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/4x16-aarch64-neonfp16arith-ld64.S.in -D INC=0 -o src/f16-gemm/gen/4x16-minmax-aarch64-neonfp16arith-ld64.S & 11tools/xngen src/f16-gemm/6x16-aarch64-neonfp16arith-ld32.S.in -D INC=0 -o src/f16-gemm/gen/6x16-minmax-aarch64-neonfp16arith-ld32.S & 12tools/xngen src/f16-gemm/6x16-aarch64-neonfp16arith-ld64.S.in -D INC=0 -o src/f16-gemm/gen/6x16-minmax-aarch64-neonfp16arith-ld64.S & 13tools/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 & 14tools/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 & 15tools/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 & 16 17tools/xngen src/f16-gemm/1x8-aarch64-neonfp16arith-ld64.S.in -D INC=0 -o src/f16-gemm/gen/1x8-minmax-aarch64-neonfp16arith-ld64.S & 18tools/xngen src/f16-gemm/4x8-aarch64-neonfp16arith-ld64.S.in -D INC=0 -o src/f16-gemm/gen/4x8-minmax-aarch64-neonfp16arith-ld64.S & 19tools/xngen src/f16-gemm/6x8-aarch64-neonfp16arith-ld64.S.in -D INC=0 -o src/f16-gemm/gen/6x8-minmax-aarch64-neonfp16arith-ld64.S & 20tools/xngen src/f16-gemm/8x8-aarch64-neonfp16arith-ld64.S.in -D INC=0 -o src/f16-gemm/gen/8x8-minmax-aarch64-neonfp16arith-ld64.S & 21tools/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 & 22tools/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 & 23tools/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 & 24tools/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 & 25 26### Cortex A55r0 micro-kernel 27tools/xngen src/f16-gemm/6x16-aarch64-neonfp16arith-cortex-a55r0.S.in -D INC=0 -o src/f16-gemm/gen/6x16-minmax-aarch64-neonfp16arith-cortex-a55r0.S & 28 29### Cortex A55 micro-kernels 30tools/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 & 31tools/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 & 32 33### Cortex A75 micro-kernels 34tools/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 & 35tools/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 & 36 37########################## ARM NEON with FP16 compute ######################### 38### LD64 micro-kernels 39tools/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 & 40tools/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 & 41tools/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 & 42tools/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 & 43tools/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 & 44tools/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 & 45tools/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 & 46tools/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 & 47 48tools/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 & 49tools/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 & 50tools/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 & 51tools/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 & 52tools/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 & 53tools/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 & 54tools/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 & 55tools/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 & 56 57################################### x86 AVX2 ################################### 58tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=1 -D NR=8 -o src/f16-gemm/gen/1x8-minmax-avx2-broadcast.c & 59tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=4 -D NR=8 -o src/f16-gemm/gen/4x8-minmax-avx2-broadcast.c & 60tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=5 -D NR=8 -o src/f16-gemm/gen/5x8-minmax-avx2-broadcast.c & 61tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=6 -D NR=8 -o src/f16-gemm/gen/6x8-minmax-avx2-broadcast.c & 62tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=7 -D NR=8 -o src/f16-gemm/gen/7x8-minmax-avx2-broadcast.c & 63 64tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=1 -D NR=16 -o src/f16-gemm/gen/1x16-minmax-avx2-broadcast.c & 65tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=3 -D NR=16 -o src/f16-gemm/gen/3x16-minmax-avx2-broadcast.c & 66tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=4 -D NR=16 -o src/f16-gemm/gen/4x16-minmax-avx2-broadcast.c & 67tools/xngen src/f16-gemm/avx2-broadcast.c.in -D MR=5 -D NR=16 -o src/f16-gemm/gen/5x16-minmax-avx2-broadcast.c & 68 69################################## Unit tests ################################# 70tools/generate-gemm-test.py --spec test/f16-gemm-minmax.yaml --output test/f16-gemm-minmax.cc & 71 72wait 73