[Polly] [PPCGCodeGeneration] Enable GPU code generation with invariant loads.

[bollu]

Summary

The condition that disallowed code generation in PPCGCodeGeneration with
invariant loads is not required. I haven't been able to construct a
counterexample where this generates invalid code.

[grosser]

LGTM.

[bollu]

Closed by commit r306245. For some reason, Phab did not auto-close this, even though commit contains:

Differential Revision: https://reviews.llvm.org/D34604

@grosser thanks for the fix with REQUIRES: pollyacc. I missed it :)

[singam-sanjay]

Why wasn't it supported previously ?

[grosser]

We started with a minimal set of features and then enabled more and more as we found time to test them.

lib/CodeGen/PPCGCodeGeneration.cpp

public:
   bool runOnScop(Scop &CurrentScop) override {
     S = &CurrentScop;
     LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
     DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
     SE = &getAnalysis<ScalarEvolutionWrapperPass>().getSE();
     DL = &S->getRegion().getEntry()->getModule()->getDataLayout();
     RI = &getAnalysis<RegionInfoPass>().getRegionInfo();
 
-    // We currently do not support scops with invariant loads.
-    if (S->hasInvariantAccesses())
-      return false;
-
     // We currently do not support functions inside kernels, as code
     // generation will need to offload function calls to the kernel.
     // This may lead to a kernel trying to call a function on the host.
     // This also allows us to prevent codegen from trying to take the
     // address of an intrinsic function to send to the kernel.
     if (ContainsFnPtr(CurrentScop))
       return false;
 

lib/Support/RegisterPasses.cpp

#endif
 
   if (DumpAfter)
     PM.add(polly::createDumpModulePass("-after", true));
   for (auto &Filename : DumpAfterFile)
     PM.add(polly::createDumpModulePass(Filename, false));
 
   if (CFGPrinter)
     PM.add(llvm::createCFGPrinterLegacyPassPass());
-
-  if (Target == TARGET_GPU) {
-    // Invariant load hoisting not yet supported by GPU code generation.
-    PollyInvariantLoadHoisting = false;
-  }
 }
 
 static bool shouldEnablePolly() {
   if (PollyOnlyPrinter || PollyPrinter || PollyOnlyViewer || PollyViewer)
     PollyTrackFailures = true;
 
   if (PollyOnlyPrinter || PollyPrinter || PollyOnlyViewer || PollyViewer ||
       ExportJScop || ImportJScop)

test/GPGPU/invariant-load-hoisting.ll

+; RUN: opt %loadPolly -polly-scops -analyze \
+; RUN: -polly-invariant-load-hoisting < %s | FileCheck %s -check-prefix=SCOP
+;
+; RUN: opt %loadPolly -polly-scops -S  -polly-invariant-load-hoisting \
+; RUN: -polly-codegen-ppcg < %s | FileCheck %s -check-prefix=HOST-IR
+; 
+; RUN: opt %loadPolly -polly-scops -analyze  -polly-invariant-load-hoisting \
+; RUN: -polly-codegen-ppcg -polly-acc-dump-kernel-ir < %s | FileCheck %s -check-prefix=KERNEL-IR
+;
+; SCOP:       Function: f
+; SCOP-NEXT:  Region: %entry.split---%for.end26
+; SCOP-NEXT:  Max Loop Depth:  3
+; SCOP-NEXT:  Invariant Accesses: {
+; SCOP-NEXT:          ReadAccess := [Reduction Type: NONE] [Scalar: 0]
+; SCOP-NEXT:              [n, tmp12] -> { Stmt_for_body6[i0, i1, i2] -> MemRef_invariant[0] };
+; SCOP-NEXT:          Execution Context: [n, tmp12] -> {  : n > 0 }
+; SCOP-NEXT:  }
+;
+; HOST-IR:      call void @polly_launchKernel(i8* %215, i32 %221, i32 1, i32 32, i32 1, i32 1, i8* %polly_launch_0_params_i8ptr)
+; HOST-IR-NEXT: call void @polly_freeKernel(i8* %215)
+;
+; KERNEL-IR: define ptx_kernel void @kernel_0(i8 addrspace(1)* %MemRef_B, i8 addrspace(1)* %MemRef_A, i32 %n, i32 %tmp12) #0 {
+; 
+; Check that we generate correct GPU code in case of invariant load hoisting.
+; 
+;
+;    static const int N = 3000;
+;
+;    void f(int A[N][N], int *invariant, int B[N][N], int n) {
+;      for (int i = 0; i < n; i++) {
+;        for (int j = 0; j < n; j++) {
+;          for (int k = 0; k < n; k++) {
+;
+;            A[*invariant][k] = B[k][k];
+;            A[k][*invariant] += B[k][k];
+;          }
+;        }
+;      }
+;    }
+;
+
+define void @f([3000 x i32]* %A, i32* %invariant, [3000 x i32]* %B, i32 %n) {
+entry:
+  br label %entry.split
+
+entry.split:                                      ; preds = %entry
+  %cmp6 = icmp sgt i32 %n, 0
+  br i1 %cmp6, label %for.cond1.preheader.lr.ph, label %for.end26
+
+for.cond1.preheader.lr.ph:                        ; preds = %entry.split
+  br label %for.cond1.preheader
+
+for.cond1.preheader:                              ; preds = %for.cond1.preheader.lr.ph, %for.inc24
+  %i.07 = phi i32 [ 0, %for.cond1.preheader.lr.ph ], [ %inc25, %for.inc24 ]
+  %cmp23 = icmp sgt i32 %n, 0
+  br i1 %cmp23, label %for.cond4.preheader.lr.ph, label %for.inc24
+
+for.cond4.preheader.lr.ph:                        ; preds = %for.cond1.preheader
+  br label %for.cond4.preheader
+
+for.cond4.preheader:                              ; preds = %for.cond4.preheader.lr.ph, %for.inc21
+  %j.04 = phi i32 [ 0, %for.cond4.preheader.lr.ph ], [ %inc22, %for.inc21 ]
+  %cmp51 = icmp sgt i32 %n, 0
+  br i1 %cmp51, label %for.body6.lr.ph, label %for.inc21
+
+for.body6.lr.ph:                                  ; preds = %for.cond4.preheader
+  br label %for.body6
+
+for.body6:                                        ; preds = %for.body6.lr.ph, %for.body6
+  %k.02 = phi i32 [ 0, %for.body6.lr.ph ], [ %inc, %for.body6 ]
+  %idxprom = sext i32 %k.02 to i64
+  %idxprom7 = sext i32 %k.02 to i64
+  %arrayidx8 = getelementptr inbounds [3000 x i32], [3000 x i32]* %B, i64 %idxprom, i64 %idxprom7
+  %tmp9 = load i32, i32* %arrayidx8, align 4
+  %tmp12 = load i32, i32* %invariant, align 4
+  %idxprom9 = sext i32 %tmp12 to i64
+  %idxprom11 = sext i32 %k.02 to i64
+  %arrayidx12 = getelementptr inbounds [3000 x i32], [3000 x i32]* %A, i64 %idxprom9, i64 %idxprom11
+  store i32 %tmp9, i32* %arrayidx12, align 4
+  %idxprom13 = sext i32 %k.02 to i64
+  %idxprom15 = sext i32 %k.02 to i64
+  %arrayidx16 = getelementptr inbounds [3000 x i32], [3000 x i32]* %B, i64 %idxprom13, i64 %idxprom15
+  %tmp17 = load i32, i32* %arrayidx16, align 4
+  %idxprom17 = sext i32 %k.02 to i64
+  %tmp21 = load i32, i32* %invariant, align 4
+  %idxprom19 = sext i32 %tmp21 to i64
+  %arrayidx20 = getelementptr inbounds [3000 x i32], [3000 x i32]* %A, i64 %idxprom17, i64 %idxprom19
+  %tmp22 = load i32, i32* %arrayidx20, align 4
+  %add = add nsw i32 %tmp22, %tmp17
+  store i32 %add, i32* %arrayidx20, align 4
+  %inc = add nuw nsw i32 %k.02, 1
+  %cmp5 = icmp slt i32 %inc, %n
+  br i1 %cmp5, label %for.body6, label %for.cond4.for.inc21_crit_edge
+
+for.cond4.for.inc21_crit_edge:                    ; preds = %for.body6
+  br label %for.inc21
+
+for.inc21:                                        ; preds = %for.cond4.for.inc21_crit_edge, %for.cond4.preheader
+  %inc22 = add nuw nsw i32 %j.04, 1
+  %cmp2 = icmp slt i32 %inc22, %n
+  br i1 %cmp2, label %for.cond4.preheader, label %for.cond1.for.inc24_crit_edge
+
+for.cond1.for.inc24_crit_edge:                    ; preds = %for.inc21
+  br label %for.inc24
+
+for.inc24:                                        ; preds = %for.cond1.for.inc24_crit_edge, %for.cond1.preheader
+  %inc25 = add nuw nsw i32 %i.07, 1
+  %cmp = icmp slt i32 %inc25, %n
+  br i1 %cmp, label %for.cond1.preheader, label %for.cond.for.end26_crit_edge
+
+for.cond.for.end26_crit_edge:                     ; preds = %for.inc24
+  br label %for.end26
+
+for.end26:                                        ; preds = %for.cond.for.end26_crit_edge, %entry.split
+  ret void
+}