clr/hipamd/include/hip/amd_detail/amd_surface_functions.h Source File

• clr
• hipamd
• include
• hip
• amd_detail

amd_surface_functions.h

Go to the documentation of this file.

1/*

2Copyright (c) 2018 - 2023 Advanced Micro Devices, Inc. All rights reserved.

3

4Permission is hereby granted, free of charge, to any person obtaining a copy

5of this software and associated documentation files (the "Software"), to deal

6in the Software without restriction, including without limitation the rights

7to use, copy, modify, merge, publish, distribute, sublicense, and/or sell

8copies of the Software, and to permit persons to whom the Software is

9furnished to do so, subject to the following conditions:

10

11The above copyright notice and this permission notice shall be included in

12all copies or substantial portions of the Software.

13

14THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR

15IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

16FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE

17AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER

18LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,

19OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN

20THE SOFTWARE.

21*/

22

23#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_SURFACE_FUNCTIONS_H

24#define HIP_INCLUDE_HIP_AMD_DETAIL_SURFACE_FUNCTIONS_H

25

26#if defined(__cplusplus)

27

28#if !defined(__HIPCC_RTC__)

29#include <hip/surface_types.h>

30#include <hip/hip_vector_types.h>

31#include <hip/amd_detail/texture_fetch_functions.h>

32#include <hip/amd_detail/ockl_image.h>

33#endif

34

35#if defined(__HIPCC_RTC__)

36#define __HOST_DEVICE__ __device__

37#else

38#define __HOST_DEVICE__ __host__ __device__

39#endif

40

41#define __HIP_SURFACE_OBJECT_PARAMETERS_INIT \

42 unsigned int ADDRESS_SPACE_CONSTANT* i = (unsigned int ADDRESS_SPACE_CONSTANT*)surfObj;

43

49// CUDA is using byte address, need map to pixel address for HIP

50static __HOST_DEVICE__ __forceinline__ int __hipGetPixelAddr(int x, int format, int order) {

51 /*

52 * use below format index to generate format LUT

53 typedef enum {

54 HSA_EXT_IMAGE_CHANNEL_TYPE_SNORM_INT8 = 0,

55 HSA_EXT_IMAGE_CHANNEL_TYPE_SNORM_INT16 = 1,

56 HSA_EXT_IMAGE_CHANNEL_TYPE_UNORM_INT8 = 2,

57 HSA_EXT_IMAGE_CHANNEL_TYPE_UNORM_INT16 = 3,

58 HSA_EXT_IMAGE_CHANNEL_TYPE_UNORM_INT24 = 4,

59 HSA_EXT_IMAGE_CHANNEL_TYPE_UNORM_SHORT_555 = 5,

60 HSA_EXT_IMAGE_CHANNEL_TYPE_UNORM_SHORT_565 = 6,

61 HSA_EXT_IMAGE_CHANNEL_TYPE_UNORM_SHORT_101010 = 7,

62 HSA_EXT_IMAGE_CHANNEL_TYPE_SIGNED_INT8 = 8,

63 HSA_EXT_IMAGE_CHANNEL_TYPE_SIGNED_INT16 = 9,

64 HSA_EXT_IMAGE_CHANNEL_TYPE_SIGNED_INT32 = 10,

65 HSA_EXT_IMAGE_CHANNEL_TYPE_UNSIGNED_INT8 = 11,

66 HSA_EXT_IMAGE_CHANNEL_TYPE_UNSIGNED_INT16 = 12,

67 HSA_EXT_IMAGE_CHANNEL_TYPE_UNSIGNED_INT32 = 13,

68 HSA_EXT_IMAGE_CHANNEL_TYPE_HALF_FLOAT = 14,

69 HSA_EXT_IMAGE_CHANNEL_TYPE_FLOAT = 15

70 } hsa_ext_image_channel_type_t;

71 */

72 static const int FormatLUT[] = { 0, 1, 0, 1, 3, 1, 1, 1, 0, 1, 2, 0, 1, 2, 1, 2 };

73 x = FormatLUT[format] == 3 ? x / FormatLUT[format] : x >> FormatLUT[format];

74

75 /*

76 * use below order index to generate order LUT

77 typedef enum {

78 HSA_EXT_IMAGE_CHANNEL_ORDER_A = 0,

79 HSA_EXT_IMAGE_CHANNEL_ORDER_R = 1,

80 HSA_EXT_IMAGE_CHANNEL_ORDER_RX = 2,

81 HSA_EXT_IMAGE_CHANNEL_ORDER_RG = 3,

82 HSA_EXT_IMAGE_CHANNEL_ORDER_RGX = 4,

83 HSA_EXT_IMAGE_CHANNEL_ORDER_RA = 5,

84 HSA_EXT_IMAGE_CHANNEL_ORDER_RGB = 6,

85 HSA_EXT_IMAGE_CHANNEL_ORDER_RGBX = 7,

86 HSA_EXT_IMAGE_CHANNEL_ORDER_RGBA = 8,

87 HSA_EXT_IMAGE_CHANNEL_ORDER_BGRA = 9,

88 HSA_EXT_IMAGE_CHANNEL_ORDER_ARGB = 10,

89 HSA_EXT_IMAGE_CHANNEL_ORDER_ABGR = 11,

90 HSA_EXT_IMAGE_CHANNEL_ORDER_SRGB = 12,

91 HSA_EXT_IMAGE_CHANNEL_ORDER_SRGBX = 13,

92 HSA_EXT_IMAGE_CHANNEL_ORDER_SRGBA = 14,

93 HSA_EXT_IMAGE_CHANNEL_ORDER_SBGRA = 15,

94 HSA_EXT_IMAGE_CHANNEL_ORDER_INTENSITY = 16,

95 HSA_EXT_IMAGE_CHANNEL_ORDER_LUMINANCE = 17,

96 HSA_EXT_IMAGE_CHANNEL_ORDER_DEPTH = 18,

97 HSA_EXT_IMAGE_CHANNEL_ORDER_DEPTH_STENCIL = 19

98 } hsa_ext_image_channel_order_t;

99 */

100 static const int OrderLUT[] = { 0, 0, 1, 1, 3, 1, 3, 2, 2, 2, 2, 2, 3, 2, 2, 2, 0, 0, 0, 0 };

101 return x = OrderLUT[order] == 3 ? x / OrderLUT[order] : x >> OrderLUT[order];

102}

103

112template <

113 typename T,

114 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

115static __device__ __hip_img_chk__ void surf1Dread(T* data, hipSurfaceObject_t surfObj, int x,

116 int boundaryMode = hipBoundaryModeZero) {

117 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

118 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_1D(i), __ockl_image_channel_order_1D(i));

119 auto tmp = __ockl_image_load_1D(i, x);

120 *data = __hipMapFrom<T>(tmp);

121}

122

130template <

131 typename T,

132 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

133static __device__ __hip_img_chk__ void surf1Dwrite(T data, hipSurfaceObject_t surfObj, int x) {

134 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

135 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_1D(i), __ockl_image_channel_order_1D(i));

136 auto tmp = __hipMapTo<float4::Native_vec_>(data);

137 __ockl_image_store_1D(i, x, tmp);

138}

139

140

149template <

150 typename T,

151 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

152static __device__ __hip_img_chk__ void surf2Dread(T* data, hipSurfaceObject_t surfObj, int x, int y) {

153 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

154 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_2D(i), __ockl_image_channel_order_2D(i));

155 auto tmp = __ockl_image_load_2D(i, int2(x, y).data);

156 *data = __hipMapFrom<T>(tmp);

157}

158

168template <

169 typename T,

170 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

171static __device__ __hip_img_chk__ void surf2Dwrite(T data, hipSurfaceObject_t surfObj, int x, int y) {

172 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

173 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_2D(i), __ockl_image_channel_order_2D(i));

174 auto tmp = __hipMapTo<float4::Native_vec_>(data);

175 __ockl_image_store_2D(i, int2(x, y).data, tmp);

176}

177

188template <

189 typename T,

190 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

191static __device__ __hip_img_chk__ void surf3Dread(T* data, hipSurfaceObject_t surfObj, int x, int y, int z) {

192 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

193 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_3D(i), __ockl_image_channel_order_3D(i));

194 auto tmp = __ockl_image_load_3D(i, int4(x, y, z, 0).data);

195 *data = __hipMapFrom<T>(tmp);

196}

197

208template <

209 typename T,

210 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

211static __device__ __hip_img_chk__ void surf3Dwrite(T data, hipSurfaceObject_t surfObj, int x, int y, int z) {

212 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

213 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_3D(i), __ockl_image_channel_order_3D(i));

214 auto tmp = __hipMapTo<float4::Native_vec_>(data);

215 __ockl_image_store_3D(i, int4(x, y, z, 0).data, tmp);

216}

217

227template <

228 typename T,

229 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

230static __device__ __hip_img_chk__ void surf1DLayeredread(T* data, hipSurfaceObject_t surfObj, int x, int layer) {

231 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

232 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_1D(i), __ockl_image_channel_order_1D(i));

233 auto tmp = __ockl_image_load_lod_1D(i, x, layer);

234 *data = __hipMapFrom<T>(tmp);

235}

236

246template <

247 typename T,

248 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

249static __device__ __hip_img_chk__ void surf1DLayeredwrite(T data, hipSurfaceObject_t surfObj, int x, int layer) {

250 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

251 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_1D(i), __ockl_image_channel_order_1D(i));

252 auto tmp = __hipMapTo<float4::Native_vec_>(data);

253 __ockl_image_store_lod_1D(i, x, layer, tmp);

254}

255

266template <

267 typename T,

268 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

269static __device__ __hip_img_chk__ void surf2DLayeredread(T* data, hipSurfaceObject_t surfObj, int x, int y, int layer) {

270 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

271 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_2D(i), __ockl_image_channel_order_2D(i));

272 auto tmp = __ockl_image_load_lod_2D(i, int2(x, y).data, layer);

273 *data = __hipMapFrom<T>(tmp);

274}

275

286template <

287 typename T,

288 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

289static __device__ __hip_img_chk__ void surf2DLayeredwrite(T data, hipSurfaceObject_t surfObj, int x, int y, int layer) {

290 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

291 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_2D(i), __ockl_image_channel_order_2D(i));

292 auto tmp = __hipMapTo<float4::Native_vec_>(data);

293 __ockl_image_store_lod_2D(i, int2(x, y).data, layer, tmp);

294}

295

306template <

307 typename T,

308 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

309static __device__ __hip_img_chk__ void surfCubemapread(T* data, hipSurfaceObject_t surfObj, int x, int y, int face) {

310 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

311 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_2D(i), __ockl_image_channel_order_2D(i));

312 auto tmp = __ockl_image_load_CM(i, int2(x, y).data, face);

313 *data = __hipMapFrom<T>(tmp);

314}

315

326template <

327 typename T,

328 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

329static __device__ __hip_img_chk__ void surfCubemapwrite(T data, hipSurfaceObject_t surfObj, int x, int y, int face) {

330 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

331 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_2D(i), __ockl_image_channel_order_2D(i));

332 auto tmp = __hipMapTo<float4::Native_vec_>(data);

333 __ockl_image_store_CM(i, int2(x, y).data, face, tmp);

334}

335

347template <

348 typename T,

349 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

350static __device__ __hip_img_chk__ void surfCubemapLayeredread(T* data, hipSurfaceObject_t surfObj, int x, int y, int face,

351 int layer) {

352 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

353 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_2D(i), __ockl_image_channel_order_2D(i));

354 auto tmp = __ockl_image_load_lod_CM(i, int2(x, y).data, face, layer);

355 *data = __hipMapFrom<T>(tmp);

356}

357

369template <

370 typename T,

371 typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value>::type* = nullptr>

372static __device__ __hip_img_chk__ void surfCubemapLayeredwrite(T* data, hipSurfaceObject_t surfObj, int x, int y, int face,

373 int layer) {

374 __HIP_SURFACE_OBJECT_PARAMETERS_INIT

375 x = __hipGetPixelAddr(x, __ockl_image_channel_data_type_2D(i), __ockl_image_channel_order_2D(i));

376 auto tmp = __hipMapTo<float4::Native_vec_>(data);

377 __ockl_image_store_lod_CM(i, int2(x, y).data, face, layer, tmp);

378}

379

380// Doxygen end group SurfaceAPI

385#endif

386

387#endif

__HIP_SURFACE_OBJECT_PARAMETERS_INIT

#define __HIP_SURFACE_OBJECT_PARAMETERS_INIT

Definition amd_surface_functions.h:41

__HOST_DEVICE__

#define __HOST_DEVICE__

Definition amd_surface_functions.h:38

hip_vector_types.h

surface_types.h

Defines surface types for HIP runtime.

hipBoundaryModeZero

@ hipBoundaryModeZero

Definition surface_types.h:54

hipSurfaceObject_t

struct __hip_surface * hipSurfaceObject_t

Definition surface_types.h:41