|
| 1 | +# Copyright 2026 The OpenXLA Authors. |
| 2 | +# |
| 3 | +# Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | +# you may not use this file except in compliance with the License. |
| 5 | +# You may obtain a copy of the License at |
| 6 | +# |
| 7 | +# http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | +# |
| 9 | +# Unless required by applicable law or agreed to in writing, software |
| 10 | +# distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | +# See the License for the specific language governing permissions and |
| 13 | +# limitations under the License. |
| 14 | + |
| 15 | +config { block_m: 32 block_n: 32 block_k: 256 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 16 | +config { block_m: 64 block_n: 32 block_k: 32 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 17 | +config { block_m: 32 block_n: 64 block_k: 64 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 18 | +config { block_m: 128 block_n: 128 block_k: 64 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 19 | +config { block_m: 16 block_n: 16 block_k: 256 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 20 | +config { block_m: 16 block_n: 128 block_k: 32 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 21 | +config { block_m: 256 block_n: 256 block_k: 32 num_stages: 2 num_warps: 8 num_ctas: 1 waves_per_eu: 0 } |
| 22 | +config { block_m: 128 block_n: 256 block_k: 64 num_stages: 2 num_warps: 8 num_ctas: 1 waves_per_eu: 0 } |
| 23 | +config { block_m: 128 block_n: 256 block_k: 32 num_stages: 2 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 24 | +config { block_m: 256 block_n: 128 block_k: 64 num_stages: 2 num_warps: 8 num_ctas: 1 waves_per_eu: 0 } |
| 25 | +config { block_m: 128 block_n: 128 block_k: 64 num_stages: 2 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 26 | +config { block_m: 32 block_n: 8 block_k: 16 num_stages: 2 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 27 | +config { block_m: 128 block_n: 32 block_k: 16 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 4 } |
| 28 | +config { block_m: 32 block_n: 8 block_k: 32 num_stages: 1 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 29 | +config { block_m: 128 block_n: 32 block_k: 16 num_stages: 1 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 30 | +config { block_m: 256 block_n: 256 block_k: 16 num_stages: 4 num_warps: 8 num_ctas: 1 waves_per_eu: 0 } |
| 31 | +config { block_m: 256 block_n: 128 block_k: 64 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 32 | +config { block_m: 256 block_n: 128 block_k: 64 num_stages: 2 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 33 | +config { block_m: 256 block_n: 128 block_k: 16 num_stages: 2 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 34 | +config { block_m: 16 block_n: 16 block_k: 128 num_stages: 3 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 35 | +config { block_m: 32 block_n: 16 block_k: 128 num_stages: 2 num_warps: 4 num_ctas: 1 waves_per_eu: 4 } |
| 36 | +config { block_m: 64 block_n: 8 block_k: 128 num_stages: 4 num_warps: 2 num_ctas: 1 waves_per_eu: 2 } |
| 37 | +config { block_m: 64 block_n: 16 block_k: 64 num_stages: 2 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 38 | +config { block_m: 128 block_n: 32 block_k: 16 num_stages: 3 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 39 | +config { block_m: 64 block_n: 32 block_k: 16 num_stages: 1 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 40 | +config { block_m: 32 block_n: 16 block_k: 32 num_stages: 2 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 41 | +config { block_m: 64 block_n: 8 block_k: 16 num_stages: 3 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 42 | +config { block_m: 16 block_n: 8 block_k: 256 num_stages: 1 num_warps: 2 num_ctas: 1 waves_per_eu: 1 } |
| 43 | +config { block_m: 16 block_n: 16 block_k: 128 num_stages: 2 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 44 | +config { block_m: 16 block_n: 16 block_k: 128 num_stages: 1 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 45 | +config { block_m: 16 block_n: 64 block_k: 128 num_stages: 2 num_warps: 2 num_ctas: 1 waves_per_eu: 4 } |
| 46 | +config { block_m: 32 block_n: 16 block_k: 64 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 47 | +config { block_m: 64 block_n: 8 block_k: 16 num_stages: 3 num_warps: 2 num_ctas: 1 waves_per_eu: 4 } |
| 48 | +config { block_m: 64 block_n: 8 block_k: 64 num_stages: 2 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 49 | +config { block_m: 64 block_n: 8 block_k: 256 num_stages: 2 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 50 | +config { block_m: 64 block_n: 16 block_k: 256 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 51 | +config { block_m: 128 block_n: 8 block_k: 32 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 52 | +config { block_m: 128 block_n: 32 block_k: 64 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 53 | +config { block_m: 128 block_n: 64 block_k: 16 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 54 | +config { block_m: 32 block_n: 32 block_k: 128 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 0 } |
| 55 | +config { block_m: 32 block_n: 32 block_k: 128 num_stages: 1 num_warps: 8 num_ctas: 1 waves_per_eu: 4 } |
| 56 | +config { block_m: 32 block_n: 64 block_k: 128 num_stages: 4 num_warps: 2 num_ctas: 1 waves_per_eu: 1 } |
| 57 | +config { block_m: 64 block_n: 32 block_k: 32 num_stages: 3 num_warps: 8 num_ctas: 1 waves_per_eu: 0 } |
| 58 | +config { block_m: 64 block_n: 64 block_k: 32 num_stages: 4 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 59 | +config { block_m: 64 block_n: 64 block_k: 128 num_stages: 4 num_warps: 4 num_ctas: 1 waves_per_eu: 2 } |
| 60 | +config { block_m: 64 block_n: 64 block_k: 128 num_stages: 3 num_warps: 4 num_ctas: 1 waves_per_eu: 1 } |
| 61 | +config { block_m: 64 block_n: 64 block_k: 128 num_stages: 3 num_warps: 4 num_ctas: 1 waves_per_eu: 4 } |
| 62 | +config { block_m: 64 block_n: 64 block_k: 128 num_stages: 4 num_warps: 4 num_ctas: 1 waves_per_eu: 4 } |
| 63 | +config { block_m: 64 block_n: 128 block_k: 128 num_stages: 1 num_warps: 2 num_ctas: 1 waves_per_eu: 1 } |
| 64 | +config { block_m: 128 block_n: 64 block_k: 64 num_stages: 2 num_warps: 4 num_ctas: 1 waves_per_eu: 2 } |
| 65 | +config { block_m: 128 block_n: 64 block_k: 128 num_stages: 2 num_warps: 2 num_ctas: 1 waves_per_eu: 0 } |
| 66 | +config { block_m: 128 block_n: 128 block_k: 32 num_stages: 2 num_warps: 8 num_ctas: 1 waves_per_eu: 1 } |
| 67 | +config { block_m: 128 block_n: 128 block_k: 128 num_stages: 3 num_warps: 8 num_ctas: 1 waves_per_eu: 0 } |
| 68 | +config { block_m: 128 block_n: 128 block_k: 128 num_stages: 3 num_warps: 8 num_ctas: 1 waves_per_eu: 2 } |
| 69 | +config { block_m: 128 block_n: 128 block_k: 128 num_stages: 1 num_warps: 8 num_ctas: 1 waves_per_eu: 4 } |
| 70 | +config { block_m: 128 block_n: 256 block_k: 64 num_stages: 1 num_warps: 4 num_ctas: 1 waves_per_eu: 2 } |
| 71 | +config { block_m: 256 block_n: 256 block_k: 64 num_stages: 3 num_warps: 8 num_ctas: 1 waves_per_eu: 0 } |
| 72 | +config { block_m: 256 block_n: 256 block_k: 64 num_stages: 3 num_warps: 8 num_ctas: 1 waves_per_eu: 1 } |
0 commit comments