whisper.rn 0.5.4 → 0.5.5

package/ios/rnwhisper.xcframework/tvos-arm64/rnwhisper.framework/Headers/ggml-alloc.h

@@ -53,7 +53,14 @@ WSP_GGML_API void           wsp_ggml_gallocr_free(wsp_ggml_gallocr_t galloc);
 // call with a worst-case graph to avoid buffer reallocations
 // not strictly required for single buffer usage: wsp_ggml_gallocr_alloc_graph will reallocate the buffers automatically if needed
 // returns false if the buffer allocation failed
+// wsp_ggml_gallocr_resrve_n_size writes the buffer sizes per galloc buffer that would be allocated by wsp_ggml_gallocr_reserve_n to sizes
 WSP_GGML_API bool wsp_ggml_gallocr_reserve(wsp_ggml_gallocr_t galloc, struct wsp_ggml_cgraph * graph);
+WSP_GGML_API void wsp_ggml_gallocr_reserve_n_size(
+    wsp_ggml_gallocr_t galloc,
+    struct wsp_ggml_cgraph * graph,
+    const int * node_buffer_ids,
+    const int * leaf_buffer_ids,
+    size_t * sizes);
 WSP_GGML_API bool wsp_ggml_gallocr_reserve_n(
     wsp_ggml_gallocr_t galloc,
     struct wsp_ggml_cgraph * graph,
@@ -68,6 +75,8 @@ WSP_GGML_API size_t wsp_ggml_gallocr_get_buffer_size(wsp_ggml_gallocr_t galloc,
 
 // Utils
 // Create a buffer and allocate all the tensors in a wsp_ggml_context
+// wsp_ggml_backend_alloc_ctx_tensors_from_buft_size returns the size of the buffer that would be allocated by wsp_ggml_backend_alloc_ctx_tensors_from_buft
+WSP_GGML_API size_t                       wsp_ggml_backend_alloc_ctx_tensors_from_buft_size(struct wsp_ggml_context * ctx, wsp_ggml_backend_buffer_type_t buft);
 WSP_GGML_API struct wsp_ggml_backend_buffer * wsp_ggml_backend_alloc_ctx_tensors_from_buft(struct wsp_ggml_context * ctx, wsp_ggml_backend_buffer_type_t buft);
 WSP_GGML_API struct wsp_ggml_backend_buffer * wsp_ggml_backend_alloc_ctx_tensors(struct wsp_ggml_context * ctx, wsp_ggml_backend_t backend);
 

package/ios/rnwhisper.xcframework/tvos-arm64/rnwhisper.framework/Headers/ggml-backend-impl.h

@@ -144,7 +144,7 @@ extern "C" {
         // device description: short informative description of the device, could be the model name
         const char * (*get_description)(wsp_ggml_backend_dev_t dev);
 
-        // device memory in bytes
+        // device memory in bytes: 0 bytes to indicate no memory to report
         void         (*get_memory)(wsp_ggml_backend_dev_t dev, size_t * free, size_t * total);
 
         // device type

package/ios/rnwhisper.xcframework/tvos-arm64/rnwhisper.framework/Headers/ggml-backend.h

@@ -307,6 +307,7 @@ extern "C" {
     WSP_GGML_API void                 wsp_ggml_backend_sched_free(wsp_ggml_backend_sched_t sched);
 
     // Initialize backend buffers from a measure graph
+    WSP_GGML_API void                 wsp_ggml_backend_sched_reserve_size(wsp_ggml_backend_sched_t sched, struct wsp_ggml_cgraph * measure_graph, size_t * sizes);
     WSP_GGML_API bool                 wsp_ggml_backend_sched_reserve(wsp_ggml_backend_sched_t sched, struct wsp_ggml_cgraph * measure_graph); // returns success
 
     WSP_GGML_API int                  wsp_ggml_backend_sched_get_n_backends(wsp_ggml_backend_sched_t sched);
@@ -357,7 +358,7 @@ extern "C" {
     typedef bool (*wsp_ggml_backend_eval_callback)(int node_index, struct wsp_ggml_tensor * t1, struct wsp_ggml_tensor * t2, void * user_data);
 
     // Compare the output of two backends
-    WSP_GGML_API bool wsp_ggml_backend_compare_graph_backend(wsp_ggml_backend_t backend1, wsp_ggml_backend_t backend2, struct wsp_ggml_cgraph * graph, wsp_ggml_backend_eval_callback callback, void * user_data, struct wsp_ggml_tensor * test_node);
+    WSP_GGML_API bool wsp_ggml_backend_compare_graph_backend(wsp_ggml_backend_t backend1, wsp_ggml_backend_t backend2, struct wsp_ggml_cgraph * graph, wsp_ggml_backend_eval_callback callback, void * user_data, struct wsp_ggml_tensor const * const * test_nodes, size_t num_test_nodes);
 
     // Tensor initialization
     WSP_GGML_API enum wsp_ggml_status wsp_ggml_backend_tensor_alloc(wsp_ggml_backend_buffer_t buffer, struct wsp_ggml_tensor * tensor, void * addr);

package/ios/rnwhisper.xcframework/tvos-arm64/rnwhisper.framework/Headers/ggml-cpu.h

@@ -99,6 +99,7 @@ extern "C" {
     WSP_GGML_BACKEND_API int wsp_ggml_cpu_has_sme        (void);
     // other
     WSP_GGML_BACKEND_API int wsp_ggml_cpu_has_riscv_v    (void);
+    WSP_GGML_BACKEND_API int wsp_ggml_cpu_get_rvv_vlen   (void);  // risc-v vector length in bytes
     WSP_GGML_BACKEND_API int wsp_ggml_cpu_has_vsx        (void);
     WSP_GGML_BACKEND_API int wsp_ggml_cpu_has_vxe        (void);
     WSP_GGML_BACKEND_API int wsp_ggml_cpu_has_wasm_simd  (void);

package/ios/rnwhisper.xcframework/tvos-arm64/rnwhisper.framework/Headers/ggml-impl.h

@@ -24,10 +24,6 @@
 #include <arm_neon.h>
 #endif
 
-#if defined(__F16C__)
-#include <immintrin.h>
-#endif
-
 #ifdef __cplusplus
 extern "C" {
 #endif

package/ios/rnwhisper.xcframework/tvos-arm64/rnwhisper.framework/Headers/ggml.h

@@ -204,6 +204,10 @@
 #    define WSP_GGML_ATTRIBUTE_FORMAT(...) __attribute__((format(printf, __VA_ARGS__)))
 #endif
 
+#if defined(_WIN32) && !defined(_WIN32_WINNT)
+#    define _WIN32_WINNT 0x0A00
+#endif
+
 #include <stdbool.h>
 #include <stddef.h>
 #include <stdint.h>
@@ -230,6 +234,11 @@
 
 #if UINTPTR_MAX == 0xFFFFFFFF
     #define WSP_GGML_MEM_ALIGN 4
+#elif defined(__EMSCRIPTEN__)
+// emscripten uses max_align_t == 8, so we need WSP_GGML_MEM_ALIGN == 8 for 64-bit wasm.
+// (for 32-bit wasm, the first conditional is true and WSP_GGML_MEM_ALIGN stays 4.)
+// ref: https://github.com/ggml-org/llama.cpp/pull/18628
+    #define WSP_GGML_MEM_ALIGN 8
 #else
     #define WSP_GGML_MEM_ALIGN 16
 #endif
@@ -530,6 +539,7 @@ extern "C" {
         WSP_GGML_OP_ARANGE,
         WSP_GGML_OP_TIMESTEP_EMBEDDING,
         WSP_GGML_OP_ARGSORT,
+        WSP_GGML_OP_TOP_K,
         WSP_GGML_OP_LEAKY_RELU,
         WSP_GGML_OP_TRI,
         WSP_GGML_OP_FILL,
@@ -2147,7 +2157,8 @@ extern "C" {
     };
 
     enum wsp_ggml_scale_flag {
-        WSP_GGML_SCALE_FLAG_ALIGN_CORNERS = (1 << 8)
+        WSP_GGML_SCALE_FLAG_ALIGN_CORNERS = (1 << 8),
+        WSP_GGML_SCALE_FLAG_ANTIALIAS     = (1 << 9),
     };
 
     // interpolate
@@ -2190,6 +2201,15 @@ extern "C" {
             int                  p2,
             int                  p3);
 
+    // pad each dimension with values on the other side of the torus (looping around)
+    WSP_GGML_API struct wsp_ggml_tensor * wsp_ggml_pad_circular(
+            struct wsp_ggml_context * ctx,
+            struct wsp_ggml_tensor  * a,
+            int                   p0,
+            int                   p1,
+            int                   p2,
+            int                   p3);
+
     WSP_GGML_API struct wsp_ggml_tensor * wsp_ggml_pad_ext(
             struct wsp_ggml_context * ctx,
             struct wsp_ggml_tensor  * a,
@@ -2203,6 +2223,19 @@ extern "C" {
             int                  rp3
             );
 
+    // pad each dimension with values on the other side of the torus (looping around)
+    WSP_GGML_API struct wsp_ggml_tensor * wsp_ggml_pad_ext_circular(
+            struct wsp_ggml_context * ctx,
+            struct wsp_ggml_tensor  * a,
+            int                   lp0,
+            int                   rp0,
+            int                   lp1,
+            int                   rp1,
+            int                   lp2,
+            int                   rp2,
+            int                   lp3,
+            int                   rp3);
+
     // pad each dimension with reflection: [a, b, c, d] -> [b, a, b, c, d, c]
     WSP_GGML_API struct wsp_ggml_tensor * wsp_ggml_pad_reflect_1d(
             struct wsp_ggml_context * ctx,
@@ -2258,25 +2291,30 @@ extern "C" {
             struct wsp_ggml_tensor  * a,
             enum wsp_ggml_sort_order  order);
 
-    WSP_GGML_API struct wsp_ggml_tensor * wsp_ggml_arange(
+    // similar to wsp_ggml_top_k but implemented as `argsort` + `view`
+    WSP_GGML_API struct wsp_ggml_tensor * wsp_ggml_argsort_top_k(
             struct wsp_ggml_context * ctx,
-            float                 start,
-            float                 stop,
-            float                 step);
+            struct wsp_ggml_tensor  * a,
+            int                   k);
 
     // top k elements per row
+    // note: the resulting top k indices are in no particular order
     WSP_GGML_API struct wsp_ggml_tensor * wsp_ggml_top_k(
             struct wsp_ggml_context * ctx,
             struct wsp_ggml_tensor  * a,
             int                   k);
 
-#define WSP_GGML_KQ_MASK_PAD 64
+    WSP_GGML_API struct wsp_ggml_tensor * wsp_ggml_arange(
+            struct wsp_ggml_context * ctx,
+            float                 start,
+            float                 stop,
+            float                 step);
 
-    // q:    [n_embd_k, n_batch,     n_head,    ne3 ]
-    // k:    [n_embd_k, n_kv,        n_head_kv, ne3 ]
-    // v:    [n_embd_v, n_kv,        n_head_kv, ne3 ] !! not transposed !!
-    // mask: [n_kv,     n_batch_pad, ne32,      ne33] !! n_batch_pad = WSP_GGML_PAD(n_batch, WSP_GGML_KQ_MASK_PAD) !!
-    // res:  [n_embd_v, n_head,      n_batch,   ne3 ] !! permuted !!
+    // q:    [n_embd_k, n_batch, n_head,    ne3 ]
+    // k:    [n_embd_k, n_kv,    n_head_kv, ne3 ]
+    // v:    [n_embd_v, n_kv,    n_head_kv, ne3 ] !! not transposed !!
+    // mask: [n_kv,     n_batch, ne32,      ne33]
+    // res:  [n_embd_v, n_head,  n_batch,   ne3 ] !! permuted !!
     //
     // broadcast:
     //   n_head % n_head_kv == 0
@@ -2582,7 +2620,8 @@ extern "C" {
 
     // Set callback for all future logging events.
     // If this is not called, or NULL is supplied, everything is output on stderr.
-    WSP_GGML_API void wsp_ggml_log_set(wsp_ggml_log_callback log_callback, void * user_data);
+    WSP_GGML_API void wsp_ggml_log_get(wsp_ggml_log_callback * log_callback, void ** user_data);
+    WSP_GGML_API void wsp_ggml_log_set(wsp_ggml_log_callback   log_callback, void *  user_data);
 
     WSP_GGML_API struct wsp_ggml_tensor * wsp_ggml_set_zero(struct wsp_ggml_tensor * tensor);