AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SWDEV-288707 - Merge HIP refactored code to staging

Browse Source 
These changes move-
- rocclr based implementation inside src/hipamd/src
- platform specific (both nvidia and amd) headers inside
src/hipamd/include/hip

Change-Id: Ia29791a727244952591fe1d813dcef0303b73a9e
This commit is contained in:
Rahul Garg
2021-05-28 23:15:18 +00:00
parent 792415f960
commit ce76ec8b88
119 changed files with 3943 additions and 97 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/hip/amd_detail/channel_descriptor.h
-348
View File
@@ -1,348 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_CHANNEL_DESCRIPTOR_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_CHANNEL_DESCRIPTOR_H
#include <hip/hip_common.h>
#include <hip/amd_detail/driver_types.h>
#include <hip/amd_detail/hip_vector_types.h>
#ifdef __cplusplus
extern "C" HIP_PUBLIC_API
hipChannelFormatDesc hipCreateChannelDesc(int x, int y, int z, int w, hipChannelFormatKind f);
static inline hipChannelFormatDesc hipCreateChannelDescHalf() {
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindFloat);
}
static inline hipChannelFormatDesc hipCreateChannelDescHalf1() {
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindFloat);
}
static inline hipChannelFormatDesc hipCreateChannelDescHalf2() {
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindFloat);
}
template <typename T>
static inline hipChannelFormatDesc hipCreateChannelDesc() {
return hipCreateChannelDesc(0, 0, 0, 0, hipChannelFormatKindNone);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<char>() {
int e = (int)sizeof(char) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<signed char>() {
int e = (int)sizeof(signed char) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<unsigned char>() {
int e = (int)sizeof(unsigned char) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<uchar1>() {
int e = (int)sizeof(unsigned char) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<char1>() {
int e = (int)sizeof(signed char) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<uchar2>() {
int e = (int)sizeof(unsigned char) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<char2>() {
int e = (int)sizeof(signed char) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindSigned);
}
#ifndef __GNUC__ // vector3 is the same as vector4
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<uchar3>() {
int e = (int)sizeof(unsigned char) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<char3>() {
int e = (int)sizeof(signed char) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindSigned);
}
#endif
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<uchar4>() {
int e = (int)sizeof(unsigned char) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<char4>() {
int e = (int)sizeof(signed char) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<unsigned short>() {
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<signed short>() {
int e = (int)sizeof(signed short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<ushort1>() {
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<short1>() {
int e = (int)sizeof(signed short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<ushort2>() {
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<short2>() {
int e = (int)sizeof(signed short) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindSigned);
}
#ifndef __GNUC__
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<ushort3>() {
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<short3>() {
int e = (int)sizeof(signed short) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindSigned);
}
#endif
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<ushort4>() {
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<short4>() {
int e = (int)sizeof(signed short) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<unsigned int>() {
int e = (int)sizeof(unsigned int) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<signed int>() {
int e = (int)sizeof(signed int) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<uint1>() {
int e = (int)sizeof(unsigned int) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<int1>() {
int e = (int)sizeof(signed int) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<uint2>() {
int e = (int)sizeof(unsigned int) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<int2>() {
int e = (int)sizeof(signed int) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindSigned);
}
#ifndef __GNUC__
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<uint3>() {
int e = (int)sizeof(unsigned int) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<int3>() {
int e = (int)sizeof(signed int) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindSigned);
}
#endif
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<uint4>() {
int e = (int)sizeof(unsigned int) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<int4>() {
int e = (int)sizeof(signed int) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<float>() {
int e = (int)sizeof(float) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindFloat);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<float1>() {
int e = (int)sizeof(float) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindFloat);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<float2>() {
int e = (int)sizeof(float) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindFloat);
}
#ifndef __GNUC__
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<float3>() {
int e = (int)sizeof(float) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindFloat);
}
#endif
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<float4>() {
int e = (int)sizeof(float) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindFloat);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<unsigned long>() {
int e = (int)sizeof(unsigned long) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<signed long>() {
int e = (int)sizeof(signed long) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<ulong1>() {
int e = (int)sizeof(unsigned long) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<long1>() {
int e = (int)sizeof(signed long) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<ulong2>() {
int e = (int)sizeof(unsigned long) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<long2>() {
int e = (int)sizeof(signed long) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindSigned);
}
#ifndef __GNUC__
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<ulong3>() {
int e = (int)sizeof(unsigned long) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<long3>() {
int e = (int)sizeof(signed long) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindSigned);
}
#endif
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<ulong4>() {
int e = (int)sizeof(unsigned long) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindUnsigned);
}
template <>
inline hipChannelFormatDesc hipCreateChannelDesc<long4>() {
int e = (int)sizeof(signed long) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindSigned);
}
#else
struct hipChannelFormatDesc hipCreateChannelDesc(int x, int y, int z, int w,
enum hipChannelFormatKind f);
#endif
#endif
include/hip/amd_detail/concepts.hpp
-30
View File
@@ -1,30 +0,0 @@
/*
Copyright (c) 2015-present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
namespace hip_impl // Documentation only.
{
#define requires(...)
#define FunctionalProcedure typename
} // namespace hip_impl
include/hip/amd_detail/cuda/cuda.h
-1
View File
@@ -1 +0,0 @@
include/hip/amd_detail/cuda/math_functions.h
-1
View File
@@ -1 +0,0 @@
include/hip/amd_detail/device_functions.h
-1369
View File
File diff suppressed because it is too large Load Diff
include/hip/amd_detail/device_library_decls.h
-118
View File
@@ -1,118 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/**
* @file amd_detail/device_library_decls.h
* @brief Contains declarations for types and functions in device library.
* Uses int64_t and uint64_t instead of long, long long, unsigned
* long and unsigned long long types for device library API
* declarations.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_DEVICE_LIBRARY_DECLS_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_DEVICE_LIBRARY_DECLS_H
#include "hip/amd_detail/host_defines.h"
typedef unsigned char uchar;
typedef unsigned short ushort;
typedef unsigned int uint;
typedef unsigned long ulong;
typedef unsigned long long ullong;
extern "C" __device__ __attribute__((const)) bool __ockl_wfany_i32(int);
extern "C" __device__ __attribute__((const)) bool __ockl_wfall_i32(int);
extern "C" __device__ uint __ockl_activelane_u32(void);
extern "C" __device__ __attribute__((const)) uint __ockl_mul24_u32(uint, uint);
extern "C" __device__ __attribute__((const)) int __ockl_mul24_i32(int, int);
extern "C" __device__ __attribute__((const)) uint __ockl_mul_hi_u32(uint, uint);
extern "C" __device__ __attribute__((const)) int __ockl_mul_hi_i32(int, int);
extern "C" __device__ __attribute__((const)) uint __ockl_sadd_u32(uint, uint, uint);
extern "C" __device__ __attribute__((const)) uchar __ockl_clz_u8(uchar);
extern "C" __device__ __attribute__((const)) ushort __ockl_clz_u16(ushort);
extern "C" __device__ __attribute__((const)) uint __ockl_clz_u32(uint);
extern "C" __device__ __attribute__((const)) uint64_t __ockl_clz_u64(uint64_t);
extern "C" __device__ __attribute__((const)) float __ocml_floor_f32(float);
extern "C" __device__ __attribute__((const)) float __ocml_rint_f32(float);
extern "C" __device__ __attribute__((const)) float __ocml_ceil_f32(float);
extern "C" __device__ __attribute__((const)) float __ocml_trunc_f32(float);
extern "C" __device__ __attribute__((const)) float __ocml_fmin_f32(float, float);
extern "C" __device__ __attribute__((const)) float __ocml_fmax_f32(float, float);
extern "C" __device__ __attribute__((convergent)) void __ockl_gws_init(uint nwm1, uint rid);
extern "C" __device__ __attribute__((convergent)) void __ockl_gws_barrier(uint nwm1, uint rid);
extern "C" __device__ __attribute__((const)) uint32_t __ockl_lane_u32();
extern "C" __device__ __attribute__((const)) int __ockl_grid_is_valid(void);
extern "C" __device__ __attribute__((convergent)) void __ockl_grid_sync(void);
extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_num_grids(void);
extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_grid_rank(void);
extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_size(void);
extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_thread_rank(void);
extern "C" __device__ __attribute__((const)) int __ockl_multi_grid_is_valid(void);
extern "C" __device__ __attribute__((convergent)) void __ockl_multi_grid_sync(void);
extern "C" __device__ void __ockl_atomic_add_noret_f32(float*, float);
extern "C" __device__ __attribute__((convergent)) int __ockl_wgred_add_i32(int a);
extern "C" __device__ __attribute__((convergent)) int __ockl_wgred_and_i32(int a);
extern "C" __device__ __attribute__((convergent)) int __ockl_wgred_or_i32(int a);
// Introduce local address space
#define __local __attribute__((address_space(3)))
#ifdef __HIP_DEVICE_COMPILE__
__device__ inline static __local void* __to_local(unsigned x) { return (__local void*)x; }
#endif //__HIP_DEVICE_COMPILE__
// Using hip.amdgcn.bc - sync threads
#define __CLK_LOCAL_MEM_FENCE 0x01
typedef unsigned __cl_mem_fence_flags;
typedef enum __memory_scope {
__memory_scope_work_item = __OPENCL_MEMORY_SCOPE_WORK_ITEM,
__memory_scope_work_group = __OPENCL_MEMORY_SCOPE_WORK_GROUP,
__memory_scope_device = __OPENCL_MEMORY_SCOPE_DEVICE,
__memory_scope_all_svm_devices = __OPENCL_MEMORY_SCOPE_ALL_SVM_DEVICES,
__memory_scope_sub_group = __OPENCL_MEMORY_SCOPE_SUB_GROUP
} __memory_scope;
// enum values aligned with what clang uses in EmitAtomicExpr()
typedef enum __memory_order
{
__memory_order_relaxed = __ATOMIC_RELAXED,
__memory_order_acquire = __ATOMIC_ACQUIRE,
__memory_order_release = __ATOMIC_RELEASE,
__memory_order_acq_rel = __ATOMIC_ACQ_REL,
__memory_order_seq_cst = __ATOMIC_SEQ_CST
} __memory_order;
// Linked from hip.amdgcn.bc
extern "C" __device__ void
__atomic_work_item_fence(__cl_mem_fence_flags, __memory_order, __memory_scope);
#endif
include/hip/amd_detail/driver_types.h
-478
View File
@@ -1,478 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_DRIVER_TYPES_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_DRIVER_TYPES_H
// The follow macro should be removed after upstream updation.
// It's defined here for workarround of rocThrust building failure.
#define HIP_INCLUDE_HIP_HCC_DETAIL_DRIVER_TYPES_H
#if !defined(__HIPCC_RTC__)
#ifndef __cplusplus
#include <stdbool.h>
#endif
#endif // !defined(__HIPCC_RTC__)
typedef void* hipDeviceptr_t;
typedef enum hipChannelFormatKind {
hipChannelFormatKindSigned = 0,
hipChannelFormatKindUnsigned = 1,
hipChannelFormatKindFloat = 2,
hipChannelFormatKindNone = 3
}hipChannelFormatKind;
typedef struct hipChannelFormatDesc {
int x;
int y;
int z;
int w;
enum hipChannelFormatKind f;
}hipChannelFormatDesc;
#define HIP_TRSA_OVERRIDE_FORMAT 0x01
#define HIP_TRSF_READ_AS_INTEGER 0x01
#define HIP_TRSF_NORMALIZED_COORDINATES 0x02
#define HIP_TRSF_SRGB 0x10
typedef enum hipArray_Format {
HIP_AD_FORMAT_UNSIGNED_INT8 = 0x01,
HIP_AD_FORMAT_UNSIGNED_INT16 = 0x02,
HIP_AD_FORMAT_UNSIGNED_INT32 = 0x03,
HIP_AD_FORMAT_SIGNED_INT8 = 0x08,
HIP_AD_FORMAT_SIGNED_INT16 = 0x09,
HIP_AD_FORMAT_SIGNED_INT32 = 0x0a,
HIP_AD_FORMAT_HALF = 0x10,
HIP_AD_FORMAT_FLOAT = 0x20
}hipArray_Format;
typedef struct HIP_ARRAY_DESCRIPTOR {
size_t Width;
size_t Height;
enum hipArray_Format Format;
unsigned int NumChannels;
}HIP_ARRAY_DESCRIPTOR;
typedef struct HIP_ARRAY3D_DESCRIPTOR {
size_t Width;
size_t Height;
size_t Depth;
enum hipArray_Format Format;
unsigned int NumChannels;
unsigned int Flags;
}HIP_ARRAY3D_DESCRIPTOR;
typedef struct hipArray {
void* data; // FIXME: generalize this
struct hipChannelFormatDesc desc;
unsigned int type;
unsigned int width;
unsigned int height;
unsigned int depth;
enum hipArray_Format Format;
unsigned int NumChannels;
bool isDrv;
unsigned int textureType;
}hipArray;
#if !defined(__HIPCC_RTC__)
typedef struct hip_Memcpy2D {
size_t srcXInBytes;
size_t srcY;
hipMemoryType srcMemoryType;
const void* srcHost;
hipDeviceptr_t srcDevice;
hipArray* srcArray;
size_t srcPitch;
size_t dstXInBytes;
size_t dstY;
hipMemoryType dstMemoryType;
void* dstHost;
hipDeviceptr_t dstDevice;
hipArray* dstArray;
size_t dstPitch;
size_t WidthInBytes;
size_t Height;
} hip_Memcpy2D;
#endif // !defined(__HIPCC_RTC__)
typedef struct hipArray* hipArray_t;
typedef hipArray_t hiparray;
typedef const struct hipArray* hipArray_const_t;
typedef struct hipMipmappedArray {
void* data;
struct hipChannelFormatDesc desc;
unsigned int type;
unsigned int width;
unsigned int height;
unsigned int depth;
unsigned int min_mipmap_level;
unsigned int max_mipmap_level;
unsigned int flags;
enum hipArray_Format format;
} hipMipmappedArray;
typedef struct hipMipmappedArray* hipMipmappedArray_t;
typedef const struct hipMipmappedArray* hipMipmappedArray_const_t;
/**
* hip resource types
*/
typedef enum hipResourceType {
hipResourceTypeArray = 0x00,
hipResourceTypeMipmappedArray = 0x01,
hipResourceTypeLinear = 0x02,
hipResourceTypePitch2D = 0x03
}hipResourceType;
typedef enum HIPresourcetype_enum {
HIP_RESOURCE_TYPE_ARRAY = 0x00, /**< Array resoure */
HIP_RESOURCE_TYPE_MIPMAPPED_ARRAY = 0x01, /**< Mipmapped array resource */
HIP_RESOURCE_TYPE_LINEAR = 0x02, /**< Linear resource */
HIP_RESOURCE_TYPE_PITCH2D = 0x03 /**< Pitch 2D resource */
} HIPresourcetype;
/**
* hip address modes
*/
typedef enum HIPaddress_mode_enum {
HIP_TR_ADDRESS_MODE_WRAP = 0,
HIP_TR_ADDRESS_MODE_CLAMP = 1,
HIP_TR_ADDRESS_MODE_MIRROR = 2,
HIP_TR_ADDRESS_MODE_BORDER = 3
} HIPaddress_mode;
/**
* hip filter modes
*/
typedef enum HIPfilter_mode_enum {
HIP_TR_FILTER_MODE_POINT = 0,
HIP_TR_FILTER_MODE_LINEAR = 1
} HIPfilter_mode;
/**
* Texture descriptor
*/
typedef struct HIP_TEXTURE_DESC_st {
HIPaddress_mode addressMode[3]; /**< Address modes */
HIPfilter_mode filterMode; /**< Filter mode */
unsigned int flags; /**< Flags */
unsigned int maxAnisotropy; /**< Maximum anisotropy ratio */
HIPfilter_mode mipmapFilterMode; /**< Mipmap filter mode */
float mipmapLevelBias; /**< Mipmap level bias */
float minMipmapLevelClamp; /**< Mipmap minimum level clamp */
float maxMipmapLevelClamp; /**< Mipmap maximum level clamp */
float borderColor[4]; /**< Border Color */
int reserved[12];
} HIP_TEXTURE_DESC;
/**
* hip texture resource view formats
*/
typedef enum hipResourceViewFormat {
hipResViewFormatNone = 0x00,
hipResViewFormatUnsignedChar1 = 0x01,
hipResViewFormatUnsignedChar2 = 0x02,
hipResViewFormatUnsignedChar4 = 0x03,
hipResViewFormatSignedChar1 = 0x04,
hipResViewFormatSignedChar2 = 0x05,
hipResViewFormatSignedChar4 = 0x06,
hipResViewFormatUnsignedShort1 = 0x07,
hipResViewFormatUnsignedShort2 = 0x08,
hipResViewFormatUnsignedShort4 = 0x09,
hipResViewFormatSignedShort1 = 0x0a,
hipResViewFormatSignedShort2 = 0x0b,
hipResViewFormatSignedShort4 = 0x0c,
hipResViewFormatUnsignedInt1 = 0x0d,
hipResViewFormatUnsignedInt2 = 0x0e,
hipResViewFormatUnsignedInt4 = 0x0f,
hipResViewFormatSignedInt1 = 0x10,
hipResViewFormatSignedInt2 = 0x11,
hipResViewFormatSignedInt4 = 0x12,
hipResViewFormatHalf1 = 0x13,
hipResViewFormatHalf2 = 0x14,
hipResViewFormatHalf4 = 0x15,
hipResViewFormatFloat1 = 0x16,
hipResViewFormatFloat2 = 0x17,
hipResViewFormatFloat4 = 0x18,
hipResViewFormatUnsignedBlockCompressed1 = 0x19,
hipResViewFormatUnsignedBlockCompressed2 = 0x1a,
hipResViewFormatUnsignedBlockCompressed3 = 0x1b,
hipResViewFormatUnsignedBlockCompressed4 = 0x1c,
hipResViewFormatSignedBlockCompressed4 = 0x1d,
hipResViewFormatUnsignedBlockCompressed5 = 0x1e,
hipResViewFormatSignedBlockCompressed5 = 0x1f,
hipResViewFormatUnsignedBlockCompressed6H = 0x20,
hipResViewFormatSignedBlockCompressed6H = 0x21,
hipResViewFormatUnsignedBlockCompressed7 = 0x22
}hipResourceViewFormat;
typedef enum HIPresourceViewFormat_enum
{
HIP_RES_VIEW_FORMAT_NONE = 0x00, /**< No resource view format (use underlying resource format) */
HIP_RES_VIEW_FORMAT_UINT_1X8 = 0x01, /**< 1 channel unsigned 8-bit integers */
HIP_RES_VIEW_FORMAT_UINT_2X8 = 0x02, /**< 2 channel unsigned 8-bit integers */
HIP_RES_VIEW_FORMAT_UINT_4X8 = 0x03, /**< 4 channel unsigned 8-bit integers */
HIP_RES_VIEW_FORMAT_SINT_1X8 = 0x04, /**< 1 channel signed 8-bit integers */
HIP_RES_VIEW_FORMAT_SINT_2X8 = 0x05, /**< 2 channel signed 8-bit integers */
HIP_RES_VIEW_FORMAT_SINT_4X8 = 0x06, /**< 4 channel signed 8-bit integers */
HIP_RES_VIEW_FORMAT_UINT_1X16 = 0x07, /**< 1 channel unsigned 16-bit integers */
HIP_RES_VIEW_FORMAT_UINT_2X16 = 0x08, /**< 2 channel unsigned 16-bit integers */
HIP_RES_VIEW_FORMAT_UINT_4X16 = 0x09, /**< 4 channel unsigned 16-bit integers */
HIP_RES_VIEW_FORMAT_SINT_1X16 = 0x0a, /**< 1 channel signed 16-bit integers */
HIP_RES_VIEW_FORMAT_SINT_2X16 = 0x0b, /**< 2 channel signed 16-bit integers */
HIP_RES_VIEW_FORMAT_SINT_4X16 = 0x0c, /**< 4 channel signed 16-bit integers */
HIP_RES_VIEW_FORMAT_UINT_1X32 = 0x0d, /**< 1 channel unsigned 32-bit integers */
HIP_RES_VIEW_FORMAT_UINT_2X32 = 0x0e, /**< 2 channel unsigned 32-bit integers */
HIP_RES_VIEW_FORMAT_UINT_4X32 = 0x0f, /**< 4 channel unsigned 32-bit integers */
HIP_RES_VIEW_FORMAT_SINT_1X32 = 0x10, /**< 1 channel signed 32-bit integers */
HIP_RES_VIEW_FORMAT_SINT_2X32 = 0x11, /**< 2 channel signed 32-bit integers */
HIP_RES_VIEW_FORMAT_SINT_4X32 = 0x12, /**< 4 channel signed 32-bit integers */
HIP_RES_VIEW_FORMAT_FLOAT_1X16 = 0x13, /**< 1 channel 16-bit floating point */
HIP_RES_VIEW_FORMAT_FLOAT_2X16 = 0x14, /**< 2 channel 16-bit floating point */
HIP_RES_VIEW_FORMAT_FLOAT_4X16 = 0x15, /**< 4 channel 16-bit floating point */
HIP_RES_VIEW_FORMAT_FLOAT_1X32 = 0x16, /**< 1 channel 32-bit floating point */
HIP_RES_VIEW_FORMAT_FLOAT_2X32 = 0x17, /**< 2 channel 32-bit floating point */
HIP_RES_VIEW_FORMAT_FLOAT_4X32 = 0x18, /**< 4 channel 32-bit floating point */
HIP_RES_VIEW_FORMAT_UNSIGNED_BC1 = 0x19, /**< Block compressed 1 */
HIP_RES_VIEW_FORMAT_UNSIGNED_BC2 = 0x1a, /**< Block compressed 2 */
HIP_RES_VIEW_FORMAT_UNSIGNED_BC3 = 0x1b, /**< Block compressed 3 */
HIP_RES_VIEW_FORMAT_UNSIGNED_BC4 = 0x1c, /**< Block compressed 4 unsigned */
HIP_RES_VIEW_FORMAT_SIGNED_BC4 = 0x1d, /**< Block compressed 4 signed */
HIP_RES_VIEW_FORMAT_UNSIGNED_BC5 = 0x1e, /**< Block compressed 5 unsigned */
HIP_RES_VIEW_FORMAT_SIGNED_BC5 = 0x1f, /**< Block compressed 5 signed */
HIP_RES_VIEW_FORMAT_UNSIGNED_BC6H = 0x20, /**< Block compressed 6 unsigned half-float */
HIP_RES_VIEW_FORMAT_SIGNED_BC6H = 0x21, /**< Block compressed 6 signed half-float */
HIP_RES_VIEW_FORMAT_UNSIGNED_BC7 = 0x22 /**< Block compressed 7 */
} HIPresourceViewFormat;
/**
* HIP resource descriptor
*/
typedef struct hipResourceDesc {
enum hipResourceType resType;
union {
struct {
hipArray_t array;
} array;
struct {
hipMipmappedArray_t mipmap;
} mipmap;
struct {
void* devPtr;
struct hipChannelFormatDesc desc;
size_t sizeInBytes;
} linear;
struct {
void* devPtr;
struct hipChannelFormatDesc desc;
size_t width;
size_t height;
size_t pitchInBytes;
} pitch2D;
} res;
}hipResourceDesc;
typedef struct HIP_RESOURCE_DESC_st
{
HIPresourcetype resType; /**< Resource type */
union {
struct {
hipArray_t hArray; /**< HIP array */
} array;
struct {
hipMipmappedArray_t hMipmappedArray; /**< HIP mipmapped array */
} mipmap;
struct {
hipDeviceptr_t devPtr; /**< Device pointer */
hipArray_Format format; /**< Array format */
unsigned int numChannels; /**< Channels per array element */
size_t sizeInBytes; /**< Size in bytes */
} linear;
struct {
hipDeviceptr_t devPtr; /**< Device pointer */
hipArray_Format format; /**< Array format */
unsigned int numChannels; /**< Channels per array element */
size_t width; /**< Width of the array in elements */
size_t height; /**< Height of the array in elements */
size_t pitchInBytes; /**< Pitch between two rows in bytes */
} pitch2D;
struct {
int reserved[32];
} reserved;
} res;
unsigned int flags; /**< Flags (must be zero) */
} HIP_RESOURCE_DESC;
/**
* hip resource view descriptor
*/
struct hipResourceViewDesc {
enum hipResourceViewFormat format;
size_t width;
size_t height;
size_t depth;
unsigned int firstMipmapLevel;
unsigned int lastMipmapLevel;
unsigned int firstLayer;
unsigned int lastLayer;
};
/**
* Resource view descriptor
*/
typedef struct HIP_RESOURCE_VIEW_DESC_st
{
HIPresourceViewFormat format; /**< Resource view format */
size_t width; /**< Width of the resource view */
size_t height; /**< Height of the resource view */
size_t depth; /**< Depth of the resource view */
unsigned int firstMipmapLevel; /**< First defined mipmap level */
unsigned int lastMipmapLevel; /**< Last defined mipmap level */
unsigned int firstLayer; /**< First layer index */
unsigned int lastLayer; /**< Last layer index */
unsigned int reserved[16];
} HIP_RESOURCE_VIEW_DESC;
/**
* Memory copy types
*
*/
#if !defined(__HIPCC_RTC__)
typedef enum hipMemcpyKind {
hipMemcpyHostToHost = 0, ///< Host-to-Host Copy
hipMemcpyHostToDevice = 1, ///< Host-to-Device Copy
hipMemcpyDeviceToHost = 2, ///< Device-to-Host Copy
hipMemcpyDeviceToDevice = 3, ///< Device-to-Device Copy
hipMemcpyDefault =
4 ///< Runtime will automatically determine copy-kind based on virtual addresses.
} hipMemcpyKind;
typedef struct hipPitchedPtr {
void* ptr;
size_t pitch;
size_t xsize;
size_t ysize;
}hipPitchedPtr;
typedef struct hipExtent {
size_t width; // Width in elements when referring to array memory, in bytes when referring to
// linear memory
size_t height;
size_t depth;
}hipExtent;
typedef struct hipPos {
size_t x;
size_t y;
size_t z;
}hipPos;
typedef struct hipMemcpy3DParms {
hipArray_t srcArray;
struct hipPos srcPos;
struct hipPitchedPtr srcPtr;
hipArray_t dstArray;
struct hipPos dstPos;
struct hipPitchedPtr dstPtr;
struct hipExtent extent;
enum hipMemcpyKind kind;
} hipMemcpy3DParms;
typedef struct HIP_MEMCPY3D {
unsigned int srcXInBytes;
unsigned int srcY;
unsigned int srcZ;
unsigned int srcLOD;
hipMemoryType srcMemoryType;
const void* srcHost;
hipDeviceptr_t srcDevice;
hipArray_t srcArray;
unsigned int srcPitch;
unsigned int srcHeight;
unsigned int dstXInBytes;
unsigned int dstY;
unsigned int dstZ;
unsigned int dstLOD;
hipMemoryType dstMemoryType;
void* dstHost;
hipDeviceptr_t dstDevice;
hipArray_t dstArray;
unsigned int dstPitch;
unsigned int dstHeight;
unsigned int WidthInBytes;
unsigned int Height;
unsigned int Depth;
} HIP_MEMCPY3D;
static inline struct hipPitchedPtr make_hipPitchedPtr(void* d, size_t p, size_t xsz,
size_t ysz) {
struct hipPitchedPtr s;
s.ptr = d;
s.pitch = p;
s.xsize = xsz;
s.ysize = ysz;
return s;
}
static inline struct hipPos make_hipPos(size_t x, size_t y, size_t z) {
struct hipPos p;
p.x = x;
p.y = y;
p.z = z;
return p;
}
static inline struct hipExtent make_hipExtent(size_t w, size_t h, size_t d) {
struct hipExtent e;
e.width = w;
e.height = h;
e.depth = d;
return e;
}
typedef enum hipFunction_attribute {
HIP_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK,
HIP_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES,
HIP_FUNC_ATTRIBUTE_CONST_SIZE_BYTES,
HIP_FUNC_ATTRIBUTE_LOCAL_SIZE_BYTES,
HIP_FUNC_ATTRIBUTE_NUM_REGS,
HIP_FUNC_ATTRIBUTE_PTX_VERSION,
HIP_FUNC_ATTRIBUTE_BINARY_VERSION,
HIP_FUNC_ATTRIBUTE_CACHE_MODE_CA,
HIP_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES,
HIP_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT,
HIP_FUNC_ATTRIBUTE_MAX
}hipFunction_attribute;
#endif // !defined(__HIPCC_RTC__)
#endif
include/hip/amd_detail/functional_grid_launch.hpp
-218
View File
@@ -1,218 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#include "concepts.hpp"
#include "helpers.hpp"
#include "program_state.hpp"
#include "hip_runtime_api.h"
#include <cstdint>
#include <cstring>
#include <stdexcept>
#include <tuple>
#include <type_traits>
#include <utility>
hipError_t ihipExtLaunchMultiKernelMultiDevice(hipLaunchParams* launchParamsList, int numDevices,
unsigned int flags, hip_impl::program_state& ps);
hipError_t hipLaunchCooperativeKernel(const void* f, dim3 gridDim,
dim3 blockDim, void** args,
size_t sharedMem, hipStream_t stream,
hip_impl::program_state& ps);
hipError_t hipLaunchCooperativeKernelMultiDevice(hipLaunchParams* launchParamsList,
int numDevices,
unsigned int flags,
hip_impl::program_state& ps);
#pragma GCC visibility push(hidden)
namespace hip_impl {
template <typename T, typename std::enable_if<std::is_integral<T>{}>::type* = nullptr>
inline T round_up_to_next_multiple_nonnegative(T x, T y) {
T tmp = x + y - 1;
return tmp - tmp % y;
}
template <
std::size_t n,
typename... Ts,
typename std::enable_if<n == sizeof...(Ts)>::type* = nullptr>
inline hip_impl::kernarg make_kernarg(
const std::tuple<Ts...>&,
const kernargs_size_align&,
hip_impl::kernarg kernarg) {
return kernarg;
}
template <
std::size_t n,
typename... Ts,
typename std::enable_if<n != sizeof...(Ts)>::type* = nullptr>
inline hip_impl::kernarg make_kernarg(
const std::tuple<Ts...>& formals,
const kernargs_size_align& size_align,
hip_impl::kernarg kernarg) {
using T = typename std::tuple_element<n, std::tuple<Ts...>>::type;
static_assert(
!std::is_reference<T>{},
"A __global__ function cannot have a reference as one of its "
"arguments.");
#if defined(HIP_STRICT)
static_assert(
std::is_trivially_copyable<T>{},
"Only TriviallyCopyable types can be arguments to a __global__ "
"function");
#endif
kernarg.resize(round_up_to_next_multiple_nonnegative(
kernarg.size(), size_align.alignment(n)) + size_align.size(n));
std::memcpy(
kernarg.data() + kernarg.size() - size_align.size(n),
&std::get<n>(formals),
size_align.size(n));
return make_kernarg<n + 1>(formals, size_align, std::move(kernarg));
}
template <typename... Formals, typename... Actuals>
inline hip_impl::kernarg make_kernarg(
void (*kernel)(Formals...), std::tuple<Actuals...> actuals) {
static_assert(sizeof...(Formals) == sizeof...(Actuals),
"The count of formal arguments must match the count of actuals.");
if (sizeof...(Formals) == 0) return {};
std::tuple<Formals...> to_formals{std::move(actuals)};
hip_impl::kernarg kernarg;
kernarg.reserve(sizeof(to_formals));
auto& ps = hip_impl::get_program_state();
return make_kernarg<0>(to_formals,
ps.get_kernargs_size_align(
reinterpret_cast<std::uintptr_t>(kernel)),
std::move(kernarg));
}
HIP_INTERNAL_EXPORTED_API hsa_agent_t target_agent(hipStream_t stream);
inline
__attribute__((visibility("hidden")))
void hipLaunchKernelGGLImpl(
std::uintptr_t function_address,
const dim3& numBlocks,
const dim3& dimBlocks,
std::uint32_t sharedMemBytes,
hipStream_t stream,
void** kernarg) {
const auto& kd = hip_impl::get_program_state().kernel_descriptor(function_address,
target_agent(stream));
hipModuleLaunchKernel(kd, numBlocks.x, numBlocks.y, numBlocks.z,
dimBlocks.x, dimBlocks.y, dimBlocks.z, sharedMemBytes,
stream, nullptr, kernarg);
}
} // Namespace hip_impl.
template <class T>
inline
hipError_t hipOccupancyMaxPotentialBlockSize(int* gridSize, int* blockSize,
T kernel, size_t dynSharedMemPerBlk = 0, int blockSizeLimit = 0) {
using namespace hip_impl;
hip_impl::hip_init();
auto f = get_program_state().kernel_descriptor(reinterpret_cast<std::uintptr_t>(kernel),
target_agent(0));
return hipModuleOccupancyMaxPotentialBlockSize(gridSize, blockSize, f,
dynSharedMemPerBlk, blockSizeLimit);
}
template <class T>
inline
hipError_t hipOccupancyMaxPotentialBlockSizeWithFlags(int* gridSize, int* blockSize,
T kernel, size_t dynSharedMemPerBlk = 0, int blockSizeLimit = 0, unsigned int flags = 0 ) {
using namespace hip_impl;
hip_impl::hip_init();
if(flags != hipOccupancyDefault) return hipErrorNotSupported;
auto f = get_program_state().kernel_descriptor(reinterpret_cast<std::uintptr_t>(kernel),
target_agent(0));
return hipModuleOccupancyMaxPotentialBlockSize(gridSize, blockSize, f,
dynSharedMemPerBlk, blockSizeLimit);
}
template <typename... Args, typename F = void (*)(Args...)>
inline
void hipLaunchKernelGGL(F kernel, const dim3& numBlocks, const dim3& dimBlocks,
std::uint32_t sharedMemBytes, hipStream_t stream,
Args... args) {
hip_impl::hip_init();
auto kernarg = hip_impl::make_kernarg(kernel, std::tuple<Args...>{std::move(args)...});
std::size_t kernarg_size = kernarg.size();
void* config[]{
HIP_LAUNCH_PARAM_BUFFER_POINTER,
kernarg.data(),
HIP_LAUNCH_PARAM_BUFFER_SIZE,
&kernarg_size,
HIP_LAUNCH_PARAM_END};
hip_impl::hipLaunchKernelGGLImpl(reinterpret_cast<std::uintptr_t>(kernel),
numBlocks, dimBlocks, sharedMemBytes,
stream, &config[0]);
}
template <typename F>
inline
__attribute__((visibility("hidden")))
hipError_t hipLaunchCooperativeKernel(F f, dim3 gridDim, dim3 blockDim,
void** args, size_t sharedMem,
hipStream_t stream) {
hip_impl::hip_init();
auto& ps = hip_impl::get_program_state();
return hipLaunchCooperativeKernel(reinterpret_cast<void*>(f), gridDim,
blockDim, args, sharedMem, stream, ps);
}
inline
__attribute__((visibility("hidden")))
hipError_t hipLaunchCooperativeKernelMultiDevice(hipLaunchParams* launchParamsList,
int numDevices,
unsigned int flags) {
hip_impl::hip_init();
auto& ps = hip_impl::get_program_state();
return hipLaunchCooperativeKernelMultiDevice(launchParamsList, numDevices, flags, ps);
}
#pragma GCC visibility pop
include/hip/amd_detail/grid_launch.h
-67
View File
@@ -1,67 +0,0 @@
#pragma once
#include <stdint.h>
#include <hc_defines.h>
#define GRID_LAUNCH_VERSION 20
// Extern definitions
namespace hc{
class completion_future;
class accelerator_view;
}
// 3 dim structure for groups and grids.
typedef struct gl_dim3
{
int x,y,z;
gl_dim3(uint32_t _x=1, uint32_t _y=1, uint32_t _z=1) : x(_x), y(_y), z(_z) {};
} gl_dim3;
typedef enum gl_barrier_bit {
barrier_bit_queue_default,
barrier_bit_none,
barrier_bit_wait,
} gl_barrier_bit;
// grid_launch_parm contains information used to launch the kernel.
typedef struct grid_launch_parm
{
//! Grid dimensions
gl_dim3 grid_dim;
//! Group dimensions
gl_dim3 group_dim;
//! Amount of dynamic group memory to use with the kernel launch.
//! This memory is in addition to the amount used statically in the kernel.
unsigned int dynamic_group_mem_bytes;
//! Control setting of barrier bit on per-packet basis:
//! See gl_barrier_bit description.
//! Placeholder, is not used to control packet dispatch yet
enum gl_barrier_bit barrier_bit;
//! Value of packet fences to apply to launch.
//! The correspond to the value of bits 9:14 in the AQL packet,
//! see HSA_PACKET_HEADER_ACQUIRE_FENCE_SCOPE and hsa_fence_scope_t.
unsigned int launch_fence;
//! Pointer to the accelerator_view where the kernel should execute.
//! If NULL, the default view on the default accelerator is used.
hc::accelerator_view *av;
//! Pointer to the completion_future used to track the status of the command.
//! If NULL, the command does not write status. In this case,
//! synchronization can be enforced with queue-level waits or
//! waiting on younger commands.
hc::completion_future *cf;
grid_launch_parm() = default;
} grid_launch_parm;
extern void init_grid_launch(grid_launch_parm *gl);
include/hip/amd_detail/grid_launch.hpp
-50
View File
@@ -1,50 +0,0 @@
#pragma once
#include "grid_launch.h"
#include "hc.hpp"
class grid_launch_parm_cxx : public grid_launch_parm
{
public:
grid_launch_parm_cxx() = default;
// customized serialization: don't need av and cf in kernel
__attribute__((annotate("serialize")))
void __cxxamp_serialize(Kalmar::Serialize& s) const {
s.Append(sizeof(int), &grid_dim.x);
s.Append(sizeof(int), &grid_dim.y);
s.Append(sizeof(int), &grid_dim.z);
s.Append(sizeof(int), &group_dim.x);
s.Append(sizeof(int), &group_dim.y);
s.Append(sizeof(int), &group_dim.z);
}
__attribute__((annotate("user_deserialize")))
grid_launch_parm_cxx(int grid_dim_x, int grid_dim_y, int grid_dim_z,
int group_dim_x, int group_dim_y, int group_dim_z) {
grid_dim.x = grid_dim_x;
grid_dim.y = grid_dim_y;
grid_dim.z = grid_dim_z;
group_dim.x = group_dim_x;
group_dim.y = group_dim_y;
group_dim.z = group_dim_z;
}
};
extern inline void grid_launch_init(grid_launch_parm *lp) {
lp->grid_dim.x = lp->grid_dim.y = lp->grid_dim.z = 1;
lp->group_dim.x = lp->group_dim.y = lp->group_dim.z = 1;
lp->dynamic_group_mem_bytes = 0;
lp->barrier_bit = barrier_bit_queue_default;
lp->launch_fence = -1;
// TODO - set to NULL?
static hc::accelerator_view av = hc::accelerator().get_default_view();
lp->av = &av;
lp->cf = NULL;
}
include/hip/amd_detail/grid_launch_GGL.hpp
-26
View File
@@ -1,26 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#if GENERIC_GRID_LAUNCH == 1
#include "macro_based_grid_launch.hpp"
#endif // GENERIC_GRID_LAUNCH
include/hip/amd_detail/helpers.hpp
-137
View File
@@ -1,137 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#include "concepts.hpp"
#include <type_traits> // For std::conditional, std::decay, std::enable_if,
// std::false_type, std result_of and std::true_type.
#include <utility> // For std::declval.
#ifdef __has_include // Check if __has_include is present
# if __has_include(<version>) // Check for version header
# include <version>
# if defined(__cpp_lib_is_invocable) && !defined(HIP_HAS_INVOCABLE)
# define HIP_HAS_INVOCABLE __cpp_lib_is_invocable
# endif
# if defined(__cpp_lib_result_of_sfinae) && !defined(HIP_HAS_RESULT_OF_SFINAE)
# define HIP_HAS_RESULT_OF_SFINAE __cpp_lib_result_of_sfinae
# endif
# endif
#endif
#ifndef HIP_HAS_INVOCABLE
#define HIP_HAS_INVOCABLE 0
#endif
#ifndef HIP_HAS_RESULT_OF_SFINAE
#define HIP_HAS_RESULT_OF_SFINAE 0
#endif
namespace std { // TODO: these should be removed as soon as possible.
#if (__cplusplus < 201406L)
#if (__cplusplus < 201402L)
template <bool cond, typename T = void>
using enable_if_t = typename enable_if<cond, T>::type;
template <bool cond, typename T, typename U>
using conditional_t = typename conditional<cond, T, U>::type;
template <typename T>
using decay_t = typename decay<T>::type;
template <FunctionalProcedure F, typename... Ts>
using result_of_t = typename result_of<F(Ts...)>::type;
template <typename T>
using remove_reference_t = typename remove_reference<T>::type;
#endif
#endif
} // namespace std
namespace hip_impl {
template <typename...>
using void_t_ = void;
#if HIP_HAS_INVOCABLE
template <typename, typename = void>
struct is_callable_impl;
template <FunctionalProcedure F, typename... Ts>
struct is_callable_impl<F(Ts...)> : std::is_invocable<F, Ts...> {};
#elif HIP_HAS_RESULT_OF_SFINAE
template <typename, typename = void>
struct is_callable_impl : std::false_type {};
template <FunctionalProcedure F, typename... Ts>
struct is_callable_impl<F(Ts...), void_t_<typename std::result_of<F(Ts...)>::type > > : std::true_type {};
#else
template <class Base, class T, class Derived>
auto simple_invoke(T Base::*pmd, Derived&& ref)
-> decltype(static_cast<Derived&&>(ref).*pmd);
template <class PMD, class Pointer>
auto simple_invoke(PMD&& pmd, Pointer&& ptr)
-> decltype((*static_cast<Pointer&&>(ptr)).*static_cast<PMD&&>(pmd));
template <class Base, class T, class Derived>
auto simple_invoke(T Base::*pmd, const std::reference_wrapper<Derived>& ref)
-> decltype(ref.get().*pmd);
template <class Base, class T, class Derived, class... Args>
auto simple_invoke(T Base::*pmf, Derived&& ref, Args&&... args)
-> decltype((static_cast<Derived&&>(ref).*pmf)(static_cast<Args&&>(args)...));
template <class PMF, class Pointer, class... Args>
auto simple_invoke(PMF&& pmf, Pointer&& ptr, Args&&... args)
-> decltype(((*static_cast<Pointer&&>(ptr)).*static_cast<PMF&&>(pmf))(static_cast<Args&&>(args)...));
template <class Base, class T, class Derived, class... Args>
auto simple_invoke(T Base::*pmf, const std::reference_wrapper<Derived>& ref, Args&&... args)
-> decltype((ref.get().*pmf)(static_cast<Args&&>(args)...));
template<class F, class... Ts>
auto simple_invoke(F&& f, Ts&&... xs)
-> decltype(f(static_cast<Ts&&>(xs)...));
template <typename, typename = void>
struct is_callable_impl : std::false_type {};
template <FunctionalProcedure F, typename... Ts>
struct is_callable_impl<F(Ts...), void_t_<decltype(simple_invoke(std::declval<F>(), std::declval<Ts>()...))> >
: std::true_type {};
#endif
template <typename Call>
struct is_callable : is_callable_impl<Call> {};
#define count_macro_args_impl_hip_(_0, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, \
_14, _15, _16, _17, _18, _19, _20, _21, _22, _23, _24, _25, \
_26, _27, _28, _29, _30, _31, _n, ...) \
_n
#define count_macro_args_hip_(...) \
count_macro_args_impl_hip_(, ##__VA_ARGS__, 31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, \
19, 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, \
0)
#define overloaded_macro_expand_hip_(macro, arg_cnt) macro##arg_cnt
#define overload_macro_impl_hip_(macro, arg_cnt) overloaded_macro_expand_hip_(macro, arg_cnt)
#define overload_macro_hip_(macro, ...) \
overload_macro_impl_hip_(macro, count_macro_args_hip_(__VA_ARGS__))(__VA_ARGS__)
} // namespace hip_impl
include/hip/amd_detail/hip_atomic.h
-691
View File
@@ -1,691 +0,0 @@
#include "device_functions.h"
#if __has_builtin(__hip_atomic_compare_exchange_strong)
#if !__HIP_DEVICE_COMPILE__
//TODO: Remove this after compiler pre-defines the following Macros.
#define __HIP_MEMORY_SCOPE_SINGLETHREAD 1
#define __HIP_MEMORY_SCOPE_WAVEFRONT 2
#define __HIP_MEMORY_SCOPE_WORKGROUP 3
#define __HIP_MEMORY_SCOPE_AGENT 4
#define __HIP_MEMORY_SCOPE_SYSTEM 5
#endif
__device__
inline
int atomicCAS(int* address, int compare, int val) {
__hip_atomic_compare_exchange_strong(address, &compare, val, __ATOMIC_RELAXED, __ATOMIC_RELAXED,
__HIP_MEMORY_SCOPE_AGENT);
return compare;
}
__device__
inline
int atomicCAS_system(int* address, int compare, int val) {
__hip_atomic_compare_exchange_strong(address, &compare, val, __ATOMIC_RELAXED, __ATOMIC_RELAXED,
__HIP_MEMORY_SCOPE_SYSTEM);
return compare;
}
__device__
inline
unsigned int atomicCAS(unsigned int* address, unsigned int compare, unsigned int val) {
__hip_atomic_compare_exchange_strong(address, &compare, val, __ATOMIC_RELAXED, __ATOMIC_RELAXED,
__HIP_MEMORY_SCOPE_AGENT);
return compare;
}
__device__
inline
unsigned int atomicCAS_system(unsigned int* address, unsigned int compare, unsigned int val) {
__hip_atomic_compare_exchange_strong(address, &compare, val, __ATOMIC_RELAXED, __ATOMIC_RELAXED,
__HIP_MEMORY_SCOPE_SYSTEM);
return compare;
}
__device__
inline
unsigned long long atomicCAS(unsigned long long* address, unsigned long long compare,
unsigned long long val) {
__hip_atomic_compare_exchange_strong(address, &compare, val, __ATOMIC_RELAXED, __ATOMIC_RELAXED,
__HIP_MEMORY_SCOPE_AGENT);
return compare;
}
__device__
inline
unsigned long long atomicCAS_system(unsigned long long* address, unsigned long long compare,
unsigned long long val) {
__hip_atomic_compare_exchange_strong(address, &compare, val, __ATOMIC_RELAXED, __ATOMIC_RELAXED,
__HIP_MEMORY_SCOPE_SYSTEM);
return compare;
}
__device__
inline
int atomicAdd(int* address, int val) {
return __hip_atomic_fetch_add(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
int atomicAdd_system(int* address, int val) {
return __hip_atomic_fetch_add(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned int atomicAdd(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_add(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned int atomicAdd_system(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_add(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned long long atomicAdd(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_add(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned long long atomicAdd_system(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_add(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
float atomicAdd(float* address, float val) {
return __hip_atomic_fetch_add(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
float atomicAdd_system(float* address, float val) {
return __hip_atomic_fetch_add(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
#if !defined(__HIPCC_RTC__)
DEPRECATED("use atomicAdd instead")
#endif // !defined(__HIPCC_RTC__)
__device__
inline
void atomicAddNoRet(float* address, float val)
{
__ockl_atomic_add_noret_f32(address, val);
}
__device__
inline
double atomicAdd(double* address, double val) {
return __hip_atomic_fetch_add(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
double atomicAdd_system(double* address, double val) {
return __hip_atomic_fetch_add(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
int atomicSub(int* address, int val) {
return __hip_atomic_fetch_add(address, -val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
int atomicSub_system(int* address, int val) {
return __hip_atomic_fetch_add(address, -val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned int atomicSub(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_add(address, -val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned int atomicSub_system(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_add(address, -val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
int atomicExch(int* address, int val) {
return __hip_atomic_exchange(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
int atomicExch_system(int* address, int val) {
return __hip_atomic_exchange(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned int atomicExch(unsigned int* address, unsigned int val) {
return __hip_atomic_exchange(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned int atomicExch_system(unsigned int* address, unsigned int val) {
return __hip_atomic_exchange(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned long long atomicExch(unsigned long long* address, unsigned long long val) {
return __hip_atomic_exchange(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned long long atomicExch_system(unsigned long long* address, unsigned long long val) {
return __hip_atomic_exchange(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
float atomicExch(float* address, float val) {
return __hip_atomic_exchange(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
float atomicExch_system(float* address, float val) {
return __hip_atomic_exchange(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
int atomicMin(int* address, int val) {
return __hip_atomic_fetch_min(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
int atomicMin_system(int* address, int val) {
return __hip_atomic_fetch_min(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned int atomicMin(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_min(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned int atomicMin_system(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_min(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned long long atomicMin(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_min(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned long long atomicMin_system(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_min(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
int atomicMax(int* address, int val) {
return __hip_atomic_fetch_max(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
int atomicMax_system(int* address, int val) {
return __hip_atomic_fetch_max(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned int atomicMax(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_max(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned int atomicMax_system(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_max(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned long long atomicMax(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_max(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned long long atomicMax_system(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_max(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned int atomicInc(unsigned int* address, unsigned int val)
{
__device__
extern
unsigned int __builtin_amdgcn_atomic_inc(
unsigned int*,
unsigned int,
unsigned int,
unsigned int,
bool) __asm("llvm.amdgcn.atomic.inc.i32.p0i32");
return __builtin_amdgcn_atomic_inc(
address, val, __ATOMIC_RELAXED, 1 /* Device scope */, false);
}
__device__
inline
unsigned int atomicDec(unsigned int* address, unsigned int val)
{
__device__
extern
unsigned int __builtin_amdgcn_atomic_dec(
unsigned int*,
unsigned int,
unsigned int,
unsigned int,
bool) __asm("llvm.amdgcn.atomic.dec.i32.p0i32");
return __builtin_amdgcn_atomic_dec(
address, val, __ATOMIC_RELAXED, 1 /* Device scope */, false);
}
__device__
inline
int atomicAnd(int* address, int val) {
return __hip_atomic_fetch_and(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
int atomicAnd_system(int* address, int val) {
return __hip_atomic_fetch_and(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned int atomicAnd(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_and(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned int atomicAnd_system(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_and(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned long long atomicAnd(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_and(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned long long atomicAnd_system(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_and(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
int atomicOr(int* address, int val) {
return __hip_atomic_fetch_or(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
int atomicOr_system(int* address, int val) {
return __hip_atomic_fetch_or(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned int atomicOr(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_or(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned int atomicOr_system(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_or(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned long long atomicOr(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_or(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned long long atomicOr_system(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_or(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
int atomicXor(int* address, int val) {
return __hip_atomic_fetch_xor(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
int atomicXor_system(int* address, int val) {
return __hip_atomic_fetch_xor(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned int atomicXor(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_xor(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned int atomicXor_system(unsigned int* address, unsigned int val) {
return __hip_atomic_fetch_xor(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
__device__
inline
unsigned long long atomicXor(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_xor(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
}
__device__
inline
unsigned long long atomicXor_system(unsigned long long* address, unsigned long long val) {
return __hip_atomic_fetch_xor(address, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
}
#else
__device__
inline
int atomicCAS(int* address, int compare, int val)
{
__atomic_compare_exchange_n(
address, &compare, val, false, __ATOMIC_RELAXED, __ATOMIC_RELAXED);
return compare;
}
__device__
inline
unsigned int atomicCAS(
unsigned int* address, unsigned int compare, unsigned int val)
{
__atomic_compare_exchange_n(
address, &compare, val, false, __ATOMIC_RELAXED, __ATOMIC_RELAXED);
return compare;
}
__device__
inline
unsigned long long atomicCAS(
unsigned long long* address,
unsigned long long compare,
unsigned long long val)
{
__atomic_compare_exchange_n(
address, &compare, val, false, __ATOMIC_RELAXED, __ATOMIC_RELAXED);
return compare;
}
__device__
inline
int atomicAdd(int* address, int val)
{
return __atomic_fetch_add(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned int atomicAdd(unsigned int* address, unsigned int val)
{
return __atomic_fetch_add(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned long long atomicAdd(
unsigned long long* address, unsigned long long val)
{
return __atomic_fetch_add(address, val, __ATOMIC_RELAXED);
}
__device__
inline
float atomicAdd(float* address, float val)
{
return __atomic_fetch_add(address, val, __ATOMIC_RELAXED);
}
#if !defined(__HIPCC_RTC__)
DEPRECATED("use atomicAdd instead")
#endif // !defined(__HIPCC_RTC__)
__device__
inline
void atomicAddNoRet(float* address, float val)
{
__ockl_atomic_add_noret_f32(address, val);
}
__device__
inline
double atomicAdd(double* address, double val)
{
return __atomic_fetch_add(address, val, __ATOMIC_RELAXED);
}
__device__
inline
int atomicSub(int* address, int val)
{
return __atomic_fetch_sub(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned int atomicSub(unsigned int* address, unsigned int val)
{
return __atomic_fetch_sub(address, val, __ATOMIC_RELAXED);
}
__device__
inline
int atomicExch(int* address, int val)
{
return __atomic_exchange_n(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned int atomicExch(unsigned int* address, unsigned int val)
{
return __atomic_exchange_n(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned long long atomicExch(unsigned long long* address, unsigned long long val)
{
return __atomic_exchange_n(address, val, __ATOMIC_RELAXED);
}
__device__
inline
float atomicExch(float* address, float val)
{
return __uint_as_float(__atomic_exchange_n(
reinterpret_cast<unsigned int*>(address),
__float_as_uint(val),
__ATOMIC_RELAXED));
}
__device__
inline
int atomicMin(int* address, int val)
{
return __atomic_fetch_min(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned int atomicMin(unsigned int* address, unsigned int val)
{
return __atomic_fetch_min(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned long long atomicMin(
unsigned long long* address, unsigned long long val)
{
unsigned long long tmp{__atomic_load_n(address, __ATOMIC_RELAXED)};
while (val < tmp) {
const auto tmp1 = __atomic_load_n(address, __ATOMIC_RELAXED);
if (tmp1 != tmp) { tmp = tmp1; continue; }
tmp = atomicCAS(address, tmp, val);
}
return tmp;
}
__device__
inline
int atomicMax(int* address, int val)
{
return __atomic_fetch_max(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned int atomicMax(unsigned int* address, unsigned int val)
{
return __atomic_fetch_max(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned long long atomicMax(
unsigned long long* address, unsigned long long val)
{
unsigned long long tmp{__atomic_load_n(address, __ATOMIC_RELAXED)};
while (tmp < val) {
const auto tmp1 = __atomic_load_n(address, __ATOMIC_RELAXED);
if (tmp1 != tmp) { tmp = tmp1; continue; }
tmp = atomicCAS(address, tmp, val);
}
return tmp;
}
__device__
inline
unsigned int atomicInc(unsigned int* address, unsigned int val)
{
__device__
extern
unsigned int __builtin_amdgcn_atomic_inc(
unsigned int*,
unsigned int,
unsigned int,
unsigned int,
bool) __asm("llvm.amdgcn.atomic.inc.i32.p0i32");
return __builtin_amdgcn_atomic_inc(
address, val, __ATOMIC_RELAXED, 1 /* Device scope */, false);
}
__device__
inline
unsigned int atomicDec(unsigned int* address, unsigned int val)
{
__device__
extern
unsigned int __builtin_amdgcn_atomic_dec(
unsigned int*,
unsigned int,
unsigned int,
unsigned int,
bool) __asm("llvm.amdgcn.atomic.dec.i32.p0i32");
return __builtin_amdgcn_atomic_dec(
address, val, __ATOMIC_RELAXED, 1 /* Device scope */, false);
}
__device__
inline
int atomicAnd(int* address, int val)
{
return __atomic_fetch_and(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned int atomicAnd(unsigned int* address, unsigned int val)
{
return __atomic_fetch_and(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned long long atomicAnd(
unsigned long long* address, unsigned long long val)
{
return __atomic_fetch_and(address, val, __ATOMIC_RELAXED);
}
__device__
inline
int atomicOr(int* address, int val)
{
return __atomic_fetch_or(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned int atomicOr(unsigned int* address, unsigned int val)
{
return __atomic_fetch_or(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned long long atomicOr(
unsigned long long* address, unsigned long long val)
{
return __atomic_fetch_or(address, val, __ATOMIC_RELAXED);
}
__device__
inline
int atomicXor(int* address, int val)
{
return __atomic_fetch_xor(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned int atomicXor(unsigned int* address, unsigned int val)
{
return __atomic_fetch_xor(address, val, __ATOMIC_RELAXED);
}
__device__
inline
unsigned long long atomicXor(
unsigned long long* address, unsigned long long val)
{
return __atomic_fetch_xor(address, val, __ATOMIC_RELAXED);
}
#endif
include/hip/amd_detail/hip_common.h
-32
View File
@@ -1,32 +0,0 @@
/*
Copyright (c) 2019 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy of
this software and associated documentation files (the "Software"), to deal in
the Software without restriction, including without limitation the rights to
use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
of the Software, and to permit persons to whom the Software is furnished to do
so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COMMON_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COMMON_H
#if defined(__clang__) && defined(__HIP__)
#define __HIP_CLANG_ONLY__ 1
#else
#define __HIP_CLANG_ONLY__ 0
#endif
#endif // HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COMMON_H
include/hip/amd_detail/hip_complex.h
-309
View File
@@ -1,309 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COMPLEX_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COMPLEX_H
#include "hip/amd_detail/hip_vector_types.h"
#if defined(__HIPCC_RTC__)
#define __HOST_DEVICE__ __device__
#else
#define __HOST_DEVICE__ __host__ __device__
// TODO: Clang has a bug which allows device functions to call std functions
// when std functions are introduced into default namespace by using statement.
// math.h may be included after this bug is fixed.
#if __cplusplus
#include <cmath>
#else
#include "math.h"
#endif
#endif // !defined(__HIPCC_RTC__)
#if __cplusplus
#define COMPLEX_NEG_OP_OVERLOAD(type) \
__HOST_DEVICE__ static inline type operator-(const type& op) { \
type ret; \
ret.x = -op.x; \
ret.y = -op.y; \
return ret; \
}
#define COMPLEX_EQ_OP_OVERLOAD(type) \
__HOST_DEVICE__ static inline bool operator==(const type& lhs, const type& rhs) { \
return lhs.x == rhs.x && lhs.y == rhs.y; \
}
#define COMPLEX_NE_OP_OVERLOAD(type) \
__HOST_DEVICE__ static inline bool operator!=(const type& lhs, const type& rhs) { \
return !(lhs == rhs); \
}
#define COMPLEX_ADD_OP_OVERLOAD(type) \
__HOST_DEVICE__ static inline type operator+(const type& lhs, const type& rhs) { \
type ret; \
ret.x = lhs.x + rhs.x; \
ret.y = lhs.y + rhs.y; \
return ret; \
}
#define COMPLEX_SUB_OP_OVERLOAD(type) \
__HOST_DEVICE__ static inline type operator-(const type& lhs, const type& rhs) { \
type ret; \
ret.x = lhs.x - rhs.x; \
ret.y = lhs.y - rhs.y; \
return ret; \
}
#define COMPLEX_MUL_OP_OVERLOAD(type) \
__HOST_DEVICE__ static inline type operator*(const type& lhs, const type& rhs) { \
type ret; \
ret.x = lhs.x * rhs.x - lhs.y * rhs.y; \
ret.y = lhs.x * rhs.y + lhs.y * rhs.x; \
return ret; \
}
#define COMPLEX_DIV_OP_OVERLOAD(type) \
__HOST_DEVICE__ static inline type operator/(const type& lhs, const type& rhs) { \
type ret; \
ret.x = (lhs.x * rhs.x + lhs.y * rhs.y); \
ret.y = (rhs.x * lhs.y - lhs.x * rhs.y); \
ret.x = ret.x / (rhs.x * rhs.x + rhs.y * rhs.y); \
ret.y = ret.y / (rhs.x * rhs.x + rhs.y * rhs.y); \
return ret; \
}
#define COMPLEX_ADD_PREOP_OVERLOAD(type) \
__HOST_DEVICE__ static inline type& operator+=(type& lhs, const type& rhs) { \
lhs.x += rhs.x; \
lhs.y += rhs.y; \
return lhs; \
}
#define COMPLEX_SUB_PREOP_OVERLOAD(type) \
__HOST_DEVICE__ static inline type& operator-=(type& lhs, const type& rhs) { \
lhs.x -= rhs.x; \
lhs.y -= rhs.y; \
return lhs; \
}
#define COMPLEX_MUL_PREOP_OVERLOAD(type) \
__HOST_DEVICE__ static inline type& operator*=(type& lhs, const type& rhs) { \
lhs = lhs * rhs; \
return lhs; \
}
#define COMPLEX_DIV_PREOP_OVERLOAD(type) \
__HOST_DEVICE__ static inline type& operator/=(type& lhs, const type& rhs) { \
lhs = lhs / rhs; \
return lhs; \
}
#define COMPLEX_SCALAR_PRODUCT(type, type1) \
__HOST_DEVICE__ static inline type operator*(const type& lhs, type1 rhs) { \
type ret; \
ret.x = lhs.x * rhs; \
ret.y = lhs.y * rhs; \
return ret; \
}
#endif
typedef float2 hipFloatComplex;
__HOST_DEVICE__ static inline float hipCrealf(hipFloatComplex z) { return z.x; }
__HOST_DEVICE__ static inline float hipCimagf(hipFloatComplex z) { return z.y; }
__HOST_DEVICE__ static inline hipFloatComplex make_hipFloatComplex(float a, float b) {
hipFloatComplex z;
z.x = a;
z.y = b;
return z;
}
__HOST_DEVICE__ static inline hipFloatComplex hipConjf(hipFloatComplex z) {
hipFloatComplex ret;
ret.x = z.x;
ret.y = -z.y;
return ret;
}
__HOST_DEVICE__ static inline float hipCsqabsf(hipFloatComplex z) {
return z.x * z.x + z.y * z.y;
}
__HOST_DEVICE__ static inline hipFloatComplex hipCaddf(hipFloatComplex p, hipFloatComplex q) {
return make_hipFloatComplex(p.x + q.x, p.y + q.y);
}
__HOST_DEVICE__ static inline hipFloatComplex hipCsubf(hipFloatComplex p, hipFloatComplex q) {
return make_hipFloatComplex(p.x - q.x, p.y - q.y);
}
__HOST_DEVICE__ static inline hipFloatComplex hipCmulf(hipFloatComplex p, hipFloatComplex q) {
return make_hipFloatComplex(p.x * q.x - p.y * q.y, p.y * q.x + p.x * q.y);
}
__HOST_DEVICE__ static inline hipFloatComplex hipCdivf(hipFloatComplex p, hipFloatComplex q) {
float sqabs = hipCsqabsf(q);
hipFloatComplex ret;
ret.x = (p.x * q.x + p.y * q.y) / sqabs;
ret.y = (p.y * q.x - p.x * q.y) / sqabs;
return ret;
}
__HOST_DEVICE__ static inline float hipCabsf(hipFloatComplex z) { return sqrtf(hipCsqabsf(z)); }
typedef double2 hipDoubleComplex;
__HOST_DEVICE__ static inline double hipCreal(hipDoubleComplex z) { return z.x; }
__HOST_DEVICE__ static inline double hipCimag(hipDoubleComplex z) { return z.y; }
__HOST_DEVICE__ static inline hipDoubleComplex make_hipDoubleComplex(double a, double b) {
hipDoubleComplex z;
z.x = a;
z.y = b;
return z;
}
__HOST_DEVICE__ static inline hipDoubleComplex hipConj(hipDoubleComplex z) {
hipDoubleComplex ret;
ret.x = z.x;
ret.y = -z.y;
return ret;
}
__HOST_DEVICE__ static inline double hipCsqabs(hipDoubleComplex z) {
return z.x * z.x + z.y * z.y;
}
__HOST_DEVICE__ static inline hipDoubleComplex hipCadd(hipDoubleComplex p, hipDoubleComplex q) {
return make_hipDoubleComplex(p.x + q.x, p.y + q.y);
}
__HOST_DEVICE__ static inline hipDoubleComplex hipCsub(hipDoubleComplex p, hipDoubleComplex q) {
return make_hipDoubleComplex(p.x - q.x, p.y - q.y);
}
__HOST_DEVICE__ static inline hipDoubleComplex hipCmul(hipDoubleComplex p, hipDoubleComplex q) {
return make_hipDoubleComplex(p.x * q.x - p.y * q.y, p.y * q.x + p.x * q.y);
}
__HOST_DEVICE__ static inline hipDoubleComplex hipCdiv(hipDoubleComplex p, hipDoubleComplex q) {
double sqabs = hipCsqabs(q);
hipDoubleComplex ret;
ret.x = (p.x * q.x + p.y * q.y) / sqabs;
ret.y = (p.y * q.x - p.x * q.y) / sqabs;
return ret;
}
__HOST_DEVICE__ static inline double hipCabs(hipDoubleComplex z) { return sqrt(hipCsqabs(z)); }
#if __cplusplus
COMPLEX_NEG_OP_OVERLOAD(hipFloatComplex)
COMPLEX_EQ_OP_OVERLOAD(hipFloatComplex)
COMPLEX_NE_OP_OVERLOAD(hipFloatComplex)
COMPLEX_ADD_OP_OVERLOAD(hipFloatComplex)
COMPLEX_SUB_OP_OVERLOAD(hipFloatComplex)
COMPLEX_MUL_OP_OVERLOAD(hipFloatComplex)
COMPLEX_DIV_OP_OVERLOAD(hipFloatComplex)
COMPLEX_ADD_PREOP_OVERLOAD(hipFloatComplex)
COMPLEX_SUB_PREOP_OVERLOAD(hipFloatComplex)
COMPLEX_MUL_PREOP_OVERLOAD(hipFloatComplex)
COMPLEX_DIV_PREOP_OVERLOAD(hipFloatComplex)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, unsigned short)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, signed short)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, unsigned int)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, signed int)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, float)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, unsigned long)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, signed long)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, double)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, signed long long)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, unsigned long long)
COMPLEX_NEG_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_EQ_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_NE_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_ADD_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_SUB_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_MUL_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_DIV_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_ADD_PREOP_OVERLOAD(hipDoubleComplex)
COMPLEX_SUB_PREOP_OVERLOAD(hipDoubleComplex)
COMPLEX_MUL_PREOP_OVERLOAD(hipDoubleComplex)
COMPLEX_DIV_PREOP_OVERLOAD(hipDoubleComplex)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, unsigned short)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, signed short)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, unsigned int)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, signed int)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, float)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, unsigned long)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, signed long)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, double)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, signed long long)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, unsigned long long)
#endif
typedef hipFloatComplex hipComplex;
__HOST_DEVICE__ static inline hipComplex make_hipComplex(float x, float y) {
return make_hipFloatComplex(x, y);
}
__HOST_DEVICE__ static inline hipFloatComplex hipComplexDoubleToFloat(hipDoubleComplex z) {
return make_hipFloatComplex((float)z.x, (float)z.y);
}
__HOST_DEVICE__ static inline hipDoubleComplex hipComplexFloatToDouble(hipFloatComplex z) {
return make_hipDoubleComplex((double)z.x, (double)z.y);
}
__HOST_DEVICE__ static inline hipComplex hipCfmaf(hipComplex p, hipComplex q, hipComplex r) {
float real = (p.x * q.x) + r.x;
float imag = (q.x * p.y) + r.y;
real = -(p.y * q.y) + real;
imag = (p.x * q.y) + imag;
return make_hipComplex(real, imag);
}
__HOST_DEVICE__ static inline hipDoubleComplex hipCfma(hipDoubleComplex p, hipDoubleComplex q,
hipDoubleComplex r) {
double real = (p.x * q.x) + r.x;
double imag = (q.x * p.y) + r.y;
real = -(p.y * q.y) + real;
imag = (p.x * q.y) + imag;
return make_hipDoubleComplex(real, imag);
}
#endif //HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COMPLEX_H
include/hip/amd_detail/hip_cooperative_groups.h
-510
View File
@@ -1,510 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/**
* @file amd_detail/hip_cooperative_groups.h
*
* @brief Device side implementation of `Cooperative Group` feature.
*
* Defines new types and device API wrappers related to `Cooperative Group`
* feature, which the programmer can directly use in his kernel(s) in order to
* make use of this feature.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COOPERATIVE_GROUPS_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COOPERATIVE_GROUPS_H
#if __cplusplus
#include <hip/amd_detail/hip_cooperative_groups_helper.h>
namespace cooperative_groups {
/** \brief The base type of all cooperative group types
*
* \details Holds the key properties of a constructed cooperative group types
* object, like the group type, its size, etc
*/
class thread_group {
protected:
uint32_t _type; // thread_group type
uint32_t _size; // total number of threads in the tread_group
uint64_t _mask; // Lanemask for coalesced and tiled partitioned group types,
// LSB represents lane 0, and MSB represents lane 63
// Construct a thread group, and set thread group type and other essential
// thread group properties. This generic thread group is directly constructed
// only when the group is supposed to contain only the calling the thread
// (throurh the API - `this_thread()`), and in all other cases, this thread
// group object is a sub-object of some other derived thread group object
__CG_QUALIFIER__ thread_group(internal::group_type type, uint32_t size,
uint64_t mask = (uint64_t)0) {
_type = type;
_size = size;
_mask = mask;
}
struct _tiled_info {
bool is_tiled;
unsigned int size;
} tiled_info;
friend __CG_QUALIFIER__ thread_group tiled_partition(const thread_group& parent,
unsigned int tile_size);
friend class thread_block;
public:
// Total number of threads in the thread group, and this serves the purpose
// for all derived cooperative group types since their `size` is directly
// saved during the construction
__CG_QUALIFIER__ uint32_t size() const { return _size; }
__CG_QUALIFIER__ unsigned int cg_type() const { return _type; }
// Rank of the calling thread within [0, size())
__CG_QUALIFIER__ uint32_t thread_rank() const;
// Is this cooperative group type valid?
__CG_QUALIFIER__ bool is_valid() const;
// synchronize the threads in the thread group
__CG_QUALIFIER__ void sync() const;
};
/** \brief The multi-grid cooperative group type
*
* \details Represents an inter-device cooperative group type where the
* participating threads within the group spans across multple
* devices, running the (same) kernel on these devices
*/
class multi_grid_group : public thread_group {
// Only these friend functions are allowed to construct an object of this class
// and access its resources
friend __CG_QUALIFIER__ multi_grid_group this_multi_grid();
protected:
// Construct mutli-grid thread group (through the API this_multi_grid())
explicit __CG_QUALIFIER__ multi_grid_group(uint32_t size)
: thread_group(internal::cg_multi_grid, size) {}
public:
// Number of invocations participating in this multi-grid group. In other
// words, the number of GPUs
__CG_QUALIFIER__ uint32_t num_grids() { return internal::multi_grid::num_grids(); }
// Rank of this invocation. In other words, an ID number within the range
// [0, num_grids()) of the GPU, this kernel is running on
__CG_QUALIFIER__ uint32_t grid_rank() { return internal::multi_grid::grid_rank(); }
__CG_QUALIFIER__ uint32_t thread_rank() const { return internal::multi_grid::thread_rank(); }
__CG_QUALIFIER__ bool is_valid() const { return internal::multi_grid::is_valid(); }
__CG_QUALIFIER__ void sync() const { internal::multi_grid::sync(); }
};
/** \brief User exposed API interface to construct multi-grid cooperative
* group type object - `multi_grid_group`
*
* \details User is not allowed to directly construct an object of type
* `multi_grid_group`. Instead, he should construct it through this
* API function
*/
__CG_QUALIFIER__ multi_grid_group this_multi_grid() {
return multi_grid_group(internal::multi_grid::size());
}
/** \brief The grid cooperative group type
*
* \details Represents an inter-workgroup cooperative group type where the
* participating threads within the group spans across multiple
* workgroups running the (same) kernel on the same device
*/
class grid_group : public thread_group {
// Only these friend functions are allowed to construct an object of this class
// and access its resources
friend __CG_QUALIFIER__ grid_group this_grid();
protected:
// Construct grid thread group (through the API this_grid())
explicit __CG_QUALIFIER__ grid_group(uint32_t size) : thread_group(internal::cg_grid, size) {}
public:
__CG_QUALIFIER__ uint32_t thread_rank() const { return internal::grid::thread_rank(); }
__CG_QUALIFIER__ bool is_valid() const { return internal::grid::is_valid(); }
__CG_QUALIFIER__ void sync() const { internal::grid::sync(); }
};
/** \brief User exposed API interface to construct grid cooperative group type
* object - `grid_group`
*
* \details User is not allowed to directly construct an object of type
* `multi_grid_group`. Instead, he should construct it through this
* API function
*/
__CG_QUALIFIER__ grid_group this_grid() { return grid_group(internal::grid::size()); }
/** \brief The workgroup (thread-block in CUDA terminology) cooperative group
* type
*
* \details Represents an intra-workgroup cooperative group type where the
* participating threads within the group are exactly the same threads
* which are participated in the currently executing `workgroup`
*/
class thread_block : public thread_group {
// Only these friend functions are allowed to construct an object of thi
// class and access its resources
friend __CG_QUALIFIER__ thread_block this_thread_block();
friend __CG_QUALIFIER__ thread_group tiled_partition(const thread_group& parent,
unsigned int tile_size);
friend __CG_QUALIFIER__ thread_group tiled_partition(const thread_block& parent,
unsigned int tile_size);
protected:
// Construct a workgroup thread group (through the API this_thread_block())
explicit __CG_QUALIFIER__ thread_block(uint32_t size)
: thread_group(internal::cg_workgroup, size) {}
__CG_QUALIFIER__ thread_group new_tiled_group(unsigned int tile_size) const {
const bool pow2 = ((tile_size & (tile_size - 1)) == 0);
// Invalid tile size, assert
if (!tile_size || (tile_size > WAVEFRONT_SIZE) || !pow2) {
assert(false && "invalid tile size");
}
thread_group tiledGroup = thread_group(internal::cg_tiled_group, tile_size);
tiledGroup.tiled_info.size = tile_size;
tiledGroup.tiled_info.is_tiled = true;
return tiledGroup;
}
public:
// 3-dimensional block index within the grid
__CG_QUALIFIER__ dim3 group_index() { return internal::workgroup::group_index(); }
// 3-dimensional thread index within the block
__CG_QUALIFIER__ dim3 thread_index() { return internal::workgroup::thread_index(); }
__CG_QUALIFIER__ uint32_t thread_rank() const { return internal::workgroup::thread_rank(); }
__CG_QUALIFIER__ bool is_valid() const { return internal::workgroup::is_valid(); }
__CG_QUALIFIER__ void sync() const { internal::workgroup::sync(); }
};
/** \brief User exposed API interface to construct workgroup cooperative
* group type object - `thread_block`.
*
* \details User is not allowed to directly construct an object of type
* `thread_block`. Instead, he should construct it through this API
* function.
*/
__CG_QUALIFIER__ thread_block this_thread_block() {
return thread_block(internal::workgroup::size());
}
/** \brief The tiled_group cooperative group type
*
* \details Represents one tiled thread group in a wavefront.
* This group type also supports sub-wave level intrinsics.
*/
class tiled_group : public thread_group {
private:
friend __CG_QUALIFIER__ thread_group tiled_partition(const thread_group& parent,
unsigned int tile_size);
friend __CG_QUALIFIER__ tiled_group tiled_partition(const tiled_group& parent,
unsigned int tile_size);
__CG_QUALIFIER__ tiled_group new_tiled_group(unsigned int tile_size) const {
const bool pow2 = ((tile_size & (tile_size - 1)) == 0);
if (!tile_size || (tile_size > WAVEFRONT_SIZE) || !pow2) {
assert(false && "invalid tile size");
}
if (size() <= tile_size) {
return (*this);
}
tiled_group tiledGroup = tiled_group(tile_size);
tiledGroup.tiled_info.is_tiled = true;
return tiledGroup;
}
protected:
explicit __CG_QUALIFIER__ tiled_group(unsigned int tileSize)
: thread_group(internal::cg_tiled_group, tileSize) {
tiled_info.size = tileSize;
tiled_info.is_tiled = true;
}
public:
__CG_QUALIFIER__ unsigned int size() const { return (tiled_info.size); }
__CG_QUALIFIER__ unsigned int thread_rank() const {
return (internal::workgroup::thread_rank() & (tiled_info.size - 1));
}
__CG_QUALIFIER__ void sync() const {
// enforce memory ordering for memory instructions.
__builtin_amdgcn_fence(__ATOMIC_ACQ_REL, "agent");
}
};
/**
* Implemenation of all publicly exposed base class APIs
*/
__CG_QUALIFIER__ uint32_t thread_group::thread_rank() const {
switch (this->_type) {
case internal::cg_multi_grid: {
return (static_cast<const multi_grid_group*>(this)->thread_rank());
}
case internal::cg_grid: {
return (static_cast<const grid_group*>(this)->thread_rank());
}
case internal::cg_workgroup: {
return (static_cast<const thread_block*>(this)->thread_rank());
}
case internal::cg_tiled_group: {
return (static_cast<const tiled_group*>(this)->thread_rank());
}
default: {
assert(false && "invalid cooperative group type");
return -1;
}
}
}
__CG_QUALIFIER__ bool thread_group::is_valid() const {
switch (this->_type) {
case internal::cg_multi_grid: {
return (static_cast<const multi_grid_group*>(this)->is_valid());
}
case internal::cg_grid: {
return (static_cast<const grid_group*>(this)->is_valid());
}
case internal::cg_workgroup: {
return (static_cast<const thread_block*>(this)->is_valid());
}
case internal::cg_tiled_group: {
return (static_cast<const tiled_group*>(this)->is_valid());
}
default: {
assert(false && "invalid cooperative group type");
return false;
}
}
}
__CG_QUALIFIER__ void thread_group::sync() const {
switch (this->_type) {
case internal::cg_multi_grid: {
static_cast<const multi_grid_group*>(this)->sync();
break;
}
case internal::cg_grid: {
static_cast<const grid_group*>(this)->sync();
break;
}
case internal::cg_workgroup: {
static_cast<const thread_block*>(this)->sync();
break;
}
case internal::cg_tiled_group: {
static_cast<const tiled_group*>(this)->sync();
break;
}
default: {
assert(false && "invalid cooperative group type");
}
}
}
/**
* Implemenation of publicly exposed `wrapper` APIs on top of basic cooperative
* group type APIs
*/
template <class CGTy> __CG_QUALIFIER__ uint32_t group_size(CGTy const& g) { return g.size(); }
template <class CGTy> __CG_QUALIFIER__ uint32_t thread_rank(CGTy const& g) {
return g.thread_rank();
}
template <class CGTy> __CG_QUALIFIER__ bool is_valid(CGTy const& g) { return g.is_valid(); }
template <class CGTy> __CG_QUALIFIER__ void sync(CGTy const& g) { g.sync(); }
template <unsigned int tileSize> class tile_base {
protected:
_CG_STATIC_CONST_DECL_ unsigned int numThreads = tileSize;
public:
// Rank of the thread within this tile
_CG_STATIC_CONST_DECL_ unsigned int thread_rank() {
return (internal::workgroup::thread_rank() & (numThreads - 1));
}
// Number of threads within this tile
__CG_STATIC_QUALIFIER__ unsigned int size() { return numThreads; }
};
template <unsigned int size> class thread_block_tile_base : public tile_base<size> {
static_assert(is_valid_tile_size<size>::value,
"Tile size is either not a power of 2 or greater than the wavefront size");
using tile_base<size>::numThreads;
public:
__CG_STATIC_QUALIFIER__ void sync() {
// enforce ordering for memory instructions
__builtin_amdgcn_fence(__ATOMIC_ACQ_REL, "agent");
}
template <class T> __CG_QUALIFIER__ T shfl(T var, int srcRank) const {
static_assert(is_valid_type<T>::value, "Neither an integer or float type.");
return (__shfl(var, srcRank, numThreads));
}
template <class T> __CG_QUALIFIER__ T shfl_down(T var, unsigned int lane_delta) const {
static_assert(is_valid_type<T>::value, "Neither an integer or float type.");
return (__shfl_down(var, lane_delta, numThreads));
}
template <class T> __CG_QUALIFIER__ T shfl_up(T var, unsigned int lane_delta) const {
static_assert(is_valid_type<T>::value, "Neither an integer or float type.");
return (__shfl_up(var, lane_delta, numThreads));
}
template <class T> __CG_QUALIFIER__ T shfl_xor(T var, unsigned int laneMask) const {
static_assert(is_valid_type<T>::value, "Neither an integer or float type.");
return (__shfl_xor(var, laneMask, numThreads));
}
};
/** \brief Group type - thread_block_tile
*
* \details Represents one tile of thread group.
*/
template <unsigned int tileSize, class ParentCGTy = void>
class thread_block_tile_type : public thread_block_tile_base<tileSize>, public tiled_group {
_CG_STATIC_CONST_DECL_ unsigned int numThreads = tileSize;
friend class thread_block_tile_type<tileSize, ParentCGTy>;
typedef thread_block_tile_base<numThreads> tbtBase;
protected:
__CG_QUALIFIER__ thread_block_tile_type() : tiled_group(numThreads) {
tiled_info.size = numThreads;
tiled_info.is_tiled = true;
}
public:
using tbtBase::size;
using tbtBase::sync;
using tbtBase::thread_rank;
};
/** \brief User exposed API to partition groups.
*
* \details A collective operation that partitions the parent group into a one-dimensional,
* row-major, tiling of subgroups.
*/
__CG_QUALIFIER__ thread_group tiled_partition(const thread_group& parent, unsigned int tile_size) {
if (parent.cg_type() == internal::cg_tiled_group) {
const tiled_group* cg = static_cast<const tiled_group*>(&parent);
return cg->new_tiled_group(tile_size);
} else {
const thread_block* tb = static_cast<const thread_block*>(&parent);
return tb->new_tiled_group(tile_size);
}
}
// Thread block type overload
__CG_QUALIFIER__ thread_group tiled_partition(const thread_block& parent, unsigned int tile_size) {
return (parent.new_tiled_group(tile_size));
}
// Coalesced group type overload
__CG_QUALIFIER__ tiled_group tiled_partition(const tiled_group& parent, unsigned int tile_size) {
return (parent.new_tiled_group(tile_size));
}
template <unsigned int size, class ParentCGTy> class thread_block_tile;
namespace impl {
template <unsigned int size, class ParentCGTy> class thread_block_tile_internal;
template <unsigned int size, class ParentCGTy>
class thread_block_tile_internal : public thread_block_tile_type<size, ParentCGTy> {
protected:
template <unsigned int tbtSize, class tbtParentT>
__CG_QUALIFIER__ thread_block_tile_internal(
const thread_block_tile_internal<tbtSize, tbtParentT>& g)
: thread_block_tile_type<size, ParentCGTy>() {}
__CG_QUALIFIER__ thread_block_tile_internal(const thread_block& g)
: thread_block_tile_type<size, ParentCGTy>() {}
};
} // namespace impl
template <unsigned int size, class ParentCGTy>
class thread_block_tile : public impl::thread_block_tile_internal<size, ParentCGTy> {
protected:
__CG_QUALIFIER__ thread_block_tile(const ParentCGTy& g)
: impl::thread_block_tile_internal<size, ParentCGTy>(g) {}
public:
__CG_QUALIFIER__ operator thread_block_tile<size, void>() const {
return thread_block_tile<size, void>(*this);
}
};
template <unsigned int size>
class thread_block_tile<size, void> : public impl::thread_block_tile_internal<size, void> {
template <unsigned int, class ParentCGTy> friend class thread_block_tile;
protected:
public:
template <class ParentCGTy>
__CG_QUALIFIER__ thread_block_tile(const thread_block_tile<size, ParentCGTy>& g)
: impl::thread_block_tile_internal<size, void>(g) {}
};
template <unsigned int size, class ParentCGTy = void> class thread_block_tile;
namespace impl {
template <unsigned int size, class ParentCGTy = void> struct tiled_partition_internal;
template <unsigned int size>
struct tiled_partition_internal<size, thread_block> : public thread_block_tile<size, thread_block> {
__CG_QUALIFIER__ tiled_partition_internal(const thread_block& g)
: thread_block_tile<size, thread_block>(g) {}
};
} // namespace impl
/** \brief User exposed API to partition groups.
*
* \details This constructs a templated class derieved from thread_group.
* The template defines tile size of the new thread group at compile time.
*/
template <unsigned int size, class ParentCGTy>
__CG_QUALIFIER__ thread_block_tile<size, ParentCGTy> tiled_partition(const ParentCGTy& g) {
static_assert(is_valid_tile_size<size>::value,
"Tiled partition with size > wavefront size. Currently not supported ");
return impl::tiled_partition_internal<size, ParentCGTy>(g);
}
} // namespace cooperative_groups
#endif // __cplusplus
#endif // HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COOPERATIVE_GROUPS_H
include/hip/amd_detail/hip_cooperative_groups_helper.h
-180
View File
@@ -1,180 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/**
* @file amd_detail/hip_cooperative_groups_helper.h
*
* @brief Device side implementation of cooperative group feature.
*
* Defines helper constructs and APIs which aid the types and device API
* wrappers defined within `amd_detail/hip_cooperative_groups.h`.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COOPERATIVE_GROUPS_HELPER_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COOPERATIVE_GROUPS_HELPER_H
#if __cplusplus
#include <hip/amd_detail/hip_runtime_api.h>
#include <hip/amd_detail/device_functions.h>
#if !defined(__align__)
#define __align__(x) __attribute__((aligned(x)))
#endif
#if !defined(__CG_QUALIFIER__)
#define __CG_QUALIFIER__ __device__ __forceinline__
#endif
#if !defined(__CG_STATIC_QUALIFIER__)
#define __CG_STATIC_QUALIFIER__ __device__ static __forceinline__
#endif
#if !defined(_CG_STATIC_CONST_DECL_)
#define _CG_STATIC_CONST_DECL_ static constexpr
#endif
#if !defined(WAVEFRONT_SIZE)
#if __gfx1010__ || __gfx1011__ || __gfx1012__ || __gfx1030__ || __gfx1031__
#define WAVEFRONT_SIZE 32
#else
#define WAVEFRONT_SIZE 64
#endif
namespace cooperative_groups {
/* Global scope */
template <unsigned int size>
using is_power_of_2 = std::integral_constant<bool, (size & (size - 1)) == 0>;
template <unsigned int size>
using is_valid_wavefront = std::integral_constant<bool, (size <= WAVEFRONT_SIZE)>;
template <unsigned int size>
using is_valid_tile_size =
std::integral_constant<bool, is_power_of_2<size>::value && is_valid_wavefront<size>::value>;
template <typename T>
using is_valid_type =
std::integral_constant<bool, std::is_integral<T>::value || std::is_floating_point<T>::value>;
namespace internal {
/** \brief Enums representing different cooperative group types
*/
typedef enum {
cg_invalid,
cg_multi_grid,
cg_grid,
cg_workgroup,
cg_tiled_group
} group_type;
/**
* Functionalities related to multi-grid cooperative group type
*/
namespace multi_grid {
__CG_STATIC_QUALIFIER__ uint32_t num_grids() { return (uint32_t)__ockl_multi_grid_num_grids(); }
__CG_STATIC_QUALIFIER__ uint32_t grid_rank() { return (uint32_t)__ockl_multi_grid_grid_rank(); }
__CG_STATIC_QUALIFIER__ uint32_t size() { return (uint32_t)__ockl_multi_grid_size(); }
__CG_STATIC_QUALIFIER__ uint32_t thread_rank() { return (uint32_t)__ockl_multi_grid_thread_rank(); }
__CG_STATIC_QUALIFIER__ bool is_valid() { return (bool)__ockl_multi_grid_is_valid(); }
__CG_STATIC_QUALIFIER__ void sync() { __ockl_multi_grid_sync(); }
} // namespace multi_grid
/**
* Functionalities related to grid cooperative group type
*/
namespace grid {
__CG_STATIC_QUALIFIER__ uint32_t size() {
return (uint32_t)((hipBlockDim_z * hipGridDim_z) * (hipBlockDim_y * hipGridDim_y) *
(hipBlockDim_x * hipGridDim_x));
}
__CG_STATIC_QUALIFIER__ uint32_t thread_rank() {
// Compute global id of the workgroup to which the current thread belongs to
uint32_t blkIdx = (uint32_t)((hipBlockIdx_z * hipGridDim_y * hipGridDim_x) +
(hipBlockIdx_y * hipGridDim_x) + (hipBlockIdx_x));
// Compute total number of threads being passed to reach current workgroup
// within grid
uint32_t num_threads_till_current_workgroup =
(uint32_t)(blkIdx * (hipBlockDim_x * hipBlockDim_y * hipBlockDim_z));
// Compute thread local rank within current workgroup
uint32_t local_thread_rank = (uint32_t)((hipThreadIdx_z * hipBlockDim_y * hipBlockDim_x) +
(hipThreadIdx_y * hipBlockDim_x) + (hipThreadIdx_x));
return (num_threads_till_current_workgroup + local_thread_rank);
}
__CG_STATIC_QUALIFIER__ bool is_valid() { return (bool)__ockl_grid_is_valid(); }
__CG_STATIC_QUALIFIER__ void sync() { __ockl_grid_sync(); }
} // namespace grid
/**
* Functionalities related to `workgroup` (thread_block in CUDA terminology)
* cooperative group type
*/
namespace workgroup {
__CG_STATIC_QUALIFIER__ dim3 group_index() {
return (dim3((uint32_t)hipBlockIdx_x, (uint32_t)hipBlockIdx_y, (uint32_t)hipBlockIdx_z));
}
__CG_STATIC_QUALIFIER__ dim3 thread_index() {
return (dim3((uint32_t)hipThreadIdx_x, (uint32_t)hipThreadIdx_y, (uint32_t)hipThreadIdx_z));
}
__CG_STATIC_QUALIFIER__ uint32_t size() {
return ((uint32_t)(hipBlockDim_x * hipBlockDim_y * hipBlockDim_z));
}
__CG_STATIC_QUALIFIER__ uint32_t thread_rank() {
return ((uint32_t)((hipThreadIdx_z * hipBlockDim_y * hipBlockDim_x) +
(hipThreadIdx_y * hipBlockDim_x) + (hipThreadIdx_x)));
}
__CG_STATIC_QUALIFIER__ bool is_valid() {
// TODO(mahesha) any functionality need to be added here? I believe not
return true;
}
__CG_STATIC_QUALIFIER__ void sync() { __syncthreads(); }
} // namespace workgroup
} // namespace internal
} // namespace cooperative_groups
#endif // __cplusplus
#endif // HIP_INCLUDE_HIP_AMD_DETAIL_HIP_COOPERATIVE_GROUPS_HELPER_H
#endif
include/hip/amd_detail/hip_fp16.h
-1662
View File
File diff suppressed because it is too large Load Diff
include/hip/amd_detail/hip_fp16_gcc.h
-254
View File
@@ -1,254 +0,0 @@
#pragma once
#if defined(__cplusplus)
#include <cstring>
#endif
struct __half_raw {
unsigned short x;
};
struct __half2_raw {
unsigned short x;
unsigned short y;
};
#if defined(__cplusplus)
struct __half;
__half __float2half(float);
float __half2float(__half);
// BEGIN STRUCT __HALF
struct __half {
protected:
unsigned short __x;
public:
// CREATORS
__half() = default;
__half(const __half_raw& x) : __x{x.x} {}
#if !defined(__HIP_NO_HALF_CONVERSIONS__)
__half(float x) : __x{__float2half(x).__x} {}
__half(double x) : __x{__float2half(x).__x} {}
#endif
__half(const __half&) = default;
__half(__half&&) = default;
~__half() = default;
// MANIPULATORS
__half& operator=(const __half&) = default;
__half& operator=(__half&&) = default;
__half& operator=(const __half_raw& x) { __x = x.x; return *this; }
#if !defined(__HIP_NO_HALF_CONVERSIONS__)
__half& operator=(float x)
{
__x = __float2half(x).__x;
return *this;
}
__half& operator=(double x)
{
return *this = static_cast<float>(x);
}
#endif
// ACCESSORS
operator float() const { return __half2float(*this); }
operator __half_raw() const { return __half_raw{__x}; }
};
// END STRUCT __HALF
// BEGIN STRUCT __HALF2
struct __half2 {
public:
__half x;
__half y;
// CREATORS
__half2() = default;
__half2(const __half2_raw& ix)
:
x{reinterpret_cast<const __half&>(ix.x)},
y{reinterpret_cast<const __half&>(ix.y)}
{}
__half2(const __half& ix, const __half& iy) : x{ix}, y{iy} {}
__half2(const __half2&) = default;
__half2(__half2&&) = default;
~__half2() = default;
// MANIPULATORS
__half2& operator=(const __half2&) = default;
__half2& operator=(__half2&&) = default;
__half2& operator=(const __half2_raw& ix)
{
x = reinterpret_cast<const __half_raw&>(ix.x);
y = reinterpret_cast<const __half_raw&>(ix.y);
return *this;
}
// ACCESSORS
operator __half2_raw() const
{
return __half2_raw{
reinterpret_cast<const unsigned short&>(x),
reinterpret_cast<const unsigned short&>(y)};
}
};
// END STRUCT __HALF2
inline
unsigned short __internal_float2half(
float flt, unsigned int& sgn, unsigned int& rem)
{
unsigned int x{};
std::memcpy(&x, &flt, sizeof(flt));
unsigned int u = (x & 0x7fffffffU);
sgn = ((x >> 16) & 0x8000U);
// NaN/+Inf/-Inf
if (u >= 0x7f800000U) {
rem = 0;
return static_cast<unsigned short>(
(u == 0x7f800000U) ? (sgn | 0x7c00U) : 0x7fffU);
}
// Overflows
if (u > 0x477fefffU) {
rem = 0x80000000U;
return static_cast<unsigned short>(sgn | 0x7bffU);
}
// Normal numbers
if (u >= 0x38800000U) {
rem = u << 19;
u -= 0x38000000U;
return static_cast<unsigned short>(sgn | (u >> 13));
}
// +0/-0
if (u < 0x33000001U) {
rem = u;
return static_cast<unsigned short>(sgn);
}
// Denormal numbers
unsigned int exponent = u >> 23;
unsigned int mantissa = (u & 0x7fffffU);
unsigned int shift = 0x7eU - exponent;
mantissa |= 0x800000U;
rem = mantissa << (32 - shift);
return static_cast<unsigned short>(sgn | (mantissa >> shift));
}
inline
__half __float2half(float x)
{
__half_raw r;
unsigned int sgn{};
unsigned int rem{};
r.x = __internal_float2half(x, sgn, rem);
if (rem > 0x80000000U || (rem == 0x80000000U && (r.x & 0x1))) ++r.x;
return r;
}
inline
__half __float2half_rn(float x) { return __float2half(x); }
inline
__half __float2half_rz(float x)
{
__half_raw r;
unsigned int sgn{};
unsigned int rem{};
r.x = __internal_float2half(x, sgn, rem);
return r;
}
inline
__half __float2half_rd(float x)
{
__half_raw r;
unsigned int sgn{};
unsigned int rem{};
r.x = __internal_float2half(x, sgn, rem);
if (rem && sgn) ++r.x;
return r;
}
inline
__half __float2half_ru(float x)
{
__half_raw r;
unsigned int sgn{};
unsigned int rem{};
r.x = __internal_float2half(x, sgn, rem);
if (rem && !sgn) ++r.x;
return r;
}
inline
__half2 __float2half2_rn(float x)
{
return __half2{__float2half_rn(x), __float2half_rn(x)};
}
inline
__half2 __floats2half2_rn(float x, float y)
{
return __half2{__float2half_rn(x), __float2half_rn(y)};
}
inline
float __internal_half2float(unsigned short x)
{
unsigned int sign = ((x >> 15) & 1);
unsigned int exponent = ((x >> 10) & 0x1f);
unsigned int mantissa = ((x & 0x3ff) << 13);
if (exponent == 0x1fU) { /* NaN or Inf */
mantissa = (mantissa ? (sign = 0, 0x7fffffU) : 0);
exponent = 0xffU;
} else if (!exponent) { /* Denorm or Zero */
if (mantissa) {
unsigned int msb;
exponent = 0x71U;
do {
msb = (mantissa & 0x400000U);
mantissa <<= 1; /* normalize */
--exponent;
} while (!msb);
mantissa &= 0x7fffffU; /* 1.mantissa is implicit */
}
} else {
exponent += 0x70U;
}
unsigned int u = ((sign << 31) | (exponent << 23) | mantissa);
float f;
memcpy(&f, &u, sizeof(u));
return f;
}
inline
float __half2float(__half x)
{
return __internal_half2float(static_cast<__half_raw>(x).x);
}
inline
float __low2float(__half2 x)
{
return __internal_half2float(static_cast<__half2_raw>(x).x);
}
inline
float __high2float(__half2 x)
{
return __internal_half2float(static_cast<__half2_raw>(x).y);
}
#if !defined(HIP_NO_HALF)
using half = __half;
using half2 = __half2;
#endif
#endif // defined(__cplusplus)
include/hip/amd_detail/hip_fp16_math_fwd.h
-86
View File
@@ -1,86 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
// /*
// Half Math Functions
// */
#include "host_defines.h"
#if !__CLANG_HIP_RUNTIME_WRAPPER_INCLUDED__
extern "C"
{
__device__ __attribute__((const)) _Float16 __ocml_ceil_f16(_Float16);
__device__ _Float16 __ocml_cos_f16(_Float16);
__device__ __attribute__((pure)) _Float16 __ocml_exp_f16(_Float16);
__device__ __attribute__((pure)) _Float16 __ocml_exp10_f16(_Float16);
__device__ __attribute__((pure)) _Float16 __ocml_exp2_f16(_Float16);
__device__ __attribute__((const)) _Float16 __ocml_floor_f16(_Float16);
__device__ __attribute__((const))
_Float16 __ocml_fma_f16(_Float16, _Float16, _Float16);
__device__ __attribute__((const)) _Float16 __ocml_fabs_f16(_Float16);
__device__ __attribute__((const)) int __ocml_isinf_f16(_Float16);
__device__ __attribute__((const)) int __ocml_isnan_f16(_Float16);
__device__ __attribute__((pure)) _Float16 __ocml_log_f16(_Float16);
__device__ __attribute__((pure)) _Float16 __ocml_log10_f16(_Float16);
__device__ __attribute__((pure)) _Float16 __ocml_log2_f16(_Float16);
__device__ __attribute__((pure)) _Float16 __ocml_pown_f16(_Float16, int);
__device__ __attribute__((const)) _Float16 __llvm_amdgcn_rcp_f16(_Float16);
__device__ __attribute__((const)) _Float16 __ocml_rint_f16(_Float16);
__device__ __attribute__((const)) _Float16 __ocml_rsqrt_f16(_Float16);
__device__ _Float16 __ocml_sin_f16(_Float16);
__device__ __attribute__((const)) _Float16 __ocml_sqrt_f16(_Float16);
__device__ __attribute__((const)) _Float16 __ocml_trunc_f16(_Float16);
typedef _Float16 __2f16 __attribute__((ext_vector_type(2)));
typedef short __2i16 __attribute__((ext_vector_type(2)));
#if __HIP_CLANG_ONLY__
__device__ __attribute__((const)) float __ockl_fdot2(__2f16 a, __2f16 b, float c, bool s);
#endif
__device__ __attribute__((const)) __2f16 __ocml_ceil_2f16(__2f16);
__device__ __attribute__((const)) __2f16 __ocml_fabs_2f16(__2f16);
__device__ __2f16 __ocml_cos_2f16(__2f16);
__device__ __attribute__((pure)) __2f16 __ocml_exp_2f16(__2f16);
__device__ __attribute__((pure)) __2f16 __ocml_exp10_2f16(__2f16);
__device__ __attribute__((pure)) __2f16 __ocml_exp2_2f16(__2f16);
__device__ __attribute__((const)) __2f16 __ocml_floor_2f16(__2f16);
__device__ __attribute__((const)) __2f16 __ocml_fma_2f16(__2f16, __2f16, __2f16);
__device__ __attribute__((const)) __2i16 __ocml_isinf_2f16(__2f16);
__device__ __attribute__((const)) __2i16 __ocml_isnan_2f16(__2f16);
__device__ __attribute__((pure)) __2f16 __ocml_log_2f16(__2f16);
__device__ __attribute__((pure)) __2f16 __ocml_log10_2f16(__2f16);
__device__ __attribute__((pure)) __2f16 __ocml_log2_2f16(__2f16);
__device__ inline
__2f16 __llvm_amdgcn_rcp_2f16(__2f16 x) // Not currently exposed by ROCDL.
{
return __2f16{__llvm_amdgcn_rcp_f16(x.x), __llvm_amdgcn_rcp_f16(x.y)};
}
__device__ __attribute__((const)) __2f16 __ocml_rint_2f16(__2f16);
__device__ __attribute__((const)) __2f16 __ocml_rsqrt_2f16(__2f16);
__device__ __2f16 __ocml_sin_2f16(__2f16);
__device__ __attribute__((const)) __2f16 __ocml_sqrt_2f16(__2f16);
__device__ __attribute__((const)) __2f16 __ocml_trunc_2f16(__2f16);
}
#endif // !__CLANG_HIP_RUNTIME_WRAPPER_INCLUDED__
include/hip/amd_detail/hip_ldg.h
-100
View File
@@ -1,100 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_HIP_LDG_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_HIP_LDG_H
#if __HIP_CLANG_ONLY__
#include "hip_vector_types.h"
#include "host_defines.h"
__device__ inline static char __ldg(const char* ptr) { return *ptr; }
__device__ inline static char2 __ldg(const char2* ptr) { return *ptr; }
__device__ inline static char4 __ldg(const char4* ptr) { return *ptr; }
__device__ inline static signed char __ldg(const signed char* ptr) { return ptr[0]; }
__device__ inline static unsigned char __ldg(const unsigned char* ptr) { return ptr[0]; }
__device__ inline static short __ldg(const short* ptr) { return ptr[0]; }
__device__ inline static short2 __ldg(const short2* ptr) { return ptr[0]; }
__device__ inline static short4 __ldg(const short4* ptr) { return ptr[0]; }
__device__ inline static unsigned short __ldg(const unsigned short* ptr) { return ptr[0]; }
__device__ inline static int __ldg(const int* ptr) { return ptr[0]; }
__device__ inline static int2 __ldg(const int2* ptr) { return ptr[0]; }
__device__ inline static int4 __ldg(const int4* ptr) { return ptr[0]; }
__device__ inline static unsigned int __ldg(const unsigned int* ptr) { return ptr[0]; }
__device__ inline static long __ldg(const long* ptr) { return ptr[0]; }
__device__ inline static unsigned long __ldg(const unsigned long* ptr) { return ptr[0]; }
__device__ inline static long long __ldg(const long long* ptr) { return ptr[0]; }
__device__ inline static longlong2 __ldg(const longlong2* ptr) { return ptr[0]; }
__device__ inline static unsigned long long __ldg(const unsigned long long* ptr) { return ptr[0]; }
__device__ inline static uchar2 __ldg(const uchar2* ptr) { return ptr[0]; }
__device__ inline static uchar4 __ldg(const uchar4* ptr) { return ptr[0]; }
__device__ inline static ushort2 __ldg(const ushort2* ptr) { return ptr[0]; }
__device__ inline static uint2 __ldg(const uint2* ptr) { return ptr[0]; }
__device__ inline static uint4 __ldg(const uint4* ptr) { return ptr[0]; }
__device__ inline static ulonglong2 __ldg(const ulonglong2* ptr) { return ptr[0]; }
__device__ inline static float __ldg(const float* ptr) { return ptr[0]; }
__device__ inline static float2 __ldg(const float2* ptr) { return ptr[0]; }
__device__ inline static float4 __ldg(const float4* ptr) { return ptr[0]; }
__device__ inline static double __ldg(const double* ptr) { return ptr[0]; }
__device__ inline static double2 __ldg(const double2* ptr) { return ptr[0]; }
#endif // __HIP_CLANG_ONLY__
#endif // HIP_LDG_H
include/hip/amd_detail/hip_memory.h
-114
View File
@@ -1,114 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_HIP_MEMORY_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_HIP_MEMORY_H
// Implementation of malloc and free device functions.
// HIP heap is implemented as a global array with fixed size. Users may define
// __HIP_SIZE_OF_PAGE and __HIP_NUM_PAGES to have a larger heap.
#if __HIP__ && __HIP_ENABLE_DEVICE_MALLOC__
// Size of page in bytes.
#ifndef __HIP_SIZE_OF_PAGE
#define __HIP_SIZE_OF_PAGE 64
#endif
// Total number of pages
#ifndef __HIP_NUM_PAGES
#define __HIP_NUM_PAGES (16 * 64 * 64)
#endif
#define __HIP_SIZE_OF_HEAP (__HIP_NUM_PAGES * __HIP_SIZE_OF_PAGE)
#if __HIP_DEVICE_COMPILE__
__attribute__((weak)) __device__ char __hip_device_heap[__HIP_SIZE_OF_HEAP];
__attribute__((weak)) __device__
uint32_t __hip_device_page_flag[__HIP_NUM_PAGES];
#else
extern __device__ char __hip_device_heap[];
extern __device__ uint32_t __hip_device_page_flag[];
#endif
extern "C" inline __device__ void* __hip_malloc(size_t size) {
char* heap = (char*)__hip_device_heap;
if (size > __HIP_SIZE_OF_HEAP) {
return (void*)nullptr;
}
uint32_t totalThreads =
hipBlockDim_x * hipGridDim_x * hipBlockDim_y
* hipGridDim_y * hipBlockDim_z * hipGridDim_z;
uint32_t currentWorkItem = hipThreadIdx_x + hipBlockDim_x * hipBlockIdx_x
+ (hipThreadIdx_y + hipBlockDim_y * hipBlockIdx_y) * hipBlockDim_x
+ (hipThreadIdx_z + hipBlockDim_z * hipBlockIdx_z) * hipBlockDim_x
* hipBlockDim_y;
uint32_t numHeapsPerWorkItem = __HIP_NUM_PAGES / totalThreads;
uint32_t heapSizePerWorkItem = __HIP_SIZE_OF_HEAP / totalThreads;
uint32_t stride = size / __HIP_SIZE_OF_PAGE;
uint32_t start = numHeapsPerWorkItem * currentWorkItem;
uint32_t k = 0;
while (__hip_device_page_flag[k] > 0) {
k++;
}
for (uint32_t i = 0; i < stride - 1; i++) {
__hip_device_page_flag[i + start + k] = 1;
}
__hip_device_page_flag[start + stride - 1 + k] = 2;
void* ptr = (void*)(heap
+ heapSizePerWorkItem * currentWorkItem + k * __HIP_SIZE_OF_PAGE);
return ptr;
}
extern "C" inline __device__ void* __hip_free(void* ptr) {
if (ptr == nullptr) {
return nullptr;
}
uint32_t offsetByte = (uint64_t)ptr - (uint64_t)__hip_device_heap;
uint32_t offsetPage = offsetByte / __HIP_SIZE_OF_PAGE;
while (__hip_device_page_flag[offsetPage] != 0) {
if (__hip_device_page_flag[offsetPage] == 2) {
__hip_device_page_flag[offsetPage] = 0;
offsetPage++;
break;
} else {
__hip_device_page_flag[offsetPage] = 0;
offsetPage++;
}
}
return nullptr;
}
#endif
#endif // HIP_INCLUDE_HIP_AMD_DETAIL_HIP_MEMORY_H
include/hip/amd_detail/hip_runtime.h
-417
View File
@@ -1,417 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/**
* @file amd_detail/hip_runtime.h
* @brief Contains definitions of APIs for HIP runtime.
*/
//#pragma once
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_HIP_RUNTIME_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_HIP_RUNTIME_H
#include <hip/amd_detail/hip_common.h>
//---
// Top part of file can be compiled with any compiler
#if !defined(__HIPCC_RTC__)
//#include <cstring>
#if __cplusplus
#include <cmath>
#include <cstdint>
#else
#include <math.h>
#include <string.h>
#include <stddef.h>
#endif // __cplusplus
#endif // !defined(__HIPCC_RTC__)
// __hip_malloc is not working. Disable it by default.
#ifndef __HIP_ENABLE_DEVICE_MALLOC__
#define __HIP_ENABLE_DEVICE_MALLOC__ 0
#endif
#if __HIP_CLANG_ONLY__
#if !defined(__align__)
#define __align__(x) __attribute__((aligned(x)))
#endif
#define CUDA_SUCCESS hipSuccess
#if !defined(__HIPCC_RTC__)
#include <hip/hip_runtime_api.h>
extern int HIP_TRACE_API;
#endif // !defined(__HIPCC_RTC__)
#ifdef __cplusplus
#include <hip/amd_detail/hip_ldg.h>
#endif
#include <hip/amd_detail/hip_atomic.h>
#include <hip/amd_detail/host_defines.h>
#include <hip/amd_detail/device_functions.h>
#include <hip/amd_detail/surface_functions.h>
#include <hip/amd_detail/texture_fetch_functions.h>
#include <hip/amd_detail/texture_indirect_functions.h>
// TODO-HCC remove old definitions ; ~1602 hcc supports __HCC_ACCELERATOR__ define.
#if defined(__KALMAR_ACCELERATOR__) && !defined(__HCC_ACCELERATOR__)
#define __HCC_ACCELERATOR__ __KALMAR_ACCELERATOR__
#endif
// Feature tests:
#if (defined(__HCC_ACCELERATOR__) && (__HCC_ACCELERATOR__ != 0)) || __HIP_DEVICE_COMPILE__
// Device compile and not host compile:
// 32-bit Atomics:
#define __HIP_ARCH_HAS_GLOBAL_INT32_ATOMICS__ (1)
#define __HIP_ARCH_HAS_GLOBAL_FLOAT_ATOMIC_EXCH__ (1)
#define __HIP_ARCH_HAS_SHARED_INT32_ATOMICS__ (1)
#define __HIP_ARCH_HAS_SHARED_FLOAT_ATOMIC_EXCH__ (1)
#define __HIP_ARCH_HAS_FLOAT_ATOMIC_ADD__ (1)
// 64-bit Atomics:
#define __HIP_ARCH_HAS_GLOBAL_INT64_ATOMICS__ (1)
#define __HIP_ARCH_HAS_SHARED_INT64_ATOMICS__ (1)
// Doubles
#define __HIP_ARCH_HAS_DOUBLES__ (1)
// warp cross-lane operations:
#define __HIP_ARCH_HAS_WARP_VOTE__ (1)
#define __HIP_ARCH_HAS_WARP_BALLOT__ (1)
#define __HIP_ARCH_HAS_WARP_SHUFFLE__ (1)
#define __HIP_ARCH_HAS_WARP_FUNNEL_SHIFT__ (0)
// sync
#define __HIP_ARCH_HAS_THREAD_FENCE_SYSTEM__ (1)
#define __HIP_ARCH_HAS_SYNC_THREAD_EXT__ (0)
// misc
#define __HIP_ARCH_HAS_SURFACE_FUNCS__ (0)
#define __HIP_ARCH_HAS_3DGRID__ (1)
#define __HIP_ARCH_HAS_DYNAMIC_PARALLEL__ (0)
#endif /* Device feature flags */
#define launch_bounds_impl0(requiredMaxThreadsPerBlock) \
__attribute__((amdgpu_flat_work_group_size(1, requiredMaxThreadsPerBlock)))
#define launch_bounds_impl1(requiredMaxThreadsPerBlock, minBlocksPerMultiprocessor) \
__attribute__((amdgpu_flat_work_group_size(1, requiredMaxThreadsPerBlock), \
amdgpu_waves_per_eu(minBlocksPerMultiprocessor)))
#define select_impl_(_1, _2, impl_, ...) impl_
#define __launch_bounds__(...) \
select_impl_(__VA_ARGS__, launch_bounds_impl1, launch_bounds_impl0)(__VA_ARGS__)
#if !defined(__HIPCC_RTC__)
__host__ inline void* __get_dynamicgroupbaseptr() { return nullptr; }
#endif // !defined(__HIPCC_RTC__)
#if __HIP_ARCH_GFX701__ == 0
__device__ unsigned __hip_ds_bpermute(int index, unsigned src);
__device__ float __hip_ds_bpermutef(int index, float src);
__device__ unsigned __hip_ds_permute(int index, unsigned src);
__device__ float __hip_ds_permutef(int index, float src);
template <int pattern>
__device__ unsigned __hip_ds_swizzle_N(unsigned int src);
template <int pattern>
__device__ float __hip_ds_swizzlef_N(float src);
template <int dpp_ctrl, int row_mask, int bank_mask, bool bound_ctrl>
__device__ int __hip_move_dpp_N(int src);
#endif //__HIP_ARCH_GFX803__ == 1
#ifndef __OPENMP_AMDGCN__
#if !__CLANG_HIP_RUNTIME_WRAPPER_INCLUDED__
#if __HIP_ENABLE_DEVICE_MALLOC__
extern "C" __device__ void* __hip_malloc(size_t);
extern "C" __device__ void* __hip_free(void* ptr);
static inline __device__ void* malloc(size_t size) { return __hip_malloc(size); }
static inline __device__ void* free(void* ptr) { return __hip_free(ptr); }
#else
static inline __device__ void* malloc(size_t size) { __builtin_trap(); return nullptr; }
static inline __device__ void* free(void* ptr) { __builtin_trap(); return nullptr; }
#endif
#endif // !__CLANG_HIP_RUNTIME_WRAPPER_INCLUDED__
#endif // !__OPENMP_AMDGCN__
// End doxygen API:
/**
* @}
*/
//
// hip-clang functions
//
#if !defined(__HIPCC_RTC__)
#define HIP_KERNEL_NAME(...) __VA_ARGS__
#define HIP_SYMBOL(X) X
typedef int hipLaunchParm;
template <std::size_t n, typename... Ts,
typename std::enable_if<n == sizeof...(Ts)>::type* = nullptr>
void pArgs(const std::tuple<Ts...>&, void*) {}
template <std::size_t n, typename... Ts,
typename std::enable_if<n != sizeof...(Ts)>::type* = nullptr>
void pArgs(const std::tuple<Ts...>& formals, void** _vargs) {
using T = typename std::tuple_element<n, std::tuple<Ts...> >::type;
static_assert(!std::is_reference<T>{},
"A __global__ function cannot have a reference as one of its "
"arguments.");
#if defined(HIP_STRICT)
static_assert(std::is_trivially_copyable<T>{},
"Only TriviallyCopyable types can be arguments to a __global__ "
"function");
#endif
_vargs[n] = const_cast<void*>(reinterpret_cast<const void*>(&std::get<n>(formals)));
return pArgs<n + 1>(formals, _vargs);
}
template <typename... Formals, typename... Actuals>
std::tuple<Formals...> validateArgsCountType(void (*kernel)(Formals...), std::tuple<Actuals...>(actuals)) {
static_assert(sizeof...(Formals) == sizeof...(Actuals), "Argument Count Mismatch");
std::tuple<Formals...> to_formals{std::move(actuals)};
return to_formals;
}
#if defined(HIP_TEMPLATE_KERNEL_LAUNCH)
template <typename... Args, typename F = void (*)(Args...)>
void hipLaunchKernelGGL(F kernel, const dim3& numBlocks, const dim3& dimBlocks,
std::uint32_t sharedMemBytes, hipStream_t stream, Args... args) {
constexpr size_t count = sizeof...(Args);
auto tup_ = std::tuple<Args...>{args...};
auto tup = validateArgsCountType(kernel, tup_);
void* _Args[count];
pArgs<0>(tup, _Args);
auto k = reinterpret_cast<void*>(kernel);
hipLaunchKernel(k, numBlocks, dimBlocks, _Args, sharedMemBytes, stream);
}
#else
#define hipLaunchKernelGGLInternal(kernelName, numBlocks, numThreads, memPerBlock, streamId, ...) \
do { \
kernelName<<<(numBlocks), (numThreads), (memPerBlock), (streamId)>>>(__VA_ARGS__); \
} while (0)
#define hipLaunchKernelGGL(kernelName, ...) hipLaunchKernelGGLInternal((kernelName), __VA_ARGS__)
#endif
#include <hip/hip_runtime_api.h>
#endif // !defined(__HIPCC_RTC__)
extern "C" __device__ __attribute__((const)) size_t __ockl_get_local_id(uint);
extern "C" __device__ __attribute__((const)) size_t __ockl_get_group_id(uint);
extern "C" __device__ __attribute__((const)) size_t __ockl_get_local_size(uint);
extern "C" __device__ __attribute__((const)) size_t __ockl_get_num_groups(uint);
struct __HIP_BlockIdx {
__device__
std::uint32_t operator()(std::uint32_t x) const noexcept { return __ockl_get_group_id(x); }
};
struct __HIP_BlockDim {
__device__
std::uint32_t operator()(std::uint32_t x) const noexcept {
return __ockl_get_local_size(x);
}
};
struct __HIP_GridDim {
__device__
std::uint32_t operator()(std::uint32_t x) const noexcept {
return __ockl_get_num_groups(x);
}
};
struct __HIP_ThreadIdx {
__device__
std::uint32_t operator()(std::uint32_t x) const noexcept {
return __ockl_get_local_id(x);
}
};
#if defined(__HIPCC_RTC__)
typedef struct dim3 {
uint32_t x; ///< x
uint32_t y; ///< y
uint32_t z; ///< z
#ifdef __cplusplus
constexpr __device__ dim3(uint32_t _x = 1, uint32_t _y = 1, uint32_t _z = 1) : x(_x), y(_y), z(_z){};
#endif
} dim3;
#endif // !defined(__HIPCC_RTC__)
template <typename F>
struct __HIP_Coordinates {
using R = decltype(F{}(0));
struct __X { __device__ operator R() const noexcept { return F{}(0); } };
struct __Y { __device__ operator R() const noexcept { return F{}(1); } };
struct __Z { __device__ operator R() const noexcept { return F{}(2); } };
static constexpr __X x{};
static constexpr __Y y{};
static constexpr __Z z{};
#ifdef __cplusplus
__device__ operator dim3() const { return dim3(x, y, z); }
#endif
};
template <typename F>
#if !defined(_MSC_VER)
__attribute__((weak))
#endif
constexpr typename __HIP_Coordinates<F>::__X __HIP_Coordinates<F>::x;
template <typename F>
#if !defined(_MSC_VER)
__attribute__((weak))
#endif
constexpr typename __HIP_Coordinates<F>::__Y __HIP_Coordinates<F>::y;
template <typename F>
#if !defined(_MSC_VER)
__attribute__((weak))
#endif
constexpr typename __HIP_Coordinates<F>::__Z __HIP_Coordinates<F>::z;
extern "C" __device__ __attribute__((const)) size_t __ockl_get_global_size(uint);
inline
__device__
std::uint32_t operator*(__HIP_Coordinates<__HIP_GridDim>::__X,
__HIP_Coordinates<__HIP_BlockDim>::__X) noexcept {
return __ockl_get_global_size(0);
}
inline
__device__
std::uint32_t operator*(__HIP_Coordinates<__HIP_BlockDim>::__X,
__HIP_Coordinates<__HIP_GridDim>::__X) noexcept {
return __ockl_get_global_size(0);
}
inline
__device__
std::uint32_t operator*(__HIP_Coordinates<__HIP_GridDim>::__Y,
__HIP_Coordinates<__HIP_BlockDim>::__Y) noexcept {
return __ockl_get_global_size(1);
}
inline
__device__
std::uint32_t operator*(__HIP_Coordinates<__HIP_BlockDim>::__Y,
__HIP_Coordinates<__HIP_GridDim>::__Y) noexcept {
return __ockl_get_global_size(1);
}
inline
__device__
std::uint32_t operator*(__HIP_Coordinates<__HIP_GridDim>::__Z,
__HIP_Coordinates<__HIP_BlockDim>::__Z) noexcept {
return __ockl_get_global_size(2);
}
inline
__device__
std::uint32_t operator*(__HIP_Coordinates<__HIP_BlockDim>::__Z,
__HIP_Coordinates<__HIP_GridDim>::__Z) noexcept {
return __ockl_get_global_size(2);
}
static constexpr __HIP_Coordinates<__HIP_BlockDim> blockDim{};
static constexpr __HIP_Coordinates<__HIP_BlockIdx> blockIdx{};
static constexpr __HIP_Coordinates<__HIP_GridDim> gridDim{};
static constexpr __HIP_Coordinates<__HIP_ThreadIdx> threadIdx{};
extern "C" __device__ __attribute__((const)) size_t __ockl_get_local_id(uint);
#define hipThreadIdx_x (__ockl_get_local_id(0))
#define hipThreadIdx_y (__ockl_get_local_id(1))
#define hipThreadIdx_z (__ockl_get_local_id(2))
extern "C" __device__ __attribute__((const)) size_t __ockl_get_group_id(uint);
#define hipBlockIdx_x (__ockl_get_group_id(0))
#define hipBlockIdx_y (__ockl_get_group_id(1))
#define hipBlockIdx_z (__ockl_get_group_id(2))
extern "C" __device__ __attribute__((const)) size_t __ockl_get_local_size(uint);
#define hipBlockDim_x (__ockl_get_local_size(0))
#define hipBlockDim_y (__ockl_get_local_size(1))
#define hipBlockDim_z (__ockl_get_local_size(2))
extern "C" __device__ __attribute__((const)) size_t __ockl_get_num_groups(uint);
#define hipGridDim_x (__ockl_get_num_groups(0))
#define hipGridDim_y (__ockl_get_num_groups(1))
#define hipGridDim_z (__ockl_get_num_groups(2))
#include <hip/amd_detail/math_functions.h>
#if __HIP_HCC_COMPAT_MODE__
// Define HCC work item functions in terms of HIP builtin variables.
#pragma push_macro("__DEFINE_HCC_FUNC")
#define __DEFINE_HCC_FUNC(hc_fun,hip_var) \
inline __device__ __attribute__((always_inline)) uint hc_get_##hc_fun(uint i) { \
if (i==0) \
return hip_var.x; \
else if(i==1) \
return hip_var.y; \
else \
return hip_var.z; \
}
__DEFINE_HCC_FUNC(workitem_id, threadIdx)
__DEFINE_HCC_FUNC(group_id, blockIdx)
__DEFINE_HCC_FUNC(group_size, blockDim)
__DEFINE_HCC_FUNC(num_groups, gridDim)
#pragma pop_macro("__DEFINE_HCC_FUNC")
extern "C" __device__ __attribute__((const)) size_t __ockl_get_global_id(uint);
inline __device__ __attribute__((always_inline)) uint
hc_get_workitem_absolute_id(int dim)
{
return (uint)__ockl_get_global_id(dim);
}
#endif
#if !__CLANG_HIP_RUNTIME_WRAPPER_INCLUDED__
#if !defined(__HIPCC_RTC__)
// Support std::complex.
#if !_OPENMP || __HIP_ENABLE_CUDA_WRAPPER_FOR_OPENMP__
#pragma push_macro("__CUDA__")
#define __CUDA__
#include <__clang_cuda_math_forward_declares.h>
#include <__clang_cuda_complex_builtins.h>
// Workaround for using libc++ with HIP-Clang.
// The following headers requires clang include path before standard C++ include path.
// However libc++ include path requires to be before clang include path.
// To workaround this, we pass -isystem with the parent directory of clang include
// path instead of the clang include path itself.
#include <include/cuda_wrappers/algorithm>
#include <include/cuda_wrappers/complex>
#include <include/cuda_wrappers/new>
#undef __CUDA__
#pragma pop_macro("__CUDA__")
#endif // !_OPENMP || __HIP_ENABLE_CUDA_WRAPPER_FOR_OPENMP__
#endif // !defined(__HIPCC_RTC__)
#endif // !__CLANG_HIP_RUNTIME_WRAPPER_INCLUDED__
#endif // __HIP_CLANG_ONLY__
#include <hip/amd_detail/hip_memory.h>
#endif // HIP_AMD_DETAIL_RUNTIME_H
include/hip/amd_detail/hip_runtime_prof.h
-77
View File
@@ -1,77 +0,0 @@
/*
Copyright (c) 2019 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_HIP_RUNTIME_PROF_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_HIP_RUNTIME_PROF_H
// HIP ROCclr Op IDs enumeration
enum HipVdiOpId {
kHipVdiOpIdDispatch = 0,
kHipVdiOpIdCopy = 1,
kHipVdiOpIdBarrier = 2,
kHipVdiOpIdNumber = 3
};
// Types of ROCclr commands
enum HipVdiCommandKind {
kHipVdiCommandKernel = 0x11F0,
kHipVdiMemcpyDeviceToHost = 0x11F3,
kHipHipVdiMemcpyHostToDevice = 0x11F4,
kHipVdiMemcpyDeviceToDevice = 0x11F5,
kHipVidMemcpyDeviceToHostRect = 0x1201,
kHipVdiMemcpyHostToDeviceRect = 0x1202,
kHipVdiMemcpyDeviceToDeviceRect = 0x1203,
kHipVdiFillMemory = 0x1207,
};
/**
* @brief Initializes activity callback
*
* @param [input] id_callback Event ID callback function
* @param [input] op_callback Event operation callback function
* @param [input] arg Arguments passed into callback
*
* @returns None
*/
void hipInitActivityCallback(void* id_callback, void* op_callback, void* arg);
/**
* @brief Enables activity callback
*
* @param [input] op Operation, which will trigger a callback (@see HipVdiOpId)
* @param [input] enable Enable state for the callback
*
* @returns True if successful
*/
bool hipEnableActivityCallback(uint32_t op, bool enable);
/**
* @brief Returns the description string for the operation kind
*
* @param [input] id Command kind id (@see HipVdiCommandKind)
*
* @returns A pointer to a const string with the command description
*/
const char* hipGetCmdName(uint32_t id);
#endif // HIP_INCLUDE_HIP_AMD_DETAIL_HIP_RUNTIME_PROF_H
include/hip/amd_detail/hip_surface_types.h
-54
View File
@@ -1,54 +0,0 @@
/*
Copyright (c) 2015- present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/**
* @file amd_detail/hip_surface_types.h
* @brief Defines surface types for HIP runtime.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_HIP_SURFACE_TYPES_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_HIP_SURFACE_TYPES_H
#include <hip/amd_detail/driver_types.h>
/**
* An opaque value that represents a hip surface object
*/
typedef unsigned long long hipSurfaceObject_t;
/**
* hip surface reference
*/
struct surfaceReference {
hipSurfaceObject_t surfaceObject;
};
/**
* hip surface boundary modes
*/
enum hipSurfaceBoundaryMode {
hipBoundaryModeZero = 0,
hipBoundaryModeTrap = 1,
hipBoundaryModeClamp = 2
};
#endif /* !HIP_INCLUDE_HIP_AMD_DETAIL_HIP_SURFACE_TYPES_H */
include/hip/amd_detail/hip_texture_types.h
-97
View File
@@ -1,97 +0,0 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/**
* @file amd_detail/hip_texture_types.h
* @brief Defines the different newt vector types for HIP runtime.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_HIP_TEXTURE_TYPES_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_HIP_TEXTURE_TYPES_H
/*******************************************************************************
* *
* *
* *
*******************************************************************************/
#if !defined(__HIPCC_RTC__)
#include <limits.h>
#include <hip/amd_detail/channel_descriptor.h>
#endif // !defined(__HIPCC_RTC__)
#include <hip/amd_detail/texture_types.h>
#if __cplusplus
/*******************************************************************************
* *
* *
* *
*******************************************************************************/
#if __HIP__
#define __HIP_TEXTURE_ATTRIB __attribute__((device_builtin_texture_type))
#else
#define __HIP_TEXTURE_ATTRIB
#endif
typedef textureReference* hipTexRef;
template <class T, int texType = hipTextureType1D,
enum hipTextureReadMode mode = hipReadModeElementType>
struct __HIP_TEXTURE_ATTRIB texture : public textureReference {
texture(int norm = 0, enum hipTextureFilterMode fMode = hipFilterModePoint,
enum hipTextureAddressMode aMode = hipAddressModeClamp) {
normalized = norm;
readMode = mode;
filterMode = fMode;
addressMode[0] = aMode;
addressMode[1] = aMode;
addressMode[2] = aMode;
channelDesc = hipCreateChannelDesc<T>();
sRGB = 0;
textureObject = nullptr;
maxAnisotropy = 0;
mipmapLevelBias = 0;
minMipmapLevelClamp = 0;
maxMipmapLevelClamp = 0;
}
texture(int norm, enum hipTextureFilterMode fMode, enum hipTextureAddressMode aMode,
struct hipChannelFormatDesc desc) {
normalized = norm;
readMode = mode;
filterMode = fMode;
addressMode[0] = aMode;
addressMode[1] = aMode;
addressMode[2] = aMode;
channelDesc = desc;
sRGB = 0;
textureObject = nullptr;
maxAnisotropy = 0;
mipmapLevelBias = 0;
minMipmapLevelClamp = 0;
maxMipmapLevelClamp = 0;
}
};
#endif /* __cplusplus */
#endif /* !HIP_INCLUDE_HIP_AMD_DETAIL_HIP_TEXTURE_TYPES_H */
include/hip/amd_detail/hip_vector_types.h
-1598
View File
File diff suppressed because it is too large Load Diff
include/hip/amd_detail/hiprtc.h
-94
View File
@@ -1,94 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIPRTC_H
#define HIPRTC_H
#ifdef __cplusplus
extern "C" {
#endif /* __cplusplus */
#include <stdlib.h>
#if !defined(_WIN32)
#pragma GCC visibility push (default)
#endif
enum hiprtcResult {
HIPRTC_SUCCESS = 0,
HIPRTC_ERROR_OUT_OF_MEMORY = 1,
HIPRTC_ERROR_PROGRAM_CREATION_FAILURE = 2,
HIPRTC_ERROR_INVALID_INPUT = 3,
HIPRTC_ERROR_INVALID_PROGRAM = 4,
HIPRTC_ERROR_INVALID_OPTION = 5,
HIPRTC_ERROR_COMPILATION = 6,
HIPRTC_ERROR_BUILTIN_OPERATION_FAILURE = 7,
HIPRTC_ERROR_NO_NAME_EXPRESSIONS_AFTER_COMPILATION = 8,
HIPRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION = 9,
HIPRTC_ERROR_NAME_EXPRESSION_NOT_VALID = 10,
HIPRTC_ERROR_INTERNAL_ERROR = 11
};
const char* hiprtcGetErrorString(hiprtcResult result);
hiprtcResult hiprtcVersion(int* major, int* minor);
typedef struct _hiprtcProgram* hiprtcProgram;
hiprtcResult hiprtcAddNameExpression(hiprtcProgram prog,
const char* name_expression);
hiprtcResult hiprtcCompileProgram(hiprtcProgram prog,
int numOptions,
const char** options);
hiprtcResult hiprtcCreateProgram(hiprtcProgram* prog,
const char* src,
const char* name,
int numHeaders,
const char** headers,
const char** includeNames);
hiprtcResult hiprtcDestroyProgram(hiprtcProgram* prog);
hiprtcResult hiprtcGetLoweredName(hiprtcProgram prog,
const char* name_expression,
const char** lowered_name);
hiprtcResult hiprtcGetProgramLog(hiprtcProgram prog, char* log);
hiprtcResult hiprtcGetProgramLogSize(hiprtcProgram prog,
size_t* logSizeRet);
hiprtcResult hiprtcGetCode(hiprtcProgram prog, char* code);
hiprtcResult hiprtcGetCodeSize(hiprtcProgram prog, size_t* codeSizeRet);
#if !defined(_WIN32)
#pragma GCC visibility pop
#endif
#ifdef __cplusplus
}
#endif /* __cplusplus */
#endif //HIPRTC_H
include/hip/amd_detail/host_defines.h
-72
View File
@@ -1,72 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/**
* @file amd_detail/host_defines.h
* @brief TODO-doc
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_HOST_DEFINES_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_HOST_DEFINES_H
// The follow macro should be removed after upstream updation.
// It's defined here for workarround of rocThrust building failure.
#define HIP_INCLUDE_HIP_HCC_DETAIL_HOST_DEFINES_H
// Add guard to Generic Grid Launch method
#ifndef GENERIC_GRID_LAUNCH
#define GENERIC_GRID_LAUNCH 1
#endif
#if defined(__clang__) && defined(__HIP__)
#if !__CLANG_HIP_RUNTIME_WRAPPER_INCLUDED__
#define __host__ __attribute__((host))
#define __device__ __attribute__((device))
#define __global__ __attribute__((global))
#define __shared__ __attribute__((shared))
#define __constant__ __attribute__((constant))
#endif // !__CLANG_HIP_RUNTIME_WRAPPER_INCLUDED__
#define __noinline__ __attribute__((noinline))
#define __forceinline__ inline __attribute__((always_inline))
#else
// Non-HCC compiler
/**
* Function and kernel markers
*/
#define __host__
#define __device__
#define __global__
#define __noinline__
#define __forceinline__ inline
#define __shared__
#define __constant__
#endif
#endif
include/hip/amd_detail/hsa_helpers.hpp
-102
View File
@@ -1,102 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#include <hsa/hsa.h>
#include <cstdint>
#include <functional>
#include <string>
namespace hip_impl {
inline void* address(hsa_executable_symbol_t x) {
void* r = nullptr;
hsa_executable_symbol_get_info(x, HSA_EXECUTABLE_SYMBOL_INFO_VARIABLE_ADDRESS, &r);
return r;
}
inline hsa_agent_t agent(hsa_executable_symbol_t x) {
hsa_agent_t r = {};
hsa_executable_symbol_get_info(x, HSA_EXECUTABLE_SYMBOL_INFO_AGENT, &r);
return r;
}
inline std::uint32_t group_size(hsa_executable_symbol_t x) {
std::uint32_t r = 0u;
hsa_executable_symbol_get_info(x, HSA_EXECUTABLE_SYMBOL_INFO_KERNEL_GROUP_SEGMENT_SIZE, &r);
return r;
}
inline hsa_isa_t isa(hsa_agent_t x) {
hsa_isa_t r = {};
hsa_agent_iterate_isas(x,
[](hsa_isa_t i, void* o) {
*static_cast<hsa_isa_t*>(o) = i; // Pick the first.
return HSA_STATUS_INFO_BREAK;
},
&r);
return r;
}
inline std::uint64_t kernel_object(hsa_executable_symbol_t x) {
std::uint64_t r = 0u;
hsa_executable_symbol_get_info(x, HSA_EXECUTABLE_SYMBOL_INFO_KERNEL_OBJECT, &r);
return r;
}
inline std::string name(hsa_executable_symbol_t x) {
std::uint32_t sz = 0u;
hsa_executable_symbol_get_info(x, HSA_EXECUTABLE_SYMBOL_INFO_NAME_LENGTH, &sz);
std::string r(sz, '\0');
hsa_executable_symbol_get_info(x, HSA_EXECUTABLE_SYMBOL_INFO_NAME, &r.front());
return r;
}
inline std::uint32_t private_size(hsa_executable_symbol_t x) {
std::uint32_t r = 0u;
hsa_executable_symbol_get_info(x, HSA_EXECUTABLE_SYMBOL_INFO_KERNEL_PRIVATE_SEGMENT_SIZE, &r);
return r;
}
inline std::uint32_t size(hsa_executable_symbol_t x) {
std::uint32_t r = 0;
hsa_executable_symbol_get_info(x, HSA_EXECUTABLE_SYMBOL_INFO_VARIABLE_SIZE, &r);
return r;
}
inline hsa_symbol_kind_t type(hsa_executable_symbol_t x) {
hsa_symbol_kind_t r = {};
hsa_executable_symbol_get_info(x, HSA_EXECUTABLE_SYMBOL_INFO_TYPE, &r);
return r;
}
} // namespace hip_impl
include/hip/amd_detail/library_types.h
-41
View File
@@ -1,41 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_LIBRARY_TYPES_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_LIBRARY_TYPES_H
typedef enum hipDataType {
HIP_R_16F = 2,
HIP_R_32F = 0,
HIP_R_64F = 1,
HIP_C_16F = 6,
HIP_C_32F = 4,
HIP_C_64F = 5
} hipDataType;
typedef enum hipLibraryPropertyType {
HIP_LIBRARY_MAJOR_VERSION,
HIP_LIBRARY_MINOR_VERSION,
HIP_LIBRARY_PATCH_LEVEL
} hipLibraryPropertyType;
#endif
include/hip/amd_detail/llvm_intrinsics.h
-41
View File
@@ -1,41 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/**
* @file amd_detail/llvm_intrinsics.h
* @brief Contains declarations for wrapper functions for llvm intrinsics
* like llvm.amdgcn.s.barrier.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_LLVM_INTRINSICS_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_LLVM_INTRINSICS_H
#include "hip/amd_detail/host_defines.h"
// FIXME: These should all be removed and proper builtins used.
__device__
unsigned __llvm_amdgcn_groupstaticsize() __asm("llvm.amdgcn.groupstaticsize");
__device__
int __llvm_amdgcn_ds_swizzle(int index, int pattern) __asm("llvm.amdgcn.ds.swizzle");
#endif
include/hip/amd_detail/macro_based_grid_launch.hpp
-798
View File
@@ -1,798 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#include "concepts.hpp"
#include "helpers.hpp"
#include "hc.hpp"
#include "hip/hip_ext.h"
#include "hip_runtime.h"
#include <functional>
#include <iostream>
#include <stdexcept>
#include <type_traits>
#include <utility>
namespace hip_impl {
namespace {
struct New_grid_launch_tag {};
struct Old_grid_launch_tag {};
template <typename C, typename D>
class RAII_guard {
D dtor_;
public:
RAII_guard() = default;
RAII_guard(const C& ctor, D dtor) : dtor_{std::move(dtor)} { ctor(); }
RAII_guard(const RAII_guard&) = default;
RAII_guard(RAII_guard&&) = default;
RAII_guard& operator=(const RAII_guard&) = default;
RAII_guard& operator=(RAII_guard&&) = default;
~RAII_guard() { dtor_(); }
};
template <typename C, typename D>
RAII_guard<C, D> make_RAII_guard(const C& ctor, D dtor) {
return RAII_guard<C, D>{ctor, std::move(dtor)};
}
template <FunctionalProcedure F, typename... Ts>
using is_new_grid_launch_t = typename std::conditional<is_callable<F(Ts...)>{}, New_grid_launch_tag,
Old_grid_launch_tag>::type;
} // namespace
// TODO: - dispatch rank should be derived from the domain dimensions passed
// in, and not always assumed to be 3;
template <FunctionalProcedure K, typename... Ts>
requires(Domain<K> ==
{Ts...}) inline void grid_launch_hip_impl_(New_grid_launch_tag, dim3 num_blocks,
dim3 dim_blocks, int group_mem_bytes,
const hc::accelerator_view& acc_v, K k) {
const auto d =
hc::extent<3>{num_blocks.z * dim_blocks.z, num_blocks.y * dim_blocks.y,
num_blocks.x * dim_blocks.x}
.tile_with_dynamic(dim_blocks.z, dim_blocks.y, dim_blocks.x, group_mem_bytes);
try {
hc::parallel_for_each(acc_v, d, k);
} catch (std::exception& ex) {
std::cerr << "Failed in " << __func__ << ", with exception: " << ex.what() << std::endl;
hip_throw(ex);
}
}
// TODO: these are workarounds, they should be removed.
hc::accelerator_view lock_stream_hip_(hipStream_t&, void*&);
void print_prelaunch_trace_(const char*, dim3, dim3, int, hipStream_t);
void unlock_stream_hip_(hipStream_t, void*, const char*, hc::accelerator_view*);
template <FunctionalProcedure K, typename... Ts>
requires(Domain<K> == {Ts...}) inline void grid_launch_hip_impl_(New_grid_launch_tag,
dim3 num_blocks, dim3 dim_blocks,
int group_mem_bytes,
hipStream_t stream,
const char* kernel_name, K k) {
void* lck_stream = nullptr;
auto acc_v = lock_stream_hip_(stream, lck_stream);
auto stream_guard =
make_RAII_guard(std::bind(print_prelaunch_trace_, kernel_name, num_blocks, dim_blocks,
group_mem_bytes, stream),
std::bind(unlock_stream_hip_, stream, lck_stream, kernel_name, &acc_v));
try {
grid_launch_hip_impl_(New_grid_launch_tag{}, std::move(num_blocks), std::move(dim_blocks),
group_mem_bytes, acc_v, std::move(k));
} catch (std::exception& ex) {
std::cerr << "Failed in " << __func__ << ", with exception: " << ex.what() << std::endl;
hip_throw(ex);
}
}
template <FunctionalProcedure K, typename... Ts>
requires(Domain<K> ==
{hipLaunchParm, Ts...}) inline void grid_launch_hip_impl_(Old_grid_launch_tag,
dim3 num_blocks, dim3 dim_blocks,
int group_mem_bytes,
hipStream_t stream, K k) {
grid_launch_hip_impl_(New_grid_launch_tag{}, std::move(num_blocks), std::move(dim_blocks),
group_mem_bytes, std::move(stream), std::move(k));
}
template <FunctionalProcedure K, typename... Ts>
requires(Domain<K> == {hipLaunchParm, Ts...}) inline void grid_launch_hip_impl_(
Old_grid_launch_tag, dim3 num_blocks, dim3 dim_blocks, int group_mem_bytes, hipStream_t stream,
const char* kernel_name, K k) {
grid_launch_hip_impl_(New_grid_launch_tag{}, std::move(num_blocks), std::move(dim_blocks),
group_mem_bytes, std::move(stream), kernel_name, std::move(k));
}
template <FunctionalProcedure K, typename... Ts>
requires(Domain<K> == {Ts...}) inline std::enable_if_t<
!std::is_function<K>::value> grid_launch_hip_(dim3 num_blocks, dim3 dim_blocks,
int group_mem_bytes, hipStream_t stream,
const char* kernel_name, K k) {
grid_launch_hip_impl_(is_new_grid_launch_t<K, Ts...>{}, std::move(num_blocks),
std::move(dim_blocks), group_mem_bytes, std::move(stream), kernel_name,
std::move(k));
}
template <FunctionalProcedure K, typename... Ts>
requires(Domain<K> == {Ts...}) inline std::enable_if_t<
!std::is_function<K>::value> grid_launch_hip_(dim3 num_blocks, dim3 dim_blocks,
int group_mem_bytes, hipStream_t stream, K k) {
grid_launch_hip_impl_(is_new_grid_launch_t<K, Ts...>{}, std::move(num_blocks),
std::move(dim_blocks), group_mem_bytes, std::move(stream), std::move(k));
}
// TODO: these are temporary and purposefully noisy and disruptive.
#define make_kernel_name_hip(k, n) \
HIP_kernel_functor_name_begin##_##k##_##HIP_kernel_functor_name_end##_##n
#define make_kernel_functor_hip_30(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16, p17, p18, p19, p20, p21, \
p22, p23, p24, p25, p26, p27) \
struct make_kernel_name_hip(function_name, 28) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
std::decay_t<decltype(p17)> _p17_; \
std::decay_t<decltype(p18)> _p18_; \
std::decay_t<decltype(p19)> _p19_; \
std::decay_t<decltype(p20)> _p20_; \
std::decay_t<decltype(p21)> _p21_; \
std::decay_t<decltype(p22)> _p22_; \
std::decay_t<decltype(p23)> _p23_; \
std::decay_t<decltype(p24)> _p24_; \
std::decay_t<decltype(p25)> _p25_; \
std::decay_t<decltype(p26)> _p26_; \
std::decay_t<decltype(p27)> _p27_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_, _p17_, _p18_, _p19_, _p20_, _p21_, \
_p22_, _p23_, _p24_, _p25_, _p26_, _p27_); \
} \
}
#define make_kernel_functor_hip_29(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16, p17, p18, p19, p20, p21, \
p22, p23, p24, p25, p26) \
struct make_kernel_name_hip(function_name, 27) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
std::decay_t<decltype(p17)> _p17_; \
std::decay_t<decltype(p18)> _p18_; \
std::decay_t<decltype(p19)> _p19_; \
std::decay_t<decltype(p20)> _p20_; \
std::decay_t<decltype(p21)> _p21_; \
std::decay_t<decltype(p22)> _p22_; \
std::decay_t<decltype(p23)> _p23_; \
std::decay_t<decltype(p24)> _p24_; \
std::decay_t<decltype(p25)> _p25_; \
std::decay_t<decltype(p26)> _p26_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_, _p17_, _p18_, _p19_, _p20_, _p21_, \
_p22_, _p23_, _p24_, _p25_, _p26_); \
} \
}
#define make_kernel_functor_hip_28(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16, p17, p18, p19, p20, p21, \
p22, p23, p24, p25) \
struct make_kernel_name_hip(function_name, 26) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
std::decay_t<decltype(p17)> _p17_; \
std::decay_t<decltype(p18)> _p18_; \
std::decay_t<decltype(p19)> _p19_; \
std::decay_t<decltype(p20)> _p20_; \
std::decay_t<decltype(p21)> _p21_; \
std::decay_t<decltype(p22)> _p22_; \
std::decay_t<decltype(p23)> _p23_; \
std::decay_t<decltype(p24)> _p24_; \
std::decay_t<decltype(p25)> _p25_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_, _p17_, _p18_, _p19_, _p20_, _p21_, \
_p22_, _p23_, _p24_, _p25_); \
} \
}
#define make_kernel_functor_hip_27(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16, p17, p18, p19, p20, p21, \
p22, p23, p24) \
struct make_kernel_name_hip(function_name, 25) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
std::decay_t<decltype(p17)> _p17_; \
std::decay_t<decltype(p18)> _p18_; \
std::decay_t<decltype(p19)> _p19_; \
std::decay_t<decltype(p20)> _p20_; \
std::decay_t<decltype(p21)> _p21_; \
std::decay_t<decltype(p22)> _p22_; \
std::decay_t<decltype(p23)> _p23_; \
std::decay_t<decltype(p24)> _p24_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_, _p17_, _p18_, _p19_, _p20_, _p21_, \
_p22_, _p23_, _p24_); \
} \
}
#define make_kernel_functor_hip_26(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16, p17, p18, p19, p20, p21, \
p22, p23) \
struct make_kernel_name_hip(function_name, 24) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
std::decay_t<decltype(p17)> _p17_; \
std::decay_t<decltype(p18)> _p18_; \
std::decay_t<decltype(p19)> _p19_; \
std::decay_t<decltype(p20)> _p20_; \
std::decay_t<decltype(p21)> _p21_; \
std::decay_t<decltype(p22)> _p22_; \
std::decay_t<decltype(p23)> _p23_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_, _p17_, _p18_, _p19_, _p20_, _p21_, \
_p22_, _p23_); \
} \
}
#define make_kernel_functor_hip_25(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16, p17, p18, p19, p20, p21, \
p22) \
struct make_kernel_name_hip(function_name, 23) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
std::decay_t<decltype(p17)> _p17_; \
std::decay_t<decltype(p18)> _p18_; \
std::decay_t<decltype(p19)> _p19_; \
std::decay_t<decltype(p20)> _p20_; \
std::decay_t<decltype(p21)> _p21_; \
std::decay_t<decltype(p22)> _p22_; \
__attribute__((used, flatten)) void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_, _p17_, _p18_, _p19_, _p20_, _p21_, \
_p22_); \
} \
}
#define make_kernel_functor_hip_24(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16, p17, p18, p19, p20, p21) \
struct make_kernel_name_hip(function_name, 22) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
std::decay_t<decltype(p17)> _p17_; \
std::decay_t<decltype(p18)> _p18_; \
std::decay_t<decltype(p19)> _p19_; \
std::decay_t<decltype(p20)> _p20_; \
std::decay_t<decltype(p21)> _p21_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_, _p17_, _p18_, _p19_, _p20_, _p21_); \
} \
}
#define make_kernel_functor_hip_23(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16, p17, p18, p19, p20) \
struct make_kernel_name_hip(function_name, 21) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
std::decay_t<decltype(p17)> _p17_; \
std::decay_t<decltype(p18)> _p18_; \
std::decay_t<decltype(p19)> _p19_; \
std::decay_t<decltype(p20)> _p20_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_, _p17_, _p18_, _p19_, _p20_); \
} \
}
#define make_kernel_functor_hip_22(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16, p17, p18, p19) \
struct make_kernel_name_hip(function_name, 20) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
std::decay_t<decltype(p17)> _p17_; \
std::decay_t<decltype(p18)> _p18_; \
std::decay_t<decltype(p19)> _p19_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_, _p17_, _p18_, _p19_); \
} \
}
#define make_kernel_functor_hip_21(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16, p17, p18) \
struct make_kernel_name_hip(function_name, 19) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
std::decay_t<decltype(p17)> _p17_; \
std::decay_t<decltype(p18)> _p18_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_, _p17_, _p18_); \
} \
}
#define make_kernel_functor_hip_20(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16, p17) \
struct make_kernel_name_hip(function_name, 18) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
std::decay_t<decltype(p17)> _p17_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_, _p17_); \
} \
}
#define make_kernel_functor_hip_19(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15, p16) \
struct make_kernel_name_hip(function_name, 17) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
std::decay_t<decltype(p16)> _p16_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_, _p16_); \
} \
}
#define make_kernel_functor_hip_18(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14, p15) \
struct make_kernel_name_hip(function_name, 16) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
std::decay_t<decltype(p15)> _p15_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_, _p15_); \
} \
}
#define make_kernel_functor_hip_17(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13, p14) \
struct make_kernel_name_hip(function_name, 15) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
std::decay_t<decltype(p14)> _p14_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_, _p14_); \
} \
}
#define make_kernel_functor_hip_16(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12, p13) \
struct make_kernel_name_hip(function_name, 14) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
std::decay_t<decltype(p13)> _p13_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_, _p13_); \
} \
}
#define make_kernel_functor_hip_15(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11, p12) \
struct make_kernel_name_hip(function_name, 13) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
std::decay_t<decltype(p12)> _p12_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_, \
_p12_); \
} \
}
#define make_kernel_functor_hip_14(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10, p11) \
struct make_kernel_name_hip(function_name, 12) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
std::decay_t<decltype(p11)> _p11_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_, _p11_); \
} \
}
#define make_kernel_functor_hip_13(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9, p10) \
struct make_kernel_name_hip(function_name, 11) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
std::decay_t<decltype(p10)> _p10_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { \
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_, _p10_); \
} \
}
#define make_kernel_functor_hip_12(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, \
p9) \
struct make_kernel_name_hip(function_name, 10) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
std::decay_t<decltype(p9)> _p9_; \
void operator()(const hc::tiled_index<3>&) const \
[[hc]] { kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_); } \
}
#define make_kernel_functor_hip_11(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8) \
struct make_kernel_name_hip(function_name, 9) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
std::decay_t<decltype(p8)> _p8_; \
void operator()(const hc::tiled_index<3>&) const \
[[hc]] { kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_); } \
}
#define make_kernel_functor_hip_10(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6, p7) \
struct make_kernel_name_hip(function_name, 8) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
std::decay_t<decltype(p7)> _p7_; \
void operator()(const hc::tiled_index<3>&) const \
[[hc]] { kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_); } \
}
#define make_kernel_functor_hip_9(function_name, kernel_name, p0, p1, p2, p3, p4, p5, p6) \
struct make_kernel_name_hip(function_name, 7) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
std::decay_t<decltype(p6)> _p6_; \
void operator()(const hc::tiled_index<3>&) const \
[[hc]] { kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_); } \
}
#define make_kernel_functor_hip_8(function_name, kernel_name, p0, p1, p2, p3, p4, p5) \
struct make_kernel_name_hip(function_name, 6) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
std::decay_t<decltype(p5)> _p5_; \
void operator()(const hc::tiled_index<3>&) const \
[[hc]] { kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_); } \
}
#define make_kernel_functor_hip_7(function_name, kernel_name, p0, p1, p2, p3, p4) \
struct make_kernel_name_hip(function_name, 5) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
std::decay_t<decltype(p4)> _p4_; \
void operator()(const hc::tiled_index<3>&) const \
[[hc]] { kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_); } \
}
#define make_kernel_functor_hip_6(function_name, kernel_name, p0, p1, p2, p3) \
struct make_kernel_name_hip(function_name, 4) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
std::decay_t<decltype(p3)> _p3_; \
void operator()(const hc::tiled_index<3>&) const \
[[hc]] { kernel_name(_p0_, _p1_, _p2_, _p3_); } \
}
#define make_kernel_functor_hip_5(function_name, kernel_name, p0, p1, p2) \
struct make_kernel_name_hip(function_name, 3) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
std::decay_t<decltype(p2)> _p2_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { kernel_name(_p0_, _p1_, _p2_); } \
}
#define make_kernel_functor_hip_4(function_name, kernel_name, p0, p1) \
struct make_kernel_name_hip(function_name, 2) { \
std::decay_t<decltype(p0)> _p0_; \
std::decay_t<decltype(p1)> _p1_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { kernel_name(_p0_, _p1_); } \
}
#define fofo(f, n) kernel_prefix_hip##f##kernel_suffix_hip##n
#define make_kernel_functor_hip_3(function_name, kernel_name, p0) \
struct make_kernel_name_hip(function_name, 1) { \
std::decay_t<decltype(p0)> _p0_; \
void operator()(const hc::tiled_index<3>&) const [[hc]] { kernel_name(_p0_); } \
}
#define make_kernel_functor_hip_2(function_name, kernel_name) \
struct make_kernel_name_hip(function_name, 0) { \
void operator()(const hc::tiled_index<3>&)[[hc]] { return kernel_name(hipLaunchParm{}); } \
}
#define make_kernel_functor_hip_1(...)
#define make_kernel_functor_hip_0(...)
#define make_kernel_functor_hip_(...) overload_macro_hip_(make_kernel_functor_hip_, __VA_ARGS__)
#define hipLaunchNamedKernelGGL(function_name, kernel_name, num_blocks, dim_blocks, \
group_mem_bytes, stream, ...) \
do { \
make_kernel_functor_hip_(function_name, kernel_name, __VA_ARGS__) \
hip_kernel_functor_impl_{__VA_ARGS__}; \
hip_impl::grid_launch_hip_(num_blocks, dim_blocks, group_mem_bytes, stream, #kernel_name, \
hip_kernel_functor_impl_); \
} while (0)
#define hipLaunchKernelGGL(kernel_name, num_blocks, dim_blocks, group_mem_bytes, stream, ...) \
do { \
hipLaunchNamedKernelGGL(unnamed, kernel_name, num_blocks, dim_blocks, group_mem_bytes, \
stream, ##__VA_ARGS__); \
} while (0)
#define hipLaunchKernel(kernel_name, num_blocks, dim_blocks, group_mem_bytes, stream, ...) \
do { \
hipLaunchKernelGGL(kernel_name, num_blocks, dim_blocks, group_mem_bytes, stream, \
hipLaunchParm{}, ##__VA_ARGS__); \
} while (0)
} // namespace hip_impl
include/hip/amd_detail/math_functions.h
-1502
View File
File diff suppressed because it is too large Load Diff
include/hip/amd_detail/math_fwd.h
-714
View File
@@ -1,714 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#include "host_defines.h"
#if defined(__cplusplus)
extern "C" {
#endif
// DOT FUNCTIONS
#if __HIP_CLANG_ONLY__
__device__
__attribute__((const))
int __ockl_sdot2(
HIP_vector_base<short, 2>::Native_vec_,
HIP_vector_base<short, 2>::Native_vec_,
int, bool);
__device__
__attribute__((const))
unsigned int __ockl_udot2(
HIP_vector_base<unsigned short, 2>::Native_vec_,
HIP_vector_base<unsigned short, 2>::Native_vec_,
unsigned int, bool);
__device__
__attribute__((const))
int __ockl_sdot4(
HIP_vector_base<char, 4>::Native_vec_,
HIP_vector_base<char, 4>::Native_vec_,
int, bool);
__device__
__attribute__((const))
unsigned int __ockl_udot4(
HIP_vector_base<unsigned char, 4>::Native_vec_,
HIP_vector_base<unsigned char, 4>::Native_vec_,
unsigned int, bool);
__device__
__attribute__((const))
int __ockl_sdot8(int, int, int, bool);
__device__
__attribute__((const))
unsigned int __ockl_udot8(unsigned int, unsigned int, unsigned int, bool);
#endif
#if !__CLANG_HIP_RUNTIME_WRAPPER_INCLUDED__
// BEGIN FLOAT
__device__
__attribute__((const))
float __ocml_acos_f32(float);
__device__
__attribute__((pure))
float __ocml_acosh_f32(float);
__device__
__attribute__((const))
float __ocml_asin_f32(float);
__device__
__attribute__((pure))
float __ocml_asinh_f32(float);
__device__
__attribute__((const))
float __ocml_atan2_f32(float, float);
__device__
__attribute__((const))
float __ocml_atan_f32(float);
__device__
__attribute__((pure))
float __ocml_atanh_f32(float);
__device__
__attribute__((pure))
float __ocml_cbrt_f32(float);
__device__
__attribute__((const))
float __ocml_ceil_f32(float);
__device__
__attribute__((const))
__device__
float __ocml_copysign_f32(float, float);
__device__
float __ocml_cos_f32(float);
__device__
float __ocml_native_cos_f32(float);
__device__
__attribute__((pure))
__device__
float __ocml_cosh_f32(float);
__device__
float __ocml_cospi_f32(float);
__device__
float __ocml_i0_f32(float);
__device__
float __ocml_i1_f32(float);
__device__
__attribute__((pure))
float __ocml_erfc_f32(float);
__device__
__attribute__((pure))
float __ocml_erfcinv_f32(float);
__device__
__attribute__((pure))
float __ocml_erfcx_f32(float);
__device__
__attribute__((pure))
float __ocml_erf_f32(float);
__device__
__attribute__((pure))
float __ocml_erfinv_f32(float);
__device__
__attribute__((pure))
float __ocml_exp10_f32(float);
__device__
__attribute__((pure))
float __ocml_native_exp10_f32(float);
__device__
__attribute__((pure))
float __ocml_exp2_f32(float);
__device__
__attribute__((pure))
float __ocml_exp_f32(float);
__device__
__attribute__((pure))
float __ocml_native_exp_f32(float);
__device__
__attribute__((pure))
float __ocml_expm1_f32(float);
__device__
__attribute__((const))
float __ocml_fabs_f32(float);
__device__
__attribute__((const))
float __ocml_fdim_f32(float, float);
__device__
__attribute__((const))
float __ocml_floor_f32(float);
__device__
__attribute__((const))
float __ocml_fma_f32(float, float, float);
__device__
__attribute__((const))
float __ocml_fmax_f32(float, float);
__device__
__attribute__((const))
float __ocml_fmin_f32(float, float);
__device__
__attribute__((const))
__device__
float __ocml_fmod_f32(float, float);
__device__
float __ocml_frexp_f32(float, __attribute__((address_space(5))) int*);
__device__
__attribute__((const))
float __ocml_hypot_f32(float, float);
__device__
__attribute__((const))
int __ocml_ilogb_f32(float);
__device__
__attribute__((const))
int __ocml_isfinite_f32(float);
__device__
__attribute__((const))
int __ocml_isinf_f32(float);
__device__
__attribute__((const))
int __ocml_isnan_f32(float);
__device__
float __ocml_j0_f32(float);
__device__
float __ocml_j1_f32(float);
__device__
__attribute__((const))
float __ocml_ldexp_f32(float, int);
__device__
float __ocml_lgamma_f32(float);
__device__
__attribute__((pure))
float __ocml_log10_f32(float);
__device__
__attribute__((pure))
float __ocml_native_log10_f32(float);
__device__
__attribute__((pure))
float __ocml_log1p_f32(float);
__device__
__attribute__((pure))
float __ocml_log2_f32(float);
__device__
__attribute__((pure))
float __ocml_native_log2_f32(float);
__device__
__attribute__((const))
float __ocml_logb_f32(float);
__device__
__attribute__((pure))
float __ocml_log_f32(float);
__device__
__attribute__((pure))
float __ocml_native_log_f32(float);
__device__
float __ocml_modf_f32(float, __attribute__((address_space(5))) float*);
__device__
__attribute__((const))
float __ocml_nearbyint_f32(float);
__device__
__attribute__((const))
float __ocml_nextafter_f32(float, float);
__device__
__attribute__((const))
float __ocml_len3_f32(float, float, float);
__device__
__attribute__((const))
float __ocml_len4_f32(float, float, float, float);
__device__
__attribute__((pure))
float __ocml_ncdf_f32(float);
__device__
__attribute__((pure))
float __ocml_ncdfinv_f32(float);
__device__
__attribute__((pure))
float __ocml_pow_f32(float, float);
__device__
__attribute__((pure))
float __ocml_pown_f32(float, int);
__device__
__attribute__((pure))
float __ocml_rcbrt_f32(float);
__device__
__attribute__((const))
float __ocml_remainder_f32(float, float);
__device__
float __ocml_remquo_f32(float, float, __attribute__((address_space(5))) int*);
__device__
__attribute__((const))
float __ocml_rhypot_f32(float, float);
__device__
__attribute__((const))
float __ocml_rint_f32(float);
__device__
__attribute__((const))
float __ocml_rlen3_f32(float, float, float);
__device__
__attribute__((const))
float __ocml_rlen4_f32(float, float, float, float);
__device__
__attribute__((const))
float __ocml_round_f32(float);
__device__
__attribute__((pure))
float __ocml_rsqrt_f32(float);
__device__
__attribute__((const))
float __ocml_scalb_f32(float, float);
__device__
__attribute__((const))
float __ocml_scalbn_f32(float, int);
__device__
__attribute__((const))
int __ocml_signbit_f32(float);
__device__
float __ocml_sincos_f32(float, __attribute__((address_space(5))) float*);
__device__
float __ocml_sincospi_f32(float, __attribute__((address_space(5))) float*);
__device__
float __ocml_sin_f32(float);
__device__
float __ocml_native_sin_f32(float);
__device__
__attribute__((pure))
float __ocml_sinh_f32(float);
__device__
float __ocml_sinpi_f32(float);
__device__
__attribute__((const))
float __ocml_sqrt_f32(float);
__device__
__attribute__((const))
float __ocml_native_sqrt_f32(float);
__device__
float __ocml_tan_f32(float);
__device__
__attribute__((pure))
float __ocml_tanh_f32(float);
__device__
float __ocml_tgamma_f32(float);
__device__
__attribute__((const))
float __ocml_trunc_f32(float);
__device__
float __ocml_y0_f32(float);
__device__
float __ocml_y1_f32(float);
// BEGIN INTRINSICS
__device__
__attribute__((const))
float __ocml_add_rte_f32(float, float);
__device__
__attribute__((const))
float __ocml_add_rtn_f32(float, float);
__device__
__attribute__((const))
float __ocml_add_rtp_f32(float, float);
__device__
__attribute__((const))
float __ocml_add_rtz_f32(float, float);
__device__
__attribute__((const))
float __ocml_sub_rte_f32(float, float);
__device__
__attribute__((const))
float __ocml_sub_rtn_f32(float, float);
__device__
__attribute__((const))
float __ocml_sub_rtp_f32(float, float);
__device__
__attribute__((const))
float __ocml_sub_rtz_f32(float, float);
__device__
__attribute__((const))
float __ocml_mul_rte_f32(float, float);
__device__
__attribute__((const))
float __ocml_mul_rtn_f32(float, float);
__device__
__attribute__((const))
float __ocml_mul_rtp_f32(float, float);
__device__
__attribute__((const))
float __ocml_mul_rtz_f32(float, float);
__device__
__attribute__((const))
float __ocml_div_rte_f32(float, float);
__device__
__attribute__((const))
float __ocml_div_rtn_f32(float, float);
__device__
__attribute__((const))
float __ocml_div_rtp_f32(float, float);
__device__
__attribute__((const))
float __ocml_div_rtz_f32(float, float);
__device__
__attribute__((const))
float __ocml_sqrt_rte_f32(float);
__device__
__attribute__((const))
float __ocml_sqrt_rtn_f32(float);
__device__
__attribute__((const))
float __ocml_sqrt_rtp_f32(float);
__device__
__attribute__((const))
float __ocml_sqrt_rtz_f32(float);
__device__
__attribute__((const))
float __ocml_fma_rte_f32(float, float, float);
__device__
__attribute__((const))
float __ocml_fma_rtn_f32(float, float, float);
__device__
__attribute__((const))
float __ocml_fma_rtp_f32(float, float, float);
__device__
__attribute__((const))
float __ocml_fma_rtz_f32(float, float, float);
__device__
__attribute__((const))
float __llvm_amdgcn_cos_f32(float) __asm("llvm.amdgcn.cos.f32");
__device__
__attribute__((const))
float __llvm_amdgcn_rcp_f32(float) __asm("llvm.amdgcn.rcp.f32");
__device__
__attribute__((const))
float __llvm_amdgcn_rsq_f32(float) __asm("llvm.amdgcn.rsq.f32");
__device__
__attribute__((const))
float __llvm_amdgcn_sin_f32(float) __asm("llvm.amdgcn.sin.f32");
// END INTRINSICS
// END FLOAT
// BEGIN DOUBLE
__device__
__attribute__((const))
double __ocml_acos_f64(double);
__device__
__attribute__((pure))
double __ocml_acosh_f64(double);
__device__
__attribute__((const))
double __ocml_asin_f64(double);
__device__
__attribute__((pure))
double __ocml_asinh_f64(double);
__device__
__attribute__((const))
double __ocml_atan2_f64(double, double);
__device__
__attribute__((const))
double __ocml_atan_f64(double);
__device__
__attribute__((pure))
double __ocml_atanh_f64(double);
__device__
__attribute__((pure))
double __ocml_cbrt_f64(double);
__device__
__attribute__((const))
double __ocml_ceil_f64(double);
__device__
__attribute__((const))
double __ocml_copysign_f64(double, double);
__device__
double __ocml_cos_f64(double);
__device__
__attribute__((pure))
double __ocml_cosh_f64(double);
__device__
double __ocml_cospi_f64(double);
__device__
double __ocml_i0_f64(double);
__device__
double __ocml_i1_f64(double);
__device__
__attribute__((pure))
double __ocml_erfc_f64(double);
__device__
__attribute__((pure))
double __ocml_erfcinv_f64(double);
__device__
__attribute__((pure))
double __ocml_erfcx_f64(double);
__device__
__attribute__((pure))
double __ocml_erf_f64(double);
__device__
__attribute__((pure))
double __ocml_erfinv_f64(double);
__device__
__attribute__((pure))
double __ocml_exp10_f64(double);
__device__
__attribute__((pure))
double __ocml_exp2_f64(double);
__device__
__attribute__((pure))
double __ocml_exp_f64(double);
__device__
__attribute__((pure))
double __ocml_expm1_f64(double);
__device__
__attribute__((const))
double __ocml_fabs_f64(double);
__device__
__attribute__((const))
double __ocml_fdim_f64(double, double);
__device__
__attribute__((const))
double __ocml_floor_f64(double);
__device__
__attribute__((const))
double __ocml_fma_f64(double, double, double);
__device__
__attribute__((const))
double __ocml_fmax_f64(double, double);
__device__
__attribute__((const))
double __ocml_fmin_f64(double, double);
__device__
__attribute__((const))
double __ocml_fmod_f64(double, double);
__device__
double __ocml_frexp_f64(double, __attribute__((address_space(5))) int*);
__device__
__attribute__((const))
double __ocml_hypot_f64(double, double);
__device__
__attribute__((const))
int __ocml_ilogb_f64(double);
__device__
__attribute__((const))
int __ocml_isfinite_f64(double);
__device__
__attribute__((const))
int __ocml_isinf_f64(double);
__device__
__attribute__((const))
int __ocml_isnan_f64(double);
__device__
double __ocml_j0_f64(double);
__device__
double __ocml_j1_f64(double);
__device__
__attribute__((const))
double __ocml_ldexp_f64(double, int);
__device__
double __ocml_lgamma_f64(double);
__device__
__attribute__((pure))
double __ocml_log10_f64(double);
__device__
__attribute__((pure))
double __ocml_log1p_f64(double);
__device__
__attribute__((pure))
double __ocml_log2_f64(double);
__device__
__attribute__((const))
double __ocml_logb_f64(double);
__device__
__attribute__((pure))
double __ocml_log_f64(double);
__device__
double __ocml_modf_f64(double, __attribute__((address_space(5))) double*);
__device__
__attribute__((const))
double __ocml_nearbyint_f64(double);
__device__
__attribute__((const))
double __ocml_nextafter_f64(double, double);
__device__
__attribute__((const))
double __ocml_len3_f64(double, double, double);
__device__
__attribute__((const))
double __ocml_len4_f64(double, double, double, double);
__device__
__attribute__((pure))
double __ocml_ncdf_f64(double);
__device__
__attribute__((pure))
double __ocml_ncdfinv_f64(double);
__device__
__attribute__((pure))
double __ocml_pow_f64(double, double);
__device__
__attribute__((pure))
double __ocml_pown_f64(double, int);
__device__
__attribute__((pure))
double __ocml_rcbrt_f64(double);
__device__
__attribute__((const))
double __ocml_remainder_f64(double, double);
__device__
double __ocml_remquo_f64(
double, double, __attribute__((address_space(5))) int*);
__device__
__attribute__((const))
double __ocml_rhypot_f64(double, double);
__device__
__attribute__((const))
double __ocml_rint_f64(double);
__device__
__attribute__((const))
double __ocml_rlen3_f64(double, double, double);
__device__
__attribute__((const))
double __ocml_rlen4_f64(double, double, double, double);
__device__
__attribute__((const))
double __ocml_round_f64(double);
__device__
__attribute__((pure))
double __ocml_rsqrt_f64(double);
__device__
__attribute__((const))
double __ocml_scalb_f64(double, double);
__device__
__attribute__((const))
double __ocml_scalbn_f64(double, int);
__device__
__attribute__((const))
int __ocml_signbit_f64(double);
__device__
double __ocml_sincos_f64(double, __attribute__((address_space(5))) double*);
__device__
double __ocml_sincospi_f64(double, __attribute__((address_space(5))) double*);
__device__
double __ocml_sin_f64(double);
__device__
__attribute__((pure))
double __ocml_sinh_f64(double);
__device__
double __ocml_sinpi_f64(double);
__device__
__attribute__((const))
double __ocml_sqrt_f64(double);
__device__
double __ocml_tan_f64(double);
__device__
__attribute__((pure))
double __ocml_tanh_f64(double);
__device__
double __ocml_tgamma_f64(double);
__device__
__attribute__((const))
double __ocml_trunc_f64(double);
__device__
double __ocml_y0_f64(double);
__device__
double __ocml_y1_f64(double);
// BEGIN INTRINSICS
__device__
__attribute__((const))
double __ocml_add_rte_f64(double, double);
__device__
__attribute__((const))
double __ocml_add_rtn_f64(double, double);
__device__
__attribute__((const))
double __ocml_add_rtp_f64(double, double);
__device__
__attribute__((const))
double __ocml_add_rtz_f64(double, double);
__device__
__attribute__((const))
double __ocml_sub_rte_f64(double, double);
__device__
__attribute__((const))
double __ocml_sub_rtn_f64(double, double);
__device__
__attribute__((const))
double __ocml_sub_rtp_f64(double, double);
__device__
__attribute__((const))
double __ocml_sub_rtz_f64(double, double);
__device__
__attribute__((const))
double __ocml_mul_rte_f64(double, double);
__device__
__attribute__((const))
double __ocml_mul_rtn_f64(double, double);
__device__
__attribute__((const))
double __ocml_mul_rtp_f64(double, double);
__device__
__attribute__((const))
double __ocml_mul_rtz_f64(double, double);
__device__
__attribute__((const))
double __ocml_div_rte_f64(double, double);
__device__
__attribute__((const))
double __ocml_div_rtn_f64(double, double);
__device__
__attribute__((const))
double __ocml_div_rtp_f64(double, double);
__device__
__attribute__((const))
double __ocml_div_rtz_f64(double, double);
__device__
__attribute__((const))
double __ocml_sqrt_rte_f64(double);
__device__
__attribute__((const))
double __ocml_sqrt_rtn_f64(double);
__device__
__attribute__((const))
double __ocml_sqrt_rtp_f64(double);
__device__
__attribute__((const))
double __ocml_sqrt_rtz_f64(double);
__device__
__attribute__((const))
double __ocml_fma_rte_f64(double, double, double);
__device__
__attribute__((const))
double __ocml_fma_rtn_f64(double, double, double);
__device__
__attribute__((const))
double __ocml_fma_rtp_f64(double, double, double);
__device__
__attribute__((const))
double __ocml_fma_rtz_f64(double, double, double);
__device__
__attribute__((const))
double __llvm_amdgcn_rcp_f64(double) __asm("llvm.amdgcn.rcp.f64");
__device__
__attribute__((const))
double __llvm_amdgcn_rsq_f64(double) __asm("llvm.amdgcn.rsq.f64");
// END INTRINSICS
// END DOUBLE
#endif // !__CLANG_HIP_RUNTIME_WRAPPER_INCLUDED__
#if defined(__cplusplus)
} // extern "C"
#endif
include/hip/amd_detail/ockl_image.h
-135
View File
@@ -1,135 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#include <hip/hip_vector_types.h>
extern "C" {
#define ADDRESS_SPACE_CONSTANT __attribute__((address_space(4)))
__device__ float4::Native_vec_ __ockl_image_load_1D(unsigned int ADDRESS_SPACE_CONSTANT*i, int c);
__device__ float4::Native_vec_ __ockl_image_load_1Db(unsigned int ADDRESS_SPACE_CONSTANT*i, int c);
__device__ float4::Native_vec_ __ockl_image_load_1Da(unsigned int ADDRESS_SPACE_CONSTANT*i, int2::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_load_2D(unsigned int ADDRESS_SPACE_CONSTANT*i, int2::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_load_2Da(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_load_3D(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_load_CM(unsigned int ADDRESS_SPACE_CONSTANT*i, int2::Native_vec_ c, int f);
__device__ float4::Native_vec_ __ockl_image_load_CMa(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, int f);
__device__ float4::Native_vec_ __ockl_image_load_lod_1D(unsigned int ADDRESS_SPACE_CONSTANT*i, int c, int l);
__device__ float4::Native_vec_ __ockl_image_load_lod_1Da(unsigned int ADDRESS_SPACE_CONSTANT*i, int2::Native_vec_ c, int l);
__device__ float4::Native_vec_ __ockl_image_load_lod_2D(unsigned int ADDRESS_SPACE_CONSTANT*i, int2::Native_vec_ c, int l);
__device__ float4::Native_vec_ __ockl_image_load_lod_2Da(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, int l);
__device__ float4::Native_vec_ __ockl_image_load_lod_3D(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, int l);
__device__ float4::Native_vec_ __ockl_image_load_lod_CM(unsigned int ADDRESS_SPACE_CONSTANT*i, int2::Native_vec_ c, int f, int l);
__device__ float4::Native_vec_ __ockl_image_load_lod_CMa(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, int f, int l);
__device__ void __ockl_image_store_1D(unsigned int ADDRESS_SPACE_CONSTANT*i, int c, float4::Native_vec_ p);
__device__ void __ockl_image_store_1Da(unsigned int ADDRESS_SPACE_CONSTANT*i, int2::Native_vec_ c, float4::Native_vec_ p);
__device__ void __ockl_image_store_2D(unsigned int ADDRESS_SPACE_CONSTANT*i, int2::Native_vec_ c, float4::Native_vec_ p);
__device__ void __ockl_image_store_2Da(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, float4::Native_vec_ p);
__device__ void __ockl_image_store_3D(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, float4::Native_vec_ p);
__device__ void __ockl_image_store_CM(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, float4::Native_vec_ p);
__device__ void __ockl_image_store_CMa(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, float4::Native_vec_ p);
__device__ void __ockl_image_store_lod_1D(unsigned int ADDRESS_SPACE_CONSTANT*i, int c, int l, float4::Native_vec_ p);
__device__ void __ockl_image_store_lod_1Da(unsigned int ADDRESS_SPACE_CONSTANT*i, int2::Native_vec_ c, int l, float4::Native_vec_ p);
__device__ void __ockl_image_store_lod_2D(unsigned int ADDRESS_SPACE_CONSTANT*i, int2::Native_vec_ c, int l, float4::Native_vec_ p);
__device__ void __ockl_image_store_lod_2Da(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, int l, float4::Native_vec_ p);
__device__ void __ockl_image_store_lod_3D(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, int l, float4::Native_vec_ p);
__device__ void __ockl_image_store_lod_CM(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, int l, float4::Native_vec_ p);
__device__ void __ockl_image_store_lod_CMa(unsigned int ADDRESS_SPACE_CONSTANT*i, int4::Native_vec_ c, int l, float4::Native_vec_ p);
__device__ float4::Native_vec_ __ockl_image_sample_1D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float c);
__device__ float4::Native_vec_ __ockl_image_sample_1Da(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float2::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_sample_2D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float2::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_sample_2Da(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float4::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_sample_3D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float4::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_sample_CM(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float4::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_sample_CMa(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float4::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_sample_grad_1D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float c, float dx, float dy);
__device__ float4::Native_vec_ __ockl_image_sample_grad_1Da(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float2::Native_vec_ c, float dx, float dy);
__device__ float4::Native_vec_ __ockl_image_sample_grad_2D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float2::Native_vec_ c, float2::Native_vec_ dx, float2::Native_vec_ dy);
__device__ float4::Native_vec_ __ockl_image_sample_grad_2Da(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float4::Native_vec_ c, float2::Native_vec_ dx, float2::Native_vec_ dy);
__device__ float4::Native_vec_ __ockl_image_sample_grad_3D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float4::Native_vec_ c, float4::Native_vec_ dx, float4::Native_vec_ dy);
__device__ float4::Native_vec_ __ockl_image_sample_lod_1D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float c, float l);
__device__ float4::Native_vec_ __ockl_image_sample_lod_1Da(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float2::Native_vec_ c, float l);
__device__ float4::Native_vec_ __ockl_image_sample_lod_2D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float2::Native_vec_ c, float l);
__device__ float4::Native_vec_ __ockl_image_sample_lod_2Da(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float4::Native_vec_ c, float l);
__device__ float4::Native_vec_ __ockl_image_sample_lod_3D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float4::Native_vec_ c, float l);
__device__ float4::Native_vec_ __ockl_image_sample_lod_CM(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float4::Native_vec_ c, float l);
__device__ float4::Native_vec_ __ockl_image_sample_lod_CMa(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float4::Native_vec_ c, float l);
__device__ float4::Native_vec_ __ockl_image_gather4r_2D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float2::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_gather4g_2D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float2::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_gather4b_2D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float2::Native_vec_ c);
__device__ float4::Native_vec_ __ockl_image_gather4a_2D(unsigned int ADDRESS_SPACE_CONSTANT*i, unsigned int ADDRESS_SPACE_CONSTANT*s, float2::Native_vec_ c);
};
include/hip/amd_detail/program_state.hpp
-107
View File
@@ -1,107 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#include <hsa/amd_hsa_kernel_code.h>
#include <hsa/hsa.h>
#include <hsa/hsa_ext_amd.h>
#include <hsa/hsa_ven_amd_loader.h>
#include <cstddef>
#include <cstdint>
#include <cstdlib>
#include <hip/hip_common.h>
struct ihipModuleSymbol_t;
using hipFunction_t = ihipModuleSymbol_t*;
namespace hip_impl {
// This section contains internal APIs that
// needs to be exported
#ifdef __GNUC__
#pragma GCC visibility push (default)
#endif
struct kernarg_impl;
class kernarg {
public:
kernarg();
kernarg(kernarg&&);
~kernarg();
std::uint8_t* data();
std::size_t size();
void reserve(std::size_t);
void resize(std::size_t);
private:
kernarg_impl* impl;
};
class kernargs_size_align;
class program_state_impl;
class program_state {
public:
program_state();
~program_state();
program_state(const program_state&) = delete;
hipFunction_t kernel_descriptor(std::uintptr_t,
hsa_agent_t);
kernargs_size_align get_kernargs_size_align(std::uintptr_t);
hsa_executable_t load_executable(const char*, const size_t,
hsa_executable_t,
hsa_agent_t);
hsa_executable_t load_executable_no_copy(const char*, const size_t,
hsa_executable_t,
hsa_agent_t);
void* global_addr_by_name(const char* name);
private:
friend class agent_globals_impl;
program_state_impl* impl;
};
class kernargs_size_align {
public:
std::size_t size(std::size_t n) const;
std::size_t alignment(std::size_t n) const;
const void* getHandle() const {return handle;};
private:
const void* handle;
friend kernargs_size_align program_state::get_kernargs_size_align(std::uintptr_t);
};
#ifdef __GNUC__
#pragma GCC visibility pop
#endif
inline
__attribute__((visibility("hidden")))
program_state& get_program_state() {
static program_state ps;
return ps;
}
} // Namespace hip_impl.
include/hip/amd_detail/surface_functions.h
-59
View File
@@ -1,59 +0,0 @@
/*
Copyright (c) 2018 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_SURFACE_FUNCTIONS_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_SURFACE_FUNCTIONS_H
#include <hip/amd_detail/hip_surface_types.h>
#define __SURFACE_FUNCTIONS_DECL__ static inline __device__
template <class T>
__SURFACE_FUNCTIONS_DECL__ void surf2Dread(T* data, hipSurfaceObject_t surfObj, int x, int y,
int boundaryMode = hipBoundaryModeZero) {
hipArray* arrayPtr = (hipArray*)surfObj;
size_t width = arrayPtr->width;
size_t height = arrayPtr->height;
int32_t xOffset = x / sizeof(T);
T* dataPtr = (T*)arrayPtr->data;
if ((xOffset > width) || (xOffset < 0) || (y > height) || (y < 0)) {
if (boundaryMode == hipBoundaryModeZero) {
*data = 0;
}
} else {
*data = *(dataPtr + y * width + xOffset);
}
}
template <class T>
__SURFACE_FUNCTIONS_DECL__ void surf2Dwrite(T data, hipSurfaceObject_t surfObj, int x, int y,
int boundaryMode = hipBoundaryModeZero) {
hipArray* arrayPtr = (hipArray*)surfObj;
size_t width = arrayPtr->width;
size_t height = arrayPtr->height;
int32_t xOffset = x / sizeof(T);
T* dataPtr = (T*)arrayPtr->data;
if (!((xOffset > width) || (xOffset < 0) || (y > height) || (y < 0))) {
*(dataPtr + y * width + xOffset) = data;
}
}
#endif
include/hip/amd_detail/texture_fetch_functions.h
-388
View File
@@ -1,388 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#if defined(__cplusplus)
#include <hip/hip_vector_types.h>
#include <hip/hip_texture_types.h>
#include <hip/amd_detail/ockl_image.h>
#if !defined(__HIPCC_RTC__)
#include <type_traits>
#endif // !defined(__HIPCC_RTC__)
#define TEXTURE_PARAMETERS_INIT \
unsigned int ADDRESS_SPACE_CONSTANT* i = (unsigned int ADDRESS_SPACE_CONSTANT*)t.textureObject; \
unsigned int ADDRESS_SPACE_CONSTANT* s = i + HIP_SAMPLER_OBJECT_OFFSET_DWORD;
template<typename T>
struct __hip_is_tex_channel_type
{
static constexpr bool value =
std::is_same<T, char>::value ||
std::is_same<T, unsigned char>::value ||
std::is_same<T, short>::value ||
std::is_same<T, unsigned short>::value ||
std::is_same<T, int>::value ||
std::is_same<T, unsigned int>::value ||
std::is_same<T, float>::value;
};
template<
typename T,
unsigned int rank>
struct __hip_is_tex_channel_type<HIP_vector_type<T, rank>>
{
static constexpr bool value =
__hip_is_tex_channel_type<T>::value &&
((rank == 1) ||
(rank == 2) ||
(rank == 4));
};
template<typename T>
struct __hip_is_tex_normalized_channel_type
{
static constexpr bool value =
std::is_same<T, char>::value ||
std::is_same<T, unsigned char>::value ||
std::is_same<T, short>::value ||
std::is_same<T, unsigned short>::value;
};
template<
typename T,
unsigned int rank>
struct __hip_is_tex_normalized_channel_type<HIP_vector_type<T, rank>>
{
static constexpr bool value =
__hip_is_tex_normalized_channel_type<T>::value &&
((rank == 1) ||
(rank == 2) ||
(rank == 4));
};
template <
typename T,
hipTextureReadMode readMode,
typename Enable = void>
struct __hip_tex_ret
{
static_assert(std::is_same<Enable, void>::value, "Invalid channel type!");
};
template <
typename T,
hipTextureReadMode readMode>
using __hip_tex_ret_t = typename __hip_tex_ret<T, readMode, bool>::type;
template <typename T>
struct __hip_tex_ret<
T,
hipReadModeElementType,
typename std::enable_if<__hip_is_tex_channel_type<T>::value, bool>::type>
{
using type = T;
};
template<
typename T,
unsigned int rank>
struct __hip_tex_ret<
HIP_vector_type<T, rank>,
hipReadModeElementType,
typename std::enable_if<__hip_is_tex_channel_type<HIP_vector_type<T, rank>>::value, bool>::type>
{
using type = HIP_vector_type<__hip_tex_ret_t<T, hipReadModeElementType>, rank>;
};
template<typename T>
struct __hip_tex_ret<
T,
hipReadModeNormalizedFloat,
typename std::enable_if<__hip_is_tex_normalized_channel_type<T>::value, bool>::type>
{
using type = float;
};
template<
typename T,
unsigned int rank>
struct __hip_tex_ret<
HIP_vector_type<T, rank>,
hipReadModeNormalizedFloat,
typename std::enable_if<__hip_is_tex_normalized_channel_type<HIP_vector_type<T, rank>>::value, bool>::type>
{
using type = HIP_vector_type<__hip_tex_ret_t<T, hipReadModeNormalizedFloat>, rank>;
};
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex1Dfetch(texture<T, hipTextureType1D, readMode> t, int x)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_load_1Db(i, x);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex1D(texture<T, hipTextureType1D, readMode> t, float x)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_1D(i, s, x);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex2D(texture<T, hipTextureType2D, readMode> t, float x, float y)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_2D(i, s, float2(x, y).data);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex1DLayered(texture<T, hipTextureType1DLayered, readMode> t, float x, int layer)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_1Da(i, s, float2(x, layer).data);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex2DLayered(texture<T, hipTextureType2DLayered, readMode> t, float x, float y, int layer)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_2Da(i, s, float4(x, y, layer, 0.0f).data);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex3D(texture<T, hipTextureType3D, readMode> t, float x, float y, float z)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_3D(i, s, float4(x, y, z, 0.0f).data);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> texCubemap(texture<T, hipTextureTypeCubemap, readMode> t, float x, float y, float z)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_CM(i, s, float4(x, y, z, 0.0f).data);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex1DLod(texture<T, hipTextureType1D, readMode> t, float x, float level)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_lod_1D(i, s, x, level);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex2DLod(texture<T, hipTextureType2D, readMode> t, float x, float y, float level)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_lod_2D(i, s, float2(x, y).data, level);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex1DLayeredLod(texture<T, hipTextureType1DLayered, readMode> t, float x, int layer, float level)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_lod_1Da(i, s, float2(x, layer).data, level);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex2DLayeredLod(texture<T, hipTextureType2DLayered, readMode> t, float x, float y, int layer, float level)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_lod_2Da(i, s, float4(x, y, layer, 0.0f).data, level);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex3DLod(texture<T, hipTextureType3D, readMode> t, float x, float y, float z, float level)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_lod_3D(i, s, float4(x, y, z, 0.0f).data, level);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> texCubemapLod(texture<T, hipTextureTypeCubemap, readMode> t, float x, float y, float z, float level)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_lod_CM(i, s, float4(x, y, z, 0.0f).data, level);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> texCubemapLayered(texture<T, hipTextureTypeCubemapLayered, readMode> t, float x, float y, float z, int layer)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_CMa(i, s, float4(x, y, z, layer).data);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> texCubemapLayeredLod(texture<T, hipTextureTypeCubemapLayered, readMode> t, float x, float y, float z, int layer, float level)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_lod_CMa(i, s, float4(x, y, z, layer).data, level);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> texCubemapGrad(texture<T, hipTextureTypeCubemap, readMode> t, float x, float y, float z, float4 dPdx, float4 dPdy)
{
TEXTURE_PARAMETERS_INIT;
// TODO missing in device libs.
// auto tmp = __ockl_image_sample_grad_CM(i, s, float4(x, y, z, 0.0f).data, float4(dPdx.x, dPdx.y, dPdx.z, 0.0f).data, float4(dPdy.x, dPdy.y, dPdy.z, 0.0f).data);
// return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
return {};
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> texCubemapLayeredGrad(texture<T, hipTextureTypeCubemapLayered, readMode> t, float x, float y, float z, int layer, float4 dPdx, float4 dPdy)
{
TEXTURE_PARAMETERS_INIT;
// TODO missing in device libs.
// auto tmp = __ockl_image_sample_grad_CMa(i, s, float4(x, y, z, layer).data, float4(dPdx.x, dPdx.y, dPdx.z, 0.0f).data, float4(dPdy.x, dPdy.y, dPdy.z, 0.0f).data);
// return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
return {};
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex1DGrad(texture<T, hipTextureType1D, readMode> t, float x, float dPdx, float dPdy)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_grad_1D(i, s, x, dPdx, dPdy);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex2DGrad(texture<T, hipTextureType2D, readMode> t, float x, float y, float2 dPdx, float2 dPdy)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_grad_2D(i, s, float2(x, y).data, float2(dPdx.x, dPdx.y).data, float2(dPdy.x, dPdy.y).data);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex1DLayeredGrad(texture<T, hipTextureType1DLayered, readMode> t, float x, int layer, float dPdx, float dPdy)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_grad_1Da(i, s, float2(x, layer).data, dPdx, dPdy);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex2DLayeredGrad(texture<T, hipTextureType2DLayered, readMode> t, float x, float y, int layer, float2 dPdx, float2 dPdy)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_grad_2Da(i, s, float4(x, y, layer, 0.0f).data, float2(dPdx.x, dPdx.y).data, float2(dPdy.x, dPdy.y).data);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex_ret_t<T, readMode> tex3DGrad(texture<T, hipTextureType3D, readMode> t, float x, float y, float z, float4 dPdx, float4 dPdy)
{
TEXTURE_PARAMETERS_INIT;
auto tmp = __ockl_image_sample_grad_3D(i, s, float4(x, y, z, 0.0f).data, float4(dPdx.x, dPdx.y, dPdx.z, 0.0f).data, float4(dPdy.x, dPdy.y, dPdy.z, 0.0f).data);
return *reinterpret_cast<__hip_tex_ret_t<T, readMode>*>(&tmp);
}
template <
typename T,
hipTextureReadMode readMode,
typename Enable = void>
struct __hip_tex2dgather_ret
{
static_assert(std::is_same<Enable, void>::value, "Invalid channel type!");
};
template <
typename T,
hipTextureReadMode readMode>
using __hip_tex2dgather_ret_t = typename __hip_tex2dgather_ret<T, readMode, bool>::type;
template <typename T>
struct __hip_tex2dgather_ret<
T,
hipReadModeElementType,
typename std::enable_if<__hip_is_tex_channel_type<T>::value, bool>::type>
{
using type = HIP_vector_type<T, 4>;
};
template<
typename T,
unsigned int rank>
struct __hip_tex2dgather_ret<
HIP_vector_type<T, rank>,
hipReadModeElementType,
typename std::enable_if<__hip_is_tex_channel_type<HIP_vector_type<T, rank>>::value, bool>::type>
{
using type = HIP_vector_type<T, 4>;
};
template <typename T>
struct __hip_tex2dgather_ret<
T,
hipReadModeNormalizedFloat,
typename std::enable_if<__hip_is_tex_normalized_channel_type<T>::value, bool>::type>
{
using type = float4;
};
template <typename T, hipTextureReadMode readMode>
static __forceinline__ __device__ __hip_tex2dgather_ret_t<T, readMode> tex2Dgather(texture<T, hipTextureType2D, readMode> t, float x, float y, int comp=0)
{
TEXTURE_PARAMETERS_INIT;
switch (comp) {
case 1: {
auto tmp = __ockl_image_gather4g_2D(i, s, float2(x, y).data);
return *reinterpret_cast<__hip_tex2dgather_ret_t<T, readMode>*>(&tmp);
}
case 2: {
auto tmp = __ockl_image_gather4b_2D(i, s, float2(x, y).data);
return *reinterpret_cast<__hip_tex2dgather_ret_t<T, readMode>*>(&tmp);
}
case 3: {
auto tmp = __ockl_image_gather4a_2D(i, s, float2(x, y).data);
return *reinterpret_cast<__hip_tex2dgather_ret_t<T, readMode>*>(&tmp);
}
default: {
auto tmp = __ockl_image_gather4r_2D(i, s, float2(x, y).data);
return *reinterpret_cast<__hip_tex2dgather_ret_t<T, readMode>*>(&tmp);
}
}
return {};
}
#endif
include/hip/amd_detail/texture_functions.h
-11102
View File
File diff suppressed because it is too large Load Diff
include/hip/amd_detail/texture_indirect_functions.h
-503
View File
@@ -1,503 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#if defined(__cplusplus)
#include <hip/hip_vector_types.h>
#include <hip/hip_texture_types.h>
#include <hip/amd_detail/ockl_image.h>
#if !defined(__HIPCC_RTC__)
#include <type_traits>
#endif // !defined(__HIPCC_RTC__)
#define TEXTURE_OBJECT_PARAMETERS_INIT \
unsigned int ADDRESS_SPACE_CONSTANT* i = (unsigned int ADDRESS_SPACE_CONSTANT*)textureObject; \
unsigned int ADDRESS_SPACE_CONSTANT* s = i + HIP_SAMPLER_OBJECT_OFFSET_DWORD;
template<typename T>
struct __hip_is_itex_channel_type
{
static constexpr bool value =
std::is_same<T, char>::value ||
std::is_same<T, unsigned char>::value ||
std::is_same<T, short>::value ||
std::is_same<T, unsigned short>::value ||
std::is_same<T, int>::value ||
std::is_same<T, unsigned int>::value ||
std::is_same<T, float>::value;
};
template<
typename T,
unsigned int rank>
struct __hip_is_itex_channel_type<HIP_vector_type<T, rank>>
{
static constexpr bool value =
__hip_is_itex_channel_type<T>::value &&
((rank == 1) ||
(rank == 2) ||
(rank == 4));
};
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex1Dfetch(hipTextureObject_t textureObject, int x)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_load_1Db(i, x);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex1Dfetch(T *ptr, hipTextureObject_t textureObject, int x)
{
*ptr = tex1Dfetch<T>(textureObject, x);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex1D(hipTextureObject_t textureObject, float x)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_1D(i, s, x);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex1D(T *ptr, hipTextureObject_t textureObject, float x)
{
*ptr = tex1D<T>(textureObject, x);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex2D(hipTextureObject_t textureObject, float x, float y)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_2D(i, s, float2(x, y).data);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex2D(T *ptr, hipTextureObject_t textureObject, float x, float y)
{
*ptr = tex2D<T>(textureObject, x, y);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex3D(hipTextureObject_t textureObject, float x, float y, float z)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_3D(i, s, float4(x, y, z, 0.0f).data);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex3D(T *ptr, hipTextureObject_t textureObject, float x, float y, float z)
{
*ptr = tex3D<T>(textureObject, x, y, z);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex1DLayered(hipTextureObject_t textureObject, float x, int layer)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_1Da(i, s, float2(x, layer).data);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex1DLayered(T *ptr, hipTextureObject_t textureObject, float x, int layer)
{
*ptr = tex1DLayered<T>(textureObject, x, layer);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex2DLayered(hipTextureObject_t textureObject, float x, float y, int layer)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_2Da(i, s, float4(x, y, layer, 0.0f).data);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex2DLayered(T *ptr, hipTextureObject_t textureObject, float x, float y, int layer)
{
*ptr = tex1DLayered<T>(textureObject, x, y, layer);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T texCubemap(hipTextureObject_t textureObject, float x, float y, float z)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_CM(i, s, float4(x, y, z, 0.0f).data);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void texCubemap(T *ptr, hipTextureObject_t textureObject, float x, float y, float z)
{
*ptr = texCubemap<T>(textureObject, x, y, z);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T texCubemapLayered(hipTextureObject_t textureObject, float x, float y, float z, int layer)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_CMa(i, s, float4(x, y, z, layer).data);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void texCubemapLayered(T *ptr, hipTextureObject_t textureObject, float x, float y, float z, int layer)
{
*ptr = texCubemapLayered<T>(textureObject, x, y, z, layer);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex2Dgather(hipTextureObject_t textureObject, float x, float y, int comp = 0)
{
TEXTURE_OBJECT_PARAMETERS_INIT
switch (comp) {
case 1: {
auto tmp = __ockl_image_gather4r_2D(i, s, float2(x, y).data);
return *reinterpret_cast<T*>(&tmp);
break;
}
case 2: {
auto tmp = __ockl_image_gather4g_2D(i, s, float2(x, y).data);
return *reinterpret_cast<T*>(&tmp);
break;
}
case 3: {
auto tmp = __ockl_image_gather4b_2D(i, s, float2(x, y).data);
return *reinterpret_cast<T*>(&tmp);
break;
}
default: {
auto tmp = __ockl_image_gather4a_2D(i, s, float2(x, y).data);
return *reinterpret_cast<T*>(&tmp);
break;
}
};
return {};
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex2Dgather(T *ptr, hipTextureObject_t textureObject, float x, float y, int comp = 0)
{
*ptr = texCubemapLayered<T>(textureObject, x, y, comp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex1DLod(hipTextureObject_t textureObject, float x, float level)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_lod_1D(i, s, x, level);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex1DLod(T *ptr, hipTextureObject_t textureObject, float x, float level)
{
*ptr = tex1DLod<T>(textureObject, x, level);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex2DLod(hipTextureObject_t textureObject, float x, float y, float level)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_lod_2D(i, s, float2(x, y).data, level);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex2DLod(T *ptr, hipTextureObject_t textureObject, float x, float y, float level)
{
*ptr = tex2DLod<T>(textureObject, x, y, level);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex3DLod(hipTextureObject_t textureObject, float x, float y, float z, float level)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_lod_3D(i, s, float4(x, y, z, 0.0f).data, level);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex3DLod(T *ptr, hipTextureObject_t textureObject, float x, float y, float z, float level)
{
*ptr = tex3DLod<T>(textureObject, x, y, z, level);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex1DLayeredLod(hipTextureObject_t textureObject, float x, int layer, float level)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_1Da(i, s, float2(x, layer).data);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex1DLayeredLod(T *ptr, hipTextureObject_t textureObject, float x, int layer, float level)
{
*ptr = tex1DLayeredLod<T>(textureObject, x, layer, level);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex2DLayeredLod(hipTextureObject_t textureObject, float x, float y, int layer, float level)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_2Da(i, s, float4(x, y, layer, 0.0f).data);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex2DLayeredLod(T *ptr, hipTextureObject_t textureObject, float x, float y, int layer, float level)
{
*ptr = tex2DLayeredLod<T>(textureObject, x, y, layer, level);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T texCubemapLod(hipTextureObject_t textureObject, float x, float y, float z, float level)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_lod_CM(i, s, float4(x, y, z, 0.0f).data, level);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void texCubemapLod(T *ptr, hipTextureObject_t textureObject, float x, float y, float z, float level)
{
*ptr = texCubemapLod<T>(textureObject, x, y, z, level);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T texCubemapGrad(hipTextureObject_t textureObject, float x, float y, float z, float4 dPdx, float4 dPdy)
{
TEXTURE_OBJECT_PARAMETERS_INIT
// TODO missing in device libs.
// auto tmp = __ockl_image_sample_grad_CM(i, s, float4(x, y, z, 0.0f).data, float4(dPdx.x, dPdx.y, dPdx.z, 0.0f).data, float4(dPdy.x, dPdy.y, dPdy.z, 0.0f).data);
// return *reinterpret_cast<T*>(&tmp);
return {};
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void texCubemapGrad(T *ptr, hipTextureObject_t textureObject, float x, float y, float z, float4 dPdx, float4 dPdy)
{
*ptr = texCubemapGrad<T>(textureObject, x, y, z, dPdx, dPdy);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T texCubemapLayeredLod(hipTextureObject_t textureObject, float x, float y, float z, int layer, float level)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_lod_CMa(i, s, float4(x, y, z, layer).data, level);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void texCubemapLayeredLod(T *ptr, hipTextureObject_t textureObject, float x, float y, float z, int layer, float level)
{
*ptr = texCubemapLayeredLod<T>(textureObject, x, y, z, layer, level);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex1DGrad(hipTextureObject_t textureObject, float x, float dPdx, float dPdy)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_grad_1D(i, s, x, dPdx, dPdy);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex1DGrad(T *ptr, hipTextureObject_t textureObject, float x, float dPdx, float dPdy)
{
*ptr = tex1DGrad<T>(textureObject, x, dPdx, dPdy);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex2DGrad(hipTextureObject_t textureObject, float x, float y, float2 dPdx, float2 dPdy)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_grad_2D(i, s, float2(x, y).data, float2(dPdx.x, dPdx.y).data, float2(dPdy.x, dPdy.y).data);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex2DGrad(T *ptr, hipTextureObject_t textureObject, float x, float y, float2 dPdx, float2 dPdy)
{
*ptr = tex2DGrad<T>(textureObject, x, y, dPdx, dPdy);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex3DGrad(hipTextureObject_t textureObject, float x, float y, float z, float4 dPdx, float4 dPdy)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_grad_3D(i, s, float4(x, y, z, 0.0f).data, float4(dPdx.x, dPdx.y, dPdx.z, 0.0f).data, float4(dPdy.x, dPdy.y, dPdy.z, 0.0f).data);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex3DGrad(T *ptr, hipTextureObject_t textureObject, float x, float y, float z, float4 dPdx, float4 dPdy)
{
*ptr = tex3DGrad<T>(textureObject, x, y, z, dPdx, dPdy);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex1DLayeredGrad(hipTextureObject_t textureObject, float x, int layer, float dPdx, float dPdy)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_grad_1Da(i, s, float2(x, layer).data, dPdx, dPdy);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex1DLayeredGrad(T *ptr, hipTextureObject_t textureObject, float x, int layer, float dPdx, float dPdy)
{
*ptr = tex1DLayeredGrad<T>(textureObject, x, layer, dPdx, dPdy);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T tex2DLayeredGrad(hipTextureObject_t textureObject, float x, float y, int layer, float2 dPdx, float2 dPdy)
{
TEXTURE_OBJECT_PARAMETERS_INIT
auto tmp = __ockl_image_sample_grad_2Da(i, s, float4(x, y, layer, 0.0f).data, float2(dPdx.x, dPdx.y).data, float2(dPdy.x, dPdy.y).data);
return *reinterpret_cast<T*>(&tmp);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void tex2DLayeredGrad(T *ptr, hipTextureObject_t textureObject, float x, float y, int layer, float2 dPdx, float2 dPdy)
{
*ptr = tex2DLayeredGrad<T>(textureObject, x, y, layer, dPdx, dPdy);
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ T texCubemapLayeredGrad(hipTextureObject_t textureObject, float x, float y, float z, int layer, float4 dPdx, float4 dPdy)
{
TEXTURE_OBJECT_PARAMETERS_INIT
// TODO missing in device libs.
// auto tmp = __ockl_image_sample_grad_CMa(i, s, float4(x, y, z, layer).data, float4(dPdx.x, dPdx.y, dPdx.z, 0.0f).data, float4(dPdy.x, dPdy.y, dPdy.z, 0.0f).data);
// return *reinterpret_cast<T*>(&tmp);
return {};
}
template <
typename T,
typename std::enable_if<__hip_is_itex_channel_type<T>::value>::type* = nullptr>
static __device__ void texCubemapLayeredGrad(T *ptr, hipTextureObject_t textureObject, float x, float y, float z, int layer, float4 dPdx, float4 dPdy)
{
*ptr = texCubemapLayeredGrad<T>(textureObject, x, y, z, layer, dPdx, dPdy);
}
#endif
include/hip/amd_detail/texture_types.h
-109
View File
@@ -1,109 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_TEXTURE_TYPES_H
#define HIP_INCLUDE_HIP_AMD_DETAIL_TEXTURE_TYPES_H
#include <hip/amd_detail/driver_types.h>
#define hipTextureType1D 0x01
#define hipTextureType2D 0x02
#define hipTextureType3D 0x03
#define hipTextureTypeCubemap 0x0C
#define hipTextureType1DLayered 0xF1
#define hipTextureType2DLayered 0xF2
#define hipTextureTypeCubemapLayered 0xFC
/**
* Should be same as HSA_IMAGE_OBJECT_SIZE_DWORD/HSA_SAMPLER_OBJECT_SIZE_DWORD
*/
#define HIP_IMAGE_OBJECT_SIZE_DWORD 12
#define HIP_SAMPLER_OBJECT_SIZE_DWORD 8
#define HIP_SAMPLER_OBJECT_OFFSET_DWORD HIP_IMAGE_OBJECT_SIZE_DWORD
#define HIP_TEXTURE_OBJECT_SIZE_DWORD (HIP_IMAGE_OBJECT_SIZE_DWORD + HIP_SAMPLER_OBJECT_SIZE_DWORD)
/**
* An opaque value that represents a hip texture object
*/
struct __hip_texture;
typedef struct __hip_texture* hipTextureObject_t;
/**
* hip texture address modes
*/
enum hipTextureAddressMode {
hipAddressModeWrap = 0,
hipAddressModeClamp = 1,
hipAddressModeMirror = 2,
hipAddressModeBorder = 3
};
/**
* hip texture filter modes
*/
enum hipTextureFilterMode { hipFilterModePoint = 0, hipFilterModeLinear = 1 };
/**
* hip texture read modes
*/
enum hipTextureReadMode { hipReadModeElementType = 0, hipReadModeNormalizedFloat = 1 };
/**
* hip texture reference
*/
typedef struct textureReference {
int normalized;
enum hipTextureReadMode readMode;// used only for driver API's
enum hipTextureFilterMode filterMode;
enum hipTextureAddressMode addressMode[3]; // Texture address mode for up to 3 dimensions
struct hipChannelFormatDesc channelDesc;
int sRGB; // Perform sRGB->linear conversion during texture read
unsigned int maxAnisotropy; // Limit to the anisotropy ratio
enum hipTextureFilterMode mipmapFilterMode;
float mipmapLevelBias;
float minMipmapLevelClamp;
float maxMipmapLevelClamp;
hipTextureObject_t textureObject;
int numChannels;
enum hipArray_Format format;
}textureReference;
/**
* hip texture descriptor
*/
typedef struct hipTextureDesc {
enum hipTextureAddressMode addressMode[3]; // Texture address mode for up to 3 dimensions
enum hipTextureFilterMode filterMode;
enum hipTextureReadMode readMode;
int sRGB; // Perform sRGB->linear conversion during texture read
float borderColor[4];
int normalizedCoords;
unsigned int maxAnisotropy;
enum hipTextureFilterMode mipmapFilterMode;
float mipmapLevelBias;
float minMipmapLevelClamp;
float maxMipmapLevelClamp;
}hipTextureDesc;
#endif
include/hip/channel_descriptor.h
+2 -2
View File
@@ -29,9 +29,9 @@ THE SOFTWARE.
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/amd_detail/channel_descriptor.h>
#include <hip/amd_detail/amd_channel_descriptor.h>
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/nvidia_detail/channel_descriptor.h>
#include <hip/nvidia_detail/nvidia_channel_descriptor.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
include/hip/device_functions.h
+1 -1
View File
@@ -26,7 +26,7 @@ THE SOFTWARE.
#include <hip/hip_common.h>
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/amd_detail/device_functions.h>
#include <hip/amd_detail/amd_device_functions.h>
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <device_functions.h>
#else
include/hip/hcc_detail
-1
View File
@@ -1 +0,0 @@
amd_detail
include/hip/hip_complex.h
+2 -2
View File
@@ -26,9 +26,9 @@ THE SOFTWARE.
#include <hip/hip_common.h>
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/amd_detail/hip_complex.h>
#include <hip/amd_detail/amd_hip_complex.h>
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/nvidia_detail/hip_complex.h>
#include <hip/nvidia_detail/nvidia_hip_complex.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
include/hip/hip_cooperative_groups.h
+2 -2
View File
@@ -35,10 +35,10 @@ THE SOFTWARE.
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#if __cplusplus && defined(__clang__) && defined(__HIP__)
#include <hip/amd_detail/hip_cooperative_groups.h>
#include <hip/amd_detail/amd_hip_cooperative_groups.h>
#endif
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/nvidia_detail/hip_cooperative_groups.h>
#include <hip/nvidia_detail/nvidia_hip_cooperative_groups.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
include/hip/hip_fp16.h
+1 -1
View File
@@ -26,7 +26,7 @@ THE SOFTWARE.
#include <hip/hip_common.h>
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/amd_detail/hip_fp16.h>
#include <hip/amd_detail/amd_hip_fp16.h>
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include "cuda_fp16.h"
#else
include/hip/hip_runtime.h
+2 -2
View File
@@ -59,9 +59,9 @@ THE SOFTWARE.
#include <hip/hip_common.h>
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/amd_detail/hip_runtime.h>
#include <hip/amd_detail/amd_hip_runtime.h>
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/nvidia_detail/hip_runtime.h>
#include <hip/nvidia_detail/nvidia_hip_runtime.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_AMD__ or __HIP_PLATFORM_NVIDIA__");
#endif
include/hip/hip_runtime_api.h
+3845 -3
View File
File diff suppressed because it is too large Load Diff
include/hip/hip_texture_types.h
+1 -1
View File
@@ -25,7 +25,7 @@ THE SOFTWARE.
#define HIP_INCLUDE_HIP_HIP_TEXTURE_TYPES_H
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/amd_detail/hip_texture_types.h>
#include <hip/amd_detail/amd_hip_texture_types.h>
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/nvidia_detail/hip_texture_types.h>
#else
include/hip/hip_vector_types.h
+1 -1
View File
@@ -30,7 +30,7 @@ THE SOFTWARE.
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#if __cplusplus
#include <hip/amd_detail/hip_vector_types.h>
#include <hip/amd_detail/amd_hip_vector_types.h>
#endif
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <vector_types.h>
include/hip/hiprtc.h
+1 -1
View File
@@ -24,7 +24,7 @@ THE SOFTWARE.
#include <hip/hip_common.h>
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/amd_detail/hiprtc.h>
#include <hip/amd_detail/amd_hiprtc.h>
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/nvidia_detail/hiprtc.h>
#else
include/hip/library_types.h
+1 -1
View File
@@ -26,7 +26,7 @@ THE SOFTWARE.
#include <hip/hip_common.h>
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/amd_detail/library_types.h>
#include <hip/amd_detail/amd_library_types.h>
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include "library_types.h"
#else
include/hip/math_functions.h
+1 -1
View File
@@ -30,7 +30,7 @@ THE SOFTWARE.
#include <hip/hip_common.h>
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/amd_detail/math_functions.h>
#include <hip/amd_detail/amd_math_functions.h>
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
//#include <hip/nvidia_detail/math_functions.h>
#else
include/hip/nvidia_detail/channel_descriptor.h
-28
View File
@@ -1,28 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_NVIDIA_DETAIL_CHANNEL_DESCRIPTOR_H
#define HIP_INCLUDE_HIP_NVIDIA_DETAIL_CHANNEL_DESCRIPTOR_H
#include "channel_descriptor.h"
#endif
include/hip/nvidia_detail/hip_complex.h
-119
View File
@@ -1,119 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_NVIDIA_DETAIL_HIP_COMPLEX_H
#define HIP_INCLUDE_HIP_NVIDIA_DETAIL_HIP_COMPLEX_H
#include "cuComplex.h"
typedef cuFloatComplex hipFloatComplex;
__device__ __host__ static inline float hipCrealf(hipFloatComplex z) { return cuCrealf(z); }
__device__ __host__ static inline float hipCimagf(hipFloatComplex z) { return cuCimagf(z); }
__device__ __host__ static inline hipFloatComplex make_hipFloatComplex(float a, float b) {
return make_cuFloatComplex(a, b);
}
__device__ __host__ static inline hipFloatComplex hipConjf(hipFloatComplex z) { return cuConjf(z); }
__device__ __host__ static inline float hipCsqabsf(hipFloatComplex z) {
return cuCabsf(z) * cuCabsf(z);
}
__device__ __host__ static inline hipFloatComplex hipCaddf(hipFloatComplex p, hipFloatComplex q) {
return cuCaddf(p, q);
}
__device__ __host__ static inline hipFloatComplex hipCsubf(hipFloatComplex p, hipFloatComplex q) {
return cuCsubf(p, q);
}
__device__ __host__ static inline hipFloatComplex hipCmulf(hipFloatComplex p, hipFloatComplex q) {
return cuCmulf(p, q);
}
__device__ __host__ static inline hipFloatComplex hipCdivf(hipFloatComplex p, hipFloatComplex q) {
return cuCdivf(p, q);
}
__device__ __host__ static inline float hipCabsf(hipFloatComplex z) { return cuCabsf(z); }
typedef cuDoubleComplex hipDoubleComplex;
__device__ __host__ static inline double hipCreal(hipDoubleComplex z) { return cuCreal(z); }
__device__ __host__ static inline double hipCimag(hipDoubleComplex z) { return cuCimag(z); }
__device__ __host__ static inline hipDoubleComplex make_hipDoubleComplex(double a, double b) {
return make_cuDoubleComplex(a, b);
}
__device__ __host__ static inline hipDoubleComplex hipConj(hipDoubleComplex z) { return cuConj(z); }
__device__ __host__ static inline double hipCsqabs(hipDoubleComplex z) {
return cuCabs(z) * cuCabs(z);
}
__device__ __host__ static inline hipDoubleComplex hipCadd(hipDoubleComplex p, hipDoubleComplex q) {
return cuCadd(p, q);
}
__device__ __host__ static inline hipDoubleComplex hipCsub(hipDoubleComplex p, hipDoubleComplex q) {
return cuCsub(p, q);
}
__device__ __host__ static inline hipDoubleComplex hipCmul(hipDoubleComplex p, hipDoubleComplex q) {
return cuCmul(p, q);
}
__device__ __host__ static inline hipDoubleComplex hipCdiv(hipDoubleComplex p, hipDoubleComplex q) {
return cuCdiv(p, q);
}
__device__ __host__ static inline double hipCabs(hipDoubleComplex z) { return cuCabs(z); }
typedef cuFloatComplex hipComplex;
__device__ __host__ static inline hipComplex make_Complex(float x, float y) {
return make_cuComplex(x, y);
}
__device__ __host__ static inline hipFloatComplex hipComplexDoubleToFloat(hipDoubleComplex z) {
return cuComplexDoubleToFloat(z);
}
__device__ __host__ static inline hipDoubleComplex hipComplexFloatToDouble(hipFloatComplex z) {
return cuComplexFloatToDouble(z);
}
__device__ __host__ static inline hipComplex hipCfmaf(hipComplex p, hipComplex q, hipComplex r) {
return cuCfmaf(p, q, r);
}
__device__ __host__ static inline hipDoubleComplex hipCfma(hipDoubleComplex p, hipDoubleComplex q,
hipDoubleComplex r) {
return cuCfma(p, q, r);
}
#endif
include/hip/nvidia_detail/hip_cooperative_groups.h
-12
View File
@@ -1,12 +0,0 @@
#ifndef HIP_INCLUDE_HIP_NVIDIA_DETAIL_HIP_COOPERATIVE_GROUPS_H
#define HIP_INCLUDE_HIP_NVIDIA_DETAIL_HIP_COOPERATIVE_GROUPS_H
// Include CUDA headers
#include <cuda_runtime.h>
#include <cooperative_groups.h>
// Include HIP wrapper headers around CUDA
#include <hip/hip_runtime.h>
#include <hip/hip_runtime_api.h>
#endif // HIP_INCLUDE_HIP_NVIDIA_DETAIL_HIP_COOPERATIVE_GROUPS_H
include/hip/nvidia_detail/hip_runtime.h
-122
View File
@@ -1,122 +0,0 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_NVIDIA_DETAIL_HIP_RUNTIME_H
#define HIP_INCLUDE_HIP_NVIDIA_DETAIL_HIP_RUNTIME_H
#include <cuda_runtime.h>
#include <hip/hip_runtime_api.h>
#define HIP_KERNEL_NAME(...) __VA_ARGS__
typedef int hipLaunchParm;
#define hipLaunchKernelGGLInternal(kernelName, numBlocks, numThreads, memPerBlock, streamId, ...) \
do { \
kernelName<<<numBlocks, numThreads, memPerBlock, streamId>>>(__VA_ARGS__); \
} while (0)
#define hipLaunchKernelGGL(kernelName, ...) hipLaunchKernelGGLInternal((kernelName), __VA_ARGS__)
#define hipReadModeElementType cudaReadModeElementType
#ifdef __CUDA_ARCH__
// 32-bit Atomics:
#define __HIP_ARCH_HAS_GLOBAL_INT32_ATOMICS__ (__CUDA_ARCH__ >= 110)
#define __HIP_ARCH_HAS_GLOBAL_FLOAT_ATOMIC_EXCH__ (__CUDA_ARCH__ >= 110)
#define __HIP_ARCH_HAS_SHARED_INT32_ATOMICS__ (__CUDA_ARCH__ >= 120)
#define __HIP_ARCH_HAS_SHARED_FLOAT_ATOMIC_EXCH__ (__CUDA_ARCH__ >= 120)
#define __HIP_ARCH_HAS_FLOAT_ATOMIC_ADD__ (__CUDA_ARCH__ >= 200)
// 64-bit Atomics:
#define __HIP_ARCH_HAS_GLOBAL_INT64_ATOMICS__ (__CUDA_ARCH__ >= 200)
#define __HIP_ARCH_HAS_SHARED_INT64_ATOMICS__ (__CUDA_ARCH__ >= 120)
// Doubles
#define __HIP_ARCH_HAS_DOUBLES__ (__CUDA_ARCH__ >= 120)
// warp cross-lane operations:
#define __HIP_ARCH_HAS_WARP_VOTE__ (__CUDA_ARCH__ >= 120)
#define __HIP_ARCH_HAS_WARP_BALLOT__ (__CUDA_ARCH__ >= 200)
#define __HIP_ARCH_HAS_WARP_SHUFFLE__ (__CUDA_ARCH__ >= 300)
#define __HIP_ARCH_HAS_WARP_FUNNEL_SHIFT__ (__CUDA_ARCH__ >= 350)
// sync
#define __HIP_ARCH_HAS_THREAD_FENCE_SYSTEM__ (__CUDA_ARCH__ >= 200)
#define __HIP_ARCH_HAS_SYNC_THREAD_EXT__ (__CUDA_ARCH__ >= 200)
// misc
#define __HIP_ARCH_HAS_SURFACE_FUNCS__ (__CUDA_ARCH__ >= 200)
#define __HIP_ARCH_HAS_3DGRID__ (__CUDA_ARCH__ >= 200)
#define __HIP_ARCH_HAS_DYNAMIC_PARALLEL__ (__CUDA_ARCH__ >= 350)
#endif
#ifdef __CUDACC__
#define hipThreadIdx_x threadIdx.x
#define hipThreadIdx_y threadIdx.y
#define hipThreadIdx_z threadIdx.z
#define hipBlockIdx_x blockIdx.x
#define hipBlockIdx_y blockIdx.y
#define hipBlockIdx_z blockIdx.z
#define hipBlockDim_x blockDim.x
#define hipBlockDim_y blockDim.y
#define hipBlockDim_z blockDim.z
#define hipGridDim_x gridDim.x
#define hipGridDim_y gridDim.y
#define hipGridDim_z gridDim.z
#define HIP_SYMBOL(X) &X
/**
* Map HIP_DYNAMIC_SHARED to "extern __shared__" for compatibility with old HIP applications
* To be removed in a future release.
*/
#define HIP_DYNAMIC_SHARED(type, var) extern __shared__ type var[];
#define HIP_DYNAMIC_SHARED_ATTRIBUTE
#ifdef __HIP_DEVICE_COMPILE__
#define abort_() \
{ asm("trap;"); }
#undef assert
#define assert(COND) \
{ \
if (!COND) { \
abort_(); \
} \
}
#endif
#define __clock() clock()
#define __clock64() clock64()
#endif
#endif
include/hip/nvidia_detail/hip_runtime_api.h
-2195
View File
File diff suppressed because it is too large Load Diff
include/hip/nvidia_detail/hip_texture_types.h
-6
View File
@@ -1,6 +0,0 @@
#ifndef HIP_INCLUDE_HIP_NVIDIA_DETAIL_HIP_TEXTURE_TYPES_H
#define HIP_INCLUDE_HIP_NVIDIA_DETAIL_HIP_TEXTURE_TYPES_H
#include <texture_types.h>
#endif
include/hip/nvidia_detail/hiprtc.h
-168
View File
@@ -1,168 +0,0 @@
/*
Copyright (c) 2021 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIPRTC_H
#define HIPRTC_H
#include <cuda.h>
#include <nvrtc.h>
#ifdef __cplusplus
extern "C" {
#endif /* __cplusplus */
#include <stdlib.h>
#if !defined(_WIN32)
#pragma GCC visibility push(default)
#endif
typedef enum hiprtcResult {
HIPRTC_SUCCESS = 0,
HIPRTC_ERROR_OUT_OF_MEMORY = 1,
HIPRTC_ERROR_PROGRAM_CREATION_FAILURE = 2,
HIPRTC_ERROR_INVALID_INPUT = 3,
HIPRTC_ERROR_INVALID_PROGRAM = 4,
HIPRTC_ERROR_INVALID_OPTION = 5,
HIPRTC_ERROR_COMPILATION = 6,
HIPRTC_ERROR_BUILTIN_OPERATION_FAILURE = 7,
HIPRTC_ERROR_NO_NAME_EXPRESSIONS_AFTER_COMPILATION = 8,
HIPRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION = 9,
HIPRTC_ERROR_NAME_EXPRESSION_NOT_VALID = 10,
HIPRTC_ERROR_INTERNAL_ERROR = 11
} hiprtcResult;
inline static nvrtcResult hiprtcResultTonvrtcResult(hiprtcResult result) {
switch (result) {
case HIPRTC_SUCCESS:
return NVRTC_SUCCESS;
case HIPRTC_ERROR_OUT_OF_MEMORY:
return NVRTC_ERROR_OUT_OF_MEMORY;
case HIPRTC_ERROR_PROGRAM_CREATION_FAILURE:
return NVRTC_ERROR_PROGRAM_CREATION_FAILURE;
case HIPRTC_ERROR_INVALID_INPUT:
return NVRTC_ERROR_INVALID_INPUT;
case HIPRTC_ERROR_INVALID_PROGRAM:
return NVRTC_ERROR_INVALID_PROGRAM;
case HIPRTC_ERROR_INVALID_OPTION:
return NVRTC_ERROR_INVALID_OPTION;
case HIPRTC_ERROR_COMPILATION:
return NVRTC_ERROR_COMPILATION;
case HIPRTC_ERROR_BUILTIN_OPERATION_FAILURE:
return NVRTC_ERROR_BUILTIN_OPERATION_FAILURE;
case HIPRTC_ERROR_NO_NAME_EXPRESSIONS_AFTER_COMPILATION:
return NVRTC_ERROR_NO_NAME_EXPRESSIONS_AFTER_COMPILATION;
case HIPRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION:
return NVRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION;
case HIPRTC_ERROR_NAME_EXPRESSION_NOT_VALID:
return NVRTC_ERROR_NAME_EXPRESSION_NOT_VALID;
case HIPRTC_ERROR_INTERNAL_ERROR:
return NVRTC_ERROR_INTERNAL_ERROR;
}
}
inline static hiprtcResult nvrtcResultTohiprtcResult(nvrtcResult result) {
switch (result) {
case NVRTC_SUCCESS:
return HIPRTC_SUCCESS;
case NVRTC_ERROR_OUT_OF_MEMORY:
return HIPRTC_ERROR_OUT_OF_MEMORY;
case NVRTC_ERROR_PROGRAM_CREATION_FAILURE:
return HIPRTC_ERROR_PROGRAM_CREATION_FAILURE;
case NVRTC_ERROR_INVALID_INPUT:
return HIPRTC_ERROR_INVALID_INPUT;
case NVRTC_ERROR_INVALID_PROGRAM:
return HIPRTC_ERROR_INVALID_PROGRAM;
case NVRTC_ERROR_INVALID_OPTION:
return HIPRTC_ERROR_INVALID_OPTION;
case NVRTC_ERROR_COMPILATION:
return HIPRTC_ERROR_COMPILATION;
case NVRTC_ERROR_BUILTIN_OPERATION_FAILURE:
return HIPRTC_ERROR_BUILTIN_OPERATION_FAILURE;
case NVRTC_ERROR_NO_NAME_EXPRESSIONS_AFTER_COMPILATION:
return HIPRTC_ERROR_NO_NAME_EXPRESSIONS_AFTER_COMPILATION;
case NVRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION:
return HIPRTC_ERROR_NO_LOWERED_NAMES_BEFORE_COMPILATION;
case NVRTC_ERROR_NAME_EXPRESSION_NOT_VALID:
return HIPRTC_ERROR_NAME_EXPRESSION_NOT_VALID;
case NVRTC_ERROR_INTERNAL_ERROR:
return HIPRTC_ERROR_INTERNAL_ERROR;
}
}
const char* hiprtcGetErrorString(hiprtcResult result) {
return nvrtcGetErrorString(hiprtcResultTonvrtcResult(result));
}
hiprtcResult hiprtcVersion(int* major, int* minor) {
return nvrtcResultTohiprtcResult(nvrtcVersion(major, minor));
}
typedef nvrtcProgram hiprtcProgram;
hiprtcResult hiprtcAddNameExpression(hiprtcProgram prog, const char* name_expression) {
return nvrtcResultTohiprtcResult(nvrtcAddNameExpression(prog, name_expression));
}
hiprtcResult hiprtcCompileProgram(hiprtcProgram prog, int numOptions, const char** options) {
return nvrtcResultTohiprtcResult(nvrtcCompileProgram(prog, numOptions, options));
}
hiprtcResult hiprtcCreateProgram(hiprtcProgram* prog, const char* src, const char* name,
int numHeaders, const char** headers, const char** includeNames) {
return nvrtcResultTohiprtcResult(
nvrtcCreateProgram(prog, src, name, numHeaders, headers, includeNames));
}
hiprtcResult hiprtcDestroyProgram(hiprtcProgram* prog) {
return nvrtcResultTohiprtcResult(nvrtcDestroyProgram(prog));
}
hiprtcResult hiprtcGetLoweredName(hiprtcProgram prog, const char* name_expression,
const char** lowered_name) {
return nvrtcResultTohiprtcResult(nvrtcGetLoweredName(prog, name_expression, lowered_name));
}
hiprtcResult hiprtcGetProgramLog(hiprtcProgram prog, char* log) {
return nvrtcResultTohiprtcResult(nvrtcGetProgramLog(prog, log));
}
hiprtcResult hiprtcGetProgramLogSize(hiprtcProgram prog, size_t* logSizeRet) {
return nvrtcResultTohiprtcResult(nvrtcGetProgramLogSize(prog, logSizeRet));
}
hiprtcResult hiprtcGetCode(hiprtcProgram prog, char* code) {
return nvrtcResultTohiprtcResult(nvrtcGetPTX(prog, code));
}
hiprtcResult hiprtcGetCodeSize(hiprtcProgram prog, size_t* codeSizeRet) {
return nvrtcResultTohiprtcResult(nvrtcGetPTXSize(prog, codeSizeRet));
}
#if !defined(_WIN32)
#pragma GCC visibility pop
#endif
#ifdef __cplusplus
}
#endif /* __cplusplus */
#endif // HIPRTC_H
include/hip/texture_types.h
+1 -1
View File
@@ -26,7 +26,7 @@ THE SOFTWARE.
#include <hip/hip_common.h>
#if (defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && !(defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include <hip/amd_detail/texture_types.h>
#include <hip/amd_detail/amd_texture_types.h>
#elif !(defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__)) && (defined(__HIP_PLATFORM_NVCC__) || defined(__HIP_PLATFORM_NVIDIA__))
#include "texture_types.h"
#else