Add Gemm instances for performance improvement (#1018)

* improve kpad

* more tuning parameters

* f16_f8_fp16

* cut test time

* add f16_f8_fp16

* add f16_f8_f16

* testing instances for skinny cases

* format

* clean

* add fp16_f8_fp16

* clang-format

* add grouped gemm instalces

* fixed profile grouped_gemm

* clean

* clean

* clean

* clean

* clean

* add missing instance func

* fixed inferface

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
Co-authored-by: root <root@sh5-1e707-rc06-38.mkm.dcgpu>
This commit is contained in:
zjing14
2023-11-07 09:09:58 -06:00
committed by GitHub
parent aa0b979887
commit 98fd41f597
28 changed files with 1346 additions and 338 deletions

View File

@@ -16,6 +16,57 @@ namespace element_wise {
extern "C" __device__ float __ocml_native_recip_f32(float);
#endif
struct PassThroughPack2
{
template <typename Y, typename X>
__host__ __device__ void operator()(Y& y, const X& x) const;
__host__ __device__ constexpr void operator()(ck::f8x2_t& y, const ck::half2_t& x) const
{
// fake conversion
uint16_t t = ck::bit_cast<uint32_t>(x);
y = ck::bit_cast<ck::f8x2_t>(t);
}
__host__ __device__ constexpr void operator()(ck::half2_t& y, const ck::f8x2_t& x) const
{
auto t = type_convert<float2_t>(x);
y = type_convert<half2_t>(t);
}
__host__ __device__ constexpr void operator()(ck::half2_t& y, const ck::half2_t& x) const
{
y = x;
}
__host__ __device__ constexpr void operator()(ck::f8x2_t& y, const ck::f8x2_t& x) const
{
y = x;
}
__host__ __device__ constexpr void operator()(ck::float2_t& y, const ck::float2_t& x) const
{
y = x;
}
__host__ __device__ constexpr void operator()(ck::int8x2_t& y, const ck::int8x2_t& x) const
{
y = x;
}
__host__ __device__ constexpr void operator()(ck::bhalf2_t& y, const ck::bhalf2_t& x) const
{
y = x;
}
__host__ __device__ constexpr void operator()(ck::double2_t& y, const ck::double2_t& x) const
{
y = x;
}
constexpr const static bool is_pack2_invocable = true;
};
struct PassThrough
{
template <typename Y, typename X>