Merge branch 'main' into pers-tflite-greater

apps/microtvm/reference-vm/base-box-tool.py

@@ -52,20 +52,20 @@
 
 # Extra scripts required to execute on provisioning
 # in [platform]/base-box/base_box_provision.sh
+COMMON_SCRIPTS = [
+    "apps/microtvm/reference-vm/base_box_setup_common.sh",
+    "docker/install/ubuntu_install_core.sh",
+    "docker/install/ubuntu_install_python.sh",
+    "docker/utils/apt-install-and-clear.sh",
+]
+
 EXTRA_SCRIPTS = {
-    "arduino": (
-        "apps/microtvm/reference-vm/base_box_setup_common.sh",
-        "docker/install/ubuntu_install_core.sh",
-        "docker/install/ubuntu_install_python.sh",
-    ),
-    "zephyr": (
-        "apps/microtvm/reference-vm/base_box_setup_common.sh",
-        "docker/install/ubuntu_install_core.sh",
-        "docker/install/ubuntu_install_python.sh",
+    "arduino": [],
+    "zephyr": [
         "docker/install/ubuntu_init_zephyr_project.sh",
         "docker/install/ubuntu_install_zephyr_sdk.sh",
         "docker/install/ubuntu_install_cmsis.sh",
-    ),
+    ],
 }
 
 PACKER_FILE_NAME = "packer.json"
@@ -251,7 +251,9 @@ def generate_packer_config(platform, file_path, providers):
     repo_root = subprocess.check_output(
         ["git", "rev-parse", "--show-toplevel"], encoding="utf-8"
     ).strip()
-    for script in EXTRA_SCRIPTS[platform]:
+
+    scripts_to_copy = COMMON_SCRIPTS + EXTRA_SCRIPTS[platform]
+    for script in scripts_to_copy:
         script_path = os.path.join(repo_root, script)
         filename = os.path.basename(script_path)
         provisioners.append({"type": "file", "source": script_path, "destination": f"~/{filename}"})

apps/microtvm/reference-vm/base_box_setup_common.sh

@@ -16,22 +16,24 @@
 # specific language governing permissions and limitations
 # under the License.
 
-set -e
 set -x
 
 # Fix network DNS issue
 sudo sed -i 's/DNSSEC=yes/DNSSEC=no/' /etc/systemd/resolved.conf
 sudo systemctl restart systemd-resolved
 
+sudo cp ~/apt-install-and-clear.sh /usr/local/bin/apt-install-and-clear
+rm -f ~/apt-install-and-clear.sh
+
 sudo apt update
-sudo apt install -y build-essential
+sudo apt-install-and-clear -y build-essential
 sudo apt-get --purge remove modemmanager  # required to access serial ports.
 
 # Core
 sudo ~/ubuntu_install_core.sh
 rm -f ~/ubuntu_install_core.sh
 
-sudo apt install -y --no-install-recommends git \
+sudo apt-install-and-clear -y --no-install-recommends git \
      gperf ccache dfu-util device-tree-compiler xz-utils file \
      gcc gcc-multilib g++-multilib libsdl2-dev
 
@@ -40,22 +42,22 @@ wget --no-verbose https://apt.kitware.com/keys/kitware-archive-latest.asc
 sudo apt-key add kitware-archive-latest.asc
 sudo apt-add-repository 'deb https://apt.kitware.com/ubuntu/ bionic main'
 sudo apt update
-sudo apt install -y --no-install-recommends \
+sudo apt-install-and-clear -y --no-install-recommends \
      cmake=3.22.2-0kitware1ubuntu18.04.1 cmake-data=3.22.2-0kitware1ubuntu18.04.1 \
 
 # Python
 sudo ~/ubuntu_install_python.sh
 rm -f ~/ubuntu_install_python.sh
 
 # Poetry deps
-sudo apt install -y python3-venv
+sudo apt-install-and-clear -y python3-venv
 
 # TVM deps
 # TODO(mehrdadh): replace with ubuntu_install_llvm.sh
-sudo apt install -y llvm
+sudo apt-install-and-clear -y llvm
 
 # ONNX deps
-sudo apt install -y protobuf-compiler libprotoc-dev
+sudo apt-install-and-clear -y protobuf-compiler libprotoc-dev
 
 # Poetry
 curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/get-poetry.py | python3

apps/microtvm/zephyr/template_project/src/aot_standalone_demo/main.c

@@ -20,9 +20,9 @@
 #include <assert.h>
 #include <float.h>
 #include <kernel.h>
-#include <power/reboot.h>
 #include <stdio.h>
 #include <string.h>
+#include <sys/reboot.h>
 #include <tvm/runtime/c_runtime_api.h>
 #include <tvm/runtime/crt/logging.h>
 #include <tvm/runtime/crt/stack_allocator.h>

apps/microtvm/zephyr/template_project/src/host_driven/main.c

@@ -33,10 +33,10 @@
 #include <drivers/uart.h>
 #include <fatal.h>
 #include <kernel.h>
-#include <power/reboot.h>
 #include <random/rand32.h>
 #include <stdio.h>
 #include <sys/printk.h>
+#include <sys/reboot.h>
 #include <sys/ring_buffer.h>
 #include <tvm/runtime/crt/logging.h>
 #include <tvm/runtime/crt/microtvm_rpc_server.h>

docker/Dockerfile.ci_qemu

@@ -110,5 +110,11 @@ RUN bash /install/ubuntu_install_ethosu_driver_stack.sh
 COPY install/ubuntu_install_vela.sh /install/ubuntu_install_vela.sh
 RUN bash /install/ubuntu_install_vela.sh
 
+#Install CSI-NN2
+COPY install/ubuntu_download_csinn2_compute_lib.sh /install/ubuntu_download_csinn2_compute_lib.sh
+RUN bash /install/ubuntu_download_csinn2_compute_lib.sh
+
 # Update PATH
 ENV PATH /opt/arm/gcc-arm-none-eabi/bin:/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:$PATH
+ENV PATH /opt/csi-nn2/csi-nn2/tools/gcc-toolchain/bin:$PATH
+ENV PATH /opt/csi-nn2/csi-nn2/tools/qemu/bin:$PATH

docker/install/ubuntu_download_csinn2_compute_lib.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+set -e
+
+install_path="/opt/csi-nn2"
+
+# Clone CSI-NN2 Compute Library source code
+git clone --depth 1 --branch 1.12.2 https://github.com/T-head-Semi/csi-nn2.git ${install_path}
+
+# download cross-compiler when not building natively.
+# riscv gcc toolchain will be downloaded to "/path/csi-nn2/tools/gcc-toolchain".
+cd ${install_path}
+./script/download_toolchain.sh
+
+# download custom QEMU to "/path/csi-nn2/tools/qemu".
+./script/download_qemu.sh
+
+# build csinn2 lib for x86 and c906
+# lib will be installed in /path/csi-nn2/install
+# for x86
+make -j4; cd x86_build; make install; cd -
+# for c906
+mkdir -p riscv_build; cd riscv_build
+cmake ../ -DBUILD_RISCV=ON; make -j4; make install; cd -
+

docs/README.md

@@ -60,13 +60,14 @@ This folder contains the source of TVM's documentation, hosted at https://tvm.ap
    # work on most environments (e.g. MacOS).
    export TVM_TUTORIAL_EXEC_PATTERN=none
 
+   cd docs
    make html
    ```
 
 4. Run an HTTP server and visit http://localhost:8000 in your browser
 
    ```bash
-   cd docs/_build/html && python3 -m http.server
+   cd _build/html && python3 -m http.server
    ```
 
 ## Only Execute Specified Tutorials

gallery/how_to/work_with_microtvm/micro_train.py

@@ -165,19 +165,20 @@
 import urllib.request
 
 # Download datasets
+os.makedirs(f"{FOLDER}/downloads")
 os.makedirs(f"{FOLDER}/images")
 urllib.request.urlretrieve(
-    "http://ai.stanford.edu/~jkrause/car196/cars_train.tgz", f"{FOLDER}/images/target.tgz"
+    "https://data.deepai.org/stanfordcars.zip", f"{FOLDER}/downloads/target.zip"
 )
 urllib.request.urlretrieve(
-    "http://images.cocodataset.org/zips/val2017.zip", f"{FOLDER}/images/random.zip"
+    "http://images.cocodataset.org/zips/val2017.zip", f"{FOLDER}/downloads/random.zip"
 )
 
 # Extract them and rename their folders
-shutil.unpack_archive(f"{FOLDER}/images/target.tgz", f"{FOLDER}/images")
-shutil.unpack_archive(f"{FOLDER}/images/random.zip", f"{FOLDER}/images")
-shutil.move(f"{FOLDER}/images/cars_train", f"{FOLDER}/images/target")
-shutil.move(f"{FOLDER}/images/val2017", f"{FOLDER}/images/random")
+shutil.unpack_archive(f"{FOLDER}/downloads/target.zip", f"{FOLDER}/downloads")
+shutil.unpack_archive(f"{FOLDER}/downloads/random.zip", f"{FOLDER}/downloads")
+shutil.move(f"{FOLDER}/downloads/cars_train/cars_train", f"{FOLDER}/images/target")
+shutil.move(f"{FOLDER}/downloads/val2017", f"{FOLDER}/images/random")
 
 ######################################################################
 # Loading the Data

include/tvm/relay/attrs/nn.h

@@ -124,7 +124,8 @@ struct Conv2DAttrs : public tvm::AttrsNode<Conv2DAttrs> {
   tvm::String data_layout;
   tvm::String kernel_layout;
   tvm::String out_layout;
-  tvm::String auto_scheduler_rewritten_layout;  // The layout after auto-scheduler's layout rewrite
+  tvm::String auto_scheduler_rewritten_layout;   // The layout after auto-scheduler's layout rewrite
+  Array<PrimExpr> meta_schedule_original_shape;  // The original shape of the weights
   DataType out_dtype;
 
   TVM_DECLARE_ATTRS(Conv2DAttrs, "relay.attrs.Conv2DAttrs") {
@@ -217,7 +218,8 @@ struct Conv2DWinogradAttrs : public tvm::AttrsNode<Conv2DWinogradAttrs> {
   tvm::String data_layout;
   tvm::String kernel_layout;
   tvm::String out_layout;
-  tvm::String auto_scheduler_rewritten_layout;  // The layout after auto-scheduler's layout rewrite
+  tvm::String auto_scheduler_rewritten_layout;   // The layout after auto-scheduler's layout rewrite
+  Array<PrimExpr> meta_schedule_original_shape;  // The original shape of the weights
   DataType out_dtype;
 
   TVM_DECLARE_ATTRS(Conv2DWinogradAttrs, "relay.attrs.Conv2DWinogradAttrs") {
@@ -308,7 +310,8 @@ struct Conv3DAttrs : public tvm::AttrsNode<Conv3DAttrs> {
   tvm::String data_layout;
   tvm::String kernel_layout;
   tvm::String out_layout;
-  tvm::String auto_scheduler_rewritten_layout;  // The layout after auto-scheduler's layout rewrite
+  tvm::String auto_scheduler_rewritten_layout;   // The layout after auto-scheduler's layout rewrite
+  Array<PrimExpr> meta_schedule_original_shape;  // The original shape of the weights
   DataType out_dtype;
 
   TVM_DECLARE_ATTRS(Conv3DAttrs, "relay.attrs.Conv3DAttrs") {
@@ -1049,7 +1052,8 @@ struct MatmulAttrs : public tvm::AttrsNode<MatmulAttrs> {
   DataType out_dtype;
   bool transpose_a;
   bool transpose_b;
-  tvm::String auto_scheduler_rewritten_layout;  // The layout after auto-scheduler's layout rewrite
+  tvm::String auto_scheduler_rewritten_layout;   // The layout after auto-scheduler's layout rewrite
+  Array<PrimExpr> meta_schedule_original_shape;  // The original shape of the weights
 
   TVM_DECLARE_ATTRS(MatmulAttrs, "relay.attrs.MatmulAttrs") {
     TVM_ATTR_FIELD(units).describe("Number of hidden units of the dense transformation.");
@@ -1072,7 +1076,8 @@ struct MatmulAttrs : public tvm::AttrsNode<MatmulAttrs> {
 /*! \brief Attributes for dense operator */
 struct DenseAttrs : public tvm::AttrsNode<DenseAttrs> {
   IndexExpr units;
-  tvm::String auto_scheduler_rewritten_layout;  // The layout after auto-scheduler's layout rewrite
+  tvm::String auto_scheduler_rewritten_layout;   // The layout after auto-scheduler's layout rewrite
+  Array<PrimExpr> meta_schedule_original_shape;  // The original shape of the weights
   DataType out_dtype;
 
   TVM_DECLARE_ATTRS(DenseAttrs, "relay.attrs.DenseAttrs") {
@@ -1109,7 +1114,8 @@ struct BatchMatmulAttrs : public tvm::AttrsNode<BatchMatmulAttrs> {
   DataType out_dtype;
   bool transpose_a;
   bool transpose_b;
-  tvm::String auto_scheduler_rewritten_layout;  // The layout after auto-scheduler's layout rewrite
+  tvm::String auto_scheduler_rewritten_layout;   // The layout after auto-scheduler's layout rewrite
+  Array<PrimExpr> meta_schedule_original_shape;  // The original shape of the weights
 
   TVM_DECLARE_ATTRS(BatchMatmulAttrs, "relay.attrs.BatchMatmulAttrs") {
     // use 0 bits to indicate none.

include/tvm/tir/index_map.h

@@ -135,8 +135,19 @@ class IndexMapNode : public Object {
     v->Visit("final_indices", &final_indices);
   }
 
-  static constexpr const char* _type_key = "tir.IndexMap";
+  bool SEqualReduce(const IndexMapNode* other, SEqualReducer equal) const {
+    return equal.DefEqual(initial_indices, other->initial_indices) &&
+           equal(final_indices, other->final_indices);
+  }
 
+  void SHashReduce(SHashReducer hash_reduce) const {
+    hash_reduce.DefHash(initial_indices);
+    hash_reduce(final_indices);
+  }
+
+  static constexpr const char* _type_key = "tir.IndexMap";
+  static constexpr const bool _type_has_method_sequal_reduce = true;
+  static constexpr const bool _type_has_method_shash_reduce = true;
   TVM_DECLARE_FINAL_OBJECT_INFO(IndexMapNode, Object);
 };
 

jenkins/Jenkinsfile.j2

@@ -52,7 +52,7 @@ import org.jenkinsci.plugins.pipeline.modeldefinition.Utils
 
 // NOTE: these lines are scanned by docker/dev_common.sh. Please update the regex as needed. -->
 ci_lint = 'tlcpack/ci-lint:20220513-055910-fa834f67e'
-ci_gpu = 'tlcpack/ci-gpu:20220606-055910-bf4b8f5c7'
+ci_gpu = 'tlcpack/ci-gpu:20220619-055908-9bba7580b'
 ci_cpu = 'tlcpack/ci-cpu:20220519-055908-ddfa1da69'
 ci_wasm = 'tlcpack/ci-wasm:20220513-055910-fa834f67e'
 ci_i386 = 'tlcpack/ci-i386:20220513-055910-fa834f67e'

python/tvm/relay/frontend/tflite.py

@@ -1443,11 +1443,7 @@ def convert_less_equal(self, op):
 
     def convert_equal(self, op):
         """Convert TFLite EQUAL"""
-        if self.is_quantized(op):
-            raise tvm.error.OpNotImplemented(
-                "TFlite quantized EQUAL operator is not supported yet."
-            )
-        return self._convert_elemwise(_op.equal, op)
+        return self._convert_elemwise(_op.equal, op, self.is_quantized(op))
 
     def convert_not_equal(self, op):
         """Convert TFLite NOT_EQUAL"""

python/tvm/runtime/module.py

@@ -436,7 +436,7 @@ def export_library(self, file_name, fcompile=None, addons=None, workspace_dir=No
         files = addons if addons else []
         is_system_lib = False
         has_c_module = False
-        llvm_target_triple = None
+        llvm_target_string = None
         for index, module in enumerate(modules):
             if fcompile is not None and hasattr(fcompile, "object_format"):
                 if module.type_key == "c":
@@ -475,25 +475,27 @@ def export_library(self, file_name, fcompile=None, addons=None, workspace_dir=No
             is_system_lib = (
                 module.type_key == "llvm" and module.get_function("__tvm_is_system_module")()
             )
-            llvm_target_triple = (
-                module.type_key == "llvm" and module.get_function("_get_target_triple")()
+            llvm_target_string = (
+                module.type_key == "llvm" and module.get_function("_get_target_string")()
             )
         if not fcompile:
             if file_name.endswith(".tar"):
                 fcompile = _tar.tar
             else:
                 fcompile = _cc.create_shared
 
-        if llvm_target_triple is None and hasattr(fcompile, "get_target_triple"):
-            llvm_target_triple = fcompile.get_target_triple()
+        if llvm_target_string is None and hasattr(fcompile, "get_target_triple"):
+            triple = fcompile.get_target_triple()
+            assert triple, "Target triple should not be empty"
+            llvm_target_string = "llvm -mtriple " + triple
 
         if getattr(fcompile, "need_system_lib", False) and not is_system_lib:
             raise ValueError("%s need --system-lib option" % str(fcompile))
 
         if self.imported_modules:
-            if enabled("llvm") and llvm_target_triple:
+            if enabled("llvm") and llvm_target_string:
                 path_obj = os.path.join(workspace_dir, f"devc.{object_format}")
-                m = _ffi_api.ModulePackImportsToLLVM(self, is_system_lib, llvm_target_triple)
+                m = _ffi_api.ModulePackImportsToLLVM(self, is_system_lib, llvm_target_string)
                 m.save(path_obj)
                 files.append(path_obj)
             else: