Skip to content

[CIR] Introduce syntax for scalable vectors #172683

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
banach-space merged 3 commits into from
Dec 18, 2025
Merged

[CIR] Introduce syntax for scalable vectors #172683

banach-space merged 3 commits into from
Dec 18, 2025
+90 −4

Conversation

@banach-space
Copy link
Contributor

@banach-space banach-space commented Dec 17, 2025

As of #172346, CIR supports scalable vectors. This patch updates the
assembly format to introduce syntax for representing scalable
dimensions.

The proposed syntax follows the format used by the builtin vector type:

  # Builtin scalable vector
  cir.vector<[16] x !cir.int<u, 8>>
  # CIR scalable vector
  !cir.vector<[16] x !cir.int<u, 8>>

This contrasts with LLVM IR, where scalable dimensions are modeled using
the vscale keyword:

  ; LLVM scalable vector
  <vscale x 16 x i8>

To support this change, cir::VectorType gains a custom parser and
printer, which are small modifications of the auto-generated ones.

@banach-space
[CIR] Introduce syntax for scalable vectors
69cff0c 
As of llvm#172346, CIR supports scalable vectors. This patch updates the
assembly format to introduce syntax for representing scalable
dimensions.

The proposed syntax follows the format used by the builtin vector type:
```mlir
  # Builtin scalable vector
  cir.vector<[16] x !cir.int<u, 8>>
  # CIR scalable vector
  !cir.vector<[16] x !cir.int<u, 8>>
```

This contrasts with LLVM IR, where scalable dimensions are modeled using
the `vscale` keyword:
```llvm
  ; LLVM scalable vector
  <vscale x 16 x i8>
```

To support this change, `cir::VectorType` gains a custom parser and
printer, which are small modifications of the auto-generated ones.
@llvmbot llvmbot added clang Clang issues not falling into any other category ClangIR Anything related to the ClangIR project labels Dec 17, 2025
@llvmbot
Copy link
Member

llvmbot commented Dec 17, 2025

@llvm/pr-subscribers-clangir

Author: Andrzej Warzyński (banach-space)

Changes

As of #172346, CIR supports scalable vectors. This patch updates the
assembly format to introduce syntax for representing scalable
dimensions.

The proposed syntax follows the format used by the builtin vector type:

  # Builtin scalable vector
  cir.vector&lt;[16] x !cir.int&lt;u, 8&gt;&gt;
  # CIR scalable vector
  !cir.vector&lt;[16] x !cir.int&lt;u, 8&gt;&gt;

This contrasts with LLVM IR, where scalable dimensions are modeled using
the vscale keyword:

  ; LLVM scalable vector
  &lt;vscale x 16 x i8&gt;

To support this change, cir::VectorType gains a custom parser and
printer, which are small modifications of the auto-generated ones.

Full diff: https://github.com/llvm/llvm-project/pull/172683.diff

3 Files Affected:

  • (modified) clang/include/clang/CIR/Dialect/IR/CIRTypes.td (+9-4)
  • (modified) clang/lib/CIR/Dialect/IR/CIRTypes.cpp (+70)
  • (modified) clang/test/CIR/IR/vector.cir (+2)
diff --git a/clang/include/clang/CIR/Dialect/IR/CIRTypes.td b/clang/include/clang/CIR/Dialect/IR/CIRTypes.td
index ce64bef3270ed..fe79e3a086d4e 100644
--- a/clang/include/clang/CIR/Dialect/IR/CIRTypes.td
+++ b/clang/include/clang/CIR/Dialect/IR/CIRTypes.td
@@ -430,6 +430,7 @@ def CIR_VectorType : CIR_Type<"Vector", "vector", [
 
     ```mlir
     vector-type ::= !cir.vector<size x element-type>
+    size ::= (decimal-literal | `[` decimal-literal `]`)
     element-type ::= float-type | integer-type | pointer-type
     ```
 
@@ -442,6 +443,13 @@ def CIR_VectorType : CIR_Type<"Vector", "vector", [
     !cir.vector<4 x !cir.int<u, 8>>
     !cir.vector<2 x !cir.float>
     ```
+
+    Scalable vectors are indicated by enclosing size in square brackets.
+
+    Example:
+    ```mlir
+    !cir.vector<[4] x !cir.int<u, 8>>
+    ```
   }];
 
   let parameters = (ins
@@ -450,10 +458,6 @@ def CIR_VectorType : CIR_Type<"Vector", "vector", [
     OptionalParameter<"bool">:$is_scalable
   );
 
-  let assemblyFormat = [{
-    `<` $size `x` $element_type `>`
-  }];
-
   let builders = [
     TypeBuilderWithInferredContext<(ins
       "mlir::Type":$element_type, "uint64_t":$size, CArg<"bool",
@@ -471,6 +475,7 @@ def CIR_VectorType : CIR_Type<"Vector", "vector", [
 
   let genVerifyDecl = 1;
   let skipDefaultBuilders = 1;
+  let hasCustomAssemblyFormat = 1;
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/CIR/Dialect/IR/CIRTypes.cpp b/clang/lib/CIR/Dialect/IR/CIRTypes.cpp
index c7531022fdfb8..182714ddad9e9 100644
--- a/clang/lib/CIR/Dialect/IR/CIRTypes.cpp
+++ b/clang/lib/CIR/Dialect/IR/CIRTypes.cpp
@@ -828,6 +828,76 @@ mlir::LogicalResult cir::VectorType::verify(
   return success();
 }
 
+mlir::Type cir::VectorType::parse(::mlir::AsmParser &odsParser) {
+
+  llvm::SMLoc odsLoc = odsParser.getCurrentLocation();
+  mlir::Builder odsBuilder(odsParser.getContext());
+  mlir::FailureOr<::mlir::Type> elementType;
+  mlir::FailureOr<uint64_t> size;
+  bool isScalabe = false;
+
+  // Parse literal '<'
+  if (odsParser.parseLess())
+    return {};
+
+  // Parse literal '[', if present, and set the scalability flag accordingly
+  if (odsParser.parseOptionalLSquare().succeeded()) {
+    isScalabe = true;
+  }
+
+  // Parse variable 'size'
+  size = mlir::FieldParser<uint64_t>::parse(odsParser);
+  if (mlir::failed(size)) {
+    odsParser.emitError(odsParser.getCurrentLocation(),
+                        "failed to parse CIR_VectorType parameter 'size' which "
+                        "is to be a `uint64_t`");
+    return {};
+  }
+
+  // Parse literal ']', which is expected when dealing with scalable
+  // dim sizes
+  if (isScalabe && odsParser.parseRSquare().failed()) {
+    odsParser.emitError(odsParser.getCurrentLocation(),
+                        "missing closing `]` for scalable dim size");
+    return {};
+  }
+
+  // Parse literal 'x'
+  if (odsParser.parseKeyword("x"))
+    return {};
+
+  // Parse variable 'elementType'
+  elementType = mlir::FieldParser<::mlir::Type>::parse(odsParser);
+  if (mlir::failed(elementType)) {
+    odsParser.emitError(odsParser.getCurrentLocation(),
+                        "failed to parse CIR_VectorType parameter "
+                        "'elementType' which is to be a `mlir::Type`");
+    return {};
+  }
+
+  // Parse literal '>'
+  if (odsParser.parseGreater())
+    return {};
+  return odsParser.getChecked<VectorType>(odsLoc, odsParser.getContext(),
+                                          mlir::Type((*elementType)),
+                                          uint64_t((*size)), isScalabe);
+}
+
+void cir::VectorType::print(mlir::AsmPrinter &odsPrinter) const {
+  mlir::Builder odsBuilder(getContext());
+  odsPrinter << "<";
+  if (this->getIsScalable())
+    odsPrinter << "[";
+
+  odsPrinter.printStrippedAttrOrType(getSize());
+  if (this->getIsScalable())
+    odsPrinter << "]";
+  odsPrinter << ' ' << "x";
+  odsPrinter << ' ';
+  odsPrinter.printStrippedAttrOrType(getElementType());
+  odsPrinter << ">";
+}
+
 //===----------------------------------------------------------------------===//
 // TargetAddressSpace definitions
 //===----------------------------------------------------------------------===//
diff --git a/clang/test/CIR/IR/vector.cir b/clang/test/CIR/IR/vector.cir
index d274c35099ee5..ac5d0453b1b7e 100644
--- a/clang/test/CIR/IR/vector.cir
+++ b/clang/test/CIR/IR/vector.cir
@@ -23,6 +23,7 @@ cir.func @vec_int_test() {
   %0 = cir.alloca !cir.vector<4 x !s32i>, !cir.ptr<!cir.vector<4 x !s32i>>, ["a"]
   %1 = cir.alloca !cir.vector<3 x !s32i>, !cir.ptr<!cir.vector<3 x !s32i>>, ["b"]
   %2 = cir.alloca !cir.vector<2 x !s32i>, !cir.ptr<!cir.vector<2 x !s32i>>, ["c"]
+  %3 = cir.alloca !cir.vector<[1] x !s32i>, !cir.ptr<!cir.vector<[1] x !s32i>>, ["d"]
   cir.return
 }
 
@@ -30,6 +31,7 @@ cir.func @vec_int_test() {
 // CHECK:  %0 = cir.alloca !cir.vector<4 x !s32i>, !cir.ptr<!cir.vector<4 x !s32i>>, ["a"]
 // CHECK:  %1 = cir.alloca !cir.vector<3 x !s32i>, !cir.ptr<!cir.vector<3 x !s32i>>, ["b"]
 // CHECK:  %2 = cir.alloca !cir.vector<2 x !s32i>, !cir.ptr<!cir.vector<2 x !s32i>>, ["c"]
+// CHECK:  %3 = cir.alloca !cir.vector<[1] x !s32i>, !cir.ptr<!cir.vector<[1] x !s32i>>, ["d"]
 // CHECK:  cir.return
 // CHECK: }

@llvmbot
Copy link
Member

llvmbot commented Dec 17, 2025

@llvm/pr-subscribers-clang

Author: Andrzej Warzyński (banach-space)

Changes

As of #172346, CIR supports scalable vectors. This patch updates the
assembly format to introduce syntax for representing scalable
dimensions.

The proposed syntax follows the format used by the builtin vector type:

  # Builtin scalable vector
  cir.vector&lt;[16] x !cir.int&lt;u, 8&gt;&gt;
  # CIR scalable vector
  !cir.vector&lt;[16] x !cir.int&lt;u, 8&gt;&gt;

This contrasts with LLVM IR, where scalable dimensions are modeled using
the vscale keyword:

  ; LLVM scalable vector
  &lt;vscale x 16 x i8&gt;

To support this change, cir::VectorType gains a custom parser and
printer, which are small modifications of the auto-generated ones.

Full diff: https://github.com/llvm/llvm-project/pull/172683.diff

3 Files Affected:

  • (modified) clang/include/clang/CIR/Dialect/IR/CIRTypes.td (+9-4)
  • (modified) clang/lib/CIR/Dialect/IR/CIRTypes.cpp (+70)
  • (modified) clang/test/CIR/IR/vector.cir (+2)
diff --git a/clang/include/clang/CIR/Dialect/IR/CIRTypes.td b/clang/include/clang/CIR/Dialect/IR/CIRTypes.td
index ce64bef3270ed..fe79e3a086d4e 100644
--- a/clang/include/clang/CIR/Dialect/IR/CIRTypes.td
+++ b/clang/include/clang/CIR/Dialect/IR/CIRTypes.td
@@ -430,6 +430,7 @@ def CIR_VectorType : CIR_Type<"Vector", "vector", [
 
     ```mlir
     vector-type ::= !cir.vector<size x element-type>
+    size ::= (decimal-literal | `[` decimal-literal `]`)
     element-type ::= float-type | integer-type | pointer-type
     ```
 
@@ -442,6 +443,13 @@ def CIR_VectorType : CIR_Type<"Vector", "vector", [
     !cir.vector<4 x !cir.int<u, 8>>
     !cir.vector<2 x !cir.float>
     ```
+
+    Scalable vectors are indicated by enclosing size in square brackets.
+
+    Example:
+    ```mlir
+    !cir.vector<[4] x !cir.int<u, 8>>
+    ```
   }];
 
   let parameters = (ins
@@ -450,10 +458,6 @@ def CIR_VectorType : CIR_Type<"Vector", "vector", [
     OptionalParameter<"bool">:$is_scalable
   );
 
-  let assemblyFormat = [{
-    `<` $size `x` $element_type `>`
-  }];
-
   let builders = [
     TypeBuilderWithInferredContext<(ins
       "mlir::Type":$element_type, "uint64_t":$size, CArg<"bool",
@@ -471,6 +475,7 @@ def CIR_VectorType : CIR_Type<"Vector", "vector", [
 
   let genVerifyDecl = 1;
   let skipDefaultBuilders = 1;
+  let hasCustomAssemblyFormat = 1;
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/CIR/Dialect/IR/CIRTypes.cpp b/clang/lib/CIR/Dialect/IR/CIRTypes.cpp
index c7531022fdfb8..182714ddad9e9 100644
--- a/clang/lib/CIR/Dialect/IR/CIRTypes.cpp
+++ b/clang/lib/CIR/Dialect/IR/CIRTypes.cpp
@@ -828,6 +828,76 @@ mlir::LogicalResult cir::VectorType::verify(
   return success();
 }
 
+mlir::Type cir::VectorType::parse(::mlir::AsmParser &odsParser) {
+
+  llvm::SMLoc odsLoc = odsParser.getCurrentLocation();
+  mlir::Builder odsBuilder(odsParser.getContext());
+  mlir::FailureOr<::mlir::Type> elementType;
+  mlir::FailureOr<uint64_t> size;
+  bool isScalabe = false;
+
+  // Parse literal '<'
+  if (odsParser.parseLess())
+    return {};
+
+  // Parse literal '[', if present, and set the scalability flag accordingly
+  if (odsParser.parseOptionalLSquare().succeeded()) {
+    isScalabe = true;
+  }
+
+  // Parse variable 'size'
+  size = mlir::FieldParser<uint64_t>::parse(odsParser);
+  if (mlir::failed(size)) {
+    odsParser.emitError(odsParser.getCurrentLocation(),
+                        "failed to parse CIR_VectorType parameter 'size' which "
+                        "is to be a `uint64_t`");
+    return {};
+  }
+
+  // Parse literal ']', which is expected when dealing with scalable
+  // dim sizes
+  if (isScalabe && odsParser.parseRSquare().failed()) {
+    odsParser.emitError(odsParser.getCurrentLocation(),
+                        "missing closing `]` for scalable dim size");
+    return {};
+  }
+
+  // Parse literal 'x'
+  if (odsParser.parseKeyword("x"))
+    return {};
+
+  // Parse variable 'elementType'
+  elementType = mlir::FieldParser<::mlir::Type>::parse(odsParser);
+  if (mlir::failed(elementType)) {
+    odsParser.emitError(odsParser.getCurrentLocation(),
+                        "failed to parse CIR_VectorType parameter "
+                        "'elementType' which is to be a `mlir::Type`");
+    return {};
+  }
+
+  // Parse literal '>'
+  if (odsParser.parseGreater())
+    return {};
+  return odsParser.getChecked<VectorType>(odsLoc, odsParser.getContext(),
+                                          mlir::Type((*elementType)),
+                                          uint64_t((*size)), isScalabe);
+}
+
+void cir::VectorType::print(mlir::AsmPrinter &odsPrinter) const {
+  mlir::Builder odsBuilder(getContext());
+  odsPrinter << "<";
+  if (this->getIsScalable())
+    odsPrinter << "[";
+
+  odsPrinter.printStrippedAttrOrType(getSize());
+  if (this->getIsScalable())
+    odsPrinter << "]";
+  odsPrinter << ' ' << "x";
+  odsPrinter << ' ';
+  odsPrinter.printStrippedAttrOrType(getElementType());
+  odsPrinter << ">";
+}
+
 //===----------------------------------------------------------------------===//
 // TargetAddressSpace definitions
 //===----------------------------------------------------------------------===//
diff --git a/clang/test/CIR/IR/vector.cir b/clang/test/CIR/IR/vector.cir
index d274c35099ee5..ac5d0453b1b7e 100644
--- a/clang/test/CIR/IR/vector.cir
+++ b/clang/test/CIR/IR/vector.cir
@@ -23,6 +23,7 @@ cir.func @vec_int_test() {
   %0 = cir.alloca !cir.vector<4 x !s32i>, !cir.ptr<!cir.vector<4 x !s32i>>, ["a"]
   %1 = cir.alloca !cir.vector<3 x !s32i>, !cir.ptr<!cir.vector<3 x !s32i>>, ["b"]
   %2 = cir.alloca !cir.vector<2 x !s32i>, !cir.ptr<!cir.vector<2 x !s32i>>, ["c"]
+  %3 = cir.alloca !cir.vector<[1] x !s32i>, !cir.ptr<!cir.vector<[1] x !s32i>>, ["d"]
   cir.return
 }
 
@@ -30,6 +31,7 @@ cir.func @vec_int_test() {
 // CHECK:  %0 = cir.alloca !cir.vector<4 x !s32i>, !cir.ptr<!cir.vector<4 x !s32i>>, ["a"]
 // CHECK:  %1 = cir.alloca !cir.vector<3 x !s32i>, !cir.ptr<!cir.vector<3 x !s32i>>, ["b"]
 // CHECK:  %2 = cir.alloca !cir.vector<2 x !s32i>, !cir.ptr<!cir.vector<2 x !s32i>>, ["c"]
+// CHECK:  %3 = cir.alloca !cir.vector<[1] x !s32i>, !cir.ptr<!cir.vector<[1] x !s32i>>, ["d"]
 // CHECK:  cir.return
 // CHECK: }

andykaylor
andykaylor approved these changes Dec 17, 2025
View reviewed changes
Copy link
Contributor

@andykaylor andykaylor left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This looks great. I have only the smallest possible nit.

clang/lib/CIR/Dialect/IR/CIRTypes.cpp Outdated
return {};

// Parse literal '[', if present, and set the scalability flag accordingly
if (odsParser.parseOptionalLSquare().succeeded()) {
Copy link
Contributor

@andykaylor andykaylor Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This shouldn't have braces.

AmrDeveloper
AmrDeveloper reviewed Dec 17, 2025
View reviewed changes
AmrDeveloper
AmrDeveloper reviewed Dec 17, 2025
View reviewed changes
clang/lib/CIR/Dialect/IR/CIRTypes.cpp
// Parse variable 'elementType'
elementType = mlir::FieldParser<::mlir::Type>::parse(odsParser);
if (mlir::failed(elementType)) {
odsParser.emitError(odsParser.getCurrentLocation(),
Copy link
Member

@AmrDeveloper AmrDeveloper Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we add an invalid vector cir test to make sure errors are emitted correctly

Copy link
Contributor Author

@banach-space banach-space Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This might be different in CIR, but in MLIR we don't really test the parser and tests in invalid.mlir (and similar) are reserved for verification errors.

I am happy to add a test if you think that that would be helpful, but we’d probably want to add a dedicated file for parser errors - perhaps one already exists? I didn’t find any.

Copy link
Member

@AmrDeveloper AmrDeveloper Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Usually, for the handwritten parser or verifier, we have a invalid-<>.cir file to test it, for example, clang/test/CIR/IR/invalid-vector.cir, I think we can add a small test for scalable vector type syntax. What do you think? @andykaylor

Copy link
Member

@AmrDeveloper AmrDeveloper Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

BTW, Only this error need to be covered "missing closing ] for scalable dim size", the other already covered in this test :D

Copy link
Contributor Author

@banach-space banach-space Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added a test in this commit.

IMO, it's not great :/ (suggestions for improvement are welcome) .Testing !cir.vector<[1 x !s32i> (instead of !cir.vector<[1 x] !s32i>) would be better, but the former is captured by Parser::parseDialectSymbolBody with:

/llvm-project/clang/test/CIR/IR/invalid-vector.cir:17:30: error: unbalanced '[' character in pretty dialect name
  %3 = cir.alloca !cir.vector<[1 x !s32i>, !cir.ptr<!cir.vector<[1] x !s32i>>
                             ^

That error is hit before getting into cir::VectorType::parse.

Btw, I don't want to come across as nit-picking or pushing back, but I see quite a few CIR parser errors that are not tested, e.g.

llvm-project/clang/lib/CIR/Dialect/IR/CIRTypes.cpp

Line 154 in fa511cd

parser.emitError(loc, "invalid self-reference within record");

Perhaps it would be better to skip testing in this case as well? WDYT?

Copy link
Member

@AmrDeveloper AmrDeveloper Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for adding it. I am fine with both options, but I think it will be better to keep this test case 🤔

Copy link
Contributor

@andykaylor andykaylor Dec 17, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Btw, I don't want to come across as nit-picking or pushing back, but I see quite a few CIR parser errors that are not tested, e.g.

I'm sure we've been inconsistent about this. I generally only look for tests for verifier errors and have been satisfied with round-trip tests for printing/parsing. My view is that the tests for the verifier check that we are correctly catching incorrectly constructed IR, which can occur anywhere during IR generation or processing, whereas the printing and parsing are paired so they test each other for correctness.

Copy link
Contributor Author

@banach-space banach-space Dec 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Lets me keep the test for now, we can always remove it later.

AmrDeveloper
AmrDeveloper approved these changes Dec 17, 2025
View reviewed changes
Copy link
Member

@AmrDeveloper AmrDeveloper left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM with nits, Thanks

@banach-space banach-space merged commit 0af137c into llvm:main Dec 18, 2025
10 checks passed
@banach-space banach-space deleted the andrzej/cir/print_scalable branch December 18, 2025 10:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@AmrDeveloper AmrDeveloper AmrDeveloper approved these changes

@andykaylor andykaylor andykaylor approved these changes

@lanza lanza Awaiting requested review from lanza lanza is a code owner

@bcardosolopes bcardosolopes Awaiting requested review from bcardosolopes bcardosolopes is a code owner

@xlauko xlauko Awaiting requested review from xlauko xlauko is a code owner

Assignees

No one assigned

Labels

clang Clang issues not falling into any other category ClangIR Anything related to the ClangIR project

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@banach-space @llvmbot @AmrDeveloper @andykaylor