diff --git a/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp b/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
index e6dd6f135884e..8002d08fdbd27 100644
--- a/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
+++ b/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
@@ -149,10 +149,13 @@ struct GPUShuffleOpLowering : public ConvertOpToLLVMPattern<gpu::ShuffleOp> {
     Value widthOrZeroIfOutside =
         rewriter.create<LLVM::AndOp>(loc, int32Type, add, negwidth);
     Value dstLane;
-    // TODO: Add support for gpu::ShuffleMode::UP and gpu::ShuffleMode::DOWN.
     // TODO: Use ds_swizzle for XOR when step/offsets are constants for better
     // perf.
     switch (op.getMode()) {
+    case gpu::ShuffleMode::UP:
+      dstLane = rewriter.create<LLVM::SubOp>(loc, int32Type, srcLaneId,
+                                             adaptor.getOffset());
+      break;
     case gpu::ShuffleMode::DOWN:
       dstLane = rewriter.create<LLVM::AddOp>(loc, int32Type, srcLaneId,
                                              adaptor.getOffset());
diff --git a/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir b/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
index 071cae9d5789f..999b383a3b8db 100644
--- a/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
+++ b/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
@@ -660,7 +660,7 @@ gpu.module @test_module {
 
 gpu.module @test_module {
   // CHECK-LABEL: func @gpu_shuffle()
-  func.func @gpu_shuffle() -> (f32, f32, f32) {
+  func.func @gpu_shuffle() -> (f32, f32, f32, f32) {
     // CHECK: %[[#VALUE:]] = llvm.mlir.constant(1.000000e+00 : f32) : f32
     %arg0 = arith.constant 1.0 : f32
     // CHECK: %[[#OFFSET:]] = llvm.mlir.constant(4 : i32) : i32
@@ -693,7 +693,22 @@ gpu.module @test_module {
     // CHECK: %[[#CAST_VALUE:]] = llvm.bitcast %[[#VALUE]] : f32 to i32
     // CHECK: %[[#PERMUTE:]] = rocdl.ds_bpermute %[[#ALIGNED_DST_LANE]], %[[#CAST_VALUE]] : (i32, i32) -> i32
     // CHECK: %[[#CAST_SHFL_VALUE:]] = llvm.bitcast %[[#PERMUTE]] : i32 to f32
-    %shfli, %predi = gpu.shuffle idx %arg0, %arg1, %arg2 : f32
+    %shfli, %predi = gpu.shuffle idx %arg0, %arg1, %arg2 : f32 
+    // *** UP mode shuffle ***
+    // CHECK: %[[#LANE_ID:]] = rocdl.mbcnt.hi
+    // CHECK: %[[#ZERO:]] = llvm.mlir.constant(0 : i32) : i32
+    // CHECK: %[[#NEG_WIDTH:]] = llvm.sub %[[#ZERO]], %[[#WIDTH]] : i32
+    // CHECK: %[[#ADD:]] = llvm.add %[[#LANE_ID]], %[[#WIDTH]] : i32
+    // CHECK: %[[#WARP_OR_ZERO:]] = llvm.and %[[#ADD]], %[[#NEG_WIDTH]] : i32
+    // CHECK: %[[#UP:]] = llvm.sub %[[#LANE_ID]], %{{.*}} : i32
+    // CHECK: %[[#CMP:]] = llvm.icmp "slt" %[[#UP]], %[[#WARP_OR_ZERO]] : i32
+    // CHECK: %[[#DST_LANE:]] = llvm.select %[[#CMP]], %[[#UP]], %{{.*}} : i1, i32
+    // CHECK: %[[#TWO:]] = llvm.mlir.constant(2 : i32) : i32
+    // CHECK: %[[#ALIGNED_DST_LANE:]] = llvm.shl %[[#DST_LANE]], %[[#TWO]] : i32
+    // CHECK: %[[#CAST_VALUE:]] = llvm.bitcast %[[#VALUE]] : f32 to i32
+    // CHECK: %[[#PERMUTE:]] = rocdl.ds_bpermute %[[#ALIGNED_DST_LANE]], %[[#CAST_VALUE]] : (i32, i32) -> i32
+    // CHECK: %[[#CAST_SHFL_VALUE:]] = llvm.bitcast %[[#PERMUTE]] : i32 to f32
+    %shflu, %predu = gpu.shuffle up  %arg0, %arg1, %arg2 : f32
     // CHECK: %[[#LANE_ID:]] = rocdl.mbcnt.hi
     // CHECK: %[[#ZERO:]] = llvm.mlir.constant(0 : i32) : i32
     // CHECK: %[[#NEG_WIDTH:]] = llvm.sub %[[#ZERO]], %[[#WIDTH]] : i32
@@ -708,7 +723,7 @@ gpu.module @test_module {
     // CHECK: %[[#PERMUTE:]] = rocdl.ds_bpermute %[[#ALIGNED_DST_LANE]], %[[#CAST_VALUE]] : (i32, i32) -> i32
     // CHECK: %[[#CAST_SHFL_VALUE:]] = llvm.bitcast %[[#PERMUTE]] : i32 to f32
     %shfld, %predd = gpu.shuffle down %arg0, %arg1, %arg2 : f32
-    func.return %shfl, %shfli, %shfld : f32, f32, f32
+    func.return %shfl, %shfli, %shflu, %shfld : f32, f32, f32, f32
   }
 
   // CHECK-LABEL: func @gpu_shuffle_vec