Skip to content

GitLab

Commit 44a12304 authored by turneram's avatar turneram
Browse files

Rough draft working

parent cfbe4da6
Hide whitespace changes
Inline Side-by-side
Showing with 43 additions and 69 deletions
src/targets/gpu/kernels/include/migraphx/kernels/ck_elementwise.hpp
View file @ 44a12304
...@@ -27,6 +27,7 @@ ...@@ -27,6 +27,7 @@
#include <migraphx/kernels/index.hpp> #include <migraphx/kernels/index.hpp>
#include <migraphx/kernels/algorithm.hpp> #include <migraphx/kernels/algorithm.hpp>
#include <migraphx/kernels/integral_constant.hpp> #include <migraphx/kernels/integral_constant.hpp>
#include <migraphx/kernels/tensor_view.hpp>
#include "ck/device_utility/device_prop.hpp" #include "ck/device_utility/device_prop.hpp"
#include "ck/device_utility/kernel_launch.hpp" #include "ck/device_utility/kernel_launch.hpp"
...@@ -42,20 +43,32 @@ using CDataType = float; ...@@ -42,20 +43,32 @@ using CDataType = float;
using ElementwiseFunctor = float; using ElementwiseFunctor = float;
static constexpr auto I0 = ck::Number<0>{}; static constexpr auto I0 = ck::Number<0>{};
using index_t = index_int;
template <class L, class S> template <class L, class S, class N>
__host__ __device__ constexpr auto MakeDescriptor_M(const L& lengths, const S& strides) constexpr auto MakeDescriptor_M(const L& lengths, const S& strides, const N& /* ndim */)
{ {
auto idx = make_index(); auto gridSize = 72;
auto blockSize = 1024;
constexpr auto ndim = 1;
//auto idx = make_index();
auto tupleOfShape = generate_tuple([&](auto I) { return static_cast<ck::index_t>(lengths[I]); }, auto tupleOfShape = generate_tuple([&](auto I) { return static_cast<ck::index_t>(lengths[I]); },
ck::Number<1>{}); ck::Number<ndim>{});
auto tupleOfStride = generate_tuple( auto tupleOfStride = generate_tuple(
[&](auto I) { return static_cast<ck::index_t>(strides[I]); }, ck::Number<1>{}); [&](auto I) { return static_cast<ck::index_t>(strides[I]); }, ck::Number<1>{});
const auto desc_m = make_naive_tensor_descriptor(tupleOfShape, tupleOfStride); const auto desc = make_naive_tensor_descriptor(tupleOfShape, tupleOfStride);
auto desc_m = desc;
// merge nd to 1d desc - [s0 * s1 * ...]
if constexpr(ndim > 1)
{
desc_m = transform_tensor_descriptor(
desc,
make_tuple(make_merge_transform(tupleOfShape)),
make_tuple(generate_sequence_v2([&](auto I) { return I; }, ck::Number<ndim>{})),
make_tuple(ck::Sequence<0>{}));
}
const auto M = desc_m.GetLength(I0); const auto M = desc_m.GetLength(I0);
const index_t loop_step = idx.nglobal(); // gridSize * blockSize * MPerThread; const ck::index_t loop_step = /* idx.nglobal(); // */ gridSize * blockSize/* * MPerThread */;
const auto pad = ck::math::integer_least_multiple(M, loop_step) - M; const auto pad = ck::math::integer_least_multiple(M, loop_step) - M;
const auto desc_m_pad = const auto desc_m_pad =
transform_tensor_descriptor(desc_m, transform_tensor_descriptor(desc_m,
...@@ -68,18 +81,7 @@ __host__ __device__ constexpr auto MakeDescriptor_M(const L& lengths, const S& s ...@@ -68,18 +81,7 @@ __host__ __device__ constexpr auto MakeDescriptor_M(const L& lengths, const S& s
struct Add struct Add
{ {
template <typename Y, typename X0, typename X1> template <typename Y, typename X0, typename X1>
__host__ __device__ constexpr void operator()(Y& y, const X0& x0, const X1& x1) const; __device__ constexpr void operator()(Y& y, const X0& x0, const X1& x1) const
template <>
__host__ __device__ constexpr void
operator()<float>(float& y, const float& x0, const float& x1) const
{
y = x0 + x1;
};
template <>
__host__ __device__ constexpr void
operator()<double>(double& y, const double& x0, const double& x1) const
{ {
y = x0 + x1; y = x0 + x1;
}; };
...@@ -88,57 +90,29 @@ struct Add ...@@ -88,57 +90,29 @@ struct Add
template <class T, class U, class V> template <class T, class U, class V>
__device__ void ck_elementwise(const T& a_t, const U& b_t, const V& c_t) __device__ void ck_elementwise(const T& a_t, const U& b_t, const V& c_t)
{ {
// auto add = [](auto a, auto b) { return a + b; }; auto idx = make_index();
auto lengths = a_t.get_shape().lens; if (idx.global == 0)
auto strides = a_t.get_shape().strides; {
auto a_desc = MakeDescriptor_M(lengths, strides); constexpr auto lengths = get_shape_c<T>{}.lens;
constexpr auto strides = get_shape_c<T>{}.strides;
using AGridDesc_M = decltype(a_desc); constexpr auto a_desc = MakeDescriptor_M(lengths, strides, 1);
// using Add = ck::tensor_operation::element_wise::Add;
using GridwiseBinEltwise = ck::GridwiseBinaryElementwise_1D<ADataType,
BDataType,
CDataType,
CDataType,
AGridDesc_M,
AGridDesc_M,
AGridDesc_M,
Add,
8,
8,
8,
8>;
auto op = Add{};
GridwiseBinEltwise::Run(a_t.data(), b_t.data(), c_t.data(), a_desc, a_desc, a_desc, op);
// auto kernel = ck::kernel_binary_elementwise_1d<GridwiseBinEltwise,
// ADataType,
// BDataType,
// CDataType,
// AGridDesc_M,
// AGridDesc_M,
// AGridDesc_M,
// Add>;
// kernel(a_t.data(), b_t.data(), c_t.data(), a_desc, a_desc, a_desc, Add);
// Argument arg{a_t.data(), b_t.data(), c_t.data(), c_t.get_shape().lens,
// a_t.get_shape().strides, b_t.get_shape().strides, c_t.get_shape().strides,
// add};
// auto lengths = a_t.get_shape().lens;
// auto strides = a_t.get_shape().strides;
// auto idx = make_index();
// b_t.get_shape();
// c_t.get_shape();
// auto tupleOfShape = generate_tuple([&](auto I) { return lengths[I]; }, ck::Number<1>{});
// auto tupleOfStride = generate_tuple([&](auto I) { return strides[I]; }, ck::Number<1>{});
// const auto desc_m = make_naive_tensor_descriptor(tupleOfShape, tupleOfStride);
// const auto M = desc_m.GetLength(I0); using AGridDesc_M = decltype(a_desc);
// const ck::index_t loop_step = idx.nglobal();//gridSize * blockSize * MPerThread; using GridwiseBinEltwise = ck::GridwiseBinaryElementwise_1D<ADataType,
// const auto pad = ck::math::integer_least_multiple(M, loop_step) - M; BDataType,
// const auto desc_m_pad = CDataType,
// transform_tensor_descriptor(desc_m, CDataType,
// make_tuple(ck::make_right_pad_transform(M, pad)), AGridDesc_M,
// make_tuple(ck::Sequence<0>{}), AGridDesc_M,
// make_tuple(ck::Sequence<0>{})); AGridDesc_M,
Add,
1,
1,
1,
1>;
auto op = Add{};
GridwiseBinEltwise::Run(a_t.data(), b_t.data(), c_t.data(), a_desc, a_desc, a_desc, op);
}
} }
} // namespace migraphx } // namespace migraphx
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment