[PATCH] D19676: [MemorySSA] Special case for assume intrinsics.

[Daniel Berlin via llvm-commits]

So, should we not just move this into getModRefInfo?

(It seems like general applicable alias knowledge that requires no
computation)

On Thu, Apr 28, 2016 at 11:49 AM, Geoff Berry <gberry at codeaurora.org> wrote:

> gberry created this revision.
> gberry added reviewers: dberlin, george.burgess.iv.
> gberry added a subscriber: llvm-commits.
> Herald added a subscriber: mcrosier.
>
> Treat assume intrinsics as not reading/writing memory despite
> AA->getModRefInfo() saying they do, as is done in BasicAliasAnalysis.
>
> http://reviews.llvm.org/D19676
>
> Files:
>   lib/Transforms/Utils/MemorySSA.cpp
>   test/Transforms/Util/MemorySSA/assume.ll
>
> Index: test/Transforms/Util/MemorySSA/assume.ll
> ===================================================================
> --- /dev/null
> +++ test/Transforms/Util/MemorySSA/assume.ll
> @@ -0,0 +1,18 @@
> +; RUN: opt -basicaa -print-memoryssa -verify-memoryssa -analyze < %s 2>&1
> | FileCheck %s
> +; RUN: opt -aa-pipeline=basic-aa -passes='print<memoryssa>'
> -verify-memoryssa -disable-output < %s 2>&1 | FileCheck %s
> +;
> +; Ensures that assumes are treated as not reading or writing memory.
> +
> +declare void @llvm.assume(i1)
> +
> +define i32 @foo(i32* %a, i32* %b, i1 %c) {
> +; CHECK: 1 = MemoryDef(liveOnEntry)
> +; CHECK-NEXT: store i32 4
> +  store i32 4, i32* %a, align 4
> +; CHECK: call void @llvm.assume
> +  call void @llvm.assume(i1 %c)
> +; CHECK: MemoryUse(1)
> +; CHECK-NEXT: %1 = load i32
> +  %1 = load i32, i32* %a, align 4
> +  ret i32 %1
> +}
> Index: lib/Transforms/Utils/MemorySSA.cpp
> ===================================================================
> --- lib/Transforms/Utils/MemorySSA.cpp
> +++ lib/Transforms/Utils/MemorySSA.cpp
> @@ -364,6 +364,14 @@
>    bool Def = bool(ModRef & MRI_Mod);
>    bool Use = bool(ModRef & MRI_Ref);
>
> +  // While the assume intrinsic is marked as arbitrarily writing so that
> +  // proper control dependencies will be maintained, it never aliases any
> +  // particular memory location.
> +  if (match(I, PatternMatch::m_Intrinsic<Intrinsic::assume>())) {
> +    Def = false;
> +    Use = false;
> +  }
> +
>    // It's possible for an instruction to not modify memory at all. During
>    // construction, we ignore them.
>    if (IgnoreNonMemory && !Def && !Use)
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160428/f5f4b1fd/attachment.html>