test/Analysis/LoopAccessAnalysis/clamped-access-pattern.ll

; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 6
; RUN: opt -passes='print<access-info>' -disable-output  < %s 2>&1 | FileCheck %s

target datalayout = "e-m:o-i64:64-p1:128:128-f80:128-n8:16:32:64-S128"

; for (i = 0; i < 1024; i++)
;   A[i % 16] = A[i % 16] + 1;
define void @load_store_same_clamped_i32(ptr %a) {
; CHECK-LABEL: 'load_store_same_clamped_i32'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            %lv = load i32, ptr %gep, align 4 ->
; CHECK-NEXT:            store i32 %add, ptr %gep, 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-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep = getelementptr inbounds i32, ptr %a, i64 %idx:
; CHECK-NEXT:        ((4 * (zext i4 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 16
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  %lv = load i32, ptr %gep, align 4
  %add = add i32 %lv, 1
  store i32 %add, ptr %gep, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[i % 8] = B[i];
define void @store_clamped_load_linear(ptr %a, ptr %b) {
; CHECK-LABEL: 'store_clamped_load_linear'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe with run-time checks
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:      Run-time memory checks:
; CHECK-NEXT:      Check 0:
; CHECK-NEXT:        Comparing group GRP0:
; CHECK-NEXT:          %gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
; CHECK-NEXT:        Against group GRP1:
; CHECK-NEXT:          %gep.a = getelementptr inbounds i32, ptr %a, i64 %idx.mod
; CHECK-NEXT:      Grouped accesses:
; CHECK-NEXT:        Group GRP0:
; CHECK-NEXT:          (Low: %b High: (4096 + %b))
; CHECK-NEXT:            Member: {%b,+,4}<nuw><%loop>
; CHECK-NEXT:        Group GRP1:
; CHECK-NEXT:          (Low: %a High: (4096 + %a))
; CHECK-NEXT:            Member: {%a,+,4}<nw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT:      SCEV assumptions:
; CHECK-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep.a = getelementptr inbounds i32, ptr %a, i64 %idx.mod:
; CHECK-NEXT:        ((4 * (zext i3 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx.mod = urem i64 %iv, 8
  %gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
  %lv = load i32, ptr %gep.b, align 4
  %gep.a = getelementptr inbounds i32, ptr %a, i64 %idx.mod
  store i32 %lv, ptr %gep.a, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[i % 4] = i;
define void @only_store_clamped(ptr %a) {
; CHECK-LABEL: 'only_store_clamped'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; 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 %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 4
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  %val = trunc i64 %iv to i32
  store i32 %val, ptr %gep, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[i % 8] = A[i % 8] + 1;
define void @clamped_power_of_2(ptr %a) {
; CHECK-LABEL: 'clamped_power_of_2'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            %lv = load i32, ptr %gep, align 4 ->
; CHECK-NEXT:            store i32 %add, ptr %gep, 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-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep = getelementptr inbounds i32, ptr %a, i64 %idx:
; CHECK-NEXT:        ((4 * (zext i3 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 8
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  %lv = load i32, ptr %gep, align 4
  %add = add i32 %lv, 1
  store i32 %add, ptr %gep, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[i % 7] = A[i % 7] + 1;
define void @clamped_non_power_of_2(ptr %a) {
; CHECK-LABEL: 'clamped_non_power_of_2'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
; CHECK-NEXT:  Unsafe indirect dependence.
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        IndirectUnsafe:
; CHECK-NEXT:            %lv = load i32, ptr %gep, align 4 ->
; CHECK-NEXT:            store i32 %add, ptr %gep, 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 %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 7
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  %lv = load i32, ptr %gep, align 4
  %add = add i32 %lv, 1
  store i32 %add, ptr %gep, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++) {
;   A[i % 8] = x;
;   A[i % 8] = y;
; }
define void @two_stores_same_clamped(ptr %a, i32 %x, i32 %y) {
; CHECK-LABEL: 'two_stores_same_clamped'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; 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 %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 8
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  store i32 %x, ptr %gep, align 4
  store i32 %y, ptr %gep, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[i % 8] = A[(i + 1) % 8];
define void @clamped_offset_load(ptr %a) {
; CHECK-LABEL: 'clamped_offset_load'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            %lv = load i32, ptr %gep.load, align 4 ->
; CHECK-NEXT:            store i32 %lv, ptr %gep.store, align 4
; CHECK-EMPTY:
; CHECK-NEXT:      Run-time memory checks:
; CHECK-NEXT:      Grouped accesses:
; CHECK-NEXT:        Group GRP0:
; CHECK-NEXT:          (Low: %a High: (4100 + %a))
; CHECK-NEXT:            Member: {(4 + %a),+,4}<nw><%loop>
; CHECK-NEXT:            Member: {%a,+,4}<nw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT:      SCEV assumptions:
; CHECK-NEXT:      {1,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep.load = getelementptr inbounds i32, ptr %a, i64 %idx.load:
; CHECK-NEXT:        ((4 * (zext i3 {1,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {(4 + %a),+,4}<nw><%loop>
; CHECK-NEXT:      [PSE] %gep.store = getelementptr inbounds i32, ptr %a, i64 %idx.store:
; CHECK-NEXT:        ((4 * (zext i3 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %iv.plus1 = add nuw nsw i64 %iv, 1
  %idx.store = urem i64 %iv, 8
  %idx.load = urem i64 %iv.plus1, 8
  %gep.load = getelementptr inbounds i32, ptr %a, i64 %idx.load
  %lv = load i32, ptr %gep.load, align 4
  %gep.store = getelementptr inbounds i32, ptr %a, i64 %idx.store
  store i32 %lv, ptr %gep.store, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[i % 2] += 1;
define void @clamped_small_bound(ptr %a) {
; CHECK-LABEL: 'clamped_small_bound'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            %lv = load i32, ptr %gep, align 4 ->
; CHECK-NEXT:            store i32 %add, ptr %gep, 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-NEXT:      {false,+,true}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep = getelementptr inbounds i32, ptr %a, i64 %idx:
; CHECK-NEXT:        ((4 * (zext i1 {false,+,true}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,-4}<%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 2
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  %lv = load i32, ptr %gep, align 4
  %add = add i32 %lv, 1
  store i32 %add, ptr %gep, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   sum += A[i % 8];
define i32 @clamped_read_only(ptr noalias %a) {
; CHECK-LABEL: 'clamped_read_only'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; 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 %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %sum = phi i32 [ 0, %entry ], [ %sum.next, %loop ]
  %idx = urem i64 %iv, 8
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  %lv = load i32, ptr %gep, align 4
  %sum.next = add i32 %sum, %lv
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret i32 %sum.next
}

; for (i = 0; i < 1024; i++)
;   A[i % 8] = A[i] + 1;
define void @clamped_and_linear_same_array(ptr %a) {
; CHECK-LABEL: 'clamped_and_linear_same_array'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            %lv = load i32, ptr %gep.load, align 4 ->
; CHECK-NEXT:            store i32 %add, ptr %gep.store, align 4
; CHECK-EMPTY:
; CHECK-NEXT:      Run-time memory checks:
; CHECK-NEXT:      Grouped accesses:
; CHECK-NEXT:        Group GRP0:
; CHECK-NEXT:          (Low: %a High: (4096 + %a))
; CHECK-NEXT:            Member: {%a,+,4}<nuw><%loop>
; CHECK-NEXT:            Member: {%a,+,4}<nuw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT:      SCEV assumptions:
; CHECK-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep.store = getelementptr inbounds i32, ptr %a, i64 %idx.mod:
; CHECK-NEXT:        ((4 * (zext i3 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nuw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx.mod = urem i64 %iv, 8
  %gep.load = getelementptr inbounds i32, ptr %a, i64 %iv
  %lv = load i32, ptr %gep.load, align 4
  %add = add i32 %lv, 1
  %gep.store = getelementptr inbounds i32, ptr %a, i64 %idx.mod
  store i32 %add, ptr %gep.store, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[i % 1] = i;
define void @clamped_one(ptr %a) {
; CHECK-LABEL: 'clamped_one'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; 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 %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 1
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  %val = trunc i64 %iv to i32
  store i32 %val, ptr %gep, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[i % 4] = A[i % 8] + 1;
define void @different_clamped_bounds_same_array(ptr %a) {
; CHECK-LABEL: 'different_clamped_bounds_same_array'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            %lv = load i32, ptr %gep.load, align 4 ->
; CHECK-NEXT:            store i32 %add, ptr %gep.store, align 4
; CHECK-EMPTY:
; CHECK-NEXT:      Run-time memory checks:
; CHECK-NEXT:      Grouped accesses:
; CHECK-NEXT:        Group GRP0:
; CHECK-NEXT:          (Low: %a High: (4096 + %a))
; CHECK-NEXT:            Member: {%a,+,4}<nw><%loop>
; CHECK-NEXT:            Member: {%a,+,4}<nw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT:      SCEV assumptions:
; CHECK-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep.load = getelementptr inbounds i32, ptr %a, i64 %idx.load:
; CHECK-NEXT:        ((4 * (zext i3 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
; CHECK-NEXT:      [PSE] %gep.store = getelementptr inbounds i32, ptr %a, i64 %idx.store:
; CHECK-NEXT:        ((4 * (zext i2 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx.load = urem i64 %iv, 8
  %idx.store = urem i64 %iv, 4
  %gep.load = getelementptr inbounds i32, ptr %a, i64 %idx.load
  %lv = load i32, ptr %gep.load, align 4
  %add = add i32 %lv, 1
  %gep.store = getelementptr inbounds i32, ptr %a, i64 %idx.store
  store i32 %add, ptr %gep.store, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[(2*i) % 8] = i;
define void @clamped_non_unit_step(ptr %a) {
; CHECK-LABEL: 'clamped_non_unit_step'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; 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 %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %iv2 = shl i64 %iv, 1
  %idx = urem i64 %iv2, 8
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  %val = trunc i64 %iv to i32
  store i32 %val, ptr %gep, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[i % 8] = i;   // i8 type (1 byte)
define void @clamped_i8_type(ptr %a) {
; CHECK-LABEL: 'clamped_i8_type'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; 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 %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 8
  %gep = getelementptr inbounds i8, ptr %a, i64 %idx
  %val = trunc i64 %iv to i8
  store i8 %val, ptr %gep, align 1
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[i % 4] += 1;   // i64 type (8 bytes)
define void @clamped_i64_type(ptr %a) {
; CHECK-LABEL: 'clamped_i64_type'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            %lv = load i64, ptr %gep, align 8 ->
; CHECK-NEXT:            store i64 %add, ptr %gep, align 8
; 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-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep = getelementptr inbounds i64, ptr %a, i64 %idx:
; CHECK-NEXT:        ((8 * (zext i2 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,8}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 4
  %gep = getelementptr inbounds i64, ptr %a, i64 %idx
  %lv = load i64, ptr %gep, align 8
  %add = add i64 %lv, 1
  store i64 %add, ptr %gep, align 8
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < N; i++)
;   A[i % C] = B[i];
define void @clamped_non_constant(ptr %a, ptr %b, i64 %c) {
; CHECK-LABEL: 'clamped_non_constant'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Report: cannot identify array bounds
; CHECK-NEXT:      Dependences:
; 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 %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, %c
  %gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
  %lv = load i32, ptr %gep.b, align 4
  %gep.a = getelementptr inbounds i32, ptr %a, i64 %idx
  store i32 %lv, ptr %gep.a, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; for (i = 0; i < 1024; i++)
;   A[i % 16] = B[i % 16] + 1;
define void @clamped_different_arrays_same_bound(ptr %a, ptr %b) {
; CHECK-LABEL: 'clamped_different_arrays_same_bound'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe with run-time checks
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:      Run-time memory checks:
; CHECK-NEXT:      Check 0:
; CHECK-NEXT:        Comparing group GRP0:
; CHECK-NEXT:          %gep.a = getelementptr inbounds i32, ptr %a, i64 %idx
; CHECK-NEXT:        Against group GRP1:
; CHECK-NEXT:          %gep.b = getelementptr inbounds i32, ptr %b, i64 %idx
; CHECK-NEXT:      Grouped accesses:
; CHECK-NEXT:        Group GRP0:
; CHECK-NEXT:          (Low: %a High: (4096 + %a))
; CHECK-NEXT:            Member: {%a,+,4}<nw><%loop>
; CHECK-NEXT:        Group GRP1:
; CHECK-NEXT:          (Low: %b High: (4096 + %b))
; CHECK-NEXT:            Member: {%b,+,4}<nw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT:      SCEV assumptions:
; CHECK-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep.b = getelementptr inbounds i32, ptr %b, i64 %idx:
; CHECK-NEXT:        ((4 * (zext i4 {0,+,1}<%loop> to i64))<nuw><nsw> + %b)<nuw>
; CHECK-NEXT:        --> {%b,+,4}<nw><%loop>
; CHECK-NEXT:      [PSE] %gep.a = getelementptr inbounds i32, ptr %a, i64 %idx:
; CHECK-NEXT:        ((4 * (zext i4 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 16
  %gep.b = getelementptr inbounds i32, ptr %b, i64 %idx
  %lv = load i32, ptr %gep.b, align 4
  %add = add i32 %lv, 1
  %gep.a = getelementptr inbounds i32, ptr %a, i64 %idx
  store i32 %add, ptr %gep.a, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

; Two stores with different modulo bounds to same array.
; for (i = 0; i < 1024; i++) {
;   A[i % 4] = x;
;   A[i % 8] = y;
; }
define void @two_stores_different_clamped(ptr %a, i32 %x, i32 %y) {
; CHECK-LABEL: 'two_stores_different_clamped'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            store i32 %x, ptr %gep4, align 4 ->
; CHECK-NEXT:            store i32 %y, ptr %gep8, align 4
; CHECK-EMPTY:
; CHECK-NEXT:      Run-time memory checks:
; CHECK-NEXT:      Grouped accesses:
; CHECK-NEXT:        Group GRP0:
; CHECK-NEXT:          (Low: %a High: (4096 + %a))
; CHECK-NEXT:            Member: {%a,+,4}<nw><%loop>
; CHECK-NEXT:            Member: {%a,+,4}<nw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT:      SCEV assumptions:
; CHECK-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep4 = getelementptr inbounds i32, ptr %a, i64 %idx4:
; CHECK-NEXT:        ((4 * (zext i2 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
; CHECK-NEXT:      [PSE] %gep8 = getelementptr inbounds i32, ptr %a, i64 %idx8:
; CHECK-NEXT:        ((4 * (zext i3 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx4 = urem i64 %iv, 4
  %idx8 = urem i64 %iv, 8
  %gep4 = getelementptr inbounds i32, ptr %a, i64 %idx4
  store i32 %x, ptr %gep4, align 4
  %gep8 = getelementptr inbounds i32, ptr %a, i64 %idx8
  store i32 %y, ptr %gep8, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

define void @srem_not_matched_as_clamped(ptr %a) {
; CHECK-LABEL: 'srem_not_matched_as_clamped'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; 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 %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = srem i64 %iv, 8
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  %val = trunc i64 %iv to i32
  store i32 %val, ptr %gep, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

define void @clamped_multi_index_gep(ptr %a) {
; CHECK-LABEL: 'clamped_multi_index_gep'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; 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 %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 7
  %gep = getelementptr inbounds [16 x i32], ptr %a, i64 0, i64 %idx
  %val = trunc i64 %iv to i32
  store i32 %val, ptr %gep, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

define void @clamped_mod_256_narrow_cutoff(ptr %a) {
; CHECK-LABEL: 'clamped_mod_256_narrow_cutoff'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            %lv = load i32, ptr %gep, align 4 ->
; CHECK-NEXT:            store i32 %add, ptr %gep, 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-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep = getelementptr inbounds i32, ptr %a, i64 %idx:
; CHECK-NEXT:        ((4 * (zext i8 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 256
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  %lv = load i32, ptr %gep, align 4
  %add = add i32 %lv, 1
  store i32 %add, ptr %gep, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

define void @clamped_with_const_offset(ptr %a) {
; CHECK-LABEL: 'clamped_with_const_offset'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            %lv = load i32, ptr %gep.off, align 4 ->
; CHECK-NEXT:            store i32 %add, ptr %gep.off, 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-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep.off = getelementptr inbounds i8, ptr %gep, i64 16:
; CHECK-NEXT:        (16 + (4 * (zext i2 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {(16 + %a),+,4}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 4
  %gep = getelementptr inbounds i32, ptr %a, i64 %idx
  %gep.off = getelementptr inbounds i8, ptr %gep, i64 16
  %lv = load i32, ptr %gep.off, align 4
  %add = add i32 %lv, 1
  store i32 %add, ptr %gep.off, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

define void @clamped_load_clamped_store_same(ptr %a) {
; CHECK-LABEL: 'clamped_load_clamped_store_same'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
; CHECK-NEXT:  Backward loop carried data dependence.
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Backward:
; CHECK-NEXT:            %lv = load i32, ptr %gep.l, align 4 ->
; CHECK-NEXT:            store i32 %add, ptr %gep.s, align 4
; CHECK-EMPTY:
; CHECK-NEXT:      Run-time memory checks:
; CHECK-NEXT:      Grouped accesses:
; CHECK-NEXT:        Group GRP0:
; CHECK-NEXT:          (Low: %a High: (4100 + %a))
; CHECK-NEXT:            Member: {%a,+,4}<nw><%loop>
; CHECK-NEXT:            Member: {(4 + %a),+,4}<nw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT:      SCEV assumptions:
; CHECK-NEXT:      {1,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep.l = getelementptr inbounds i32, ptr %a, i64 %idx.l:
; CHECK-NEXT:        ((4 * (zext i2 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
; CHECK-NEXT:      [PSE] %gep.s = getelementptr inbounds i32, ptr %a, i64 %idx.s:
; CHECK-NEXT:        ((4 * (zext i2 {1,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {(4 + %a),+,4}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx.l = urem i64 %iv, 4
  %iv.p1 = add i64 %iv, 1
  %idx.s = urem i64 %iv.p1, 4
  %gep.l = getelementptr inbounds i32, ptr %a, i64 %idx.l
  %gep.s = getelementptr inbounds i32, ptr %a, i64 %idx.s
  %lv = load i32, ptr %gep.l, align 4
  %add = add i32 %lv, 1
  store i32 %add, ptr %gep.s, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

define void @clamped_i24_store_size(ptr %a, ptr %b) {
; CHECK-LABEL: 'clamped_i24_store_size'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Memory dependences are safe with run-time checks
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        Forward:
; CHECK-NEXT:            %lv = load i24, ptr %gep.a, align 4 ->
; CHECK-NEXT:            store i24 %add, ptr %gep.a, align 4
; CHECK-EMPTY:
; CHECK-NEXT:      Run-time memory checks:
; CHECK-NEXT:      Check 0:
; CHECK-NEXT:        Comparing group GRP0:
; CHECK-NEXT:          %gep.b = getelementptr inbounds i24, ptr %b, i64 %iv
; CHECK-NEXT:        Against group GRP1:
; CHECK-NEXT:          %gep.a = getelementptr inbounds i24, ptr %a, i64 %idx
; CHECK-NEXT:          %gep.a = getelementptr inbounds i24, ptr %a, i64 %idx
; CHECK-NEXT:      Grouped accesses:
; CHECK-NEXT:        Group GRP0:
; CHECK-NEXT:          (Low: %b High: (4095 + %b))
; CHECK-NEXT:            Member: {%b,+,4}<nuw><%loop>
; CHECK-NEXT:        Group GRP1:
; CHECK-NEXT:          (Low: %a High: (4095 + %a))
; CHECK-NEXT:            Member: {%a,+,4}<nw><%loop>
; CHECK-NEXT:            Member: {%a,+,4}<nw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT:      SCEV assumptions:
; CHECK-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep.a = getelementptr inbounds i24, ptr %a, i64 %idx:
; CHECK-NEXT:        ((4 * (zext i2 {0,+,1}<%loop> to i64))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,4}<nw><%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 4
  %gep.a = getelementptr inbounds i24, ptr %a, i64 %idx
  %lv = load i24, ptr %gep.a, align 4
  %add = add i24 %lv, 1
  store i24 %add, ptr %gep.a, align 4
  %gep.b = getelementptr inbounds i24, ptr %b, i64 %iv
  store i24 %add, ptr %gep.b, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

define void @clamped_scalable_with_dep(ptr %a, ptr %b) {
; CHECK-LABEL: 'clamped_scalable_with_dep'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
; CHECK-NEXT:  Unsafe indirect dependence.
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        IndirectUnsafe:
; CHECK-NEXT:            store <vscale x 4 x i32> zeroinitializer, ptr %gep.a, align 16 ->
; CHECK-NEXT:            store i32 %lv, ptr %gep.b2, align 4
; CHECK-EMPTY:
; CHECK-NEXT:      Run-time memory checks:
; CHECK-NEXT:      Check 0:
; CHECK-NEXT:        Comparing group GRP0:
; CHECK-NEXT:          %gep.b2 = getelementptr inbounds i32, ptr %a, i64 %iv
; CHECK-NEXT:        Against group GRP2:
; CHECK-NEXT:          %gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
; CHECK-NEXT:      Check 1:
; CHECK-NEXT:        Comparing group GRP1:
; CHECK-NEXT:          %gep.a = getelementptr inbounds <vscale x 4 x i32>, ptr %a, i64 %idx
; CHECK-NEXT:        Against group GRP2:
; CHECK-NEXT:          %gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
; CHECK-NEXT:      Grouped accesses:
; CHECK-NEXT:        Group GRP0:
; CHECK-NEXT:          (Low: %a High: (4096 + %a))
; CHECK-NEXT:            Member: {%a,+,4}<nuw><%loop>
; CHECK-NEXT:        Group GRP1:
; CHECK-NEXT:          (Low: (((16368 * vscale) + %a) umin %a) High: ((16 * vscale) + (((16368 * vscale) + %a) umax %a)))
; CHECK-NEXT:            Member: {%a,+,(16 * vscale)}<%loop>
; CHECK-NEXT:        Group GRP2:
; CHECK-NEXT:          (Low: %b High: (4096 + %b))
; CHECK-NEXT:            Member: {%b,+,4}<nuw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT:      SCEV assumptions:
; CHECK-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE] %gep.a = getelementptr inbounds <vscale x 4 x i32>, ptr %a, i64 %idx:
; CHECK-NEXT:        ((16 * vscale * (zext i2 {0,+,1}<%loop> to i64)) + %a)
; CHECK-NEXT:        --> {%a,+,(16 * vscale)}<%loop>
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i64 %iv, 4
  %gep.a = getelementptr inbounds <vscale x 4 x i32>, ptr %a, i64 %idx
  store <vscale x 4 x i32> zeroinitializer, ptr %gep.a, align 16
  %gep.b = getelementptr inbounds i32, ptr %b, i64 %iv
  %lv = load i32, ptr %gep.b, align 4
  %gep.b2 = getelementptr inbounds i32, ptr %a, i64 %iv
  store i32 %lv, ptr %gep.b2, align 4
  %iv.next = add nuw nsw i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 1024
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

define void @clamped_mul_huge_scale_as1(ptr addrspace(1) %a) {
; CHECK-LABEL: 'clamped_mul_huge_scale_as1'
; CHECK-NEXT:    loop:
; CHECK-NEXT:      Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
; CHECK-NEXT:    Unsafe indirect dependence.
; CHECK-NEXT:      Dependences:
; CHECK-NEXT:        IndirectUnsafe:
; CHECK-NEXT:            %ld = load i64, ptr addrspace(1) %gep, align 8 ->
; CHECK-NEXT:            store i64 %add, ptr addrspace(1) %gep, align 8
; 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-NEXT:      {0,+,1}<%loop> Added Flags: <nusw>
; CHECK-EMPTY:
; CHECK-NEXT:      Expressions re-written:
; CHECK-NEXT:      [PSE]  %gep = getelementptr inbounds i8, ptr addrspace(1) %a, i128 %off:
; CHECK-NEXT:        ((36893488147419103232 * (zext i2 {0,+,1}<%loop> to i128))<nuw><nsw> + %a)<nuw>
; CHECK-NEXT:        --> {%a,+,36893488147419103232}<nw><%loop>
;
entry:
  br label %loop
loop:
  %iv = phi i128 [ 0, %entry ], [ %iv.next, %loop ]
  %idx = urem i128 %iv, 4
  %off = mul i128 %idx, 36893488147419103232
  %gep = getelementptr inbounds i8, ptr addrspace(1) %a, i128 %off
  %ld = load i64, ptr addrspace(1) %gep, align 8
  %add = add i64 %ld, 1
  store i64 %add, ptr addrspace(1) %gep, align 8
  %iv.next = add nuw nsw i128 %iv, 1
  %ec = icmp eq i128 %iv.next, 1024
  br i1 %ec, label %exit, label %loop
exit:
  ret void
}