *pollyBuildAstExprForStmt(
     void *StmtT, isl_ast_build *Build,
     isl_multi_pw_aff *(*FunctionIndex)(__isl_take isl_multi_pw_aff *MPA,
                                        isl_id *Id, void *User),
public:
   /// We do not use Polly's dependence analysis for now, to ensure we match
   /// the PPCG default behaviour more closely.
   ///
   /// @returns A new ppcg scop.
   ppcg_scop *createPPCGScop() {
     auto PPCGScop = (ppcg_scop *)malloc(sizeof(ppcg_scop));
 
     PPCGScop->options = createPPCGOptions();
+    // enable live range reordering
+    PPCGScop->options->live_range_reordering = 1;
 
     PPCGScop->start = 0;
     PPCGScop->end = 0;
 
     PPCGScop->context = S->getContext();
     PPCGScop->domain = S->getDomains();
     PPCGScop->call = nullptr;
     PPCGScop->tagged_reads = getTaggedReads();
     PPCGScop->reads = S->getReads();
     PPCGScop->live_in = nullptr;
     PPCGScop->tagged_may_writes = getTaggedMayWrites();
     PPCGScop->may_writes = S->getWrites();
     PPCGScop->tagged_must_writes = getTaggedMustWrites();
     PPCGScop->must_writes = S->getMustWrites();
     PPCGScop->live_out = nullptr;
-    PPCGScop->tagged_must_kills = isl_union_map_empty(S->getParamSpace());
     PPCGScop->tagger = nullptr;
-
-    PPCGScop->independence = nullptr;
+    PPCGScop->independence =
+        isl_union_map_empty(isl_set_get_space(PPCGScop->context));
     PPCGScop->dep_flow = nullptr;
     PPCGScop->tagged_dep_flow = nullptr;
     PPCGScop->dep_false = nullptr;
     PPCGScop->dep_forced = nullptr;
     PPCGScop->dep_order = nullptr;
     PPCGScop->tagged_dep_order = nullptr;
 
     PPCGScop->schedule = S->getScheduleTree();
-    PPCGScop->names = getNames();
 
+    LiveRangeReorderingInfo LiveRangeInfo = computeLiveRangeReorderingInfo(*S);
+    // If we have something non-trivial to kill, add it to the schedule
+    if (LiveRangeInfo.KillsSchedule)
+      PPCGScop->schedule = isl_schedule_sequence(PPCGScop->schedule,
+                                                 LiveRangeInfo.KillsSchedule);
+    PPCGScop->tagged_must_kills = LiveRangeInfo.TaggedMustKills;
+
+    PPCGScop->names = getNames();
     PPCGScop->pet = nullptr;
 
     compute_tagger(PPCGScop);
     compute_dependences(PPCGScop);
 
     return PPCGScop;
   }
 
gpu_prog *createPPCGProg(ppcg_scop *PPCGScop) {
     PPCGProg->read = isl_union_map_copy(PPCGScop->reads);
     PPCGProg->may_write = isl_union_map_copy(PPCGScop->may_writes);
     PPCGProg->must_write = isl_union_map_copy(PPCGScop->must_writes);
     PPCGProg->tagged_must_kill =
         isl_union_map_copy(PPCGScop->tagged_must_kills);
     PPCGProg->to_inner = getArrayIdentity();
     PPCGProg->to_outer = getArrayIdentity();
     PPCGProg->any_to_outer = nullptr;
-    PPCGProg->array_order = nullptr;
+
+    // this needs to be set when live range reordering is enabled.
+    // NOTE: I believe that is conservatively correct. I'm not sure
+    //       what the semantics of this is.
+    // Quoting PPCG/gpu.h: "Order dependences on non-scalars."
+    PPCGProg->array_order =
+        isl_union_map_empty(isl_set_get_space(PPCGScop->context));
     PPCGProg->n_stmts = std::distance(S->begin(), S->end());
     PPCGProg->stmts = getStatements();
     PPCGProg->n_array = std::distance(S->array_begin(), S->array_end());
     PPCGProg->array = isl_calloc_array(S->getIslCtx(), struct gpu_array_info,
                                        PPCGProg->n_array);
 
     createArrays(PPCGProg);
 
     PPCGProg->may_persist = compute_may_persist(PPCGProg);
-
     return PPCGProg;
   }
 
   struct PrintGPUUserData {
     struct cuda_info *CudaInfo;
     struct gpu_prog *PPCGProg;
     std::vector<ppcg_kernel *> Kernels;
   };

test/GPGPU/non-read-only-scalars.ll

 ; CODE-NEXT:         dim3 k3_dimBlock;
 ; CODE-NEXT:         dim3 k3_dimGrid;
 ; CODE-NEXT:         kernel3 <<<k3_dimGrid, k3_dimBlock>>> (dev_MemRef_A, dev_MemRef_sum_0__phi, dev_MemRef_sum_0, c0);
 ; CODE-NEXT:         cudaCheckKernel();
 ; CODE-NEXT:       }
 
 ; CODE:   }
 ; CODE-NEXT:   cudaCheckReturn(cudaMemcpy(MemRef_A, dev_MemRef_A, (32) * sizeof(float), cudaMemcpyDeviceToHost));
-; CODE-NEXT:   cudaCheckReturn(cudaMemcpy(&MemRef_sum_0__phi, dev_MemRef_sum_0__phi, sizeof(float), cudaMemcpyDeviceToHost));
 ; CODE-NEXT:   cudaCheckReturn(cudaMemcpy(&MemRef_sum_0, dev_MemRef_sum_0, sizeof(float), cudaMemcpyDeviceToHost));
 ; CODE-NEXT: }
 
 ; CODE: # kernel0
 ; CODE-NEXT: {
 ; CODE-NEXT:   Stmt_bb4(t0);
 ; CODE-NEXT:   Stmt_bb10(t0);
 ; CODE-NEXT: }

test/GPGPU/phi-nodes-in-kernel.ll

 ; CODE-NEXT:   cudaCheckReturn(cudaMemcpy(dev_MemRef_out_l_055__phi, &MemRef_out_l_055__phi, sizeof(i32), cudaMemcpyHostToDevice));
 ; CODE-NEXT:   {
 ; CODE-NEXT:     dim3 k0_dimBlock(32);
 ; CODE-NEXT:     dim3 k0_dimGrid(2);
 ; CODE-NEXT:     kernel0 <<<k0_dimGrid, k0_dimBlock>>> (dev_MemRef_out_l_055__phi, dev_MemRef_out_l_055, dev_MemRef_c);
 ; CODE-NEXT:     cudaCheckKernel();
 ; CODE-NEXT:   }
 
-; CODE:   cudaCheckReturn(cudaMemcpy(&MemRef_out_l_055__phi, dev_MemRef_out_l_055__phi, sizeof(i32), cudaMemcpyDeviceToHost));
-; CODE-NEXT:   cudaCheckReturn(cudaMemcpy(&MemRef_out_l_055, dev_MemRef_out_l_055, sizeof(i32), cudaMemcpyDeviceToHost));
-; CODE-NEXT:   cudaCheckReturn(cudaMemcpy(MemRef_c, dev_MemRef_c, (50) * sizeof(i32), cudaMemcpyDeviceToHost));
+; CODE:       cudaCheckReturn(cudaMemcpy(&MemRef_out_l_055, dev_MemRef_out_l_055, sizeof(i32), cudaMemcpyDeviceToHost));
+; CODE-NEXT:  cudaCheckReturn(cudaMemcpy(MemRef_c, dev_MemRef_c, (50) * sizeof(i32), cudaMemcpyDeviceToHost));
 ; CODE-NEXT: }
 
 ; CODE: # kernel0
 ; CODE-NEXT: if (32 * b0 + t0 <= 48) {
 ; CODE-NEXT:   if (b0 == 1 && t0 == 16)
 ; CODE-NEXT:     Stmt_for_cond1_preheader(0);
 ; CODE-NEXT:   Stmt_for_body17(0, 32 * b0 + t0);
 ; CODE-NEXT:   if (b0 == 1 && t0 == 16)
 ; CODE-NEXT:     Stmt_for_cond15_for_cond12_loopexit_crit_edge(0);
 ; CODE-NEXT: }
 
 ; IR:      [[REGA:%.+]] = bitcast i32* %out_l.055.phiops to i8*
 ; IR-NEXT: call void @polly_copyFromHostToDevice(i8* [[REGA]], i8* %p_dev_array_MemRef_out_l_055__phi, i64 4)
 
-; IR:      [[REGB:%.+]] = bitcast i32* %out_l.055.phiops to i8*
-; IR-NEXT: call void @polly_copyFromDeviceToHost(i8* %p_dev_array_MemRef_out_l_055__phi, i8* [[REGB]], i64 4)
-; IR-NEXT: [[REGC:%.+]] = bitcast i32* %out_l.055.s2a to i8*
+; IR: [[REGC:%.+]] = bitcast i32* %out_l.055.s2a to i8*
 ; IR-NEXT: call void @polly_copyFromDeviceToHost(i8* %p_dev_array_MemRef_out_l_055, i8* [[REGC]], i64 4)
 
 ; KERNEL-IR: entry:
 ; KERNEL-IR-NEXT:   %out_l.055.s2a = alloca i32
 ; KERNEL-IR-NEXT:   %out_l.055.phiops = alloca i32
 ; KERNEL-IR-NEXT:   %1 = addrspacecast i8 addrspace(1)* %MemRef_out_l_055__phi to i32*
 ; KERNEL-IR-NEXT:   %2 = load i32, i32* %1
 ; KERNEL-IR-NEXT:   store i32 %2, i32* %out_l.055.phiops

test/GPGPU/privatization-simple.ll

+; RUN: opt %loadPolly -analyze -polly-scops < %s | FileCheck %s -check-prefix=SCOP
+; RUN: opt %loadPolly -S -polly-codegen-ppcg < %s | FileCheck %s -check-prefix=HOST-IR
+
+; SCOP:      Function: f
+; SCOP-NEXT: Region: %for.body---%for.end
+; SCOP-NEXT: Max Loop Depth:  1
+
+; Check that kernel launch is generated in host IR.
+; the declare would not be generated unless a call to a kernel exists.
+; HOST-IR: declare void @polly_launchKernel(i8*, i32, i32, i32, i32, i32, i8*)
+
+; void f(int A[], int B[], int control, int C[]) {
+;     int x;
+; #pragma scop
+;     for(int i = 0; i < 1000; i ++) {
+;         x = 0;
+;         if(control) x = C[i];
+;         B[i] = x * A[i];
+; 
+;     }
+; #pragma endscop
+; }
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+
+define void @f(i32* %A, i32* %B, i32 %control, i32* %C) {
+entry:
+  br label %entry.split
+
+entry.split:                                      ; preds = %entry
+  br label %for.body
+
+for.body:                                         ; preds = %entry.split, %if.end
+  %indvars.iv = phi i64 [ 0, %entry.split ], [ %indvars.iv.next, %if.end ]
+  %tobool = icmp eq i32 %control, 0
+  br i1 %tobool, label %if.end, label %if.then
+
+if.then:                                          ; preds = %for.body
+  %arrayidx = getelementptr inbounds i32, i32* %C, i64 %indvars.iv
+  %tmp4 = load i32, i32* %arrayidx, align 4
+  br label %if.end
+
+if.end:                                           ; preds = %for.body, %if.then
+  %x.0 = phi i32 [ %tmp4, %if.then ], [ 0, %for.body ]
+  %arrayidx2 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
+  %tmp8 = load i32, i32* %arrayidx2, align 4
+  %mul = mul nsw i32 %tmp8, %x.0
+  %arrayidx4 = getelementptr inbounds i32, i32* %B, i64 %indvars.iv
+  store i32 %mul, i32* %arrayidx4, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp ne i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.body, label %for.end
+
+for.end:                                          ; preds = %if.end
+  ret void
+}

test/GPGPU/privatization.ll

+; RUN: opt %loadPolly -analyze -polly-scops < %s | FileCheck %s -check-prefix=SCOP
+; RUN: opt %loadPolly -S -polly-codegen-ppcg < %s | FileCheck %s -check-prefix=HOST-IR
+
+; SCOP:      Function: checkPrivatization
+; SCOP-NEXT: Region: %for.body---%for.end
+; SCOP-NEXT: Max Loop Depth:  1
+
+
+; Check that kernel launch is generated in host IR.
+; the declare would not be generated unless a call to a kernel exists.
+; HOST-IR: declare void @polly_launchKernel(i8*, i32, i32, i32, i32, i32, i8*)
+
+; 
+;
+;    void checkPrivatization(int A[], int B[], int C[], int control) {
+;      int x;
+;    #pragma scop
+;      for (int i = 0; i < 1000; i++) {
+;        x = 0;
+;        if (control)
+;          x += C[i];
+;
+;        B[i] = x * A[i];
+;      }
+;    #pragma endscop
+;    }
+;
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+
+define void @checkPrivatization(i32* %A, i32* %B, i32* %C, i32 %control) {
+entry:
+  br label %entry.split
+
+entry.split:                                      ; preds = %entry
+  br label %for.body
+
+for.body:                                         ; preds = %entry.split, %if.end
+  %indvars.iv = phi i64 [ 0, %entry.split ], [ %indvars.iv.next, %if.end ]
+  %tobool = icmp eq i32 %control, 0
+  br i1 %tobool, label %if.end, label %if.then
+
+if.then:                                          ; preds = %for.body
+  %arrayidx = getelementptr inbounds i32, i32* %C, i64 %indvars.iv
+  %tmp4 = load i32, i32* %arrayidx, align 4
+  br label %if.end
+
+if.end:                                           ; preds = %for.body, %if.then
+  %x.0 = phi i32 [ %tmp4, %if.then ], [ 0, %for.body ]
+  %arrayidx2 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
+  %tmp9 = load i32, i32* %arrayidx2, align 4
+  %mul = mul nsw i32 %tmp9, %x.0
+  %arrayidx4 = getelementptr inbounds i32, i32* %B, i64 %indvars.iv
+  store i32 %mul, i32* %arrayidx4, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp ne i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.body, label %for.end
+
+for.end:                                          ; preds = %if.end
+  ret void
+}

unittests/ScopPassManager/PassManagerTest.cpp

-#include "llvm/IR/PassManager.h"
 #include "polly/CodeGen/IslAst.h"
 #include "polly/DependenceInfo.h"
 #include "polly/ScopPass.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/Analysis/CGSCCPassManager.h"
+#include "llvm/IR/PassManager.h"
 #include "llvm/Passes/PassBuilder.h"
 #include "llvm/Transforms/Scalar/LoopPassManager.h"
 #include "gtest/gtest.h"
 
 using namespace polly;
 using namespace llvm;
 
 namespace {