Skip to content

GitLab

"example/36_permute/common.hpp" did not exist on "7ebb1cbfdf9c5946dbcb16124746b7575a705b1d"
Commit 7f65ac05 authored by Jun Liu's avatar Jun Liu
Browse files

Merge branch 'develop' into amd-develop

parents 687d2b7e 7e5c81fe
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 3853 additions and 1716 deletions
include/ck/tensor_operation/gpu/element/combined_element_wise_operation.hpp 0 → 100644
View file @ 7f65ac05
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck/utility/data_type.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
namespace ck {
namespace tensor_operation {
namespace element_wise {
// y = UnaryOp0(UnaryOp1(...(x)))
template <typename... UnaryOpsSet>
struct UnaryCombinedOp
{
__host__ __device__ UnaryCombinedOp(UnaryOpsSet... unary_ops) : unary_ops_(unary_ops...) {}
template <typename Y, typename X>
__host__ __device__ void operator()(Y& y, const X& x) const
{
// Execute first unary op to copy data to y
unary_ops_.At(Number<0>{})(y, x);
static_for<1, Tuple<UnaryOpsSet...>::Size(), 1>{}([&](auto i) { unary_ops_.At(i)(y, y); });
};
Tuple<UnaryOpsSet...> unary_ops_;
};
// y = BinaryOp(UnaryOp0(x0), UnaryOp1(x1))
template <typename BinaryOp, typename UnaryOp0, typename UnaryOp1>
struct BinaryWithUnaryCombinedOp
{
__host__ __device__ BinaryWithUnaryCombinedOp(BinaryOp binary_op,
UnaryOp0 unary_op0,
UnaryOp1 unary_op1)
: binary_op_(binary_op), unary_op0_(unary_op0), unary_op1_(unary_op1)
{
}
template <typename Y, typename X0, typename X1>
__host__ __device__ void operator()(Y& y, const X0& x0, const X1& x1) const
{
Y unary_x0_tmp_result;
Y unary_x1_tmp_result;
unary_op0_(unary_x0_tmp_result, x0);
unary_op1_(unary_x1_tmp_result, x1);
binary_op_(y, unary_x0_tmp_result, unary_x1_tmp_result);
};
private:
BinaryOp binary_op_;
UnaryOp0 unary_op0_;
UnaryOp1 unary_op1_;
};
// y = BinaryOp0(BinaryOp1(UnaryOp0(x0), UnaryOp1(x1)), UnaryOp2(x2))
template <typename BinaryOp0,
typename BinaryOp1,
typename UnaryOp0,
typename UnaryOp1,
typename UnaryOp2>
struct TrinaryWithUnaryCombinedOp
{
__host__ __device__ TrinaryWithUnaryCombinedOp(BinaryOp0 binary_op0,
BinaryOp0 binary_op1,
UnaryOp0 unary_op0,
UnaryOp1 unary_op1,
UnaryOp2 unary_op2)
: binary_op0_(binary_op0),
binary_op1_(binary_op1),
unary_op0_(unary_op0),
unary_op1_(unary_op1),
unary_op2_(unary_op2)
{
}
template <typename Y, typename X0, typename X1, typename X2>
__host__ __device__ void operator()(Y& y, const X0& x0, const X1& x1, const X2& x2) const
{
Y unary_x0_tmp_result;
Y unary_x1_tmp_result;
Y unary_x2_tmp_result;
unary_op0_(unary_x0_tmp_result, x0);
unary_op1_(unary_x1_tmp_result, x1);
unary_op2_(unary_x2_tmp_result, x2);
binary_op0_(unary_x0_tmp_result, unary_x0_tmp_result, unary_x1_tmp_result);
binary_op1_(y, unary_x0_tmp_result, unary_x2_tmp_result);
};
private:
BinaryOp0 binary_op0_{};
BinaryOp1 binary_op1_{};
UnaryOp0 unary_op0_{};
UnaryOp1 unary_op1_{};
UnaryOp2 unary_op2_{};
};
} // namespace element_wise
} // namespace tensor_operation
} // namespace ck
include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
View file @ 7f65ac05
......@@ -12,10 +12,6 @@ namespace ck {
namespace tensor_operation {
namespace element_wise {
#if CK_WORKAROUND_SWDEV_383542
extern "C" __device__ float __ocml_native_recip_f32(float);
#endif
struct PassThroughPack2
{
template <typename Y, typename X>
......@@ -449,11 +445,7 @@ struct FastGelu
const float u = x * (c1 * x * x + c2);
const float emu = __expf(u);
#if !CK_WORKAROUND_SWDEV_383542
y = x * __frcp_rn(1.f + emu);
#else
y = x * __ocml_native_recip_f32(1.f + emu);
#endif
y = x * ck::math::rcp(1.f + emu);
}
template <>
......@@ -559,6 +551,244 @@ struct TanH
};
};
struct ACos
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::acos(x);
};
};
struct Neg
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::neg(x);
};
};
struct ATan
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::atan(x);
};
};
struct Sin
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::sin(x);
};
};
struct ASinH
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::asinh(x);
};
};
struct Cos
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::cos(x);
};
};
struct ACosH
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::acosh(x);
};
};
struct Tan
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::tan(x);
};
};
struct ATanH
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::atanh(x);
};
};
struct SinH
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::sinh(x);
};
};
struct Ceil
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::ceil(x);
};
};
struct Exp
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::exp(x);
};
};
struct CosH
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::cosh(x);
};
};
struct Floor
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::floor(x);
};
};
struct Log
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::log(x);
};
};
struct ASin
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::asin(x);
};
};
struct Rcp
{
template <typename T>
__host__ __device__ void operator()(T& y, const T& x) const
{
static_assert(is_same<T, float>::value || is_same<T, double>::value ||
is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
is_same<T, int32_t>::value,
"Data type is not supported by this operation!");
y = ck::math::rcp(x);
};
};
struct Swish
{
Swish(float beta = 1.0f) : beta_(beta) {}
......
include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
View file @ 7f65ac05
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
......@@ -73,7 +73,7 @@ template <typename ADataType,
index_t CDEShuffleBlockTransferScalarPerVector_NPerBlock,
LoopScheduler LoopSched,
PipelineVersion PipelineVer = PipelineVersion::v1,
typename BComputeDataType = AComputeDataType_>
typename BComputeDataType_ = AComputeDataType_>
struct GridwiseGemmMultipleD_xdl_cshuffle
{
static constexpr index_t NumDTensor = DsDataType::Size();
......@@ -103,8 +103,11 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
#if CK_WORKAROUND_DENORM_FIX
using AComputeDataType =
conditional_t<is_same_v<AComputeDataType_, ck::half_t>, ck::bhalf_t, AComputeDataType_>;
using BComputeDataType =
conditional_t<is_same_v<BComputeDataType_, ck::half_t>, ck::bhalf_t, BComputeDataType_>;
#else
using AComputeDataType = AComputeDataType_;
using BComputeDataType = BComputeDataType_;
#endif
__host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
......
include/ck/utility/math_v2.hpp
View file @ 7f65ac05
This diff is collapsed.
library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
View file @ 7f65ac05
......@@ -16,7 +16,7 @@ namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
#ifdef CK_ENABLE_FP16
#if defined(CK_ENABLE_FP16) && defined(CK_USE_XDL)
void add_device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances(
std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
Row,
......@@ -69,7 +69,7 @@ void add_device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance
PassThrough,
Bilinear>>>& instances);
#endif
#ifdef CK_ENABLE_INT8
#if defined(CK_ENABLE_INT8) && defined(CK_USE_WMMA)
void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances(
std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
Row,
......@@ -159,7 +159,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
static auto GetInstances()
{
std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
#ifdef CK_ENABLE_FP16
#if defined(CK_ENABLE_FP16) && defined(CK_USE_XDL)
if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
is_same_v<DDataType, half_t> && is_same_v<EDataType, half_t>)
{
......@@ -189,7 +189,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
}
}
#endif
#ifdef CK_ENABLE_INT8
#if defined(CK_ENABLE_INT8) && defined(CK_USE_WMMA)
if constexpr(is_same_v<ADataType, std::int8_t> && is_same_v<BDataType, std::int8_t> &&
is_same_v<DDataType, std::int8_t> && is_same_v<EDataType, std::int8_t>)
{
......
library/include/ck/library/tensor_operation_instance/gpu/gemm_wmma.inc 0 → 100644
View file @ 7f65ac05
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_gemm_wmma_f16_f16_f16_km_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_wmma_f16_f16_f16_km_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_wmma_f16_f16_f16_mk_kn_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
void add_device_gemm_wmma_f16_f16_f16_mk_nk_mn_instances(
std::vector<std::unique_ptr<
DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
instances);
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
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