triton-windows 3.3.0a0.post14__cp39-cp39-win_amd64.whl → 3.3.0a0.post16__cp39-cp39-win_amd64.whl

triton/backends/nvidia/compiler.py

@@ -403,11 +403,10 @@ class CUDABackend(BaseBackend):
 
     def make_cubin(self, src, metadata, opt, capability):
         ptxas, _ = get_ptxas(self.target.arch)
-        # On Windows, we need to set delete=False, close the temp file before reading it, and manually remove it
         with tempfile.NamedTemporaryFile(delete=False, mode='w', suffix='.ptx') as fsrc, \
             tempfile.NamedTemporaryFile(delete=False, mode='r', suffix='.log') as flog:
             fsrc.write(src)
-            fsrc.close()
+            fsrc.flush()
             fbin = fsrc.name + '.o'
 
             line_info = ["-lineinfo", "-suppress-debug-info"] if os.environ.get("TRITON_DISABLE_LINE_INFO",
@@ -417,10 +416,8 @@ class CUDABackend(BaseBackend):
             opt_level = ['--opt-level', '0'] if os.environ.get("DISABLE_PTXAS_OPT", "0") == "1" else []
             ptxas_cmd = [ptxas, *line_info, *fmad, '-v', *opt_level, f'--gpu-name={arch}', fsrc.name, '-o', fbin]
             try:
-                subprocess.run(ptxas_cmd, check=True, close_fds=False, stdout=flog, stderr=flog)
-                flog.close()
+                subprocess.run(ptxas_cmd, check=True, close_fds=True, stdout=flog, stderr=flog)
             except subprocess.CalledProcessError as e:
-                flog.close()
                 with open(flog.name) as log_file:
                     log = log_file.read()
 
@@ -434,13 +431,14 @@ class CUDABackend(BaseBackend):
                 raise PTXASError(f"{error}\n"
                                  f"`ptxas` stderr:\n{log}\n"
                                  f'Repro command: {" ".join(ptxas_cmd)}\n')
-            finally:
-                try_remove(fsrc.name)
-                try_remove(flog.name)
 
-            with open(fbin, 'rb') as f:
-                cubin = f.read()
-            try_remove(fbin)
+        with open(fbin, 'rb') as f:
+            cubin = f.read()
+        try_remove(fbin)
+
+        # It's better to remove the temp files outside the context managers
+        try_remove(fsrc.name)
+        try_remove(flog.name)
         return cubin
 
     def add_stages(self, stages, options):

triton/backends/nvidia/driver.c

@@ -9,6 +9,7 @@
 
 #include <stdbool.h>
 #define PY_SSIZE_T_CLEAN
+#define Py_LIMITED_API 0x03090000
 #include <Python.h>
 
 // Raises a Python exception and returns false if code is not CUDA_SUCCESS.

triton/backends/nvidia/driver.py

@@ -206,6 +206,8 @@ def make_launcher(constants, signature):
     src = f"""
 #include \"cuda.h\"
 #include <stdbool.h>
+#define PY_SSIZE_T_CLEAN
+#define Py_LIMITED_API 0x03090000
 #include <Python.h>
 
 #ifndef _WIN32

{triton_windows-3.3.0a0.post14.dist-info → triton_windows-3.3.0a0.post16.dist-info}/METADATA

@@ -1,6 +1,6 @@
-Metadata-Version: 2.2
+Metadata-Version: 2.4
 Name: triton-windows
-Version: 3.3.0a0.post14
+Version: 3.3.0a0.post16
 Summary: A language and compiler for custom Deep Learning operations
 Home-page: https://github.com/woct0rdho/triton-windows
 Author: Philippe Tillet, Dian Wu

{triton_windows-3.3.0a0.post14.dist-info → triton_windows-3.3.0a0.post16.dist-info}/RECORD

@@ -4,7 +4,7 @@ triton/_utils.py,sha256=5RiCLwW14w0Q3mdZ-9yz-VO5KiSexNj9xeDt4gaNsvE,1014
 triton/errors.py,sha256=8WfnuRKLG578mgY6cBA3ECruVMf9ULEKFNgRcJ6IhWM,89
 triton/testing.py,sha256=ivFf1Fq9frmfVahaVUp0bgJxmvVZNACZfj3Sai6zfAs,20048
 triton/windows_utils.py,sha256=aQMItmuZNXaki8zSB7HMvhy1RsXlmz1GcO-lORm1IIk,10852
-triton/_C/libtriton.pyd,sha256=hyIqWiSBaVnnxLUjjZF23t8FCfWOX0hGOYvQcVviTCo,86823424
+triton/_C/libtriton.pyd,sha256=Mo1TfRFiaem1-G90baea4QQ3JpOU_9j-HP5903E0Y0Q,86823424
 triton/backends/__init__.py,sha256=opAo_vgEMt3tLO_bYFrYGksnIu0qohbmyuu_s3-rNAs,1595
 triton/backends/compiler.py,sha256=ymaG0kpveAuESbQ9QZ0RyXjr0Aq4el_G5XGYogJ2gNA,3588
 triton/backends/driver.py,sha256=AN60upJlPgia0JwvZ8vIVgLITNPuI0fdz8zMIIHPpF4,1450
@@ -112,9 +112,9 @@ triton/backends/amd/lib/asanrtl.bc,sha256=1xv2RlU3WvbdsghHlmhwiHewGM2B5dKts5bERM
 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=R0WKIM_JsMcd3nlwnrlB8pXPjyDj6egaaFxNwi2hneM,18588
-triton/backends/nvidia/driver.c,sha256=-O-Q00SFcCHJiHvArIC7yB10Ma1W8jw7ORFKp7spxHE,18846
-triton/backends/nvidia/driver.py,sha256=9ZhZTs19M5IU4fAD4JBNhh24MYcP6QLmEMW3SfGgfAQ,20843
+triton/backends/nvidia/compiler.py,sha256=iHbIl-HXSgD3S5eJIXbQhr-Cyk9Gq_g8RXDFZxAGUF4,18444
+triton/backends/nvidia/driver.c,sha256=RWzdn9_RWaaVxXKtdrvxoRg4yR7WSH7ELRyjGDYKwBM,18880
+triton/backends/nvidia/driver.py,sha256=BcEzr0_ONM8H_DKoCbNuO1jRpBM_XHKBgt_Y2bG9aZw,20902
 triton/backends/nvidia/bin/ptxas.exe,sha256=iCva9hIYg-Q2NybchwaQJFkwDzNiliFOTDdZrHPLb6A,24732160
 triton/backends/nvidia/include/cuda.h,sha256=Fn44OjeRImxegJ39apYUspseEfTWNGwpqSGUOnHj5WY,1183268
 triton/backends/nvidia/lib/libdevice.10.bc,sha256=XC-uN8huaMOjhgWpX1EtfRLV89uYYxC-R_VzBKpype4,473728
@@ -247,7 +247,7 @@ triton/tools/link.py,sha256=u7qtfZRLriZkAMEGNvj8YF-k1cthmLL7BwHYqBgT63E,11871
 triton/tools/mxfp.py,sha256=YQdpBrGkOVNOtnLeRjMCeVFHWkSwUubGeWsItIjO8TU,11737
 triton/tools/extra/cuda/compile.c,sha256=Me7beHPc6WNTwjg85H84DUMCRu4KJdVK2hNNgvlhBZ4,2126
 triton/tools/extra/cuda/compile.h,sha256=n9QKIFZTL4RSsiXtAxBP9XGSnxjyaevQQ9bBpwDsvAg,332
-triton_windows-3.3.0a0.post14.dist-info/METADATA,sha256=tYF6HEgycHirPm3FRqIWkuxccUaU5kG2duUj8DEUCAM,1629
-triton_windows-3.3.0a0.post14.dist-info/WHEEL,sha256=fvNoTPbNdRAfsv1hWsmn2TKqULUpllfaSGBE1Ye0N8Q,99
-triton_windows-3.3.0a0.post14.dist-info/top_level.txt,sha256=KhMzHYsArnZ3IkjAQ-xLnx1n_FjvEpJNelg2xPiDl-U,254
-triton_windows-3.3.0a0.post14.dist-info/RECORD,,
+triton_windows-3.3.0a0.post16.dist-info/METADATA,sha256=soxppI1GtZ5SPZoHEK6khzoUWAnqHdzSINwOc5AsKLs,1629
+triton_windows-3.3.0a0.post16.dist-info/WHEEL,sha256=DiyQlQBAA8vWmmwFRDPSwjrGdTC0GCx3phqzmWJ5fvE,99
+triton_windows-3.3.0a0.post16.dist-info/top_level.txt,sha256=KhMzHYsArnZ3IkjAQ-xLnx1n_FjvEpJNelg2xPiDl-U,254
+triton_windows-3.3.0a0.post16.dist-info/RECORD,,

{triton_windows-3.3.0a0.post14.dist-info → triton_windows-3.3.0a0.post16.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (76.0.0)
+Generator: setuptools (77.0.1)
 Root-Is-Purelib: false
 Tag: cp39-cp39-win_amd64