| Marat Dukhan | bdb56f5 | 2020-02-05 21:42:49 -0800 | [diff] [blame] | 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 | ################################### ARM NEON ################################## |
| 8 | ### Microkernels without unrolling |
| 9 | tools/xngen src/f16-spmm/neonfp16arith.c.in -D MR=8 -D NR=1 -D UNROLL=1 -o src/f16-spmm/gen/8x1-neonfp16arith.c |
| 10 | tools/xngen src/f16-spmm/neonfp16arith.c.in -D MR=16 -D NR=1 -D UNROLL=1 -o src/f16-spmm/gen/16x1-neonfp16arith.c |
| 11 | tools/xngen src/f16-spmm/neonfp16arith.c.in -D MR=24 -D NR=1 -D UNROLL=1 -o src/f16-spmm/gen/24x1-neonfp16arith.c |
| 12 | tools/xngen src/f16-spmm/neonfp16arith.c.in -D MR=32 -D NR=1 -D UNROLL=1 -o src/f16-spmm/gen/32x1-neonfp16arith.c |
| 13 | ### Microkernels with 2X unrolling |
| 14 | tools/xngen src/f16-spmm/neonfp16arith.c.in -D MR=8 -D NR=1 -D UNROLL=2 -o src/f16-spmm/gen/8x1-neonfp16arith-unroll2.c |
| 15 | tools/xngen src/f16-spmm/neonfp16arith.c.in -D MR=16 -D NR=1 -D UNROLL=2 -o src/f16-spmm/gen/16x1-neonfp16arith-unroll2.c |
| 16 | tools/xngen src/f16-spmm/neonfp16arith.c.in -D MR=24 -D NR=1 -D UNROLL=2 -o src/f16-spmm/gen/24x1-neonfp16arith-unroll2.c |
| 17 | tools/xngen src/f16-spmm/neonfp16arith.c.in -D MR=32 -D NR=1 -D UNROLL=2 -o src/f16-spmm/gen/32x1-neonfp16arith-unroll2.c |
| 18 | |
| 19 | ################################## Unit tests ################################# |
| 20 | tools/generate-spmm-test.py --spec test/f16-spmm.yaml --output test/f16-spmm.cc |