1#!/bin/sh 2# Copyright 2019 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############################### AArch64 assembly ############################## 8tools/xngen src/f16-gemm/1x16-aarch64-neonfp16arith-ld32.S.in -D INC=0 -o src/f16-gemm/gen/1x16-minmax-aarch64-neonfp16arith-ld32.S & 9tools/xngen src/f16-gemm/4x16-aarch64-neonfp16arith-ld32.S.in -D INC=0 -o src/f16-gemm/gen/4x16-minmax-aarch64-neonfp16arith-ld32.S & 10tools/xngen src/f16-gemm/6x16-aarch64-neonfp16arith-ld32.S.in -D INC=0 -o src/f16-gemm/gen/6x16-minmax-aarch64-neonfp16arith-ld32.S & 11tools/xngen src/f16-gemm/1x16-aarch64-neonfp16arith-ld32.S.in -D INC=1 -o src/f16-gemm/gen-inc/1x16inc-minmax-aarch64-neonfp16arith-ld32.S & 12tools/xngen src/f16-gemm/4x16-aarch64-neonfp16arith-ld32.S.in -D INC=1 -o src/f16-gemm/gen-inc/4x16inc-minmax-aarch64-neonfp16arith-ld32.S & 13tools/xngen src/f16-gemm/6x16-aarch64-neonfp16arith-ld32.S.in -D INC=1 -o src/f16-gemm/gen-inc/6x16inc-minmax-aarch64-neonfp16arith-ld32.S & 14 15tools/xngen src/f16-gemm/1x8-aarch64-neonfp16arith-ld64.S.in -D INC=0 -o src/f16-gemm/gen/1x8-minmax-aarch64-neonfp16arith-ld64.S & 16tools/xngen src/f16-gemm/4x8-aarch64-neonfp16arith-ld64.S.in -D INC=0 -o src/f16-gemm/gen/4x8-minmax-aarch64-neonfp16arith-ld64.S & 17tools/xngen src/f16-gemm/6x8-aarch64-neonfp16arith-ld64.S.in -D INC=0 -o src/f16-gemm/gen/6x8-minmax-aarch64-neonfp16arith-ld64.S & 18tools/xngen src/f16-gemm/8x8-aarch64-neonfp16arith-ld64.S.in -D INC=0 -o src/f16-gemm/gen/8x8-minmax-aarch64-neonfp16arith-ld64.S & 19tools/xngen src/f16-gemm/1x8-aarch64-neonfp16arith-ld64.S.in -D INC=1 -o src/f16-gemm/gen-inc/1x8inc-minmax-aarch64-neonfp16arith-ld64.S & 20tools/xngen src/f16-gemm/4x8-aarch64-neonfp16arith-ld64.S.in -D INC=1 -o src/f16-gemm/gen-inc/4x8inc-minmax-aarch64-neonfp16arith-ld64.S & 21tools/xngen src/f16-gemm/6x8-aarch64-neonfp16arith-ld64.S.in -D INC=1 -o src/f16-gemm/gen-inc/6x8inc-minmax-aarch64-neonfp16arith-ld64.S & 22tools/xngen src/f16-gemm/8x8-aarch64-neonfp16arith-ld64.S.in -D INC=1 -o src/f16-gemm/gen-inc/8x8inc-minmax-aarch64-neonfp16arith-ld64.S & 23 24### Cortex A55 micro-kernels 25tools/xngen src/f16-gemm/6x16-aarch64-neonfp16arith-cortex-a55.S.in -D INC=0 -o src/f16-gemm/gen/6x16-minmax-aarch64-neonfp16arith-cortex-a55.S & 26tools/xngen src/f16-gemm/6x16-aarch64-neonfp16arith-cortex-a55.S.in -D INC=1 -o src/f16-gemm/gen-inc/6x16inc-minmax-aarch64-neonfp16arith-cortex-a55.S & 27 28### Cortex A75 micro-kernels 29tools/xngen src/f16-gemm/6x16-aarch64-neonfp16arith-cortex-a75.S.in -D INC=0 -o src/f16-gemm/gen/6x16-minmax-aarch64-neonfp16arith-cortex-a75.S & 30tools/xngen src/f16-gemm/6x16-aarch64-neonfp16arith-cortex-a75.S.in -D INC=1 -o src/f16-gemm/gen-inc/6x16inc-minmax-aarch64-neonfp16arith-cortex-a75.S & 31 32########################## ARM NEON with FP16 compute ######################### 33### LD64 micro-kernels 34tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=1 -D NR=8 -D INC=0 -o src/f16-gemm/gen/1x8-minmax-neonfp16arith-ld64.c & 35tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=4 -D NR=8 -D INC=0 -o src/f16-gemm/gen/4x8-minmax-neonfp16arith-ld64.c & 36tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=6 -D NR=8 -D INC=0 -o src/f16-gemm/gen/6x8-minmax-neonfp16arith-ld64.c & 37tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=8 -D NR=8 -D INC=0 -o src/f16-gemm/gen/8x8-minmax-neonfp16arith-ld64.c & 38tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=1 -D NR=8 -D INC=1 -o src/f16-gemm/gen-inc/1x8inc-minmax-neonfp16arith-ld64.c & 39tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=4 -D NR=8 -D INC=1 -o src/f16-gemm/gen-inc/4x8inc-minmax-neonfp16arith-ld64.c & 40tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=6 -D NR=8 -D INC=1 -o src/f16-gemm/gen-inc/6x8inc-minmax-neonfp16arith-ld64.c & 41tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=8 -D NR=8 -D INC=1 -o src/f16-gemm/gen-inc/8x8inc-minmax-neonfp16arith-ld64.c & 42 43tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=1 -D NR=16 -D INC=0 -o src/f16-gemm/gen/1x16-minmax-neonfp16arith-ld64.c & 44tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=4 -D NR=16 -D INC=0 -o src/f16-gemm/gen/4x16-minmax-neonfp16arith-ld64.c & 45tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=6 -D NR=16 -D INC=0 -o src/f16-gemm/gen/6x16-minmax-neonfp16arith-ld64.c & 46tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=8 -D NR=16 -D INC=0 -o src/f16-gemm/gen/8x16-minmax-neonfp16arith-ld64.c & 47tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=1 -D NR=16 -D INC=1 -o src/f16-gemm/gen-inc/1x16inc-minmax-neonfp16arith-ld64.c & 48tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=4 -D NR=16 -D INC=1 -o src/f16-gemm/gen-inc/4x16inc-minmax-neonfp16arith-ld64.c & 49tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=6 -D NR=16 -D INC=1 -o src/f16-gemm/gen-inc/6x16inc-minmax-neonfp16arith-ld64.c & 50tools/xngen src/f16-gemm/neonfp16arith-ld64.c.in -D MR=8 -D NR=16 -D INC=1 -o src/f16-gemm/gen-inc/8x16inc-minmax-neonfp16arith-ld64.c & 51 52################################### x86 AVX2 ################################### 53tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=1 -D NR=8 -o src/f16-gemm/gen/1x8-minmax-avx2-broadcast.c & 54tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=4 -D NR=8 -o src/f16-gemm/gen/4x8-minmax-avx2-broadcast.c & 55tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=5 -D NR=8 -o src/f16-gemm/gen/5x8-minmax-avx2-broadcast.c & 56tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=6 -D NR=8 -o src/f16-gemm/gen/6x8-minmax-avx2-broadcast.c & 57tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=7 -D NR=8 -o src/f16-gemm/gen/7x8-minmax-avx2-broadcast.c & 58 59tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=1 -D NR=16 -o src/f16-gemm/gen/1x16-minmax-avx2-broadcast.c & 60tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=3 -D NR=16 -o src/f16-gemm/gen/3x16-minmax-avx2-broadcast.c & 61tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=4 -D NR=16 -o src/f16-gemm/gen/4x16-minmax-avx2-broadcast.c & 62tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=5 -D NR=16 -o src/f16-gemm/gen/5x16-minmax-avx2-broadcast.c & 63 64################################## Unit tests ################################# 65tools/generate-gemm-test.py --spec test/f16-gemm-minmax.yaml --output test/f16-gemm-minmax.cc & 66 67wait 68