mirror of
https://github.com/LostRuins/koboldcpp.git
synced 2025-09-10 17:14:36 +00:00
a0535ffa0d
* implement unary REGLU/GEGLU/SWIGLU cpu ops * relax constraints * duplicate shape of source * fix ggml_vec_geglu_f16 * special case gated ops * implement unary REGLU/GEGLU/SWIGLU cuda ops * tighten constraints again * refactor into GGML_GLU_OP * metal : add glu kernels ggml-ci * add CUDA_GLU_BLOCK_SIZE [no ci] * more constraints and use 64bit ints ggml-ci * 64bit multiplication [no ci] * implement swapped variants (cpu/cuda) * update comment [no ci] ggml-ci * Vulkan: Add GLU ops and shaders * SYCL: Implement fused kernel GEGLU, SWIGLU and REGLU for single up+gate * ggml : implement GLU for split up/gate (#14181) * implement GLU for split up/gate * add tests for ggml_glu_split * Vulkan: Implement glu_split logic and shader support * add split to logging [no ci] * SYCL: refactor element_size ops and add split up and gate support to gated kernels * SYCL: switch GEGLU to use tanh approximation --------- Co-authored-by: 0cc4m <picard12@live.de> Co-authored-by: Akarshan <akarshan@menlo.ai> * GGML: increase OP count in assertion * Refactor: Optimize SYCL element-wise operations with unary function inlining This commit refactors the SYCL element-wise operations to improve performance by: - Inlining unary operations (sgn, abs, elu, gelu, silu, etc.) to reduce kernel launch overhead. - Introducing helper functions `op_xxx` for each unary operation to encapsulate the logic. - Replacing direct kernel calls with calls to these inlined functions. - Using `__dpct_inline__` to encourage compiler inlining. - Minor code cleanup and consistency improvements. The changes aim to reduce kernel launch overhead and improve the overall efficiency of element-wise operations on SYCL devices. * vulkan: Increase workgroup size for GLU, for performance (#14345) * vulkan: Increase workgroup size for GLU, for performance * vulkan: change GLU shaders to do one element per invocation rather than one row per workgroup * merge fix * metal : add support for split and swap ggml-ci --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> Co-authored-by: 0cc4m <picard12@live.de> Co-authored-by: Akarshan <akarshan@menlo.ai> Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
113 lines
8.3 KiB
C++
113 lines
8.3 KiB
C++
#pragma once
|
|
|
|
#include "ggml.h"
|
|
|
|
//
|
|
// cache line
|
|
//
|
|
|
|
#if defined(__cpp_lib_hardware_interference_size)
|
|
#define CACHE_LINE_SIZE std::hardware_destructive_interference_size
|
|
#else
|
|
#if defined(__POWER9_VECTOR__)
|
|
#define CACHE_LINE_SIZE 128
|
|
#elif defined(__VXE__) || defined(__VXE2__)
|
|
#define CACHE_LINE_SIZE 256
|
|
#else
|
|
#define CACHE_LINE_SIZE 64
|
|
#endif
|
|
#endif
|
|
|
|
static const size_t CACHE_LINE_SIZE_F32 = CACHE_LINE_SIZE/sizeof(float);
|
|
|
|
#ifdef __cplusplus
|
|
extern "C" {
|
|
#endif
|
|
|
|
void ggml_compute_forward_dup(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_add(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_add1(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_acc(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_sum(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_sum_rows(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_mean(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_argmax(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_count_equal(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_repeat(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_repeat_back(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_concat(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_silu_back(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_norm(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_rms_norm(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_rms_norm_back(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_group_norm(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_l2_norm(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_out_prod(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_scale(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_set(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_cpy(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_cont(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_reshape(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_view(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_permute(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_transpose(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_get_rows(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_get_rows_back(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_set_rows(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_diag(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_diag_mask_inf(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_diag_mask_zero(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_soft_max(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_soft_max_ext_back(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_rope(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_rope_back(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_clamp(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_conv_transpose_1d(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_im2col(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_im2col_back_f32(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_conv_transpose_2d(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_conv_2d_dw(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_pool_1d(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_pool_2d(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_pool_2d_back(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_upscale(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_pad(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_pad_reflect_1d(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_roll(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_arange(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_timestep_embedding(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_argsort(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_leaky_relu(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_flash_attn_ext(
|
|
const struct ggml_compute_params * params,
|
|
const struct ggml_tensor * q,
|
|
const struct ggml_tensor * k,
|
|
const struct ggml_tensor * v,
|
|
const struct ggml_tensor * mask,
|
|
struct ggml_tensor * dst);
|
|
void ggml_compute_forward_flash_attn_back(
|
|
const struct ggml_compute_params * params,
|
|
const bool masked,
|
|
struct ggml_tensor * dst);
|
|
void ggml_compute_forward_ssm_conv(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_ssm_scan(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_win_part(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_win_unpart(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_unary(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_glu(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_get_rel_pos(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_add_rel_pos(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_rwkv_wkv6(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_rwkv_wkv7(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_gla(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_map_custom1(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_map_custom2(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_map_custom3(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_custom(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_cross_entropy_loss(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_cross_entropy_loss_back(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
void ggml_compute_forward_opt_step_adamw(const struct ggml_compute_params * params, struct ggml_tensor * dst);
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|