whispercpp 1.3.5 → 1.3.7

data/ext/sources/ggml/src/ggml-cpu/amx/mmq.cpp

@@ -1,4 +1,3 @@
-
 #if defined(__GNUC__)
 #pragma GCC diagnostic ignored "-Wpedantic"
 #pragma GCC diagnostic ignored "-Wunused-local-typedefs"
@@ -196,41 +195,33 @@ struct tile_config_t{
 // will be needed.
 //
 // Here another commonly used pattern 1-3-3 is skipped, as it is mostly used when m <=16;
-// and the sinlge batch gemm (m=1) has a special fast path with `avx512-vnni`.
+// and the single batch gemm (m=1) has a special fast path with `avx512-vnni`.
 //
 // ref: https://www.intel.com/content/www/us/en/developer/articles/code-sample/
 //    advanced-matrix-extensions-intrinsics-functions.html
 //
 
-#define TC_CONFIG_TILE(i, r, cb) tc.rows[i] = r; tc.colsb[i] = cb
-void ggml_tile_config_init(void) {
-    static thread_local bool is_first_time = true;
+inline void ggml_tile_config_init(void) {
+    static thread_local bool done = false;
 
-    if (!is_first_time) {
+    if (done) {
         return;
     }
 
-    static thread_local tile_config_t tc;
-    tile_config_t current_tc;
-    _tile_storeconfig(&current_tc);
-
-    // load only when config changes
-    if (tc.palette_id == 0 || (memcmp(&current_tc.colsb, &tc.colsb, sizeof(uint16_t) * 8) != 0 &&
-                               memcmp(&current_tc.rows, &tc.rows, sizeof(uint8_t) * 8) != 0)) {
-        tc.palette_id = 1;
-        tc.start_row = 0;
-        TC_CONFIG_TILE(TMM0, 8, 64);
-        TC_CONFIG_TILE(TMM1, 8, 64);
-        TC_CONFIG_TILE(TMM2, 16, 32);
-        TC_CONFIG_TILE(TMM3, 16, 32);
-        TC_CONFIG_TILE(TMM4, 16, 64);
-        TC_CONFIG_TILE(TMM5, 16, 64);
-        TC_CONFIG_TILE(TMM6, 16, 64);
-        TC_CONFIG_TILE(TMM7, 16, 64);
-        _tile_loadconfig(&tc);
-    }
-
-    is_first_time = false;
+    alignas(64) tile_config_t tc = {};
+    tc.palette_id = 1;
+    tc.start_row = 0;
+    tc.rows[0] = 8;   tc.colsb[0] = 64;
+    tc.rows[1] = 8;   tc.colsb[1] = 64;
+    tc.rows[2] = 16;  tc.colsb[2] = 32;
+    tc.rows[3] = 16;  tc.colsb[3] = 32;
+    tc.rows[4] = 16;  tc.colsb[4] = 64;
+    tc.rows[5] = 16;  tc.colsb[5] = 64;
+    tc.rows[6] = 16;  tc.colsb[6] = 64;
+    tc.rows[7] = 16;  tc.colsb[7] = 64;
+
+    _tile_loadconfig(&tc);
+    done = true;
 }
 
 // we need an extra 16 * 4B (TILE_N * int32_t) for each NB/KB block for compensation.
@@ -268,33 +259,6 @@ int get_row_size(int K) {
     return row_size;
 }
 
-// vectorized dtype conversion
-inline float FP16_TO_FP32(ggml_half val) {
-    __m256i v = _mm256_setr_epi16(
-        val, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
-    __m512 o = _mm512_cvtph_ps(v);
-    return _mm512_cvtss_f32(o);
-}
-
-inline __m512 FP16_TO_FP32_VEC(ggml_half val) {
-    __m256i v = _mm256_set1_epi16(val);
-    return _mm512_cvtph_ps(v);
-}
-
-// horizontal reduce
-inline float _mm512_reduce_max_ps(const __m512 x) {
-    __m512 v = x;
-    __m512 v1 = _mm512_shuffle_f32x4(v, v, 0x4E);
-    v = _mm512_max_ps(v, v1);
-    v1 = _mm512_shuffle_f32x4(v, v, 0xB1);
-    v = _mm512_max_ps(v, v1);
-    v1 = _mm512_shuffle_ps(v, v, 0x4E);
-    v = _mm512_max_ps(v, v1);
-    v1 = _mm512_shuffle_ps(v, v, 0xB1);
-    v = _mm512_max_ps(v, v1);
-    return _mm512_cvtss_f32(v);
-}
-
 // transpose utils
 #define SHUFFLE_EPI32(a, b, mask) \
     _mm256_castps_si256(_mm256_shuffle_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b), mask))
@@ -1370,9 +1334,9 @@ struct tinygemm_kernel_avx<float, ggml_fp16_t, float, BLOCK_M, BLOCK_N, BLOCK_K>
 
 #define LAUNCH_TINYGEMM_KERNEL_AVX(MB_SIZE, NB_SIZE)                                \
     tinygemm_kernel_avx<float, type, float, MB_SIZE, NB_SIZE, blck_size>::apply(    \
-        K, (const float *)src1->data + mb_start * K,                                \
-        (const type *)src0->data + nb_start * K,                                    \
-        (float *)dst->data + mb_start * ldc + nb_start, ldc);
+        K, (const float *)src1->data + src1_offset + mb_start * K,                  \
+        (const type *)src0->data + src0_offset + nb_start * K,                      \
+        (float *)dst->data + dst_offset + mb_start * ldc + nb_start, ldc)
 
 
 // re-organize in the format {NB, KB, TILE_SIZE}:
@@ -1415,8 +1379,8 @@ struct tinygemm_kernel_vnni<block_q8_0, block_q4_0, float, BLOCK_M, BLOCK_N, BLO
         // sum of offsets, shared across COLS
         //
         // avx512-vnni does not have `_mm512_dpbssd_epi32`,
-        // need to transfrom ss to us:
-        //   a * (b - 8) is equavilent to b * a - 8 * a
+        // need to transform ss to us:
+        //   a * (b - 8) is equivalent to b * a - 8 * a
         //   s    u   u                   u   s   u   s
         //
         __m512i vcomp;
@@ -2019,11 +1983,11 @@ struct tinygemm_kernel_vnni<block_q8_K, block_iq4_xs, float, BLOCK_M, BLOCK_N, B
     }
 };
 
-#define LAUNCH_TINYGEMM_KERNEL_VNNI(NB_SIZE)                                         \
-    tinygemm_kernel_vnni<vec_dot_type, type, float, 1, NB_SIZE, blck_size>::apply(   \
-        KB, (const char *)wdata + 0 * row_size_A,                                    \
-        (const char *)src0->data + PACKED_INDEX(nb * kTilesN, 0, KB, TILE_SIZE),     \
-        (float *) dst->data + 0 * N + nb_start, ldc)
+#define LAUNCH_TINYGEMM_KERNEL_VNNI(NB_SIZE)                                                   \
+    tinygemm_kernel_vnni<vec_dot_type, type, float, 1, NB_SIZE, blck_size>::apply(             \
+        KB, wdata_batch,                                                                       \
+        (const char *)src0->data + src0_offset + PACKED_INDEX(nb * kTilesN, 0, KB, TILE_SIZE), \
+        (float *) dst->data + dst_offset + nb_start, ldc)
 
 template <typename TA, typename TB, typename TC, int BLOCK_K,
           typename std::enable_if<!is_type_qkk<TB>::value, int>::type = 0>
@@ -2041,12 +2005,12 @@ void tinygemm_kernel_amx(int M, int N, int KB, const void * RESTRICT _A, const v
     const int lda = KB * sizeof(TA);
     //const int ldb = KB * sizeof(TB);
 
-    static thread_local packed_B_t Tile0[TILE_N * TILE_K];
-    static thread_local packed_B_t Tile1[TILE_N * TILE_K];
-    static thread_local int8_t Tile23[TILE_M * TILE_K];
+    alignas(64) static thread_local packed_B_t Tile0[TILE_N * TILE_K];
+    alignas(64) static thread_local packed_B_t Tile1[TILE_N * TILE_K];
+    alignas(64) static thread_local int8_t Tile23[TILE_M * TILE_K];
 
-    static thread_local int32_t TileC0[TILE_M * TILE_N * 4];
-    static thread_local int32_t TileC1[TILE_M * TILE_N * 4];
+    alignas(64) static thread_local int32_t TileC0[TILE_M * TILE_N * 4];
+    alignas(64) static thread_local int32_t TileC1[TILE_M * TILE_N * 4];
 
     // double buffering C to interleave avx512 and amx
     int32_t * C_cur = TileC0;
@@ -2079,7 +2043,7 @@ void tinygemm_kernel_amx(int M, int N, int KB, const void * RESTRICT _A, const v
         _tile_stored(TMM5, Tile5(C_pre), TILE_N * sizeof(int32_t));
 
         if (need_unpack) {
-            unpack_B<TB>(Tile1, B_blk0);
+            unpack_B<TB>(Tile1, B_blk1);
             _tile_loadd(TMM1, Tile1, TILE_N * VNNI_BLK);
         } else {
             _tile_loadd(TMM1, B_blk1, TILE_N * VNNI_BLK);
@@ -2223,21 +2187,21 @@ void tinygemm_kernel_amx(int M, int N, int KB, const void * RESTRICT _A, const v
     const int m1 = std::max(M - TILE_M, 0);
     //const int lda = KB * sizeof(TA);
 
-    static thread_local int8_t Tile0[TILE_N * TILE_K];
-    static thread_local int8_t Tile1[TILE_N * TILE_K];
-    static thread_local int8_t Tile23[TILE_M * TILE_K];
+    alignas(64) static thread_local int8_t Tile0[TILE_N * TILE_K];
+    alignas(64) static thread_local int8_t Tile1[TILE_N * TILE_K];
+    alignas(64) static thread_local int8_t Tile23[TILE_M * TILE_K];
 
     // mat mul result for each group
-    static thread_local int32_t Tile4[TILE_M * TILE_N];
-    static thread_local int32_t Tile5[TILE_M * TILE_N];
-    static thread_local int32_t Tile6[TILE_M * TILE_N];
-    static thread_local int32_t Tile7[TILE_M * TILE_N];
+    alignas(64) static thread_local int32_t Tile4[TILE_M * TILE_N];
+    alignas(64) static thread_local int32_t Tile5[TILE_M * TILE_N];
+    alignas(64) static thread_local int32_t Tile6[TILE_M * TILE_N];
+    alignas(64) static thread_local int32_t Tile7[TILE_M * TILE_N];
 
     // sum of each QK_K block, contains 8 groups, int32
-    static thread_local int32_t Sumi4[TILE_M * TILE_N];
-    static thread_local int32_t Sumi5[TILE_M * TILE_N];
-    static thread_local int32_t Sumi6[TILE_M * TILE_N];
-    static thread_local int32_t Sumi7[TILE_M * TILE_N];
+    alignas(64) static thread_local int32_t Sumi4[TILE_M * TILE_N];
+    alignas(64) static thread_local int32_t Sumi5[TILE_M * TILE_N];
+    alignas(64) static thread_local int32_t Sumi6[TILE_M * TILE_N];
+    alignas(64) static thread_local int32_t Sumi7[TILE_M * TILE_N];
 
     const int k_group_size = std::is_same<TB, block_q6_K>::value ? 16 : 32;
     for (int i = 0; i < KB; ++i) {
@@ -2336,6 +2300,13 @@ void ggml_backend_amx_convert_weight(struct ggml_tensor * tensor, const void * d
     });
 }
 
+// ne2 is passed explicitly to help compiler optimize repeated calls
+inline int64_t ggml_batch_offset(const ggml_tensor * t, int64_t batch_idx, int64_t ne2) {
+    const int64_t i2 = batch_idx % ne2;
+    const int64_t i3 = batch_idx / ne2;
+    return i3 * t->nb[3] + i2 * t->nb[2];
+}
+
 size_t ggml_backend_amx_desired_wsize(const struct ggml_tensor * dst) {
     struct ggml_tensor * src0 = dst->src[0];
 
@@ -2348,12 +2319,13 @@ size_t ggml_backend_amx_desired_wsize(const struct ggml_tensor * dst) {
 
     const int M = dst->ne[1];
     const int K = src0->ne[0];
+    const int64_t n_batch = dst->ne[2] * dst->ne[3];
 
     size_t desired_wsize = 0;
 
     GGML_DISPATCH_QTYPES(TYPE, [&] {
         const size_t row_size_A = K / blck_size * sizeof(vec_dot_type);
-        desired_wsize = M * row_size_A;
+        desired_wsize = n_batch * M * row_size_A;
     });
 
     return desired_wsize;
@@ -2365,7 +2337,7 @@ size_t ggml_backend_amx_desired_wsize(const struct ggml_tensor * dst) {
 // src1: input  in shape of {M, K}, float32
 // dst:  output in shape of {M, N}, float32
 //
-// the function performs: dst = src1 @ src0.T
+// the function performs: dst = src1 @ src0.T for each batch
 //
 void ggml_backend_amx_mul_mat(const ggml_compute_params * params, struct ggml_tensor * dst) {
     struct ggml_tensor * src0 = dst->src[0];
@@ -2382,17 +2354,26 @@ void ggml_backend_amx_mul_mat(const ggml_compute_params * params, struct ggml_te
     const int K = src0->ne[0];
     const int ldc = dst->nb[1] / dst->nb[0];
 
+    const int64_t ne2 = dst->ne[2];
+    const int64_t n_batch = ne2 * dst->ne[3];
+
     if (is_floating_type) {
         constexpr int BLOCK_M = 4;
         constexpr int BLOCK_N = 6;
         const int MB = div_up(M, BLOCK_M);
         const int NB = div_up(N, BLOCK_N);
 
-        parallel_for_ggml(params, MB * NB, [&](int begin, int end) {
+        parallel_for_ggml(params, n_batch * MB * NB, [&](int begin, int end) {
             GGML_DISPATCH_FLOATING_TYPES(TYPE, [&] {
                 for (int i = begin; i < end; ++i) {
-                    int mb = i / NB;
-                    int nb = i % NB;
+                    int batch_idx = i / (MB * NB);
+                    int remaining = i % (MB * NB);
+                    int mb = remaining / NB;
+                    int nb = remaining % NB;
+
+                    int64_t src0_offset = ggml_batch_offset(src0, batch_idx, ne2);
+                    int64_t src1_offset = ggml_batch_offset(src1, batch_idx, ne2);
+                    int64_t dst_offset  = ggml_batch_offset(dst,  batch_idx, ne2);
 
                     int mb_start = mb * BLOCK_M;
                     int mb_size = std::min(BLOCK_M, M - mb_start);
@@ -2424,10 +2405,10 @@ void ggml_backend_amx_mul_mat(const ggml_compute_params * params, struct ggml_te
     void * wdata = params->wdata;
 
     //TODO: performance improvement: merge quant A
-    if (params->ith == 0) {
+ // if (params->ith == 0) {
         GGML_DISPATCH_QTYPES(TYPE, [&] {
             const size_t row_size_A = K / blck_size * sizeof(vec_dot_type);
-            const size_t desired_wsize = M * row_size_A;
+            const size_t desired_wsize = n_batch * M * row_size_A;
             if (params->wsize < desired_wsize) {
                 GGML_ABORT("insufficient work space size");
             }
@@ -2436,12 +2417,19 @@ void ggml_backend_amx_mul_mat(const ggml_compute_params * params, struct ggml_te
             // Q4_K, Q5_K, Q6_K, IQ4_XS handles 8 TILE_K per blck_size
             GGML_ASSERT(TILE_K == blck_size || TILE_K * 8 == blck_size);
 
-            const float * A_data = static_cast<const float *>(src1->data);
-            for (int m = 0; m < M; ++m) {
-                from_float<vec_dot_type>(A_data + m * K, (char *)wdata + m * row_size_A, K);
-            }
+            parallel_for_ggml(params, n_batch, [&](int begin, int end) {
+                for (int batch_idx = begin; batch_idx < end; ++batch_idx) {
+                    int64_t src1_offset = ggml_batch_offset(src1, batch_idx, ne2);
+                    const float * A_data = (const float *)((const char *)src1->data + src1_offset);
+                    char * wdata_batch = (char *)wdata + batch_idx * M * row_size_A;
+
+                    for (int m = 0; m < M; ++m) {
+                        from_float<vec_dot_type>(A_data + m * K, wdata_batch + m * row_size_A, K);
+                    }
+                }
+            });
         });
-    }
+ // }
 
     ggml_barrier(params->threadpool);
 
@@ -2451,13 +2439,19 @@ void ggml_backend_amx_mul_mat(const ggml_compute_params * params, struct ggml_te
         constexpr int BLOCK_N = TILE_N * kTilesN;
         const int NB = div_up(N, BLOCK_N);
 
-        parallel_for_ggml(params, NB, [&](int begin, int end) {
+        parallel_for_ggml(params, n_batch * NB, [&](int begin, int end) {
             GGML_DISPATCH_QTYPES(TYPE, [&] {
                 const int KB = K / blck_size;
                 const int TILE_SIZE = get_tile_size<type>();
                 const int row_size_A = KB * sizeof(vec_dot_type);
                 for (int i = begin; i < end; ++i) {
-                    int nb = i;
+                    int batch_idx = i / NB;
+                    int nb = i % NB;
+
+                    int64_t src0_offset = ggml_batch_offset(src0, batch_idx, ne2);
+                    int64_t dst_offset  = ggml_batch_offset(dst,  batch_idx, ne2);
+                    const char * wdata_batch = (const char *)wdata + batch_idx * row_size_A;
+
                     int nb_start = nb * BLOCK_N;
                     int nb_size = std::min(BLOCK_N, N - nb_start); // 32, 64, 96
 
@@ -2481,7 +2475,7 @@ void ggml_backend_amx_mul_mat(const ggml_compute_params * params, struct ggml_te
     const int MB = div_up(M, BLOCK_M);
     const int NB = div_up(N, BLOCK_N);
 
-    parallel_for_ggml(params, MB * NB, [&](int begin, int end) {
+    parallel_for_ggml(params, n_batch * MB * NB, [&](int begin, int end) {
         // init tile config for each thread
         ggml_tile_config_init();
 
@@ -2491,8 +2485,14 @@ void ggml_backend_amx_mul_mat(const ggml_compute_params * params, struct ggml_te
             const int row_size_A = KB * sizeof(vec_dot_type);
 
             for (int i = begin; i < end; ++i) {
-                int mb = i / NB;
-                int nb = i % NB;
+                int batch_idx = i / (MB * NB);
+                int remaining = i % (MB * NB);
+                int mb = remaining / NB;
+                int nb = remaining % NB;
+
+                int64_t src0_offset = ggml_batch_offset(src0, batch_idx, ne2);
+                int64_t dst_offset  = ggml_batch_offset(dst,  batch_idx, ne2);
+                const char * wdata_batch = (const char *)wdata + batch_idx * M * row_size_A;
 
                 int mb_start = mb * BLOCK_M;
                 int mb_size = std::min(BLOCK_M, M - mb_start);
@@ -2501,9 +2501,9 @@ void ggml_backend_amx_mul_mat(const ggml_compute_params * params, struct ggml_te
 
                 tinygemm_kernel_amx<vec_dot_type, type, float, blck_size>(
                     mb_size, nb_size, KB,
-                    (const char *)wdata + mb_start * row_size_A,
-                    (const char *)src0->data + PACKED_INDEX(nb * 2, 0, KB, TILE_SIZE),
-                    (float *) dst->data + mb_start * N + nb_start, ldc);
+                    wdata_batch + mb_start * row_size_A,
+                    (const char *)src0->data + src0_offset + PACKED_INDEX(nb * 2, 0, KB, TILE_SIZE),
+                    (float *) dst->data + dst_offset + mb_start * N + nb_start, ldc);
             }
         });
     });

data/ext/sources/ggml/src/ggml-cpu/arch/arm/quants.c

@@ -137,6 +137,89 @@ void quantize_row_q8_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, in
 
 //===================================== Dot products =================================
 
+void ggml_vec_dot_q1_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc) {
+    const int qk = QK1_0;  // 128
+    const int nb = n / qk;
+
+    assert(n % qk == 0);
+    assert(nrc == 1);
+    UNUSED(nrc);
+    UNUSED(bx);
+    UNUSED(by);
+    UNUSED(bs);
+
+    const block_q1_0 * GGML_RESTRICT x = vx;
+    const block_q8_0 * GGML_RESTRICT y = vy;
+
+#if defined(__ARM_NEON)
+    float32x4_t sumv = vdupq_n_f32(0.0f);
+
+    for (int i = 0; i < nb; i++) {
+        const float d0 = GGML_CPU_FP16_TO_FP32(x[i].d);
+
+        // Process 4 Q8_0 blocks (each has 32 elements)
+        for (int k = 0; k < 4; k++) {
+            const block_q8_0 * GGML_RESTRICT yb = &y[i * 4 + k];
+            const float d1 = GGML_CPU_FP16_TO_FP32(yb->d);
+
+            // Get the 4 bytes of bits for this Q8_0 block (32 bits = 4 bytes)
+            // Bits are at offset k*4 bytes in x[i].qs
+            const uint8_t * bits = &x[i].qs[k * 4];
+
+            // Load 32 int8 values from y
+            const int8x16_t y0 = vld1q_s8(yb->qs);
+            const int8x16_t y1 = vld1q_s8(yb->qs + 16);
+
+            // Byte 0-1: bits for y0[0..15]
+            const uint64_t expand0 = table_b2b_0[bits[0]];
+            const uint64_t expand1 = table_b2b_0[bits[1]];
+            // Byte 2-3: bits for y1[0..15]
+            const uint64_t expand2 = table_b2b_0[bits[2]];
+            const uint64_t expand3 = table_b2b_0[bits[3]];
+
+            // Build the sign vectors by reinterpreting the table values
+            uint8x8_t e0 = vcreate_u8(expand0);
+            uint8x8_t e1 = vcreate_u8(expand1);
+            uint8x8_t e2 = vcreate_u8(expand2);
+            uint8x8_t e3 = vcreate_u8(expand3);
+
+            // Shift right by 4 to get 0 or 1
+            int8x8_t s0 = vreinterpret_s8_u8(vshr_n_u8(e0, 4));
+            int8x8_t s1 = vreinterpret_s8_u8(vshr_n_u8(e1, 4));
+            int8x8_t s2 = vreinterpret_s8_u8(vshr_n_u8(e2, 4));
+            int8x8_t s3 = vreinterpret_s8_u8(vshr_n_u8(e3, 4));
+
+            // Convert 0/1 to -1/+1: sign = 2*val - 1
+            int8x8_t one = vdup_n_s8(1);
+            s0 = vsub_s8(vadd_s8(s0, s0), one);  // 2*s0 - 1
+            s1 = vsub_s8(vadd_s8(s1, s1), one);
+            s2 = vsub_s8(vadd_s8(s2, s2), one);
+            s3 = vsub_s8(vadd_s8(s3, s3), one);
+
+            // Combine into 16-element vectors
+            int8x16_t signs0 = vcombine_s8(s0, s1);
+            int8x16_t signs1 = vcombine_s8(s2, s3);
+
+            // Multiply signs with y values and accumulate
+            // dot(signs, y) where signs are +1/-1
+            int32x4_t p0 = ggml_vdotq_s32(vdupq_n_s32(0), signs0, y0);
+            int32x4_t p1 = ggml_vdotq_s32(p0, signs1, y1);
+
+            // Scale by d1 and accumulate
+            sumv = vmlaq_n_f32(sumv, vcvtq_f32_s32(p1), d0 * d1);
+        }
+    }
+
+    *s = vaddvq_f32(sumv);
+#else
+    UNUSED(nb);
+    UNUSED(x);
+    UNUSED(y);
+    ggml_vec_dot_q1_0_q8_0_generic(n, s, bs, vx, bx, vy, by, nrc);
+#endif
+}
+
+
 void ggml_vec_dot_q4_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc) {
     const int qk = QK8_0;
     const int nb = n / qk;
@@ -650,6 +733,116 @@ void ggml_vec_dot_mxfp4_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const vo
     *s = sumf;
 }
 
+void ggml_vec_dot_nvfp4_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc) {
+    assert(nrc == 1);
+    UNUSED(nrc);
+    UNUSED(bx);
+    UNUSED(by);
+    UNUSED(bs);
+    assert(n % QK_NVFP4 == 0);
+
+    const block_nvfp4 * GGML_RESTRICT x = vx;
+    const block_q8_0 * GGML_RESTRICT y = vy;
+
+    // Each NVFP4 super-block (64 elements) spans 2 q8_0 blocks
+    const int nb = n / QK_NVFP4;
+
+    float sumf = 0;
+
+#if defined(__ARM_NEON) && defined(__ARM_FEATURE_FMA)
+    const int8x16_t values = vld1q_s8(kvalues_mxfp4);
+    const uint8x16_t m4b = vdupq_n_u8(0x0f);
+    float32x4_t acc = vdupq_n_f32(0.0f);
+
+    for (int ib = 0; ib < nb; ++ib) {
+        const uint8x16_t q4bits_0 = vld1q_u8(x[ib].qs);
+        const uint8x16_t q4bits_1 = vld1q_u8(x[ib].qs + 16);
+
+        const int8x16_t q4_lo_0 = ggml_vqtbl1q_s8(values, vandq_u8  (q4bits_0, m4b));
+        const int8x16_t q4_hi_0 = ggml_vqtbl1q_s8(values, vshrq_n_u8(q4bits_0, 4));
+        const int8x16_t q4_lo_1 = ggml_vqtbl1q_s8(values, vandq_u8  (q4bits_1, m4b));
+        const int8x16_t q4_hi_1 = ggml_vqtbl1q_s8(values, vshrq_n_u8(q4bits_1, 4));
+
+#if defined(__ARM_FEATURE_DOTPROD)
+        const int8x16_t q8_0a = vld1q_s8(y[2*ib].qs);
+        const int8x16_t q8_0b = vld1q_s8(y[2*ib].qs + 16);
+        const int8x16_t q8_lo_0 = vcombine_s8(vget_low_s8(q8_0a), vget_low_s8(q8_0b));
+        const int8x16_t q8_hi_0 = vcombine_s8(vget_high_s8(q8_0a), vget_high_s8(q8_0b));
+
+        const int8x16_t q8_1a = vld1q_s8(y[2*ib+1].qs);
+        const int8x16_t q8_1b = vld1q_s8(y[2*ib+1].qs + 16);
+        const int8x16_t q8_lo_1 = vcombine_s8(vget_low_s8(q8_1a), vget_low_s8(q8_1b));
+        const int8x16_t q8_hi_1 = vcombine_s8(vget_high_s8(q8_1a), vget_high_s8(q8_1b));
+
+        const int32x4_t p0 = vaddq_s32(
+            vdotq_s32(vdupq_n_s32(0), q4_lo_0, q8_lo_0),
+            vdotq_s32(vdupq_n_s32(0), q4_hi_0, q8_hi_0));
+        const int32x4_t p1 = vaddq_s32(
+            vdotq_s32(vdupq_n_s32(0), q4_lo_1, q8_lo_1),
+            vdotq_s32(vdupq_n_s32(0), q4_hi_1, q8_hi_1));
+
+        const int32x4_t sumi = vpaddq_s32(p0, p1);
+#else
+        const int8x8_t q4_0_lo = vget_low_s8(q4_lo_0);
+        const int8x8_t q4_0_hi = vget_low_s8(q4_hi_0);
+        const int8x8_t q4_1_lo = vget_high_s8(q4_lo_0);
+        const int8x8_t q4_1_hi = vget_high_s8(q4_hi_0);
+        const int8x8_t q4_2_lo = vget_low_s8(q4_lo_1);
+        const int8x8_t q4_2_hi = vget_low_s8(q4_hi_1);
+        const int8x8_t q4_3_lo = vget_high_s8(q4_lo_1);
+        const int8x8_t q4_3_hi = vget_high_s8(q4_hi_1);
+
+        const int8x8_t q8_0_lo = vld1_s8(y[2*ib].qs);
+        const int8x8_t q8_0_hi = vld1_s8(y[2*ib].qs + 8);
+        const int8x8_t q8_1_lo = vld1_s8(y[2*ib].qs + 16);
+        const int8x8_t q8_1_hi = vld1_s8(y[2*ib].qs + 24);
+        const int8x8_t q8_2_lo = vld1_s8(y[2*ib+1].qs);
+        const int8x8_t q8_2_hi = vld1_s8(y[2*ib+1].qs + 8);
+        const int8x8_t q8_3_lo = vld1_s8(y[2*ib+1].qs + 16);
+        const int8x8_t q8_3_hi = vld1_s8(y[2*ib+1].qs + 24);
+
+        const int32x4_t sumi = (int32x4_t){
+            vaddvq_s32(ggml_nvfp4_dot8(q4_0_lo, q8_0_lo, q4_0_hi, q8_0_hi)),
+            vaddvq_s32(ggml_nvfp4_dot8(q4_1_lo, q8_1_lo, q4_1_hi, q8_1_hi)),
+            vaddvq_s32(ggml_nvfp4_dot8(q4_2_lo, q8_2_lo, q4_2_hi, q8_2_hi)),
+            vaddvq_s32(ggml_nvfp4_dot8(q4_3_lo, q8_3_lo, q4_3_hi, q8_3_hi)),
+        };
+#endif
+
+        const float dy0 = GGML_CPU_FP16_TO_FP32(y[2*ib].d);
+        const float dy1 = GGML_CPU_FP16_TO_FP32(y[2*ib+1].d);
+        const float32x4_t nvsc = {
+            ggml_ue4m3_to_fp32(x[ib].d[0]),
+            ggml_ue4m3_to_fp32(x[ib].d[1]),
+            ggml_ue4m3_to_fp32(x[ib].d[2]),
+            ggml_ue4m3_to_fp32(x[ib].d[3])
+        };
+        const float32x4_t scales = vmulq_f32(nvsc, (float32x4_t){dy0, dy0, dy1, dy1});
+
+        acc = vfmaq_f32(acc, vcvtq_f32_s32(sumi), scales);
+    }
+    sumf = vaddvq_f32(acc);
+#else
+    for (int ib = 0; ib < nb; ++ib) {
+        for (int si = 0; si < 4; ++si) {
+            const float d = ggml_ue4m3_to_fp32(x[ib].d[si]);
+            const int q8b = si / 2;
+            const int q8o = (si % 2) * QK_NVFP4_SUB;
+            const float dy = GGML_CPU_FP16_TO_FP32(y[2*ib + q8b].d);
+
+            int sumi_lo = 0, sumi_hi = 0;
+            for (int j = 0; j < QK_NVFP4_SUB/2; ++j) {
+                const uint8_t qv = x[ib].qs[si*(QK_NVFP4_SUB/2) + j];
+                sumi_lo += y[2*ib + q8b].qs[q8o + j +               0] * kvalues_mxfp4[qv & 0xf];
+                sumi_hi += y[2*ib + q8b].qs[q8o + j + QK_NVFP4_SUB/2] * kvalues_mxfp4[qv >>  4];
+            }
+            sumf += dy * d * (sumi_lo + sumi_hi);
+        }
+    }
+#endif
+    *s = sumf;
+}
+
 void ggml_vec_dot_q5_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc) {
     const int qk = QK8_0;
     const int nb = n / qk;
@@ -968,7 +1161,7 @@ void ggml_vec_dot_q8_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const voi
 
     const int vector_length = ggml_cpu_get_sve_cnt()*8;
 
-    //VLA Implemenation for SVE
+    //VLA Implementation for SVE
     switch (vector_length) {
         case 128:
             {