cactus-react-native 0.0.1 → 0.1.1

package/ios/cactus.xcframework/ios-arm64/cactus.framework/Headers/ggml-cpu-quants.h

@@ -1,63 +0,0 @@
-#pragma once
-
-#define LM_GGML_COMMON_DECL_C
-#include "ggml-common.h"
-
-#include "ggml.h"
-
-// GGML CPU internal header
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-// Quantization
-void quantize_row_q4_0(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_q4_1(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_q5_0(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_q5_1(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_q8_0(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_q8_1(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-
-void quantize_row_q2_K(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_q3_K(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_q4_K(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_q5_K(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_q6_K(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_q8_K(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-
-void quantize_row_tq1_0(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_tq2_0(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-
-void quantize_row_iq4_nl (const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-void quantize_row_iq4_xs (const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
-
-// Dot product
-void lm_ggml_vec_dot_q4_0_q8_0(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_q4_1_q8_1(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_q5_0_q8_0(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_q5_1_q8_1(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_q8_0_q8_0(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-
-void lm_ggml_vec_dot_q2_K_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_q3_K_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_q4_K_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_q5_K_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_q6_K_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-
-void lm_ggml_vec_dot_tq1_0_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_tq2_0_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-
-void lm_ggml_vec_dot_iq2_xxs_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_iq2_xs_q8_K (int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_iq2_s_q8_K  (int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_iq3_xxs_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_iq1_s_q8_K  (int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_iq1_m_q8_K  (int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_iq4_nl_q8_0 (int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_iq4_xs_q8_K (int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-void lm_ggml_vec_dot_iq3_s_q8_K  (int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
-
-#ifdef __cplusplus
-}
-#endif

package/ios/cactus.xcframework/ios-arm64/cactus.framework/Headers/ggml-cpu-traits.h

@@ -1,38 +0,0 @@
-#pragma once
-#include "ggml-backend-impl.h"
-#include "ggml-cpu-impl.h"
-#include "ggml.h"
-
-#ifdef __cplusplus
-#    include <vector>
-extern "C" {
-#endif
-
-// return true if op part of extra "accelerator"
-bool lm_ggml_cpu_extra_compute_forward(struct lm_ggml_compute_params * params, struct lm_ggml_tensor * op);
-bool lm_ggml_cpu_extra_work_size(int n_threads, const struct lm_ggml_tensor * op, size_t * size);
-
-#ifdef __cplusplus
-}
-
-namespace ggml::cpu {
-// register in tensor->extra
-class tensor_traits {
-  public:
-    virtual ~tensor_traits();
-    virtual bool work_size(int n_threads, const struct lm_ggml_tensor * op, size_t & size)        = 0;
-    virtual bool compute_forward(struct lm_ggml_compute_params * params, struct lm_ggml_tensor * op) = 0;
-};
-
-class extra_buffer_type {
-  public:
-    virtual ~extra_buffer_type();
-    virtual bool            supports_op(lm_ggml_backend_dev_t dev, const struct lm_ggml_tensor * op) = 0;
-    virtual tensor_traits * get_tensor_traits(const struct lm_ggml_tensor * op)                   = 0;
-};
-}  // namespace ggml::cpu
-
-// implemented in ggml-cpu.cpp.
-std::vector<lm_ggml_backend_buffer_type_t> & lm_ggml_backend_cpu_get_extra_buffers_type();
-
-#endif

package/ios/cactus.xcframework/ios-arm64/cactus.framework/Headers/sgemm.h

@@ -1,14 +0,0 @@
-#pragma once
-#include <stdint.h>
-#include <stdbool.h>
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-bool llamafile_sgemm(const struct lm_ggml_compute_params * params, int64_t, int64_t, int64_t,
-                     const void *, int64_t, const void *, int64_t, void *, int64_t,
-                     int, int, int);
-
-#ifdef __cplusplus
-}
-#endif

package/ios/cactus.xcframework/ios-arm64_x86_64-simulator/cactus.framework/Headers/ggml-cpu-aarch64.h

@@ -1,8 +0,0 @@
-#pragma once
-
-#include "ggml-cpu-traits.h"
-#include "ggml.h"
-
-// GGML internal header
-
-lm_ggml_backend_buffer_type_t lm_ggml_backend_cpu_aarch64_buffer_type(void);

package/ios/cactus.xcframework/ios-arm64_x86_64-simulator/cactus.framework/Headers/ggml-cpu-impl.h

@@ -1,531 +0,0 @@
-#pragma once
-
-// GGML CPU internal header
-
-#include "ggml.h"
-#include "ggml-impl.h"
-#include <stdlib.h> // load `stdlib.h` before other headers to work around MinGW bug: https://sourceforge.net/p/mingw-w64/bugs/192/
-//#include <stddef.h>
-#include <stdbool.h>
-#include <string.h> // memcpy
-#include <math.h>   // fabsf
-
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-struct lm_ggml_compute_params {
-    // ith = thread index, nth = number of threads
-    int ith, nth;
-
-    // work buffer for all threads
-    size_t wsize;
-    void * wdata;
-
-    struct lm_ggml_threadpool * threadpool;
-};
-
-
-#if defined(_MSC_VER)
-
-#define m512bh(p) p
-#define m512i(p) p
-
-#else
-
-#define m512bh(p) (__m512bh)(p)
-#define m512i(p) (__m512i)(p)
-
-#endif
-
-// __FMA__ and __F16C__ are not defined in MSVC, however they are implied with AVX2/AVX512
-#if defined(_MSC_VER) && (defined(__AVX2__) || defined(__AVX512F__))
-#ifndef __FMA__
-#define __FMA__
-#endif
-#ifndef __F16C__
-#define __F16C__
-#endif
-#endif
-
-// __SSE3__ and __SSSE3__ are not defined in MSVC, but SSE3/SSSE3 are present when AVX/AVX2/AVX512 are available
-#if defined(_MSC_VER) && (defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__))
-#ifndef __SSE3__
-#define __SSE3__
-#endif
-#ifndef __SSSE3__
-#define __SSSE3__
-#endif
-#endif
-
-#if defined(__s390x__) && defined(__VEC__)
-#ifndef __VXE__
-#define __VXE__
-#endif
-#ifndef __VXE2__
-#define __VXE2__
-#endif
-#endif
-
-#if defined(__ARM_FEATURE_SVE)
-#include <arm_sve.h>
-#include <sys/prctl.h>
-#endif
-
-// 16-bit float
-// on Arm, we use __fp16
-// on x86, we use uint16_t
-#if defined(__ARM_NEON)
-
-// if YCM cannot find <arm_neon.h>, make a symbolic link to it, for example:
-//
-//   $ ln -sfn /Library/Developer/CommandLineTools/usr/lib/clang/13.1.6/include/arm_neon.h ./src/
-//
-#include <arm_neon.h>
-
-#ifdef _MSC_VER
-
-typedef uint16_t lm_ggml_fp16_internal_t;
-
-#define lm_ggml_vld1q_u32(w,x,y,z) { ((w) + ((uint64_t)(x) << 32)), ((y) + ((uint64_t)(z) << 32)) }
-
-#else
-
-typedef __fp16 lm_ggml_fp16_internal_t;
-
-#define lm_ggml_vld1q_u32(w,x,y,z) { (w), (x), (y), (z) }
-
-#endif // _MSC_VER
-
-#if !defined(__aarch64__)
-
-// 32-bit ARM compatibility
-
-// vaddlvq_s16
-// vpaddq_s16
-// vpaddq_s32
-// vaddvq_s32
-// vaddvq_f32
-// vmaxvq_f32
-// vcvtnq_s32_f32
-// vzip1_u8
-// vzip2_u8
-
-inline static int32_t vaddlvq_s16(int16x8_t v) {
-    int32x4_t v0 = vreinterpretq_s32_s64(vpaddlq_s32(vpaddlq_s16(v)));
-    return vgetq_lane_s32(v0, 0) + vgetq_lane_s32(v0, 2);
-}
-
-inline static int16x8_t vpaddq_s16(int16x8_t a, int16x8_t b) {
-    int16x4_t a0 = vpadd_s16(vget_low_s16(a), vget_high_s16(a));
-    int16x4_t b0 = vpadd_s16(vget_low_s16(b), vget_high_s16(b));
-    return vcombine_s16(a0, b0);
-}
-
-inline static int32x4_t vpaddq_s32(int32x4_t a, int32x4_t b) {
-    int32x2_t a0 = vpadd_s32(vget_low_s32(a), vget_high_s32(a));
-    int32x2_t b0 = vpadd_s32(vget_low_s32(b), vget_high_s32(b));
-    return vcombine_s32(a0, b0);
-}
-
-inline static int32_t vaddvq_s32(int32x4_t v) {
-    return vgetq_lane_s32(v, 0) + vgetq_lane_s32(v, 1) + vgetq_lane_s32(v, 2) + vgetq_lane_s32(v, 3);
-}
-
-inline static float vaddvq_f32(float32x4_t v) {
-    return vgetq_lane_f32(v, 0) + vgetq_lane_f32(v, 1) + vgetq_lane_f32(v, 2) + vgetq_lane_f32(v, 3);
-}
-
-inline static float vmaxvq_f32(float32x4_t v) {
-    return
-        MAX(MAX(vgetq_lane_f32(v, 0), vgetq_lane_f32(v, 1)),
-            MAX(vgetq_lane_f32(v, 2), vgetq_lane_f32(v, 3)));
-}
-
-inline static int32x4_t vcvtnq_s32_f32(float32x4_t v) {
-    int32x4_t res;
-
-    res[0] = roundf(vgetq_lane_f32(v, 0));
-    res[1] = roundf(vgetq_lane_f32(v, 1));
-    res[2] = roundf(vgetq_lane_f32(v, 2));
-    res[3] = roundf(vgetq_lane_f32(v, 3));
-
-    return res;
-}
-
-inline static uint8x8_t vzip1_u8(uint8x8_t a, uint8x8_t b) {
-    uint8x8_t res;
-
-    res[0] = a[0]; res[1] = b[0];
-    res[2] = a[1]; res[3] = b[1];
-    res[4] = a[2]; res[5] = b[2];
-    res[6] = a[3]; res[7] = b[3];
-
-    return res;
-}
-
-inline static uint8x8_t vzip2_u8(uint8x8_t a, uint8x8_t b) {
-    uint8x8_t res;
-
-    res[0] = a[4]; res[1] = b[4];
-    res[2] = a[5]; res[3] = b[5];
-    res[4] = a[6]; res[5] = b[6];
-    res[6] = a[7]; res[7] = b[7];
-
-    return res;
-}
-
-// vld1q_s16_x2
-// vld1q_u8_x2
-// vld1q_u8_x4
-// vld1q_s8_x2
-// vld1q_s8_x4
-// TODO: double-check these work correctly
-
-typedef struct lm_ggml_int16x8x2_t {
-    int16x8_t val[2];
-} lm_ggml_int16x8x2_t;
-
-inline static lm_ggml_int16x8x2_t lm_ggml_vld1q_s16_x2(const int16_t * ptr) {
-    lm_ggml_int16x8x2_t res;
-
-    res.val[0] = vld1q_s16(ptr + 0);
-    res.val[1] = vld1q_s16(ptr + 8);
-
-    return res;
-}
-
-typedef struct lm_ggml_uint8x16x2_t {
-    uint8x16_t val[2];
-} lm_ggml_uint8x16x2_t;
-
-inline static lm_ggml_uint8x16x2_t lm_ggml_vld1q_u8_x2(const uint8_t * ptr) {
-    lm_ggml_uint8x16x2_t res;
-
-    res.val[0] = vld1q_u8(ptr + 0);
-    res.val[1] = vld1q_u8(ptr + 16);
-
-    return res;
-}
-
-typedef struct lm_ggml_uint8x16x4_t {
-    uint8x16_t val[4];
-} lm_ggml_uint8x16x4_t;
-
-inline static lm_ggml_uint8x16x4_t lm_ggml_vld1q_u8_x4(const uint8_t * ptr) {
-    lm_ggml_uint8x16x4_t res;
-
-    res.val[0] = vld1q_u8(ptr + 0);
-    res.val[1] = vld1q_u8(ptr + 16);
-    res.val[2] = vld1q_u8(ptr + 32);
-    res.val[3] = vld1q_u8(ptr + 48);
-
-    return res;
-}
-
-typedef struct lm_ggml_int8x16x2_t {
-    int8x16_t val[2];
-} lm_ggml_int8x16x2_t;
-
-inline static lm_ggml_int8x16x2_t lm_ggml_vld1q_s8_x2(const int8_t * ptr) {
-    lm_ggml_int8x16x2_t res;
-
-    res.val[0] = vld1q_s8(ptr + 0);
-    res.val[1] = vld1q_s8(ptr + 16);
-
-    return res;
-}
-
-typedef struct lm_ggml_int8x16x4_t {
-    int8x16_t val[4];
-} lm_ggml_int8x16x4_t;
-
-inline static lm_ggml_int8x16x4_t lm_ggml_vld1q_s8_x4(const int8_t * ptr) {
-    lm_ggml_int8x16x4_t res;
-
-    res.val[0] = vld1q_s8(ptr + 0);
-    res.val[1] = vld1q_s8(ptr + 16);
-    res.val[2] = vld1q_s8(ptr + 32);
-    res.val[3] = vld1q_s8(ptr + 48);
-
-    return res;
-}
-
-// NOTE: not tested
-inline static int8x16_t lm_ggml_vqtbl1q_s8(int8x16_t a, uint8x16_t b) {
-    int8x16_t res;
-
-    res[ 0] = a[b[ 0]];
-    res[ 1] = a[b[ 1]];
-    res[ 2] = a[b[ 2]];
-    res[ 3] = a[b[ 3]];
-    res[ 4] = a[b[ 4]];
-    res[ 5] = a[b[ 5]];
-    res[ 6] = a[b[ 6]];
-    res[ 7] = a[b[ 7]];
-    res[ 8] = a[b[ 8]];
-    res[ 9] = a[b[ 9]];
-    res[10] = a[b[10]];
-    res[11] = a[b[11]];
-    res[12] = a[b[12]];
-    res[13] = a[b[13]];
-    res[14] = a[b[14]];
-    res[15] = a[b[15]];
-
-    return res;
-}
-
-// NOTE: not tested
-inline static uint8x16_t lm_ggml_vqtbl1q_u8(uint8x16_t a, uint8x16_t b) {
-    uint8x16_t res;
-
-    res[ 0] = a[b[ 0]];
-    res[ 1] = a[b[ 1]];
-    res[ 2] = a[b[ 2]];
-    res[ 3] = a[b[ 3]];
-    res[ 4] = a[b[ 4]];
-    res[ 5] = a[b[ 5]];
-    res[ 6] = a[b[ 6]];
-    res[ 7] = a[b[ 7]];
-    res[ 8] = a[b[ 8]];
-    res[ 9] = a[b[ 9]];
-    res[10] = a[b[10]];
-    res[11] = a[b[11]];
-    res[12] = a[b[12]];
-    res[13] = a[b[13]];
-    res[14] = a[b[14]];
-    res[15] = a[b[15]];
-
-    return res;
-}
-
-#else
-
-#define lm_ggml_int16x8x2_t  int16x8x2_t
-#define lm_ggml_uint8x16x2_t uint8x16x2_t
-#define lm_ggml_uint8x16x4_t uint8x16x4_t
-#define lm_ggml_int8x16x2_t  int8x16x2_t
-#define lm_ggml_int8x16x4_t  int8x16x4_t
-
-#define lm_ggml_vld1q_s16_x2 vld1q_s16_x2
-#define lm_ggml_vld1q_u8_x2  vld1q_u8_x2
-#define lm_ggml_vld1q_u8_x4  vld1q_u8_x4
-#define lm_ggml_vld1q_s8_x2  vld1q_s8_x2
-#define lm_ggml_vld1q_s8_x4  vld1q_s8_x4
-#define lm_ggml_vqtbl1q_s8   vqtbl1q_s8
-#define lm_ggml_vqtbl1q_u8   vqtbl1q_u8
-
-#endif // !defined(__aarch64__)
-
-#if !defined(__ARM_FEATURE_DOTPROD)
-
-inline static int32x4_t lm_ggml_vdotq_s32(int32x4_t acc, int8x16_t a, int8x16_t b) {
-    const int16x8_t p0 = vmull_s8(vget_low_s8 (a), vget_low_s8 (b));
-    const int16x8_t p1 = vmull_s8(vget_high_s8(a), vget_high_s8(b));
-
-    return vaddq_s32(acc, vaddq_s32(vpaddlq_s16(p0), vpaddlq_s16(p1)));
-}
-
-#else
-
-#define lm_ggml_vdotq_s32(a, b, c) vdotq_s32(a, b, c)
-
-#endif // !defined(__ARM_FEATURE_DOTPROD)
-
-#endif // defined(__ARM_NEON)
-
-#ifdef __wasm_simd128__
-#include <wasm_simd128.h>
-#else
-#ifdef __POWER9_VECTOR__
-#include <altivec.h>
-#undef bool
-#define bool _Bool
-#else
-#if defined(_MSC_VER) || defined(__MINGW32__)
-#include <intrin.h>
-#else
-#if defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__) || defined(__SSE3__) || defined(__SSE__)
-#if !defined(__riscv)
-#include <immintrin.h>
-#endif
-#endif
-#endif
-#endif
-#endif
-
-#ifdef __riscv_v_intrinsic
-#include <riscv_vector.h>
-#endif
-
-#if defined(__loongarch64)
-#if defined(__loongarch_asx)
-#include <lasxintrin.h>
-#endif
-#if defined(__loongarch_sx)
-#include <lsxintrin.h>
-#endif
-#endif
-
-#if defined(__VXE__) || defined(__VXE2__)
-#include <vecintrin.h>
-
-#define vec_neg(a)    (-(a))                // Vector Negate
-#define vec_add(a, b) ((a) + (b))           // Vector Add
-#define vec_sub(a, b) ((a) - (b))           // Vector Subtract
-#define vec_mul(a, b) ((a) * (b))           // Vector Multiply
-#define vec_div(a, b) ((a) / (b))           // Vector Divide
-#define vec_sl(a, b)  ((a) << (b))          // Vector Shift Left
-#define vec_sra(a, b) ((a) >> (b))          // Vector Shift Right
-#define vec_sr(a, b)  ((a) >> (b))          // Vector Shift Right Algebraic
-#define vec_slo(a, b) vec_slb(a, (b) << 64) // Vector Shift Left by Octet
-#define vec_sro(a, b) vec_srb(a, (b) << 64) // Vector Shift Right by Octet
-
-#ifndef vec_and
-#define vec_and(a, b) ((a) & (b)) // Vector AND
-#endif
-
-#ifndef vec_or
-#define vec_or(a, b)  ((a) | (b)) // Vector OR
-#endif
-
-#ifndef vec_xor
-#define vec_xor(a, b) ((a) ^ (b)) // Vector XOR
-#endif
-
-typedef signed char char8x16_t __attribute__((vector_size(16)));
-typedef unsigned char uchar8x16_t __attribute__((vector_size(16)));
-
-typedef int8_t  int8x16_t __attribute__((vector_size(16)));
-typedef int16_t int16x8_t __attribute__((vector_size(16)));
-typedef int32_t int32x4_t __attribute__((vector_size(16)));
-
-typedef uint8_t  uint8x16_t __attribute__((vector_size(16)));
-typedef uint16_t uint16x8_t __attribute__((vector_size(16)));
-typedef uint32_t uint32x4_t __attribute__((vector_size(16)));
-
-typedef float float32x4_t __attribute__((vector_size(16)));
-typedef double double64x2_t __attribute((vector_size(16)));
-
-typedef signed long long long64x2_t __attribute((vector_size(16)));
-typedef unsigned long long ulong64x2_t __attribute__((vector_size(16)));
-
-typedef struct lm_ggml_uint8x16x2_t {
-    uint8x16_t val[2];
-} lm_ggml_uint8x16x2_t;
-
-inline static lm_ggml_uint8x16x2_t lm_ggml_vec_xl_u8x2(const uint8_t * ptr) {
-    lm_ggml_uint8x16x2_t res;
-
-    res.val[0] = vec_xl( 0, ptr);
-    res.val[1] = vec_xl(16, ptr);
-
-    return res;
-}
-
-typedef struct lm_ggml_uint8x16x4_t {
-    uint8x16_t val[4];
-} lm_ggml_uint8x16x4_t;
-
-inline static lm_ggml_uint8x16x4_t lm_ggml_vec_xl_u8x4(const uint8_t * ptr) {
-    lm_ggml_uint8x16x4_t res;
-
-    res.val[0] = vec_xl( 0, ptr);
-    res.val[1] = vec_xl(16, ptr);
-    res.val[2] = vec_xl(32, ptr);
-    res.val[3] = vec_xl(48, ptr);
-
-    return res;
-}
-
-typedef struct lm_ggml_int8x16x4_t {
-    int8x16_t val[4];
-} lm_ggml_int8x16x4_t;
-
-inline static lm_ggml_int8x16x4_t lm_ggml_vec_xl_s8x4(const int8_t * ptr) {
-    lm_ggml_int8x16x4_t res;
-
-    res.val[0] = vec_xl( 0, ptr);
-    res.val[1] = vec_xl(16, ptr);
-    res.val[2] = vec_xl(32, ptr);
-    res.val[3] = vec_xl(48, ptr);
-
-    return res;
-}
-
-typedef struct lm_ggml_int16x8x2_t {
-    int16x8_t val[2];
-} lm_ggml_int16x8x2_t;
-
-inline static lm_ggml_int16x8x2_t lm_ggml_vec_xl_s16x2(const int16_t * ptr) {
-    lm_ggml_int16x8x2_t res;
-
-    res.val[0] = vec_xl( 0, ptr);
-    res.val[1] = vec_xl(16, ptr);
-
-    return res;
-}
-
-/*
-    ! WARNING: Very slow. Use vec_perm if possible. Refer to iq4_xs
-    !          or iq4_nl for example implementation.
-*/
-inline static int8x16_t lm_ggml_vec_tbl(int8x16_t a, uint8x16_t b) {
-    int8x16_t res;
-
-    res[ 0] = a[b[ 0]];
-    res[ 1] = a[b[ 1]];
-    res[ 2] = a[b[ 2]];
-    res[ 3] = a[b[ 3]];
-    res[ 4] = a[b[ 4]];
-    res[ 5] = a[b[ 5]];
-    res[ 6] = a[b[ 6]];
-    res[ 7] = a[b[ 7]];
-    res[ 8] = a[b[ 8]];
-    res[ 9] = a[b[ 9]];
-    res[10] = a[b[10]];
-    res[11] = a[b[11]];
-    res[12] = a[b[12]];
-    res[13] = a[b[13]];
-    res[14] = a[b[14]];
-    res[15] = a[b[15]];
-
-    return res;
-}
-
-inline static int16x8_t vec_padd_s16(int16x8_t a, int16x8_t b) {
-    const uchar8x16_t v_maske = {  0,  1,  4,  5,  8,  9, 12, 13,
-                                  16, 17, 20, 21, 24, 25, 28, 29 };
-
-    const int16x8_t v_abo = vec_pack((int32x4_t)a, (int32x4_t)b);
-    const int16x8_t v_abe = vec_perm(a, b, v_maske);
-    return v_abo + v_abe;
-}
-
-inline static int32x4_t lm_ggml_vec_dot(int32x4_t acc, int8x16_t a, int8x16_t b) {
-    const int16x8_t p = vec_mule(a, b) + vec_mulo(a, b);
-    return acc + (vec_unpackh(p) + vec_unpackl(p));
-}
-
-#endif
-
-#if defined(__loongarch_asx)
-/* float type data load instructions */
-static __m128 __lsx_vreplfr2vr_s(const float val) {
-    v4f32 res = {val, val, val, val};
-    return (__m128)res;
-}
-
-static __m256 __lasx_xvreplfr2vr_s(const float val) {
-    v8f32 res = {val, val, val, val, val, val, val, val};
-    return (__m256)res;
-}
-#endif
-
-// TODO: move to ggml-threading
-void lm_ggml_barrier(struct lm_ggml_threadpool * tp);
-
-#ifdef __cplusplus
-}
-#endif