triton-windows 3.3.0.post19__cp39-cp39-win_amd64.whl → 3.4.0.post20__cp39-cp39-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 (173) hide show
  1. triton/_C/libtriton.pyd +0 -0
  2. triton/__init__.py +4 -1
  3. triton/_filecheck.py +87 -0
  4. triton/_internal_testing.py +26 -15
  5. triton/_utils.py +110 -21
  6. triton/backends/__init__.py +20 -23
  7. triton/backends/amd/__init__.py +0 -0
  8. triton/backends/amd/compiler.py +112 -78
  9. triton/backends/amd/driver.c +5 -2
  10. triton/backends/amd/driver.py +149 -47
  11. triton/backends/compiler.py +7 -21
  12. triton/backends/nvidia/bin/ptxas.exe +0 -0
  13. triton/backends/nvidia/compiler.py +92 -93
  14. triton/backends/nvidia/driver.c +90 -98
  15. triton/backends/nvidia/driver.py +303 -128
  16. triton/compiler/code_generator.py +212 -111
  17. triton/compiler/compiler.py +110 -25
  18. triton/experimental/__init__.py +0 -0
  19. triton/experimental/gluon/__init__.py +4 -0
  20. triton/experimental/gluon/_compiler.py +0 -0
  21. triton/experimental/gluon/_runtime.py +99 -0
  22. triton/experimental/gluon/language/__init__.py +18 -0
  23. triton/experimental/gluon/language/_core.py +312 -0
  24. triton/experimental/gluon/language/_layouts.py +230 -0
  25. triton/experimental/gluon/language/_math.py +12 -0
  26. triton/experimental/gluon/language/_semantic.py +287 -0
  27. triton/experimental/gluon/language/_standard.py +47 -0
  28. triton/experimental/gluon/language/nvidia/__init__.py +4 -0
  29. triton/experimental/gluon/language/nvidia/blackwell/__init__.py +202 -0
  30. triton/experimental/gluon/language/nvidia/blackwell/tma.py +32 -0
  31. triton/experimental/gluon/language/nvidia/hopper/__init__.py +11 -0
  32. triton/experimental/gluon/language/nvidia/hopper/mbarrier.py +51 -0
  33. triton/experimental/gluon/language/nvidia/hopper/tma.py +96 -0
  34. triton/experimental/gluon/nvidia/__init__.py +4 -0
  35. triton/experimental/gluon/nvidia/blackwell.py +3 -0
  36. triton/experimental/gluon/nvidia/hopper.py +40 -0
  37. triton/knobs.py +481 -0
  38. triton/language/__init__.py +39 -14
  39. triton/language/core.py +794 -537
  40. triton/language/extra/cuda/__init__.py +10 -7
  41. triton/language/extra/cuda/gdc.py +42 -0
  42. triton/language/extra/cuda/libdevice.py +394 -394
  43. triton/language/extra/cuda/utils.py +21 -21
  44. triton/language/extra/hip/libdevice.py +113 -104
  45. triton/language/math.py +65 -66
  46. triton/language/random.py +12 -2
  47. triton/language/semantic.py +1706 -1770
  48. triton/language/standard.py +116 -51
  49. triton/runtime/autotuner.py +117 -59
  50. triton/runtime/build.py +76 -12
  51. triton/runtime/cache.py +18 -47
  52. triton/runtime/driver.py +32 -29
  53. triton/runtime/interpreter.py +72 -35
  54. triton/runtime/jit.py +146 -110
  55. triton/runtime/tcc/lib/python310.def +1610 -0
  56. triton/runtime/tcc/lib/python311.def +1633 -0
  57. triton/runtime/tcc/lib/python312.def +1703 -0
  58. triton/runtime/tcc/lib/python313.def +1651 -0
  59. triton/runtime/tcc/lib/python313t.def +1656 -0
  60. triton/runtime/tcc/lib/python39.def +1644 -0
  61. triton/runtime/tcc/lib/python3t.def +905 -0
  62. triton/testing.py +16 -12
  63. triton/tools/disasm.py +3 -4
  64. triton/tools/tensor_descriptor.py +36 -0
  65. triton/windows_utils.py +14 -6
  66. {triton_windows-3.3.0.post19.dist-info → triton_windows-3.4.0.post20.dist-info}/METADATA +7 -2
  67. triton_windows-3.4.0.post20.dist-info/RECORD +186 -0
  68. {triton_windows-3.3.0.post19.dist-info → triton_windows-3.4.0.post20.dist-info}/WHEEL +1 -1
  69. triton_windows-3.4.0.post20.dist-info/entry_points.txt +3 -0
  70. triton_windows-3.4.0.post20.dist-info/licenses/LICENSE +23 -0
  71. triton_windows-3.4.0.post20.dist-info/top_level.txt +1 -0
  72. triton/backends/amd/include/hip/amd_detail/amd_channel_descriptor.h +0 -358
  73. triton/backends/amd/include/hip/amd_detail/amd_device_functions.h +0 -1010
  74. triton/backends/amd/include/hip/amd_detail/amd_hip_atomic.h +0 -1638
  75. triton/backends/amd/include/hip/amd_detail/amd_hip_bf16.h +0 -1814
  76. triton/backends/amd/include/hip/amd_detail/amd_hip_bfloat16.h +0 -293
  77. triton/backends/amd/include/hip/amd_detail/amd_hip_common.h +0 -32
  78. triton/backends/amd/include/hip/amd_detail/amd_hip_complex.h +0 -174
  79. triton/backends/amd/include/hip/amd_detail/amd_hip_cooperative_groups.h +0 -835
  80. triton/backends/amd/include/hip/amd_detail/amd_hip_fp16.h +0 -1809
  81. triton/backends/amd/include/hip/amd_detail/amd_hip_fp8.h +0 -1391
  82. triton/backends/amd/include/hip/amd_detail/amd_hip_gl_interop.h +0 -108
  83. triton/backends/amd/include/hip/amd_detail/amd_hip_math_constants.h +0 -124
  84. triton/backends/amd/include/hip/amd_detail/amd_hip_runtime.h +0 -405
  85. triton/backends/amd/include/hip/amd_detail/amd_hip_runtime_pt_api.h +0 -196
  86. triton/backends/amd/include/hip/amd_detail/amd_hip_unsafe_atomics.h +0 -565
  87. triton/backends/amd/include/hip/amd_detail/amd_hip_vector_types.h +0 -2226
  88. triton/backends/amd/include/hip/amd_detail/amd_math_functions.h +0 -104
  89. triton/backends/amd/include/hip/amd_detail/amd_surface_functions.h +0 -244
  90. triton/backends/amd/include/hip/amd_detail/amd_warp_functions.h +0 -538
  91. triton/backends/amd/include/hip/amd_detail/amd_warp_sync_functions.h +0 -288
  92. triton/backends/amd/include/hip/amd_detail/concepts.hpp +0 -30
  93. triton/backends/amd/include/hip/amd_detail/device_library_decls.h +0 -133
  94. triton/backends/amd/include/hip/amd_detail/functional_grid_launch.hpp +0 -218
  95. triton/backends/amd/include/hip/amd_detail/grid_launch.h +0 -67
  96. triton/backends/amd/include/hip/amd_detail/grid_launch.hpp +0 -50
  97. triton/backends/amd/include/hip/amd_detail/grid_launch_GGL.hpp +0 -26
  98. triton/backends/amd/include/hip/amd_detail/helpers.hpp +0 -137
  99. triton/backends/amd/include/hip/amd_detail/hip_api_trace.hpp +0 -1446
  100. triton/backends/amd/include/hip/amd_detail/hip_assert.h +0 -101
  101. triton/backends/amd/include/hip/amd_detail/hip_cooperative_groups_helper.h +0 -242
  102. triton/backends/amd/include/hip/amd_detail/hip_fp16_gcc.h +0 -254
  103. triton/backends/amd/include/hip/amd_detail/hip_fp16_math_fwd.h +0 -96
  104. triton/backends/amd/include/hip/amd_detail/hip_ldg.h +0 -100
  105. triton/backends/amd/include/hip/amd_detail/hip_prof_str.h +0 -10570
  106. triton/backends/amd/include/hip/amd_detail/hip_runtime_prof.h +0 -78
  107. triton/backends/amd/include/hip/amd_detail/host_defines.h +0 -184
  108. triton/backends/amd/include/hip/amd_detail/hsa_helpers.hpp +0 -102
  109. triton/backends/amd/include/hip/amd_detail/macro_based_grid_launch.hpp +0 -798
  110. triton/backends/amd/include/hip/amd_detail/math_fwd.h +0 -698
  111. triton/backends/amd/include/hip/amd_detail/ockl_image.h +0 -177
  112. triton/backends/amd/include/hip/amd_detail/program_state.hpp +0 -107
  113. triton/backends/amd/include/hip/amd_detail/texture_fetch_functions.h +0 -491
  114. triton/backends/amd/include/hip/amd_detail/texture_indirect_functions.h +0 -478
  115. triton/backends/amd/include/hip/channel_descriptor.h +0 -39
  116. triton/backends/amd/include/hip/device_functions.h +0 -38
  117. triton/backends/amd/include/hip/driver_types.h +0 -468
  118. triton/backends/amd/include/hip/hip_bf16.h +0 -36
  119. triton/backends/amd/include/hip/hip_bfloat16.h +0 -44
  120. triton/backends/amd/include/hip/hip_common.h +0 -100
  121. triton/backends/amd/include/hip/hip_complex.h +0 -38
  122. triton/backends/amd/include/hip/hip_cooperative_groups.h +0 -46
  123. triton/backends/amd/include/hip/hip_deprecated.h +0 -95
  124. triton/backends/amd/include/hip/hip_ext.h +0 -161
  125. triton/backends/amd/include/hip/hip_fp16.h +0 -36
  126. triton/backends/amd/include/hip/hip_fp8.h +0 -33
  127. triton/backends/amd/include/hip/hip_gl_interop.h +0 -32
  128. triton/backends/amd/include/hip/hip_hcc.h +0 -24
  129. triton/backends/amd/include/hip/hip_math_constants.h +0 -36
  130. triton/backends/amd/include/hip/hip_profile.h +0 -27
  131. triton/backends/amd/include/hip/hip_runtime.h +0 -75
  132. triton/backends/amd/include/hip/hip_runtime_api.h +0 -9261
  133. triton/backends/amd/include/hip/hip_texture_types.h +0 -29
  134. triton/backends/amd/include/hip/hip_vector_types.h +0 -41
  135. triton/backends/amd/include/hip/hip_version.h +0 -17
  136. triton/backends/amd/include/hip/hiprtc.h +0 -421
  137. triton/backends/amd/include/hip/library_types.h +0 -78
  138. triton/backends/amd/include/hip/math_functions.h +0 -42
  139. triton/backends/amd/include/hip/surface_types.h +0 -63
  140. triton/backends/amd/include/hip/texture_types.h +0 -194
  141. triton/backends/amd/include/hsa/Brig.h +0 -1131
  142. triton/backends/amd/include/hsa/amd_hsa_common.h +0 -91
  143. triton/backends/amd/include/hsa/amd_hsa_elf.h +0 -462
  144. triton/backends/amd/include/hsa/amd_hsa_kernel_code.h +0 -269
  145. triton/backends/amd/include/hsa/amd_hsa_queue.h +0 -109
  146. triton/backends/amd/include/hsa/amd_hsa_signal.h +0 -80
  147. triton/backends/amd/include/hsa/hsa.h +0 -5738
  148. triton/backends/amd/include/hsa/hsa_amd_tool.h +0 -91
  149. triton/backends/amd/include/hsa/hsa_api_trace.h +0 -579
  150. triton/backends/amd/include/hsa/hsa_api_trace_version.h +0 -68
  151. triton/backends/amd/include/hsa/hsa_ext_amd.h +0 -3146
  152. triton/backends/amd/include/hsa/hsa_ext_finalize.h +0 -531
  153. triton/backends/amd/include/hsa/hsa_ext_image.h +0 -1454
  154. triton/backends/amd/include/hsa/hsa_ven_amd_aqlprofile.h +0 -488
  155. triton/backends/amd/include/hsa/hsa_ven_amd_loader.h +0 -667
  156. triton/backends/amd/include/hsa/hsa_ven_amd_pc_sampling.h +0 -416
  157. triton/backends/amd/include/roctracer/ext/prof_protocol.h +0 -107
  158. triton/backends/amd/include/roctracer/hip_ostream_ops.h +0 -4515
  159. triton/backends/amd/include/roctracer/hsa_ostream_ops.h +0 -1727
  160. triton/backends/amd/include/roctracer/hsa_prof_str.h +0 -3059
  161. triton/backends/amd/include/roctracer/roctracer.h +0 -779
  162. triton/backends/amd/include/roctracer/roctracer_ext.h +0 -81
  163. triton/backends/amd/include/roctracer/roctracer_hcc.h +0 -24
  164. triton/backends/amd/include/roctracer/roctracer_hip.h +0 -37
  165. triton/backends/amd/include/roctracer/roctracer_hsa.h +0 -112
  166. triton/backends/amd/include/roctracer/roctracer_plugin.h +0 -137
  167. triton/backends/amd/include/roctracer/roctracer_roctx.h +0 -67
  168. triton/backends/amd/include/roctracer/roctx.h +0 -229
  169. triton/language/_utils.py +0 -21
  170. triton/language/extra/cuda/_experimental_tma.py +0 -106
  171. triton/tools/experimental_descriptor.py +0 -32
  172. triton_windows-3.3.0.post19.dist-info/RECORD +0 -253
  173. triton_windows-3.3.0.post19.dist-info/top_level.txt +0 -14
@@ -1,196 +0,0 @@
1
- /*
2
- Copyright (c) 2022 - Present 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
- #ifndef HIP_INCLUDE_HIP_HIP_RUNTIME_PT_API_H
26
- #define HIP_INCLUDE_HIP_HIP_RUNTIME_PT_API_H
27
-
28
- #if defined(__HIP_PLATFORM_AMD__) && !defined(__HIP_PLATFORM_NVIDIA__)
29
-
30
- /// hipStreamPerThread implementation
31
- #if defined(HIP_API_PER_THREAD_DEFAULT_STREAM)
32
- #define __HIP_STREAM_PER_THREAD
33
- #define __HIP_API_SPT(api) api ## _spt
34
- #else
35
- #define __HIP_API_SPT(api) api
36
- #endif
37
-
38
- #if defined(__HIP_STREAM_PER_THREAD)
39
- // Memory APIs
40
- #define hipMemcpy __HIP_API_SPT(hipMemcpy)
41
- #define hipMemcpyToSymbol __HIP_API_SPT(hipMemcpyToSymbol)
42
- #define hipMemcpyFromSymbol __HIP_API_SPT(hipMemcpyFromSymbol)
43
- #define hipMemcpy2D __HIP_API_SPT(hipMemcpy2D)
44
- #define hipMemcpy2DFromArray __HIP_API_SPT(hipMemcpy2DFromArray)
45
- #define hipMemcpy3D __HIP_API_SPT(hipMemcpy3D)
46
- #define hipMemset __HIP_API_SPT(hipMemset)
47
- #define hipMemset2D __HIP_API_SPT(hipMemset2D)
48
- #define hipMemset3D __HIP_API_SPT(hipMemset3D)
49
- #define hipMemcpyAsync __HIP_API_SPT(hipMemcpyAsync)
50
- #define hipMemset3DAsync __HIP_API_SPT(hipMemset3DAsync)
51
- #define hipMemset2DAsync __HIP_API_SPT(hipMemset2DAsync)
52
- #define hipMemsetAsync __HIP_API_SPT(hipMemsetAsync)
53
- #define hipMemcpy3DAsync __HIP_API_SPT(hipMemcpy3DAsync)
54
- #define hipMemcpy2DAsync __HIP_API_SPT(hipMemcpy2DAsync)
55
- #define hipMemcpyFromSymbolAsync __HIP_API_SPT(hipMemcpyFromSymbolAsync)
56
- #define hipMemcpyToSymbolAsync __HIP_API_SPT(hipMemcpyToSymbolAsync)
57
- #define hipMemcpyFromArray __HIP_API_SPT(hipMemcpyFromArray)
58
- #define hipMemcpy2DToArray __HIP_API_SPT(hipMemcpy2DToArray)
59
- #define hipMemcpy2DFromArrayAsync __HIP_API_SPT(hipMemcpy2DFromArrayAsync)
60
- #define hipMemcpy2DToArrayAsync __HIP_API_SPT(hipMemcpy2DToArrayAsync)
61
-
62
- // Stream APIs
63
- #define hipStreamSynchronize __HIP_API_SPT(hipStreamSynchronize)
64
- #define hipStreamQuery __HIP_API_SPT(hipStreamQuery)
65
- #define hipStreamGetFlags __HIP_API_SPT(hipStreamGetFlags)
66
- #define hipStreamGetPriority __HIP_API_SPT(hipStreamGetPriority)
67
- #define hipStreamWaitEvent __HIP_API_SPT(hipStreamWaitEvent)
68
- #define hipStreamAddCallback __HIP_API_SPT(hipStreamAddCallback)
69
- #define hipLaunchHostFunc __HIP_API_SPT(hipLaunchHostFunc)
70
-
71
- // Event APIs
72
- #define hipEventRecord __HIP_API_SPT(hipEventRecord)
73
-
74
- // Launch APIs
75
- #define hipLaunchKernel __HIP_API_SPT(hipLaunchKernel)
76
- #define hipLaunchCooperativeKernel __HIP_API_SPT(hipLaunchCooperativeKernel)
77
-
78
- // Graph APIs
79
- #define hipGraphLaunch __HIP_API_SPT(hipGraphLaunch)
80
- #define hipStreamBeginCapture __HIP_API_SPT(hipStreamBeginCapture)
81
- #define hipStreamEndCapture __HIP_API_SPT(hipStreamEndCapture)
82
- #define hipStreamIsCapturing __HIP_API_SPT(hipStreamIsCapturing)
83
- #define hipStreamGetCaptureInfo __HIP_API_SPT(hipStreamGetCaptureInfo)
84
- #define hipStreamGetCaptureInfo_v2 __HIP_API_SPT(hipStreamGetCaptureInfo_v2)
85
- #endif
86
-
87
- #ifdef __cplusplus
88
- extern "C" {
89
- #endif
90
-
91
- hipError_t hipMemcpy_spt(void* dst, const void* src, size_t sizeBytes, hipMemcpyKind kind);
92
-
93
- hipError_t hipMemcpyToSymbol_spt(const void* symbol, const void* src, size_t sizeBytes,
94
- size_t offset __dparm(0),
95
- hipMemcpyKind kind __dparm(hipMemcpyHostToDevice));
96
-
97
- hipError_t hipMemcpyFromSymbol_spt(void* dst, const void* symbol,size_t sizeBytes,
98
- size_t offset __dparm(0),
99
- hipMemcpyKind kind __dparm(hipMemcpyDeviceToHost));
100
-
101
- hipError_t hipMemcpy2D_spt(void* dst, size_t dpitch, const void* src, size_t spitch, size_t width,
102
- size_t height, hipMemcpyKind kind);
103
-
104
- hipError_t hipMemcpy2DFromArray_spt( void* dst, size_t dpitch, hipArray_const_t src, size_t wOffset,
105
- size_t hOffset, size_t width, size_t height, hipMemcpyKind kind);
106
-
107
- hipError_t hipMemcpy3D_spt(const struct hipMemcpy3DParms* p);
108
-
109
- hipError_t hipMemset_spt(void* dst, int value, size_t sizeBytes);
110
-
111
- hipError_t hipMemsetAsync_spt(void* dst, int value, size_t sizeBytes, hipStream_t stream);
112
-
113
- hipError_t hipMemset2D_spt(void* dst, size_t pitch, int value, size_t width, size_t height);
114
-
115
- hipError_t hipMemset2DAsync_spt(void* dst, size_t pitch, int value,
116
- size_t width, size_t height, hipStream_t stream);
117
-
118
- hipError_t hipMemset3DAsync_spt(hipPitchedPtr pitchedDevPtr, int value, hipExtent extent, hipStream_t stream);
119
-
120
- hipError_t hipMemset3D_spt(hipPitchedPtr pitchedDevPtr, int value, hipExtent extent );
121
-
122
- hipError_t hipMemcpyAsync_spt(void* dst, const void* src, size_t sizeBytes, hipMemcpyKind kind,
123
- hipStream_t stream);
124
-
125
- hipError_t hipMemcpy3DAsync_spt(const hipMemcpy3DParms* p, hipStream_t stream);
126
-
127
- hipError_t hipMemcpy2DAsync_spt(void* dst, size_t dpitch, const void* src, size_t spitch, size_t width,
128
- size_t height, hipMemcpyKind kind, hipStream_t stream);
129
-
130
- hipError_t hipMemcpyFromSymbolAsync_spt(void* dst, const void* symbol, size_t sizeBytes,
131
- size_t offset, hipMemcpyKind kind, hipStream_t stream);
132
-
133
- hipError_t hipMemcpyToSymbolAsync_spt(const void* symbol, const void* src, size_t sizeBytes,
134
- size_t offset, hipMemcpyKind kind, hipStream_t stream);
135
-
136
- hipError_t hipMemcpyFromArray_spt(void* dst, hipArray_const_t src, size_t wOffsetSrc, size_t hOffset,
137
- size_t count, hipMemcpyKind kind);
138
-
139
- hipError_t hipMemcpy2DToArray_spt(hipArray_t dst, size_t wOffset, size_t hOffset, const void* src,
140
- size_t spitch, size_t width, size_t height, hipMemcpyKind kind);
141
-
142
- hipError_t hipMemcpy2DFromArrayAsync_spt(void* dst, size_t dpitch, hipArray_const_t src,
143
- size_t wOffsetSrc, size_t hOffsetSrc, size_t width, size_t height,
144
- hipMemcpyKind kind, hipStream_t stream);
145
-
146
- hipError_t hipMemcpy2DToArrayAsync_spt(hipArray_t dst, size_t wOffset, size_t hOffset, const void* src,
147
- size_t spitch, size_t width, size_t height, hipMemcpyKind kind,
148
- hipStream_t stream);
149
-
150
- hipError_t hipStreamQuery_spt(hipStream_t stream);
151
-
152
- hipError_t hipStreamSynchronize_spt(hipStream_t stream);
153
-
154
- hipError_t hipStreamGetPriority_spt(hipStream_t stream, int* priority);
155
-
156
- hipError_t hipStreamWaitEvent_spt(hipStream_t stream, hipEvent_t event, unsigned int flags __dparm(0));
157
-
158
- hipError_t hipStreamGetFlags_spt(hipStream_t stream, unsigned int* flags);
159
-
160
- hipError_t hipStreamAddCallback_spt(hipStream_t stream, hipStreamCallback_t callback, void* userData,
161
- unsigned int flags);
162
- #ifdef __cplusplus
163
- hipError_t hipEventRecord_spt(hipEvent_t event, hipStream_t stream = NULL);
164
- #else
165
- hipError_t hipEventRecord_spt(hipEvent_t event, hipStream_t stream);
166
- #endif
167
-
168
- hipError_t hipLaunchCooperativeKernel_spt(const void* f,
169
- dim3 gridDim, dim3 blockDim,
170
- void **kernelParams, uint32_t sharedMemBytes, hipStream_t hStream);
171
-
172
- hipError_t hipLaunchKernel_spt(const void* function_address,
173
- dim3 numBlocks,
174
- dim3 dimBlocks,
175
- void** args,
176
- size_t sharedMemBytes, hipStream_t stream);
177
-
178
- hipError_t hipGraphLaunch_spt(hipGraphExec_t graphExec, hipStream_t stream);
179
- hipError_t hipStreamBeginCapture_spt(hipStream_t stream, hipStreamCaptureMode mode);
180
- hipError_t hipStreamEndCapture_spt(hipStream_t stream, hipGraph_t* pGraph);
181
- hipError_t hipStreamIsCapturing_spt(hipStream_t stream, hipStreamCaptureStatus* pCaptureStatus);
182
- hipError_t hipStreamGetCaptureInfo_spt(hipStream_t stream, hipStreamCaptureStatus* pCaptureStatus,
183
- unsigned long long* pId);
184
- hipError_t hipStreamGetCaptureInfo_v2_spt(hipStream_t stream, hipStreamCaptureStatus* captureStatus_out,
185
- unsigned long long* id_out, hipGraph_t* graph_out,
186
- const hipGraphNode_t** dependencies_out,
187
- size_t* numDependencies_out);
188
- hipError_t hipLaunchHostFunc_spt(hipStream_t stream, hipHostFn_t fn, void* userData);
189
-
190
-
191
- #ifdef __cplusplus
192
- }
193
- #endif // extern "C"
194
-
195
- #endif //defined(__HIP_PLATFORM_AMD__) && !defined(__HIP_PLATFORM_NVIDIA__)
196
- #endif //HIP_INCLUDE_HIP_HIP_RUNTIME_PT_API_H