feat: drop support for CUDA<12.3

Signed-off-by: Zihua Wu <13583761+lucifer1004@users.noreply.github.com>

deep_gemm/jit/compiler.py

@@ -14,7 +14,7 @@ import cuda.bindings.nvrtc as nvrtc
 from torch.utils.cpp_extension import CUDA_HOME
 
 from . import interleave_ffma
-from .runtime import Runtime, Fp8GemmRuntime, RuntimeCache, get_symbol
+from .runtime import Runtime, Fp8GemmRuntime, RuntimeCache
 
 runtime_cache = RuntimeCache()
 
@@ -115,7 +115,7 @@ class Compiler(abc.ABC):
 
     @classmethod
     @abc.abstractmethod
-    def compile(cls, name: str, code: str, target_path: str, kernel_name_pattern: str) -> str:
+    def compile(cls, name: str, code: str, target_path: str) -> str:
         pass
 
     @staticmethod
@@ -132,10 +132,9 @@ class Compiler(abc.ABC):
         return [get_jit_include_dir()]
 
     @classmethod
-    def build(cls, name: str, code: str, kernel_name_pattern: str, runtime_cls: Type[Runtime] = Fp8GemmRuntime) -> Runtime:
+    def build(cls, name: str, code: str, runtime_cls: Type[Runtime] = Fp8GemmRuntime) -> Runtime:
         # Compiler flags
         flags = cls.flags()
-        include_dirs = cls.include_dirs()
 
         # Build signature
         enable_sass_opt = get_nvcc_compiler()[1] <= '12.8' and int(
@@ -158,7 +157,7 @@ class Compiler(abc.ABC):
         tmp_cubin_path = os.path.join(make_tmp_dir(), f'nvcc.tmp.{str(uuid.uuid4())}.{hash_to_hex(cubin_path)}.cubin')
 
         start_time = time.time()
-        kernel_name = cls.compile(name, code, tmp_cubin_path, kernel_name_pattern)
+        cls.compile(name, code, tmp_cubin_path)
         end_time = time.time()
         elapsed_time = end_time - start_time
         if os.getenv('DG_JIT_DEBUG', None):
@@ -169,15 +168,11 @@ class Compiler(abc.ABC):
         if enable_sass_opt:
             interleave_ffma.process(tmp_cubin_path)
             
-        # Store kernel name
-        put(f'{tmp_cubin_path}.name', kernel_name)
-
         # Atomic replace files
         os.replace(tmp_cubin_path, cubin_path)
-        os.replace(f'{tmp_cubin_path}.name', f'{cubin_path}.name')
 
         # Put cache and return
-        runtime = runtime_cls(path, kernel_name)
+        runtime = runtime_cls(path)
         runtime_cache[path] = runtime
         return runtime
 
@@ -202,7 +197,7 @@ class NvccCompiler(Compiler):
                 f'--compiler-options={",".join(cxx_flags)}']
 
     @classmethod
-    def compile(cls, name: str, code: str, target_path: str, kernel_name_pattern: str) -> str:
+    def compile(cls, name: str, code: str, target_path: str):
         # Write the code
         path = os.path.join(get_cache_dir(), name)
         src_path = os.path.join(path, 'kernel.cu')
@@ -221,9 +216,6 @@ class NvccCompiler(Compiler):
         
         assert result.returncode == 0, f'Failed to compile {src_path}'
 
-        # NVCC needs to get the symbol name from the cubin file using `cuobjdump`
-        return get_symbol(target_path, kernel_name_pattern)
-
 
 class NvrtcCompiler(Compiler):
     @staticmethod
@@ -238,7 +230,7 @@ class NvrtcCompiler(Compiler):
     def include_dirs() -> List[str]:
         if CUDA_HOME is None:
             raise RuntimeError('CUDA_HOME is required for NVRTC compilation')
-        return [get_jit_include_dir(), os.path.join(CUDA_HOME, 'include'), os.path.join(CUDA_HOME, 'targets', 'x86_64-linux', 'include')]
+        return [get_jit_include_dir(), os.path.join(CUDA_HOME, 'include')]
 
     @classmethod
     def flags(cls) -> List[str]:
@@ -251,20 +243,12 @@ class NvrtcCompiler(Compiler):
         return base_flags
 
     @classmethod
-    def compile(cls, name: str, code: str, target_path: str, kernel_name_pattern: str) -> str:
+    def compile(cls, name: str, code: str, target_path: str) -> str:
         code_bytes = bytes(code, 'utf-8')
         res, program = nvrtc.nvrtcCreateProgram(
             code_bytes, bytes(name, 'utf-8'), 0, [], [])
         if res != nvrtc.nvrtcResult.NVRTC_SUCCESS:
-            raise Exception(f"Failed to create program: {res}")
+            raise Exception(f'Failed to create program: {res}')
-
-        kernel_regex = re.compile(kernel_name_pattern, re.MULTILINE)
-        kernel_name = kernel_regex.search(code).group(
-            0).replace('\n', '').replace(' ', '')
-        res = nvrtc.nvrtcAddNameExpression(
-            program, bytes(kernel_name, 'utf-8'))[0]
-        if res != nvrtc.nvrtcResult.NVRTC_SUCCESS:
-            raise Exception(f"Failed to add name expression: {res}")
 
         options = [bytes(flag, 'utf-8') for flag in cls.flags()]
         compile_res = nvrtc.nvrtcCompileProgram(
@@ -273,43 +257,35 @@ class NvrtcCompiler(Compiler):
         if os.getenv('DG_JIT_DEBUG', None):
             res, log_size = nvrtc.nvrtcGetProgramLogSize(program)
             if res != nvrtc.nvrtcResult.NVRTC_SUCCESS:
-                raise Exception(f"Failed to get program log size: {res}")
+                raise Exception(f'Failed to get program log size: {res}')
             log_bytes = bytes(log_size)
             res = nvrtc.nvrtcGetProgramLog(program, log_bytes)[0]
             if res != nvrtc.nvrtcResult.NVRTC_SUCCESS:
-                raise Exception(f"Failed to get program log: {res}")
+                raise Exception(f'Failed to get program log: {res}')
             log_str = log_bytes.decode('utf-8')
             print(log_str)
 
         if compile_res != nvrtc.nvrtcResult.NVRTC_SUCCESS:
-            raise Exception(f"Failed to compile program: {compile_res}")
+            raise Exception(f'Failed to compile program: {compile_res}')
-
-        # NVRTC can directly get the lowered name
-        res, lowered_name = nvrtc.nvrtcGetLoweredName(
-            program, bytes(kernel_name, 'utf-8'))
-        if res != nvrtc.nvrtcResult.NVRTC_SUCCESS:
-            raise Exception(f"Failed to get lowered name: {res}")
 
         res, cubin_size = nvrtc.nvrtcGetCUBINSize(program)
         if res != nvrtc.nvrtcResult.NVRTC_SUCCESS:
-            raise Exception(f"Failed to get CUBIN size: {res}")
+            raise Exception(f'Failed to get CUBIN size: {res}')
 
         cubin_bytes = bytes(cubin_size)
         res = nvrtc.nvrtcGetCUBIN(program, cubin_bytes)[0]
         if res != nvrtc.nvrtcResult.NVRTC_SUCCESS:
-            raise Exception(f"Failed to get CUBIN: {res}")
+            raise Exception(f'Failed to get CUBIN: {res}')
 
         put(target_path, cubin_bytes)
 
         res = nvrtc.nvrtcDestroyProgram(program)[0]
         if res != nvrtc.nvrtcResult.NVRTC_SUCCESS:
-            raise Exception(f"Failed to destroy program: {res}")
+            raise Exception(f'Failed to destroy program: {res}')
-
-        return lowered_name.decode('utf-8')
 
 
-def build(name: str, code: str) -> Runtime:
+def build(name: str, code: str, runtime_cls: Type[Runtime] = Fp8GemmRuntime) -> Runtime:
     if os.getenv('DG_JIT_USE_NVRTC', '0') in ['1', 'true', 'True']:
-        return NvrtcCompiler.build(name, code, kernel_name_pattern=r'fp8_gemm_kernel<[\S\s]*?>')
+        return NvrtcCompiler.build(name, code, runtime_cls=runtime_cls)
     else:
-        return NvccCompiler.build(name, code, kernel_name_pattern='fp8_gemm_kernel')
+        return NvccCompiler.build(name, code, runtime_cls=runtime_cls)

deep_gemm/jit/runtime.py

@@ -1,31 +1,12 @@
 import os
-import platform
 import time
-import subprocess
 from typing import Any, Callable, Dict, List, Optional, Type
 
 import cuda.bindings.driver as cuda
-from torch.utils.cpp_extension import CUDA_HOME
 
 from .utils import run_gemm
 
 
-def get_symbol(file_path: str, pattern: str) -> Optional[str]:
-    if CUDA_HOME is None:
-        raise Exception("CUDA_HOME is not set")
-
-    cuobjdump_bin = 'cuobjdump.exe' if platform.system() == 'Windows' else 'cuobjdump'
-    command = [os.path.join(CUDA_HOME, 'bin', cuobjdump_bin),
-               '-symbols', file_path]
-    result = subprocess.run(command, stdout=subprocess.PIPE,
-                            stderr=subprocess.PIPE, text=True)
-    assert result.returncode == 0
-    for line in result.stdout.splitlines():
-        if pattern in line:
-            return line.split()[-1]
-    return None
-
-
 class Runtime:
     def __init__(self, path: str, kernel_name: str, caller: Callable[..., cuda.CUresult], args: List[str]) -> None:
         self.path = path
@@ -43,7 +24,7 @@ class Runtime:
             return False
 
         # Contains all necessary files
-        files = ['kernel.cubin', 'kernel.cubin.name']
+        files = ['kernel.cubin']
         return all(os.path.exists(os.path.join(path, file)) for file in files)
 
     def __call__(self, **kwargs: Dict[str, Any]) -> cuda.CUresult:
@@ -53,18 +34,28 @@ class Runtime:
             res, lib = cuda.cuLibraryLoadFromFile(
                 bytes(os.path.join(self.path, 'kernel.cubin'), 'utf-8'), [], [], 0, [], [], 0)
             if res != cuda.CUresult.CUDA_SUCCESS:
-                raise Exception(f"Failed to load library: {res}")
+                raise Exception(f'Failed to load library: {res}')
 
-            res, kernel = cuda.cuLibraryGetKernel(
+            res, kernel_count = cuda.cuLibraryGetKernelCount(lib)
-                lib, bytes(self.kernel_name, encoding='utf-8'))
             if res != cuda.CUresult.CUDA_SUCCESS:
-                raise Exception(f"Failed to get kernel: {res}")
+                raise Exception(f'Failed to get kernel count: {res}')
             
+            res, kernels = cuda.cuLibraryEnumerateKernels(kernel_count, lib)
+            if res != cuda.CUresult.CUDA_SUCCESS:
+                raise Exception(f'Failed to enumerate kernels: {res}')
+
+            for kernel in kernels:
+                res, kernel_name = cuda.cuKernelGetName(kernel)
+                if res != cuda.CUresult.CUDA_SUCCESS:
+                    raise Exception(f'Failed to get kernel name: {res}')
+                if bytes(self.kernel_name, encoding='utf-8') in kernel_name:
                     self.kernel = kernel
+                    break
+
             if self.kernel is not None:
                 self.lib = lib
             else:
-                raise Exception("Failed to find kernel")
+                raise Exception('Failed to find required kernel')
 
             end_time = time.time_ns()
             elapsed_time = (end_time - start_time) / 1000
@@ -81,12 +72,12 @@ class Runtime:
         if self.lib is not None:
             res = cuda.cuLibraryUnload(self.lib)[0]
             if res != cuda.CUresult.CUDA_SUCCESS:
-                raise Exception(f"Failed to unload library {self.path}: {res}")
+                raise Exception(f'Failed to unload library {self.path}: {res}')
 
 
 class Fp8GemmRuntime(Runtime):
-    def __init__(self, path: str, kernel_name: str) -> None:
+    def __init__(self, path: str) -> None:
-        super().__init__(path, kernel_name, run_gemm, [
+        super().__init__(path, 'fp8_gemm', run_gemm, [
             'NUM_TMA_MULTICAST',
             'M',
             'BLOCK_M',
@@ -117,9 +108,7 @@ class RuntimeCache:
 
         # Already compiled
         if os.path.exists(path) and Runtime.is_path_valid(path):
-            kernel_name = open(os.path.join(
+            runtime = runtime_cls(path)
-                path, 'kernel.cubin.name'), 'r').read()
-            runtime = runtime_cls(path, kernel_name)
             self.cache[path] = runtime
             return runtime
         return None

deep_gemm/jit/template.py

@@ -24,7 +24,6 @@ def generate(**kwargs: Dict[str, Any]) -> str:
 
 using namespace deep_gemm;
 
-#ifndef NVRTC_JIT_COMPILATION
 __global__ void dummy_kernel() {{
   void *ptr = (void *)&fp8_gemm_kernel<
     {kwargs['N']},
@@ -43,7 +42,6 @@ __global__ void dummy_kernel() {{
     GemmType::{kwargs['GEMM_TYPE']}
   >;
 }}
-#endif
 '''
 
     # Debug print

deep_gemm/jit/utils.py

@@ -53,7 +53,7 @@ def get_num_math_warpgroups(block_m: int) -> int:
     return 1 if block_m == 64 else 2
 
 def get_num_threads_per_sm(num_tma_threads: int, num_math_threads_per_group: int, block_m: int) -> int:
-    assert num_math_threads_per_group == 128, "Only support 128 threads per math group"
+    assert num_math_threads_per_group == 128, 'Only support 128 threads per math group'
     return get_num_math_warpgroups(block_m) * num_math_threads_per_group + num_tma_threads
 
 
@@ -74,7 +74,7 @@ def make_2d_tma_copy_desc(global_address: torch.Tensor, gmem_dim: Tuple[cuda.cuu
     )
 
     if res != cuda.CUresult.CUDA_SUCCESS:
-        raise Exception(f"Failed to encode tensor map: {res}")
+        raise Exception(f'Failed to encode tensor map: {res}')
 
     return tensor_map
 
@@ -118,7 +118,7 @@ def run_gemm(kernel: cuda.CUkernel, num_tma_multicast: int, shape_m: int, block_
     
     res = cuda.cuKernelSetAttribute(cuda.CUfunction_attribute.CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES, smem_size, kernel, cuda.CUdevice(gmem_d.device.index))[0]
     if res != cuda.CUresult.CUDA_SUCCESS:
-        raise Exception(f"Failed to set max dynamic shared memory size: {res}")
+        raise Exception(f'Failed to set max dynamic shared memory size: {res}')
 
     attr_val = cuda.CUlaunchAttributeValue()
     attr_val.clusterDim.x = num_tma_multicast

tests/test_jit.py

@@ -62,17 +62,15 @@ __global__ void vector_add(T* a, T* b, T* c, uint32_t N) {{
     }}
 }}
 
-#ifndef NVRTC_JIT_COMPILATION
 __global__ void dummy_kernel() {{
     void *ptr = (void *)&vector_add<{kwargs['T']}>;
 }}
-#endif
 """
 
 
 class VectorAddRuntime(jit.Runtime):
-    def __init__(self, path: str, kernel_name: str) -> None:
+    def __init__(self, path: str) -> None:
-        super().__init__(path, kernel_name, run_vector_add, [
+        super().__init__(path, 'vector_add', run_vector_add, [
             'A',
             'B',
             'C',
@@ -87,7 +85,7 @@ if __name__ == '__main__':
     code = generate_vector_add(T='float')
     print(code)
     print('Building ...')
-    func = jit.NvccCompiler.build('test_func', code, 'vector_add', VectorAddRuntime)
+    func = jit.NvccCompiler.build('test_func', code, VectorAddRuntime)
 
     a = torch.randn((1024, ), dtype=torch.float32, device='cuda')
     b = torch.randn((1024, ), dtype=torch.float32, device='cuda')
@@ -105,7 +103,7 @@ if __name__ == '__main__':
     code = generate_vector_add(T='__nv_bfloat16')
     print(code)
     print('Building ...')
-    func = jit.NvrtcCompiler.build('test_func', code, r'vector_add<[\S\s]*?>', VectorAddRuntime)
+    func = jit.NvrtcCompiler.build('test_func', code, VectorAddRuntime)
 
     a = torch.randn((1024, ), dtype=torch.bfloat16, device='cuda')
     b = torch.randn((1024, ), dtype=torch.bfloat16, device='cuda')