xref: /aosp_15_r20/external/XNNPACK/scripts/generate-f16-igemm.sh (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
1#!/bin/sh
2# Copyright 2020 Google LLC
3#
4# This source code is licensed under the BSD-style license found in the
5# LICENSE file in the root directory of this source tree.
6
7########################## ARM NEON with FP16 compute #########################
8### LD64 micro-kernels
9tools/xngen src/f16-igemm/neonfp16arith-ld64.c.in -D MR=1 -D NR=8  -o src/f16-igemm/gen/1x8-minmax-neonfp16arith-ld64.c &
10tools/xngen src/f16-igemm/neonfp16arith-ld64.c.in -D MR=4 -D NR=8  -o src/f16-igemm/gen/4x8-minmax-neonfp16arith-ld64.c &
11tools/xngen src/f16-igemm/neonfp16arith-ld64.c.in -D MR=6 -D NR=8  -o src/f16-igemm/gen/6x8-minmax-neonfp16arith-ld64.c &
12tools/xngen src/f16-igemm/neonfp16arith-ld64.c.in -D MR=8 -D NR=8  -o src/f16-igemm/gen/8x8-minmax-neonfp16arith-ld64.c &
13tools/xngen src/f16-igemm/neonfp16arith-ld64.c.in -D MR=1 -D NR=16 -o src/f16-igemm/gen/1x16-minmax-neonfp16arith-ld64.c &
14tools/xngen src/f16-igemm/neonfp16arith-ld64.c.in -D MR=4 -D NR=16 -o src/f16-igemm/gen/4x16-minmax-neonfp16arith-ld64.c &
15tools/xngen src/f16-igemm/neonfp16arith-ld64.c.in -D MR=6 -D NR=16 -o src/f16-igemm/gen/6x16-minmax-neonfp16arith-ld64.c &
16tools/xngen src/f16-igemm/neonfp16arith-ld64.c.in -D MR=8 -D NR=16 -o src/f16-igemm/gen/8x16-minmax-neonfp16arith-ld64.c &
17
18################################### x86 AVX2 ###################################
19tools/xngen src/f16-igemm/avx2-broadcast.c.in -D MR=1 -D NR=8  -o src/f16-igemm/gen/1x8-minmax-avx2-broadcast.c &
20tools/xngen src/f16-igemm/avx2-broadcast.c.in -D MR=4 -D NR=8  -o src/f16-igemm/gen/4x8-minmax-avx2-broadcast.c &
21tools/xngen src/f16-igemm/avx2-broadcast.c.in -D MR=5 -D NR=8  -o src/f16-igemm/gen/5x8-minmax-avx2-broadcast.c &
22tools/xngen src/f16-igemm/avx2-broadcast.c.in -D MR=6 -D NR=8  -o src/f16-igemm/gen/6x8-minmax-avx2-broadcast.c &
23tools/xngen src/f16-igemm/avx2-broadcast.c.in -D MR=7 -D NR=8  -o src/f16-igemm/gen/7x8-minmax-avx2-broadcast.c &
24
25tools/xngen src/f16-igemm/avx2-broadcast.c.in -D MR=1 -D NR=16 -o src/f16-igemm/gen/1x16-minmax-avx2-broadcast.c &
26tools/xngen src/f16-igemm/avx2-broadcast.c.in -D MR=3 -D NR=16 -o src/f16-igemm/gen/3x16-minmax-avx2-broadcast.c &
27tools/xngen src/f16-igemm/avx2-broadcast.c.in -D MR=4 -D NR=16 -o src/f16-igemm/gen/4x16-minmax-avx2-broadcast.c &
28tools/xngen src/f16-igemm/avx2-broadcast.c.in -D MR=5 -D NR=16 -o src/f16-igemm/gen/5x16-minmax-avx2-broadcast.c &
29
30################################## Unit tests #################################
31tools/generate-gemm-test.py --spec test/f16-igemm-minmax.yaml --output test/f16-igemm-minmax.cc &
32
33wait
34