q2 k dequantize and get_rows

ggml/src/ggml-cpu/repack.cpp

@@ -21,7 +21,6 @@
 #if defined(__GNUC__)
 #pragma GCC diagnostic ignored "-Woverlength-strings"
 #endif
-
 #define UNUSED GGML_UNUSED
 
 static inline int nearest_int(float fval) {
@@ -1576,6 +1575,11 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int64_t NB_COLS, ggml_type PAR
 
                     size += sizeof_mmid_row_mapping*ne02*(ne12 + 1);
 
+                    return true;
+                }
+            case GGML_OP_GET_ROWS:
+                {
+                    size = 0;  // GET_ROWS (standard and repacked) doesn't need a work buffer
                     return true;
                 }
             default:
@@ -1593,6 +1597,9 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int64_t NB_COLS, ggml_type PAR
             case GGML_OP_MUL_MAT_ID:
                 forward_mul_mat_id(params, op);
                 return true;
+            case GGML_OP_GET_ROWS:
+                forward_get_rows(params, op);
+                return true;
             default:
                 // GGML_ABORT("fatal error");
                 break;
@@ -1801,6 +1808,145 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int64_t NB_COLS, ggml_type PAR
 #undef MMID_MATRIX_ROW
     }
 
+    void forward_get_rows(const ggml_compute_params * params,
+                          ggml_tensor * dst) {
+        const ggml_tensor * src0 = dst->src[0];
+
+        switch (src0->type) {
+            case GGML_TYPE_Q2_K:
+                ggml_compute_forward_get_rows_q2_Kx8(params, dst);
+                break;
+            default:
+                GGML_ABORT("fatal error");
+                break;
+        }
+    }
+
+    static void ggml_compute_forward_get_rows_q2_Kx8(
+        const ggml_compute_params * params,
+        ggml_tensor * dst) {
+        const ggml_tensor * src0 = dst->src[0];
+        const ggml_tensor * src1 = dst->src[1];
+
+        GGML_TENSOR_BINARY_OP_LOCALS
+        const int64_t nc = ne00;
+        const int64_t nr = ggml_nelements(src1);
+
+        assert(ne0 == nc);
+        assert(ne02 == ne11);
+        assert(nb00 == ggml_type_size(src0->type));
+        assert(ggml_nrows(dst) == nr);
+
+        const int ith = params->ith;
+        const int nth = params->nth;
+
+        // rows per thread
+        const int dr = (nr + nth - 1) / nth;
+
+        // row range for this thread
+        const int ir0 = dr * ith;
+        const int ir1 = MIN(ir0 + dr, nr);
+
+        constexpr int nrows_interleaved = 8;
+        const size_t sizeof_one_repacked_block = sizeof(block_q2_Kx8);
+
+        const int num_repacked_blocks_per_row_width = nc / QK_K;
+
+        const size_t stride_between_actual_row_groups = num_repacked_blocks_per_row_width * sizeof_one_repacked_block;
+
+        for (int64_t i = ir0; i < ir1; ++i) {
+            const int64_t i12 = i / (ne11 * ne10);
+            const int64_t i11 = (i - i12 * ne11 * ne10) / ne10;
+            const int64_t i10 = (i - i12 * ne11 * ne10 - i11 * ne10);
+            const int64_t i01 = *(int32_t *)((char *)src1->data + i10 * nb10 + i11 * nb11 + i12 * nb12);  // original logical row
+
+            GGML_ASSERT(i01 >= 0 && i01 < ne01);
+
+            const int row_group_idx = i01 / nrows_interleaved;
+            const int row_idx_in_group = i01 % nrows_interleaved;
+
+            const char * base_ptr_for_higher_dims_in_src0 = (const char *)src0->data + i11 * nb02 + i12 * nb03;
+
+            // Pointer to the first block_q2_Kx8 of the identified row_group_idx
+            const block_q2_Kx8 * p_first_repacked_block_of_group_x8 = (const block_q2_Kx8 *)(base_ptr_for_higher_dims_in_src0 + row_group_idx * stride_between_actual_row_groups);
+
+            dequantize_row_q2_Kx8(
+                p_first_repacked_block_of_group_x8,
+                (float *)((char *)dst->data + i10 * nb1 + i11 * nb2 + i12 * nb3), nc, row_idx_in_group);
+        }
+    }
+
+    /**
+     * Dequantizes a single logical row from the repacked q2_Kx8 data format.
+     *
+     * @param p_repacked_blocks Pointer to the start of the 'block_q_Kx8' structures for the entire row.
+     * @param y                 Output buffer for the dequantized float values.
+     * @param k                 Total number of elements (columns) in the logical row.
+     * @param row_idx_in_group  The index (0-7) of the logical row to extract from the interleaved data.
+     */
+
+    static void dequantize_row_q2_Kx8(
+        const void * GGML_RESTRICT p_repacked_blocks,
+        float * GGML_RESTRICT y,
+        int64_t k,
+        int row_idx_in_group) {
+        assert(k % QK_K == 0);
+        assert(row_idx_in_group >= 0 && row_idx_in_group < 8);
+
+        const int nb = k / QK_K;
+        const block_q2_Kx8 * blocks = (const block_q2_Kx8 *)p_repacked_blocks;
+        int out_pos = 0;
+        for (int i = 0; i < nb; i++) {
+            const block_q2_Kx8 * current_block = &blocks[i];
+
+            const float d_super_block = GGML_FP16_TO_FP32(current_block->d[row_idx_in_group]);
+            const float dmin_super_block = GGML_FP16_TO_FP32(current_block->dmin[row_idx_in_group]);
+
+            const uint8_t * ptr_qs_base = current_block->qs;
+
+            uint8_t * ptr_repacked_scales = (uint8_t *)current_block->scales;
+            float dl, ml;
+            int is = 0;
+            for (int n = 0; n < QK_K; n += 128) {
+                int shift = 0;
+                for (int j = 0; j < 4; ++j) {
+                    const uint8_t sc0 = read_scale_from_repacked(ptr_repacked_scales, row_idx_in_group, is++);
+                    dl = d_super_block * (sc0 & 0xF); 
+                    ml = dmin_super_block * (sc0 >> 4);
+
+                    for (int l = 0; l < 16; ++l) {
+                        *y++ = dl * ((int8_t)((read_q_from_repacked(ptr_qs_base, row_idx_in_group, n/4 + l) >> shift) & 3)) - ml;
+                    }
+
+                    const uint8_t sc1 = read_scale_from_repacked(ptr_repacked_scales, row_idx_in_group, is++);
+                    dl = d_super_block * (sc1 & 0xF); 
+                    ml = dmin_super_block * (sc1 >> 4);
+
+                    for (int l = 0; l < 16; ++l) {
+                        *y++ = dl * ((int8_t)((read_q_from_repacked(ptr_qs_base, row_idx_in_group, n/4 + l + 16) >> shift) & 3)) - ml;
+                    }
+
+                    shift +=2;
+                }
+            }
+        }
+    }
+
+    static inline uint8_t read_scale_from_repacked(const uint8_t* ptr_repacked_scales, int row_idx_in_group, int scale_idx) {
+        const int pair_group_idx = scale_idx / 2;
+        const int sub_idx_in_pair = scale_idx % 2;
+        const int offset = pair_group_idx * 16 + row_idx_in_group * 2 + sub_idx_in_pair;
+        return ptr_repacked_scales[offset];
+    }
+
+    static inline uint8_t read_q_from_repacked(const uint8_t* ptr_q_base, int row_idx_in_group, int q_idx) {
+        const int block_size_interleave  = 8;
+        const int chunk_idx = q_idx / block_size_interleave;
+        const int offset_in_chunk = q_idx % block_size_interleave;
+        const int offset = chunk_idx * (8 * block_size_interleave) + row_idx_in_group * block_size_interleave + offset_in_chunk;
+        return ptr_q_base[offset];
+    }
+
     int repack(struct ggml_tensor * t, const void * data, size_t data_size) override {
         GGML_LOG_DEBUG("%s: repack tensor %s with %s_%dx%d\n", __func__, t->name, ggml_type_name(t->type),
                        (int) NB_COLS, (int) INTER_SIZE);
@@ -1949,12 +2095,23 @@ class extra_buffer_type : ggml::cpu::extra_buffer_type {
             //if (op->src[1]->type == GGML_TYPE_Q8_0) {
             //    return true;
             //}
+        } else if (op->op == GGML_OP_GET_ROWS
+        && op->src[0]->buffer
+        && (ggml_n_dims(op->src[0]) == 2)
+        && op->src[0]->buffer->buft == ggml_backend_cpu_repack_buffer_type()
+        && ggml_repack_get_optimal_repack_type(op->src[0])) {
+        if (op->src[1]->buffer && !ggml_backend_buft_is_host(op->src[1]->buffer->buft)) {
+            return false;
+        }
+        if (op->src[0]->type == GGML_TYPE_Q2_K) {
+            return true;
+        }
         }
         return false;
     }
 
     ggml::cpu::tensor_traits * get_tensor_traits(const struct ggml_tensor * op) override {
-        if (op->op == GGML_OP_MUL_MAT || op->op == GGML_OP_MUL_MAT_ID) {
+        if (op->op == GGML_OP_MUL_MAT || op->op == GGML_OP_MUL_MAT_ID || op->op == GGML_OP_GET_ROWS) {
             if (op->src[0]->buffer && op->src[0]->buffer->buft == ggml_backend_cpu_repack_buffer_type()) {
                 return (ggml::cpu::tensor_traits *) op->src[0]->extra;
             }