Merge pull request #45928 from ROCmSoftwarePlatform:google_upstream_rocm_switch_to_rocm40

PiperOrigin-RevId: 350140370
Change-Id: Ic33f95a7e829d06bf860c54eb0bd8c5fdcf198e8

tensorflow/compiler/xla/service/gpu/llvm_gpu_backend/gpu_backend_lib.cc

@@ -800,8 +800,16 @@ Status AMDGPUTargetModuleLinker(llvm::Module* module, GpuVersion gpu_version,
 std::unique_ptr<llvm::TargetMachine> AMDGPUGetTargetMachine(
     llvm::Triple target_triple, int amdgpu_version,
     const HloModuleConfig& hlo_module_config) {
+  string feature_str = "+code-object-v3";
+#if TF_ROCM_VERSION >= 30900
+  // code-object-v3 is default, so no need to expliticitly specify it
+  // in the feature string. Also, starting with ROCm 4.0, this feature string
+  // is deprecated, and we get a warning to that effect. So removing that
+  // feature string
+  feature_str = "";
+#endif
   return GetTargetMachine(target_triple, absl::StrCat("gfx", amdgpu_version),
-                          hlo_module_config, "+code-object-v3");
+                          hlo_module_config, feature_str);
 }
 
 void AMDGPUBackendInit(const HloModuleConfig& hlo_module_config) {

tensorflow/compiler/xla/tests/BUILD

@@ -367,7 +367,6 @@ xla_test(
         "conv_depthwise_test.cc",
     ],
     shard_count = 50,
-    tags = ["no_rocm"],  # ROCm 3.9 regression
     deps = [
         ":conv_depthwise_common",
         ":test_macros_header",
@@ -389,7 +388,6 @@ xla_test(
     timeout = "long",
     srcs = ["conv_depthwise_backprop_filter_test.cc"],
     shard_count = 40,
-    tags = ["no_rocm"],  # ROCm 3.9 regression
     deps = [
         ":test_macros_header",
         "//tensorflow/compiler/xla:execution_options_util",
@@ -414,7 +412,6 @@ xla_test(
         "cpu",
     ],
     shard_count = 50,
-    tags = ["no_rocm"],  # ROCm 3.9 regression
     deps = [
         ":client_library_test_base",
         ":hlo_test_base",
@@ -924,7 +921,6 @@ xla_test(
     srcs = ["dot_operation_test.cc"],
     shard_count = 20,
     tags = [
-        "no_rocm",  # ROCm 3.9 regression
         "optonly",
     ],
     deps = [
@@ -958,7 +954,6 @@ xla_test(
     backends = ["gpu"],
     shard_count = 20,
     tags = [
-        "no_rocm",  # ROCm 3.9 regression
         "optonly",
         # TODO(b/151340488): Timed out on 2020-03-12.
         "nozapfhahn",
@@ -1025,7 +1020,6 @@ xla_test(
     },
     shard_count = 20,
     tags = [
-        "no_rocm",  # ROCm 3.9 regression
         "optonly",
     ],
     deps = [
@@ -1253,7 +1247,6 @@ xla_test(
         "cpu": ["nomsan"],
     },
     shard_count = 30,
-    tags = ["no_rocm"],  # ROCm 3.9 regression
     deps = [
         ":test_macros_header",
         "//tensorflow/compiler/xla:array3d",
@@ -1278,7 +1271,6 @@ xla_test(
     timeout = "long",
     srcs = ["convolution_dimension_numbers_test.cc"],
     shard_count = 20,
-    tags = ["no_rocm"],  # ROCm 3.9 regression
     deps = [
         ":test_macros_header",
         "//tensorflow/compiler/xla:array4d",
@@ -2322,7 +2314,6 @@ xla_test(
     name = "multioutput_fusion_test",
     srcs = ["multioutput_fusion_test.cc"],
     backends = ["gpu"],
-    tags = ["no_rocm"],  # ROCm 3.9 regression
     deps = [
         ":test_macros_header",
         "//tensorflow/compiler/xla:literal",

tensorflow/python/ops/numpy_ops/BUILD

@@ -110,6 +110,7 @@ cuda_py_test(
 cuda_py_test(
     name = "np_interop_test",
     srcs = ["np_interop_test.py"],
+    tags = ["no_rocm"],
     deps = [
         ":numpy",
         "//tensorflow:tensorflow_py",

tensorflow/tools/ci_build/Dockerfile.rocm

@@ -3,10 +3,10 @@
 FROM ubuntu:bionic
 MAINTAINER Jeff Poznanovic <jeffrey.poznanovic@amd.com>
 
-ARG ROCM_DEB_REPO=http://repo.radeon.com/rocm/apt/3.9/
+ARG ROCM_DEB_REPO=http://repo.radeon.com/rocm/apt/4.0/
 ARG ROCM_BUILD_NAME=xenial
 ARG ROCM_BUILD_NUM=main
-ARG ROCM_PATH=/opt/rocm-3.9.0
+ARG ROCM_PATH=/opt/rocm-4.0.0
 
 ENV DEBIAN_FRONTEND noninteractive
 ENV TF_NEED_ROCM 1
@@ -78,7 +78,7 @@ ENV PATH="$ROCM_PATH/bin:${PATH}"
 ENV PATH="$OPENCL_ROOT/bin:${PATH}"
 
 # Add target file to help determine which device(s) to build for
-RUN bash -c 'echo -e "gfx803\ngfx900\ngfx906" >> ${ROCM_PATH}/bin/target.lst'
+RUN bash -c 'echo -e "gfx803\ngfx900\ngfx906\ngfx908" >> ${ROCM_PATH}/bin/target.lst'
 
 # Need to explicitly create the $ROCM_PATH/.info/version file to workaround what seems to be a bazel bug
 # The env vars being set via --action_env in .bazelrc and .tf_configure.bazelrc files are sometimes

tensorflow/tools/ci_build/linux/rocm/run_cc_core.sh

@@ -28,7 +28,7 @@ echo "Bazel will use ${N_BUILD_JOBS} concurrent build job(s) and ${N_TEST_JOBS}
 echo ""
 
 # First positional argument (if any) specifies the ROCM_INSTALL_DIR
-ROCM_INSTALL_DIR=/opt/rocm-3.9.0
+ROCM_INSTALL_DIR=/opt/rocm-4.0.0
 if [[ -n $1 ]]; then
     ROCM_INSTALL_DIR=$1
 fi

tensorflow/tools/ci_build/linux/rocm/run_csb_tests.sh

@@ -28,7 +28,7 @@ echo "Bazel will use ${N_BUILD_JOBS} concurrent build job(s) and ${N_TEST_JOBS}
 echo ""
 
 # First positional argument (if any) specifies the ROCM_INSTALL_DIR
-ROCM_INSTALL_DIR=/opt/rocm-3.9.0
+ROCM_INSTALL_DIR=/opt/rocm-4.0.0
 if [[ -n $1 ]]; then
     ROCM_INSTALL_DIR=$1
 fi

tensorflow/tools/ci_build/linux/rocm/run_py3_core.sh

@@ -28,7 +28,7 @@ echo "Bazel will use ${N_BUILD_JOBS} concurrent build job(s) and ${N_TEST_JOBS}
 echo ""
 
 # First positional argument (if any) specifies the ROCM_INSTALL_DIR
-ROCM_INSTALL_DIR=/opt/rocm-3.9.0
+ROCM_INSTALL_DIR=/opt/rocm-4.0.0
 if [[ -n $1 ]]; then
     ROCM_INSTALL_DIR=$1
 fi

tensorflow/tools/ci_build/xla/linux/rocm/run_py3.sh

@@ -28,7 +28,7 @@ echo "Bazel will use ${N_BUILD_JOBS} concurrent build job(s) and ${N_TEST_JOBS}
 echo ""
 
 # First positional argument (if any) specifies the ROCM_INSTALL_DIR
-ROCM_INSTALL_DIR=/opt/rocm-3.9.0
+ROCM_INSTALL_DIR=/opt/rocm-4.0.0
 if [[ -n $1 ]]; then
     ROCM_INSTALL_DIR=$1
 fi