Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Fix test to load autotuning results from cache instead of actually co…
…mputing it. This test wants to assume CUBLAS is always faster than Triton for this particular GEMM, and can only be so via autotuning DB. PiperOrigin-RevId: 622277357
- Loading branch information
1 parent
c4cff96
commit 7b54481
Showing
3 changed files
with
40 additions
and
0 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
25 changes: 25 additions & 0 deletions
25
third_party/xla/xla/service/gpu/gpu_compiler_test_autotune_db.textproto
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,25 @@ | ||
version: 3 | ||
results { | ||
device: "sm_9.0 with 84942979072B RAM, 132 cores, 1980000KHz clock, 2619000KHz mem clock, 52428800B L2$" | ||
hlo: "(bf16[128,1024,1024]{2,1,0}, s8[33554432]{0}) custom-call(bf16[128,1024,1024]{2,1,0}, bf16[128,1024,1024]{2,1,0}), custom_call_target=\"__cublas$gemm\", backend_config={\"operation_queue_id\":\"0\",\"wait_on_operation_queues\":[],\"gemm_backend_config\":{\"alpha_real\":1,\"beta\":0,\"dot_dimension_numbers\":{\"lhs_contracting_dimensions\":[\"2\"],\"rhs_contracting_dimensions\":[\"1\"],\"lhs_batch_dimensions\":[\"0\"],\"rhs_batch_dimensions\":[\"0\"]},\"alpha_imag\":0,\"precision_config\":{\"operand_precision\":[\"DEFAULT\",\"DEFAULT\"],\"algorithm\":\"ALG_UNSET\"},\"epilogue\":\"DEFAULT\",\"lhs_stride\":\"1048576\",\"rhs_stride\":\"1048576\",\"grad_x\":false,\"grad_y\":false},\"force_earliest_schedule\":false}" | ||
result { | ||
gemm { | ||
algorithm: -1 | ||
} | ||
run_time { | ||
nanos: 657376 | ||
} | ||
} | ||
} | ||
results { | ||
device: "sm_9.0 with 84942979072B RAM, 132 cores, 1980000KHz clock, 2619000KHz mem clock, 52428800B L2$" | ||
hlo: "{\n tmp_0 = bf16[1,4,32,1024,1024]{4,3,2,1,0} parameter(0)\n tmp_1 = bf16[] constant({...})\n tmp_2 = bf16[1,4,32,1024,1024]{4,3,2,1,0} broadcast(bf16[] tmp_1), dimensions={}\n tmp_3 = bf16[1,4,32,1024,1024]{4,3,2,1,0} multiply(bf16[1,4,32,1024,1024]{4,3,2,1,0} tmp_0, bf16[1,4,32,1024,1024]{4,3,2,1,0} tmp_2)\n tmp_4 = bf16[4,32,1024,1024]{3,2,1,0} bitcast(bf16[1,4,32,1024,1024]{4,3,2,1,0} tmp_3)\n tmp_5 = bf16[4,32,1024,1024]{3,2,1,0} transpose(bf16[4,32,1024,1024]{3,2,1,0} tmp_4), dimensions={0,1,3,2}\n tmp_6 = bf16[128,1024,1024]{2,1,0} bitcast(bf16[4,32,1024,1024]{3,2,1,0} tmp_5)\n tmp_7 = bf16[1,4,32,1024,1024]{4,3,2,1,0} parameter(1)\n tmp_8 = bf16[128,1024,1024]{2,1,0} bitcast(bf16[1,4,32,1024,1024]{4,3,2,1,0} tmp_7)\n tmp_9 = bf16[128,1024,1024]{2,1,0} dot(bf16[128,1024,1024]{2,1,0} tmp_6, bf16[128,1024,1024]{2,1,0} tmp_8), lhs_batch_dims={0}, lhs_contracting_dims={2}, rhs_batch_dims={0}, rhs_contracting_dims={1}\n ROOT tmp_10 = bf16[4,32,1024,1024]{3,2,1,0} bitcast(bf16[128,1024,1024]{2,1,0} tmp_9)\n}" | ||
result { | ||
gemm { | ||
algorithm: -1 | ||
} | ||
run_time { | ||
nanos: 854688 | ||
} | ||
} | ||
} |