diff --git a/lib/Conversion/TritonGPUToLLVM/MemoryOpToLLVM.cpp b/lib/Conversion/TritonGPUToLLVM/MemoryOpToLLVM.cpp
index b090670d955c..d1c92bc80345 100644
--- a/lib/Conversion/TritonGPUToLLVM/MemoryOpToLLVM.cpp
+++ b/lib/Conversion/TritonGPUToLLVM/MemoryOpToLLVM.cpp
@@ -22,10 +22,6 @@ void lowerDistributedToShared(
     std::pair<size_t, Type> *const llvmOpCount = nullptr) {
   auto srcTy = cast<RankedTensorType>(src.getType());
   auto dstTy = cast<MemDescType>(dst.getType());
-  auto outOrd = mlir::cast<SharedEncodingAttr>(dstTy.getEncoding()).getOrder();
-  assert(srcTy.getShape().size() <= 2 ||
-         (srcTy.getShape().size() == 3 && outOrd[2] == 0) &&
-             "Unexpected rank of ConvertLayout(blocked->shared)");
   auto elemTy = typeConverter->convertType(srcTy.getElementType());
 
   auto smemBase = smemObj.getBase();
diff --git a/lib/Dialect/TritonGPU/Transforms/Utility.cpp b/lib/Dialect/TritonGPU/Transforms/Utility.cpp
index fa8ec2b926eb..3ae8e8bbcde3 100644
--- a/lib/Dialect/TritonGPU/Transforms/Utility.cpp
+++ b/lib/Dialect/TritonGPU/Transforms/Utility.cpp
@@ -563,7 +563,8 @@ bool canFoldIntoConversion(Operation *op, Attribute targetEncoding) {
   }
   return isa<triton::gpu::ConvertLayoutOp, arith::ConstantOp,
              triton::MakeRangeOp, triton::SplatOp, triton::HistogramOp,
-             triton::gpu::LocalAllocOp, triton::gpu::LocalStoreOp>(op);
+             triton::gpu::LocalAllocOp, triton::gpu::LocalLoadOp,
+             triton::gpu::LocalStoreOp>(op);
 }
 
 scf::ForOp replaceForOpWithNewSignature(
diff --git a/test/TritonGPU/combine.mlir b/test/TritonGPU/combine.mlir
index 2ec11a24f197..b78754a6b878 100644
--- a/test/TritonGPU/combine.mlir
+++ b/test/TritonGPU/combine.mlir
@@ -2685,3 +2685,19 @@ module attributes {"triton_gpu.num-warps" = 1 : i32, "triton_gpu.threads-per-war
     tt.return
   }
 }
+
+// -----
+
+#blocked = #triton_gpu.blocked<{sizePerThread = [1, 1, 1, 1, 4], threadsPerWarp = [2, 1, 16, 1, 1], warpsPerCTA = [1, 1, 2, 2, 1], order = [4, 0, 1, 2, 3]}>
+#blocked2 = #triton_gpu.blocked<{sizePerThread = [1, 1, 1, 1, 4], threadsPerWarp = [1, 1, 32, 1, 1], warpsPerCTA = [1, 1, 1, 1, 4], order = [4, 3, 2, 1, 0]}>
+#blocked1 = #triton_gpu.blocked<{sizePerThread = [1, 1, 1, 1, 4], threadsPerWarp = [2, 1, 16, 1, 1], warpsPerCTA = [1, 2, 2, 1, 1], order = [4, 0, 3, 2, 1]}>
+#shared = #triton_gpu.shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [4, 0, 1, 2, 3], hasLeadingOffset = false}>
+module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 4 : i32, triton_gpu.target = "cuda:100", "triton_gpu.threads-per-warp" = 32 : i32} {
+  // CHECK-NOT: convert_layout
+  tt.func public @lift_convert_to_local_load(%arg0 : !tt.memdesc<2x1x32x4x4xi8, #shared, #triton_gpu.shared_memory, mutable>) -> tensor<2x4x32x1x4xi8, #blocked2> {
+    %1 = triton_gpu.local_load %arg0 : !tt.memdesc<2x1x32x4x4xi8, #shared, #triton_gpu.shared_memory, mutable> -> tensor<2x1x32x4x4xi8, #blocked>
+    %2 = tt.trans %1 {order = array<i32: 0, 3, 2, 1, 4>} : tensor<2x1x32x4x4xi8, #blocked> -> tensor<2x4x32x1x4xi8, #blocked1>
+    %3 = triton_gpu.convert_layout %2 : tensor<2x4x32x1x4xi8, #blocked1> -> tensor<2x4x32x1x4xi8, #blocked2>
+    tt.return %3 : tensor<2x4x32x1x4xi8, #blocked2>
+  }
+}
diff --git a/third_party/nvidia/lib/TritonNVIDIAGPUToLLVM/LoadStoreOpToLLVM.cpp b/third_party/nvidia/lib/TritonNVIDIAGPUToLLVM/LoadStoreOpToLLVM.cpp
index cc52507121b5..91887c68df22 100644
--- a/third_party/nvidia/lib/TritonNVIDIAGPUToLLVM/LoadStoreOpToLLVM.cpp
+++ b/third_party/nvidia/lib/TritonNVIDIAGPUToLLVM/LoadStoreOpToLLVM.cpp
@@ -913,9 +913,6 @@ struct AsyncCopyGlobalToLocalOpConversion
     assert((isa<BlockedEncodingAttr, SliceEncodingAttr>(srcLayout) &&
             "Unexpected srcLayout in AsyncCopyGlobalToLocalOpConversion"));
     auto resSharedLayout = cast<SharedEncodingAttr>(dstTy.getEncoding());
-    auto srcShape = srcTy.getShape();
-    assert((srcShape.size() <= 3) &&
-           "insert_slice_async: Unexpected rank of %src");
 
     Value llDst = adaptor.getResult();
     Value llSrc = adaptor.getSrc();