[llvm] [ARM][LSR] Exclude uses outside the loop when favoring postinc. (PR #67090)

[David Green via llvm-commits]

https://github.com/davemgreen created https://github.com/llvm/llvm-project/pull/67090

Extra uses for variables outside the loop can mess with the generation of postinc variables. This patch alters the collection of loop invariant fixups in LSR when the target is optimizing for PostInc, to exclude the collection of these extra uses. It is expected that the variable can be rematerialized, which will lead to a more optimal sequence of instructions in the loop.

>From bacd04a20486a27a2f415cb0d1a0e0dbd8dc2629 Mon Sep 17 00:00:00 2001
From: David Green <david.green at arm.com>
Date: Fri, 22 Sep 2023 07:23:22 +0100
Subject: [PATCH] [ARM][LSR] Exclude uses outside the loop when favoring
 postinc.

Extra uses for variables outside the loop can mess with the generation of
postinc variables. This patch alters the collection of loop invariant fixups in
LSR when the target is optimizing for PostInc, to exclude the collection of
these extra uses. It is expected that the variable can be rematerialized, which
will lead to a more optimal sequence of instructions in the loop.
---
 .../Transforms/Scalar/LoopStrengthReduce.cpp  |  5 ++++
 llvm/test/CodeGen/Thumb2/mve-useafterloop.ll  | 23 ++++++++-----------
 2 files changed, 14 insertions(+), 14 deletions(-)

diff --git a/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
index 3cf188ddb70c7a6..5070b8da8e77d35 100644
--- a/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
@@ -3482,6 +3482,11 @@ LSRInstance::CollectLoopInvariantFixupsAndFormulae() {
   SmallVector<const SCEV *, 8> Worklist(RegUses.begin(), RegUses.end());
   SmallPtrSet<const SCEV *, 32> Visited;
 
+  // Don't collect outside uses if we are favoring postinc - the instructions in
+  // the loop are more important than the ones outside of it.
+  if (AMK == TTI::AMK_PostIndexed)
+    return;
+
   while (!Worklist.empty()) {
     const SCEV *S = Worklist.pop_back_val();
 
diff --git a/llvm/test/CodeGen/Thumb2/mve-useafterloop.ll b/llvm/test/CodeGen/Thumb2/mve-useafterloop.ll
index 5f2d356a6d11951..17daa7d54e391aa 100644
--- a/llvm/test/CodeGen/Thumb2/mve-useafterloop.ll
+++ b/llvm/test/CodeGen/Thumb2/mve-useafterloop.ll
@@ -7,19 +7,16 @@ define nonnull ptr @useafterloop(ptr nocapture noundef readonly %pSrcA, ptr noca
 ; CHECK-NEXT:    .save {r7, lr}
 ; CHECK-NEXT:    push {r7, lr}
 ; CHECK-NEXT:    mov.w lr, #64
-; CHECK-NEXT:    mov r12, r2
-; CHECK-NEXT:    movs r3, #0
+; CHECK-NEXT:    mov r3, r2
 ; CHECK-NEXT:  .LBB0_1: @ %while.body
 ; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
 ; CHECK-NEXT:    vldrw.u32 q0, [r0], #16
 ; CHECK-NEXT:    vldrw.u32 q1, [r1], #16
-; CHECK-NEXT:    add.w r2, r12, r3
-; CHECK-NEXT:    adds r3, #16
 ; CHECK-NEXT:    vadd.f32 q0, q1, q0
-; CHECK-NEXT:    vstrw.32 q0, [r2]
+; CHECK-NEXT:    vstrb.8 q0, [r3], #16
 ; CHECK-NEXT:    le lr, .LBB0_1
 ; CHECK-NEXT:  @ %bb.2: @ %while.end
-; CHECK-NEXT:    mov r0, r12
+; CHECK-NEXT:    mov r0, r2
 ; CHECK-NEXT:    pop {r7, pc}
 entry:
   br label %while.body
@@ -92,17 +89,15 @@ define nofpclass(nan inf) float @manyusesafterloop(ptr nocapture noundef readonl
 ; CHECK-NEXT:    .save {r4, lr}
 ; CHECK-NEXT:    push {r4, lr}
 ; CHECK-NEXT:    mov.w lr, #64
-; CHECK-NEXT:    movs r3, #0
+; CHECK-NEXT:    mov r12, r0
+; CHECK-NEXT:    mov r3, r1
+; CHECK-NEXT:    mov r4, r2
 ; CHECK-NEXT:  .LBB2_1: @ %while.body
 ; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    add.w r12, r0, r3
-; CHECK-NEXT:    adds r4, r1, r3
-; CHECK-NEXT:    vldrw.u32 q1, [r4]
-; CHECK-NEXT:    vldrw.u32 q0, [r12]
-; CHECK-NEXT:    adds r4, r2, r3
-; CHECK-NEXT:    adds r3, #16
+; CHECK-NEXT:    vldrw.u32 q0, [r12], #16
+; CHECK-NEXT:    vldrw.u32 q1, [r3], #16
 ; CHECK-NEXT:    vadd.f32 q0, q1, q0
-; CHECK-NEXT:    vstrw.32 q0, [r4]
+; CHECK-NEXT:    vstrb.8 q0, [r4], #16
 ; CHECK-NEXT:    le lr, .LBB2_1
 ; CHECK-NEXT:  @ %bb.2: @ %while.end
 ; CHECK-NEXT:    vldr s0, [r2]