ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-17 19:40:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

Resolution of issue #153: Add compiler warning on comparing int and size_t (#212)

Browse Source 
* Turning compare warnings on

* Cleaning part I

* Cleaning part II

* Explicit static_cast to ck::type_convert

* Resolving large tensor size issue.

* format

* revert change to tensor descriptor; promote lementSpaceSize to 64bit

* use integer value for GEMM test

* Review remarks

* Review remarks + issues with (un)signed arithmetic

* Format fix

* Format

* Clang-format.

* fix 2gb limit issue

Co-authored-by: Chao Liu <chao.liu2@amd.com>
Co-authored-by: Adam Osewski <aosewski@amd.com>

[ROCm/composable_kernel commit: f03a1738d9]
This commit is contained in:
myamlak
2022-05-09 22:06:49 +02:00
committed by GitHub
parent 0bde7aaef1
commit 060938c9a5
30 changed files with 261 additions and 165 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
include/ck/tensor_description/tensor_descriptor_helper.hpp
View File

@@ -1,6 +1,4 @@
#ifndef CK_TENSOR_DESCRIPTOR_HELPER_HPP
#define CK_TENSOR_DESCRIPTOR_HELPER_HPP
#pragma once
#include "common_header.hpp"
#include "tensor_descriptor.hpp"
#include "multi_index_transform_helper.hpp"
@@ -35,6 +33,12 @@ __host__ __device__ constexpr auto calculate_element_space_size_impl(const Lengt
}
#endif
// Lengths..., Strides... could be:
// 1) index_t, which is known at run-time, or
// 2) Number<>, which is known at compile-time
// element_space_size could be:
// 1) long_index_t, or
// 2) LongNumber<>
template <typename... Lengths,
typename... Strides,
typename enable_if<sizeof...(Lengths) == sizeof...(Strides), bool>::type = false>
@@ -68,10 +72,10 @@ __host__ __device__ constexpr auto make_naive_tensor_descriptor(const Tuple<Leng
}
};
const auto element_space_size = f(f, Number<0>{}, Number<1>{});
const auto element_space_size = f(f, Number<0>{}, LongNumber<1>{});
#else
const auto element_space_size =
calculate_element_space_size_impl(lengths, strides, Number<0>{}, Number<1>{});
calculate_element_space_size_impl(lengths, strides, Number<0>{}, LongNumber<1>{});
#endif
return TensorDescriptor<remove_cv_t<decltype(transforms)>,
@@ -82,9 +86,12 @@ __host__ __device__ constexpr auto make_naive_tensor_descriptor(const Tuple<Leng
element_space_size};
}
// Lengths... can be:
// 1) index_t, which is known at run-time
// Lengths... could be:
// 1) index_t, which is known at run-time, or
// 2) Number<>, which is known at compile-time
// element_space_size could be:
// 1) long_index_t, or
// 2) LongNumber<>
template <typename... Lengths>
__host__ __device__ constexpr auto
make_naive_tensor_descriptor_packed(const Tuple<Lengths...>& lengths)
@@ -100,7 +107,7 @@ make_naive_tensor_descriptor_packed(const Tuple<Lengths...>& lengths)
constexpr auto visible_dim_hidden_ids = typename arithmetic_sequence_gen<1, N + 1, 1>::type{};
const auto element_space_size = container_reduce(lengths, math::multiplies{}, Number<1>{});
const auto element_space_size = container_reduce(lengths, math::multiplies{}, LongNumber<1>{});
return TensorDescriptor<remove_cv_t<decltype(transforms)>,
remove_cv_t<decltype(low_dim_hidden_idss)>,
@@ -110,6 +117,12 @@ make_naive_tensor_descriptor_packed(const Tuple<Lengths...>& lengths)
element_space_size};
}
// Lengths... could be:
// 1) index_t, which is known at run-time, or
// 2) Number<>, which is known at compile-time
// align could be:
// 1) index_t, or
// 2) Number<>
template <typename... Lengths, typename Align>
__host__ __device__ constexpr auto
make_naive_tensor_descriptor_aligned(const Tuple<Lengths...>& lengths, Align align)
@@ -146,4 +159,3 @@ make_naive_tensor_descriptor_aligned(const Tuple<Lengths...>& lengths, Align ali
}
} // namespace ck
#endif