[llvm] [LV] Emit better debug and opt-report messages when vectorization is disallowed in the LoopVectorizer (PR #158513)

[Florian Hahn via llvm-commits]

Tibor =?utf-8?q?Győri?= <tibor.gyori at chem.u-szeged.hu>,GYT
 <tiborgyri at gmail.com>,GYT <tiborgyri at gmail.com>,GYT <tiborgyri at gmail.com>,GYT
 <tiborgyri at gmail.com>,GYT <tiborgyri at gmail.com>,GYT <tiborgyri at gmail.com>,GYT
 <tiborgyri at gmail.com>,GYT <tiborgyri at gmail.com>,GYT <tiborgyri at gmail.com>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/158513 at github.com>


================
@@ -0,0 +1,67 @@
+; TEST 1
+; This test checks that we emit only the correct debug messages and
+; optimization remark when the loop vectorizer is disabled by loop metadata.
+
+; REQUIRES: asserts
+; RUN: opt -passes=loop-vectorize -pass-remarks=loop-vectorize \
+; RUN:     -pass-remarks-missed=loop-vectorize \
+; RUN:     -pass-remarks-analysis=loop-vectorize -debug -disable-output \
+; RUN:     < %s 2>&1 | FileCheck --check-prefix=METADATA %s
+; METADATA-NOT: LV: We can vectorize this loop
+; METADATA-NOT: LV: Not vectorizing: loop hasDisableAllTransformsHint
+; METADATA-NOT: LV: Not vectorizing: disabled for an unknown reason
+; METADATA-NOT: LV: Not vectorizing: VectorizeOnlyWhenForced is set
+; METADATA-NOT: LV: Not vectorizing: Disabled/already vectorized
+; METADATA-NOT: LV: Not vectorizing: Cannot prove legality
+; METADATA: LV: Loop hints: force=disabled
+; METADATA: LV: Not vectorizing: #pragma vectorize disable.
+; METADATA: remark:
+; METADATA-SAME: loop not vectorized: vectorization is explicitly disabled
+; METADATA: LV: Loop hints prevent vectorization
+
+; TEST 2
+; This test checks that we emit only the correct debug messages and
+; optimization remark when the loop is not vectorized due to the 
+; vectorize-forced-only pass option being set.
+
+; Strip metadata for FORCEDONLY run, keep it for METADATA run
+; RUN: sed 's/,[[:space:]]*!llvm\.loop[[:space:]]*!0//' %s | \
+; RUN: opt -passes='loop-vectorize<vectorize-forced-only>' \
+; RUN:   -pass-remarks=loop-vectorize \
+; RUN:   -pass-remarks-missed=loop-vectorize \
+; RUN:   -pass-remarks-analysis=loop-vectorize -debug -disable-output \
+; RUN:   2>&1 | FileCheck --check-prefix=FORCEDONLY %s
+; FORCEDONLY-NOT: LV: We can vectorize this loop
+; FORCEDONLY-NOT: LV: Not vectorizing: loop hasDisableAllTransformsHint
+; FORCEDONLY-NOT: LV: Not vectorizing: disabled for an unknown reason
+; FORCEDONLY-NOT: LV: Not vectorizing: #pragma vectorize disable
+; FORCEDONLY-NOT: LV: Not vectorizing: Disabled/already vectorized
+; FORCEDONLY-NOT: LV: Not vectorizing: Cannot prove legality
+; FORCEDONLY: LV: Loop hints: force=?
+; FORCEDONLY: LV: Not vectorizing: VectorizeOnlyWhenForced is set, and no #pragma vectorize enable
+; FORCEDONLY: remark:
+; FORCEDONLY-SAME: loop not vectorized: only vectorizing loops that explicitly request it
+; FORCEDONLY: LV: Loop hints prevent vectorization
+
+define double @disabled_loop_vectorization(ptr %distmat1){
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i64 [ 0, %entry ], [ %inc, %loop ]
+  %RMSD = phi double [ 0.000000e+00, %entry ], [ %add, %loop ]
----------------
fhahn wrote:

```suggestion
  %rdx = phi double [ 0.000000e+00, %entry ], [ %add, %loop ]
```

is the floating point reduction needed or could the loop just store constants or something simple like that?

https://github.com/llvm/llvm-project/pull/158513