ggml : introduce GGML_CALL function annotation (#4850)

This change makes it possible to build ggml-cuda.cu and ggml-metal.m as
independent dynamic shared objects, that may be conditionally linked at
runtime in a multiplatform binary. It introduces a GGML_CALL annotation
that documents which functions have a cyclic call relationship, between
the application code and GPU modules.

This change does nothing, unless the build defines -DGGML_MULTIPLATFORM
which causes back-references and function pointers to conform to MS ABI
which is supported by NVCC, ROCm, XCode, GCC and Clang across platforms

ggml.c

@@ -1990,19 +1990,19 @@ void ggml_print_objects(const struct ggml_context * ctx) {
     GGML_PRINT("%s: --- end ---\n", __func__);
 }
 
-int64_t ggml_nelements(const struct ggml_tensor * tensor) {
+GGML_CALL int64_t ggml_nelements(const struct ggml_tensor * tensor) {
     static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
 
     return tensor->ne[0]*tensor->ne[1]*tensor->ne[2]*tensor->ne[3];
 }
 
-int64_t ggml_nrows(const struct ggml_tensor * tensor) {
+GGML_CALL int64_t ggml_nrows(const struct ggml_tensor * tensor) {
     static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
 
     return tensor->ne[1]*tensor->ne[2]*tensor->ne[3];
 }
 
-size_t ggml_nbytes(const struct ggml_tensor * tensor) {
+GGML_CALL size_t ggml_nbytes(const struct ggml_tensor * tensor) {
     size_t nbytes;
     size_t blck_size = ggml_blck_size(tensor->type);
     if (blck_size == 1) {
@@ -2025,15 +2025,15 @@ size_t ggml_nbytes_pad(const struct ggml_tensor * tensor) {
     return GGML_PAD(ggml_nbytes(tensor), GGML_MEM_ALIGN);
 }
 
-int ggml_blck_size(enum ggml_type type) {
+GGML_CALL int ggml_blck_size(enum ggml_type type) {
     return type_traits[type].blck_size;
 }
 
-size_t ggml_type_size(enum ggml_type type) {
+GGML_CALL size_t ggml_type_size(enum ggml_type type) {
     return type_traits[type].type_size;
 }
 
-size_t ggml_row_size(enum ggml_type type, int64_t ne) {
+GGML_CALL size_t ggml_row_size(enum ggml_type type, int64_t ne) {
     assert(ne % ggml_blck_size(type) == 0);
     return ggml_type_size(type)*ne/ggml_blck_size(type);
 }
@@ -2042,15 +2042,15 @@ double ggml_type_sizef(enum ggml_type type) {
     return ((double)(type_traits[type].type_size))/type_traits[type].blck_size;
 }
 
-const char * ggml_type_name(enum ggml_type type) {
+GGML_CALL const char * ggml_type_name(enum ggml_type type) {
     return type_traits[type].type_name;
 }
 
-bool ggml_is_quantized(enum ggml_type type) {
+GGML_CALL bool ggml_is_quantized(enum ggml_type type) {
     return type_traits[type].is_quantized;
 }
 
-const char * ggml_op_name(enum ggml_op op) {
+GGML_CALL const char * ggml_op_name(enum ggml_op op) {
     return GGML_OP_NAME[op];
 }
 
@@ -2062,7 +2062,7 @@ const char * ggml_unary_op_name(enum ggml_unary_op op) {
     return GGML_UNARY_OP_NAME[op];
 }
 
-const char * ggml_op_desc(const struct ggml_tensor * t) {
+GGML_CALL const char * ggml_op_desc(const struct ggml_tensor * t) {
     if (t->op == GGML_OP_UNARY) {
         enum ggml_unary_op uop = ggml_get_unary_op(t);
         return ggml_unary_op_name(uop);
@@ -2072,7 +2072,7 @@ const char * ggml_op_desc(const struct ggml_tensor * t) {
     }
 }
 
-size_t ggml_element_size(const struct ggml_tensor * tensor) {
+GGML_CALL size_t ggml_element_size(const struct ggml_tensor * tensor) {
     return ggml_type_size(tensor->type);
 }
 
@@ -2154,11 +2154,11 @@ size_t ggml_tensor_overhead(void) {
     return GGML_OBJECT_SIZE + GGML_TENSOR_SIZE;
 }
 
-bool ggml_is_transposed(const struct ggml_tensor * tensor) {
+GGML_CALL bool ggml_is_transposed(const struct ggml_tensor * tensor) {
     return tensor->nb[0] > tensor->nb[1];
 }
 
-bool ggml_is_contiguous(const struct ggml_tensor * tensor) {
+GGML_CALL bool ggml_is_contiguous(const struct ggml_tensor * tensor) {
     static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
 
     return
@@ -2177,7 +2177,7 @@ static inline bool ggml_is_contiguous_except_dim_1(const struct ggml_tensor * te
         tensor->nb[3] == tensor->nb[2]*tensor->ne[2];
 }
 
-bool ggml_is_permuted(const struct ggml_tensor * tensor) {
+GGML_CALL bool ggml_is_permuted(const struct ggml_tensor * tensor) {
     static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
 
     return tensor->nb[0] > tensor->nb[1] || tensor->nb[1] > tensor->nb[2] || tensor->nb[2] > tensor->nb[3];
@@ -3079,7 +3079,7 @@ float * ggml_get_data_f32(const struct ggml_tensor * tensor) {
     return (float *)(tensor->data);
 }
 
-enum ggml_unary_op ggml_get_unary_op(const struct ggml_tensor * tensor) {
+GGML_CALL enum ggml_unary_op ggml_get_unary_op(const struct ggml_tensor * tensor) {
     GGML_ASSERT(tensor->op == GGML_OP_UNARY);
     return (enum ggml_unary_op) ggml_get_op_params_i32(tensor, 0);
 }
@@ -11653,7 +11653,7 @@ static void ggml_rope_cache_init(
     }
 }
 
-void ggml_rope_yarn_corr_dims(
+GGML_CALL void ggml_rope_yarn_corr_dims(
     int n_dims, int n_orig_ctx, float freq_base, float beta_fast, float beta_slow, float dims[2]
 ) {
     // start and end correction dims