CUDA: add unused vars to mmvf and mmvq (llama/16807)

diff --git a/src/ggml-cuda/mmvf.cu b/src/ggml-cuda/mmvf.cu
index c2c31cdaf231bc7f1a72426f88af6df4a74ccb07..4e31783436d80a4abc06645ecab14b3b3c1a0aee 100644 (file)
--- a/src/ggml-cuda/mmvf.cu
+++ b/src/ggml-cuda/mmvf.cu
@@ -343,6 +343,10 @@ static __global__ void mul_mat_vec_f(
     }
 
     dst[tid*stride_col_dst + row] = value;
+
+    if constexpr (!has_fusion) {
+        GGML_UNUSED_VARS(use_gate, use_bias, use_gate_bias, glu_op, gate_x, x_bias, gate_bias, sumf_gate);
+    }
 }
 
 template<typename T, typename type_acc, int ncols_dst, int block_size>

diff --git a/src/ggml-cuda/mmvq.cu b/src/ggml-cuda/mmvq.cu
index 7a783e4fcf9b45cbf206e17f9d9dcc274881892e..be04a85cc55154cb10963c6069b4b9eafd6db5c8 100644 (file)
--- a/src/ggml-cuda/mmvq.cu
+++ b/src/ggml-cuda/mmvq.cu
@@ -310,6 +310,10 @@ static __global__ void mul_mat_vec_q(
             dst[j*stride_col_dst + threadIdx.x] = result;
         }
     }
+
+    if constexpr (!has_fusion) {
+        GGML_UNUSED_VARS(use_gate, use_bias, use_gate_bias, active_glu, gate_bias, x_bias, tmp_gate);
+    }
 }
 
 static std::pair<dim3, dim3> calc_launch_params(