[LAA] Improve convergent tests (#136758)

LoopAccessAnalysis has code for handling function calls where the
function is marked with the 'convergent' attribute, but the test
coverage is insufficient. Fix this by adding a test showing the case of
no-runtime-checks adapted from LoopDistribute, and clean up the existing
test with runtime-checks. Also regenerate the test file with
UpdateTestChecks.

Author: artagnon

llvm/test/Analysis/LoopAccessAnalysis/unsafe-and-rt-checks-convergent.ll

@@ -1,41 +1,52 @@
-; RUN: opt -passes='print<access-info>' -disable-output  < %s 2>&1 | FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes='print<access-info>' -disable-output %s 2>&1 | FileCheck %s
 
 ; Analyze this loop:
 ;   for (i = 0; i < n; i++)
 ;    A[i + 1] = A[i] * B[i] * C[i];
 
-target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
-
-; CHECK: for.body:
-; CHECK: Has convergent operation in loop
-; CHECK: Report: cannot add control dependency to convergent operation
-; CHECK-NEXT: Dependences:
-; CHECK-NEXT:   Backward:
-; CHECK-NEXT:     %loadA = load i16, ptr %arrayidxA, align 2 ->
-; CHECK-NEXT:     store i16 %mul1, ptr %arrayidxA_plus_2, align 2
-; CHECK: Run-time memory checks:
-; CHECK-NEXT: 0:
-; CHECK-NEXT: Comparing group
-; CHECK-NEXT:   %arrayidxA = getelementptr inbounds i16, ptr %a, i64 %storemerge3
-; CHECK-NEXT:   %arrayidxA_plus_2 = getelementptr inbounds i16, ptr %a, i64 %add
-; CHECK-NEXT: Against group
-; CHECK-NEXT:   %arrayidxB = getelementptr inbounds i16, ptr %b, i64 %storemerge3
-; CHECK-NEXT: 1:
-; CHECK-NEXT: Comparing group
-; CHECK-NEXT:   %arrayidxA = getelementptr inbounds i16, ptr %a, i64 %storemerge3
-; CHECK-NEXT:   %arrayidxA_plus_2 = getelementptr inbounds i16, ptr %a, i64 %add
-; CHECK-NEXT: Against group
-; CHECK-NEXT:   %arrayidxC = getelementptr inbounds i16, ptr %c, i64 %storemerge3
-
-@B = common global ptr null, align 8
-@A = common global ptr null, align 8
-@C = common global ptr null, align 8
-
-define void @f() #1 {
+; In this case, runtime checks are needed, and there is a convergent operation.
+define void @rtchecks_needed(ptr %a, ptr %b, ptr %c) {
+; CHECK-LABEL: 'rtchecks_needed'
+; CHECK-NEXT:    for.body:
+; CHECK-NEXT:      Has convergent operation in loop
+; CHECK-NEXT:      Report: cannot add control dependency to convergent operation
+; CHECK-NEXT:      Dependences:
+; CHECK-NEXT:        Backward:
+; CHECK-NEXT:            %loadA = load i16, ptr %arrayidxA, align 2 ->
+; CHECK-NEXT:            store i16 %mul1, ptr %arrayidxA_plus_2, align 2
+; CHECK-EMPTY:
+; CHECK-NEXT:      Run-time memory checks:
+; CHECK-NEXT:      Check 0:
+; CHECK-NEXT:        Comparing group ([[GRP1:0x[0-9a-f]+]]):
+; CHECK-NEXT:          %arrayidxA = getelementptr inbounds i16, ptr %a, i64 %storemerge3
+; CHECK-NEXT:          %arrayidxA_plus_2 = getelementptr inbounds i16, ptr %a, i64 %add
+; CHECK-NEXT:        Against group ([[GRP2:0x[0-9a-f]+]]):
+; CHECK-NEXT:          %arrayidxB = getelementptr inbounds i16, ptr %b, i64 %storemerge3
+; CHECK-NEXT:      Check 1:
+; CHECK-NEXT:        Comparing group ([[GRP1]]):
+; CHECK-NEXT:          %arrayidxA = getelementptr inbounds i16, ptr %a, i64 %storemerge3
+; CHECK-NEXT:          %arrayidxA_plus_2 = getelementptr inbounds i16, ptr %a, i64 %add
+; CHECK-NEXT:        Against group ([[GRP3:0x[0-9a-f]+]]):
+; CHECK-NEXT:          %arrayidxC = getelementptr inbounds i16, ptr %c, i64 %storemerge3
+; CHECK-NEXT:      Grouped accesses:
+; CHECK-NEXT:        Group [[GRP1]]:
+; CHECK-NEXT:          (Low: %a High: (42 + %a))
+; CHECK-NEXT:            Member: {%a,+,2}<nuw><%for.body>
+; CHECK-NEXT:            Member: {(2 + %a),+,2}<nw><%for.body>
+; CHECK-NEXT:        Group [[GRP2]]:
+; CHECK-NEXT:          (Low: %b High: (40 + %b))
+; CHECK-NEXT:            Member: {%b,+,2}<nuw><%for.body>
+; CHECK-NEXT:        Group [[GRP3]]:
+; CHECK-NEXT:          (Low: %c High: (40 + %c))
+; CHECK-NEXT:            Member: {%c,+,2}<nuw><%for.body>
+; CHECK-EMPTY:
+; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
+; CHECK-NEXT:      SCEV assumptions:
+; CHECK-EMPTY:
+; CHECK-NEXT:      Expressions re-written:
+;
 entry:
-  %a = load ptr, ptr @A, align 8
-  %b = load ptr, ptr @B, align 8
-  %c = load ptr, ptr @C, align 8
   br label %for.body
 
 for.body:                                         ; preds = %for.body, %entry
@@ -50,7 +61,7 @@ for.body:                                         ; preds = %for.body, %entry
   %arrayidxC = getelementptr inbounds i16, ptr %c, i64 %storemerge3
   %loadC = load i16, ptr %arrayidxC, align 2
 
-  call void @llvm.convergent()
+  call i16 @llvm.convergent(i16 %loadC)
 
   %mul = mul i16 %loadB, %loadA
   %mul1 = mul i16 %mul, %loadC
@@ -66,7 +77,62 @@ for.end:                                          ; preds = %for.body
   ret void
 }
 
-declare void @llvm.convergent() #0
+; In this case, no runtime checks are needed, and there is a convergent operation.
+define void @no_rtchecks(ptr noalias %a, ptr noalias %b, ptr noalias %c, ptr noalias %d, ptr noalias %e) {
+; CHECK-LABEL: 'no_rtchecks'
+; CHECK-NEXT:    for.body:
+; CHECK-NEXT:      Has convergent operation in loop
+; CHECK-NEXT:      Report: cannot add control dependency to convergent operation
+; CHECK-NEXT:      Dependences:
+; CHECK-NEXT:        Backward:
+; CHECK-NEXT:            %loadA = load i16, ptr %arrayidxA, align 4 ->
+; CHECK-NEXT:            store i16 %mulA, ptr %arrayidxA_plus_4, align 4
+; CHECK-EMPTY:
+; CHECK-NEXT:      Run-time memory checks:
+; CHECK-NEXT:      Grouped accesses:
+; CHECK-EMPTY:
+; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
+; CHECK-NEXT:      SCEV assumptions:
+; CHECK-EMPTY:
+; CHECK-NEXT:      Expressions re-written:
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
+
+  %arrayidxA = getelementptr inbounds i16, ptr %a, i64 %ind
+  %loadA = load i16, ptr %arrayidxA, align 4
+
+  %arrayidxB = getelementptr inbounds i16, ptr %b, i64 %ind
+  %loadB = load i16, ptr %arrayidxB, align 4
+
+  %mulA = mul i16 %loadB, %loadA
+
+  %add = add nuw nsw i64 %ind, 1
+  %arrayidxA_plus_4 = getelementptr inbounds i16, ptr %a, i64 %add
+  store i16 %mulA, ptr %arrayidxA_plus_4, align 4
+
+  %arrayidxD = getelementptr inbounds i16, ptr %d, i64 %ind
+  %loadD = load i16, ptr %arrayidxD, align 4
+
+  %arrayidxE = getelementptr inbounds i16, ptr %e, i64 %ind
+  %loadE = load i16, ptr %arrayidxE, align 4
+
+  %convergentD = call i16 @llvm.convergent(i16 %loadD)
+  %mulC = mul i16 %convergentD, %loadE
+
+  %arrayidxC = getelementptr inbounds i16, ptr %c, i64 %ind
+  store i16 %mulC, ptr %arrayidxC, align 4
+
+  %exitcond = icmp eq i64 %add, 20
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+declare i16 @llvm.convergent(i16) #0
 
 attributes #0 = { nounwind readnone convergent }
-attributes #1 = { nounwind convergent }