Up Fused RMS Norm

Update ggml-cuda.cu
Update ggml-alloc.c

Author: Nexesenex

ggml/src/ggml-alloc.c

@@ -53,6 +53,7 @@ static bool ggml_op_can_inplace(enum ggml_op op) {
         case GGML_OP_UNARY:
         case GGML_OP_ROPE:
         case GGML_OP_RMS_NORM:
+        case GGML_OP_FUSED_RMS_NORM:
         case GGML_OP_SOFT_MAX:
             return true;
 

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

@@ -3141,6 +3141,8 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
             return ggml_is_contiguous(op->src[0]) && op->ne[0] % WARP_SIZE == 0;
             break;
         case GGML_OP_FUSED_RMS_NORM:
+            return ggml_is_contiguous(op->src[0]) && op->ne[0] % WARP_SIZE == 0;
+            break;
         case GGML_OP_NONE:
         case GGML_OP_RESHAPE:
         case GGML_OP_VIEW: