Add DIAG for CUDA (#17873)

* Add DIAG for CUDA

* Refactor parameters

ggml/src/ggml-cuda/ggml-cuda.cu

@@ -20,6 +20,7 @@
 #include "ggml-cuda/cpy.cuh"
 #include "ggml-cuda/cross-entropy-loss.cuh"
 #include "ggml-cuda/diagmask.cuh"
+#include "ggml-cuda/diag.cuh"
 #include "ggml-cuda/fattn.cuh"
 #include "ggml-cuda/getrows.cuh"
 #include "ggml-cuda/im2col.cuh"
@@ -2641,6 +2642,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
         case GGML_OP_PERMUTE:
         case GGML_OP_TRANSPOSE:
                 break;
+        case GGML_OP_DIAG:
+            ggml_cuda_op_diag(ctx, dst);
+            break;
         case GGML_OP_DIAG_MASK_INF:
             ggml_cuda_op_diag_mask_inf(ctx, dst);
             break;
@@ -4624,6 +4628,7 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
         case GGML_OP_FILL:
         case GGML_OP_CUMSUM:
         case GGML_OP_TRI:
+        case GGML_OP_DIAG:
             return true;
         case GGML_OP_SOLVE_TRI:
             return op->src[0]->ne[0] <= 64 && op->src[1]->ne[0] <= 32;