fix check that was looking into output_tensor instead of the input_te…

…nsor_b
tenstorrent · Nov 22, 2024 · 9eff1ee · 9eff1ee
1 parent 1b5f624
commit 9eff1ee
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/ttnn/cpp/ttnn/operations/matmul/device/matmul_op.cpp b/ttnn/cpp/ttnn/operations/matmul/device/matmul_op.cpp
@@ -1177,7 +1177,7 @@ void Matmul::validate(
                         TT_FATAL(program_config.out_subblock_w == per_core_N || program_config.out_subblock_h == 1, "Error");
                     }
                     if (input_tensor_b.buffer()->buffer_type() == tt_metal::BufferType::L1 && input_tensor_b.memory_config().is_sharded()) {
-                        TT_FATAL(this->output_mem_config.memory_layout == TensorMemoryLayout::WIDTH_SHARDED, "Operand B can only be interleaved or L1 width sharded.");
+                        TT_FATAL(input_tensor_b.memory_config().memory_layout == TensorMemoryLayout::WIDTH_SHARDED, "Operand B can only be interleaved or L1 width sharded.");
                         TT_FATAL(program_config.per_core_N == (input_tensor_b.shard_spec().value().shape[1] / in1_tile_shape[1]), "Shard width must match per core N.");
                         if (optional_bias.has_value()) {
                             TT_FATAL(