fix build and nvrtc problem

CHANGELOG.md

 # Changelog
+## [2.2.1] - 2022-9-25
+### Fixed 
+- Fix build problem
+- Fix nvrtc problem
+
 ## [2.2.0] - 2022-9-24
 ### Added 
 - Add Ampere support. faster fp16, faster tf32 and greatly faster int8 kernels in Ampere GPUs.

pyproject.toml

 [build-system]
-requires = ["setuptools>=41.0", "wheel", "pccm>=0.4.0", "cumm>=0.3.1"]
+requires = ["setuptools>=41.0", "wheel", "pccm>=0.4.0", "cumm>=0.3.2"]
 build-backend = "setuptools.build_meta"

setup.py

@@ -38,9 +38,9 @@ if cuda_ver:
     cuda_ver = cuda_ver.replace(".", "") # 10.2 to 102
 
     RELEASE_NAME += "-cu{}".format(cuda_ver)
-    deps = ["cumm-cu{}>=0.3.1".format(cuda_ver)]
+    deps = ["cumm-cu{}>=0.3.2".format(cuda_ver)]
 else:
-    deps = ["cumm>=0.3.1"]
+    deps = ["cumm>=0.3.2"]
 
 
 

spconv/algo.py

@@ -17,7 +17,7 @@ import time
 from enum import Enum
 from threading import Lock
 from typing import Dict, List, Optional, Set, Tuple, Union
-
+from spconv.core_cc.cumm.common import CompileInfo
 import numpy as np
 from cumm import tensorview as tv
 from cumm.conv.bases import ConvLayout, ConvLayoutType, ConvOpType
@@ -337,9 +337,20 @@ class SimpleGemm:
             ldb = b.stride[0]
             ldc = c.stride[0]
             if desp.supported_ldx(lda, ldb, ldc):
-                if arch not in COMPILED_CUDA_GEMM_ARCHS:
+                if desp.is_nvrtc:
+                    if not CompileInfo.algo_can_be_nvrtc_compiled(desp.min_arch):
+                        continue
+                if not CompileInfo.arch_is_compiled_gemm(arch):
+                    # use PTX of possible
+                    if not CompileInfo.gemm_algo_can_use_ptx(desp.min_arch, arch):
+                        if CompileInfo.algo_can_be_nvrtc_compiled(desp.min_arch):
+                            # compiled kernel can't use PTX, for example, desp need at least sm_80 and only sm_75+PTX is compiled
+                            # all sm_80 code of this desp is invalid, we must use nvrtc.
+                            # only desp <= sm_75 can use virtual PTX code to generate sm_80 code.
                             desp = desp.copy()
                             desp.is_nvrtc = True
+                        else:
+                            continue
                 if SPCONV_DEBUG_NVRTC_KERNELS:
                     desp.is_nvrtc = True
                 finally_algos.append(desp)
@@ -455,7 +466,7 @@ class SimpleGemm:
             if desp.split_k_serial and hint & AlgoHint.BackwardWeight.value:
                 split_k_slices = max(min(32, k // 128), 1)
             params = GemmParams()
-            if desp.is_nvrtc and str(desp) not in self.prebuilt_desp_names:
+            if desp.is_nvrtc or str(desp) not in self.prebuilt_desp_names:
                 params.nvrtc_params = self._cached_get_nvrtc_params(desp, arch)
             params.a = a
             params.b = b
@@ -550,7 +561,7 @@ class SimpleGemm:
             split_k_slices = profile_res.splitk
         params = GemmParams()
         is_not_static = str(algo_desp) not in self.prebuilt_desp_names
-        if algo_desp.is_nvrtc and (is_not_static or force_nvrtc):
+        if algo_desp.is_nvrtc or is_not_static or force_nvrtc:
             params.nvrtc_params = self._cached_get_nvrtc_params(
                 algo_desp, profile_res.arch)
 
@@ -720,9 +731,20 @@ class SimpleConv:
                 assert mask_width > 0
                 mask_width_valid = mask_width % desp.tile_shape[2] == 0
             if desp.supported_ldx_conv(ldi, ldw, ldo) and mask_width_valid:
-                if arch not in COMPILED_CUDA_GEMM_ARCHS:
+                if desp.is_nvrtc:
+                    if not CompileInfo.algo_can_be_nvrtc_compiled(desp.min_arch):
+                        continue
+                if not CompileInfo.arch_is_compiled_gemm(arch):
+                    # use PTX of possible
+                    if not CompileInfo.gemm_algo_can_use_ptx(desp.min_arch, arch):
+                        if CompileInfo.algo_can_be_nvrtc_compiled(desp.min_arch):
+                            # compiled kernel can't use PTX, for example, desp need at least sm_80 and only sm_75+PTX is compiled
+                            # all sm_80 code of this desp is invalid, we must use nvrtc.
+                            # only desp <= sm_75 can use virtual PTX code to generate sm_80 code.
                             desp = desp.copy()
                             desp.is_nvrtc = True
+                        else:
+                            continue
                 if SPCONV_DEBUG_NVRTC_KERNELS:
                     desp.is_nvrtc = True
                 finally_algos.append(desp)
@@ -826,7 +848,7 @@ class SimpleConv:
         for desp in avail:
             # for sparse conv, ndim isn't used, so we just provide a constant value.
             params = ConvParams(NDIM_DONT_CARE, ConvOpTypeCpp(op_type.value))
-            if desp.is_nvrtc and str(desp) not in self.prebuilt_desp_names:
+            if desp.is_nvrtc or str(desp) not in self.prebuilt_desp_names:
                 params.nvrtc_params = self._cached_get_nvrtc_params(desp, arch)
 
             params.conv_algo_desp = desp
@@ -935,7 +957,7 @@ class SimpleConv:
         params = ConvParams(NDIM_DONT_CARE, ConvOpTypeCpp(op_type_value))
         is_not_static = str(
                 algo_desp) not in self.prebuilt_desp_names
-        if force_nvrtc or (algo_desp.is_nvrtc and is_not_static):
+        if force_nvrtc or algo_desp.is_nvrtc or is_not_static:
             params.nvrtc_params = self._cached_get_nvrtc_params(
                 algo_desp, profile_res.arch)
         params.conv_algo_desp = profile_res.algo_desp

spconv/core_cc/csrc/sparse/convops/__init__.pyi

@@ -3,7 +3,6 @@ from pccm.stubs import EnumValue, EnumClassValue
 from cumm.tensorview.gemm import GemmAlgoDesp
 from cumm.tensorview.gemm import ConvAlgoDesp
 from cumm.tensorview import Tensor
-from ...csrc.sparse.convops import ExternalSpconvMatmul
 class GemmTuneResult:
     algo_desp: GemmAlgoDesp
     arch: Tuple[int, int]

spconv/core_cc/cumm/common.pyi

 from typing import overload, Any, Callable, Dict, List, Optional, Set, Tuple, Type, Union
 from pccm.stubs import EnumValue, EnumClassValue
 class CompileInfo:
+    @staticmethod
+    def get_compiled_cuda_version() -> Tuple[int, int]: ...
     @staticmethod
     def get_compiled_cuda_arch() -> List[Tuple[int, int]]: ...
     @staticmethod
@@ -19,3 +21,40 @@ class CompileInfo:
             arch: 
         """
         ...
+    @staticmethod
+    def arch_is_compatible(arch: Tuple[int, int]) -> bool: 
+        """
+        Args:
+            arch: 
+        """
+        ...
+    @staticmethod
+    def arch_is_compatible_gemm(arch: Tuple[int, int]) -> bool: 
+        """
+        Args:
+            arch: 
+        """
+        ...
+    @staticmethod
+    def algo_can_use_ptx(min_arch: Tuple[int, int], arch: Tuple[int, int]) -> bool: 
+        """
+        Args:
+            min_arch: 
+            arch: 
+        """
+        ...
+    @staticmethod
+    def gemm_algo_can_use_ptx(min_arch: Tuple[int, int], arch: Tuple[int, int]) -> bool: 
+        """
+        Args:
+            min_arch: 
+            arch: 
+        """
+        ...
+    @staticmethod
+    def algo_can_be_nvrtc_compiled(min_arch: Tuple[int, int]) -> bool: 
+        """
+        Args:
+            min_arch: 
+        """
+        ...

spconv/cppconstants.py

@@ -14,13 +14,13 @@
 
 import spconv.core_cc as _ext
 from spconv.core_cc.csrc.sparse.all import SpconvOps
+from spconv.core_cc.csrc.utils.boxops import BoxOps
+from spconv.core_cc.cumm.common import CompileInfo
 
 CPU_ONLY_BUILD = SpconvOps.is_cpu_only_build()
 
 BUILD_CUMM_VERSION = SpconvOps.cumm_version()
 BUILD_PCCM_VERSION = SpconvOps.pccm_version()
-from spconv.core_cc.csrc.utils.boxops import BoxOps
-from spconv.core_cc.cumm.common import CompileInfo
 HAS_BOOST = BoxOps.has_boost()
 
 COMPILED_CUDA_ARCHS = set(CompileInfo.get_compiled_cuda_arch())

spconv/csrc/sparse/all.py

@@ -131,6 +131,8 @@ class SpconvOps(pccm.Class):
         define_str = "\n".join(defines)
         self.add_global_code(define_str)
         self.build_meta.add_global_cflags("cl", "/DNOMINMAX")
+        # self.build_meta.add_global_cflags("nvcc", "-w")
+
         # for name in dir(AllocKeys):
         #     if not name.startswith("__"):
         #         v = getattr(AllocKeys, name)

spconv/csrc/sparse/convops.py

@@ -550,7 +550,6 @@ class GemmTunerSimple(pccm.ParameterizedClass):
         }}
         // auto avail_algos = get_available_algo_str_from_arch(arch);
         std::vector<tv::gemm::GemmAlgoDesp> finally_algos;
-        auto is_arch_compiled = CompileInfo::arch_is_compiled_gemm(arch);
         static_key_t static_key = std::make_tuple(trans_a, trans_b, trans_c, int(a.dtype()),
             int(b.dtype()), int(c.dtype()), shuffle_type);
         if (static_key_to_desps_.find(static_key) == static_key_to_desps_.end()){{
@@ -574,14 +573,23 @@ class GemmTunerSimple(pccm.ParameterizedClass):
             auto ldb = b.stride(0);
             auto ldc = c.stride(0);
             if (desp.supported_ldx(lda, ldb, ldc)){{
-                if (!is_arch_compiled){{
+                if (desp.is_nvrtc){{
+                    if (!CompileInfo::algo_can_be_nvrtc_compiled(desp.min_arch)){{
+                        continue;
+                    }}
+                }}
+                if (!CompileInfo::arch_is_compiled_gemm(arch)){{
+                    if (!CompileInfo::gemm_algo_can_use_ptx(desp.min_arch, arch)){{
+                        if (CompileInfo::algo_can_be_nvrtc_compiled(desp.min_arch)){{
                             auto desp2 = desp;
                             desp2.is_nvrtc = true;
-                    finally_algos.push_back(desp2);
                         }}else{{
-                    finally_algos.push_back(desp);
+                            continue;
+                        }}
                     }}
                 }}
+                finally_algos.push_back(desp);
+            }}
         }}
         return finally_algos;
         """)
@@ -699,7 +707,7 @@ class GemmTunerSimple(pccm.ParameterizedClass):
 
         for (auto& desp : avail){{
             tv::gemm::GemmParams params;
-            if (desp.is_nvrtc && prebuilt_names_.find(desp.__repr__()) == prebuilt_names_.end()){{
+            if (desp.is_nvrtc || prebuilt_names_.find(desp.__repr__()) == prebuilt_names_.end()){{
                 params.nvrtc_params = cached_get_nvrtc_params(desp, arch, stream_int);
             }}
             params.a = a;
@@ -865,7 +873,7 @@ class GemmTunerSimple(pccm.ParameterizedClass):
 
         tv::gemm::GemmParams params;
         bool desp_is_static = prebuilt_names_.find(desp.__repr__()) == prebuilt_names_.end();
-        if (force_nvrtc || (desp.is_nvrtc && desp_is_static)){{
+        if (force_nvrtc || (desp.is_nvrtc || desp_is_static)){{
             params.nvrtc_params = cached_get_nvrtc_params(desp, profile_res.arch, stream_int);
         }}
         params.a = a;
@@ -1008,7 +1016,6 @@ class ConvTunerSimple(pccm.ParameterizedClass):
         use_f32_as_accum = false;
 
         std::vector<tv::gemm::ConvAlgoDesp> finally_algos;
-        auto is_arch_compiled = CompileInfo::arch_is_compiled_gemm(arch);
         static_key_t static_key = std::make_tuple(
             layout_i, layout_w, layout_o,
             interleave_i, interleave_w, interleave_o, inp.dtype(),
@@ -1053,14 +1060,23 @@ class ConvTunerSimple(pccm.ParameterizedClass):
                 mask_width_valid = mask_width % desp.tile_shape[2] == 0;
             }}
             if (desp.supported_ldx_conv(ldi, ldw, ldo) && mask_width_valid){{
-                if (!is_arch_compiled){{
+                if (desp.is_nvrtc){{
+                    if (!CompileInfo::algo_can_be_nvrtc_compiled(desp.min_arch)){{
+                        continue;
+                    }}
+                }}
+                if (!CompileInfo::arch_is_compiled_gemm(arch)){{
+                    if (!CompileInfo::gemm_algo_can_use_ptx(desp.min_arch, arch)){{
+                        if (CompileInfo::algo_can_be_nvrtc_compiled(desp.min_arch)){{
                             auto desp2 = desp;
                             desp2.is_nvrtc = true;
-                    finally_algos.push_back(desp2);
                         }}else{{
-                    finally_algos.push_back(desp);
+                            continue;
+                        }}
                     }}
                 }}
+                finally_algos.push_back(desp);
+            }}
         }}
         return finally_algos;
         """)
@@ -1134,7 +1150,7 @@ class ConvTunerSimple(pccm.ParameterizedClass):
         tv::gemm::ConvOpType op_type_cpp = static_cast<tv::gemm::ConvOpType>(op_type);
         for (auto& desp : avail){{
             tv::gemm::ConvParams params({NDIM_DONT_CARE}, op_type_cpp, tv::CUDAKernelTimer(false));
-            if (desp.is_nvrtc && prebuilt_names_.find(desp.__repr__()) == prebuilt_names_.end()){{
+            if (desp.is_nvrtc || prebuilt_names_.find(desp.__repr__()) == prebuilt_names_.end()){{
                 params.nvrtc_params = cached_get_nvrtc_params(desp, arch, stream_int);
             }}
             params.conv_algo_desp = desp;
@@ -1311,7 +1327,7 @@ class ConvTunerSimple(pccm.ParameterizedClass):
         auto arch = profile_res.arch;
         tv::gemm::ConvParams params({NDIM_DONT_CARE}, op_type_cpp, timer);
         bool desp_is_static = prebuilt_names_.find(desp.__repr__()) == prebuilt_names_.end();
-        if (force_nvrtc || (desp.is_nvrtc && desp_is_static)){{
+        if (force_nvrtc || (desp.is_nvrtc || desp_is_static)){{
             params.nvrtc_params = cached_get_nvrtc_params(desp, arch, stream_int);
         }}
         params.conv_algo_desp = desp;

spconv/pytorch/cppcore.py

@@ -20,6 +20,8 @@ from spconv.cppconstants import COMPILED_CUDA_ARCHS
 import sys
 from spconv.core_cc.csrc.sparse.alloc import ExternalAllocator
 from spconv.core_cc.csrc.sparse.convops import ExternalSpconvMatmul
+from spconv.core_cc.cumm.common import CompileInfo
+import warnings
 
 import numpy as np
 
@@ -93,12 +95,11 @@ def get_current_stream():
 
 def get_arch():
     arch = torch.cuda.get_device_capability()
-    if arch not in COMPILED_CUDA_ARCHS:
-        print(
+    if not CompileInfo.arch_is_compatible(arch) and not CompileInfo.algo_can_use_ptx((0, 0), arch):
+        warnings.warn(
             f"[WARNING]your gpu arch {arch} isn't compiled in prebuilt, "
-            f"may cause invalid device function. "
-            f"available: {COMPILED_CUDA_ARCHS}",
-            file=sys.stderr)
+            f"may cause invalid device function error. "
+            f"available: {COMPILED_CUDA_ARCHS}")
     return arch
 
 

test/dev.py

 import spconv 
 
 
+from spconv.core_cc.cumm.common import CompileInfo
+if __name__ == "__main__":
+    print(CompileInfo.arch_is_compatible_gemm((9, 0)), CompileInfo.arch_is_compiled_gemm((9, 0)))
+    print(CompileInfo.arch_is_compatible_gemm((8, 6)), CompileInfo.arch_is_compiled_gemm((8, 6)))
\ No newline at end of file

version.txt

-2.2.0
+2.2.1