[Analysis] Teach ScalarEvolution::getRangeRef about more dereferenceable objects

Whilst dealing with review comments on

llvm#96752

I discovered that SCEV does not know about the dereferenceable
attribute on function arguments so I have updated getRangeRef
to make use of it by calling getPointerDereferenceableBytes.

Author: david-arm

llvm/lib/Analysis/ScalarEvolution.cpp

@@ -6858,10 +6858,9 @@ const ConstantRange &ScalarEvolution::getRangeRef(
       ObjectSizeOpts Opts;
       Opts.RoundToAlign = false;
       Opts.NullIsUnknownSize = true;
-      uint64_t ObjSize;
-      if ((isa<GlobalVariable>(V) || isa<AllocaInst>(V) ||
-           isAllocationFn(V, &TLI)) &&
-          getObjectSize(V, ObjSize, DL, &TLI, Opts) && ObjSize > 1) {
+      bool CanBeNull, CanBeFreed;
+      uint64_t ObjSize = V->getPointerDereferenceableBytes(DL, CanBeNull, CanBeFreed);
+      if (ObjSize > 1) {
         // The highest address the object can start is ObjSize bytes before the
         // end (unsigned max value). If this value is not a multiple of the
         // alignment, the last possible start value is the next lowest multiple

llvm/test/Analysis/ScalarEvolution/different-loops-recs.ll

@@ -457,7 +457,7 @@ define void @test_05(i32 %N) {
 ; CHECK-NEXT:    %"alloca point" = bitcast i32 0 to i32
 ; CHECK-NEXT:    --> 0 U: [0,1) S: [0,1)
 ; CHECK-NEXT:    %tmp = getelementptr [1000 x i32], ptr @A, i32 0, i32 %i.0
-; CHECK-NEXT:    --> {(8 + @A)<nuw><nsw>,+,4}<nw><%bb3> U: [0,-3) S: [-9223372036854775808,9223372036854775805) Exits: (408 + @A) LoopDispositions: { %bb3: Computable }
+; CHECK-NEXT:    --> {(8 + @A)<nuw><nsw>,+,4}<nw><%bb3> U: [40,-3623) S: [-9223372036854775808,9223372036854775805) Exits: (408 + @A)<nuw> LoopDispositions: { %bb3: Computable }
 ; CHECK-NEXT:    %tmp2 = add i32 %i.0, 1
 ; CHECK-NEXT:    --> {3,+,1}<nuw><nsw><%bb3> U: [3,104) S: [3,104) Exits: 103 LoopDispositions: { %bb3: Computable }
 ; CHECK-NEXT:    %i.0 = phi i32 [ 2, %entry ], [ %tmp2, %bb ]

llvm/test/Analysis/ScalarEvolution/no-wrap-add-exprs.ll

@@ -183,15 +183,15 @@ define void @f3(ptr %x_addr, ptr %y_addr, ptr %tmp_addr) {
 ; CHECK-NEXT:    %s3.zext = zext i8 %s3 to i16
 ; CHECK-NEXT:    --> (1 + (zext i8 (4 + (32 * %x) + (36 * %y)) to i16))<nuw><nsw> U: [1,254) S: [1,257)
 ; CHECK-NEXT:    %ptr = bitcast ptr @z_addr to ptr
-; CHECK-NEXT:    --> @z_addr U: [0,-3) S: [-9223372036854775808,9223372036854775805)
+; CHECK-NEXT:    --> @z_addr U: [4,-19) S: [-9223372036854775808,9223372036854775805)
 ; CHECK-NEXT:    %int0 = ptrtoint ptr %ptr to i32
 ; CHECK-NEXT:    --> (trunc i64 (ptrtoint ptr @z_addr to i64) to i32) U: [0,-3) S: [-2147483648,2147483645)
 ; CHECK-NEXT:    %int5 = add i32 %int0, 5
 ; CHECK-NEXT:    --> (5 + (trunc i64 (ptrtoint ptr @z_addr to i64) to i32)) U: [5,2) S: [-2147483643,-2147483646)
 ; CHECK-NEXT:    %int.zext = zext i32 %int5 to i64
 ; CHECK-NEXT:    --> (1 + (zext i32 (4 + (trunc i64 (ptrtoint ptr @z_addr to i64) to i32)) to i64))<nuw><nsw> U: [1,4294967294) S: [1,4294967297)
 ; CHECK-NEXT:    %ptr_noalign = bitcast ptr @z_addr_noalign to ptr
-; CHECK-NEXT:    --> @z_addr_noalign U: full-set S: full-set
+; CHECK-NEXT:    --> @z_addr_noalign U: [1,-16) S: full-set
 ; CHECK-NEXT:    %int0_na = ptrtoint ptr %ptr_noalign to i32
 ; CHECK-NEXT:    --> (trunc i64 (ptrtoint ptr @z_addr_noalign to i64) to i32) U: full-set S: full-set
 ; CHECK-NEXT:    %int5_na = add i32 %int0_na, 5
@@ -362,3 +362,20 @@ loop:
 exit2:
   ret i1 false
 }
+
+
+define void @dereferenceable_arg(ptr dereferenceable(128) %len_addr, ptr dereferenceable(128) align(8) %len_addr2) {
+; CHECK-LABEL: 'dereferenceable_arg'
+; CHECK-NEXT:  Classifying expressions for: @dereferenceable_arg
+; CHECK-NEXT:    %ptr = bitcast ptr %len_addr to ptr
+; CHECK-NEXT:    --> %len_addr U: [1,-128) S: full-set
+; CHECK-NEXT:    %ptr2 = bitcast ptr %len_addr2 to ptr
+; CHECK-NEXT:    --> %len_addr2 U: [8,-135) S: [-9223372036854775808,9223372036854775801)
+; CHECK-NEXT:  Determining loop execution counts for: @dereferenceable_arg
+;
+  entry:
+  %ptr = bitcast ptr %len_addr to ptr
+  %ptr2 = bitcast ptr %len_addr2 to ptr
+
+  ret void
+}