ggml : extend ggml_pool_1d + metal (#16429)

* chore: resolve conflicts

* feat: ggml metal impl

* fix: ggml_metal_kargs_pool_1d struct

* fix: require contiguous input

* chore: test pool_1d

* chore: limit pool1d test cases to p0=0 and s0=k0 to conform with asserts

* chore: add p0 and s0 to testing

* fix: allow padding for cpu and metal

* Update ggml/src/ggml-metal/ggml-metal.metal

* fix: correct single-threaded loop

* ggml : cleanup

* tests : add ne[1] != 1 tests

* fix: ne[1] handling in np

* cont : fixes

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

tests/test-backend-ops.cpp

@@ -4679,6 +4679,37 @@ struct test_pool2d : public test_case {
     }
 };
 
+// GGML_OP_POOL1D
+struct test_pool1d : public test_case {
+    enum ggml_op_pool pool_type;
+    const ggml_type type_input;
+    const std::array<int64_t, 4> ne_input;
+    const int k0;
+    const int s0;
+    const int p0;
+
+    std::string vars() override {
+        return VARS_TO_STR6(pool_type, type_input, ne_input, k0, s0, p0);
+    }
+
+    test_pool1d(ggml_op_pool pool_type = GGML_OP_POOL_AVG,
+                ggml_type type_input = GGML_TYPE_F32,
+                std::array<int64_t,4> ne_input = {10, 1, 1, 1},
+                int k0 = 3, int s0 = 3, int p0 = 0)
+        : pool_type(pool_type), type_input(type_input), ne_input(ne_input), k0(k0), s0(s0), p0(p0) {}
+
+    ggml_tensor * build_graph(ggml_context * ctx) override {
+        ggml_tensor * input = ggml_new_tensor(ctx, type_input, 4, ne_input.data());
+        ggml_set_param(input);
+        ggml_set_name(input, "input");
+
+        ggml_tensor * out = ggml_pool_1d(ctx, input, pool_type, k0, s0, p0);
+        ggml_set_name(out, "out");
+
+        return out;
+    }
+};
+
 // GGML_OP_CONV_TRANSPOSE_1D
 struct test_conv_transpose_1d : public test_case {
     const std::array<int64_t, 4> ne_input;
@@ -7058,6 +7089,20 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
         }
     }
 
+    for (ggml_type type_input : {GGML_TYPE_F32}) {
+        for (ggml_op_pool pool_type : {GGML_OP_POOL_AVG, GGML_OP_POOL_MAX}) {
+            for (int k0 : {1, 3}) {
+                for (int s0 : {1, 2}) {
+                    for (int p0 : {0, 1}) {
+                        test_cases.emplace_back(new test_pool1d(pool_type, type_input, { 10,  3, 2, 1 }, k0, s0, p0));
+                        test_cases.emplace_back(new test_pool1d(pool_type, type_input, { 11,  1, 3, 2 }, k0, s0, p0));
+                        test_cases.emplace_back(new test_pool1d(pool_type, type_input, { 128, 2, 1, 3 }, k0, s0, p0));
+                    }
+                }
+            }
+        }
+    }
+
 #if 0
     // >4GB im2col destination. Too slow to run by default.
     // Test cases taken from Wan2.1 T2V 1.3B.