[mlir][LLVM] Add exact flag #115327

lfrenot · 2024-11-07T14:52:52Z

The implementation is mostly based on the one existing for the nsw and nuw flags.

If the exact flag is present, the corresponding operation returns a poison value when the result is not exact. (For a division, if rounding happens; for a right shift, if a non-zero bit is shifted out.)

@zero9178, @gysit and @Dinistro, could you take a look?

github-actions · 2024-11-07T14:53:14Z

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

llvmbot · 2024-11-07T15:05:28Z

@llvm/pr-subscribers-mlir-llvm

@llvm/pr-subscribers-mlir

Author: None (lfrenot)

Changes

The implementation is mostly based on the one existing for the nsw and nuw flags.

If the exact flag is present, the corresponding operation returns a poison value when the result is not exact. (For a division, if rounding happens; for a right shift, if a non-zero bit is shifted out.)

@zero9178, @gysit and @Dinistro, could you take a look?

Full diff: https://github.com/llvm/llvm-project/pull/115327.diff

8 Files Affected:

(modified) mlir/include/mlir/Dialect/LLVMIR/LLVMInterfaces.td (+27)
(modified) mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td (+22-4)
(modified) mlir/include/mlir/Target/LLVMIR/ModuleImport.h (+5)
(modified) mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp (+3)
(modified) mlir/lib/Target/LLVMIR/ModuleImport.cpp (+8)
(modified) mlir/test/Dialect/LLVMIR/roundtrip.mlir (+10)
(added) mlir/test/Target/LLVMIR/Import/exact.ll (+14)
(added) mlir/test/Target/LLVMIR/exact.mlir (+14)

diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMInterfaces.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMInterfaces.td
index 7e38e0b27fd96b..12c430df208925 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMInterfaces.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMInterfaces.td
@@ -87,6 +87,33 @@ def IntegerOverflowFlagsInterface : OpInterface<"IntegerOverflowFlagsInterface">
   ];
 }
 
+def ExactFlagInterface : OpInterface<"ExactFlagInterface"> {
+  let description = [{
+    This interface defines an LLVM operation with an exact flag and
+    provides a uniform API for accessing it.
+  }];
+
+  let cppNamespace = "::mlir::LLVM";
+
+  let methods = [
+    InterfaceMethod<[{
+      Get the exact flag for the operation.
+    }], "bool", "getIsExact", (ins), [{}], [{
+      return $_op.getProperties().isExact;
+    }]>,
+    InterfaceMethod<[{
+      Set the exact flag for the operation.
+    }], "void", "setIsExact", (ins "bool":$isExact), [{}], [{
+      $_op.getProperties().isExact = isExact;
+    }]>,
+    StaticInterfaceMethod<[{
+      Get the attribute name of the isExact property.
+    }], "StringRef", "getIsExactName", (ins), [{}], [{
+      return "isExact";
+    }]>,
+  ];
+}
+
 def BranchWeightOpInterface : OpInterface<"BranchWeightOpInterface"> {
   let description = [{
     An interface for operations that can carry branch weights metadata. It
diff --git a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
index d5def510a904d3..b7ce126dbf54dd 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
@@ -76,6 +76,24 @@ class LLVM_IntArithmeticOpWithOverflowFlag<string mnemonic, string instName,
     "$res = builder.Create" # instName #
     "($lhs, $rhs, /*Name=*/\"\", op.hasNoUnsignedWrap(), op.hasNoSignedWrap());";
 }
+class LLVM_IntArithmeticOpWithIsExact<string mnemonic, string instName,
+                                   list<Trait> traits = []> :
+    LLVM_ArithmeticOpBase<AnySignlessInteger, mnemonic, instName,
+    !listconcat([DeclareOpInterfaceMethods<ExactFlagInterface>], traits)> {
+  let arguments = !con(commonArgs, (ins UnitAttr:$isExact));
+
+  string mlirBuilder = [{
+    auto op = $_builder.create<$_qualCppClassName>($_location, $lhs, $rhs);
+    moduleImport.setExactFlag(inst, op);
+    $res = op;
+  }];
+  let assemblyFormat = [{
+    (`exact` $isExact^)? $lhs `,` $rhs custom<LLVMOpAttrs>(attr-dict) `:` type($res)
+  }];
+  string llvmBuilder =
+    "$res = builder.Create" # instName #
+    "($lhs, $rhs, /*Name=*/\"\", op.getIsExact());";
+}
 class LLVM_FloatArithmeticOp<string mnemonic, string instName,
                              list<Trait> traits = []> :
     LLVM_ArithmeticOpBase<LLVM_AnyFloat, mnemonic, instName,
@@ -116,8 +134,8 @@ def LLVM_AddOp : LLVM_IntArithmeticOpWithOverflowFlag<"add", "Add",
 def LLVM_SubOp : LLVM_IntArithmeticOpWithOverflowFlag<"sub", "Sub", []>;
 def LLVM_MulOp : LLVM_IntArithmeticOpWithOverflowFlag<"mul", "Mul",
     [Commutative]>;
-def LLVM_UDivOp : LLVM_IntArithmeticOp<"udiv", "UDiv">;
-def LLVM_SDivOp : LLVM_IntArithmeticOp<"sdiv", "SDiv">;
+def LLVM_UDivOp : LLVM_IntArithmeticOpWithIsExact<"udiv", "UDiv">;
+def LLVM_SDivOp : LLVM_IntArithmeticOpWithIsExact<"sdiv", "SDiv">;
 def LLVM_URemOp : LLVM_IntArithmeticOp<"urem", "URem">;
 def LLVM_SRemOp : LLVM_IntArithmeticOp<"srem", "SRem">;
 def LLVM_AndOp : LLVM_IntArithmeticOp<"and", "And">;
@@ -128,8 +146,8 @@ def LLVM_XOrOp : LLVM_IntArithmeticOp<"xor", "Xor">;
 def LLVM_ShlOp : LLVM_IntArithmeticOpWithOverflowFlag<"shl", "Shl", []> {
   let hasFolder = 1;
 }
-def LLVM_LShrOp : LLVM_IntArithmeticOp<"lshr", "LShr">;
-def LLVM_AShrOp : LLVM_IntArithmeticOp<"ashr", "AShr">;
+def LLVM_LShrOp : LLVM_IntArithmeticOpWithIsExact<"lshr", "LShr">;
+def LLVM_AShrOp : LLVM_IntArithmeticOpWithIsExact<"ashr", "AShr">;
 
 // Base class for compare operations. A compare operation takes two operands
 // of the same type and returns a boolean result. If the operands are
diff --git a/mlir/include/mlir/Target/LLVMIR/ModuleImport.h b/mlir/include/mlir/Target/LLVMIR/ModuleImport.h
index bbb7af58d27393..6c3a500f20e3a9 100644
--- a/mlir/include/mlir/Target/LLVMIR/ModuleImport.h
+++ b/mlir/include/mlir/Target/LLVMIR/ModuleImport.h
@@ -187,6 +187,11 @@ class ModuleImport {
   /// operation does not implement the integer overflow flag interface.
   void setIntegerOverflowFlags(llvm::Instruction *inst, Operation *op) const;
 
+  /// Sets the exact flag attribute for the imported operation `op` given
+  /// the original instruction `inst`. Asserts if the operation does not
+  /// implement the exact flag interface.
+  void setExactFlag(llvm::Instruction *inst, Operation *op) const;
+
   /// Sets the fastmath flags attribute for the imported operation `op` given
   /// the original instruction `inst`. Asserts if the operation does not
   /// implement the fastmath interface.
diff --git a/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp b/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp
index c9bc9533ca2a6b..6b2d8943bf4885 100644
--- a/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp
+++ b/mlir/lib/Dialect/LLVMIR/IR/LLVMDialect.cpp
@@ -143,6 +143,9 @@ static void printLLVMOpAttrs(OpAsmPrinter &printer, Operation *op,
   if (auto iface = dyn_cast<IntegerOverflowFlagsInterface>(op)) {
     printer.printOptionalAttrDict(
         filteredAttrs, /*elidedAttrs=*/{iface.getOverflowFlagsAttrName()});
+  } else if (auto iface = dyn_cast<ExactFlagInterface>(op)) {
+    printer.printOptionalAttrDict(filteredAttrs,
+                                  /*elidedAttrs=*/{iface.getIsExactName()});
   } else {
     printer.printOptionalAttrDict(filteredAttrs);
   }
diff --git a/mlir/lib/Target/LLVMIR/ModuleImport.cpp b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
index 1f63519373ecab..ccec2034a298b2 100644
--- a/mlir/lib/Target/LLVMIR/ModuleImport.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleImport.cpp
@@ -683,6 +683,14 @@ void ModuleImport::setIntegerOverflowFlags(llvm::Instruction *inst,
   iface.setOverflowFlags(value);
 }
 
+void ModuleImport::setExactFlag(llvm::Instruction *inst, Operation *op) const {
+  auto iface = cast<ExactFlagInterface>(op);
+
+  bool value = inst->isExact();
+
+  iface.setIsExact(value);
+}
+
 void ModuleImport::setFastmathFlagsAttr(llvm::Instruction *inst,
                                         Operation *op) const {
   auto iface = cast<FastmathFlagsInterface>(op);
diff --git a/mlir/test/Dialect/LLVMIR/roundtrip.mlir b/mlir/test/Dialect/LLVMIR/roundtrip.mlir
index b8ce7db795a1d1..9daad2ef5b0b1b 100644
--- a/mlir/test/Dialect/LLVMIR/roundtrip.mlir
+++ b/mlir/test/Dialect/LLVMIR/roundtrip.mlir
@@ -49,6 +49,16 @@ func.func @ops(%arg0: i32, %arg1: f32,
   %mul_flag = llvm.mul %arg0, %arg0 overflow<nsw, nuw> : i32
   %shl_flag = llvm.shl %arg0, %arg0 overflow<nuw, nsw> : i32
 
+// Integer exact
+// CHECK: {{.*}} = llvm.sdiv exact %[[I32]], %[[I32]] : i32
+// CHECK: {{.*}} = llvm.udiv exact %[[I32]], %[[I32]] : i32
+// CHECK: {{.*}} = llvm.ashr exact %[[I32]], %[[I32]] : i32
+// CHECK: {{.*}} = llvm.lshr exact %[[I32]], %[[I32]] : i32
+  %sdiv_flag = llvm.sdiv exact %arg0, %arg0 : i32
+  %udiv_flag = llvm.udiv exact %arg0, %arg0 : i32
+  %ashr_flag = llvm.ashr exact %arg0, %arg0 : i32
+  %lshr_flag = llvm.lshr exact %arg0, %arg0 : i32
+
 // Floating point binary operations.
 //
 // CHECK: {{.*}} = llvm.fadd %[[FLOAT]], %[[FLOAT]] : f32
diff --git a/mlir/test/Target/LLVMIR/Import/exact.ll b/mlir/test/Target/LLVMIR/Import/exact.ll
new file mode 100644
index 00000000000000..528fee5091d2da
--- /dev/null
+++ b/mlir/test/Target/LLVMIR/Import/exact.ll
@@ -0,0 +1,14 @@
+; RUN: mlir-translate -import-llvm -split-input-file %s | FileCheck %s
+
+; CHECK-LABEL: @exactflag_inst
+define void @exactflag_inst(i64 %arg1, i64 %arg2) {
+  ; CHECK: llvm.udiv exact %{{.*}}, %{{.*}} : i64
+  %1 = udiv exact i64 %arg1, %arg2
+  ; CHECK: llvm.sdiv exact %{{.*}}, %{{.*}} : i64
+  %2 = sdiv exact i64 %arg1, %arg2
+  ; CHECK: llvm.lshr exact %{{.*}}, %{{.*}} : i64
+  %3 = lshr exact i64 %arg1, %arg2
+  ; CHECK: llvm.ashr exact %{{.*}}, %{{.*}} : i64
+  %4 = ashr exact i64 %arg1, %arg2
+  ret void
+}
diff --git a/mlir/test/Target/LLVMIR/exact.mlir b/mlir/test/Target/LLVMIR/exact.mlir
new file mode 100644
index 00000000000000..b6c378c2fdcc94
--- /dev/null
+++ b/mlir/test/Target/LLVMIR/exact.mlir
@@ -0,0 +1,14 @@
+// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s
+
+// CHECK-LABEL: define void @exactflag_func
+llvm.func @exactflag_func(%arg0: i64, %arg1: i64) {
+  // CHECK: %{{.*}} = udiv exact i64 %{{.*}}, %{{.*}}
+  %0 = llvm.udiv exact %arg0, %arg1 : i64
+  // CHECK: %{{.*}} = sdiv exact i64 %{{.*}}, %{{.*}}
+  %1 = llvm.sdiv exact %arg0, %arg1 : i64
+  // CHECK: %{{.*}} = lshr exact i64 %{{.*}}, %{{.*}}
+  %2 = llvm.lshr exact %arg0, %arg1 : i64
+  // CHECK: %{{.*}} = ashr exact i64 %{{.*}}, %{{.*}}
+  %3 = llvm.ashr exact %arg0, %arg1 : i64
+  llvm.return
+}

Dinistro

Thanks for this contribution 🙂
Very nice work, LGTM % a small nit comment.

Dinistro · 2024-11-07T16:38:57Z

mlir/lib/Target/LLVMIR/ModuleImport.cpp

+  bool value = inst->isExact();
+
+  iface.setIsExact(value);


Suggested change

bool value = inst->isExact();

iface.setIsExact(value);

iface.setIsExact(inst->isExact());

gysit

Thanks for adding the flag!

LGTM modulo nits.

gysit · 2024-11-07T18:12:15Z

mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td

@@ -76,6 +76,24 @@ class LLVM_IntArithmeticOpWithOverflowFlag<string mnemonic, string instName,
    "$res = builder.Create" # instName #
    "($lhs, $rhs, /*Name=*/\"\", op.hasNoUnsignedWrap(), op.hasNoSignedWrap());";
 }
+class LLVM_IntArithmeticOpWithIsExact<string mnemonic, string instName,


Suggested change

class LLVM_IntArithmeticOpWithIsExact<string mnemonic, string instName,

class LLVM_IntArithmeticOpWithExactFlag<string mnemonic, string instName,

nit: Should we use ExactFlag to follow the naming of LLVM_IntArithmeticOpWithIsExact?

gysit · 2024-11-07T18:13:50Z

mlir/test/Dialect/LLVMIR/roundtrip.mlir

@@ -49,6 +49,16 @@ func.func @ops(%arg0: i32, %arg1: f32,
  %mul_flag = llvm.mul %arg0, %arg0 overflow<nsw, nuw> : i32
  %shl_flag = llvm.shl %arg0, %arg0 overflow<nuw, nsw> : i32

+// Integer exact


Suggested change

// Integer exact

// Integer exact flag.

ultra nit:

This changes the representation of the exact attribute from a mandatory bool to an optional Unit. This is to follow the syntax introduced by llvm/llvm-project#115327

lfrenot · 2024-11-08T12:04:26Z

If everything is fixed, is it possible for someone with write access to merge the PR?

tobiasgrosser · 2024-11-08T12:15:30Z

@Dinistro @gysit @zero9178, thank you for helping with this PR (and potentially merging it when ready).

github-actions · 2024-11-08T12:57:05Z

@lfrenot Congratulations on having your first Pull Request (PR) merged into the LLVM Project!

Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR.

Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues.

How to do this, and the rest of the post-merge process, is covered in detail here.

If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again.

If you don't get any reports, no action is required from you. Your changes are working as expected, well done!

The implementation is mostly based on the one existing for the nsw and nuw flags. If the exact flag is present, the corresponding operation returns a poison value when the result is not exact. (For a division, if rounding happens; for a right shift, if a non-zero bit is shifted out.)

…ct#3408) This changes the representation of the exact attribute from a mandatory bool to an optional Unit. This is to follow the syntax introduced by llvm/llvm-project#115327

[mlir][LLVM] Add exact flag

85ce594

lfrenot force-pushed the mlir-llvm-exact branch from 1028933 to 85ce594 Compare November 7, 2024 15:04

llvmbot added mlir:llvm mlir labels Nov 7, 2024

lfrenot mentioned this pull request Nov 7, 2024

dialects: (llvm) changed the exact attribute to a UnitAttr xdslproject/xdsl#3408

Merged

Dinistro requested review from gysit, Dinistro and zero9178 November 7, 2024 16:35

Dinistro approved these changes Nov 7, 2024

View reviewed changes

gysit approved these changes Nov 7, 2024

View reviewed changes

nit fixes

b8dcb5f

gysit merged commit afa178d into llvm:main Nov 8, 2024
8 checks passed

lfrenot deleted the mlir-llvm-exact branch November 8, 2024 14:40

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[mlir][LLVM] Add exact flag #115327

[mlir][LLVM] Add exact flag #115327

Uh oh!

lfrenot commented Nov 7, 2024

Uh oh!

github-actions bot commented Nov 7, 2024

Uh oh!

llvmbot commented Nov 7, 2024 •

edited

Loading

Uh oh!

Dinistro left a comment

Uh oh!

Dinistro Nov 7, 2024

Uh oh!

lfrenot Nov 8, 2024

Uh oh!

gysit left a comment

Uh oh!

gysit Nov 7, 2024

Uh oh!

lfrenot Nov 8, 2024

Uh oh!

gysit Nov 7, 2024

Uh oh!

lfrenot Nov 8, 2024

Uh oh!

lfrenot commented Nov 8, 2024

Uh oh!

tobiasgrosser commented Nov 8, 2024

Uh oh!

Uh oh!

github-actions bot commented Nov 8, 2024

Uh oh!

Uh oh!

	class LLVM_IntArithmeticOpWithIsExact<string mnemonic, string instName,
	class LLVM_IntArithmeticOpWithExactFlag<string mnemonic, string instName,

[mlir][LLVM] Add exact flag #115327

[mlir][LLVM] Add exact flag #115327

Uh oh!

Conversation

lfrenot commented Nov 7, 2024

Uh oh!

github-actions bot commented Nov 7, 2024

Uh oh!

llvmbot commented Nov 7, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Dinistro left a comment

Choose a reason for hiding this comment

Uh oh!

Dinistro Nov 7, 2024

Choose a reason for hiding this comment

Uh oh!

lfrenot Nov 8, 2024

Choose a reason for hiding this comment

Uh oh!

gysit left a comment

Choose a reason for hiding this comment

Uh oh!

gysit Nov 7, 2024

Choose a reason for hiding this comment

Uh oh!

lfrenot Nov 8, 2024

Choose a reason for hiding this comment

Uh oh!

gysit Nov 7, 2024

Choose a reason for hiding this comment

Uh oh!

lfrenot Nov 8, 2024

Choose a reason for hiding this comment

Uh oh!

lfrenot commented Nov 8, 2024

Uh oh!

tobiasgrosser commented Nov 8, 2024

Uh oh!

Uh oh!

github-actions bot commented Nov 8, 2024

Uh oh!

Uh oh!

llvmbot commented Nov 7, 2024 •

edited

Loading