[Polly] [PPCGCodeGeneration] Make sure invariant loads are generated before trying to generate LLVM from GPUNodeBuilder. [NFC]

[bollu]

Summary

We need to call preloadInvariantLoads to make sure that invariant loads are available for
BlockGenerators during copying.

Old hypothesis (left for context)

• It would appear that we were not skipping invariant loads in BlockGenerators::copyInstruction. We should skip these, since we create new instructions that load the value at the hoisted location.

rename test again

triim trailing whitespace

[bollu]

lib/CodeGen/BlockGenerators.cpp

198 ↗

(On Diff #106621)

@grosser, @Meinersbur - I'm not sure about this change. I'm pinging Michael since he wrote the VirtualUse stuff :).

[bollu]

Nope, this is definitely incorrect, we generate incorrect IR in the kernel module.

[grosser]

lib/CodeGen/BlockGenerators.cpp

198 ↗

(On Diff #106621)

Why do you need this at all. Is skipping the load instruction not enough?

[bollu]

• Call preloadInvariantLoads() before invoking GPUNodeBuilder::create.

[bollu]

here's the generated code for this example.

[bollu]

• [NFC wrt patch] remove debug metadata from testcase.

[grosser]

LGTM

lib/CodeGen/PPCGCodeGeneration.cpp

void generateCode(__isl_take isl_ast_node *Root, gpu_prog *Prog) {
     isl_ast_build_free(Build);
 
     Value *RTC = NodeBuilder.createRTC(Condition);
     Builder.GetInsertBlock()->getTerminator()->setOperand(0, RTC);
 
     Builder.SetInsertPoint(&*StartBlock->begin());
 
     NodeBuilder.initializeAfterRTH();
+    NodeBuilder.preloadInvariantLoads();
     NodeBuilder.create(Root);
     NodeBuilder.finalize();
 
     /// In case a sequential kernel has more surrounding loops as any parallel
     /// kernel, the SCoP is probably mostly sequential. Hence, there is no
     /// point in running it on a GPU.
     if (NodeBuilder.DeepestSequential > NodeBuilder.DeepestParallel)
       CondBr->setOperand(0, Builder.getFalse());

test/GPGPU/invariant-load-array-access.ll

+; RUN: opt %loadPolly -analyze -polly-scops \
+; RUN: -polly-invariant-load-hoisting < %s| FileCheck %s -check-prefix=SCOP
+
+; RUN: opt %loadPolly -S -polly-codegen-ppcg \
+; RUN: -polly-invariant-load-hoisting < %s | FileCheck %s -check-prefix=HOST-IR
+
+
+; REQUIRES: pollyacc
+
+; Check that we detect a scop.
+; SCOP:      Function: f
+; SCOP-NEXT: Region: %for.body---%for.end
+; SCOP-NEXT: Max Loop Depth:  1
+; SCOP-NEXT: Invariant Accesses: {
+; SCOP-NEXT:         ReadAccess :=  [Reduction Type: NONE] [Scalar: 0]
+; SCOP-NEXT:             [tmp] -> { Stmt_for_body[i0] -> MemRef_control[0] };
+; SCOP-NEXT:         Execution Context: [tmp] -> {  :  }
+; SCOP-NEXT:         ReadAccess :=  [Reduction Type: NONE] [Scalar: 0]
+; SCOP-NEXT:             [tmp] -> { Stmt_if_then[i0] -> MemRef_readarr[0] };
+; SCOP-NEXT:         Execution Context: [tmp] -> {  : tmp >= 4 }
+; SCOP-NEXT: }
+
+; 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*)
+
+; This test makes sure that such an access pattern is handled correctly
+; by PPCGCodeGeneration. It appears that not calling `preloadInvariantLoads`
+; was the main reason that caused this test case to crash.
+;
+; void f(int *arr, const int *control, const int *readarr) {
+;     for(int i = 0; i < 1000; i++) {
+;         int t = 0;
+;         if (*control > 3) {
+;             t += *readarr;
+;         }
+;         arr[i] = t;
+;     }
+; }
+
+
+target datalayout = "e-m:o-p:32:32-f64:32:64-f80:128-n8:16:32-S128"
+target triple = "i386-apple-macosx10.12.0"
+define void @f(i32* %arr, i32* %control, i32* %readarr) {
+entry:
+  br label %entry.split
+
+entry.split:                                      ; preds = %entry
+  br label %for.body
+
+for.body:                                         ; preds = %entry.split, %if.end
+  %i.01 = phi i32 [ 0, %entry.split ], [ %inc, %if.end ]
+  %tmp = load i32, i32* %control, align 4, !tbaa !4
+  %cmp1 = icmp sgt i32 %tmp, 3
+  br i1 %cmp1, label %if.then, label %if.end
+
+if.then:                                          ; preds = %for.body
+  %tmp1 = load i32, i32* %readarr, align 4, !tbaa !4
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %for.body
+  %t.0 = phi i32 [ %tmp1, %if.then ], [ 0, %for.body ]
+  %arrayidx = getelementptr inbounds i32, i32* %arr, i32 %i.01
+  store i32 %t.0, i32* %arrayidx, align 4, !tbaa !4
+  %inc = add nuw nsw i32 %i.01, 1
+  %exitcond = icmp eq i32 %inc, 1000
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %if.end
+  ret void
+}
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.start.p0i8(i64, i8* nocapture) #0
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.lifetime.end.p0i8(i64, i8* nocapture) #0
+
+attributes #0 = { argmemonly nounwind }
+
+!llvm.module.flags = !{!0, !1, !2}
+!llvm.ident = !{!3}
+
+!0 = !{i32 1, !"NumRegisterParameters", i32 0}
+!1 = !{i32 1, !"wchar_size", i32 4}
+!2 = !{i32 7, !"PIC Level", i32 2}
+!3 = !{!"clang version 5.0.0 (http://llvm.org/git/clang.git a5e65877526e030c5488a30cedb5fb743c42ad7d) (http://llvm.org/git/llvm.git 287aa279ce5f8ede095a9d423c1271cd769625be)"}
+!4 = !{!5, !5, i64 0}
+!5 = !{!"int", !6, i64 0}
+!6 = !{!"omnipotent char", !7, i64 0}
+!7 = !{!"Simple C/C++ TBAA"}