[llvm] 4f6fc7a - [Test] Add IRCE tests with non-canonical range check

[Max Kazantsev via llvm-commits]

Author: Max Kazantsev
Date: 2023-04-18T20:10:22+07:00
New Revision: 4f6fc7a30bde9a036a7774d76bcdc15238fe2378

URL: https://github.com/llvm/llvm-project/commit/4f6fc7a30bde9a036a7774d76bcdc15238fe2378
DIFF: https://github.com/llvm/llvm-project/commit/4f6fc7a30bde9a036a7774d76bcdc15238fe2378.diff

LOG: [Test] Add IRCE tests with non-canonical range check

Patch by Aleksandr Popov!

Differential Revision: https://reviews.llvm.org/D148224

Added: 
    

Modified: 
    llvm/test/Transforms/IRCE/stride_more_than_1.ll

Removed: 
    


################################################################################
diff  --git a/llvm/test/Transforms/IRCE/stride_more_than_1.ll b/llvm/test/Transforms/IRCE/stride_more_than_1.ll
index 678cbe6f7cebe..8b795269d0352 100644
--- a/llvm/test/Transforms/IRCE/stride_more_than_1.ll
+++ b/llvm/test/Transforms/IRCE/stride_more_than_1.ll
@@ -701,9 +701,116 @@ out_of_bounds:
   ret i32 -1
 }
 
+; Same as test_09 but range check comparison is inversed.
+; TODO: IRCE is allowed.
+define i32 @test_10(ptr %p, ptr %capacity_p, ptr %num_elements_p) {
+; CHECK-LABEL: define i32 @test_10
+; CHECK-SAME: (ptr [[P:%.*]], ptr [[CAPACITY_P:%.*]], ptr [[NUM_ELEMENTS_P:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CAPACITY:%.*]] = load i32, ptr [[CAPACITY_P]], align 4, !range [[RNG16]]
+; CHECK-NEXT:    [[NUM_ELEMENTS:%.*]] = load i32, ptr [[NUM_ELEMENTS_P]], align 4, !range [[RNG16]]
+; CHECK-NEXT:    [[LIMIT:%.*]] = sub i32 [[CAPACITY]], 3
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[BACKEDGE:%.*]] ]
+; CHECK-NEXT:    [[CAPACITY_CHECK:%.*]] = icmp sge i32 [[IV]], [[LIMIT]]
+; CHECK-NEXT:    br i1 [[CAPACITY_CHECK]], label [[OUT_OF_BOUNDS:%.*]], label [[BACKEDGE]], !prof [[PROF19:![0-9]+]]
+; CHECK:       backedge:
+; CHECK-NEXT:    [[IV_WIDE:%.*]] = zext i32 [[IV]] to i64
+; CHECK-NEXT:    [[EL_PTR:%.*]] = getelementptr i32, ptr [[P]], i64 [[IV_WIDE]]
+; CHECK-NEXT:    store i32 1, ptr [[EL_PTR]], align 4
+; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 4
+; CHECK-NEXT:    [[LOOP_COND:%.*]] = icmp slt i32 [[IV_NEXT]], [[NUM_ELEMENTS]]
+; CHECK-NEXT:    br i1 [[LOOP_COND]], label [[LOOP]], label [[EXIT:%.*]]
+; CHECK:       exit:
+; CHECK-NEXT:    [[IV_LCSSA1:%.*]] = phi i32 [ [[IV]], [[BACKEDGE]] ]
+; CHECK-NEXT:    ret i32 [[IV_LCSSA1]]
+; CHECK:       out_of_bounds:
+; CHECK-NEXT:    ret i32 -1
+;
+entry:
+  %capacity = load i32, ptr %capacity_p, !range !4
+  %num_elements = load i32, ptr %num_elements_p, !range !4
+  %limit = sub i32 %capacity, 3
+  br label %loop
+
+loop:
+  %iv = phi i32 [0, %entry], [%iv.next, %backedge]
+  %capacity_check = icmp sge i32 %iv, %limit
+  br i1 %capacity_check, label %out_of_bounds, label %backedge, !prof !6
+
+backedge:
+  %iv.wide = zext i32 %iv to i64
+  %el.ptr = getelementptr i32, ptr %p, i64 %iv.wide
+  store i32 1, ptr %el.ptr
+  %iv.next = add nuw nsw i32 %iv, 4
+  %loop_cond = icmp slt i32 %iv.next, %num_elements
+  br i1 %loop_cond, label %loop, label %exit
+
+exit:
+  ret i32 %iv
+
+out_of_bounds:
+  ret i32 -1
+}
+
+; Same as test_09 but range check comparison is non-strict:
+; TODO: IRCE is allowed.
+define i32 @test_11(ptr %p, ptr %capacity_p, ptr %num_elements_p) {
+; CHECK-LABEL: define i32 @test_11
+; CHECK-SAME: (ptr [[P:%.*]], ptr [[CAPACITY_P:%.*]], ptr [[NUM_ELEMENTS_P:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CAPACITY:%.*]] = load i32, ptr [[CAPACITY_P]], align 4, !range [[RNG16]]
+; CHECK-NEXT:    [[NUM_ELEMENTS:%.*]] = load i32, ptr [[NUM_ELEMENTS_P]], align 4, !range [[RNG16]]
+; CHECK-NEXT:    [[LIMIT:%.*]] = sub i32 [[CAPACITY]], 4
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[BACKEDGE:%.*]] ]
+; CHECK-NEXT:    [[CAPACITY_CHECK:%.*]] = icmp sle i32 [[IV]], [[LIMIT]]
+; CHECK-NEXT:    br i1 [[CAPACITY_CHECK]], label [[BACKEDGE]], label [[OUT_OF_BOUNDS:%.*]], !prof [[PROF17]]
+; CHECK:       backedge:
+; CHECK-NEXT:    [[IV_WIDE:%.*]] = zext i32 [[IV]] to i64
+; CHECK-NEXT:    [[EL_PTR:%.*]] = getelementptr i32, ptr [[P]], i64 [[IV_WIDE]]
+; CHECK-NEXT:    store i32 1, ptr [[EL_PTR]], align 4
+; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 4
+; CHECK-NEXT:    [[LOOP_COND:%.*]] = icmp slt i32 [[IV_NEXT]], [[NUM_ELEMENTS]]
+; CHECK-NEXT:    br i1 [[LOOP_COND]], label [[LOOP]], label [[EXIT:%.*]]
+; CHECK:       exit:
+; CHECK-NEXT:    [[IV_LCSSA1:%.*]] = phi i32 [ [[IV]], [[BACKEDGE]] ]
+; CHECK-NEXT:    ret i32 [[IV_LCSSA1]]
+; CHECK:       out_of_bounds:
+; CHECK-NEXT:    ret i32 -1
+;
+entry:
+  %capacity = load i32, ptr %capacity_p, !range !4
+  %num_elements = load i32, ptr %num_elements_p, !range !4
+  %limit = sub i32 %capacity, 4
+  br label %loop
+
+loop:
+  %iv = phi i32 [0, %entry], [%iv.next, %backedge]
+  %capacity_check = icmp sle i32 %iv, %limit
+  br i1 %capacity_check, label %backedge, label %out_of_bounds, !prof !5
+
+backedge:
+  %iv.wide = zext i32 %iv to i64
+  %el.ptr = getelementptr i32, ptr %p, i64 %iv.wide
+  store i32 1, ptr %el.ptr
+  %iv.next = add nuw nsw i32 %iv, 4
+  %loop_cond = icmp slt i32 %iv.next, %num_elements
+  br i1 %loop_cond, label %loop, label %exit
+
+exit:
+  ret i32 %iv
+
+out_of_bounds:
+  ret i32 -1
+}
+
 !0 = !{i32 0, i32 50}
 !1 = !{i32 0, i32 2147483640}
 !2 = !{i32 0, i32 2147483641}
 !3 = !{i32 10, i32 50}
 !4 = !{i32 1, i32 2147483648}
 !5 = !{!"branch_weights", i32 1000, i32 1}
+!6 = !{!"branch_weights", i32 1, i32 1000}