[llvm] [llvm][loop-rotate] Allow forcing loop-rotation (PR #82828)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-transforms

Author: Paul Kirth (ilovepi)

<details>
<summary>Changes</summary>

Many profitable optimizations cannot be performed at -Oz, due to
unrotated loops. While this is worse for size (minimally), many of the
optimizations significantly reduce code size, such as memcpy
optimizations and other patters found by loop idiom recognition.

This patch adds an experimental, backend-only flag to allow loop header
duplication, regardless of the optimization level. Downstream consumers
can experiment with this flag, and if it is profitable, we can adjust
the compiler's defaults accordingly, and expose any useful frontend
flags to opt into the new behavior.

Bug: 50308


---
Full diff: https://github.com/llvm/llvm-project/pull/82828.diff


2 Files Affected:

- (modified) llvm/lib/Transforms/Scalar/LoopRotation.cpp (+13-1) 
- (modified) llvm/test/Transforms/LoopRotate/oz-disable.ll (+3) 


``````````diff
diff --git a/llvm/lib/Transforms/Scalar/LoopRotation.cpp b/llvm/lib/Transforms/Scalar/LoopRotation.cpp
index 7036759a4eed57..c6943695ec1549 100644
--- a/llvm/lib/Transforms/Scalar/LoopRotation.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopRotation.cpp
@@ -39,8 +39,20 @@ static cl::opt<bool> PrepareForLTOOption(
     cl::desc("Run loop-rotation in the prepare-for-lto stage. This option "
              "should be used for testing only."));
 
+// Experimentally allow loop header duplication. This should allow for better
+// optimization at Oz, since loop-idiom recognition can then recognize things
+// like memcpy. If this ends up being profitable, we should drop this flag and
+// making a code gen option that can be controled independent of the opt level
+// and exposed through clang. See
+// https://github.com/llvm/llvm-project/issues/50308 for details.
+static cl::opt<bool>
+    ForceHeaderDuplication("force-loop-header-duplication", cl::init(false),
+                           cl::Hidden,
+                           cl::desc("Always enable loop header duplication"));
+
 LoopRotatePass::LoopRotatePass(bool EnableHeaderDuplication, bool PrepareForLTO)
-    : EnableHeaderDuplication(EnableHeaderDuplication),
+    : EnableHeaderDuplication(EnableHeaderDuplication ||
+                              ForceHeaderDuplication),
       PrepareForLTO(PrepareForLTO) {}
 
 void LoopRotatePass::printPipeline(
diff --git a/llvm/test/Transforms/LoopRotate/oz-disable.ll b/llvm/test/Transforms/LoopRotate/oz-disable.ll
index 6a7847ac0ff215..b933200ed80586 100644
--- a/llvm/test/Transforms/LoopRotate/oz-disable.ll
+++ b/llvm/test/Transforms/LoopRotate/oz-disable.ll
@@ -4,6 +4,9 @@
 ; RUN: opt < %s -S -passes='default<Os>' -debug -debug-only=loop-rotate 2>&1 | FileCheck %s -check-prefix=OS
 ; RUN: opt < %s -S -passes='default<Oz>' -debug -debug-only=loop-rotate 2>&1 | FileCheck %s -check-prefix=OZ
 
+;; Make sure -force-loop-header-duplication overrides the default behavior at Oz
+; RUN: opt < %s -S -passes='default<Oz>' -force-loop-header-duplication -debug -debug-only=loop-rotate 2>&1 | FileCheck %s -check-prefix=OS
+
 ; Loop should be rotated for -Os but not for -Oz.
 ; OS: rotating Loop at depth 1
 ; OZ-NOT: rotating Loop at depth 1

``````````

</details>


https://github.com/llvm/llvm-project/pull/82828