Gemm+Reduce Fusion (#128)

* add gridwise gemm v4r1

* rename

* adding gemm+reduce

* adding gemm+reduce

* adding gemm+reduce

* adding gemm+reduce

* use sfc in shuffling

* remove hardcode

* remove hardcode

* refactor

* fix build

* adding gemm+reduce

* adding gemm+reduce

* adding gemm+reduce

* adding gemm+reduce

* adding gemm+reduce

* format

* clean

* adding gemm+reduce

* adding profiler for gemm+reduce

* adding gemm+reduce profiler

* fix build

* clean up

* gemm+reduce

* fix build

* update DeviceGemm_Xdl_CShuffle; update enum to enum class

* clean up

* add test for gemm+reduce

* clean up

* refactor

* fix build

* fix build
This commit is contained in:
Chao Liu
2022-03-23 22:18:42 -05:00
committed by GitHub
parent f91579aab6
commit f95267f166
56 changed files with 4429 additions and 297 deletions

View File

@@ -0,0 +1,52 @@
#include <algorithm>
#include <cstdlib>
#include <half.hpp>
#include <iostream>
#include <numeric>
#include <tuple>
#include <vector>
#include "profile_gemm_reduce_impl.hpp"
int main()
{
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
int M = 512;
int N = 256;
int K = 128;
bool pass = true;
pass = pass &&
ck::profiler::
profile_gemm_reduce_impl<ck::half_t, ck::half_t, ck::half_t, float, Row, Row, Row>(
true, 1, false, 1, M, N, K, K, N, N);
pass = pass &&
ck::profiler::
profile_gemm_reduce_impl<ck::half_t, ck::half_t, ck::half_t, float, Row, Col, Row>(
true, 1, false, 1, M, N, K, K, K, N);
pass = pass &&
ck::profiler::
profile_gemm_reduce_impl<ck::half_t, ck::half_t, ck::half_t, float, Col, Row, Row>(
true, 1, false, 1, M, N, K, M, N, N);
pass = pass &&
ck::profiler::
profile_gemm_reduce_impl<ck::half_t, ck::half_t, ck::half_t, float, Col, Col, Row>(
true, 1, false, 1, M, N, K, M, K, N);
if(pass)
{
std::cout << "test GEMM+Reduce fp16: Pass" << std::endl;
return 0;
}
else
{
std::cout << "test GEMM+Reduce fp16: Fail" << std::endl;
return -1;
}
}