triton-windows 3.3.0.post19__cp313-cp313-win_amd64.whl → 3.4.0.post20__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 (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
triton/language/extra/cuda/utils.py CHANGED
@@ -2,25 +2,25 @@ from triton.language import core
2
2
 
3
3
 
4
4
  @core.extern
5
- def globaltimer(_builder=None):
5
+ def globaltimer(_semantic=None):
6
6
  return core.inline_asm_elementwise("mov.u64 $0, %globaltimer;", "=l", [], dtype=core.int64, is_pure=False, pack=1,
7
- _builder=_builder)
7
+ _semantic=_semantic)
8
8
 
9
9
 
10
10
  @core.extern
11
- def smid(_builder=None):
11
+ def smid(_semantic=None):
12
12
  return core.inline_asm_elementwise("mov.u32 $0, %smid;", "=r", [], dtype=core.int32, is_pure=True, pack=1,
13
- _builder=_builder)
13
+ _semantic=_semantic)
14
14
 
15
15
 
16
16
  @core.builtin
17
- def num_threads(_builder=None):
18
- return core.constexpr(_builder.options.num_warps * 32)
17
+ def num_threads(_semantic=None):
18
+ return core.constexpr(_semantic.builder.options.num_warps * 32)
19
19
 
20
20
 
21
21
  @core.builtin
22
- def num_warps(_builder=None):
23
- return core.constexpr(_builder.options.num_warps)
22
+ def num_warps(_semantic=None):
23
+ return core.constexpr(_semantic.builder.options.num_warps)
24
24
 
25
25
 
26
26
  # ----- FP8E4M3B15 ------
@@ -31,7 +31,7 @@ def num_warps(_builder=None):
31
31
  # - the exponent bias is 15 instead of 7
32
32
  # - 0xff and 0x7f are mapped to +-1.750 instead of +-nan
33
33
  @core.builtin
34
- def convert_fp8e4b15_to_float16(arg, _builder=None):
34
+ def convert_fp8e4b15_to_float16(arg, _semantic=None):
35
35
  return core.inline_asm_elementwise(
36
36
  "{ \n"
37
37
  ".reg .b32 a<2>, b<2>; \n"
@@ -44,11 +44,11 @@ def convert_fp8e4b15_to_float16(arg, _builder=None):
44
44
  "lop3.b32 $0, b0, 0x80008000, a0, 0xf8; \n"
45
45
  "shl.b32 $1, b1, 7; \n"
46
46
  "} \n", "=r,=r,r", [arg], dtype=core.float16, is_pure=True, pack=4,
47
- _builder=_builder)
47
+ _semantic=_semantic)
48
48
 
49
49
 
50
50
  @core.builtin
51
- def convert_float16_to_fp8e4b15(arg, has_minx2, _builder=None):
51
+ def convert_float16_to_fp8e4b15(arg, has_minx2, _semantic=None):
52
52
  asm = """{
53
53
  .reg .pred p<4>;
54
54
  .reg .b32 a<2>, b<2>;
@@ -80,30 +80,30 @@ def convert_float16_to_fp8e4b15(arg, has_minx2, _builder=None):
80
80
  prmt.b32 $0, b0, b1, 0x7531;
81
81
  }"""
82
82
  return core.inline_asm_elementwise(asm, "=r,r,r", [arg], dtype=core.float8e4b15, is_pure=True, pack=4,
83
- _builder=_builder)
83
+ _semantic=_semantic)
84
84
 
85
85
 
86
86
  @core.builtin
87
- def convert_custom_float8(arg, dst_ty, fp_downcast_rounding, has_minx2, _builder=None):
87
+ def convert_custom_float8(arg, dst_ty, fp_downcast_rounding, has_minx2, _semantic=None):
88
88
  if arg.type.scalar.is_fp8e4b15():
89
- upcast_val = convert_fp8e4b15_to_float16(arg, _builder=_builder)
89
+ upcast_val = convert_fp8e4b15_to_float16(arg, _semantic=_semantic)
90
90
  if dst_ty.scalar.is_fp32():
91
- upcast_val = upcast_val.to(core.float32, _builder=_builder)
91
+ upcast_val = upcast_val.to(core.float32, _semantic=_semantic)
92
92
  return upcast_val
93
93
 
94
94
  assert arg.type.scalar.is_fp16() or arg.type.scalar.is_fp32()
95
95
  downcast_val = arg
96
96
  if arg.type.scalar.is_fp32():
97
- downcast_val = downcast_val.to(core.float16, fp_downcast_rounding="rtz", _builder=_builder)
98
- downcast_val = convert_float16_to_fp8e4b15(downcast_val, has_minx2=has_minx2, _builder=_builder)
97
+ downcast_val = downcast_val.to(core.float16, fp_downcast_rounding="rtz", _semantic=_semantic)
98
+ downcast_val = convert_float16_to_fp8e4b15(downcast_val, has_minx2=has_minx2, _semantic=_semantic)
99
99
  return downcast_val
100
100
 
101
101
 
102
102
  @core.builtin
103
- def convert_custom_float8_sm80(arg, dst_ty, fp_downcast_rounding=None, _builder=None):
104
- return convert_custom_float8(arg, dst_ty, fp_downcast_rounding, has_minx2=True, _builder=_builder)
103
+ def convert_custom_float8_sm80(arg, dst_ty, fp_downcast_rounding=None, _semantic=None):
104
+ return convert_custom_float8(arg, dst_ty, fp_downcast_rounding, has_minx2=True, _semantic=_semantic)
105
105
 
106
106
 
107
107
  @core.builtin
108
- def convert_custom_float8_sm70(arg, dst_ty, fp_downcast_rounding=None, _builder=None):
109
- return convert_custom_float8(arg, dst_ty, fp_downcast_rounding, has_minx2=False, _builder=_builder)
108
+ def convert_custom_float8_sm70(arg, dst_ty, fp_downcast_rounding=None, _semantic=None):
109
+ return convert_custom_float8(arg, dst_ty, fp_downcast_rounding, has_minx2=False, _semantic=_semantic)