diff --git a/mlir/include/mlir/Conversion/Passes.td b/mlir/include/mlir/Conversion/Passes.td
index 7e7ee3a2f780..d094ee3b36ab 100644
--- a/mlir/include/mlir/Conversion/Passes.td
+++ b/mlir/include/mlir/Conversion/Passes.td
@@ -172,10 +172,6 @@ def ConvertArithToSPIRV : Pass<"convert-arith-to-spirv"> {
            "bool", /*default=*/"true",
            "Emulate narrower scalar types with 32-bit ones if not supported by "
            "the target">,
-    Option<"enableFastMath", "enable-fast-math",
-           "bool", /*default=*/"false",
-           "Enable fast math mode (assuming no NaN and infinity for floating "
-           "point values) when performing conversion">
   ];
 }
 
diff --git a/mlir/include/mlir/Dialect/SPIRV/Transforms/SPIRVConversion.h b/mlir/include/mlir/Dialect/SPIRV/Transforms/SPIRVConversion.h
index 933d62e35fce..09eecafc0c8a 100644
--- a/mlir/include/mlir/Dialect/SPIRV/Transforms/SPIRVConversion.h
+++ b/mlir/include/mlir/Dialect/SPIRV/Transforms/SPIRVConversion.h
@@ -55,11 +55,6 @@ struct SPIRVConversionOptions {
   /// values will be packed into one 32-bit value to be memory efficient.
   bool emulateLT32BitScalarTypes{true};
 
-  /// Whether to enable fast math mode during conversion. If true, various
-  /// patterns would assume no NaN/infinity numbers as inputs, and thus there
-  /// will be no special guards emitted to check and handle such cases.
-  bool enableFastMathMode{false};
-
   /// Use 64-bit integers when converting index types.
   bool use64bitIndex{false};
 };
diff --git a/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp b/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp
index edf81bd7a8f3..eb338c2da4e8 100644
--- a/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp
+++ b/mlir/lib/Conversion/ArithToSPIRV/ArithToSPIRV.cpp
@@ -995,7 +995,7 @@ public:
     auto *converter = getTypeConverter<SPIRVTypeConverter>();
 
     Value replace;
-    if (converter->getOptions().enableFastMathMode) {
+    if (bitEnumContainsAll(op.getFastmath(), arith::FastMathFlags::nnan)) {
       if (op.getPredicate() == arith::CmpFPredicate::ORD) {
         // Ordered comparsion checks if neither operand is NaN.
         replace = spirv::ConstantOp::getOne(op.getType(), loc, rewriter);
@@ -1122,7 +1122,7 @@ public:
     Value spirvOp =
         rewriter.create<SPIRVOp>(loc, dstType, adaptor.getOperands());
 
-    if (converter->getOptions().enableFastMathMode) {
+    if (bitEnumContainsAll(op.getFastmath(), arith::FastMathFlags::nnan)) {
       rewriter.replaceOp(op, spirvOp);
       return success();
     }
@@ -1177,7 +1177,7 @@ public:
         rewriter.create<SPIRVOp>(loc, dstType, adaptor.getOperands());
 
     if (!shouldInsertNanGuards<SPIRVOp>() ||
-        converter->getOptions().enableFastMathMode) {
+        bitEnumContainsAll(op.getFastmath(), arith::FastMathFlags::nnan)) {
       rewriter.replaceOp(op, spirvOp);
       return success();
     }
@@ -1286,7 +1286,6 @@ struct ConvertArithToSPIRVPass
 
     SPIRVConversionOptions options;
     options.emulateLT32BitScalarTypes = this->emulateLT32BitScalarTypes;
-    options.enableFastMathMode = this->enableFastMath;
     SPIRVTypeConverter typeConverter(targetAttr, options);
 
     // Use UnrealizedConversionCast as the bridge so that we don't need to pull
diff --git a/mlir/test/Conversion/ArithToSPIRV/fast-math.mlir b/mlir/test/Conversion/ArithToSPIRV/fast-math.mlir
index dbf0361c2ab3..9bbe28fb127a 100644
--- a/mlir/test/Conversion/ArithToSPIRV/fast-math.mlir
+++ b/mlir/test/Conversion/ArithToSPIRV/fast-math.mlir
@@ -1,4 +1,4 @@
-// RUN: mlir-opt -split-input-file -convert-arith-to-spirv=enable-fast-math -verify-diagnostics %s | FileCheck %s
+// RUN: mlir-opt -split-input-file -convert-arith-to-spirv -verify-diagnostics %s | FileCheck %s
 
 module attributes {
   spirv.target_env = #spirv.target_env<#spirv.vce<v1.0, [], []>, #spirv.resource_limits<>>
@@ -8,7 +8,7 @@ module attributes {
 // CHECK-SAME: %[[LHS:.+]]: f32, %[[RHS:.+]]: f32
 func.func @cmpf_ordered(%arg0 : f32, %arg1 : f32) -> i1 {
   // CHECK: %[[T:.+]] = spirv.Constant true
-  %0 = arith.cmpf ord, %arg0, %arg1 : f32
+  %0 = arith.cmpf ord, %arg0, %arg1 fastmath<fast> : f32
   // CHECK: return %[[T]]
   return %0: i1
 }
@@ -17,7 +17,7 @@ func.func @cmpf_ordered(%arg0 : f32, %arg1 : f32) -> i1 {
 // CHECK-SAME: %[[LHS:.+]]: vector<4xf32>, %[[RHS:.+]]: vector<4xf32>
 func.func @cmpf_unordered(%arg0 : vector<4xf32>, %arg1 : vector<4xf32>) -> vector<4xi1> {
   // CHECK: %[[F:.+]] = spirv.Constant dense<false>
-  %0 = arith.cmpf uno, %arg0, %arg1 : vector<4xf32>
+  %0 = arith.cmpf uno, %arg0, %arg1 fastmath<nnan> : vector<4xf32>
   // CHECK: return %[[F]]
   return %0: vector<4xi1>
 }
@@ -34,7 +34,7 @@ module attributes {
 // CHECK-SAME: %[[LHS:.+]]: f32, %[[RHS:.+]]: f32
 func.func @minimumf(%arg0 : f32, %arg1 : f32) -> f32 {
   // CHECK: %[[F:.+]] = spirv.GL.FMin %[[LHS]], %[[RHS]]
-  %0 = arith.minimumf %arg0, %arg1 : f32
+  %0 = arith.minimumf %arg0, %arg1 fastmath<fast> : f32
   // CHECK: return %[[F]]
   return %0: f32
 }
@@ -43,7 +43,7 @@ func.func @minimumf(%arg0 : f32, %arg1 : f32) -> f32 {
 // CHECK-SAME: %[[LHS:.+]]: vector<4xf32>, %[[RHS:.+]]: vector<4xf32>
 func.func @maximumf(%arg0 : vector<4xf32>, %arg1 : vector<4xf32>) -> vector<4xf32> {
   // CHECK: %[[F:.+]] = spirv.GL.FMax %[[LHS]], %[[RHS]]
-  %0 = arith.maximumf %arg0, %arg1 : vector<4xf32>
+  %0 = arith.maximumf %arg0, %arg1 fastmath<fast> : vector<4xf32>
   // CHECK: return %[[F]]
   return %0: vector<4xf32>
 }
@@ -52,7 +52,7 @@ func.func @maximumf(%arg0 : vector<4xf32>, %arg1 : vector<4xf32>) -> vector<4xf3
 // CHECK-SAME: %[[LHS:.+]]: f32, %[[RHS:.+]]: f32
 func.func @minnumf(%arg0 : f32, %arg1 : f32) -> f32 {
   // CHECK: %[[F:.+]] = spirv.GL.FMin %[[LHS]], %[[RHS]]
-  %0 = arith.minnumf %arg0, %arg1 : f32
+  %0 = arith.minnumf %arg0, %arg1 fastmath<fast> : f32
   // CHECK: return %[[F]]
   return %0: f32
 }
@@ -61,7 +61,7 @@ func.func @minnumf(%arg0 : f32, %arg1 : f32) -> f32 {
 // CHECK-SAME: %[[LHS:.+]]: vector<4xf32>, %[[RHS:.+]]: vector<4xf32>
 func.func @maxnumf(%arg0 : vector<4xf32>, %arg1 : vector<4xf32>) -> vector<4xf32> {
   // CHECK: %[[F:.+]] = spirv.GL.FMax %[[LHS]], %[[RHS]]
-  %0 = arith.maxnumf %arg0, %arg1 : vector<4xf32>
+  %0 = arith.maxnumf %arg0, %arg1 fastmath<fast> : vector<4xf32>
   // CHECK: return %[[F]]
   return %0: vector<4xf32>
 }