[llvm] Revert "[SLP]Support revectorization of the previously vectorized scalars" (PR #134604)
Alexey Bataev via llvm-commits
llvm-commits at lists.llvm.org
Mon Apr 7 09:30:42 PDT 2025
alexey-bataev wrote:
> Another crash:
>
> ```
> 0. Program arguments: ./clang-bad --target=aarch64-unknown-linux-gnu -c -o /dev/null -O2 reduced.ll
> 1. Optimizer
> 2. Running pass "function<eager-inv>(float2int,lower-constant-intrinsics,loop(loop-rotate<header-duplication;no-prepare-for-lto>,loop-deletion),loop-distribute,inject-tli-mappings,loop-vectorize<no-interleave-forced-only;no-vectorize-forced-only;>,infer-alignment,loop-load-elim,instcombine<max-iterations=1;no-verify-fixpoint>,simplifycfg<bonus-inst-threshold=1;forward-switch-cond;switch-range-to-icmp;switch-to-lookup;no-keep-loops;hoist-common-insts;no-hoist-loads-stores-with-cond-faulting;sink-common-insts;speculate-blocks;simplify-cond-branch;no-speculate-unpredictables>,slp-vectorizer,vector-combine,instcombine<max-iterations=1;no-verify-fixpoint>,loop-unroll<O2>,transform-warning,sroa<preserve-cfg>,infer-alignment,instcombine<max-iterations=1;no-verify-fixpoint>,loop-mssa(licm<allowspeculation>),alignment-from-assumptions,loop-sink,instsimplify,div-rem-pairs,tailcallelim,simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;hoist-loads-stores-with-cond-faulting;no-sink-common-insts;speculate-blocks;simplify-cond-branch;speculate-unpredictables>)" on module "reduced.ll"
> 3. Running pass "slp-vectorizer" on function "f"
> ```
>
> Test:
>
> ```
> target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32"
> target triple = "aarch64-unknown-linux-gnu"
>
> ; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
> declare i64 @llvm.objectsize.i64.p0(ptr, i1 immarg, i1 immarg, i1 immarg) #0
>
> define void @f(ptr %0, i32 %1, i32 %.240, i8 %.2, i64 %2, i1 %.not59, i1 %.not60, ptr %3, ptr %4, i8 %5) {
> %7 = alloca [3 x i32], align 4
> br label %8
>
> 8: ; preds = %9, %6
> br label %9
>
> 9: ; preds = %47, %34, %23, %8
> %.2402 = phi i32 [ %.240, %8 ], [ 0, %47 ], [ 0, %34 ], [ 0, %23 ]
> %10 = icmp slt i32 %.2402, %1
> br i1 %10, label %11, label %8
>
> 11: ; preds = %20, %9
> %.041 = phi i32 [ %21, %20 ], [ 0, %9 ]
> %.24 = phi i8 [ %.3, %20 ], [ 0, %9 ]
> %12 = icmp slt i32 %.041, 4
> br i1 %12, label %13, label %22
>
> 13: ; preds = %11
> br i1 %.not59, label %20, label %14
>
> 14: ; preds = %13
> %15 = zext i8 %5 to i32
> %16 = zext i8 %.2 to i32
> %17 = trunc i32 %.041 to i8
> store i32 %16, ptr %7, align 4
> %18 = getelementptr i8, ptr %7, i64 4
> store i32 %16, ptr %18, align 4
> %19 = getelementptr i8, ptr %7, i64 8
> store i32 %15, ptr %19, align 4
> br label %20
>
> 20: ; preds = %14, %13
> %.3 = phi i8 [ %17, %14 ], [ %.2, %13 ]
> %21 = add i32 %.041, 1
> br label %11
>
> 22: ; preds = %11
> switch i8 %.24, label %56 [
> i8 0, label %23
> i8 2, label %34
> i8 3, label %44
> ]
>
> 23: ; preds = %25, %22
> %.042 = phi i32 [ %33, %25 ], [ 0, %22 ]
> %24 = icmp slt i32 %.042, 3
> br i1 %24, label %25, label %9
>
> 25: ; preds = %23
> %26 = sext i32 %.042 to i64
> %27 = getelementptr [3 x i32], ptr %7, i64 0, i64 %26
> %28 = call i64 @llvm.objectsize.i64.p0(ptr %27, i1 false, i1 false, i1 false)
> %29 = icmp ne i64 %28, 0
> call void @llvm.assume(i1 %29)
> %30 = load i32, ptr %27, align 4
> %31 = mul i32 %30, 3
> %32 = getelementptr [3 x i32], ptr %3, i64 0, i64 %26
> store i32 %31, ptr %32, align 4
> %33 = add i32 %.042, 1
> br label %23
>
> 34: ; preds = %40, %22
> %.143 = phi i32 [ %43, %40 ], [ 0, %22 ]
> %35 = icmp slt i32 %.143, 3
> br i1 %35, label %36, label %9
>
> 36: ; preds = %34
> %37 = sext i32 %.143 to i64
> %38 = or i64 %2, %37
> %.not = icmp eq i64 %38, 0
> br i1 %.not, label %39, label %40
>
> 39: ; preds = %36
> store volatile i64 0, ptr null, align 8
> unreachable
>
> 40: ; preds = %36
> %41 = getelementptr [3 x i32], ptr %7, i64 0, i64 %37
> %42 = load i32, ptr %41, align 4
> store i32 %42, ptr %0, align 4
> %43 = add i32 %.143, 1
> br label %34
>
> 44: ; preds = %45, %22
> br i1 %.not60, label %45, label %47
>
> 45: ; preds = %44
> %46 = load i32, ptr %7, align 4
> store i32 %46, ptr %0, align 4
> br label %44
>
> 47: ; preds = %49, %44
> %.345 = phi i32 [ %55, %49 ], [ 0, %44 ]
> %48 = icmp slt i32 %.345, 3
> br i1 %48, label %49, label %9
>
> 49: ; preds = %47
> %50 = sext i32 %.345 to i64
> %51 = getelementptr [3 x i32], ptr %7, i64 0, i64 %50
> %52 = load i32, ptr %51, align 4
> %53 = shl i32 %52, 1
> %54 = getelementptr [3 x i32], ptr %4, i64 0, i64 %50
> store i32 %53, ptr %54, align 4
> %55 = add i32 %.345, 1
> br label %47
>
> 56: ; preds = %22
> unreachable
> }
>
> ; Function Attrs: nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write)
> declare void @llvm.assume(i1 noundef) #1
>
> attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
> attributes #1 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }
> ```
Looks like also fixed by f413772b318c2da4dfc488df81cb3c458606301f
https://github.com/llvm/llvm-project/pull/134604
More information about the llvm-commits
mailing list