[PATCH] D120070: [HIP] Support linking archive of bundled bitcode

[Yaxun Liu via Phabricator via cfe-commits]

yaxunl created this revision.
yaxunl added a reviewer: tra.
yaxunl requested review of this revision.

HIP programs compiled with -c -fgpu-rdc generate clang-offload-bundler
bundles which contain bitcode for different GPU's.

Such files can be archived to an archive file which can be linked with
HIP programs with -fgpu-rdc.

This patch adds suppor of linking archive of bundled bitcode.

When an archive of bundled bitcode is passed to clang by -l, for each
GPU specified through --offload-arch, clang extracts bitcode from
the archive and creates a new archive for that GPU and pass it
to lld.


https://reviews.llvm.org/D120070

Files:
  clang/lib/Driver/ToolChains/HIPAMD.cpp
  clang/test/Driver/clang-offload-bundler.c
  clang/test/Driver/hip-link-bundle-archive.hip


Index: clang/test/Driver/hip-link-bundle-archive.hip
===================================================================
--- /dev/null
+++ clang/test/Driver/hip-link-bundle-archive.hip
@@ -0,0 +1,14 @@
+// REQUIRES: clang-driver, x86-registered-target, amdgpu-registered-target
+
+// RUN: touch %T/libhipBundled.a
+
+// Check clang unbundle the archive and link them by lld.
+
+// RUN: %clang -### --offload-arch=gfx906 --offload-arch=gfx1030 \
+// RUN:   -nogpulib %s -fgpu-rdc -L%T -lhipBundled \
+// RUN:   2>&1 | FileCheck -check-prefix=CHECK %s
+
+// CHECK: "{{.*}}clang-offload-bundler" "-unbundle" "-type=a" "-inputs={{.*}}libhipBundled.a" "-targets=hip-amdgcn-amd-amdhsa-gfx1030" "-outputs=[[A1030:.*\.a]]" "-allow-missing-bundles"
+// CHECK: "{{.*}}lld" {{.*}}"-plugin-opt=mcpu=gfx1030" {{.*}} "[[A1030]]"
+// CHECK: "{{.*}}clang-offload-bundler" "-unbundle" "-type=a" "-inputs={{.*}}libhipBundled.a" "-targets=hip-amdgcn-amd-amdhsa-gfx906" "-outputs=[[A906:.*\.a]]" "-allow-missing-bundles"
+// CHECK: "{{.*}}lld" {{.*}}"-plugin-opt=mcpu=gfx906" {{.*}} "[[A906]]"
Index: clang/test/Driver/clang-offload-bundler.c
===================================================================
--- clang/test/Driver/clang-offload-bundler.c
+++ clang/test/Driver/clang-offload-bundler.c
@@ -365,6 +365,28 @@
 // CKLST2-NOT: openmp-powerpc64le-ibm-linux-gnu
 // CKLST2-NOT: openmp-x86_64-pc-linux-gnu
 
+//
+// Check unbundling archive for HIP.
+//
+// When the input to clang-offload-bundler is an archive of bundled bitcodes,
+// for each target, clang-offload-bundler extracts the bitcode from each
+// bundle and archives them. Therefore for each target, the output is an
+// archive of unbundled bitcodes.
+//
+// RUN: clang-offload-bundler -type=bc -targets=hip-amdgcn-amd-amdhsa--gfx900,hip-amdgcn-amd-amdhsa--gfx906 \
+// RUN:   -inputs=%t.tgt1,%t.tgt2 -outputs=%T/hip_bundle1.bc
+// RUN: clang-offload-bundler -type=bc -targets=hip-amdgcn-amd-amdhsa--gfx900,hip-amdgcn-amd-amdhsa--gfx906 \
+// RUN:   -inputs=%t.tgt1,%t.tgt2 -outputs=%T/hip_bundle2.bc
+// RUN: llvm-ar cr %T/hip_archive.a %T/hip_bundle1.bc %T/hip_bundle2.bc
+// RUN: clang-offload-bundler -unbundle -type=a -targets=hip-amdgcn-amd-amdhsa--gfx900,hip-amdgcn-amd-amdhsa--gfx906 \
+// RUN:   -outputs=%T/hip_900.a,%T/hip_906.a -inputs=%T/hip_archive.a
+// RUN: llvm-ar t %T/hip_900.a | FileCheck -check-prefix=HIP-AR-900 %s
+// RUN: llvm-ar t %T/hip_906.a | FileCheck -check-prefix=HIP-AR-906 %s
+// HIP-AR-900-DAG: hip_bundle1-hip-amdgcn-amd-amdhsa--gfx900
+// HIP-AR-900-DAG: hip_bundle2-hip-amdgcn-amd-amdhsa--gfx900
+// HIP-AR-906-DAG: hip_bundle1-hip-amdgcn-amd-amdhsa--gfx906
+// HIP-AR-906-DAG: hip_bundle2-hip-amdgcn-amd-amdhsa--gfx906
+
 //
 // Check bundling without host target is allowed for HIP.
 //
Index: clang/lib/Driver/ToolChains/HIPAMD.cpp
===================================================================
--- clang/lib/Driver/ToolChains/HIPAMD.cpp
+++ clang/lib/Driver/ToolChains/HIPAMD.cpp
@@ -121,6 +121,14 @@
   for (auto Input : Inputs)
     LldArgs.push_back(Input.getFilename());
 
+  // Look for archive of bundled bitcode in arguments, and add temporary files
+  // for the extracted archive of bitcode to inputs.
+  auto TargetID = Args.getLastArgValue(options::OPT_mcpu_EQ);
+  AddStaticDeviceLibsLinking(C, *this, JA, Inputs, Args, LldArgs, "amdgcn",
+                             TargetID,
+                             /*IsBitCodeSDL=*/true,
+                             /*PostClangLink=*/false);
+
   const char *Lld = Args.MakeArgString(getToolChain().GetProgramPath("lld"));
   C.addCommand(std::make_unique<Command>(JA, *this, ResponseFileSupport::None(),
                                          Lld, LldArgs, Inputs, Output));


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D120070.409695.patch
Type: text/x-patch
Size: 3738 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20220217/a2b260b9/attachment.bin>