[llvm] 5e4966c - [LV] Add test with an integer induction based on a ptr one.

[Florian Hahn via llvm-commits]

Author: Florian Hahn
Date: 2022-01-14T15:56:47Z
New Revision: 5e4966cbae7ba58562aa323502aaf02023bcd8ae

URL: https://github.com/llvm/llvm-project/commit/5e4966cbae7ba58562aa323502aaf02023bcd8ae
DIFF: https://github.com/llvm/llvm-project/commit/5e4966cbae7ba58562aa323502aaf02023bcd8ae.diff

LOG: [LV] Add test with an integer induction based on a ptr one.

Reduced test case from the reproducer mentioned in
073c27b5e5851f13d99.

Added: 
    llvm/test/Transforms/LoopVectorize/induction-ptrcasts.ll

Modified: 
    

Removed: 
    


################################################################################
diff  --git a/llvm/test/Transforms/LoopVectorize/induction-ptrcasts.ll b/llvm/test/Transforms/LoopVectorize/induction-ptrcasts.ll
new file mode 100644
index 0000000000000..6de60b110b05c
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/induction-ptrcasts.ll
@@ -0,0 +1,71 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -passes=loop-vectorize -force-vector-width=1 -force-vector-interleave=2 -S %s | FileCheck --check-prefix=VF1 %s
+; RUN: opt -passes=loop-vectorize -force-vector-width=2 -force-vector-interleave=1 -S %s | FileCheck --check-prefix=VF2 %s
+
+ at f = external dso_local global i32, align 4
+
+define void @int_iv_based_on_pointer_iv(i8* %A) {
+; VF1-LABEL: @int_iv_based_on_pointer_iv(
+; VF1:       vector.body:
+; VF1-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
+; VF1-NEXT:    [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 4
+; VF1-NEXT:    [[INDUCTION:%.*]] = add i64 [[OFFSET_IDX]], 0
+; VF1-NEXT:    [[INDUCTION3:%.*]] = add i64 [[OFFSET_IDX]], 4
+; VF1-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 0
+; VF1-NEXT:    [[NEXT_GEP:%.*]] = getelementptr i32, i32* null, i64 [[TMP3]]
+; VF1-NEXT:    [[TMP4:%.*]] = add i64 [[INDEX]], 1
+; VF1-NEXT:    [[NEXT_GEP4:%.*]] = getelementptr i32, i32* null, i64 [[TMP4]]
+; VF1-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[NEXT_GEP]], i64 1
+; VF1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[NEXT_GEP4]], i64 1
+; VF1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[A:%.*]], i64 [[INDUCTION]]
+; VF1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[INDUCTION3]]
+; VF1-NEXT:    store i8 0, i8* [[TMP7]], align 1
+; VF1-NEXT:    store i8 0, i8* [[TMP8]], align 1
+; VF1-NEXT:    [[TMP9:%.*]] = ptrtoint i32* [[TMP5]] to i64
+; VF1-NEXT:    [[TMP10:%.*]] = ptrtoint i32* [[TMP6]] to i64
+; VF1-NEXT:    [[TMP11:%.*]] = sub i64 ptrtoint (i32* @f to i64), [[TMP9]]
+; VF1-NEXT:    [[TMP12:%.*]] = sub i64 ptrtoint (i32* @f to i64), [[TMP10]]
+; VF1-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
+; VF1-NEXT:    [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]],
+; VF1-NEXT:    br i1 [[TMP13]], label %middle.block, label %vector.body
+;
+; VF2-LABEL: @int_iv_based_on_pointer_iv(
+; VF2:       vector.body:
+; VF2-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
+; VF2-NEXT:    [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 4
+; VF2-NEXT:    [[TMP3:%.*]] = add i64 [[OFFSET_IDX]], 0
+; VF2-NEXT:    [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 4
+; VF2-NEXT:    [[TMP5:%.*]] = add i64 [[INDEX]], 0
+; VF2-NEXT:    [[NEXT_GEP:%.*]] = getelementptr i32, i32* null, i64 [[TMP5]]
+; VF2-NEXT:    [[TMP6:%.*]] = add i64 [[INDEX]], 1
+; VF2-NEXT:    [[NEXT_GEP3:%.*]] = getelementptr i32, i32* null, i64 [[TMP6]]
+; VF2-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, i32* [[NEXT_GEP]], i64 1
+; VF2-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, i32* [[NEXT_GEP3]], i64 1
+; VF2-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8, i8* [[A:%.*]], i64 [[TMP3]]
+; VF2-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[TMP4]]
+; VF2-NEXT:    store i8 0, i8* [[TMP9]], align 1
+; VF2-NEXT:    store i8 0, i8* [[TMP10]], align 1
+; VF2-NEXT:    [[TMP11:%.*]] = ptrtoint i32* [[TMP7]] to i64
+; VF2-NEXT:    [[TMP12:%.*]] = ptrtoint i32* [[TMP8]] to i64
+; VF2-NEXT:    [[TMP13:%.*]] = sub i64 ptrtoint (i32* @f to i64), [[TMP11]]
+; VF2-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
+; VF2-NEXT:    [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]],
+; VF2-NEXT:    br i1 [[TMP14]], label %middle.block, label %vector.body
+;
+entry:
+  br label %loop
+
+loop:
+  %iv.int = phi i64 [ 0, %entry ], [ %iv.int.next, %loop ]
+  %iv.ptr = phi i32* [ null, %entry ], [ %iv.ptr.next, %loop ]
+  %iv.ptr.next = getelementptr inbounds i32, i32* %iv.ptr, i64 1
+  %gep.A = getelementptr inbounds i8, i8* %A, i64 %iv.int
+  store i8 0, i8* %gep.A
+  %iv.int.next = ptrtoint i32* %iv.ptr.next to i64
+  %sub.ptr.sub = sub i64 ptrtoint (i32* @f to i64), %iv.int.next
+  %cmp = icmp sgt i64 %sub.ptr.sub, 0
+  br i1 %cmp, label %loop, label %exit
+
+exit:
+  ret void
+}