[Clang] Default to async unwind tables for amdgcn #166464
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters. Learn more about bidirectional Unicode characters
To avoid codegen changes when enabling debug-info (see https://bugs.llvm.org/show_bug.cgi?id=37240) we want to enable unwind tables by default. There is some pessimization in post-prologepilog scheduling, and a general solution to the problem of CFI_INSTRUCTION-as-scheduling-barrier should be explored.
This was referenced Nov 4, 2025
| Warning This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
This stack of pull requests is managed by Graphite. Learn more about stacking. |
llvmbot added clang 
| @llvm/pr-subscribers-clang-driver @llvm/pr-subscribers-backend-amdgpu Author: Scott Linder (slinder1) ChangesTo avoid codegen changes when enabling debug-info (see There is some pessimization in post-prologepilog scheduling, and a Full diff: https://github.com/llvm/llvm-project/pull/166464.diff 2 Files Affected:
diff --git a/clang/lib/Driver/ToolChains/Gnu.cpp b/clang/lib/Driver/ToolChains/Gnu.cpp index 7616076847a2c..51f9ee0de5797 100644 --- a/clang/lib/Driver/ToolChains/Gnu.cpp +++ b/clang/lib/Driver/ToolChains/Gnu.cpp @@ -3032,6 +3032,7 @@ Generic_GCC::getDefaultUnwindTableLevel(const ArgList &Args) const { switch (getArch()) { case llvm::Triple::aarch64: case llvm::Triple::aarch64_be: + case llvm::Triple::amdgcn: case llvm::Triple::ppc: case llvm::Triple::ppcle: case llvm::Triple::ppc64: diff --git a/clang/test/Driver/amdgpu-unwind.cl b/clang/test/Driver/amdgpu-unwind.cl new file mode 100644 index 0000000000000..611d9735b15a4 --- /dev/null +++ b/clang/test/Driver/amdgpu-unwind.cl @@ -0,0 +1,26 @@ +// REQUIRES: amdgpu-registered-target + +// Default options +// RUN: %clang -### -x cl -target amdgcn-amd-amdhsa -c -emit-llvm --no-offloadlib %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s +// RUN: %clang -### -x hip --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s +// RUN: %clang -### -fopenmp --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s + +// Explicitly enable sync-tables (somewhat surprisingly this is still preempted by the default-on async tables) +// RUN: %clang -### -x cl -target amdgcn-amd-amdhsa -c -emit-llvm --no-offloadlib -funwind-tables %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s +// RUN: %clang -### -x hip --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -funwind-tables %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s +// RUN: %clang -### -fopenmp --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -funwind-tables %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s + +// Explicitly enable sync-tables and surpress default async-tables +// RUN: %clang -### -x cl -target amdgcn-amd-amdhsa -c -emit-llvm --no-offloadlib -funwind-tables -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=SYNC-TABLES %s +// RUN: %clang -### -x hip --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -funwind-tables -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=SYNC-TABLES %s +// RUN: %clang -### -fopenmp --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -funwind-tables -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=SYNC-TABLES %s + +// Suppress the default async-tables +// RUN: %clang -### -x cl -target amdgcn-amd-amdhsa -c -emit-llvm --no-offloadlib -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=NO-TABLES %s +// RUN: %clang -### -x hip --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=NO-TABLES %s +// RUN: %clang -### -fopenmp --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=NO-TABLES %s + + +// ASYNC-TABLES: "-triple" "amdgcn-amd-amdhsa" {{.*}} "-funwind-tables=2" +// SYNC-TABLES: "-triple" "amdgcn-amd-amdhsa" {{.*}} "-funwind-tables=1" +// NO-TABLES-NOT: "-triple" "amdgcn-amd-amdhsa" {{.*}} "-funwind-tables={{.*}}" |
| @llvm/pr-subscribers-clang Author: Scott Linder (slinder1) ChangesTo avoid codegen changes when enabling debug-info (see There is some pessimization in post-prologepilog scheduling, and a Full diff: https://github.com/llvm/llvm-project/pull/166464.diff 2 Files Affected:
diff --git a/clang/lib/Driver/ToolChains/Gnu.cpp b/clang/lib/Driver/ToolChains/Gnu.cpp index 7616076847a2c..51f9ee0de5797 100644 --- a/clang/lib/Driver/ToolChains/Gnu.cpp +++ b/clang/lib/Driver/ToolChains/Gnu.cpp @@ -3032,6 +3032,7 @@ Generic_GCC::getDefaultUnwindTableLevel(const ArgList &Args) const { switch (getArch()) { case llvm::Triple::aarch64: case llvm::Triple::aarch64_be: + case llvm::Triple::amdgcn: case llvm::Triple::ppc: case llvm::Triple::ppcle: case llvm::Triple::ppc64: diff --git a/clang/test/Driver/amdgpu-unwind.cl b/clang/test/Driver/amdgpu-unwind.cl new file mode 100644 index 0000000000000..611d9735b15a4 --- /dev/null +++ b/clang/test/Driver/amdgpu-unwind.cl @@ -0,0 +1,26 @@ +// REQUIRES: amdgpu-registered-target + +// Default options +// RUN: %clang -### -x cl -target amdgcn-amd-amdhsa -c -emit-llvm --no-offloadlib %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s +// RUN: %clang -### -x hip --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s +// RUN: %clang -### -fopenmp --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s + +// Explicitly enable sync-tables (somewhat surprisingly this is still preempted by the default-on async tables) +// RUN: %clang -### -x cl -target amdgcn-amd-amdhsa -c -emit-llvm --no-offloadlib -funwind-tables %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s +// RUN: %clang -### -x hip --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -funwind-tables %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s +// RUN: %clang -### -fopenmp --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -funwind-tables %s 2>&1 | FileCheck -check-prefix=ASYNC-TABLES %s + +// Explicitly enable sync-tables and surpress default async-tables +// RUN: %clang -### -x cl -target amdgcn-amd-amdhsa -c -emit-llvm --no-offloadlib -funwind-tables -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=SYNC-TABLES %s +// RUN: %clang -### -x hip --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -funwind-tables -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=SYNC-TABLES %s +// RUN: %clang -### -fopenmp --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -funwind-tables -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=SYNC-TABLES %s + +// Suppress the default async-tables +// RUN: %clang -### -x cl -target amdgcn-amd-amdhsa -c -emit-llvm --no-offloadlib -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=NO-TABLES %s +// RUN: %clang -### -x hip --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=NO-TABLES %s +// RUN: %clang -### -fopenmp --offload-arch=gfx90a --offload-device-only --no-offloadlib -c -emit-llvm -fno-asynchronous-unwind-tables %s 2>&1 | FileCheck -check-prefix=NO-TABLES %s + + +// ASYNC-TABLES: "-triple" "amdgcn-amd-amdhsa" {{.*}} "-funwind-tables=2" +// SYNC-TABLES: "-triple" "amdgcn-amd-amdhsa" {{.*}} "-funwind-tables=1" +// NO-TABLES-NOT: "-triple" "amdgcn-amd-amdhsa" {{.*}} "-funwind-tables={{.*}}" |
Don't really understand how this is connected to this; that's just a general bug. What does this actually do in the backend? |
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Add this suggestion to a batch that can be applied as a single commit. This suggestion is invalid because no changes were made to the code. Suggestions cannot be applied while the pull request is closed. Suggestions cannot be applied while viewing a subset of changes. Only one suggestion per line can be applied in a batch. Add this suggestion to a batch that can be applied as a single commit. Applying suggestions on deleted lines is not supported. You must change the existing code in this line in order to create a valid suggestion. Outdated suggestions cannot be applied. This suggestion has been applied or marked resolved. Suggestions cannot be applied from pending reviews. Suggestions cannot be applied on multi-line comments. Suggestions cannot be applied while the pull request is queued to merge. Suggestion cannot be applied right now. Please check back later.
To avoid codegen changes when enabling debug-info (see
https://bugs.llvm.org/show_bug.cgi?id=37240) we want to
enable unwind tables by default.
There is some pessimization in post-prologepilog scheduling, and a
general solution to the problem of CFI_INSTRUCTION-as-scheduling-barrier
should be explored.