sglang_v0.5.2/pytorch_2.8.0/third_party/XNNPACK/test/qd8-f32-qb4w-gemm-minmax.yaml

538 lines
19 KiB
YAML

# Copyright 2024 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.
# Scalar
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x2__scalar
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4__scalar
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x8__scalar
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x2__scalar
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4__scalar
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x8__scalar
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4__scalar
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
# x86 AVX
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__avx_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__avx_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__avx_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__avx_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__avx_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__avx_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__avx_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__avx_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
# x86 SSE2
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse2_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse2_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse2_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse2_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse2_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse2_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse2_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse2_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
# x86 SSE4.1
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse41_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse41_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse41_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse41_ld128
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse41_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse41_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse41_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse41_ld64
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
# ARM NEONDOT
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x8c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x8c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x16c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x8c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x16c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x8c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x16c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x8c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x8c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x16c4__neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
# x86 AVX2
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x8c8__avx2
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x8c8__avx2
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x8c8__avx2
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x8c8__avx2
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
# ARM NEON
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16__neon_mlal_lane
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16__neon_mlal_lane_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x16__neon_mlal_lane
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x16__neon_mlal_lane_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x16__neon_mlal_lane
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x16__neon_mlal_lane_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x16__neon_mlal_lane
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x16__neon_mlal_lane_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x16__neon_mlal_lane
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x16__neon_mlal_lane_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
# ARM NEONI8MM
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x8c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x32c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x8c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x16c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x32c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x8c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x16c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x32c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x8c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x16c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x32c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x8c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x32c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x8c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x16c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x32c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x8c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x32c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x8c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x32c8__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
# AVX512VNNI
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
# AVX512VNNIGFNI
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnnigfni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnnigfni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnnigfni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnnigfni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnnigfni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnnigfni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnnigfni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnnigfni
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnnigfni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnnigfni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnnigfni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnnigfni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnnigfni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnnigfni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnnigfni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True
- name: xnn_qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnnigfni_prfm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_qs8_qb4w_gemm_goi_w
k-block: 32
unsigned-inputs: True