[clang-offload-bundler] Add unbundling of archives containing bundled object files into device specific archives

[saiislam]

Summary

This patch adds unbundling support of an archive file. It takes an
archive file along with a set of offload targets as input.
Output is a device specific archive for each given offload target.
Input archive contains bundled code objects bundled using
clang-offload-bundler. Each generated device specific archive contains
a set of device code object files which are named as
<Parent Bundle Name>-<CodeObject-GPUArch>.
Entries in input archive can be of any binary type which is
supported by clang-offload-bundler, like *.bc. Output archives will
contain files in same type.
Example Usuage:
clang-offload-bundler --unbundle --inputs=lib-generic.a -type=a -targets=openmp-amdgcn-amdhsa--gfx906,openmp-amdgcn-amdhsa--gfx908 -outputs=devicelib-gfx906.a,deviceLib-gfx908.a

[saiislam]

Refer to D80816 for earlier review of this patch.

[ABataev]

clang/tools/clang-offload-bundler/ClangOffloadBundler.cpp

183–185

No need else here

1061–1065

No need for else here

1068–1070

I think llvm Support lib has all required functions for this.

1108

Do not use auto where the type is not obvious.

1156–1157

Just continue and make else if just if

[saiislam]

Modified to handle multiple targets/outputs in one run of the tool for archive unbundling. Other minor changes as requested in the review.

[saiislam]

clang/tools/clang-offload-bundler/ClangOffloadBundler.cpp

1156–1157

wasn't possible with the code flow. there is stuff to be processed in case of failure as well.

[saiislam]

Ping.

[yaxunl]

can you document this in ClangOffloadBundler.rst ? I think we need a clear description about how clang-offload-bundler knows which file in the .a file belongs to which target.

[t-tye]

In D93525#2493024, @yaxunl wrote:

can you document this in ClangOffloadBundler.rst ? I think we need a clear description about how clang-offload-bundler knows which file in the .a file belongs to which target.

How does the .a relate to bundled code objects? Does the .a have a number of bundled code objects? If so wouldn't the identity of code objects be defined by the existing bundled code object ABI already documented? If the .a is a set of non-bundled code objects then defining how they are identified is not part of the clang-offload-bundler documentation as there are no bundled code objects involved. It would seem that the documentation belongs with the OpenMP runtime/compiler that is choosing to use .a files in this manner.

[saiislam]

In D93525#2493752, @t-tye wrote:

In D93525#2493024, @yaxunl wrote:

can you document this in ClangOffloadBundler.rst ? I think we need a clear description about how clang-offload-bundler knows which file in the .a file belongs to which target.

How does the .a relate to bundled code objects? Does the .a have a number of bundled code objects? If so wouldn't the identity of code objects be defined by the existing bundled code object ABI already documented? If the .a is a set of non-bundled code objects then defining how they are identified is not part of the clang-offload-bundler documentation as there are no bundled code objects involved. It would seem that the documentation belongs with the OpenMP runtime/compiler that is choosing to use .a files in this manner.

Bundles (created using clang-offload-bundler) are passed to llvm-ar to create an archive of bundled objects (*.a file). An archive can have bundles for multiple device types. So, yes, the identity of code objects is defined by the existing bundled code object ABI.
This patch reads such an archive and produces a device-specific archive for each of the target devices given as input. Each device-specific archive contains all the code objects corresponding to that particular device and are written as per llvm archive format.

Here is a snippet of relevant lit run lines:

// RUN: %clang -O0 -target %itanium_abi_triple %s -c -o %t.o

// RUN: echo 'Content of device file 1' > %t.tgt1
// RUN: clang-offload-bundler -type=o -targets=host-%itanium_abi_triple,openmp-amdgcn-amd-amdhsa-gfx900 -inputs=%t.o,%t.tgt1 -outputs=%t.abundle1.o
 
// RUN: echo 'Content of device file 2' > %t.tgt2
// RUN: clang-offload-bundler -type=o -targets=host-%itanium_abi_triple,openmp-amdgcn-amd-amdhsa-gfx900 -inputs=%t.o,%t.tgt2 -outputs=%t.abundle2.o
 
// RUN: llvm-ar cr %t.lib.a %t.abundle1.o %t.abundle2.o

This patch ==>
// RUN: clang-offload-bundler -unbundle -type=a -targets=openmp-amdgcn-amd-amdhsa-gfx900 -inputs=%t.lib.a -outputs=%t.devicelib.a

%t.devicelib.a will contain all devices objects corresponding to gfx900

Though my interest originates from OpenMP side, Device-specific Archive Libraries created like this can be used by other offloading languages like HIP, CUDA, and OpenCL. Pelase refer D81109 for the an earlier patch in the series of patches which will enable this.

[t-tye]

In D93525#2495374, @saiislam wrote:

In D93525#2493752, @t-tye wrote:

In D93525#2493024, @yaxunl wrote:

can you document this in ClangOffloadBundler.rst ? I think we need a clear description about how clang-offload-bundler knows which file in the .a file belongs to which target.

How does the .a relate to bundled code objects? Does the .a have a number of bundled code objects? If so wouldn't the identity of code objects be defined by the existing bundled code object ABI already documented? If the .a is a set of non-bundled code objects then defining how they are identified is not part of the clang-offload-bundler documentation as there are no bundled code objects involved. It would seem that the documentation belongs with the OpenMP runtime/compiler that is choosing to use .a files in this manner.

Bundles (created using clang-offload-bundler) are passed to llvm-ar to create an archive of bundled objects (*.a file). An archive can have bundles for multiple device types. So, yes, the identity of code objects is defined by the existing bundled code object ABI.
This patch reads such an archive and produces a device-specific archive for each of the target devices given as input. Each device-specific archive contains all the code objects corresponding to that particular device and are written as per llvm archive format.

Here is a snippet of relevant lit run lines:

// RUN: %clang -O0 -target %itanium_abi_triple %s -c -o %t.o

// RUN: echo 'Content of device file 1' > %t.tgt1
// RUN: clang-offload-bundler -type=o -targets=host-%itanium_abi_triple,openmp-amdgcn-amd-amdhsa-gfx900 -inputs=%t.o,%t.tgt1 -outputs=%t.abundle1.o
 
// RUN: echo 'Content of device file 2' > %t.tgt2
// RUN: clang-offload-bundler -type=o -targets=host-%itanium_abi_triple,openmp-amdgcn-amd-amdhsa-gfx900 -inputs=%t.o,%t.tgt2 -outputs=%t.abundle2.o
 
// RUN: llvm-ar cr %t.lib.a %t.abundle1.o %t.abundle2.o

This patch ==>
// RUN: clang-offload-bundler -unbundle -type=a -targets=openmp-amdgcn-amd-amdhsa-gfx900 -inputs=%t.lib.a -outputs=%t.devicelib.a

%t.devicelib.a will contain all devices objects corresponding to gfx900

Though my interest originates from OpenMP side, Device-specific Archive Libraries created like this can be used by other offloading languages like HIP, CUDA, and OpenCL. Pelase refer D81109 for the an earlier patch in the series of patches which will enable this.

The naming of code objects in a bundled code object includes the processor name and the settings for target features (see https://clang.llvm.org/docs/ClangOffloadBundler.html#target-id and https://llvm.org/docs/AMDGPUUsage.html#target-id). The compatibility of code objects considers both target processor matching and target feature compatibility. Target features can have three settings: on, off and any. The compatibility is that each feature that is on/off must exactly match, but any will match either on or off.

So when unbundling an archive how is the desired code object being requested? How is it handling the target features? For example, if code objects that will be compatible with a feature being on is required, then matching code objects in the archive would be those that have that feature either on or any.

[saiislam]

In D93525#2495937, @t-tye wrote:

In D93525#2495374, @saiislam wrote:

In D93525#2493752, @t-tye wrote:

In D93525#2493024, @yaxunl wrote:

can you document this in ClangOffloadBundler.rst ? I think we need a clear description about how clang-offload-bundler knows which file in the .a file belongs to which target.

How does the .a relate to bundled code objects? Does the .a have a number of bundled code objects? If so wouldn't the identity of code objects be defined by the existing bundled code object ABI already documented? If the .a is a set of non-bundled code objects then defining how they are identified is not part of the clang-offload-bundler documentation as there are no bundled code objects involved. It would seem that the documentation belongs with the OpenMP runtime/compiler that is choosing to use .a files in this manner.

Bundles (created using clang-offload-bundler) are passed to llvm-ar to create an archive of bundled objects (*.a file). An archive can have bundles for multiple device types. So, yes, the identity of code objects is defined by the existing bundled code object ABI.
This patch reads such an archive and produces a device-specific archive for each of the target devices given as input. Each device-specific archive contains all the code objects corresponding to that particular device and are written as per llvm archive format.

Here is a snippet of relevant lit run lines:

// RUN: %clang -O0 -target %itanium_abi_triple %s -c -o %t.o

// RUN: echo 'Content of device file 1' > %t.tgt1
// RUN: clang-offload-bundler -type=o -targets=host-%itanium_abi_triple,openmp-amdgcn-amd-amdhsa-gfx900 -inputs=%t.o,%t.tgt1 -outputs=%t.abundle1.o
 
// RUN: echo 'Content of device file 2' > %t.tgt2
// RUN: clang-offload-bundler -type=o -targets=host-%itanium_abi_triple,openmp-amdgcn-amd-amdhsa-gfx900 -inputs=%t.o,%t.tgt2 -outputs=%t.abundle2.o
 
// RUN: llvm-ar cr %t.lib.a %t.abundle1.o %t.abundle2.o

This patch ==>
// RUN: clang-offload-bundler -unbundle -type=a -targets=openmp-amdgcn-amd-amdhsa-gfx900 -inputs=%t.lib.a -outputs=%t.devicelib.a

%t.devicelib.a will contain all devices objects corresponding to gfx900

Though my interest originates from OpenMP side, Device-specific Archive Libraries created like this can be used by other offloading languages like HIP, CUDA, and OpenCL. Pelase refer D81109 for the an earlier patch in the series of patches which will enable this.

The naming of code objects in a bundled code object includes the processor name and the settings for target features (see https://clang.llvm.org/docs/ClangOffloadBundler.html#target-id and https://llvm.org/docs/AMDGPUUsage.html#target-id). The compatibility of code objects considers both target processor matching and target feature compatibility. Target features can have three settings: on, off and any. The compatibility is that each feature that is on/off must exactly match, but any will match either on or off.

So when unbundling an archive how is the desired code object being requested? How is it handling the target features? For example, if code objects that will be compatible with a feature being on is required, then matching code objects in the archive would be those that have that feature either on or any.

At the moment this patch defines compatibility as exact string match of bundler entry ID. So, it doesn't support target ID concept fully. But, following example work.
Supporting target ID requires little more work and discussion.

// RUN: clang-offload-bundler -type=o -targets=host-%itanium_abi_triple,openmp-amdgcn-amd-amdhsa--gfx908 -inputs=%t.o,%t.tgt1 -outputs=%t.abundle1.o
// RUN: clang-offload-bundler -type=o -targets=host-%itanium_abi_triple,openmp-amdgcn-amd-amdhsa--gfx908:sramecc+:xnack+,openmp-amdgcn-amd-amdhsa--gfx908:sramecc-:xnack+ -inputs=%t.o,%t.tgt1,%t.tgt2 -outputs=%t.targetIDbundle.o
// RUN: llvm-ar cr %t.targetIDlib.a %t.abundle1.o %t.targetIDbundle.o
// RUN: clang-offload-bundler -unbundle -type=a -targets=openmp-amdgcn-amd-amdhsa--gfx908:sramecc+:xnack+ -inputs=%t.targetIDlib.a -outputs=%t.devicelibt-sramecc+.a
// RUN: llvm-ar t %t.devicelibt-sramecc+.a | FileCheck %s -check-prefix=SRAMECCplus
// SRAMECCplus: targetIDbundle.bc
// SRAMECCplus-NOT: abundle1.bc

[jdoerfert]

At the moment this patch defines compatibility as exact string match of bundler entry ID.
[...]
Supporting target ID requires little more work and discussion.

Let's get this in first, then revisit target ID support as we need it.

[t-tye]

In D93525#2509796, @jdoerfert wrote:

At the moment this patch defines compatibility as exact string match of bundler entry ID.
[...]
Supporting target ID requires little more work and discussion.

Let's get this in first, then revisit target ID support as we need it.

I do not think this patch should ignore target ID as that is now upstreamed and documented. What is involved in correcting the compatibility test to be correct by the target ID rules? There are examples of doing this in all the runtimes and I can help if that is useful.

[jdoerfert]

In D93525#2509836, @t-tye wrote:

In D93525#2509796, @jdoerfert wrote:

At the moment this patch defines compatibility as exact string match of bundler entry ID.
[...]
Supporting target ID requires little more work and discussion.

Let's get this in first, then revisit target ID support as we need it.

I do not think this patch should ignore target ID as that is now upstreamed and documented. What is involved in correcting the compatibility test to be correct by the target ID rules? There are examples of doing this in all the runtimes and I can help if that is useful.

First, there is no reason not to have multiple patches as long as they are self contained and testable. Arguably, smaller patches are better.

That said, target ID is a new feature and, as discussed in the OpenMP call today, there is a chance we have to revisit this to support more involved information. As this discussion is open ended (and hasn't started yet), it seems absolutely sensible to continue with a tested and working patch that provides features we need for sure instead of forcing some support of a feature we don't use right now anyway.

[saiislam]

Added support for optional TargetID during unbundling of archives.

[jdoerfert]

Can we split this patch now and make progress?

[saiislam]

1. Removed TargetID support, to be reviewed in a followup patch.
2. Added OffloadTargetInfo class to encapsulate handling of bundle entry ID components: OffloadKind, Triple, GPUArch.

[saiislam]

Removed unused header.

[jdoerfert]

Thanks for splitting this. I quickly went over it only.

clang/docs/ClangOffloadBundler.rst

137

not needed here.

clang/tools/clang-offload-bundler/ClangOffloadBundler.cpp

1431

By now, a proper conditional seems appropriate.

[saiislam]

Removed Triple format example from documentation and simplified conditional calling of bundling/unbundling functions.

[saiislam]

Ping.

[yaxunl]

Does this pass internal CI (ePSDB) ? I am concerned about the enforcement of the canonical format of target triple since this may break backward compatibility.

[saiislam]

Updated clang and hip tests to ensure that all 4 components of triple are mandataroly available in the bundle entry ID.

[yaxunl]

clang/lib/Driver/ToolChains/Clang.cpp

7619 ↗

(On Diff #353922)

This is not HIP specific. Other toolchain could use a non-canonical triple too. Also there may be more components of triple missing. A generic fix would be use Triple::normalize for all toolchain. same as below.

[saiislam]

Generalized padding of Triple fields of Bundle Entry ID while generating command for clang-offload-bundler.

[yaxunl]

LGTM. Thanks! Pls make sure it passes internal CI (ePSDB) before committing.

[saiislam]

In D93525#2838535, @yaxunl wrote:

LGTM. Thanks! Pls make sure it passes internal CI (ePSDB) before committing.

Sure, I will take care of it. Thanks!

Any comments @jdoerfert ?

[jdoerfert]

Looks reasonable to me. We can always refine it as we go.

clang/tools/clang-offload-bundler/ClangOffloadBundler.cpp

134

1259

Add a message to asserts. also other places.

[saiislam]

@yaxunl
this patch on its own is failing in our internal CI. I have an internal patch (542569) to integrate it cleanly there.

[grokos]

clang/docs/ClangOffloadBundler.rst

128

A bit of wordplay, but it's weird that a *triple* now has 4 elements...

clang/tools/clang-offload-bundler/ClangOffloadBundler.cpp

144

Leftover? Components is already 6 elements long.

1098

CodeObject --> CodeObjectInfo

[yaxunl]

In D93525#2849815, @saiislam wrote:

@yaxunl
this patch on its own is failing in our internal CI. I have an internal patch (542569) to integrate it cleanly there.

Fine. Thanks.

[saiislam]

clang/docs/ClangOffloadBundler.rst

128

I think llvm::Triple it is named Triple because of historical reasons. Otherwise, it already has these components (including the environment).
As llvm::Triple API's don't force presence of all components it is not an issue in most cases, but in our case of Bundle Entry ID we need a way to differentiate between 4th component of Triple and a GPU arch, hence this rule.

clang/tools/clang-offload-bundler/ClangOffloadBundler.cpp

144

Not necessarily. It is possible that target has less than 6 elements. For example all bundling/unbundling cases which do not require GPUArch field.
E.g. "openmp-powerpc64le-ibm-linux-gnu"

[grokos]

clang/tools/clang-offload-bundler/ClangOffloadBundler.cpp

144

OK, thanks!

[rsmith]

clang/test/Driver/clang-offload-bundler.c

376

This test does not depend on llvm-ar, and this change causes check-clang to fail in the case where llvm-ar has not previously been built. Please can you fix? (Might need some changes to the build rules to add a dependency on llvm-ar, if you can't avoid depending on it for this test.)

[saiislam]

@rsmith,
Thanks for pointing out this issue. I have proposed a fix in D105285. Please review.

[JonChesterfield]

This is mentioned as broken in the referenced patch and landed with @t-tye still marked as requires changes. Revert warranted?

Testing looks very sparse for something handling archives. Presumably it has the same set of bugs as D108291 e.g. implicit whole archive semantics

[saiislam]

In D93525#2952012, @JonChesterfield wrote:

This is mentioned as broken in the referenced patch and landed with @t-tye still marked as requires changes. Revert warranted?

Testing looks very sparse for something handling archives. Presumably it has the same set of bugs as D108291 e.g. implicit whole archive semantics

Tony's earlier objection/ask was to include TargetID support along with archive unbundling support in this. I had verbal consent from him to split the patch and propose TargetID support for OpenMP in a separate patch. The same was agreed upon in the multi-company meeting as well.
Also, D106870 places necessary infrastructure to support TargetID in a follow up patch. Once it lands, TargetID patch is fairly straightforward.
Whole archive semantics didn't introduce any bug anywhere, though performance is a separate issue. D108291 is required because nvlink silently fails to link cubin files inside an archive.

Any suggestions on how can I improve testing for archives?

clang/docs/ClangOffloadBundler.rst

.. table:: Bundled Code Object Offload Kind
                     to be loaded by the HIP runtime. The fat binary can be
                     loaded directly from a file, or be embedded in the host code
                     object as a data section with the name ``.hip_fatbin``.
 
       openmp        Offload code object for the OpenMP language extension.
       ============= ==============================================================
 
 **target-triple**
-  The target triple of the code object.
+    The target triple of the code object:
+
+.. code::
+
+  <Architecture>-<Vendor>-<OS>-<Environment>
+
+It is required to have all four components present, if target-id is present.
+If a component is not specified then the empty string must be used in its place.
+For example, if ``<Vendor>`` and ``<Environment>`` is unspecified the target
+triple would be:
+
+.. code::
+
+  myarch--myOS-
 
 **target-id**
   The canonical target ID of the code object. Present only if the target
   supports a target ID. See :ref:`clang-target-id`.
 
 Each entry of a bundled code object must have a different bundle entry ID. There
 can be multiple entries for the same processor provided they differ in target
 feature settings. If there is an entry with a target feature specified as *Any*,

clang/test/Driver/clang-offload-bundler.c

 // CK-ERR8B: error: invalid target 'xpenmp-x86_xx-pc-linux-gnu', unknown offloading kind 'xpenmp', unknown target triple 'x86_xx-pc-linux-gnu'
 
 // RUN: not clang-offload-bundler -type=i -targets=openmp-powerpc64le-linux,openmp-powerpc64le-ibm-linux-gnu,openmp-x86_64-pc-linux-gnu -inputs=%t.i,%t.tgt1,%t.tgt2 -outputs=%t.bundle.i 2>&1 | FileCheck %s --check-prefix CK-ERR9A
 // CK-ERR9A: error: expecting exactly one host target but got 0
 
 // RUN: not clang-offload-bundler -type=i -targets=host-%itanium_abi_triple,host-%itanium_abi_triple,openmp-x86_64-pc-linux-gnu -inputs=%t.i,%t.tgt1,%t.tgt2 -outputs=%t.bundle.i 2>&1 | FileCheck %s --check-prefix CK-ERR9B
 // CK-ERR9B: error: Duplicate targets are not allowed
 
+// RUN: not clang-offload-bundler -type=a -targets=hxst-powerpcxxle-ibm-linux-gnu,openxp-pxxerpc64le-ibm-linux-gnu,xpenmp-x86_xx-pc-linux-gnu -inputs=%t.i,%t.tgt1,%t.tgt2 -outputs=%t.bundle.i 2>&1 | FileCheck %s --check-prefix CK-ERR10A
+// CK-ERR10A: error: Archive files are only supported for unbundling
+
 //
 // Check text bundle. This is a readable format, so we check for the format we expect to find.
 //
 // RUN: clang-offload-bundler -type=i -targets=host-%itanium_abi_triple,openmp-powerpc64le-ibm-linux-gnu,openmp-x86_64-pc-linux-gnu -inputs=%t.i,%t.tgt1,%t.tgt2 -outputs=%t.bundle3.i
 // RUN: clang-offload-bundler -type=ii -targets=host-%itanium_abi_triple,openmp-powerpc64le-ibm-linux-gnu,openmp-x86_64-pc-linux-gnu -inputs=%t.ii,%t.tgt1,%t.tgt2 -outputs=%t.bundle3.ii
 // RUN: clang-offload-bundler -type=ll -targets=host-%itanium_abi_triple,openmp-powerpc64le-ibm-linux-gnu,openmp-x86_64-pc-linux-gnu -inputs=%t.ll,%t.tgt1,%t.tgt2 -outputs=%t.bundle3.ll
 // RUN: clang-offload-bundler -type=s -targets=host-%itanium_abi_triple,openmp-powerpc64le-ibm-linux-gnu,openmp-x86_64-pc-linux-gnu -inputs=%t.s,%t.tgt1,%t.tgt2 -outputs=%t.bundle3.s
 // RUN: clang-offload-bundler -type=s -targets=openmp-powerpc64le-ibm-linux-gnu,host-%itanium_abi_triple,openmp-x86_64-pc-linux-gnu -inputs=%t.tgt1,%t.s,%t.tgt2 -outputs=%t.bundle3.unordered.s

 // RUN: clang-offload-bundler -type=bc -targets=host-%itanium_abi_triple,openmp-powerpc64le-ibm-linux-gnu,openmp-x86_64-pc-linux-gnu -outputs=%t.res.bc,%t.res.tgt1,%t.res.tgt2 -inputs=%t.bundle3.bc -unbundle
 // RUN: diff %t.bc %t.res.bc
 // RUN: diff %t.tgt1 %t.res.tgt1
 // RUN: diff %t.tgt2 %t.res.tgt2
 
 //
 // Check error due to missing bundles
 //
-// RUN: clang-offload-bundler -type=bc -targets=host-%itanium_abi_triple,hip-amdgcn-amd-amdhsa-gfx900 -inputs=%t.bc,%t.tgt1 -outputs=%t.hip.bundle.bc
+// RUN: clang-offload-bundler -type=bc -targets=host-%itanium_abi_triple,hip-amdgcn-amd-amdhsa--gfx900 -inputs=%t.bc,%t.tgt1 -outputs=%t.hip.bundle.bc
 // RUN: not clang-offload-bundler -type=bc -inputs=%t.hip.bundle.bc -outputs=%t.tmp.bc -unbundle \
-// RUN:   -targets=hip-amdgcn-amd-amdhsa-gfx906 \
+// RUN:   -targets=hip-amdgcn-amd-amdhsa--gfx906 \
 // RUN:   2>&1 | FileCheck -check-prefix=MISS1 %s
 // RUN: not clang-offload-bundler -type=bc -inputs=%t.hip.bundle.bc -outputs=%t.tmp.bc,%t.tmp2.bc -unbundle \
-// RUN:   -targets=hip-amdgcn-amd-amdhsa-gfx906,hip-amdgcn-amd-amdhsa-gfx900 \
+// RUN:   -targets=hip-amdgcn-amd-amdhsa--gfx906,hip-amdgcn-amd-amdhsa--gfx900 \
 // RUN:   2>&1 | FileCheck -check-prefix=MISS1 %s
-// MISS1: error: Can't find bundles for hip-amdgcn-amd-amdhsa-gfx906
+// MISS1: error: Can't find bundles for hip-amdgcn-amd-amdhsa--gfx906
 // RUN: not clang-offload-bundler -type=bc -inputs=%t.hip.bundle.bc -outputs=%t.tmp.bc,%t.tmp2.bc -unbundle \
-// RUN:   -targets=hip-amdgcn-amd-amdhsa-gfx906,hip-amdgcn-amd-amdhsa-gfx803 \
+// RUN:   -targets=hip-amdgcn-amd-amdhsa--gfx906,hip-amdgcn-amd-amdhsa--gfx803 \
 // RUN:   2>&1 | FileCheck -check-prefix=MISS2 %s
-// MISS2: error: Can't find bundles for hip-amdgcn-amd-amdhsa-gfx803 and hip-amdgcn-amd-amdhsa-gfx906
+// MISS2: error: Can't find bundles for hip-amdgcn-amd-amdhsa--gfx803 and hip-amdgcn-amd-amdhsa--gfx906
 // RUN: not clang-offload-bundler -type=bc -inputs=%t.hip.bundle.bc -outputs=%t.tmp.bc,%t.tmp2.bc,%t.tmp3.bc -unbundle \
-// RUN:   -targets=hip-amdgcn-amd-amdhsa-gfx906,hip-amdgcn-amd-amdhsa-gfx803,hip-amdgcn-amd-amdhsa-gfx1010 \
+// RUN:   -targets=hip-amdgcn-amd-amdhsa--gfx906,hip-amdgcn-amd-amdhsa--gfx803,hip-amdgcn-amd-amdhsa--gfx1010 \
 // RUN:   2>&1 | FileCheck -check-prefix=MISS3 %s
-// MISS3: error: Can't find bundles for hip-amdgcn-amd-amdhsa-gfx1010, hip-amdgcn-amd-amdhsa-gfx803, and hip-amdgcn-amd-amdhsa-gfx906
+// MISS3: error: Can't find bundles for hip-amdgcn-amd-amdhsa--gfx1010, hip-amdgcn-amd-amdhsa--gfx803, and hip-amdgcn-amd-amdhsa--gfx906
 
 //
 // Check error due to duplicate targets
 //
-// RUN: not clang-offload-bundler -type=bc -targets=host-%itanium_abi_triple,hip-amdgcn-amd-amdhsa-gfx900,hip-amdgcn-amd-amdhsa-gfx900 \
+// RUN: not clang-offload-bundler -type=bc -targets=host-%itanium_abi_triple,hip-amdgcn-amd-amdhsa--gfx900,hip-amdgcn-amd-amdhsa--gfx900 \
 // RUN:   -inputs=%t.bc,%t.tgt1,%t.tgt1 -outputs=%t.hip.bundle.bc 2>&1 | FileCheck -check-prefix=DUP %s
 // RUN: not clang-offload-bundler -type=bc -inputs=%t.hip.bundle.bc -outputs=%t.tmp.bc,%t.tmp2.bc -unbundle \
-// RUN:   -targets=hip-amdgcn-amd-amdhsa-gfx906,hip-amdgcn-amd-amdhsa-gfx906 \
+// RUN:   -targets=hip-amdgcn-amd-amdhsa--gfx906,hip-amdgcn-amd-amdhsa--gfx906 \
 // RUN:   2>&1 | FileCheck -check-prefix=DUP %s
 // DUP: error: Duplicate targets are not allowed
 //
 // Check -list option
 //
 
 // RUN: clang-offload-bundler -bundle-align=4096 -type=bc -targets=host-%itanium_abi_triple,openmp-powerpc64le-ibm-linux-gnu,openmp-x86_64-pc-linux-gnu -inputs=%t.bc,%t.tgt1,%t.tgt2 -outputs=%t.bundle3.bc
 // RUN: not clang-offload-bundler -type=bc -inputs=%t.bundle3.bc -unbundle -list 2>&1 | FileCheck -check-prefix=CKLST-ERR %s
 // CKLST-ERR: error: -unbundle and -list cannot be used together
 // RUN: not clang-offload-bundler -type=bc -inputs=%t.bundle3.bc -targets=host-%itanium_abi_triple -list 2>&1 | FileCheck -check-prefix=CKLST-ERR2 %s
 // CKLST-ERR2: error: -targets option is invalid for -list
 // RUN: not clang-offload-bundler -type=bc -inputs=%t.bundle3.bc -outputs=out.txt -list 2>&1 | FileCheck -check-prefix=CKLST-ERR3 %s
 // CKLST-ERR3: error: -outputs option is invalid for -list
 // RUN: not clang-offload-bundler -type=bc -inputs=%t.bundle3.bc,%t.bc -list 2>&1 | FileCheck -check-prefix=CKLST-ERR4 %s
 // CKLST-ERR4: error: only one input file supported for -list
 
 // CKLST-DAG: host-
 // CKLST-DAG: openmp-powerpc64le-ibm-linux-gnu
 // CKLST-DAG: openmp-x86_64-pc-linux-gnu
 
 // CKLST2-NOT: host-
 // CKLST2-NOT: openmp-powerpc64le-ibm-linux-gnu
 // CKLST2-NOT: openmp-x86_64-pc-linux-gnu
 
+//
+// Check archive unbundling
+//
+// Create few code object bundles and archive them to create an input archive
+// RUN: clang-offload-bundler -type=o -targets=host-%itanium_abi_triple,openmp-amdgcn-amd-amdhsa--gfx906,openmp-amdgcn-amd-amdhsa--gfx908 -inputs=%t.o,%t.tgt1,%t.tgt2 -outputs=%t.simple.bundle
+// RUN: clang-offload-bundler -type=o -targets=host-%itanium_abi_triple,openmp-amdgcn-amd-amdhsa--gfx903 -inputs=%t.o,%t.tgt1 -outputs=%t.simple1.bundle
+// RUN: llvm-ar cr %t.input-archive.a %t.simple.bundle %t.simple1.bundle
+
+// RUN: clang-offload-bundler -unbundle -type=a -targets=openmp-amdgcn-amd-amdhsa--gfx906,openmp-amdgcn-amd-amdhsa--gfx908 -inputs=%t.input-archive.a -outputs=%t-archive-gfx906-simple.a,%t-archive-gfx908-simple.a
+// RUN: llvm-ar t %t-archive-gfx906-simple.a | FileCheck %s -check-prefix=GFX906
+// GFX906: simple-openmp-amdgcn-amd-amdhsa--gfx906
+// RUN: llvm-ar t %t-archive-gfx908-simple.a | FileCheck %s -check-prefix=GFX908
+// GFX908-NOT: {{gfx906}}
+
 // Some code so that we can create a binary out of this file.
 int A = 0;
 void test_func(void) {
   ++A;
 }

clang/tools/clang-offload-bundler/ClangOffloadBundler.cpp

 #include "clang/Basic/Version.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringMap.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/ADT/Triple.h"
+#include "llvm/Object/Archive.h"
+#include "llvm/Object/ArchiveWriter.h"
 #include "llvm/Object/Binary.h"
 #include "llvm/Object/ObjectFile.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
 #include "llvm/Support/Errc.h"
 #include "llvm/Support/Error.h"
 #include "llvm/Support/ErrorOr.h"
 #include "llvm/Support/FileSystem.h"
+#include "llvm/Support/Host.h"
 #include "llvm/Support/MemoryBuffer.h"
 #include "llvm/Support/Path.h"
 #include "llvm/Support/Program.h"
 #include "llvm/Support/Signals.h"
 #include "llvm/Support/StringSaver.h"
 #include "llvm/Support/WithColor.h"
 #include "llvm/Support/raw_ostream.h"
 #include <algorithm>
FilesType("type", cl::Required,
                        "  i   - cpp-output\n"
                        "  ii  - c++-cpp-output\n"
                        "  cui - cuda/hip-output\n"
                        "  d   - dependency\n"
                        "  ll  - llvm\n"
                        "  bc  - llvm-bc\n"
                        "  s   - assembler\n"
                        "  o   - object\n"
+                       "  a   - archive of objects\n"
                        "  gch - precompiled-header\n"
                        "  ast - clang AST file"),
               cl::cat(ClangOffloadBundlerCategory));
 static cl::opt<bool>
     Unbundle("unbundle",
              cl::desc("Unbundle bundled file into several output files.\n"),
              cl::init(false), cl::cat(ClangOffloadBundlerCategory));
 

 #define OFFLOAD_BUNDLER_MAGIC_STR "__CLANG_OFFLOAD_BUNDLE__"
 
 /// The index of the host input in the list of inputs.
 static unsigned HostInputIndex = ~0u;
 
 /// Path to the current binary.
 static std::string BundlerExecutable;
 
-/// Obtain the offload kind and real machine triple out of the target
-/// information specified by the user.
-static void getOffloadKindAndTriple(StringRef Target, StringRef &OffloadKind,
-                                    StringRef &Triple) {
-  auto KindTriplePair = Target.split('-');
-  OffloadKind = KindTriplePair.first;
-  Triple = KindTriplePair.second;
-}
-static bool hasHostKind(StringRef Target) {
+/// Obtain the offload kind, real machine triple, and an optional GPUArch
+/// out of the target information specified by the user.
+/// Bundle Entry ID (or, Offload Target String) has following components:
+///  * Offload Kind - Host, OpenMP, or HIP
+///  * Triple - Standard LLVM Triple
+///  * GPUArch (Optional) - Processor name, like gfx906 or sm_30
+/// In presence of Proc, the Triple should contain separator "-" for all
+/// standard four components, even if they are empty.
+struct OffloadTargetInfo {
   StringRef OffloadKind;
-  StringRef Triple;
-  getOffloadKindAndTriple(Target, OffloadKind, Triple);
-  return OffloadKind == "host";
+  llvm::Triple Triple;
+  StringRef GPUArch;
+
+  OffloadTargetInfo(const StringRef Target) {
+    SmallVector<StringRef, 6> Components;
+    Target.split(Components, '-', 5);
+    Components.resize(6);
+    this->OffloadKind = Components[0];
+    this->Triple = llvm::Triple(Components[1], Components[2], Components[3],
+                                Components[4]);
+    this->GPUArch = Components[5];
+  }
+
+  bool hasHostKind() const { return this->OffloadKind == "host"; }
+
+  bool isOffloadKindValid() const {
+    return OffloadKind == "host" || OffloadKind == "openmp" ||
+           OffloadKind == "hip" || OffloadKind == "hipv4";
+  }
+
+  bool isTripleValid() const {
+    return !Triple.str().empty() && Triple.getArch() != Triple::UnknownArch;
   }
 
+  bool operator==(const OffloadTargetInfo &Target) const {
+    return OffloadKind == Target.OffloadKind &&
+           Triple.isCompatibleWith(Target.Triple) && GPUArch == Target.GPUArch;
+  }
+
+  std::string str() {
+    return Twine(OffloadKind + "-" + Triple.str() + "-" + GPUArch).str();
+  }
+};
+
 /// Generic file handler interface.
 class FileHandler {
 public:
   struct BundleInfo {
     StringRef BundleID;
   };
 
   FileHandler() {}
 
   virtual ~FileHandler() {}
 
   /// Update the file handler with information from the header of the bundled
   /// file.
   virtual Error ReadHeader(MemoryBuffer &Input) = 0;
 
   /// Read the marker of the next bundled to be read in the file. The bundle
   /// name is returned if there is one in the file, or `None` if there are no
   /// more bundles to be read.
   virtual Expected<Optional<StringRef>>
   ReadBundleStart(MemoryBuffer &Input) = 0;
 
   /// Read the marker that closes the current bundle.
   virtual Error ReadBundleEnd(MemoryBuffer &Input) = 0;
 
   /// Read the current bundle and write the result into the stream \a OS.
-  virtual Error ReadBundle(raw_fd_ostream &OS, MemoryBuffer &Input) = 0;
+  virtual Error ReadBundle(raw_ostream &OS, MemoryBuffer &Input) = 0;
 
   /// Write the header of the bundled file to \a OS based on the information
   /// gathered from \a Inputs.
   virtual Error WriteHeader(raw_fd_ostream &OS,
                             ArrayRef<std::unique_ptr<MemoryBuffer>> Inputs) = 0;
 
   /// Write the marker that initiates a bundle for the triple \a TargetTriple to
   /// \a OS.
Expected<Optional<StringRef>> ReadBundleStart(MemoryBuffer &Input) final {
     return CurBundleInfo->first();
   }
 
   Error ReadBundleEnd(MemoryBuffer &Input) final {
     assert(CurBundleInfo != BundlesInfo.end() && "Invalid reader info!");
     return Error::success();
   }
 
-  Error ReadBundle(raw_fd_ostream &OS, MemoryBuffer &Input) final {
+  Error ReadBundle(raw_ostream &OS, MemoryBuffer &Input) final {
     assert(CurBundleInfo != BundlesInfo.end() && "Invalid reader info!");
     StringRef FC = Input.getBuffer();
     OS.write(FC.data() + CurBundleInfo->second.Offset,
              CurBundleInfo->second.Size);
     return Error::success();
   }
 
   Error WriteHeader(raw_fd_ostream &OS,
while (NextSection != Obj->section_end()) {
       if (*TripleOrErr)
         return **TripleOrErr;
     }
     return None;
   }
 
   Error ReadBundleEnd(MemoryBuffer &Input) final { return Error::success(); }
 
-  Error ReadBundle(raw_fd_ostream &OS, MemoryBuffer &Input) final {
+  Error ReadBundle(raw_ostream &OS, MemoryBuffer &Input) final {
     Expected<StringRef> ContentOrErr = CurrentSection->getContents();
     if (!ContentOrErr)
       return ContentOrErr.takeError();
     StringRef Content = *ContentOrErr;
 
     // Copy fat object contents to the output when extracting host bundle.
     if (Content.size() == 1u && Content.front() == 0)
       Content = StringRef(Input.getBufferStart(), Input.getBufferSize());
Error ReadBundleEnd(MemoryBuffer &Input) final {
     size_t TripleEnd = ReadChars = FC.find("\n", ReadChars + 1);
     if (TripleEnd != FC.npos)
       // Next time we read after the new line.
       ++ReadChars;
 
     return Error::success();
   }
 
-  Error ReadBundle(raw_fd_ostream &OS, MemoryBuffer &Input) final {
+  Error ReadBundle(raw_ostream &OS, MemoryBuffer &Input) final {
     StringRef FC = Input.getBuffer();
     size_t BundleStart = ReadChars;
 
     // Find end of the bundle.
     size_t BundleEnd = ReadChars = FC.find(BundleEndString, ReadChars);
 
     StringRef Bundle(&FC.data()[BundleStart], BundleEnd - BundleStart);
     OS << Bundle;
CreateFileHandler(MemoryBuffer &FirstInput) {
   if (FilesType == "ll")
     return std::make_unique<TextFileHandler>(/*Comment=*/";");
   if (FilesType == "bc")
     return std::make_unique<BinaryFileHandler>();
   if (FilesType == "s")
     return std::make_unique<TextFileHandler>(/*Comment=*/"#");
   if (FilesType == "o")
     return CreateObjectFileHandler(FirstInput);
+  if (FilesType == "a")
+    return CreateObjectFileHandler(FirstInput);
   if (FilesType == "gch")
     return std::make_unique<BinaryFileHandler>();
   if (FilesType == "ast")
     return std::make_unique<BinaryFileHandler>();
 
   return createStringError(errc::invalid_argument,
                            "'" + FilesType + "': invalid file type specified");
 }
if (EC)
       return createFileError(Output->second, EC);
     if (Error Err = FH->ReadBundle(OutputFile, Input))
       return Err;
     if (Error Err = FH->ReadBundleEnd(Input))
       return Err;
     Worklist.erase(Output);
 
     // Record if we found the host bundle.
-    if (hasHostKind(CurTriple))
+    auto OffloadInfo = OffloadTargetInfo(CurTriple);
+    if (OffloadInfo.hasHostKind())
       FoundHostBundle = true;
   }
 
   if (!AllowMissingBundles && !Worklist.empty()) {
     std::string ErrMsg = "Can't find bundles for";
     std::set<StringRef> Sorted;
     for (auto &E : Worklist)
       Sorted.insert(E.first());
static Error UnbundleFiles() {
   if (Worklist.size() == TargetNames.size()) {
     for (auto &E : Worklist) {
       std::error_code EC;
       raw_fd_ostream OutputFile(E.second, EC, sys::fs::OF_None);
       if (EC)
         return createFileError(E.second, EC);
 
       // If this entry has a host kind, copy the input file to the output file.
-      if (hasHostKind(E.first()))
+      auto OffloadInfo = OffloadTargetInfo(E.getKey());
+      if (OffloadInfo.hasHostKind())
         OutputFile.write(Input.getBufferStart(), Input.getBufferSize());
     }
     return Error::success();
   }
 
   // If we found elements, we emit an error if none of those were for the host
   // in case host bundle name was provided in command line.
   if (!FoundHostBundle && HostInputIndex != ~0u)
     return createStringError(inconvertibleErrorCode(),
                              "Can't find bundle for the host target");
 
   // If we still have any elements in the worklist, create empty files for them.
   for (auto &E : Worklist) {
     std::error_code EC;
     raw_fd_ostream OutputFile(E.second, EC, sys::fs::OF_None);
     if (EC)
       return createFileError(E.second, EC);
   }
 
   return Error::success();
 }
 
+static Archive::Kind getDefaultArchiveKindForHost() {
+  return Triple(sys::getDefaultTargetTriple()).isOSDarwin() ? Archive::K_DARWIN
+                                                            : Archive::K_GNU;
+}
+
+/// @brief Checks if a code object \p CodeObjectInfo is compatible with a given
+/// target \p TargetInfo.
+/// @link https://clang.llvm.org/docs/ClangOffloadBundler.html#bundle-entry-id
+bool isCodeObjectCompatible(OffloadTargetInfo &CodeObjectInfo,
+                            OffloadTargetInfo &TargetInfo) {
+
+  // Compatible in case of exact match.
+  if (CodeObjectInfo == TargetInfo) {
+    DEBUG_WITH_TYPE(
+        "CodeObjectCompatibility",
+        dbgs() << "Compatible: Exact match: " << CodeObjectInfo.str() << "\n");
+    return true;
+  }
+
+  // Incompatible if Kinds or Triples mismatch.
+  if (CodeObjectInfo.OffloadKind != TargetInfo.OffloadKind ||
+      !CodeObjectInfo.Triple.isCompatibleWith(TargetInfo.Triple)) {
+    DEBUG_WITH_TYPE(
+        "CodeObjectCompatibility",
+        dbgs() << "Incompatible: Kind/Triple mismatch \t[CodeObject: "
+               << CodeObjectInfo.str() << "]\t:\t[Target: " << TargetInfo.str()
+               << "]\n");
+    return false;
+  }
+
+  // Incompatible if GPUArch mismatch.
+  if (CodeObjectInfo.GPUArch != TargetInfo.GPUArch) {
+    DEBUG_WITH_TYPE("CodeObjectCompatibility",
+                    dbgs() << "Incompatible: GPU Arch mismatch \t[CodeObject: "
+                           << CodeObjectInfo.str()
+                           << "]\t:\t[Target: " << TargetInfo.str() << "]\n");
+    return false;
+  }
+
+  DEBUG_WITH_TYPE(
+      "CodeObjectCompatibility",
+      dbgs() << "Compatible: Code Objects are compatible \t[CodeObject: "
+             << CodeObjectInfo.str() << "]\t:\t[Target: " << TargetInfo.str()
+             << "]\n");
+  return true;
+}
+
+/// @brief Computes a list of targets among all given targets which are
+/// compatible with this code object
+/// @param [in] Code Object \p CodeObject
+/// @param [out] List of all compatible targets \p CompatibleTargets among all
+/// given targets
+/// @return false, if no compatible target is found.
+static bool
+getCompatibleOffloadTargets(OffloadTargetInfo &CodeObjectInfo,
+                            SmallVectorImpl<StringRef> &CompatibleTargets) {
+  if (!CompatibleTargets.empty()) {
+    DEBUG_WITH_TYPE("CodeObjectCompatibility",
+                    dbgs() << "CompatibleTargets list should be empty\n");
+    return false;
+  }
+  for (auto &Target : TargetNames) {
+    auto TargetInfo = OffloadTargetInfo(Target);
+    if (isCodeObjectCompatible(CodeObjectInfo, TargetInfo))
+      CompatibleTargets.push_back(Target);
+  }
+  return !CompatibleTargets.empty();
+}
+
+/// UnbundleArchive takes an archive file (".a") as input containing bundled
+/// code object files, and a list of offload targets (not host), and extracts
+/// the code objects into a new archive file for each offload target. Each
+/// resulting archive file contains all code object files corresponding to that
+/// particular offload target. The created archive file does not
+/// contain an index of the symbols and code object files are named as
+/// <<Parent Bundle Name>-<CodeObject's GPUArch>>, with ':' replaced with '_'.
+static Error UnbundleArchive() {
+  std::vector<std::unique_ptr<MemoryBuffer>> ArchiveBuffers;
+
+  /// Map of target names with list of object files that will form the device
+  /// specific archive for that target
+  StringMap<std::vector<NewArchiveMember>> OutputArchivesMap;
+
+  // Map of target names and output archive filenames
+  StringMap<StringRef> TargetOutputFileNameMap;
+
+  auto Output = OutputFileNames.begin();
+  for (auto &Target : TargetNames) {
+    TargetOutputFileNameMap[Target] = *Output;
+    ++Output;
+  }
+
+  StringRef IFName = InputFileNames.front();
+  ErrorOr<std::unique_ptr<MemoryBuffer>> BufOrErr =
+      MemoryBuffer::getFileOrSTDIN(IFName, -1, false);
+  if (std::error_code EC = BufOrErr.getError())
+    return createFileError(InputFileNames.front(), EC);
+
+  ArchiveBuffers.push_back(std::move(*BufOrErr));
+  Expected<std::unique_ptr<llvm::object::Archive>> LibOrErr =
+      Archive::create(ArchiveBuffers.back()->getMemBufferRef());
+  if (!LibOrErr)
+    return LibOrErr.takeError();
+
+  auto Archive = std::move(*LibOrErr);
+
+  Error ArchiveErr = Error::success();
+  auto ChildEnd = Archive->child_end();
+
+  /// Iterate over all bundled code object files in the input archive.
+  for (auto ArchiveIter = Archive->child_begin(ArchiveErr);
+       ArchiveIter != ChildEnd; ++ArchiveIter) {
+    if (ArchiveErr)
+      return ArchiveErr;
+    auto ArchiveChildNameOrErr = (*ArchiveIter).getName();
+    if (!ArchiveChildNameOrErr)
+      return ArchiveChildNameOrErr.takeError();
+
+    StringRef BundledObjectFile = sys::path::filename(*ArchiveChildNameOrErr);
+
+    auto CodeObjectBufferRefOrErr = (*ArchiveIter).getMemoryBufferRef();
+    if (!CodeObjectBufferRefOrErr)
+      return CodeObjectBufferRefOrErr.takeError();
+
+    auto CodeObjectBuffer =
+        MemoryBuffer::getMemBuffer(*CodeObjectBufferRefOrErr, false);
+
+    Expected<std::unique_ptr<FileHandler>> FileHandlerOrErr =
+        CreateFileHandler(*CodeObjectBuffer);
+    if (!FileHandlerOrErr)
+      return FileHandlerOrErr.takeError();
+
+    std::unique_ptr<FileHandler> &FileHandler = *FileHandlerOrErr;
+    assert(FileHandler);
+
+    if (Error ReadErr = FileHandler.get()->ReadHeader(*CodeObjectBuffer))
+      return ReadErr;
+
+    Expected<Optional<StringRef>> CurBundleIDOrErr =
+        FileHandler->ReadBundleStart(*CodeObjectBuffer);
+    if (!CurBundleIDOrErr)
+      return CurBundleIDOrErr.takeError();
+
+    Optional<StringRef> OptionalCurBundleID = *CurBundleIDOrErr;
+    // No device code in this child, skip.
+    if (!OptionalCurBundleID.hasValue())
+      continue;
+    StringRef CodeObject = *OptionalCurBundleID;
+
+    // Process all bundle entries (CodeObjects) found in this child of input
+    // archive.
+    while (!CodeObject.empty()) {
+      SmallVector<StringRef> CompatibleTargets;
+      auto CodeObjectInfo = OffloadTargetInfo(CodeObject);
+      if (CodeObjectInfo.hasHostKind()) {
+        // Do nothing, we don't extract host code yet.
+      } else if (getCompatibleOffloadTargets(CodeObjectInfo,
+                                             CompatibleTargets)) {
+        std::string BundleData;
+        raw_string_ostream DataStream(BundleData);
+        if (Error Err =
+                FileHandler.get()->ReadBundle(DataStream, *CodeObjectBuffer))
+          return Err;
+
+        for (auto &CompatibleTarget : CompatibleTargets) {
+          SmallString<128> BundledObjectFileName;
+          BundledObjectFileName.assign(BundledObjectFile);
+          auto OutputBundleName =
+              Twine(llvm::sys::path::stem(BundledObjectFileName) + "-" +
+                    CodeObject)
+                  .str();
+          // Replace ':' in optional target feature list with '_' to ensure
+          // cross-platform validity.
+          std::replace(OutputBundleName.begin(), OutputBundleName.end(), ':',
+                       '_');
+
+          std::unique_ptr<MemoryBuffer> MemBuf = MemoryBuffer::getMemBufferCopy(
+              DataStream.str(), OutputBundleName);
+          ArchiveBuffers.push_back(std::move(MemBuf));
+          llvm::MemoryBufferRef MemBufRef =
+              MemoryBufferRef(*(ArchiveBuffers.back()));
+
+          // For inserting <CompatibleTarget, list<CodeObject>> entry in
+          // OutputArchivesMap.
+          if (OutputArchivesMap.find(CompatibleTarget) ==
+              OutputArchivesMap.end()) {
+
+            std::vector<NewArchiveMember> ArchiveMembers;
+            ArchiveMembers.push_back(NewArchiveMember(MemBufRef));
+            OutputArchivesMap.insert_or_assign(CompatibleTarget,
+                                               std::move(ArchiveMembers));
+          } else {
+            OutputArchivesMap[CompatibleTarget].push_back(
+                NewArchiveMember(MemBufRef));
+          }
+        }
+      }
+
+      if (Error Err = FileHandler.get()->ReadBundleEnd(*CodeObjectBuffer))
+        return Err;
+
+      Expected<Optional<StringRef>> NextTripleOrErr =
+          FileHandler->ReadBundleStart(*CodeObjectBuffer);
+      if (!NextTripleOrErr)
+        return NextTripleOrErr.takeError();
+
+      CodeObject = ((*NextTripleOrErr).hasValue()) ? **NextTripleOrErr : "";
+    } // End of processing of all bundle entries of this child of input archive.
+  }   // End of while over children of input archive.
+
+  assert(!ArchiveErr);
+
+  /// Write out an archive for each target
+  for (auto &Target : TargetNames) {
+    StringRef FileName = TargetOutputFileNameMap[Target];
+    StringMapIterator<std::vector<llvm::NewArchiveMember>> CurArchiveMembers =
+        OutputArchivesMap.find(Target);
+    if (CurArchiveMembers != OutputArchivesMap.end()) {
+      if (Error WriteErr = writeArchive(FileName, CurArchiveMembers->getValue(),
+                                        true, getDefaultArchiveKindForHost(),
+                                        true, false, nullptr))
+        return WriteErr;
+    } else if (!AllowMissingBundles) {
+      std::string ErrMsg =
+          Twine("no compatible code object found for the target '" + Target +
+                "' in heterogenous archive library: " + IFName)
+              .str();
+      return createStringError(inconvertibleErrorCode(), ErrMsg);
+    }
+  }
+
+  return Error::success();
+}
+
 static void PrintVersion(raw_ostream &OS) {
   OS << clang::getClangToolFullVersion("clang-offload-bundler") << '\n';
 }
 
 int main(int argc, const char **argv) {
   sys::PrintStackTraceOnErrorSignal(argv[0]);
 
   cl::HideUnrelatedOptions(ClangOffloadBundlerCategory);
if (InputFileNames.size() != 1) {
           "only one input file supported in unbundling mode"));
     }
     if (OutputFileNames.size() != TargetNames.size()) {
       reportError(createStringError(errc::invalid_argument,
                                     "number of output files and targets should "
                                     "match in unbundling mode"));
     }
   } else {
+    if (FilesType == "a") {
+      reportError(createStringError(errc::invalid_argument,
+                                    "Archive files are only supported "
+                                    "for unbundling"));
+    }
     if (OutputFileNames.size() != 1) {
       reportError(createStringError(
           errc::invalid_argument,
           "only one output file supported in bundling mode"));
     }
     if (InputFileNames.size() != TargetNames.size()) {
       reportError(createStringError(
           errc::invalid_argument,
           "number of input files and targets should match in bundling mode"));
     }
   }
 
   // Verify that the offload kinds and triples are known. We also check that we
   // have exactly one host target.
   unsigned Index = 0u;
   unsigned HostTargetNum = 0u;
   llvm::DenseSet<StringRef> ParsedTargets;
   for (StringRef Target : TargetNames) {
     if (ParsedTargets.contains(Target)) {
       reportError(createStringError(errc::invalid_argument,
                                     "Duplicate targets are not allowed"));
     }
     ParsedTargets.insert(Target);
 
-    StringRef Kind;
-    StringRef Triple;
-    getOffloadKindAndTriple(Target, Kind, Triple);
-
-    bool KindIsValid = !Kind.empty();
-    KindIsValid = KindIsValid && StringSwitch<bool>(Kind)
-                                     .Case("host", true)
-                                     .Case("openmp", true)
-                                     .Case("hip", true)
-                                     .Case("hipv4", true)
-                                     .Default(false);
-
-    bool TripleIsValid = !Triple.empty();
-    llvm::Triple T(Triple);
-    TripleIsValid &= T.getArch() != Triple::UnknownArch;
+    auto OffloadInfo = OffloadTargetInfo(Target);
+    bool KindIsValid = OffloadInfo.isOffloadKindValid();
+    bool TripleIsValid = OffloadInfo.isTripleValid();
 
     if (!KindIsValid || !TripleIsValid) {
       SmallVector<char, 128u> Buf;
       raw_svector_ostream Msg(Buf);
       Msg << "invalid target '" << Target << "'";
       if (!KindIsValid)
-        Msg << ", unknown offloading kind '" << Kind << "'";
+        Msg << ", unknown offloading kind '" << OffloadInfo.OffloadKind << "'";
       if (!TripleIsValid)
-        Msg << ", unknown target triple '" << Triple << "'";
+        Msg << ", unknown target triple '" << OffloadInfo.Triple.str() << "'";
       reportError(createStringError(errc::invalid_argument, Msg.str()));
     }
 
-    if (KindIsValid && Kind == "host") {
+    if (KindIsValid && OffloadInfo.hasHostKind()) {
       ++HostTargetNum;
       // Save the index of the input that refers to the host.
       HostInputIndex = Index;
     }
 
     ++Index;
   }
 
   // Host triple is not really needed for unbundling operation, so do not
   // treat missing host triple as error if we do unbundling.
   if ((Unbundle && HostTargetNum > 1) || (!Unbundle && HostTargetNum != 1)) {
     reportError(createStringError(errc::invalid_argument,
                                   "expecting exactly one host target but got " +
                                       Twine(HostTargetNum)));
   }
 
-  doWork([]() { return Unbundle ? UnbundleFiles() : BundleFiles(); });
+  doWork([]() {
+    return (Unbundle && FilesType == "a")
+               ? UnbundleArchive()
+               : Unbundle ? UnbundleFiles() : BundleFiles();
+  });
   return 0;
 }