[ggml] fixes for cross-platform building successful

Author: GermanAizek

ggml/include/ggml-backend.h

@@ -46,11 +46,12 @@ extern "C" {
     // Backend buffer
     //
 
-    enum GGML_PACKED ggml_backend_buffer_usage {
+    GGML_PACKED_ENUM ggml_backend_buffer_usage {
         GGML_BACKEND_BUFFER_USAGE_ANY = 0,
         GGML_BACKEND_BUFFER_USAGE_WEIGHTS = 1,
         GGML_BACKEND_BUFFER_USAGE_COMPUTE = 2,
     };
+    GGML_PACKED_ENUM_END
 
     GGML_API const char *                   ggml_backend_buffer_name          (ggml_backend_buffer_t buffer);
     GGML_API void                           ggml_backend_buffer_free          (ggml_backend_buffer_t buffer);
@@ -127,14 +128,15 @@ extern "C" {
     // Backend device
     //
 
-    enum GGML_PACKED ggml_backend_dev_type {
+    GGML_PACKED_ENUM ggml_backend_dev_type {
         // CPU device using system memory
         GGML_BACKEND_DEVICE_TYPE_CPU,
         // GPU device using dedicated memory
         GGML_BACKEND_DEVICE_TYPE_GPU,
         // accelerator devices intended to be used together with the CPU backend (e.g. BLAS or AMX)
         GGML_BACKEND_DEVICE_TYPE_ACCEL
     };
+    GGML_PACKED_ENUM_END
 
     // functionality supported by the device
     struct ggml_backend_dev_caps {

ggml/include/ggml-cpu.h

@@ -22,14 +22,15 @@ extern "C" {
     };
 
     // numa strategies
-    enum GGML_PACKED ggml_numa_strategy {
+    GGML_PACKED_ENUM ggml_numa_strategy {
         GGML_NUMA_STRATEGY_DISABLED   = 0,
         GGML_NUMA_STRATEGY_DISTRIBUTE = 1,
         GGML_NUMA_STRATEGY_ISOLATE    = 2,
         GGML_NUMA_STRATEGY_NUMACTL    = 3,
         GGML_NUMA_STRATEGY_MIRROR     = 4,
         GGML_NUMA_STRATEGY_COUNT
     };
+    GGML_PACKED_ENUM_END
 
     GGML_BACKEND_API void    ggml_numa_init(enum ggml_numa_strategy numa); // call once for better performance on NUMA systems
     GGML_BACKEND_API bool    ggml_is_numa(void); // true if init detected that system has >1 NUMA node

ggml/include/ggml.h

@@ -190,11 +190,16 @@
 // TODO: support for clang
 #ifdef __GNUC__
 #    define GGML_DEPRECATED(func, hint) func __attribute__((deprecated(hint)))
-#    define GGML_PACKED __attribute__((__packed__))
+#    define GGML_PACKED_ENUM enum __attribute__((__packed__))
+#    define GGML_PACKED_ENUM_END
 #elif defined(_MSC_VER)
 #    define GGML_DEPRECATED(func, hint) __declspec(deprecated(hint)) func
+#    define GGML_PACKED_ENUM __pragma(pack(push, 1)) enum
+#    define GGML_PACKED_ENUM_END __pragma(pack(pop))
 #else
 #    define GGML_DEPRECATED(func, hint) func
+#    define GGML_PACKED_ENUM
+#    define GGML_PACKED_ENUM_END
 #endif
 
 #ifndef __GNUC__
@@ -318,12 +323,13 @@ extern "C" {
     GGML_NORETURN GGML_ATTRIBUTE_FORMAT(3, 4)
     GGML_API void ggml_abort(const char * file, int line, const char * fmt, ...);
 
-    enum GGML_PACKED ggml_status {
+    GGML_PACKED_ENUM ggml_status {
         GGML_STATUS_ALLOC_FAILED = -2,
         GGML_STATUS_FAILED = -1,
         GGML_STATUS_SUCCESS = 0,
         GGML_STATUS_ABORTED = 1,
     };
+    GGML_PACKED_ENUM_END
 
     // get ggml_status name string
     GGML_API const char * ggml_status_to_string(enum ggml_status status);
@@ -349,7 +355,7 @@ extern "C" {
     struct ggml_cgraph;
 
     // NOTE: always add types at the end of the enum to keep backward compatibility
-    enum GGML_PACKED ggml_type {
+    GGML_PACKED_ENUM ggml_type {
         GGML_TYPE_F32     = 0,
         GGML_TYPE_F16     = 1,
         GGML_TYPE_Q4_0    = 2,
@@ -391,15 +397,17 @@ extern "C" {
         // GGML_TYPE_IQ4_NL_8_8 = 38,
         GGML_TYPE_COUNT   = 39,
     };
+    GGML_PACKED_ENUM_END
 
     // precision
-    enum GGML_PACKED ggml_prec {
+    GGML_PACKED_ENUM ggml_prec {
         GGML_PREC_DEFAULT,
         GGML_PREC_F32,
     };
+    GGML_PACKED_ENUM_END
 
     // model file types
-    enum GGML_PACKED ggml_ftype {
+    GGML_PACKED_ENUM ggml_ftype {
         GGML_FTYPE_UNKNOWN        = -1,
         GGML_FTYPE_ALL_F32        = 0,
         GGML_FTYPE_MOSTLY_F16     = 1,  // except 1d tensors
@@ -425,9 +433,10 @@ extern "C" {
         GGML_FTYPE_MOSTLY_IQ1_M   = 23, // except 1d tensors
         GGML_FTYPE_MOSTLY_BF16    = 24, // except 1d tensors
     };
+    GGML_PACKED_ENUM_END
 
     // available tensor operations:
-    enum GGML_PACKED ggml_op {
+    GGML_PACKED_ENUM ggml_op {
         GGML_OP_NONE = 0,
 
         GGML_OP_DUP,
@@ -520,8 +529,9 @@ extern "C" {
 
         GGML_OP_COUNT,
     };
+    GGML_PACKED_ENUM_END
 
-    enum GGML_PACKED ggml_unary_op {
+    GGML_PACKED_ENUM ggml_unary_op {
         GGML_UNARY_OP_ABS,
         GGML_UNARY_OP_SGN,
         GGML_UNARY_OP_NEG,
@@ -539,29 +549,33 @@ extern "C" {
 
         GGML_UNARY_OP_COUNT,
     };
+    GGML_PACKED_ENUM_END
 
-    enum GGML_PACKED ggml_object_type {
+    GGML_PACKED_ENUM ggml_object_type {
         GGML_OBJECT_TYPE_TENSOR,
         GGML_OBJECT_TYPE_GRAPH,
         GGML_OBJECT_TYPE_WORK_BUFFER
     };
+    GGML_PACKED_ENUM_END
 
-    enum GGML_PACKED ggml_log_level {
+    GGML_PACKED_ENUM ggml_log_level {
         GGML_LOG_LEVEL_NONE  = 0,
         GGML_LOG_LEVEL_DEBUG = 1,
         GGML_LOG_LEVEL_INFO  = 2,
         GGML_LOG_LEVEL_WARN  = 3,
         GGML_LOG_LEVEL_ERROR = 4,
         GGML_LOG_LEVEL_CONT  = 5, // continue previous log
     };
+    GGML_PACKED_ENUM_END
 
     // this tensor...
-    enum GGML_PACKED ggml_tensor_flag {
+    GGML_PACKED_ENUM ggml_tensor_flag {
         GGML_TENSOR_FLAG_INPUT  =  1, // ...is an input for the GGML compute graph
         GGML_TENSOR_FLAG_OUTPUT =  2, // ...is an output for the GGML compute graph
         GGML_TENSOR_FLAG_PARAM  =  4, // ...contains trainable parameters
         GGML_TENSOR_FLAG_LOSS   =  8, // ...defines loss for numerical optimization (multiple loss tensors add up)
     };
+    GGML_PACKED_ENUM_END
 
     struct ggml_init_params {
         // memory pool
@@ -1679,11 +1693,12 @@ extern "C" {
             struct ggml_tensor  * b,
             int                   stride);
 
-    enum GGML_PACKED ggml_op_pool {
+    GGML_PACKED_ENUM ggml_op_pool {
         GGML_OP_POOL_MAX,
         GGML_OP_POOL_AVG,
         GGML_OP_POOL_COUNT,
     };
+    GGML_PACKED_ENUM_END
 
     GGML_API struct ggml_tensor * ggml_pool_1d(
             struct ggml_context * ctx,
@@ -1718,10 +1733,11 @@ extern "C" {
             float                 p0,
             float                 p1);
 
-    enum GGML_PACKED ggml_scale_mode {
+    GGML_PACKED_ENUM ggml_scale_mode {
         GGML_SCALE_MODE_NEAREST  = 0,
         GGML_SCALE_MODE_BILINEAR = 1,
     };
+    GGML_PACKED_ENUM_END
 
     // interpolate
     // multiplies ne0 and ne1 by scale factor
@@ -1768,10 +1784,11 @@ extern "C" {
             int                   max_period);
 
     // sort rows
-    enum GGML_PACKED ggml_sort_order {
+    GGML_PACKED_ENUM ggml_sort_order {
         GGML_SORT_ORDER_ASC,
         GGML_SORT_ORDER_DESC,
     };
+    GGML_PACKED_ENUM_END
 
     GGML_API struct ggml_tensor * ggml_argsort(
             struct ggml_context * ctx,
@@ -2138,12 +2155,13 @@ extern "C" {
     // the goal should be to create an API that other backends can use move everything to the ggml base
 
     // scheduling priorities
-    enum GGML_PACKED ggml_sched_priority {
+    GGML_PACKED_ENUM ggml_sched_priority {
         GGML_SCHED_PRIO_NORMAL,
         GGML_SCHED_PRIO_MEDIUM,
         GGML_SCHED_PRIO_HIGH,
         GGML_SCHED_PRIO_REALTIME
     };
+    GGML_PACKED_ENUM_END
 
     // threadpool params
     // Use ggml_threadpool_params_default() or ggml_threadpool_params_init() to populate the defaults

ggml/include/gguf.h

@@ -50,7 +50,7 @@ extern "C" {
 #endif
 
     // types that can be stored as GGUF KV data
-    enum GGML_PACKED gguf_type {
+    GGML_PACKED_ENUM gguf_type {
         GGUF_TYPE_UINT8   = 0,
         GGUF_TYPE_INT8    = 1,
         GGUF_TYPE_UINT16  = 2,
@@ -66,6 +66,7 @@ extern "C" {
         GGUF_TYPE_FLOAT64 = 12,
         GGUF_TYPE_COUNT,       // marks the end of the enum
     };
+    GGML_PACKED_ENUM_END
 
     struct gguf_context;
 

ggml/src/ggml-impl.h

@@ -284,11 +284,12 @@ static size_t ggml_hash_find_or_insert(struct ggml_hash_set * hash_set, struct g
 
 // computation graph
 
-enum GGML_PACKED ggml_cgraph_eval_order {
+GGML_PACKED_ENUM ggml_cgraph_eval_order {
     GGML_CGRAPH_EVAL_ORDER_LEFT_TO_RIGHT = 0,
     GGML_CGRAPH_EVAL_ORDER_RIGHT_TO_LEFT,
     GGML_CGRAPH_EVAL_ORDER_COUNT
 };
+GGML_PACKED_ENUM_END
 
 struct ggml_cgraph {
     int size;    // maximum number of nodes/leafs/grads/grad_accs

include/llama.h

@@ -66,17 +66,18 @@ extern "C" {
     typedef int32_t llama_token;
     typedef int32_t llama_seq_id;
 
-    enum GGML_PACKED llama_vocab_type {
+    GGML_PACKED_ENUM llama_vocab_type {
         LLAMA_VOCAB_TYPE_NONE = 0, // For models without vocab
         LLAMA_VOCAB_TYPE_SPM  = 1, // LLaMA tokenizer based on byte-level BPE with byte fallback
         LLAMA_VOCAB_TYPE_BPE  = 2, // GPT-2 tokenizer based on byte-level BPE
         LLAMA_VOCAB_TYPE_WPM  = 3, // BERT tokenizer based on WordPiece
         LLAMA_VOCAB_TYPE_UGM  = 4, // T5 tokenizer based on Unigram
         LLAMA_VOCAB_TYPE_RWKV = 5, // RWKV tokenizer based on greedy tokenization
     };
+    GGML_PACKED_ENUM_END
 
     // pre-tokenization types
-    enum GGML_PACKED llama_vocab_pre_type {
+    GGML_PACKED_ENUM llama_vocab_pre_type {
         LLAMA_VOCAB_PRE_TYPE_DEFAULT        = 0,
         LLAMA_VOCAB_PRE_TYPE_LLAMA3         = 1,
         LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_LLM   = 2,
@@ -112,16 +113,18 @@ extern "C" {
         LLAMA_VOCAB_PRE_TYPE_BAILINGMOE     = 32,
         LLAMA_VOCAB_PRE_TYPE_LLAMA4         = 33,
     };
+    GGML_PACKED_ENUM_END
 
-    enum GGML_PACKED llama_rope_type {
+    GGML_PACKED_ENUM llama_rope_type {
         LLAMA_ROPE_TYPE_NONE   = -1,
         LLAMA_ROPE_TYPE_NORM   = 0,
         LLAMA_ROPE_TYPE_NEOX   = GGML_ROPE_TYPE_NEOX,
         LLAMA_ROPE_TYPE_MROPE  = GGML_ROPE_TYPE_MROPE,
         LLAMA_ROPE_TYPE_VISION = GGML_ROPE_TYPE_VISION,
     };
+    GGML_PACKED_ENUM_END
 
-    enum GGML_PACKED llama_token_type { //TODO: remove, required until per token attributes are available from GGUF file
+    GGML_PACKED_ENUM llama_token_type { //TODO: remove, required until per token attributes are available from GGUF file
         LLAMA_TOKEN_TYPE_UNDEFINED    = 0,
         LLAMA_TOKEN_TYPE_NORMAL       = 1,
         LLAMA_TOKEN_TYPE_UNKNOWN      = 2,
@@ -130,8 +133,9 @@ extern "C" {
         LLAMA_TOKEN_TYPE_UNUSED       = 5,
         LLAMA_TOKEN_TYPE_BYTE         = 6,
     };
+    GGML_PACKED_ENUM_END
 
-    enum GGML_PACKED llama_token_attr {
+    GGML_PACKED_ENUM llama_token_attr {
         LLAMA_TOKEN_ATTR_UNDEFINED    = 0,
         LLAMA_TOKEN_ATTR_UNKNOWN      = 1 << 0,
         LLAMA_TOKEN_ATTR_UNUSED       = 1 << 1,
@@ -144,9 +148,10 @@ extern "C" {
         LLAMA_TOKEN_ATTR_RSTRIP       = 1 << 8,
         LLAMA_TOKEN_ATTR_SINGLE_WORD  = 1 << 9,
     };
+    GGML_PACKED_ENUM_END
 
     // model file types
-    enum GGML_PACKED llama_ftype {
+    GGML_PACKED_ENUM llama_ftype {
         LLAMA_FTYPE_ALL_F32              = 0,
         LLAMA_FTYPE_MOSTLY_F16           = 1,  // except 1d tensors
         LLAMA_FTYPE_MOSTLY_Q4_0          = 2,  // except 1d tensors
@@ -188,36 +193,41 @@ extern "C" {
 
         LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
     };
+    GGML_PACKED_ENUM_END
 
-    enum GGML_PACKED llama_rope_scaling_type {
+    GGML_PACKED_ENUM llama_rope_scaling_type {
         LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED = -1,
         LLAMA_ROPE_SCALING_TYPE_NONE        = 0,
         LLAMA_ROPE_SCALING_TYPE_LINEAR      = 1,
         LLAMA_ROPE_SCALING_TYPE_YARN        = 2,
         LLAMA_ROPE_SCALING_TYPE_LONGROPE    = 3,
         LLAMA_ROPE_SCALING_TYPE_MAX_VALUE   = LLAMA_ROPE_SCALING_TYPE_LONGROPE,
     };
+    GGML_PACKED_ENUM_END
 
-    enum GGML_PACKED llama_pooling_type {
+    GGML_PACKED_ENUM llama_pooling_type {
         LLAMA_POOLING_TYPE_UNSPECIFIED = -1,
         LLAMA_POOLING_TYPE_NONE = 0,
         LLAMA_POOLING_TYPE_MEAN = 1,
         LLAMA_POOLING_TYPE_CLS  = 2,
         LLAMA_POOLING_TYPE_LAST = 3,
         LLAMA_POOLING_TYPE_RANK = 4, // used by reranking models to attach the classification head to the graph
     };
+    GGML_PACKED_ENUM_END
 
-    enum GGML_PACKED llama_attention_type {
+    GGML_PACKED_ENUM llama_attention_type {
         LLAMA_ATTENTION_TYPE_UNSPECIFIED = -1,
         LLAMA_ATTENTION_TYPE_CAUSAL      = 0,
         LLAMA_ATTENTION_TYPE_NON_CAUSAL  = 1,
     };
+    GGML_PACKED_ENUM_END
 
-    enum GGML_PACKED llama_split_mode {
+    GGML_PACKED_ENUM llama_split_mode {
         LLAMA_SPLIT_MODE_NONE  = 0, // single GPU
         LLAMA_SPLIT_MODE_LAYER = 1, // split layers and KV across GPUs
         LLAMA_SPLIT_MODE_ROW   = 2, // split layers and KV across GPUs, use tensor parallelism if supported
     };
+    GGML_PACKED_ENUM_END
 
     // TODO: simplify (https://github.com/ggml-org/llama.cpp/pull/9294#pullrequestreview-2286561979)
     typedef struct llama_token_data {
@@ -261,15 +271,16 @@ extern "C" {
         int8_t       *  logits; // TODO: rename this to "output"
     } llama_batch;
 
-    enum GGML_PACKED llama_model_kv_override_type {
+    GGML_PACKED_ENUM llama_model_kv_override_type {
         LLAMA_KV_OVERRIDE_TYPE_INT,
         LLAMA_KV_OVERRIDE_TYPE_FLOAT,
         LLAMA_KV_OVERRIDE_TYPE_BOOL,
         LLAMA_KV_OVERRIDE_TYPE_STR,
     };
+    GGML_PACKED_ENUM_END
 
     struct llama_model_kv_override {
-        enum llama_model_kv_override_type tag;
+        GGML_PACKED_ENUM llama_model_kv_override_type tag;
 
         char key[128];