[llvm] [LV] Ignore some costs when loop gets fully unrolled (PR #106699)

[Igor Kirillov via llvm-commits]

================
@@ -0,0 +1,38 @@
+; REQUIRES: asserts
+; RUN: opt < %s -mcpu=neoverse-v2 -passes=loop-vectorize -debug-only=loop-vectorize -disable-output -S 2>&1 | FileCheck %s
+
+target triple="aarch64--linux-gnu"
+
+define i64 @test(ptr %a, ptr %b) #0 {
+; CHECK: LV: Checking a loop in 'test'
+; CHECK: LV: Found an estimated cost of 1 for VF 8 For instruction:   %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+; CHECK: LV: Found an estimated cost of 1 for VF 8 For instruction:   %exitcond.not = icmp eq i64 %indvars.iv.next, 16
+; CHECK: LV: Vector loop of width 8 costs: 3.
+; CHECK-NOT: LV: Found an estimated cost of 1 for VF 16 For instruction:   %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+; CHECK-NOT: LV: Found an estimated cost of 1 for VF 16 For instruction:   %exitcond.not = icmp eq i64 %indvars.iv.next, 16
+; CHECK: LV: Vector loop of width 16 costs: 3.
+; CHECK: LV: Selecting VF: 16
+entry:
+  br label %for.body
+
+for.cond.cleanup:                                 ; preds = %for.body
----------------
igogo-x86 wrote:

I'll pre-commit them once I'll be more sure that the patch will go through :)

https://github.com/llvm/llvm-project/pull/106699