triton-windows 3.3.1.post21__cp313-cp313-win_amd64.whl → 3.4.0.post21__cp313-cp313-win_amd64.whl

triton/windows_utils.py

@@ -54,14 +54,11 @@ def max_version(
 
 
 def check_msvc(msvc_base_path: Path, version: str) -> bool:
-    return all(
-        x.exists()
-        for x in [
-            msvc_base_path / version / "bin" / "Hostx64" / "x64" / "cl.exe",
-            msvc_base_path / version / "include" / "vcruntime.h",
-            msvc_base_path / version / "lib" / "x64" / "vcruntime.lib",
-        ]
-    )
+    return all(x.exists() for x in [
+        msvc_base_path / version / "bin" / "Hostx64" / "x64" / "cl.exe",
+        msvc_base_path / version / "include" / "vcruntime.h",
+        msvc_base_path / version / "lib" / "x64" / "vcruntime.lib",
+    ])
 
 
 def find_msvc_env() -> tuple[Optional[Path], Optional[str]]:
@@ -72,20 +69,16 @@ def find_msvc_env() -> tuple[Optional[Path], Optional[str]]:
 
     version = os.getenv("VCToolsVersion")
     if not check_msvc(msvc_base_path, version):
-        warnings.warn(
-            f"Environment variables VCINSTALLDIR = {os.getenv('VCINSTALLDIR')}, "
-            f"VCToolsVersion = {os.getenv('VCToolsVersion')} are set, "
-            "but this MSVC installation is incomplete."
-        )
+        warnings.warn(f"Environment variables VCINSTALLDIR = {os.getenv('VCINSTALLDIR')}, "
+                      f"VCToolsVersion = {os.getenv('VCToolsVersion')} are set, "
+                      "but this MSVC installation is incomplete.")
         return None, None
 
     return msvc_base_path, version
 
 
 def find_msvc_vswhere() -> tuple[Optional[Path], Optional[str]]:
-    vswhere_path = find_in_program_files(
-        r"Microsoft Visual Studio\Installer\vswhere.exe"
-    )
+    vswhere_path = find_in_program_files(r"Microsoft Visual Studio\Installer\vswhere.exe")
     if vswhere_path is None:
         return None, None
 
@@ -111,9 +104,7 @@ def find_msvc_vswhere() -> tuple[Optional[Path], Optional[str]]:
     if not msvc_base_path.exists():
         return None, None
 
-    version = max_version(
-        os.listdir(msvc_base_path), check=partial(check_msvc, msvc_base_path)
-    )
+    version = max_version(os.listdir(msvc_base_path), check=partial(check_msvc, msvc_base_path))
     if version is None:
         return None, None
 
@@ -132,9 +123,7 @@ def find_msvc_envpath() -> tuple[Optional[Path], Optional[str]]:
         if not msvc_base_path.exists():
             continue
 
-        version = max_version(
-            os.listdir(msvc_base_path), check=partial(check_msvc, msvc_base_path)
-        )
+        version = max_version(os.listdir(msvc_base_path), check=partial(check_msvc, msvc_base_path))
         if version is None:
             continue
 
@@ -153,9 +142,7 @@ def find_msvc_hardcoded() -> tuple[Optional[Path], Optional[str]]:
     paths = sorted(paths)[::-1]
     for msvc_base_path in paths:
         msvc_base_path = Path(msvc_base_path)
-        version = max_version(
-            os.listdir(msvc_base_path), check=partial(check_msvc, msvc_base_path)
-        )
+        version = max_version(os.listdir(msvc_base_path), check=partial(check_msvc, msvc_base_path))
         if version is None:
             continue
         return msvc_base_path, version
@@ -188,13 +175,10 @@ def find_msvc(env_only: bool) -> tuple[Optional[str], list[str], list[str]]:
 
 
 def check_winsdk(winsdk_base_path: Path, version: str) -> bool:
-    return all(
-        x.exists()
-        for x in [
-            winsdk_base_path / "Include" / version / "ucrt" / "stdlib.h",
-            winsdk_base_path / "Lib" / version / "ucrt" / "x64" / "ucrt.lib",
-        ]
-    )
+    return all(x.exists() for x in [
+        winsdk_base_path / "Include" / version / "ucrt" / "stdlib.h",
+        winsdk_base_path / "Lib" / version / "ucrt" / "x64" / "ucrt.lib",
+    ])
 
 
 def find_winsdk_env() -> tuple[Optional[Path], Optional[str]]:
@@ -207,18 +191,14 @@ def find_winsdk_env() -> tuple[Optional[Path], Optional[str]]:
     if version is None:
         version = os.getenv("WindowsSDKVer")
     if version is None:
-        warnings.warn(
-            f"Environment variable WindowsSdkDir = {winsdk_base_path}, "
-            "but WindowsSDKVersion (or WindowsSDKVer) is not set."
-        )
+        warnings.warn(f"Environment variable WindowsSdkDir = {winsdk_base_path}, "
+                      "but WindowsSDKVersion (or WindowsSDKVer) is not set.")
         return None, None
     version = version.rstrip("\\")
     if not check_winsdk(winsdk_base_path, version):
-        warnings.warn(
-            f"Environment variables WindowsSdkDir = {winsdk_base_path}, "
-            f"WindowsSDKVersion (or WindowsSDKVer) = {version} are set, "
-            "but this Windows SDK installation is incomplete."
-        )
+        warnings.warn(f"Environment variables WindowsSdkDir = {winsdk_base_path}, "
+                      f"WindowsSDKVersion (or WindowsSDKVer) = {version} are set, "
+                      "but this Windows SDK installation is incomplete.")
         return None, None
 
     return winsdk_base_path, version
@@ -227,9 +207,7 @@ def find_winsdk_env() -> tuple[Optional[Path], Optional[str]]:
 def find_winsdk_registry() -> tuple[Optional[Path], Optional[str]]:
     try:
         reg = winreg.ConnectRegistry(None, winreg.HKEY_LOCAL_MACHINE)
-        key = winreg.OpenKeyEx(
-            reg, r"SOFTWARE\WOW6432Node\Microsoft\Microsoft SDKs\Windows\v10.0"
-        )
+        key = winreg.OpenKeyEx(reg, r"SOFTWARE\WOW6432Node\Microsoft\Microsoft SDKs\Windows\v10.0")
         folder = winreg.QueryValueEx(key, "InstallationFolder")[0]
         winreg.CloseKey(key)
     except OSError:
@@ -295,10 +273,8 @@ def find_winsdk(env_only: bool) -> tuple[list[str], list[str]]:
     return [], []
 
 
-@functools.cache
-def find_msvc_winsdk(
-    env_only: bool = False,
-) -> tuple[Optional[str], list[str], list[str]]:
+@functools.lru_cache
+def find_msvc_winsdk(env_only: bool = False, ) -> tuple[Optional[str], list[str], list[str]]:
     msvc_bin_path, msvc_inc_dirs, msvc_lib_dirs = find_msvc(env_only)
     winsdk_inc_dirs, winsdk_lib_dirs = find_winsdk(env_only)
     return (
@@ -308,15 +284,18 @@ def find_msvc_winsdk(
     )
 
 
-@functools.cache
+@functools.lru_cache
 def find_python() -> list[str]:
+    version = sysconfig.get_python_version().replace(".", "")
+    if sysconfig.get_config_var("Py_GIL_DISABLED"):
+        version += "t"
     for python_base_path in [
-        sys.exec_prefix,
-        sys.base_exec_prefix,
-        os.path.dirname(sys.executable),
+            sys.exec_prefix,
+            sys.base_exec_prefix,
+            os.path.dirname(sys.executable),
     ]:
         python_lib_dir = Path(python_base_path) / "libs"
-        if (python_lib_dir / "python3.lib").exists():
+        if (python_lib_dir / f"python{version}.lib").exists():
             return [str(python_lib_dir)]
 
     warnings.warn("Failed to find Python libs.")
@@ -325,14 +304,11 @@ def find_python() -> list[str]:
 
 def check_and_find_cuda(base_path: Path) -> tuple[Optional[str], list[str], list[str]]:
     # pip
-    if all(
-        x.exists()
-        for x in [
+    if all(x.exists() for x in [
             base_path / "cuda_nvcc" / "bin" / "ptxas.exe",
             base_path / "cuda_runtime" / "include" / "cuda.h",
             base_path / "cuda_runtime" / "lib" / "x64" / "cuda.lib",
-        ]
-    ):
+    ]):
         return (
             str(base_path / "cuda_nvcc" / "bin"),
             [str(base_path / "cuda_runtime" / "include")],
@@ -340,14 +316,11 @@ def check_and_find_cuda(base_path: Path) -> tuple[Optional[str], list[str], list
         )
 
     # conda
-    if all(
-        x.exists()
-        for x in [
+    if all(x.exists() for x in [
             base_path / "bin" / "ptxas.exe",
             base_path / "include" / "cuda.h",
             base_path / "lib" / "cuda.lib",
-        ]
-    ):
+    ]):
         return (
             str(base_path / "bin"),
             [str(base_path / "include")],
@@ -355,14 +328,11 @@ def check_and_find_cuda(base_path: Path) -> tuple[Optional[str], list[str], list
         )
 
     # bundled or system-wide
-    if all(
-        x.exists()
-        for x in [
+    if all(x.exists() for x in [
             base_path / "bin" / "ptxas.exe",
             base_path / "include" / "cuda.h",
             base_path / "lib" / "x64" / "cuda.lib",
-        ]
-    ):
+    ]):
         return (
             str(base_path / "bin"),
             [str(base_path / "include")],
@@ -379,9 +349,7 @@ def find_cuda_env() -> tuple[Optional[str], list[str], list[str]]:
             continue
 
         cuda_base_path = Path(cuda_base_path)
-        cuda_bin_path, cuda_inc_dirs, cuda_lib_dirs = check_and_find_cuda(
-            cuda_base_path
-        )
+        cuda_bin_path, cuda_inc_dirs, cuda_lib_dirs = check_and_find_cuda(cuda_base_path)
         if cuda_bin_path:
             return cuda_bin_path, cuda_inc_dirs, cuda_lib_dirs
 
@@ -389,9 +357,7 @@ def find_cuda_env() -> tuple[Optional[str], list[str], list[str]]:
 
 
 def find_cuda_bundled() -> tuple[Optional[str], list[str], list[str]]:
-    cuda_base_path = (
-        Path(sysconfig.get_paths()["platlib"]) / "triton" / "backends" / "nvidia"
-    )
+    cuda_base_path = (Path(sysconfig.get_paths()["platlib"]) / "triton" / "backends" / "nvidia")
     return check_and_find_cuda(cuda_base_path)
 
 
@@ -415,23 +381,21 @@ def find_cuda_hardcoded() -> tuple[Optional[str], list[str], list[str]]:
     paths = sorted(paths)[::-1]
     for cuda_base_path in paths:
         cuda_base_path = Path(cuda_base_path)
-        cuda_bin_path, cuda_inc_dirs, cuda_lib_dirs = check_and_find_cuda(
-            cuda_base_path
-        )
+        cuda_bin_path, cuda_inc_dirs, cuda_lib_dirs = check_and_find_cuda(cuda_base_path)
         if cuda_bin_path:
             return cuda_bin_path, cuda_inc_dirs, cuda_lib_dirs
 
     return None, [], []
 
 
-@functools.cache
+@functools.lru_cache
 def find_cuda() -> tuple[Optional[str], list[str], list[str]]:
     for f in [
-        find_cuda_env,
-        find_cuda_bundled,
-        find_cuda_pip,
-        find_cuda_conda,
-        find_cuda_hardcoded,
+            find_cuda_env,
+            find_cuda_bundled,
+            find_cuda_pip,
+            find_cuda_conda,
+            find_cuda_hardcoded,
     ]:
         cuda_bin_path, cuda_inc_dirs, cuda_lib_dirs = f()
         if cuda_bin_path:

{triton_windows-3.3.1.post21.dist-info → triton_windows-3.4.0.post21.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: triton-windows
-Version: 3.3.1.post21
+Version: 3.4.0.post21
 Summary: A language and compiler for custom Deep Learning operations
 Home-page: https://github.com/woct0rdho/triton-windows
 Author: Philippe Tillet, Dian Wu
@@ -15,9 +15,12 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
+Requires-Python: >=3.9,<3.14
+License-File: LICENSE
 Requires-Dist: setuptools>=40.8.0
+Requires-Dist: importlib-metadata; python_version < "3.10"
 Provides-Extra: build
-Requires-Dist: cmake>=3.20; extra == "build"
+Requires-Dist: cmake<4.0,>=3.20; extra == "build"
 Requires-Dist: lit; extra == "build"
 Provides-Extra: tests
 Requires-Dist: autopep8; extra == "tests"
@@ -37,6 +40,8 @@ Dynamic: author-email
 Dynamic: classifier
 Dynamic: home-page
 Dynamic: keywords
+Dynamic: license-file
 Dynamic: provides-extra
 Dynamic: requires-dist
+Dynamic: requires-python
 Dynamic: summary

{triton_windows-3.3.1.post21.dist-info → triton_windows-3.4.0.post21.dist-info}/RECORD

@@ -1,56 +1,77 @@
-triton/__init__.py,sha256=3SQQyt-oywQBYQfubass3Nuj08o8RqPqmli6IPYthww,1392
-triton/_internal_testing.py,sha256=OBY28huiEWItqGgiukgZzHLLaSbS8yj9kdhn_u562Yg,5904
-triton/_utils.py,sha256=5RiCLwW14w0Q3mdZ-9yz-VO5KiSexNj9xeDt4gaNsvE,1014
+triton/__init__.py,sha256=CEQdOZ6zLHuEVGoJ7kl2YwcfBBIPBkOQzzDjHr6ibOM,1464
+triton/_filecheck.py,sha256=iWl8uL4LJeV4En4h4mzUbDnrmXB4jXeEgy4_uqRURH8,2845
+triton/_internal_testing.py,sha256=MpXYuQlvJUYtFAmFcfzlU7dyIDEQB_XirJ96IRujMTA,6326
+triton/_utils.py,sha256=XTYb3qDDaVmbhmXbm6ChMTYTW9jeE538jZwJE_eliQg,3539
 triton/errors.py,sha256=8WfnuRKLG578mgY6cBA3ECruVMf9ULEKFNgRcJ6IhWM,89
-triton/testing.py,sha256=ivFf1Fq9frmfVahaVUp0bgJxmvVZNACZfj3Sai6zfAs,20048
-triton/windows_utils.py,sha256=KOxhtcaTpNbs-UNtZvBJ4lCahJeMgbXUzWdv2wW3z98,13036
-triton/_C/libtriton.pyd,sha256=-9nmO4enVytCA6tMWQGbKo3UbdnkjJ0xZJhJXClm8-o,87410688
-triton/backends/__init__.py,sha256=opAo_vgEMt3tLO_bYFrYGksnIu0qohbmyuu_s3-rNAs,1595
-triton/backends/compiler.py,sha256=ymaG0kpveAuESbQ9QZ0RyXjr0Aq4el_G5XGYogJ2gNA,3588
+triton/knobs.py,sha256=VOdRM_J0TejBYP2H7QZZzLwcRJq-Eppm9hBxDt1pCgA,14916
+triton/testing.py,sha256=vbEQRNrOnnzRQvVVSaiZrUo8AC0XPV40GJxfvkKYLh0,20276
+triton/windows_utils.py,sha256=JMi6mjOApzh2-cw3Wl_nl6ji7JkwexYI7xgo2Et3ihU,12903
+triton/_C/libtriton.pyd,sha256=_TR8zPM-fGXdMumvlF2VH2hQHqeJ79qanE6SqlVgjN4,89965056
+triton/backends/__init__.py,sha256=X7290kf96Fk9QnfLScsX4UDG3zPyH_-31E4A7pVOijM,1612
+triton/backends/compiler.py,sha256=MY2_cQG26p68z8VwRv2Nlj_h2DfEhwBbN-30caMgep0,2840
 triton/backends/driver.py,sha256=AN60upJlPgia0JwvZ8vIVgLITNPuI0fdz8zMIIHPpF4,1450
-triton/backends/amd/compiler.py,sha256=27jurEV7tH5J6BHtOXrdPJyCMYIPHo5G4Op_O66gv4E,19135
-triton/backends/amd/driver.c,sha256=obiiiPndny5NyhUcJ8iyrVHrXU1ruLpLGd_LgaKQEbU,8459
-triton/backends/amd/driver.py,sha256=y7w0jkoQ-jcpWRNDc9i6ZO-gQUOV8detjmAonqBr-zo,20473
+triton/backends/amd/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+triton/backends/amd/compiler.py,sha256=yaF0MqfZ81-fL1Jb5aDxGGGXs2j-pVOVpneWQT6DOUs,19966
+triton/backends/amd/driver.c,sha256=hu5_QLMJVmeyR5zYDWfDAklZckISaAFM7kKOg9MpuWE,8612
+triton/backends/amd/driver.py,sha256=9UU3u5gdqjGl7NYdsOyFk63MjgWW0Vnx4-jJ35ROscs,23718
 triton/backends/amd/lib/asanrtl.bc,sha256=1xv2RlU3WvbdsghHlmhwiHewGM2B5dKts5bERM6S89o,24508
 triton/backends/amd/lib/ockl.bc,sha256=wQKCzkKukIHbu0lyjKUYlhndc7S27xto6L54J0Bn-C0,246124
 triton/backends/amd/lib/ocml.bc,sha256=UPNTXW0gCXUNB-c6orSYwb-mz9_mjUc7zny_vfFza44,205964
 triton/backends/nvidia/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-triton/backends/nvidia/compiler.py,sha256=IKcoPBln5j40P3heHjpPVczZ4XL2PpPkRGiAHTtoGM8,19922
-triton/backends/nvidia/driver.c,sha256=HqtiJDPRxVav9pHl_swBja28RasjHZ7oFOpWKMT627c,18912
-triton/backends/nvidia/driver.py,sha256=9BOCsdHXFgGcKiUMicPu988ay0Ixo68Sa5SAyYtr-Eo,21110
-triton/backends/nvidia/bin/ptxas.exe,sha256=iCva9hIYg-Q2NybchwaQJFkwDzNiliFOTDdZrHPLb6A,24732160
+triton/backends/nvidia/compiler.py,sha256=7Yf79DdocAgYaWLR-MSTAv_QAJjVAZkhbr1-ovsnfnQ,19452
+triton/backends/nvidia/driver.c,sha256=rH8RDtMMv_UHr7qiLnrSdNg3xojOQe_fF1zW67LFjaE,17882
+triton/backends/nvidia/driver.py,sha256=22YT4HhTPI1aJ-mjw31Lvw1mcffj17pGJry3m9W6jio,26391
+triton/backends/nvidia/bin/ptxas.exe,sha256=f28E0l5aerLAfBXk7yagfOwIEE6_6_NkMx-vqPPEQ9Y,24753152
 triton/backends/nvidia/include/cuda.h,sha256=Fn44OjeRImxegJ39apYUspseEfTWNGwpqSGUOnHj5WY,1183268
 triton/backends/nvidia/lib/libdevice.10.bc,sha256=XC-uN8huaMOjhgWpX1EtfRLV89uYYxC-R_VzBKpype4,473728
 triton/backends/nvidia/lib/x64/cuda.lib,sha256=I5DZfR8aQ9wodYo3trskSbJpJd9lHvZXsnEZ3NV30LQ,160840
 triton/compiler/__init__.py,sha256=0NEunzjGCNEVOhYZLDI4pDi_zAaYAgTXNm8U5uxbdL0,242
-triton/compiler/code_generator.py,sha256=Nkp9HdXymGenQROPg8W0UyWeJApdUfNk8k9xls7b9EI,62153
-triton/compiler/compiler.py,sha256=LiNMHZOnyulJNBtguFZ_glkUEo7VmOHYRsi8GanWWtg,18470
+triton/compiler/code_generator.py,sha256=Rj5tFwqfjMR1-cr2CxOr3nP3Ez2V4EE3ENKQAp66uXc,66887
+triton/compiler/compiler.py,sha256=Qm-71vUVkafzjYpU3ttBCTzWheIs08Z7zfBusHODEJY,21556
 triton/compiler/errors.py,sha256=I9Y15pDWcL9heY4SWWdLeMDtW6Iiq2pFXzKfJ6dY_C0,1732
 triton/compiler/make_launcher.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-triton/language/__init__.py,sha256=8gybBY5MBcoXi3K28AVXY-oSaTWxtNIyNbp6fzh-Pn0,5366
-triton/language/_utils.py,sha256=bkp98MH2y3mfSI7h1u_T33VPwYqsbnIJkjuwIsNsfE4,646
-triton/language/core.py,sha256=-YP51tb5h69SApfI_jGFYBro99YLrKuICskya9_uH2c,107314
-triton/language/math.py,sha256=thYBX3JOVTNI9bAZ9zJN7BMPe7r5hNQzAPppx8L92i8,7442
-triton/language/random.py,sha256=VP5yLL43mlVpWrlHjnPTRDd89enW4FWv980Dv2__Gyo,6906
-triton/language/semantic.py,sha256=EChVVc1RXiHKVHFPyOucFzzKawkrXpUaF65e7ijtJwc,87858
-triton/language/standard.py,sha256=8tTnMYLmjt3GI3V3shWLfYSYzLPFBkK5x0xV0_S-OUM,14442
+triton/experimental/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+triton/experimental/gluon/__init__.py,sha256=e2NX3d9SND2hKGmMDCix8_Sg12BMpK3zR3NWHy-fioQ,76
+triton/experimental/gluon/_compiler.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+triton/experimental/gluon/_runtime.py,sha256=b9KD4D-CuuYxwCwj2YPPpT_JnB9e1G4_ztaESaQjgHw,3395
+triton/experimental/gluon/language/__init__.py,sha256=EEDociOXDeGu9y2FwRKUoGNxJKu7eg2jIi94j7TQxx8,451
+triton/experimental/gluon/language/_core.py,sha256=8QyCvlTYQnqIDDrVa1AG84u7K3zaBAS_81QaJg3RfqE,8911
+triton/experimental/gluon/language/_layouts.py,sha256=H3Zkjmkl-IwJxCC4pnPxsBw-2mGNCNfn9MbY1c1l6s0,8975
+triton/experimental/gluon/language/_math.py,sha256=R9fMFusmpDy3rdA-zwxIIB5nigEg08MeTnSdslA0DQA,329
+triton/experimental/gluon/language/_semantic.py,sha256=SKRJPTtnhfWpuz-deaun845FpHvBvpeW2erGqO-UkdQ,13898
+triton/experimental/gluon/language/_standard.py,sha256=gunBiUjdix_LDVONM-OZ5wuR0i8O5fZLXHmOBEyvbLk,1063
+triton/experimental/gluon/language/nvidia/__init__.py,sha256=SFBuACK5P2XoYcutHEnKjqgRTboU4CPDmJz0hT6dFRQ,80
+triton/experimental/gluon/language/nvidia/blackwell/__init__.py,sha256=KHmHVhgmvhLQXsPfwExi_O9AfjQzpKYmqb3zWSy6W4Q,7599
+triton/experimental/gluon/language/nvidia/blackwell/tma.py,sha256=15xbiY2QmgjVPHvXPJ3MKrAkbqcaZjgx-V3B-kCgo9w,1086
+triton/experimental/gluon/language/nvidia/hopper/__init__.py,sha256=NVReVRoDY3OuXQUrKokuxY86z0QehVSje63H52u-5hs,295
+triton/experimental/gluon/language/nvidia/hopper/mbarrier.py,sha256=MezUtSQr-FzYM1kbuto1xNE4NimSBZIxP25xkn7nnp8,1603
+triton/experimental/gluon/language/nvidia/hopper/tma.py,sha256=Jih4obM2oGcUU7DYg0T7nSindQQd40zgQh0e1MeLh6A,3508
+triton/experimental/gluon/nvidia/__init__.py,sha256=ISXB4RV7RcCLsU-JhcRFeA29gCBDVk8cTwO2j99ivLc,80
+triton/experimental/gluon/nvidia/blackwell.py,sha256=cllwlUCE5_YKWqySQZk7wt7Fierz345E5VwztxNRGMs,69
+triton/experimental/gluon/nvidia/hopper.py,sha256=SKDi2fPCB87vMZAF6Em3gfZgif95U4Omeiexn7c969o,1518
+triton/language/__init__.py,sha256=XJPQq1rq0SoPcfqkw9YVln7XqP0I25nIGzTa-QmMpiY,6418
+triton/language/core.py,sha256=UsbrSmv92MdNW9JfKkyDhDOAZI8drrS1PK7Av2un7A4,116128
+triton/language/math.py,sha256=CKvuIc5iMKhz7Qgx9w-VcLfOOZadv5svKK4aGZLuHMc,7399
+triton/language/random.py,sha256=jkuFmfgZ8yvKuub9EY27zPvsC6nhkJIk05xf4y-7SR8,7102
+triton/language/semantic.py,sha256=ERQftF6yeuTMmPO722GWpfwug7r9ZDWhxThL1kwLfmI,96950
+triton/language/standard.py,sha256=EwpxORaDOiOdT4QgCnxiFXrsC652no7RptVoF9OszN0,16152
 triton/language/extra/__init__.py,sha256=XRXFvr7416pRsh_Rh-X6qV66SiEyVDVbxp4GSAE1mfc,655
 triton/language/extra/libdevice.py,sha256=Dki14elRNmQsz-Ytw9CnOaLCCnte4T6cI8bOzWjN63A,6318
-triton/language/extra/cuda/__init__.py,sha256=JqiuryHnWRkfFztXgxbiQ62XA4dEKhsjhIHGobLuzcQ,414
-triton/language/extra/cuda/_experimental_tma.py,sha256=vTZHUoUrdqC3Wp2T7Lwjl4G3MKp1AhRFoC15WqjTj2o,3420
-triton/language/extra/cuda/libdevice.py,sha256=crwXcdixYPuvzVOQ0e5styRAwQrUg0RRRlqek7QvXRw,56165
-triton/language/extra/cuda/utils.py,sha256=e1BslV7lZGhi2uVIlo5lI9dcN61HUMIU2asPaRjsyIo,4379
+triton/language/extra/cuda/__init__.py,sha256=MBBu2EUYxsp6ygjiwO4Yh1X1EswMstfaiRTMSMGtbcw,407
+triton/language/extra/cuda/gdc.py,sha256=QAqc_E1INKjYlW6ERSnb9uWoEBDAQlnxkn2yiIWHJPQ,2185
+triton/language/extra/cuda/libdevice.py,sha256=J7Kl0ejbAIus7-YBn2OSK71lkm3pC7G1J-5ZdHfS82U,56764
+triton/language/extra/cuda/utils.py,sha256=phDcXCFViaq3p4ThwHrO8-FtU-8A8I3nk4mZZJVvTio,4426
 triton/language/extra/hip/__init__.py,sha256=ieSER4LeX9_0horChGUUVwpuKAprkuka8uGAkEBDyDM,49
-triton/language/extra/hip/libdevice.py,sha256=EVraUfeXzQmN3F5Lleg2mohVcbFWOWlLaAH1nkbqtV4,16841
+triton/language/extra/hip/libdevice.py,sha256=Rf-AmBzcO6ORVzSxSuLXOy0lpoZTsnRAuTvjSF83r-E,17313
 triton/runtime/__init__.py,sha256=mKL5cqIBDUw2WO80NRCh4s1G8KYaqgM59TTAbTkPPjQ,621
 triton/runtime/_allocation.py,sha256=zaW4B7I7c-2rkVuN7IZaUB6IQSI1t4FvnTPZH-r7DTk,798
-triton/runtime/autotuner.py,sha256=ZkdEh1hFXOA9OJvaMgBzVWTNwwYcnNYrniEN1phYd24,17870
-triton/runtime/build.py,sha256=Kp-mqYUZV5YsiOlqA8f7cAaT5dtcb71l5Xg3-xvK4Cg,3869
-triton/runtime/cache.py,sha256=uoU1UH3HPxkxT0r-69HIZgLBls3T-TDxJedRULX9lbM,10583
-triton/runtime/driver.py,sha256=VZ-883Xri71R72lHB6usIpLo3gGLbZJkAlLP3ewWSpc,1509
+triton/runtime/autotuner.py,sha256=cfWBuLpL6-eBv-J2tFIbL0gE3ZGHOFSJ0e3n0GfrzLw,20244
+triton/runtime/build.py,sha256=hiFHKRV-fwDoT5lMbWpzA3hLp6LrE6ccw7zTo2AvHo0,6203
+triton/runtime/cache.py,sha256=uMV-CwCaS9cthIzKoLlTHXjhw_RoaIUaVP7zmgsdsIo,9689
+triton/runtime/driver.py,sha256=seGhU4efCFPVN0KVzd4gmZ1x5s0I_sFyM5NC8brXWF8,1798
 triton/runtime/errors.py,sha256=CwfJXciwel_-K3BfQfKUpLPDWrSyTnGsfJkqJojrdfQ,1052
-triton/runtime/interpreter.py,sha256=RVzNGSXyLlg14LgjJ5BZ3qbqfK49FccG-pXaIXXXE2g,59331
-triton/runtime/jit.py,sha256=jnMaMATWM3LU-s2QJPGeWVH5EGG78dT5aEUkGEhbpKc,34892
+triton/runtime/interpreter.py,sha256=IJ0kLHhnoRd9-lwtPk7l8IDs_reaDzLpIBJaH-h3S8g,60811
+triton/runtime/jit.py,sha256=M1bqBguYaMuztS3hH4-T5CI8UrvADyEC2mVuuwKVrvg,36766
 triton/runtime/tcc/libtcc.dll,sha256=4IVp00uvXFRsmhnF5tC1mT8Zb0Hl6uuxDlTHG1kQkrw,156160
 triton/runtime/tcc/tcc.exe,sha256=6cs-ieIKnv6tg8yeaxADFCdWNML3BQVtpx9CTqmwzfA,23552
 triton/runtime/tcc/include/_mingw.h,sha256=q0vn005_oOci8JSODJCtTZW4oexknC8Ybfo4e1e-eDM,3865
@@ -151,13 +172,15 @@ triton/runtime/tcc/lib/user32.def,sha256=EcYohyyDgmz9fLBoOR-vszLeJ2YkBUoNGvSnuXr
 triton/tools/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 triton/tools/build_extern.py,sha256=jCr-2hu3nLGBIJhCGUQ1jAyzLttughjkiPGEwRFjLR0,13673
 triton/tools/compile.py,sha256=CP_-yqEd55ejkc2_OYVE7q0Eyh9xErk8KJy2BcdCV0Y,7129
-triton/tools/disasm.py,sha256=BBO4bALdLcWgWDLhQdYHLlTx3oo8g_d8maeE_Uu-FmU,5088
-triton/tools/experimental_descriptor.py,sha256=0Wqy96Cc6YLh9o0eTknW-Lfvha6lfRSfe8bswkcPHMs,1260
+triton/tools/disasm.py,sha256=T9jiTkdK_0nI3R_4uea0zvfioYdcR-zIZwTfuucgw6g,5026
 triton/tools/link.py,sha256=u7qtfZRLriZkAMEGNvj8YF-k1cthmLL7BwHYqBgT63E,11871
 triton/tools/mxfp.py,sha256=YQdpBrGkOVNOtnLeRjMCeVFHWkSwUubGeWsItIjO8TU,11737
+triton/tools/tensor_descriptor.py,sha256=mt4iVVRcNg0gjoytb6iCP4l5vt-H2V3MGeAQfJcStJo,1289
 triton/tools/extra/cuda/compile.c,sha256=TdIENsqk6wrvv1C4Mk-sq9keXe3SJuMQcf0UpxmjNZk,2153
 triton/tools/extra/cuda/compile.h,sha256=n9QKIFZTL4RSsiXtAxBP9XGSnxjyaevQQ9bBpwDsvAg,332
-triton_windows-3.3.1.post21.dist-info/METADATA,sha256=kau-qMUZQ7G7g7qy5MvcD4KdIf8u0RngjQyogPyQaGY,1627
-triton_windows-3.3.1.post21.dist-info/WHEEL,sha256=qV0EIPljj1XC_vuSatRWjn02nZIz3N1t8jsZz7HBr2U,101
-triton_windows-3.3.1.post21.dist-info/top_level.txt,sha256=KhMzHYsArnZ3IkjAQ-xLnx1n_FjvEpJNelg2xPiDl-U,254
-triton_windows-3.3.1.post21.dist-info/RECORD,,
+triton_windows-3.4.0.post21.dist-info/licenses/LICENSE,sha256=kmQPuXIi_Qppj_KM4MN4LBcmI_jWxgm1V2NqgPKPuUY,1132
+triton_windows-3.4.0.post21.dist-info/METADATA,sha256=IHOZbw7LyV2c8dx0dvpgjzw8LmdA99209iNKB1UC01o,1794
+triton_windows-3.4.0.post21.dist-info/WHEEL,sha256=qV0EIPljj1XC_vuSatRWjn02nZIz3N1t8jsZz7HBr2U,101
+triton_windows-3.4.0.post21.dist-info/entry_points.txt,sha256=cztF9ZYXxoMhibI_OttiKCl1EBP2LQaV8naJ-BcuES4,76
+triton_windows-3.4.0.post21.dist-info/top_level.txt,sha256=WBiIZyv6n9Y7MIh-HPHSv2w1RDk7EFL__7ZgQRrmHYs,7
+triton_windows-3.4.0.post21.dist-info/RECORD,,

triton_windows-3.4.0.post21.dist-info/entry_points.txt

@@ -0,0 +1,3 @@
+[triton.backends]
+amd = triton.backends.amd
+nvidia = triton.backends.nvidia

triton_windows-3.4.0.post21.dist-info/licenses/LICENSE

@@ -0,0 +1,23 @@
+/*
+* Copyright 2018-2020 Philippe Tillet
+* Copyright 2020-2022 OpenAI
+*
+* 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.
+*/

triton_windows-3.4.0.post21.dist-info/top_level.txt

@@ -0,0 +1 @@
+triton

triton/language/_utils.py

@@ -1,21 +0,0 @@
-from typing import List
-
-TRITON_MAX_TENSOR_NUMEL = 1048576
-
-
-def is_power_of_two(x):
-    return (x & (x - 1)) == 0
-
-
-def validate_block_shape(shape: List[int]):
-    numel = 1
-    for i, d in enumerate(shape):
-        if not isinstance(d, int):
-            raise TypeError(f"Shape element {i} must have type `constexpr[int]`, got `constexpr[{type(d)}]")
-        if not is_power_of_two(d):
-            raise ValueError(f"Shape element {i} must be a power of 2")
-        numel *= d
-
-    if numel > TRITON_MAX_TENSOR_NUMEL:
-        raise ValueError(f"numel ({numel}) exceeds triton maximum tensor numel ({TRITON_MAX_TENSOR_NUMEL})")
-    return numel

triton/language/extra/cuda/_experimental_tma.py

@@ -1,106 +0,0 @@
-from typing import Sequence
-
-from triton.language import core
-from triton.language import semantic
-from triton._C.libtriton import ir
-
-__all__ = [
-    "experimental_device_tensormap_create1d",
-    "experimental_device_tensormap_create2d",
-    "experimental_tensormap_fenceproxy_acquire",
-]
-
-
-def _determine_elem_type(element_ty: core.dtype):
-    if element_ty.primitive_bitwidth == 8:
-        return 0
-    elif element_ty.primitive_bitwidth == 16:
-        return 1
-    elif element_ty.primitive_bitwidth == 32:
-        return 2
-    else:
-        raise ValueError("element_ty must be a primitive of size 1, 2, or 4 bytes but got")
-
-
-@core.builtin
-def experimental_device_tensormap_create1d(
-    desc_ptr: core.tensor,
-    global_address: core.tensor,
-    load_size: core.tensor,
-    global_size: core.tensor,
-    element_ty: core.dtype,
-    _builder: ir.builder = None,
-):
-    load_size = core._constexpr_to_value(load_size)
-    global_size = semantic.to_tensor(global_size, _builder)
-    element_ty = core._constexpr_to_value(element_ty)
-    element_stride = [core.full([], 1, core.int32, _builder=_builder)]
-
-    semantic.tensormap_create(
-        desc_ptr=desc_ptr,
-        global_address=global_address,
-        box_dim=[semantic.to_tensor(load_size, _builder)],
-        global_dim=[global_size],
-        global_stride=[],
-        element_stride=element_stride,
-        elem_type=_determine_elem_type(element_ty),
-        interleave_layout=0,
-        swizzle_mode=0,
-        fill_mode=0,
-        builder=_builder,
-    )
-
-
-@core.builtin
-def experimental_device_tensormap_create2d(
-    desc_ptr: core.tensor,
-    global_address: core.tensor,
-    load_size: Sequence[core.constexpr],
-    global_size: Sequence[core.tensor],
-    element_ty: core.dtype,
-    _builder: ir.builder = None,
-):
-    assert len(load_size) == 2
-    assert len(global_size) == 2
-    load_size = [core._constexpr_to_value(x) for x in load_size]
-    global_size = [semantic.to_tensor(x, _builder) for x in global_size]
-
-    element_size = element_ty.primitive_bitwidth // 8
-    element_size_t = core.full([], element_size, core.int64, _builder=_builder)
-    global_stride = semantic.mul(element_size_t, global_size[-1], True, _builder)
-
-    contig_dim_size_in_bytes = element_size * load_size[-1]
-    if contig_dim_size_in_bytes > 128:
-        load_size[-1] = 128 // element_size
-
-    elem_stride = core.full([], 1, core.int32, _builder=_builder)
-
-    semantic.tensormap_create(
-        desc_ptr=desc_ptr,
-        global_address=global_address,
-        box_dim=[semantic.to_tensor(x, _builder) for x in load_size[::-1]],
-        global_dim=global_size[::-1],
-        global_stride=[global_stride],
-        element_stride=[elem_stride, elem_stride],
-        elem_type=_determine_elem_type(element_ty),
-        interleave_layout=0,
-        swizzle_mode=_determine_swizzle_mode_2d(contig_dim_size_in_bytes, load_size),
-        fill_mode=0,
-        builder=_builder,
-    )
-
-
-def _determine_swizzle_mode_2d(contig_dim_size_in_bytes, load_size):
-    if contig_dim_size_in_bytes >= 128:
-        return 3
-    elif contig_dim_size_in_bytes >= 64:
-        return 2
-    elif contig_dim_size_in_bytes >= 32:
-        return 1
-    else:
-        raise ValueError("block size too small")
-
-
-@core.builtin
-def experimental_tensormap_fenceproxy_acquire(desc_ptr: core.tensor, _builder: ir.builder = None):
-    semantic.tensormap_fenceproxy_acquire(desc_ptr, _builder)

triton/tools/experimental_descriptor.py

@@ -1,32 +0,0 @@
-import torch
-
-import triton
-
-
-class TmaDescKernelParam:
-    TMA_DESC_SIZE = 128
-
-    def __init__(self, ptr, dims, block_dims, element_size):
-        self.desc = torch.empty(self.TMA_DESC_SIZE, dtype=torch.uint8, device="cpu")
-        assert len(dims) == len(block_dims)
-        assert 1 <= len(dims) <= 2
-        assert self.desc.data_ptr() % 64 == 0
-
-        if len(dims) == 1:
-            triton.runtime.driver.active.utils.fill_1d_tma_descriptor(ptr, dims[0], block_dims[0], element_size,
-                                                                      self.desc.data_ptr())
-        else:
-            triton.runtime.driver.active.utils.fill_2d_tma_descriptor(ptr, dims[0], dims[1], block_dims[0],
-                                                                      block_dims[1], element_size, self.desc.data_ptr())
-
-    # Return a CUtensorMap* pointer in host memory
-    def tma_desc_cpu_ptr(self):
-        return self.desc.data_ptr()
-
-
-def create_1d_tma_descriptor(ptr, dim, block_dim, element_size):
-    return TmaDescKernelParam(ptr, [dim], [block_dim], element_size)
-
-
-def create_2d_tma_descriptor(ptr, dim1, dim0, block_dim1, block_dim0, element_size):
-    return TmaDescKernelParam(ptr, [dim1, dim0], [block_dim1, block_dim0], element_size)

triton_windows-3.3.1.post21.dist-info/top_level.txt

@@ -1,14 +0,0 @@
-triton
-triton/_C
-triton/backends
-triton/backends/amd
-triton/backends/nvidia
-triton/compiler
-triton/language
-triton/language/extra
-triton/language/extra\cuda
-triton/language/extra\hip
-triton/runtime
-triton/tools
-triton/tools/extra
-triton/tools/extra\cuda