sync quantization code

example/mnist/mnist_qat.py

@@ -13,25 +13,37 @@
 # limitations under the License.
 
 from __future__ import print_function
+
 import argparse
+import contextlib
+import copy
+from typing import Dict, Optional
+
 import torch
-import spconv.pytorch as spconv
+import torch.ao.quantization
+import torch.ao.quantization.quantize_fx as qfx
+import torch.cuda.amp
+import torch.fx
 import torch.nn as nn
 import torch.nn.functional as F
 import torch.optim as optim
-from torchvision import datasets, transforms
+from torch.ao.quantization import (DeQuantStub, QuantStub,
+                                   get_default_qconfig_mapping)
+from torch.ao.quantization.fx._lower_to_native_backend import \
+    STATIC_LOWER_FUSED_MODULE_MAP, STATIC_LOWER_MODULE_MAP
 from torch.optim.lr_scheduler import StepLR
-import contextlib
-import torch.cuda.amp
-import torch.ao.quantization
-from torch.ao.quantization import QuantStub, DeQuantStub
-import torch.ao.quantization.quantize_fx as qfx
-from spconv.pytorch.quantization.fake_q import get_default_spconv_qconfig_mapping
+from torchvision import datasets, transforms
+
+import spconv.pytorch as spconv
 import spconv.pytorch.quantization as spconvq
 from spconv.pytorch.quantization import get_default_spconv_trt_ptq_qconfig
-from torch.ao.quantization import get_default_qconfig_mapping
-from spconv.pytorch.quantization.backend_cfg import SPCONV_STATIC_LOWER_FUSED_MODULE_MAP
-from torch.ao.quantization.fx._lower_to_native_backend import STATIC_LOWER_FUSED_MODULE_MAP
+from spconv.pytorch.quantization.backend_cfg import \
+    SPCONV_STATIC_LOWER_FUSED_MODULE_MAP, SPCONV_STATIC_LOWER_MODULE_MAP
+from spconv.pytorch.quantization.core import quantize_per_tensor
+from spconv.pytorch.quantization.fake_q import \
+    get_default_spconv_qconfig_mapping
+from spconv.pytorch.quantization.intrinsic.modules import SpconvBnAddReLUNd, SpconvAddReLUNd
+import spconv.pytorch.quantization.intrinsic.quantized as snniq
 
 @contextlib.contextmanager
 def identity_ctx():
@@ -57,6 +69,142 @@ class SparseConvBNReLU(spconv.SparseSequential):
             nn.ReLU(inplace=False)
         )
 
+class SparseBasicBlock(spconv.SparseModule):
+    """residual block that supported by spconv quantization.
+    """
+    expansion = 1
+    def __init__(self,
+                 in_planes, out_planes,
+                 stride=1,
+                 downsample=None):
+        spconv.SparseModule.__init__(self)
+        conv1 = spconv.SubMConv2d(in_planes, out_planes, 3, stride, 1, bias=False)
+        conv2 = spconv.SubMConv2d(out_planes, out_planes, 3, stride, 1, bias=False)
+
+        norm1 = nn.BatchNorm1d(out_planes, momentum=0.1)
+        norm2 = nn.BatchNorm1d(out_planes, momentum=0.1)
+
+        self.conv1_bn_relu = spconv.SparseSequential(conv=conv1, bn=norm1, relu=nn.ReLU(inplace=True))
+        self.conv2_bn = spconv.SparseSequential(conv=conv2, bn=norm2)
+        self.relu = nn.ReLU(inplace=True)
+        self.downsample = downsample
+        self.iden_for_fx_match = nn.Identity()
+
+    def forward(self, x: spconv.SparseConvTensor):
+        identity = self.iden_for_fx_match(x.features)
+        # if self.training:
+        #     assert x.features.dim() == 2, f'x.features.dim()={x.features.dim()}'
+
+        out = self.conv1_bn_relu(x)
+        out = self.conv2_bn(out)
+
+        if self.downsample is not None:
+            identity = self.downsample(x)
+
+        out = out.replace_feature(self.relu(out.features + identity))
+        return out
+
+class SparseBasicBlock1(spconv.SparseModule):
+    """residual block that supported by spconv quantization.
+    """
+    expansion = 1
+    def __init__(self,
+                 in_planes, out_planes,
+                 stride=1,
+                 downsample=None):
+        spconv.SparseModule.__init__(self)
+        self.conv1 = spconv.SubMConv2d(in_planes, out_planes, 3, stride, 1, bias=False)
+        self.conv2 = spconv.SubMConv2d(out_planes, out_planes, 3, stride, 1, bias=False)
+
+        self.norm1 = nn.BatchNorm1d(out_planes, momentum=0.1)
+        self.norm2 = nn.BatchNorm1d(out_planes, momentum=0.1)
+
+        self.relu1 = nn.ReLU(inplace=True)
+        self.relu2 = nn.ReLU(inplace=True)
+        self.downsample = downsample
+        self.iden_for_fx_match = nn.Identity()
+
+    def forward(self, x: spconv.SparseConvTensor):
+        identity = self.iden_for_fx_match(x.features)
+        # if self.training:
+        #     assert x.features.dim() == 2, f'x.features.dim()={x.features.dim()}'
+        out = self.conv1(x)
+        out = out.replace_feature(self.relu1(self.norm1(out.features)))
+        out = self.conv2(out)
+        out = out.replace_feature(self.norm2(out.features))
+
+        # if self.downsample is not None:
+        #     identity = self.downsample(x)
+
+        out = out.replace_feature(self.relu2(out.features + identity))
+        return out
+
+class SparseBasicBlock2(spconv.SparseModule):
+    """residual block that supported by spconv quantization.
+    """
+    expansion = 1
+    def __init__(self,
+                 in_planes, out_planes,
+                 stride=1,
+                 downsample=None):
+        spconv.SparseModule.__init__(self)
+        self.conv1 = spconv.SubMConv2d(in_planes, out_planes, 3, stride, 1, bias=False)
+        self.conv2 = spconv.SubMConv2d(out_planes, out_planes, 3, stride, 1, bias=False)
+
+        self.norm1 = spconv.SparseBatchNorm(out_planes, momentum=0.1)
+        self.norm2 = spconv.SparseBatchNorm(out_planes, momentum=0.1)
+
+        self.relu1 = spconv.SparseReLU(inplace=True)
+        self.relu2 = spconv.SparseReLU(inplace=True)
+        self.downsample = downsample
+        self.iden_for_fx_match = spconv.SparseIdentity()
+
+    def forward(self, x: spconv.SparseConvTensor):
+        identity = self.iden_for_fx_match(x)
+        # if self.training:
+        #     assert x.features.dim() == 2, f'x.features.dim()={x.features.dim()}'
+        out = self.conv1(x)
+        out = self.relu1(self.norm1(out))
+        out = self.conv2(out)
+        out = self.norm2(out)
+
+        if self.downsample is not None:
+            identity = self.downsample(x)
+        out = self.relu2(out + identity)
+        return out
+
+class SparseBasicBlock3(spconv.SparseModule):
+    """residual block that supported by spconv quantization.
+    """
+    expansion = 1
+    def __init__(self,
+                 in_planes, out_planes,
+                 stride=1,
+                 downsample=None):
+        spconv.SparseModule.__init__(self)
+        self.conv1 = spconv.SubMConv2d(in_planes, out_planes, 3, stride, 1, bias=False)
+        conv2 = spconv.SubMConv2d(out_planes, out_planes, 3, stride, 1, bias=False)
+
+        self.norm1 = spconv.SparseBatchNorm(out_planes, momentum=0.1)
+        norm2 = spconv.SparseBatchNorm(out_planes, momentum=0.1)
+        self.residual_conv = SpconvAddReLUNd(conv2, spconv.SparseReLU(inplace=True))
+        self.relu1 = spconv.SparseReLU(inplace=True)
+        # self.relu2 = spconv.SparseReLU(inplace=True)
+        self.downsample = downsample
+        self.iden_for_fx_match = spconv.SparseIdentity()
+
+    def forward(self, x: spconv.SparseConvTensor):
+        identity = self.iden_for_fx_match(x)
+        # if self.training:
+        #     assert x.features.dim() == 2, f'x.features.dim()={x.features.dim()}'
+        out = self.conv1(x)
+        out = self.relu1(self.norm1(out))
+
+        if self.downsample is not None:
+            identity = self.downsample(x)
+        out = self.residual_conv(out, identity)
+        return out
+
 class Net(nn.Module):
     def __init__(self):
         super(Net, self).__init__()
@@ -126,7 +274,7 @@ class NetV2(nn.Module):
 
 class NetPTQ(nn.Module):
     """pytorch currently don't support cuda int8 inference, so
-    we only use sparse ops here.
+    we build a pure sparse network here.
     """
     def __init__(self):
         super(NetPTQ, self).__init__()
@@ -138,7 +286,6 @@ class NetPTQ(nn.Module):
             SparseConvBNReLU(64, 64, 3, 2, 1), # 4x4
             spconv.SparseConv2d(64, 10, 4, 4),
             spconv.ToDense(),
-
         )
         # self.fc1 = nn.Linear(64 * 1 * 1, 128)
         # self.fc2 = nn.Linear(128, 10)
@@ -158,22 +305,47 @@ class NetPTQ(nn.Module):
         # print(x_sp.shape)
         x = x_sp
         x = torch.flatten(x, 1)
-
-        # x_res = torch.zeros_like(x)
-        # x_res[x_sp.indices[:, 0].long()] = x
-        # x = x_res
-        # x = torch.flatten(x, 1)
-        # x = self.dropout1(x)
-        # x = self.fc1(x)
-        # x = F.relu(x)
-        # x = self.dropout2(x)
-        # x = self.fc2(x)
-
-        # print(x_sp.features.shape, x_sp.spatial_shape)
         x = self.dequant(x)
         output = F.log_softmax(x, dim=1)
         return output
 
+class ResidualNetPTQ(nn.Module):
+    """pytorch currently don't support cuda int8 inference, so
+    we build a pure sparse network here.
+    """
+    def __init__(self):
+        super(ResidualNetPTQ, self).__init__()
+        self.net = spconv.SparseSequential(
+            SubMConvBNReLU(1, 32, 3),
+            SparseBasicBlock2(32, 32),
+            SubMConvBNReLU(32, 64, 3),
+            SparseConvBNReLU(64, 64, 2, 2), # 14x14
+            SparseConvBNReLU(64, 64, 2, 2), # 7x7
+            SparseConvBNReLU(64, 64, 3, 2, 1), # 4x4
+            spconv.SparseConv2d(64, 10, 4, 4),
+            spconv.ToDense(),
+        )
+        # self.fc1 = nn.Linear(64 * 1 * 1, 128)
+        # self.fc2 = nn.Linear(128, 10)
+        # self.dropout1 = nn.Dropout2d(0.25)
+        # self.dropout2 = nn.Dropout2d(0.5)
+
+        self.quant = QuantStub()
+        self.dequant = DeQuantStub()
+    
+    def forward(self, features: torch.Tensor, indices: torch.Tensor, batch_size: int):
+        # x: [N, 28, 28, 1], must be NHWC tensor
+        features = self.quant(features)
+        # x_sp = spconv.SparseConvTensor.from_dense(x.reshape(-1, 28, 28, 1))
+        x_sp = spconv.SparseConvTensor(features, indices, [28, 28], batch_size)
+        # create SparseConvTensor manually: see SparseConvTensor.from_dense
+        x_sp = self.net(x_sp)
+        # print(x_sp.shape)
+        x = x_sp
+        x = torch.flatten(x, 1)
+        x = self.dequant(x)
+        output = F.log_softmax(x, dim=1)
+        return output
 
 class NetDense(nn.Module):
     def __init__(self):
@@ -184,6 +356,8 @@ class NetDense(nn.Module):
         self.dropout2 = nn.Dropout(0.5)
         self.fc1 = nn.Linear(9216, 128)
         self.fc2 = nn.Linear(128, 10)
+        self.iden = spconv.SparseIdentity()
+
         self.quant = QuantStub()
         self.dequant = DeQuantStub()
 
@@ -195,6 +369,7 @@ class NetDense(nn.Module):
         x = F.relu(x)
         x = self.conv2(x)
         x = F.relu(x)
+        x = self.iden(x)
         x = F.max_pool2d(x, 2)
         x = self.dropout1(x)
         x = torch.flatten(x, 1)
@@ -299,6 +474,54 @@ def calibrate(args, model: torch.nn.Module, data_loader, device):
             else:
                 output = model(image)
 
+def transform_qdq(m: torch.fx.GraphModule) -> torch.fx.GraphModule:
+    """torch.quantize_per_tensor don't support SparseConvTensor, so we
+    use a custom one by fx transform.
+    """
+    for node in m.graph.nodes:
+        # Checks if we're calling a function (i.e:
+        # torch.add)
+        if node.op == 'call_function':
+            # The target attribute is the function
+            # that call_function calls.
+            if node.target == torch.quantize_per_tensor:
+                node.target = quantize_per_tensor
+    m.graph.lint() # Does some checks to make sure the
+                 # Graph is well-formed.
+    m.recompile()
+    return m
+
+
+def is_dequantize_node(node):
+    return isinstance(node, torch.fx.Node) and node.op == "call_method" and node.target == "dequantize"
+
+def _get_module(node: torch.fx.Node, modules: Dict[str, nn.Module]) -> Optional[nn.Module]:
+    """
+    Return the `torch.nn.Module` that corresponds to the specified node's target.
+    If no such node exists, return None.
+    """
+    if node.op == "call_module" and str(node.target) in modules:
+        return modules[str(node.target)]
+    else:
+        return None
+
+def remove_conv_add_dq(model: torch.fx.graph_module.GraphModule):
+    modules = dict(model.named_modules(remove_duplicate=False))
+    for n in model.graph.nodes:
+        if (n.op == "call_module" and type(_get_module(n, modules)) == snniq.SparseConvAddReLU):
+            # check second input, if it's dequantized, remove that dequantize node
+            arg1 = n.args[1]
+            if is_dequantize_node(arg1):
+                dq_node = arg1
+                assert(isinstance(dq_node, torch.fx.Node))
+                dn_input = dq_node.args[0]
+                n.replace_input_with(dq_node, dn_input)
+    model.graph.eliminate_dead_code()
+    model.recompile()
+    model.graph.lint() # Does some checks to make sure the
+                 # Graph is well-formed.
+    return model
+
 def main():
     # Training settings
     parser = argparse.ArgumentParser(description='PyTorch MNIST Example')
@@ -361,11 +584,11 @@ def main():
 
     torch.manual_seed(args.seed)
 
-    device = torch.device("cuda" if use_cuda else "cpu")
+    device = torch.device("cuda" if use_cuda and args.sparse else "cpu")
     qdevice = torch.device("cuda" if use_cuda and args.sparse else "cpu")
     kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}
     if args.sparse:
-        model = NetPTQ().to(device)
+        model = ResidualNetPTQ().to(device)
     else:
         model = NetDense().to(device)
 
@@ -401,42 +624,61 @@ def main():
         train(args, model, device, train_loader, optimizer, epoch)
         test(args, model, device, test_loader)
         scheduler.step()
-    # if args.save_model:
-    #     torch.save(model.state_dict(), "mnist_cnn.pt")
+    if args.save_model:
+        torch.save(model.state_dict(), "mnist_cnn.pt")
     model.eval()
-    STATIC_LOWER_FUSED_MODULE_MAP.update(SPCONV_STATIC_LOWER_FUSED_MODULE_MAP)
     if not args.sparse:
         model = model.cpu()
-    # qconfig_mapping_default = get_default_qconfig_mapping("x86")
+
+    model_qat = copy.deepcopy(model)
+    STATIC_LOWER_FUSED_MODULE_MAP.update(SPCONV_STATIC_LOWER_FUSED_MODULE_MAP)
+    STATIC_LOWER_MODULE_MAP.update(SPCONV_STATIC_LOWER_MODULE_MAP)
+
+    # tensorrt only support symmetric quantization, per-tensor act and per-channel weight.
     qconfig_mapping = get_default_spconv_qconfig_mapping(False)
     prepare_cfg = spconvq.get_spconv_prepare_custom_config()
     backend_cfg = spconvq.get_spconv_backend_config()
-    convert_cfg = spconvq.get_spconv_convert_custom_config()
+    # convert_cfg = spconvq.get_spconv_convert_custom_config()
     # prepare: fuse your model, all patterns such as conv-bn-relu fuse to modules in torch.ao.quantization.intrinsic / spconv.pytorch.quantization.intrinsic
     # then add observers to fused model.
     prepared_model = qfx.prepare_fx(model, qconfig_mapping, (), backend_config=backend_cfg, prepare_custom_config=prepare_cfg)
-    # prepared_model.print_readable()
-    print([type(m) for m in prepared_model.modules()])
-    print(prepared_model)
+    # print(prepared_model)
+    # breakpoint()
+
+    # print(prepared_model)
     # calibrate: run model with some inputs
-    calibrate(args, prepared_model, test_loader, qdevice)
+    # calibrate(args, prepared_model, test_loader, qdevice)
     # convert (ptq): replace intrinsic blocks with quantized modules
+    converted_model = qfx.convert_fx(prepared_model, qconfig_mapping=qconfig_mapping, backend_config=backend_cfg)
+    converted_model = transform_qdq(converted_model)
+    # test converted ptq model with int8 kernel
+    remove_conv_add_dq(converted_model)
 
-    converted_model = qfx.convert_to_reference_fx(prepared_model, convert_cfg, qconfig_mapping=qconfig_mapping, backend_config=backend_cfg)
-    print([type(m) for m in converted_model.modules()])
-    # tensorrt only support symmetric quantization, per-tensor act and per-channel weight.
-    # model.qconfig = get_default_spconv_trt_ptq_qconfig()
-    # prepare_custom_config_dict = spconvq.get_prepare_custom_config()
-    # convert_custom_config_dict = spconvq.get_convert_custom_config()
-    # torch.ao.quantization.prepare(model, inplace=True)
-    # print('Post Training Quantization Prepare: Inserting Observers')
-    # print('\n ConvBnReLUBlock:After observer insertion \n\n', model.net[0])
-    # test(args, model, device, test_loader)
     print(converted_model)
+    breakpoint()
 
     test(args, converted_model, qdevice, test_loader)
+    # do qat
+    # qconfig_mapping_qat = get_default_spconv_qconfig_mapping(True)
+    # prepared_model_qat = qfx.prepare_qat_fx(model_qat, qconfig_mapping_qat, (), backend_config=backend_cfg, prepare_custom_config=prepare_cfg)
+    # # converted_model = qfx.convert_fx(prepared_model_qat, qconfig_mapping=qconfig_mapping_qat, backend_config=backend_cfg)
+    # # breakpoint()
+    # print(prepared_model_qat)
+    # train(args, prepared_model_qat, qdevice, train_loader, optimizer, 1)
+    # converted_model = qfx.convert_fx(prepared_model_qat, qconfig_mapping=qconfig_mapping_qat, backend_config=backend_cfg)
+    # converted_model = transform_qdq(converted_model)
+    # test(args, converted_model, qdevice, test_loader)
+    # # [type(m) for m in prepared_model_qat.modules()]
+    # # model.qconfig = get_default_spconv_trt_ptq_qconfig()
+    # # prepare_custom_config_dict = spconvq.get_prepare_custom_config()
+    # # convert_custom_config_dict = spconvq.get_convert_custom_config()
+    # # torch.ao.quantization.prepare(model, inplace=True)
+    # # print('Post Training Quantization Prepare: Inserting Observers')
+    # # print('\n ConvBnReLUBlock:After observer insertion \n\n', model.net[0])
+    # # test(args, model, device, test_loader)
+    # print(converted_model)
+    # you will see some nvrtc compile log here, which means int8 kernel is used.
     breakpoint()
 
-
 if __name__ == '__main__':
     main()

spconv/algo.py

@@ -188,10 +188,16 @@ class ConvTunerSimple(ConvTunerSimpleBase):
             cudadevrt_p = get_cudadevrt_path()
             assert cudadevrt_p is not None, "DynamicParallism must have cudadevrt"
             cudadevrt = str(cudadevrt_p)
+        # mod = CummNVRTCModule([kernel],
+        #                       cudadevrt_path=cudadevrt,
+        #                       verbose=True,
+        #                       custom_names=custom_names,
+        #                       verbose_path="/home/yy/Projects/spconv-release/spconv/build/dev_nvrtc_int8")
         mod = CummNVRTCModule([kernel],
                               cudadevrt_path=cudadevrt,
                               verbose=False,
                               custom_names=custom_names)
+
         mod.load()
         return mod, kernel
 

spconv/build.py

@@ -18,10 +18,10 @@ from typing import List
 import pccm
 from pccm.utils import project_is_editable, project_is_installed
 from ccimport.compat import InWindows
-from .constants import PACKAGE_NAME, PACKAGE_ROOT, DISABLE_JIT
+from .constants import PACKAGE_NAME, PACKAGE_ROOT, DISABLE_JIT, SPCONV_INT8_DEBUG
 
 if project_is_installed(PACKAGE_NAME) and project_is_editable(
-        PACKAGE_NAME) and not DISABLE_JIT and False:
+        PACKAGE_NAME) and not DISABLE_JIT and not SPCONV_INT8_DEBUG:
     from spconv.core import SHUFFLE_SIMT_PARAMS, SHUFFLE_VOLTA_PARAMS, SHUFFLE_TURING_PARAMS, SHUFFLE_AMPERE_PARAMS
     from spconv.core import IMPLGEMM_SIMT_PARAMS, IMPLGEMM_VOLTA_PARAMS, IMPLGEMM_TURING_PARAMS, IMPLGEMM_AMPERE_PARAMS
 

spconv/constants.py

@@ -116,3 +116,5 @@ SPCONV_DIRECT_TABLE_HASH_SIZE_SCALE = 1.1
 
 SPCONV_ALLOW_TF32 = False
 
+
+SPCONV_INT8_DEBUG = False
\ No newline at end of file

spconv/pytorch/__init__.py

@@ -14,7 +14,8 @@ from spconv.pytorch.conv import (SparseConv1d, SparseConv2d, SparseConv3d,
                                  SubMConv3d, SubMConv4d)
 from spconv.pytorch.identity import Identity
 from spconv.pytorch.modules import (SparseModule, SparseSequential,
-                                    assign_name_for_sparse_modules)
+                                    assign_name_for_sparse_modules, SparseBatchNorm,
+                                    SparseReLU, SparseIdentity)
 from spconv.pytorch.ops import ConvAlgo
 from spconv.pytorch.pool import (SparseMaxPool1d, SparseMaxPool2d,
                                  SparseMaxPool3d, SparseMaxPool4d,

spconv/pytorch/core.py

@@ -233,6 +233,9 @@ class SparseConvTensor(metaclass=SpConvTensorMeta):
         features_th = x_sp.values()
         return cls(features_th, indices_th, spatial_shape, batch_size)
 
+    def dequantize(self):
+        return self.replace_feature(self.features.dequantize())
+
     @property
     def spatial_size(self):
         return np.prod(self.spatial_shape)
@@ -264,6 +267,19 @@ class SparseConvTensor(metaclass=SpConvTensorMeta):
     #     return self.indices.shape[0] / np.prod(
     #         self.spatial_shape) / self.batch_size
 
+    def __add__(self, other: "SparseConvTensor"):
+        assert isinstance(other, SparseConvTensor)
+        return self.replace_feature(self.features + other.features)
+
+    def __iadd__(self, other: "SparseConvTensor"):
+        assert isinstance(other, SparseConvTensor)
+        self.features += other.features
+        return self
+        
+    def __radd__(self, other: "SparseConvTensor"):
+        assert isinstance(other, SparseConvTensor)
+        return other.replace_feature(self.features + other.features)
+
     def shadow_copy(self) -> "SparseConvTensor":
         """create a new spconv tensor with all member unchanged"""
         tensor = SparseConvTensor(self.features, self.indices,

spconv/pytorch/modules.py

@@ -23,7 +23,7 @@ from spconv import pytorch as spconv
 
 
 def is_spconv_module(module):
-    spconv_modules = (SparseModule, )
+    spconv_modules = (SparseModule, SparseBatchNorm, SparseReLU)
     return isinstance(module, spconv_modules)
 
 
@@ -148,3 +148,37 @@ def assign_name_for_sparse_modules(module: nn.Module):
     for k, n in module.named_modules():
         if isinstance(n, SparseModule):
             n._sparse_unique_name = k
+
+
+class SparseBatchNorm(nn.BatchNorm1d):
+    """this module is exists only for torch.fx transformation for quantization.
+    """
+    def forward(self, input):
+        if isinstance(input, spconv.SparseConvTensor):
+            return input.replace_feature(super().forward(input.features))
+        return super().forward(input)
+
+class SparseSyncBatchNorm(nn.SyncBatchNorm):
+    """this module is exists only for torch.fx transformation for quantization.
+    """
+    def forward(self, input):
+        if isinstance(input, spconv.SparseConvTensor):
+            return input.replace_feature(super().forward(input.features))
+        return super().forward(input)
+
+class SparseReLU(nn.ReLU):
+    """this module is exists only for torch.fx transformation for quantization.
+    """
+    def forward(self, input):
+        if isinstance(input, spconv.SparseConvTensor):
+            return input.replace_feature(super().forward(input.features))
+        return super().forward(input)
+
+
+class SparseIdentity(nn.Identity):
+    """this module is exists only for torch.fx transformation for quantization.
+    """
+    def forward(self, input):
+        if isinstance(input, spconv.SparseConvTensor):
+            return input.replace_feature(super().forward(input.features))
+        return super().forward(input)

spconv/pytorch/ops.py

@@ -1462,14 +1462,14 @@ def implicit_gemm(features: torch.Tensor,
                   output_scale: float = 1.0,
                   scale: Optional[torch.Tensor] = None,
                   output_add: Optional[torch.Tensor] = None,
-                  output_add_scale: float = 1.0,
+                  output_add_scale: float = 0.0,
                   output_dtype: Optional[torch.dtype] = None):
     stream = get_current_stream()
     bias_tv = tv.Tensor()
     scale_tv = tv.Tensor()
     output_add_tv = tv.Tensor()
     if output_add is not None:
-        assert features.dtype == torch.int8, "fused residual add only support int8"
+        assert features.dtype == torch.qint8, "fused residual add only support int8"
     if bias is not None:
         bias_tv = torch_tensor_to_tv(bias)
     if scale is not None:
@@ -1485,7 +1485,7 @@ def implicit_gemm(features: torch.Tensor,
         output_dtype = features.dtype
 
     if SPCONV_CPP_GEMM and CONV_CPP is not None:
-        alloc = TorchAllocator(features.device)
+        alloc = TorchAllocator(features.device, features.dtype == torch.qint8)
         features_tv = torch_tensor_to_tv(features)
         pair_fwd_tv = torch_tensor_to_tv(pair_fwd)
         pair_mask_fwd_splits_tv = [
@@ -1963,6 +1963,12 @@ def indice_maxpool_implicit_gemm(features: torch.Tensor,
         features = features.contiguous()
 
     out_channel = features.shape[-1]
+    if features.is_quantized:
+        out_features = torch._empty_affine_quantized((num_activate_out, out_channel),
+                                scale=features.q_scale(),
+                                dtype=features.dtype,
+                                device=features.device)
+    else:
         out_features = torch.empty((num_activate_out, out_channel),
                                 dtype=features.dtype,
                                 device=features.device)
@@ -2016,9 +2022,16 @@ def indice_avgpool_implicit_gemm(features: torch.Tensor,
         features = features.contiguous()
 
     out_channel = features.shape[-1]
+    if features.is_quantized:
+        out_features = torch._empty_affine_quantized((num_activate_out, out_channel),
+                                scale=features.q_scale(),
+                                dtype=features.dtype,
+                                device=features.device)
+    else:
         out_features = torch.empty((num_activate_out, out_channel),
                                 dtype=features.dtype,
                                 device=features.device)
+
     assert features.is_cuda
     stream = get_current_stream()
     out_features_tv = torch_tensor_to_tv(out_features)

spconv/pytorch/pool.py

@@ -66,14 +66,14 @@ class SparseMaxPool(SparseModule):
         if algo is None:
             # keep in mind that this algorithm is set for Inverse Sparse Conv
             # maxpool itself don't need mask.
-            if kv <= 32 and not CPU_ONLY_BUILD:
+            if kv <= 128 and not CPU_ONLY_BUILD:
                 if kv < 8:
                     algo = ConvAlgo.MaskImplicitGemm
                 else:
                     algo = ConvAlgo.MaskImplicitGemm
             else:
                 algo = ConvAlgo.Native
-        if kv > 32:
+        if kv > 128:
             assert algo == ConvAlgo.Native, "implicit gemm don't support kv >= 32 for now"
         if CPU_ONLY_BUILD:
             assert algo == ConvAlgo.Native, "cpu only build only support native algorithm"
@@ -96,7 +96,10 @@ class SparseMaxPool(SparseModule):
         return None 
 
 
-    def forward(self, input):
+    def forward(self, input: spconv.SparseConvTensor):
+        is_int8 = input.is_quantized
+        if is_int8:
+            assert self.algo == ConvAlgo.MaskImplicitGemm, "only ConvAlgo.MaskImplicitGemm support int8."
         assert isinstance(input, spconv.SparseConvTensor)
         features = input.features
         device = features.device
@@ -296,6 +299,10 @@ class SparseAvgPool(SparseModule):
 
     def forward(self, input):
         assert isinstance(input, spconv.SparseConvTensor)
+        is_int8 = input.is_quantized
+        if is_int8:
+            assert self.algo == ConvAlgo.MaskImplicitGemm, "only ConvAlgo.MaskImplicitGemm support int8."
+
         features = input.features
         device = features.device
         indices = input.indices
@@ -534,3 +541,8 @@ class SparseAvgPool3d(SparseAvgPool):
                              algo=algo,
                              record_voxel_count=record_voxel_count,
                              name=name)
+
+
+ALL_POOL_LAYERS = set([
+    SparseAvgPool3d, SparseAvgPool2d, SparseAvgPool1d, SparseMaxPool1d, SparseMaxPool2d, SparseMaxPool3d, SparseMaxPool4d, SparseAvgPool, SparseMaxPool
+])
\ No newline at end of file

spconv/pytorch/quantization/__init__.py

@@ -19,3 +19,4 @@ from .fake_q import (get_default_spconv_trt_ptq_qconfig,
                      get_default_spconv_trt_qat_qconfig)
 from .qmapping import (get_spconv_fmod_to_qat_mapping,
                        get_spconv_qat_to_static_mapping)
+from .core import quantize_per_tensor
\ No newline at end of file

spconv/pytorch/quantization/core.py

+from typing import Union, List, Dict 
+
+import torch 
+
+from spconv.pytorch.core import SparseConvTensor
+
+def quantize_per_tensor(ten: Union[Union[SparseConvTensor, torch.Tensor], List[Union[SparseConvTensor, torch.Tensor]]], scale, zero_point, dtype):
+    if isinstance(ten, (list, tuple)):
+        res = []
+        for i, v in enumerate(ten):
+            if isinstance(v, SparseConvTensor):
+                res.append(v.replace_feature(torch.quantize_per_tensor(v.features, scale[i], zero_point[i], dtype)))
+            else:
+                res.append(torch.quantize_per_tensor(v, scale[i], zero_point[i], dtype))
+        return res 
+    else:
+        if isinstance(ten, SparseConvTensor):
+            return ten.replace_feature(torch.quantize_per_tensor(ten.features, scale, zero_point, dtype))
+        else:
+            return torch.quantize_per_tensor(ten, scale, zero_point, dtype)
+    
\ No newline at end of file

spconv/pytorch/quantization/fake_q.py

@@ -11,7 +11,7 @@ from torch.ao.quantization.observer import (HistogramObserver,
 from torch.ao.quantization.qconfig import QConfig, QConfigAny, default_reuse_input_qconfig
 from torch.ao.quantization.qconfig_mapping import QConfigMapping, _FIXED_QPARAMS_OP_TO_OBSERVER
 from typing import Any, Callable, Dict, Tuple, Union, List
-from torch.ao.quantization import get_default_qconfig
+from torch.ao.quantization import get_default_qconfig, get_default_qat_qconfig
 from spconv.pytorch.core import SparseConvTensor
 
 __all__ = ["get_default_spconv_trt_ptq_qconfig", "get_default_spconv_trt_qat_qconfig"]
@@ -80,13 +80,14 @@ def get_default_spconv_trt_ptq_qconfig(backend, version):
 def get_default_spconv_trt_qat_qconfig(backend, version):
     return default_symmetric_spconv_qat_qconfig
 
-def get_default_spconv_qconfig_mapping(is_qat: bool, backend: str = "x86", version: int = 0) -> QConfigMapping:
+def get_default_spconv_qconfig_mapping(is_qat: bool, backend: str = "fbgemm", version: int = 0) -> QConfigMapping:
     """
     From torch.ao.quantization.qconfig_mapping
     Return the default QConfigMapping for the given quantization type and backend.
     """
     # get_default_qconfig(backend, version)
     if is_qat:
+        # qconfig = get_default_qat_qconfig(backend, version)
         qconfig = get_default_spconv_trt_qat_qconfig(backend, version)
     else:
         # qconfig = QConfig(activation=HistogramObserver.with_args(reduce_range=False, dtype=torch.qint8),
@@ -144,3 +145,4 @@ def get_default_spconv_qconfig_mapping(is_qat: bool, backend: str = "x86", versi
                        .set_object_type(torch.nn.functional.tanh, qconfig)
 
     return qconfig_mapping
+

spconv/pytorch/quantization/fuse_mapping.py

+from functools import partial
 from typing import Union, Callable, Tuple, Dict, Optional, Type, Any
 import torch.nn as nn
 import spconv.pytorch as spconv
@@ -5,7 +6,8 @@ from .utils import fuse_spconv_bn_eval
 from . import intrinsic as snni
 from .intrinsic.qat.modules import SparseConvBn, SparseConvBnReLU, SparseConvBnAddReLU
 from spconv.pytorch.conv import DEFAULT_SPARSE_CONV_TYPES
-def fuse_conv_bn(is_qat, conv, bn):
+
+def fuse_conv_bn(is_qat, conv, bn, is_add_fuse: bool = False):
     r"""Given the conv and bn modules, fuses them and returns the fused module
 
     Args:
@@ -20,11 +22,10 @@ def fuse_conv_bn(is_qat, conv, bn):
     """
     assert(conv.training == bn.training),\
         "Conv and BN both must be in the same mode (train or eval)."
-    
+    fuse_cls = snni.SpconvAddReLUNd if is_add_fuse else snni.SpconvBnNd
     fused_module_class_map = {
-        k: snni.SpconvBnNd for k in DEFAULT_SPARSE_CONV_TYPES
+        k: fuse_cls for k in DEFAULT_SPARSE_CONV_TYPES
     }
-
     if is_qat:
         assert bn.num_features == conv.out_channels, 'Output channel of Conv2d must match num_features of BatchNorm2d'
         assert bn.affine, 'Only support fusing BatchNorm2d with affine set to True'
@@ -37,7 +38,7 @@ def fuse_conv_bn(is_qat, conv, bn):
     else:
         return fuse_spconv_bn_eval(conv, bn)
 
-def fuse_conv_bn_relu(is_qat, conv, bn, relu):
+def fuse_conv_bn_relu(is_qat, conv, bn, relu, is_add_fuse: bool = False):
     r"""Given the conv and bn modules, fuses them and returns the fused module
 
     Args:
@@ -54,8 +55,9 @@ def fuse_conv_bn_relu(is_qat, conv, bn, relu):
         "Conv and BN both must be in the same mode (train or eval)."
     fused_module : Optional[Type[spconv.SparseSequential]] = None
     if is_qat:
+        fuse_cls = snni.SpconvBnAddReLUNd if is_add_fuse else snni.SpconvBnReLUNd
         map_to_fused_module_train = {
-            k: snni.SpconvBnReLUNd for k in DEFAULT_SPARSE_CONV_TYPES
+            k: fuse_cls for k in DEFAULT_SPARSE_CONV_TYPES
         }
         assert bn.num_features == conv.out_channels, 'Output channel of Conv must match num_features of BatchNorm'
         assert bn.affine, 'Only support fusing BatchNorm with affine set to True'
@@ -66,8 +68,9 @@ def fuse_conv_bn_relu(is_qat, conv, bn, relu):
         else:
             raise NotImplementedError("Cannot fuse train modules: {}".format((conv, bn, relu)))
     else:
+        fuse_cls = snni.SpconvAddReLUNd if is_add_fuse else snni.SpconvReLUNd
         map_to_fused_module_eval = {
-            k: snni.SpconvReLUNd for k in DEFAULT_SPARSE_CONV_TYPES
+            k: fuse_cls for k in DEFAULT_SPARSE_CONV_TYPES
         }
         fused_module = map_to_fused_module_eval.get(type(conv), None)
         if fused_module is not None:
@@ -76,28 +79,21 @@ def fuse_conv_bn_relu(is_qat, conv, bn, relu):
         else:
             raise NotImplementedError("Cannot fuse eval modules: {}".format((conv, bn, relu)))
 
-# DEFAULT_SPCONV_OP_LIST_TO_FUSER_METHOD : Dict[Tuple, Union[nn.Sequential, Callable]] = {
-#     (spconv.SubMConv1d, nn.BatchNorm1d): fuse_conv_bn,
-#     (spconv.SubMConv1d, nn.BatchNorm1d, nn.ReLU): fuse_conv_bn_relu,
-#     (spconv.SparseConv1d, nn.BatchNorm1d): fuse_conv_bn,
-#     (spconv.SparseConv1d, nn.BatchNorm1d, nn.ReLU): fuse_conv_bn_relu,
-#     (spconv.SparseInverseConv1d, nn.BatchNorm1d): fuse_conv_bn,
-#     (spconv.SparseInverseConv1d, nn.BatchNorm1d, nn.ReLU): fuse_conv_bn_relu,
-#     (spconv.SubMConv2d, nn.BatchNorm1d): fuse_conv_bn,
-#     (spconv.SubMConv2d, nn.BatchNorm1d, nn.ReLU): fuse_conv_bn_relu,
-#     (spconv.SparseConv2d, nn.BatchNorm1d): fuse_conv_bn,
-#     (spconv.SparseConv2d, nn.BatchNorm1d, nn.ReLU): fuse_conv_bn_relu,
-#     (spconv.SparseInverseConv2d, nn.BatchNorm1d): fuse_conv_bn,
-#     (spconv.SparseInverseConv2d, nn.BatchNorm1d, nn.ReLU): fuse_conv_bn_relu,
-#     (spconv.SubMConv3d, nn.BatchNorm1d): fuse_conv_bn,
-#     (spconv.SubMConv3d, nn.BatchNorm1d, nn.ReLU): fuse_conv_bn_relu,
-#     (spconv.SparseConv3d, nn.BatchNorm1d): fuse_conv_bn,
-#     (spconv.SparseConv3d, nn.BatchNorm1d, nn.ReLU): fuse_conv_bn_relu,
-#     (spconv.SparseInverseConv3d, nn.BatchNorm1d): fuse_conv_bn,
-#     (spconv.SparseInverseConv3d, nn.BatchNorm1d, nn.ReLU): fuse_conv_bn_relu,
-# }
 
-# def get_spconv_fuse_method_mapping():
-#     return DEFAULT_SPCONV_OP_LIST_TO_FUSER_METHOD
+def fuse_conv_bn_add_relu(is_qat, relu, add_pattern):
+    r"""Given the conv and bn modules, fuses them and returns the fused module
+
+    Args:
+        conv: Module instance of type conv2d/conv3d
+        bn: Spatial BN instance that needs to be fused with the conv
+
+    Examples::
+
+        >>> m1 = nn.Conv2d(10, 20, 3)
+        >>> b1 = nn.BatchNorm2d(20)
+        >>> m2 = fuse_conv_bn(m1, b1)
+    """
+    _, bn_pattern, _ = add_pattern
+    bn, conv = bn_pattern
+    return fuse_conv_bn_relu(is_qat, conv, bn, relu, True)
 
-# Default map for swapping float module to qat modules

spconv/pytorch/quantization/intrinsic/__init__.py

@@ -12,4 +12,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from .modules import SpconvBnNd, SpconvBnReLUNd, SpconvBnAddReLUNd, SpconvReLUNd
+from .modules import SpconvBnNd, SpconvBnReLUNd, SpconvBnAddReLUNd, SpconvReLUNd, SpconvAddReLUNd

spconv/pytorch/quantization/intrinsic/modules.py

@@ -60,3 +60,27 @@ class SpconvBnAddReLUNd(_FusedSparseModule):
             isinstance(relu, ReLU), 'Incorrect types for input modules{}{}{}' \
             .format(type(conv), type(bn), type(relu))
         super().__init__(conv, bn, relu)
+
+    def forward(self, input, add_input):
+        conv = self[0]
+        bn = self[1]
+        relu = self[2]
+        conv_res = conv(input)
+        conv_res = conv_res.replace_feature(bn(conv_res.features))
+        return conv_res.replace_feature(relu(conv_res.features + add_input.features))
+        
+class SpconvAddReLUNd(_FusedSparseModule):
+    r"""This is a sequential container which calls the Conv 3d, Batch Norm 3d, and ReLU modules.
+    During quantization this will be replaced with the corresponding fused module."""
+    def __init__(self, conv, relu):
+        assert isinstance(conv, SparseConvolution) and isinstance(relu, ReLU), \
+            'Incorrect types for input modules{}{}'.format(
+                type(conv), type(relu))
+        super().__init__(conv, relu)
+
+    def forward(self, input, add_input):
+        conv = self[0]
+        relu = self[1]
+        conv_res = conv(input)
+        return conv_res.replace_feature(relu(conv_res.features + add_input.features))
+        

spconv/pytorch/quantization/intrinsic/qat/__init__.py

@@ -12,4 +12,5 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from .modules import SparseConvBn, SparseConvBnAddReLU, SparseConvBnReLU, SparseConv, SparseConvReLU
\ No newline at end of file
+from .modules import (SparseConv, SparseConvAddReLU, SparseConvBn,
+                      SparseConvBnAddReLU, SparseConvBnReLU, SparseConvReLU)

spconv/pytorch/quantization/intrinsic/qat/modules.py

@@ -17,7 +17,7 @@ import spconv.pytorch.quantization.intrinsic as snni
 from spconv.pytorch.quantization.utils import fuse_spconv_bn_weights
 MOD = TypeVar('MOD', bound=SparseConvolution)
 
-class _SparseConv(SparseConvolution, nni._FusedModule):
+class _SparseConv(SparseConvolution):
 
     _FLOAT_MODULE = MOD
     _FLOAT_CONV_MODULE = SparseConvolution
@@ -67,7 +67,7 @@ class _SparseConv(SparseConvolution, nni._FusedModule):
         self.weight_fake_quant = qconfig.weight(factory_kwargs=factory_kwargs)
 
     def forward(self, input):
-        return self._conv_forward(False, input, self.weight_fake_quant(self.weight), self.bias)
+        return self._conv_forward(self.training, input, self.weight_fake_quant(self.weight), self.bias)
 
     @staticmethod
     def from_float(cls, mod):
@@ -77,11 +77,12 @@ class _SparseConv(SparseConvolution, nni._FusedModule):
                `mod`: a float module, either produced by torch.ao.quantization utilities
                or directly from user
         """
-        assert type(mod) == cls._FLOAT_MODULE, (
+        assert issubclass(type(mod), cls._FLOAT_MODULE), (
             "qat."
             + cls.__name__
             + ".from_float only works for "
             + cls._FLOAT_MODULE.__name__  # type: ignore[attr-defined]
+            + f" not {type(mod).__qualname__}"
         )
         assert hasattr(mod, 'qconfig'), 'Input float module must have qconfig defined'
         assert mod.qconfig, 'Input float module must have a valid qconfig'
@@ -197,6 +198,33 @@ class SparseConvReLU(SparseConv, nni._FusedModule):
     def from_float(cls, mod):
         return super(SparseConvReLU, cls).from_float(mod)
 
+class SparseConvAddReLU(SparseConv, nni._FusedModule):
+    r"""A ConvReLU2d module is a fused module of Conv2d and ReLU, attached with
+    FakeQuantize modules for weight for
+    quantization aware training.
+
+    We combined the interface of :class:`~torch.nn.Conv2d` and
+    :class:`~torch.nn.BatchNorm2d`.
+
+    Attributes:
+        weight_fake_quant: fake quant module for weight
+
+    """
+    _FLOAT_MODULE = snni.SpconvAddReLUNd
+    _FLOAT_CONV_MODULE = SparseConvolution
+    _FLOAT_BN_MODULE = None
+    _FLOAT_RELU_MODULE = nn.ReLU
+
+    def forward(self, input, add_input):
+        x = self._conv_forward(self.training, input, self.weight_fake_quant(self.weight), self.bias,
+            add_input=add_input)
+        return x.replace_feature(F.relu(x.features))
+
+    @classmethod
+    def from_float(cls, mod):
+        return super(SparseConvAddReLU, cls).from_float(mod)
+
+
 class _SparseConvBn(SparseConvolution, nni._FusedModule):
 
     _version = 2
@@ -323,9 +351,9 @@ class _SparseConvBn(SparseConvolution, nni._FusedModule):
             zero_bias = torch.zeros(self.out_channels, device=scaled_weight.device, dtype=input.features.dtype)
         conv_spt = self._conv_forward(self.training, input, scaled_weight, zero_bias)
         conv = conv_spt.features
-        conv_orig = conv / scale_factor.reshape(bias_shape)
+        conv_orig = conv / scale_factor# .reshape(bias_shape)
         if self.bias is not None:
-            conv_orig = conv_orig + self.bias.reshape(bias_shape)
+            conv_orig = conv_orig + self.bias# .reshape(bias_shape)
         conv = self.bn(conv_orig)
         if add_input is not None:
             conv = conv + add_input.features
@@ -377,7 +405,7 @@ class _SparseConvBn(SparseConvolution, nni._FusedModule):
         conv_out = torch.Tensor()
         if self.bn.training:
             # needed to compute batch mean/std
-            conv_spt = self._conv_forward(input, self.weight, zero_bias)
+            conv_spt = self._conv_forward(self.training, input, self.weight, zero_bias)
             conv_out = conv_spt.features
             # update bn statistics
             with torch.no_grad():
@@ -393,7 +421,7 @@ class _SparseConvBn(SparseConvolution, nni._FusedModule):
             self.weight * scale_factor.reshape(weight_shape)
         )
         # fused conv without bias for inference: (r * W / running_std) * X
-        conv_bn_spt = self._conv_forward(input, scaled_weight, zero_bias)
+        conv_bn_spt = self._conv_forward(self.training, input, scaled_weight, zero_bias)
         conv_bn = conv_bn_spt.features
         if self.bn.training:
             avg_dims = [0] + list(range(2, len(self.weight.shape)))
@@ -669,12 +697,12 @@ class SparseConvBnAddReLU(_SparseConvBn):
 
     """
     # base class defines _FLOAT_MODULE as "ConvBn1d"
-    _FLOAT_MODULE = snni.SpconvBnReLUNd  # type: ignore[assignment]
+    _FLOAT_MODULE = snni.SpconvBnAddReLUNd  # type: ignore[assignment]
     _FLOAT_CONV_MODULE = SparseConvolution
     _FLOAT_BN_MODULE = nn.BatchNorm1d
     _FLOAT_RELU_MODULE = nn.ReLU  # type: ignore[assignment]
     # module class after fusing bn into conv
-    _FUSED_FLOAT_MODULE = snni.SpconvReLUNd
+    _FUSED_FLOAT_MODULE = snni.SpconvAddReLUNd
 
     def forward(self, input, add_input):
         x = _SparseConvBn._forward(self, input, add_input)