triton-windows 3.3.0.post19__cp313-cp313-win_amd64.whl → 3.3.1.post21__cp313-cp313-win_amd64.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Potentially problematic release.


This version of triton-windows might be problematic. Click here for more details.

Files changed (116) hide show
  1. triton/_C/libtriton.pyd +0 -0
  2. triton/__init__.py +1 -1
  3. triton/backends/amd/driver.py +6 -1
  4. triton/backends/nvidia/compiler.py +1 -3
  5. triton/backends/nvidia/driver.py +7 -3
  6. triton/runtime/autotuner.py +2 -2
  7. triton/runtime/build.py +5 -5
  8. triton/runtime/tcc/lib/python310.def +1610 -0
  9. triton/runtime/tcc/lib/python311.def +1633 -0
  10. triton/runtime/tcc/lib/python312.def +1703 -0
  11. triton/runtime/tcc/lib/python313.def +1651 -0
  12. triton/runtime/tcc/lib/python313t.def +1656 -0
  13. triton/runtime/tcc/lib/python39.def +1644 -0
  14. triton/runtime/tcc/lib/python3t.def +905 -0
  15. triton/windows_utils.py +11 -4
  16. {triton_windows-3.3.0.post19.dist-info → triton_windows-3.3.1.post21.dist-info}/METADATA +1 -1
  17. {triton_windows-3.3.0.post19.dist-info → triton_windows-3.3.1.post21.dist-info}/RECORD +19 -109
  18. {triton_windows-3.3.0.post19.dist-info → triton_windows-3.3.1.post21.dist-info}/WHEEL +1 -1
  19. triton/backends/amd/include/hip/amd_detail/amd_channel_descriptor.h +0 -358
  20. triton/backends/amd/include/hip/amd_detail/amd_device_functions.h +0 -1010
  21. triton/backends/amd/include/hip/amd_detail/amd_hip_atomic.h +0 -1638
  22. triton/backends/amd/include/hip/amd_detail/amd_hip_bf16.h +0 -1814
  23. triton/backends/amd/include/hip/amd_detail/amd_hip_bfloat16.h +0 -293
  24. triton/backends/amd/include/hip/amd_detail/amd_hip_common.h +0 -32
  25. triton/backends/amd/include/hip/amd_detail/amd_hip_complex.h +0 -174
  26. triton/backends/amd/include/hip/amd_detail/amd_hip_cooperative_groups.h +0 -835
  27. triton/backends/amd/include/hip/amd_detail/amd_hip_fp16.h +0 -1809
  28. triton/backends/amd/include/hip/amd_detail/amd_hip_fp8.h +0 -1391
  29. triton/backends/amd/include/hip/amd_detail/amd_hip_gl_interop.h +0 -108
  30. triton/backends/amd/include/hip/amd_detail/amd_hip_math_constants.h +0 -124
  31. triton/backends/amd/include/hip/amd_detail/amd_hip_runtime.h +0 -405
  32. triton/backends/amd/include/hip/amd_detail/amd_hip_runtime_pt_api.h +0 -196
  33. triton/backends/amd/include/hip/amd_detail/amd_hip_unsafe_atomics.h +0 -565
  34. triton/backends/amd/include/hip/amd_detail/amd_hip_vector_types.h +0 -2226
  35. triton/backends/amd/include/hip/amd_detail/amd_math_functions.h +0 -104
  36. triton/backends/amd/include/hip/amd_detail/amd_surface_functions.h +0 -244
  37. triton/backends/amd/include/hip/amd_detail/amd_warp_functions.h +0 -538
  38. triton/backends/amd/include/hip/amd_detail/amd_warp_sync_functions.h +0 -288
  39. triton/backends/amd/include/hip/amd_detail/concepts.hpp +0 -30
  40. triton/backends/amd/include/hip/amd_detail/device_library_decls.h +0 -133
  41. triton/backends/amd/include/hip/amd_detail/functional_grid_launch.hpp +0 -218
  42. triton/backends/amd/include/hip/amd_detail/grid_launch.h +0 -67
  43. triton/backends/amd/include/hip/amd_detail/grid_launch.hpp +0 -50
  44. triton/backends/amd/include/hip/amd_detail/grid_launch_GGL.hpp +0 -26
  45. triton/backends/amd/include/hip/amd_detail/helpers.hpp +0 -137
  46. triton/backends/amd/include/hip/amd_detail/hip_api_trace.hpp +0 -1446
  47. triton/backends/amd/include/hip/amd_detail/hip_assert.h +0 -101
  48. triton/backends/amd/include/hip/amd_detail/hip_cooperative_groups_helper.h +0 -242
  49. triton/backends/amd/include/hip/amd_detail/hip_fp16_gcc.h +0 -254
  50. triton/backends/amd/include/hip/amd_detail/hip_fp16_math_fwd.h +0 -96
  51. triton/backends/amd/include/hip/amd_detail/hip_ldg.h +0 -100
  52. triton/backends/amd/include/hip/amd_detail/hip_prof_str.h +0 -10570
  53. triton/backends/amd/include/hip/amd_detail/hip_runtime_prof.h +0 -78
  54. triton/backends/amd/include/hip/amd_detail/host_defines.h +0 -184
  55. triton/backends/amd/include/hip/amd_detail/hsa_helpers.hpp +0 -102
  56. triton/backends/amd/include/hip/amd_detail/macro_based_grid_launch.hpp +0 -798
  57. triton/backends/amd/include/hip/amd_detail/math_fwd.h +0 -698
  58. triton/backends/amd/include/hip/amd_detail/ockl_image.h +0 -177
  59. triton/backends/amd/include/hip/amd_detail/program_state.hpp +0 -107
  60. triton/backends/amd/include/hip/amd_detail/texture_fetch_functions.h +0 -491
  61. triton/backends/amd/include/hip/amd_detail/texture_indirect_functions.h +0 -478
  62. triton/backends/amd/include/hip/channel_descriptor.h +0 -39
  63. triton/backends/amd/include/hip/device_functions.h +0 -38
  64. triton/backends/amd/include/hip/driver_types.h +0 -468
  65. triton/backends/amd/include/hip/hip_bf16.h +0 -36
  66. triton/backends/amd/include/hip/hip_bfloat16.h +0 -44
  67. triton/backends/amd/include/hip/hip_common.h +0 -100
  68. triton/backends/amd/include/hip/hip_complex.h +0 -38
  69. triton/backends/amd/include/hip/hip_cooperative_groups.h +0 -46
  70. triton/backends/amd/include/hip/hip_deprecated.h +0 -95
  71. triton/backends/amd/include/hip/hip_ext.h +0 -161
  72. triton/backends/amd/include/hip/hip_fp16.h +0 -36
  73. triton/backends/amd/include/hip/hip_fp8.h +0 -33
  74. triton/backends/amd/include/hip/hip_gl_interop.h +0 -32
  75. triton/backends/amd/include/hip/hip_hcc.h +0 -24
  76. triton/backends/amd/include/hip/hip_math_constants.h +0 -36
  77. triton/backends/amd/include/hip/hip_profile.h +0 -27
  78. triton/backends/amd/include/hip/hip_runtime.h +0 -75
  79. triton/backends/amd/include/hip/hip_runtime_api.h +0 -9261
  80. triton/backends/amd/include/hip/hip_texture_types.h +0 -29
  81. triton/backends/amd/include/hip/hip_vector_types.h +0 -41
  82. triton/backends/amd/include/hip/hip_version.h +0 -17
  83. triton/backends/amd/include/hip/hiprtc.h +0 -421
  84. triton/backends/amd/include/hip/library_types.h +0 -78
  85. triton/backends/amd/include/hip/math_functions.h +0 -42
  86. triton/backends/amd/include/hip/surface_types.h +0 -63
  87. triton/backends/amd/include/hip/texture_types.h +0 -194
  88. triton/backends/amd/include/hsa/Brig.h +0 -1131
  89. triton/backends/amd/include/hsa/amd_hsa_common.h +0 -91
  90. triton/backends/amd/include/hsa/amd_hsa_elf.h +0 -462
  91. triton/backends/amd/include/hsa/amd_hsa_kernel_code.h +0 -269
  92. triton/backends/amd/include/hsa/amd_hsa_queue.h +0 -109
  93. triton/backends/amd/include/hsa/amd_hsa_signal.h +0 -80
  94. triton/backends/amd/include/hsa/hsa.h +0 -5738
  95. triton/backends/amd/include/hsa/hsa_amd_tool.h +0 -91
  96. triton/backends/amd/include/hsa/hsa_api_trace.h +0 -579
  97. triton/backends/amd/include/hsa/hsa_api_trace_version.h +0 -68
  98. triton/backends/amd/include/hsa/hsa_ext_amd.h +0 -3146
  99. triton/backends/amd/include/hsa/hsa_ext_finalize.h +0 -531
  100. triton/backends/amd/include/hsa/hsa_ext_image.h +0 -1454
  101. triton/backends/amd/include/hsa/hsa_ven_amd_aqlprofile.h +0 -488
  102. triton/backends/amd/include/hsa/hsa_ven_amd_loader.h +0 -667
  103. triton/backends/amd/include/hsa/hsa_ven_amd_pc_sampling.h +0 -416
  104. triton/backends/amd/include/roctracer/ext/prof_protocol.h +0 -107
  105. triton/backends/amd/include/roctracer/hip_ostream_ops.h +0 -4515
  106. triton/backends/amd/include/roctracer/hsa_ostream_ops.h +0 -1727
  107. triton/backends/amd/include/roctracer/hsa_prof_str.h +0 -3059
  108. triton/backends/amd/include/roctracer/roctracer.h +0 -779
  109. triton/backends/amd/include/roctracer/roctracer_ext.h +0 -81
  110. triton/backends/amd/include/roctracer/roctracer_hcc.h +0 -24
  111. triton/backends/amd/include/roctracer/roctracer_hip.h +0 -37
  112. triton/backends/amd/include/roctracer/roctracer_hsa.h +0 -112
  113. triton/backends/amd/include/roctracer/roctracer_plugin.h +0 -137
  114. triton/backends/amd/include/roctracer/roctracer_roctx.h +0 -67
  115. triton/backends/amd/include/roctracer/roctx.h +0 -229
  116. {triton_windows-3.3.0.post19.dist-info → triton_windows-3.3.1.post21.dist-info}/top_level.txt +0 -0
@@ -1,491 +0,0 @@
1
- /*
2
- Copyright (c) 2015 - 2023 Advanced Micro Devices, Inc. All rights reserved.
3
-
4
- Permission is hereby granted, free of charge, to any person obtaining a copy
5
- of this software and associated documentation files (the "Software"), to deal
6
- in the Software without restriction, including without limitation the rights
7
- to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
8
- copies of the Software, and to permit persons to whom the Software is
9
- furnished to do so, subject to the following conditions:
10
-
11
- The above copyright notice and this permission notice shall be included in
12
- all copies or substantial portions of the Software.
13
-
14
- THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15
- IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16
- FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
17
- AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18
- LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
19
- OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
20
- THE SOFTWARE.
21
- */
22
-
23
- #pragma once
24
-
25
- #if defined(__cplusplus)
26
-
27
- #if !defined(__HIPCC_RTC__)
28
- #include <hip/hip_vector_types.h>
29
- #include <hip/hip_texture_types.h>
30
- #include <hip/amd_detail/ockl_image.h>
31
- #include <type_traits>
32
- #endif // !defined(__HIPCC_RTC__)
33
-
34
- #define TEXTURE_PARAMETERS_INIT \
35
- unsigned int ADDRESS_SPACE_CONSTANT* i = (unsigned int ADDRESS_SPACE_CONSTANT*)t.textureObject; \
36
- unsigned int ADDRESS_SPACE_CONSTANT* s = i + HIP_SAMPLER_OBJECT_OFFSET_DWORD;
37
-
38
- template<typename T>
39
- struct __hip_is_tex_surf_scalar_channel_type
40
- {
41
- static constexpr bool value =
42
- std::is_same<T, char>::value ||
43
- std::is_same<T, unsigned char>::value ||
44
- std::is_same<T, short>::value ||
45
- std::is_same<T, unsigned short>::value ||
46
- std::is_same<T, int>::value ||
47
- std::is_same<T, unsigned int>::value ||
48
- std::is_same<T, float>::value;
49
- };
50
-
51
- template<typename T>
52
- struct __hip_is_tex_surf_channel_type
53
- {
54
- static constexpr bool value =
55
- __hip_is_tex_surf_scalar_channel_type<T>::value;
56
- };
57
-
58
- template<
59
- typename T,
60
- unsigned int rank>
61
- struct __hip_is_tex_surf_channel_type<HIP_vector_type<T, rank>>
62
- {
63
- static constexpr bool value =
64
- __hip_is_tex_surf_scalar_channel_type<T>::value &&
65
- ((rank == 1) ||
66
- (rank == 2) ||
67
- (rank == 4));
68
- };
69
-
70
- template<typename T>
71
- struct __hip_is_tex_normalized_channel_type
72
- {
73
- static constexpr bool value =
74
- std::is_same<T, char>::value ||
75
- std::is_same<T, unsigned char>::value ||
76
- std::is_same<T, short>::value ||
77
- std::is_same<T, unsigned short>::value;
78
- };
79
-
80
- template<
81
- typename T,
82
- unsigned int rank>
83
- struct __hip_is_tex_normalized_channel_type<HIP_vector_type<T, rank>>
84
- {
85
- static constexpr bool value =
86
- __hip_is_tex_normalized_channel_type<T>::value &&
87
- ((rank == 1) ||
88
- (rank == 2) ||
89
- (rank == 4));
90
- };
91
-
92
- template <
93
- typename T,
94
- hipTextureReadMode readMode,
95
- typename Enable = void>
96
- struct __hip_tex_ret
97
- {
98
- static_assert(std::is_same<Enable, void>::value, "Invalid channel type!");
99
- };
100
-
101
- /*
102
- * Map from device function return U to scalar texture type T
103
- */
104
- template<typename T, typename U>
105
- __forceinline__ __device__
106
- typename std::enable_if<
107
- __hip_is_tex_surf_scalar_channel_type<T>::value, const T>::type
108
- __hipMapFrom(const U &u) {
109
- if constexpr (sizeof(T) < sizeof(float)) {
110
- union {
111
- U u;
112
- int i;
113
- } d = { u };
114
- return static_cast<T>(d.i);
115
- } else { // sizeof(T) == sizeof(float)
116
- union {
117
- U u;
118
- T t;
119
- } d = { u };
120
- return d.t;
121
- }
122
- }
123
-
124
- /*
125
- * Map from device function return U to vector texture type T
126
- */
127
- template<typename T, typename U>
128
- __forceinline__ __device__
129
- typename std::enable_if<
130
- __hip_is_tex_surf_scalar_channel_type<typename T::value_type>::value, const T>::type
131
- __hipMapFrom(const U &u) {
132
- if constexpr (sizeof(typename T::value_type) < sizeof(float)) {
133
- union {
134
- U u;
135
- int4 i4;
136
- } d = { u };
137
- return __hipMapVector<typename T::value_type, sizeof(T)/sizeof(typename T::value_type)>(d.i4);
138
- } else { // sizeof(typename T::value_type) == sizeof(float)
139
- union {
140
- U u;
141
- T t;
142
- } d = { u };
143
- return d.t;
144
- }
145
- }
146
-
147
- /*
148
- * Map from scalar texture type T to device function input U
149
- */
150
- template<typename U, typename T>
151
- __forceinline__ __device__
152
- typename std::enable_if<
153
- __hip_is_tex_surf_scalar_channel_type<T>::value, const U>::type
154
- __hipMapTo(const T &t) {
155
- if constexpr (sizeof(T) < sizeof(float)) {
156
- union {
157
- U u;
158
- int i;
159
- } d = { 0 };
160
- d.i = static_cast<int>(t);
161
- return d.u;
162
- } else { // sizeof(T) == sizeof(float)
163
- union {
164
- U u;
165
- T t;
166
- } d = { 0 };
167
- d.t = t;
168
- return d.u;
169
- }
170
- }
171
-
172
- /*
173
- * Map from vector texture type T to device function input U
174
- */
175
- template<typename U, typename T>
176
- __forceinline__ __device__
177
- typename std::enable_if<
178
- __hip_is_tex_surf_scalar_channel_type<typename T::value_type>::value, const U>::type
179
- __hipMapTo(const T &t) {
180
- if constexpr (sizeof(typename T::value_type) < sizeof(float)) {
181
- union {
182
- U u;
183
- int4 i4;
184
- } d = { 0 };
185
- d.i4 = __hipMapVector<int, 4>(t);
186
- return d.u;
187
- } else { // sizeof(typename T::value_type) == sizeof(float)
188
- union {
189
- U u;
190
- T t;
191
- } d = { 0 };
192
- d.t = t;
193
- return d.u;
194
- }
195
- }
196
-
197
- template <
198
- typename T,
199
- hipTextureReadMode readMode>
200
- using __hip_tex_ret_t = typename __hip_tex_ret<T, readMode, bool>::type;
201
-
202
- template <typename T>
203
- struct __hip_tex_ret<
204
- T,
205
- hipReadModeElementType,
206
- typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value, bool>::type>
207
- {
208
- using type = T;
209
- };
210
-
211
- template<
212
- typename T,
213
- unsigned int rank>
214
- struct __hip_tex_ret<
215
- HIP_vector_type<T, rank>,
216
- hipReadModeElementType,
217
- typename std::enable_if<__hip_is_tex_surf_channel_type<HIP_vector_type<T, rank>>::value, bool>::type>
218
- {
219
- using type = HIP_vector_type<__hip_tex_ret_t<T, hipReadModeElementType>, rank>;
220
- };
221
-
222
- template<typename T>
223
- struct __hip_tex_ret<
224
- T,
225
- hipReadModeNormalizedFloat,
226
- typename std::enable_if<__hip_is_tex_normalized_channel_type<T>::value, bool>::type>
227
- {
228
- using type = float;
229
- };
230
-
231
- template<
232
- typename T,
233
- unsigned int rank>
234
- struct __hip_tex_ret<
235
- HIP_vector_type<T, rank>,
236
- hipReadModeNormalizedFloat,
237
- typename std::enable_if<__hip_is_tex_normalized_channel_type<HIP_vector_type<T, rank>>::value, bool>::type>
238
- {
239
- using type = HIP_vector_type<__hip_tex_ret_t<T, hipReadModeNormalizedFloat>, rank>;
240
- };
241
-
242
-
243
- template <typename T, hipTextureReadMode readMode>
244
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex1Dfetch(texture<T, hipTextureType1D, readMode> t, int x)
245
- {
246
- TEXTURE_PARAMETERS_INIT;
247
- auto tmp = __ockl_image_load_1Db(i, x);
248
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
249
- }
250
-
251
- template <typename T, hipTextureReadMode readMode>
252
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex1D(texture<T, hipTextureType1D, readMode> t, float x)
253
- {
254
- TEXTURE_PARAMETERS_INIT;
255
- auto tmp = __ockl_image_sample_1D(i, s, x);
256
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
257
- }
258
-
259
- template <typename T, hipTextureReadMode readMode>
260
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex2D(texture<T, hipTextureType2D, readMode> t, float x, float y)
261
- {
262
- TEXTURE_PARAMETERS_INIT;
263
- auto tmp = __ockl_image_sample_2D(i, s, float2(x, y).data);
264
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
265
- }
266
-
267
- template <typename T, hipTextureReadMode readMode>
268
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex1DLayered(texture<T, hipTextureType1DLayered, readMode> t, float x, int layer)
269
- {
270
- TEXTURE_PARAMETERS_INIT;
271
- auto tmp = __ockl_image_sample_1Da(i, s, float2(x, layer).data);
272
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
273
- }
274
-
275
- template <typename T, hipTextureReadMode readMode>
276
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex2DLayered(texture<T, hipTextureType2DLayered, readMode> t, float x, float y, int layer)
277
- {
278
- TEXTURE_PARAMETERS_INIT;
279
- auto tmp = __ockl_image_sample_2Da(i, s, float4(x, y, layer, 0.0f).data);
280
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
281
- }
282
-
283
- template <typename T, hipTextureReadMode readMode>
284
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex3D(texture<T, hipTextureType3D, readMode> t, float x, float y, float z)
285
- {
286
- TEXTURE_PARAMETERS_INIT;
287
- auto tmp = __ockl_image_sample_3D(i, s, float4(x, y, z, 0.0f).data);
288
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
289
- }
290
-
291
- template <typename T, hipTextureReadMode readMode>
292
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> texCubemap(texture<T, hipTextureTypeCubemap, readMode> t, float x, float y, float z)
293
- {
294
- TEXTURE_PARAMETERS_INIT;
295
- auto tmp = __ockl_image_sample_CM(i, s, float4(x, y, z, 0.0f).data);
296
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
297
- }
298
-
299
- template <typename T, hipTextureReadMode readMode>
300
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex1DLod(texture<T, hipTextureType1D, readMode> t, float x, float level)
301
- {
302
- TEXTURE_PARAMETERS_INIT;
303
- auto tmp = __ockl_image_sample_lod_1D(i, s, x, level);
304
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
305
- }
306
-
307
- template <typename T, hipTextureReadMode readMode>
308
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex2DLod(texture<T, hipTextureType2D, readMode> t, float x, float y, float level)
309
- {
310
- TEXTURE_PARAMETERS_INIT;
311
- auto tmp = __ockl_image_sample_lod_2D(i, s, float2(x, y).data, level);
312
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
313
- }
314
-
315
- template <typename T, hipTextureReadMode readMode>
316
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex1DLayeredLod(texture<T, hipTextureType1DLayered, readMode> t, float x, int layer, float level)
317
- {
318
- TEXTURE_PARAMETERS_INIT;
319
- auto tmp = __ockl_image_sample_lod_1Da(i, s, float2(x, layer).data, level);
320
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
321
- }
322
-
323
- template <typename T, hipTextureReadMode readMode>
324
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex2DLayeredLod(texture<T, hipTextureType2DLayered, readMode> t, float x, float y, int layer, float level)
325
- {
326
- TEXTURE_PARAMETERS_INIT;
327
- auto tmp = __ockl_image_sample_lod_2Da(i, s, float4(x, y, layer, 0.0f).data, level);
328
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
329
- }
330
-
331
- template <typename T, hipTextureReadMode readMode>
332
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex3DLod(texture<T, hipTextureType3D, readMode> t, float x, float y, float z, float level)
333
- {
334
- TEXTURE_PARAMETERS_INIT;
335
- auto tmp = __ockl_image_sample_lod_3D(i, s, float4(x, y, z, 0.0f).data, level);
336
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
337
- }
338
-
339
- template <typename T, hipTextureReadMode readMode>
340
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> texCubemapLod(texture<T, hipTextureTypeCubemap, readMode> t, float x, float y, float z, float level)
341
- {
342
- TEXTURE_PARAMETERS_INIT;
343
- auto tmp = __ockl_image_sample_lod_CM(i, s, float4(x, y, z, 0.0f).data, level);
344
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
345
- }
346
-
347
- template <typename T, hipTextureReadMode readMode>
348
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> texCubemapLayered(texture<T, hipTextureTypeCubemapLayered, readMode> t, float x, float y, float z, int layer)
349
- {
350
- TEXTURE_PARAMETERS_INIT;
351
- auto tmp = __ockl_image_sample_CMa(i, s, float4(x, y, z, layer).data);
352
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
353
- }
354
-
355
- template <typename T, hipTextureReadMode readMode>
356
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> texCubemapLayeredLod(texture<T, hipTextureTypeCubemapLayered, readMode> t, float x, float y, float z, int layer, float level)
357
- {
358
- TEXTURE_PARAMETERS_INIT;
359
- auto tmp = __ockl_image_sample_lod_CMa(i, s, float4(x, y, z, layer).data, level);
360
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
361
- }
362
-
363
- template <typename T, hipTextureReadMode readMode>
364
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> texCubemapGrad(texture<T, hipTextureTypeCubemap, readMode> t, float x, float y, float z, float4 dPdx, float4 dPdy)
365
- {
366
- TEXTURE_PARAMETERS_INIT;
367
- // TODO missing in device libs.
368
- // 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);
369
- // return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
370
- return {};
371
- }
372
-
373
- template <typename T, hipTextureReadMode readMode>
374
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> texCubemapLayeredGrad(texture<T, hipTextureTypeCubemapLayered, readMode> t, float x, float y, float z, int layer, float4 dPdx, float4 dPdy)
375
- {
376
- TEXTURE_PARAMETERS_INIT;
377
- // TODO missing in device libs.
378
- // 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);
379
- // return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
380
- return {};
381
- }
382
-
383
- template <typename T, hipTextureReadMode readMode>
384
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex1DGrad(texture<T, hipTextureType1D, readMode> t, float x, float dPdx, float dPdy)
385
- {
386
- TEXTURE_PARAMETERS_INIT;
387
- auto tmp = __ockl_image_sample_grad_1D(i, s, x, dPdx, dPdy);
388
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
389
- }
390
-
391
- template <typename T, hipTextureReadMode readMode>
392
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex2DGrad(texture<T, hipTextureType2D, readMode> t, float x, float y, float2 dPdx, float2 dPdy)
393
- {
394
- TEXTURE_PARAMETERS_INIT;
395
- 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);
396
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
397
- }
398
-
399
- template <typename T, hipTextureReadMode readMode>
400
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex1DLayeredGrad(texture<T, hipTextureType1DLayered, readMode> t, float x, int layer, float dPdx, float dPdy)
401
- {
402
- TEXTURE_PARAMETERS_INIT;
403
- auto tmp = __ockl_image_sample_grad_1Da(i, s, float2(x, layer).data, dPdx, dPdy);
404
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
405
- }
406
-
407
- template <typename T, hipTextureReadMode readMode>
408
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex2DLayeredGrad(texture<T, hipTextureType2DLayered, readMode> t, float x, float y, int layer, float2 dPdx, float2 dPdy)
409
- {
410
- TEXTURE_PARAMETERS_INIT;
411
- 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);
412
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
413
- }
414
-
415
- template <typename T, hipTextureReadMode readMode>
416
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex_ret_t<T, readMode> tex3DGrad(texture<T, hipTextureType3D, readMode> t, float x, float y, float z, float4 dPdx, float4 dPdy)
417
- {
418
- TEXTURE_PARAMETERS_INIT;
419
- 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);
420
- return __hipMapFrom<__hip_tex_ret_t<T, readMode>>(tmp);
421
- }
422
-
423
- template <
424
- typename T,
425
- hipTextureReadMode readMode,
426
- typename Enable = void>
427
- struct __hip_tex2dgather_ret
428
- {
429
- static_assert(std::is_same<Enable, void>::value, "Invalid channel type!");
430
- };
431
-
432
- template <
433
- typename T,
434
- hipTextureReadMode readMode>
435
- using __hip_tex2dgather_ret_t = typename __hip_tex2dgather_ret<T, readMode, bool>::type;
436
-
437
- template <typename T>
438
- struct __hip_tex2dgather_ret<
439
- T,
440
- hipReadModeElementType,
441
- typename std::enable_if<__hip_is_tex_surf_channel_type<T>::value, bool>::type>
442
- {
443
- using type = HIP_vector_type<T, 4>;
444
- };
445
-
446
- template<
447
- typename T,
448
- unsigned int rank>
449
- struct __hip_tex2dgather_ret<
450
- HIP_vector_type<T, rank>,
451
- hipReadModeElementType,
452
- typename std::enable_if<__hip_is_tex_surf_channel_type<HIP_vector_type<T, rank>>::value, bool>::type>
453
- {
454
- using type = HIP_vector_type<T, 4>;
455
- };
456
-
457
- template <typename T>
458
- struct __hip_tex2dgather_ret<
459
- T,
460
- hipReadModeNormalizedFloat,
461
- typename std::enable_if<__hip_is_tex_normalized_channel_type<T>::value, bool>::type>
462
- {
463
- using type = float4;
464
- };
465
-
466
- template <typename T, hipTextureReadMode readMode>
467
- static __forceinline__ __device__ __hip_img_chk__ __hip_tex2dgather_ret_t<T, readMode> tex2Dgather(texture<T, hipTextureType2D, readMode> t, float x, float y, int comp=0)
468
- {
469
- TEXTURE_PARAMETERS_INIT;
470
- switch (comp) {
471
- case 1: {
472
- auto tmp = __ockl_image_gather4g_2D(i, s, float2(x, y).data);
473
- return __hipMapFrom<__hip_tex2dgather_ret_t<T, readMode>>(tmp);
474
- }
475
- case 2: {
476
- auto tmp = __ockl_image_gather4b_2D(i, s, float2(x, y).data);
477
- return __hipMapFrom<__hip_tex2dgather_ret_t<T, readMode>>(tmp);
478
- }
479
- case 3: {
480
- auto tmp = __ockl_image_gather4a_2D(i, s, float2(x, y).data);
481
- return __hipMapFrom<__hip_tex2dgather_ret_t<T, readMode>>(tmp);
482
- }
483
- default: {
484
- auto tmp = __ockl_image_gather4r_2D(i, s, float2(x, y).data);
485
- return __hipMapFrom<__hip_tex2dgather_ret_t<T, readMode>>(tmp);
486
- }
487
- }
488
- return {};
489
- }
490
-
491
- #endif