[llvm] [InterleavedAccessPass] Avoid optimizing load instructions if it has dead binop users (PR #71339)

[via llvm-commits]

https://github.com/Skwoogey created https://github.com/llvm/llvm-project/pull/71339

If a load instruction qualifies to be optimized by InterleavedAccess Pass, but also has a dead binop instruction this will lead to a crash.

Binop instruction will not be deleted, because normally it would be deleted through its' users, but it has none. Later on deleting a load instruction will fail because it still has uses.

>From 5ecb692a40f1ba3badde630d88e3841b13db74a3 Mon Sep 17 00:00:00 2001
From: Denis Ivanitsa <soka00717 at gmail.com>
Date: Sun, 5 Nov 2023 23:02:55 +0100
Subject: [PATCH] [InterleavedAccessPass] Avoid optimizing load instructions if
 it has dead binop users

If a load instruction qualifies to be optimized by InterleavedAccess Pass,
but also has a dead binop instruction this will lead to a crash.

Binop instruction will not be deleted, because normally it would be
deleted through its' users, but it has none. Later on deleting a load instruction will
fail because it still has uses.
---
 llvm/lib/CodeGen/InterleavedAccessPass.cpp    |  2 +-
 .../AArch64/binopshuffles.ll                  | 21 +++++++++++++++++++
 2 files changed, 22 insertions(+), 1 deletion(-)

diff --git a/llvm/lib/CodeGen/InterleavedAccessPass.cpp b/llvm/lib/CodeGen/InterleavedAccessPass.cpp
index 6b3848531569c0b..672caea3086cd63 100644
--- a/llvm/lib/CodeGen/InterleavedAccessPass.cpp
+++ b/llvm/lib/CodeGen/InterleavedAccessPass.cpp
@@ -251,7 +251,7 @@ bool InterleavedAccess::lowerInterleavedLoad(
       continue;
     }
     if (auto *BI = dyn_cast<BinaryOperator>(User)) {
-      if (all_of(BI->users(), [](auto *U) {
+      if (!BI->users().empty() && all_of(BI->users(), [](auto *U) {
             auto *SVI = dyn_cast<ShuffleVectorInst>(U);
             return SVI && isa<UndefValue>(SVI->getOperand(1));
           })) {
diff --git a/llvm/test/Transforms/InterleavedAccess/AArch64/binopshuffles.ll b/llvm/test/Transforms/InterleavedAccess/AArch64/binopshuffles.ll
index 67910305f56d66c..2e8a7cf42ac506b 100644
--- a/llvm/test/Transforms/InterleavedAccess/AArch64/binopshuffles.ll
+++ b/llvm/test/Transforms/InterleavedAccess/AArch64/binopshuffles.ll
@@ -220,3 +220,24 @@ entry:
   store <8 x i8> %shuffled, ptr %p2
   ret void
 }
+
+define void @skip_optimizing_dead_binop(ptr %p0, ptr %p1) {
+; CHECK-LABEL: @skip_optimizing_dead_binop(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[V0:%.*]] = load <8 x double>, ptr [[P0:%.*]]
+; CHECK-NEXT:    [[SHUFFLED_1:%.*]] = shufflevector <8 x double> [[V0]], <8 x double> undef, <2 x i32> <i32 0, i32 4>
+; CHECK-NEXT:    [[SHUFFLED_2:%.*]] = shufflevector <8 x double> [[V0]], <8 x double> undef, <2 x i32> <i32 1, i32 5>
+; CHECK-NEXT:    [[SHUFFLED_3:%.*]] = shufflevector <8 x double> [[V0]], <8 x double> undef, <2 x i32> <i32 2, i32 6>
+; CHECK-NEXT:    [[SHUFFLED_4:%.*]] = shufflevector <8 x double> [[V0]], <8 x double> undef, <2 x i32> <i32 3, i32 7>
+; CHECK-NEXT:    [[DEAD_BINOP:%.*]] = fadd <8 x double> [[V0]], [[V0]]
+; CHECK-NEXT:    ret void
+;
+entry:
+  %v0 = load <8 x double>, ptr %p0
+  %shuffled_1 = shufflevector <8 x double> %v0, <8 x double> undef, <2 x i32> <i32 0, i32 4>
+  %shuffled_2 = shufflevector <8 x double> %v0, <8 x double> undef, <2 x i32> <i32 1, i32 5>
+  %shuffled_3 = shufflevector <8 x double> %v0, <8 x double> undef, <2 x i32> <i32 2, i32 6>
+  %shuffled_4 = shufflevector <8 x double> %v0, <8 x double> undef, <2 x i32> <i32 3, i32 7>
+  %dead_binop = fadd <8 x double> %v0, %v0
+  ret void
+}