commit cda3ccb052be5aa81e3a3e7f5683b9ba39a8e55b
author Han-Chung Wang <hanhan0912@gmail.com> Mon Jun 10 18:15:49 2024 -0700
committer GitHub <noreply@github.com> Mon Jun 10 18:15:49 2024 -0700
tree fb864321700b76a0512b8651f0e91741d24e9f59
parent d7744b74023c6096d6d11c97d9d0f0277edda28a

[GPU] Enable tensor.pack e2e tests for rocm backend. (#17587)

The SplitFullPartialTransferPass and WorkgroupSpecializationPass are no longer needed because we have
much more mature vector lowering. It was added long time ago.

Progress on https://github.com/iree-org/iree/issues/17186

---------

Signed-off-by: hanhanW <hanhan0912@gmail.com>

compiler/src/iree/compiler/Codegen/LLVMGPU/Passes.cpp

diff --git a/compiler/src/iree/compiler/Codegen/LLVMGPU/Passes.cpp b/compiler/src/iree/compiler/Codegen/LLVMGPU/Passes.cpp
index 8a7132a..944d2aa 100644
--- a/compiler/src/iree/compiler/Codegen/LLVMGPU/Passes.cpp
+++ b/compiler/src/iree/compiler/Codegen/LLVMGPU/Passes.cpp
@@ -811,23 +811,22 @@
 
 void addGPUPackUnPackPasses(OpPassManager &funcPassManager) {
   tileAndDistributeToWorkgroup(funcPassManager);
-
-  funcPassManager.addPass(createCanonicalizerPass());
-  funcPassManager.addPass(createWorkgroupSpecializationPass());
   funcPassManager.addPass(createCanonicalizerPass());
   funcPassManager.addPass(createCSEPass());
 
   funcPassManager.addPass(createGPUTensorTilePass());
+  funcPassManager.addPass(createCanonicalizerPass());
+  funcPassManager.addPass(createCSEPass());
+
   funcPassManager.addPass(
       createDecomposePackUnPackOpsPass(/*tileOuterToOne=*/true));
+  funcPassManager.addPass(createCanonicalizerPass());
+  funcPassManager.addPass(createCSEPass());
   addGPUVectorizationPasses(funcPassManager);
 
   addBufferizePasses(funcPassManager);
 
-  // distribute foreach threads
   funcPassManager.addPass(createGPUDistributePass());
-
-  funcPassManager.addPass(createSplitFullPartialTransferPass("linalg-copy"));
 }
 
 void addGPUSimpleDistributePassPipeline(OpPassManager &funcPassManager) {

tests/e2e/tensor_ops/CMakeLists.txt

diff --git a/tests/e2e/tensor_ops/CMakeLists.txt b/tests/e2e/tensor_ops/CMakeLists.txt
index 3a98ced..52b7932 100644
--- a/tests/e2e/tensor_ops/CMakeLists.txt
+++ b/tests/e2e/tensor_ops/CMakeLists.txt
@@ -93,3 +93,34 @@
 )
 
 ### BAZEL_TO_CMAKE_PRESERVES_ALL_CONTENT_BELOW_THIS_LINE ###
+
+if(IREE_HIP_TEST_TARGET_CHIP MATCHES "^gfx")
+
+unset(IREE_HIP_TEST_COMPILER_FLAGS)
+list(APPEND IREE_HIP_TEST_COMPILER_FLAGS
+  "--iree-rocm-target-chip=${IREE_HIP_TEST_TARGET_CHIP}"
+)
+
+iree_check_single_backend_test_suite(
+  NAME
+    check_hip_stream
+  SRCS
+    "pack.mlir"
+    "pack_i8.mlir"
+  TARGET_BACKEND
+    "rocm"
+  DRIVER
+    "hip"
+  COMPILER_FLAGS
+    ${IREE_HIP_TEST_COMPILER_FLAGS}
+  RUNNER_ARGS
+    "--hip_use_streams=true"
+  LABELS
+    "noasan"
+    "nomsan"
+    "notsan"
+    "noubsan"
+    "requires-gpu-amd"
+)
+
+endif(IREE_HIP_TEST_TARGET_CHIP MATCHES "^gfx")