[TARGET] Move target_host usage to new target style. (apache#9497)

- Add deprecation warnings to functions with target_host parameters.
- Update the build usage to new target style.

Author: tqchen

apps/android_camera/models/prepare_model.py

@@ -98,7 +98,7 @@ def main(model_str, output_path):
         pass
     print("building...")
     with tvm.transform.PassContext(opt_level=3):
-        graph, lib, params = relay.build(net, target, target_host=target_host, params=params)
+        graph, lib, params = relay.build(net, tvm.target.Target(target, target_host), params=params)
     print("dumping lib...")
     lib.export_library(output_path_str + "/" + "deploy_lib_cpu.so", ndk.create_shared)
     print("dumping graph...")

apps/android_rpc/tests/android_rpc_test.py

@@ -86,7 +86,7 @@ def test_rpc_module():
         s[B].bind(xo, te.thread_axis("blockIdx.x"))
         # Build the dynamic lib.
         # If we don't want to do metal and only use cpu, just set target to be target
-        f = tvm.build(s, [A, B], "opencl", target_host=target, name="myadd")
+        f = tvm.build(s, [A, B], tvm.target.Target("opencl", host=target), name="myadd")
         path_dso_cl = temp.relpath("dev_lib_cl.so")
         f.export_library(path_dso_cl, ndk.create_shared)
 
@@ -109,7 +109,7 @@ def test_rpc_module():
         s[B].bind(xo, te.thread_axis("blockIdx.x"))
         # Build the dynamic lib.
         # If we don't want to do metal and only use cpu, just set target to be target
-        f = tvm.build(s, [A, B], "vulkan", target_host=target, name="myadd")
+        f = tvm.build(s, [A, B], tvm.target.Target("vulkan", host=target), name="myadd")
         path_dso_vulkan = temp.relpath("dev_lib_vulkan.so")
         f.export_library(path_dso_vulkan, ndk.create_shared)
 

apps/benchmark/arm_cpu_imagenet_bench.py

@@ -40,7 +40,7 @@ def evaluate_network(network, target, target_host, repeat):
 
     print_progress("%-20s building..." % network)
     with tvm.transform.PassContext(opt_level=3):
-        lib = relay.build(net, target=target, target_host=target_host, params=params)
+        lib = relay.build(net, target=tvm.target.Target(target, host=target_host), params=params)
 
     tmp = tempdir()
     if "android" in str(target):

apps/benchmark/mobile_gpu_imagenet_bench.py

@@ -40,7 +40,7 @@ def evaluate_network(network, target, target_host, dtype, repeat):
 
     print_progress("%-20s building..." % network)
     with tvm.transform.PassContext(opt_level=3):
-        lib = relay.build(net, target=target, target_host=target_host, params=params)
+        lib = relay.build(net, target=tvm.target.Target(target, host=target_host), params=params)
 
     tmp = tempdir()
     if "android" in str(target) or "android" in str(target_host):

apps/hexagon_launcher/README.md

@@ -19,7 +19,7 @@
 ## Compilation
 
 The launcher consists of two parts: part running on Hexagon, and part running
-on Android. Each component must be compiled separately. 
+on Android. Each component must be compiled separately.
 
 The supported Snapdragon architectures are 855, 865, and 888.
 
@@ -37,9 +37,9 @@ Building the Hexagon launcher application as a component of the main TVM build
 used for Hexagon codegen can be achieved by setting `USE_HEXAGON_LAUNCHER=ON`.
 This option will compile core tvm, the android launcher binary and its corresponding
 tvm_runtime, as well as the Hexagon launcher shared library and its corresponding
-tvm_runtime. As described in the [Manual compilation](#Manual compilation) section 
-each component requires Hexagon and android dependencies. When building the launcher 
-along with TVM these configurations must be providing when invoking cmake. A minimal 
+tvm_runtime. As described in the [Manual compilation](#Manual compilation) section
+each component requires Hexagon and android dependencies. When building the launcher
+along with TVM these configurations must be providing when invoking cmake. A minimal
 example invocation for compiling TVM along with the Hexagon launcher is included below:
 
 ```
@@ -59,9 +59,9 @@ cmake -DCMAKE_C_COMPILER=/path/to/clang \
 ```
 
 where `v65|v66|v68` means "one of" these architecture versions.
-The Hexagon launcher application is an android binary and thus requires the use 
-of an android toolchain for compilation. Similarly, the Hexagon tvm runtime 
-requires the use of the Hexagon toolchain and depends on the Hexagon SDK. The 
+The Hexagon launcher application is an android binary and thus requires the use
+of an android toolchain for compilation. Similarly, the Hexagon tvm runtime
+requires the use of the Hexagon toolchain and depends on the Hexagon SDK. The
 resulting hexagon launcher binaries can be found in the `apps_hexagon_launcher`
 subdirectory of the cmake build directory. Please note that the above command
 will not build support for Hexagon codegen in the TVM library, for that please
@@ -70,7 +70,7 @@ additionally define the `USE_HEXAGON_DEVICE` variable. Also, the LLVM used in
 
 ### Manual compilation
 
-Since some source files are shared between the Hexagon and android builds, 
+Since some source files are shared between the Hexagon and android builds,
 make sure to delete all object files between compilations. Compile the Hexagon
 code first.
 
@@ -157,7 +157,7 @@ mod, params = relay.frontend.from_tflite(
 
 target = tvm.target.hexagon('v68', link_params=True)
 with tvm.transform.PassContext(opt_level=3):
-    lib = relay.build(mod, target, target_host=target, params=params, mod_name="default")
+    lib = relay.build(mod, tvm.target.Target(target, host=target), params=params, mod_name="default")
 
 # Save model.so and model.json:
 with open('model.json', 'w') as f:

apps/ios_rpc/tests/ios_rpc_mobilenet.py

@@ -93,7 +93,9 @@ def test_mobilenet(host, port, key, mode):
 
     def run(mod, target):
         with relay.build_config(opt_level=3):
-            lib = relay.build(mod, target=target, target_host=target_host, params=params)
+            lib = relay.build(
+                mod, target=tvm.target.Target(target, host=target_host), params=params
+            )
         path_dso = temp.relpath("deploy.dylib")
         lib.export_library(path_dso, xcode.create_dylib, arch=arch, sdk=sdk)
 

apps/ios_rpc/tests/ios_rpc_test.py

@@ -55,7 +55,7 @@ def test_rpc_module(host, port, key, mode):
     s[B].bind(xo, te.thread_axis("blockIdx.x"))
     # Build the dynamic lib.
     # If we don't want to do metal and only use cpu, just set target to be target
-    f = tvm.build(s, [A, B], "metal", target_host=target, name="myadd")
+    f = tvm.build(s, [A, B], tvm.target.Target("metal", host=target), name="myadd")
     path_dso1 = temp.relpath("dev_lib.dylib")
     f.export_library(path_dso1, xcode.create_dylib, arch=arch, sdk=sdk)
 

apps/topi_recipe/gemm/android_gemm_square.py

@@ -120,7 +120,7 @@ def test_gemm_gpu(N, times, bn, num_block, num_thread):
 
     print(tvm.lower(s, [A, B, C], simple_mode=True))
 
-    f = tvm.build(s, [A, B, C], "opencl", target_host=target, name="gemm_gpu")
+    f = tvm.build(s, [A, B, C], tvm.target.Target("opencl", host=target), name="gemm_gpu")
     temp = utils.tempdir()
     path_dso = temp.relpath("gemm_gpu.so")
     f.export_library(path_dso, ndk.create_shared)

docs/how_to/deploy/bnns.rst

@@ -130,7 +130,7 @@ After that you need to compile new module with target corresponding to required
 
     model = partition_for_bnns(model, params=params)  # to markup operations to be offloaded to BNNS
     with tvm.transform.PassContext(opt_level=3):
-        lib = relay.build(model, target=target, target_host=target, params=params)
+        lib = relay.build(model, target=target, params=params)
 
 Export the module.
 

docs/how_to/deploy/hls.rst

@@ -34,8 +34,7 @@ We use two python scripts for this tutorial.
       import tvm
       from tvm import te
 
-      tgt_host="llvm"
-      tgt="sdaccel"
+      tgt= tvm.target.Target("sdaccel", host="llvm")
 
       n = te.var("n")
       A = te.placeholder((n,), name='A')
@@ -47,7 +46,7 @@ We use two python scripts for this tutorial.
 
       s[C].bind(px, tvm.te.thread_axis("pipeline"))
 
-      fadd = tvm.build(s, [A, B, C], tgt, target_host=tgt_host, name="myadd")
+      fadd = tvm.build(s, [A, B, C], tgt, name="myadd")
 
       fadd.save("myadd.o")
       fadd.imported_modules[0].save("myadd.xclbin")