CUDA: add mean operation (ggml-org#14313)

* CUDA: add mean operation

* add back sum_rows_f32_cuda

* Review: early exit if col!=0

Author: am17an

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

@@ -37,6 +37,7 @@
 #include "ggml-cuda/ssm-scan.cuh"
 #include "ggml-cuda/sum.cuh"
 #include "ggml-cuda/sumrows.cuh"
+#include "ggml-cuda/mean.cuh"
 #include "ggml-cuda/tsembd.cuh"
 #include "ggml-cuda/unary.cuh"
 #include "ggml-cuda/upscale.cuh"
@@ -2357,6 +2358,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
         case GGML_OP_SUM_ROWS:
             ggml_cuda_op_sum_rows(ctx, dst);
             break;
+        case GGML_OP_MEAN:
+            ggml_cuda_op_mean(ctx, dst);
+            break;
         case GGML_OP_SSM_CONV:
             ggml_cuda_op_ssm_conv(ctx, dst);
             break;
@@ -3260,6 +3264,7 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
         case GGML_OP_POOL_2D:
         case GGML_OP_SUM:
         case GGML_OP_SUM_ROWS:
+        case GGML_OP_MEAN:
         case GGML_OP_ARGSORT:
         case GGML_OP_ACC:
             return true;

tests/test-backend-ops.cpp

@@ -4652,6 +4652,8 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_perf() {
 
     test_cases.emplace_back(new test_conv_transpose_2d({256, 256, 256, 1}, {3, 3, 16, 256}, 1));
 
+    test_cases.emplace_back(new test_mean(GGML_TYPE_F32, {256, 256, 3, 1}));
+
     return test_cases;
 }