678 lines
25 KiB
YAML
678 lines
25 KiB
YAML
# Copyright 2023 Google LLC
|
|
#
|
|
# This source code is licensed under the BSD-style license found in the
|
|
# LICENSE file in the root directory of this source tree.
|
|
|
|
# AVX2 MADD
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avx2_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x8c8__avx2_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x8c8__avx2_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x8c8__avx2_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avx2_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x8c8__avx2_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avx2_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avx2_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avx2_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x8c8__avx2_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x8c8__avx2_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x8c8__avx2_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avx2_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x8c8__avx2_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avx2_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avx2_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
|
|
# AVX256SDK MADD
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avx256skx_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avx256skx_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avx256skx_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avx256skx_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_9x8c8__avx256skx_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_10x8c8__avx256skx_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_12x8c8__avx256skx_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_14x8c8__avx256skx_madd
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avx256skx_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avx256skx_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avx256skx_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avx256skx_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_9x8c8__avx256skx_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_10x8c8__avx256skx_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_12x8c8__avx256skx_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_14x8c8__avx256skx_madd_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4uw_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
|
|
|
|
# ARM NEONI8MM
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x16c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x32c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x8c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x16c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x32c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x8c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x16c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x32c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x8c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x16c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x32c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x16c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x32c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x8c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x16c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x32c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x16c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x32c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x16c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x32c8__neoni8mm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
|
|
# ARM NEONDOT
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x16c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x8c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x16c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x8c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x16c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x8c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x16c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x16c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x8c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x16c4__neondotfp16arith
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
|
|
# ARM NEON
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x16__neonfp16arith_mlal_lane
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x16__neonfp16arith_mlal_lane_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x16__neonfp16arith_mlal_lane
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x16__neonfp16arith_mlal_lane_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x16__neonfp16arith_mlal_lane
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x16__neonfp16arith_mlal_lane_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x16__neonfp16arith_mlal_lane
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x16__neonfp16arith_mlal_lane_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x16__neonfp16arith_mlal_lane
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x16__neonfp16arith_mlal_lane_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
|
|
# AVX256 VNNI
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avx256vnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avx256vnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avx256vnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avx256vnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_9x8c8__avx256vnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_10x8c8__avx256vnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_12x8c8__avx256vnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_14x8c8__avx256vnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avx256vnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avx256vnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avx256vnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avx256vnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_9x8c8__avx256vnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_10x8c8__avx256vnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_12x8c8__avx256vnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_14x8c8__avx256vnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
|
|
# AVX256 VNNI GFNI
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avx256vnnigfni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avx256vnnigfni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avx256vnnigfni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avx256vnnigfni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_9x8c8__avx256vnnigfni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_10x8c8__avx256vnnigfni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_12x8c8__avx256vnnigfni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_14x8c8__avx256vnnigfni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avx256vnnigfni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avx256vnnigfni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avx256vnnigfni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avx256vnnigfni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_9x8c8__avx256vnnigfni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_10x8c8__avx256vnnigfni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_12x8c8__avx256vnnigfni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_14x8c8__avx256vnnigfni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
|
|
# AVXVNNI
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avxvnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x8c8__avxvnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x8c8__avxvnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x8c8__avxvnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avxvnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x8c8__avxvnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avxvnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avxvnni
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avxvnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x8c8__avxvnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x8c8__avxvnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x8c8__avxvnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avxvnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x8c8__avxvnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avxvnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
unsigned-inputs: True
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avxvnni_prfm
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 16
|
|
|
|
# x86 AVX2
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avx2
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x8c8__avx2
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x8c8__avx2
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x8c8__avx2
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avx2
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x8c8__avx2
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avx2
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avx2
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
|
|
# x86 AVX256SKX
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_1x8c8__avx256skx
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_2x8c8__avx256skx
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_3x8c8__avx256skx
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_4x8c8__avx256skx
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_5x8c8__avx256skx
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_6x8c8__avx256skx
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_7x8c8__avx256skx
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|
|
- name: xnn_qd8_f16_qc4w_gemm_minmax_ukernel_8x8c8__avx256skx
|
|
init: xnn_init_f16_qc4w_minmax_scalar_params
|
|
pack: xnn_pack_qs8_qc4w_gemm_goi_w
|
|
k-block: 8
|