[llvm] [IR] Allow llvm.experimental.memset.pattern to take any sized type as the pattern argument (PR #132026)

[Alex Bradbury via llvm-commits]

https://github.com/asb updated https://github.com/llvm/llvm-project/pull/132026

>From 73e5c12f597d1e19b517b02e5d9fefe657ecb9ed Mon Sep 17 00:00:00 2001
From: Alex Bradbury <asb at igalia.com>
Date: Wed, 19 Mar 2025 13:20:30 +0000
Subject: [PATCH 1/2] [IR] Allow llvm.experimental.memset.pattern to take any
 sized type as the pattern argument

I initially thought starting with a more narrow definition and later
expanding would make more sense. But as pointed out in review for
PR #129220, this restriction is generating additional unnecessary work.

This patch alters the intrinsic to accept patterns of any type. Future
patches will update LoopIdiomRecognize and PreISelIntrinsicLowering to
take advantage of this. The verifier will complain if an unsized type is
used. I've additionally taken the opportunity to remove a comment from
the LangRef about some bit widths potentially not being supported by the
target. I don't think this is any more true than it is for arbitrary
width loads/stores which don't carry a similar warning that I can see.
---
 llvm/docs/LangRef.rst                        |  4 ++--
 llvm/include/llvm/IR/Intrinsics.td           |  2 +-
 llvm/lib/IR/Verifier.cpp                     |  4 ++++
 llvm/test/Verifier/memset-pattern-unsized.ll | 10 ++++++++++
 4 files changed, 17 insertions(+), 3 deletions(-)
 create mode 100644 llvm/test/Verifier/memset-pattern-unsized.ll

diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index bda80d76dfaa5..844160822ccef 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -15640,8 +15640,8 @@ Syntax:
 """""""
 
 This is an overloaded intrinsic. You can use
-``llvm.experimental.memset.pattern`` on any integer bit width and for
-different address spaces. Not all targets support all bit widths however.
+``llvm.experimental.memset.pattern`` on any sized type for different address
+spaces.
 
 ::
 
diff --git a/llvm/include/llvm/IR/Intrinsics.td b/llvm/include/llvm/IR/Intrinsics.td
index 562e0714d879a..d10b07ccd91c2 100644
--- a/llvm/include/llvm/IR/Intrinsics.td
+++ b/llvm/include/llvm/IR/Intrinsics.td
@@ -1025,7 +1025,7 @@ def int_memset_inline
 def int_experimental_memset_pattern
     : Intrinsic<[],
       [llvm_anyptr_ty, // Destination.
-       llvm_anyint_ty, // Pattern value.
+       llvm_any_ty,    // Pattern value.
        llvm_anyint_ty, // Count (number of times to fill value).
        llvm_i1_ty],    // IsVolatile.
       [IntrWriteMem, IntrArgMemOnly, IntrWillReturn, IntrNoFree, IntrNoCallback,
diff --git a/llvm/lib/IR/Verifier.cpp b/llvm/lib/IR/Verifier.cpp
index 524e9647bfd35..5934f7adffb93 100644
--- a/llvm/lib/IR/Verifier.cpp
+++ b/llvm/lib/IR/Verifier.cpp
@@ -5583,7 +5583,11 @@ void Verifier::visitIntrinsicCall(Intrinsic::ID ID, CallBase &Call) {
   case Intrinsic::memmove:
   case Intrinsic::memset:
   case Intrinsic::memset_inline:
+    break;
   case Intrinsic::experimental_memset_pattern: {
+    const auto Memset = cast<MemSetPatternInst>(&Call);
+    Check(Memset->getValue()->getType()->isSized(),
+          "unsized types cannot be used as memset patterns", Call);
     break;
   }
   case Intrinsic::memcpy_element_unordered_atomic:
diff --git a/llvm/test/Verifier/memset-pattern-unsized.ll b/llvm/test/Verifier/memset-pattern-unsized.ll
new file mode 100644
index 0000000000000..71b7dca9a5a19
--- /dev/null
+++ b/llvm/test/Verifier/memset-pattern-unsized.ll
@@ -0,0 +1,10 @@
+; RUN: not opt -passes=verify < %s 2>&1 | FileCheck %s
+
+; CHECK: unsized types cannot be used as memset patterns
+
+%X = type opaque
+define void @bar(ptr %P, %X %value) {
+  call void @llvm.experimental.memset.pattern.p0.s_s.i32.0(ptr %P, %X %value, i32 4, i1 false)
+  ret void
+}
+declare void @llvm.experimental.memset.pattern.p0.s_s.i32.0(ptr nocapture, %X, i32, i1) nounwind

>From e31545d9899ca322018f6e8b241bdcd7ccdd2ca2 Mon Sep 17 00:00:00 2001
From: Alex Bradbury <asb at igalia.com>
Date: Wed, 19 Mar 2025 14:04:10 +0000
Subject: [PATCH 2/2] Missing 'and'

---
 llvm/docs/LangRef.rst | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index 844160822ccef..2ab5c491b3130 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -15640,8 +15640,8 @@ Syntax:
 """""""
 
 This is an overloaded intrinsic. You can use
-``llvm.experimental.memset.pattern`` on any sized type for different address
-spaces.
+``llvm.experimental.memset.pattern`` on any sized type and for different
+address spaces.
 
 ::