Skip to content

[CUDA] refactor fatbinary to use --image3 #20996

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
bratpiorka wants to merge 2 commits into from
Closed

[CUDA] refactor fatbinary to use --image3 #20996

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5fb55f0
fix sm
bratpiorka Jan 13, 2026
d50868c
fix fatbinary
bratpiorka Jan 14, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions clang/test/Driver/clang-linker-wrapper.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -121,7 +121,7 @@
// CHK-CMDS-AOT-NV-NEXT: sycl-post-link{{.*}} SYCL_POST_LINK_OPTIONS -o [[SYCLPOSTLINKOUT:.*]].table [[SECONDLLVMLINKOUT]].bc
// CHK-CMDS-AOT-NV-NEXT: clang{{.*}} -o [[CLANGOUT:.*]] -dumpdir a.out.nvptx64.sm_50.img. --target=nvptx64-nvidia-cuda -march={{.*}}
// CHK-CMDS-AOT-NV-NEXT: ptxas{{.*}} --output-file [[PTXASOUT:.*]] [[CLANGOUT]]
// CHK-CMDS-AOT-NV-NEXT: fatbinary{{.*}} --create [[FATBINOUT:.*]] --image=profile={{.*}},file=[[CLANGOUT]] --image=profile={{.*}},file=[[PTXASOUT]]
// CHK-CMDS-AOT-NV-NEXT: fatbinary{{.*}} --create [[FATBINOUT:.*]] --image3=kind=ptx,sm={{(compute_)?50|pute_50}},file=[[CLANGOUT]] --image3=kind=elf,sm=50,file=[[PTXASOUT]]
// CHK-CMDS-AOT-NV-NEXT: offload-wrapper: output: [[WRAPPEROUT:.*]].bc, input: [[FATBINOUT]]
// CHK-CMDS-AOT-NV-NEXT: clang{{.*}} -c -o [[LLCOUT:.*]] [[WRAPPEROUT]]
// CHK-CMDS-AOT-NV-NEXT: "{{.*}}/ld" -- HOST_LINKER_FLAGS -dynamic-linker HOST_DYN_LIB -o a.out [[LLCOUT]] HOST_LIB_PATH HOST_STAT_LIB {{.*}}.o
Expand Down Expand Up @@ -166,7 +166,7 @@
// CHK-CMDS-AOT-NV-EMBED-IR-NEXT: clang{{.*}} -c -o [[LLCOUT1:.*]] [[WRAPPEROUT1]]
// CHK-CMDS-AOT-NV-EMBED-IR-NEXT: clang{{.*}} -o [[CLANGOUT:.*]] -dumpdir a.out.nvptx64.sm_50.img. --target=nvptx64-nvidia-cuda -march={{.*}}
// CHK-CMDS-AOT-NV-EMBED-IR-NEXT: ptxas{{.*}} --output-file [[PTXASOUT:.*]] [[CLANGOUT]]
// CHK-CMDS-AOT-NV-EMBED-IR-NEXT: fatbinary{{.*}} --create [[FATBINOUT:.*]] --image=profile={{.*}},file=[[CLANGOUT]] --image=profile={{.*}},file=[[PTXASOUT]]
// CHK-CMDS-AOT-NV-EMBED-IR-NEXT: fatbinary{{.*}} --create [[FATBINOUT:.*]] --image3=kind=ptx,sm={{(compute_)?50|pute_50}},file=[[CLANGOUT]] --image3=kind=elf,sm=50,file=[[PTXASOUT]]
// CHK-CMDS-AOT-NV-EMBED-IR-NEXT: offload-wrapper: output: [[WRAPPEROUT:.*]].bc, input: [[FATBINOUT]]
// CHK-CMDS-AOT-NV-EMBED-IR-NEXT: clang{{.*}} -c -o [[LLCOUT2:.*]] [[WRAPPEROUT]]
// CHK-CMDS-AOT-NV-EMBED-IR-NEXT: "{{.*}}/ld" -- HOST_LINKER_FLAGS -dynamic-linker HOST_DYN_LIB -o a.out [[LLCOUT1]] [[LLCOUT2]] HOST_LIB_PATH HOST_STAT_LIB {{.*}}.o
Expand Down
4 changes: 2 additions & 2 deletions clang/test/Driver/linker-wrapper.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,7 +106,7 @@ __attribute__((visibility("protected"), used)) int x;

// CUDA: clang{{.*}} -o [[IMG_SM70:.+]] -dumpdir a.out.nvptx64.sm_70.img. --target=nvptx64-nvidia-cuda -march=sm_70
// CUDA: clang{{.*}} -o [[IMG_SM52:.+]] -dumpdir a.out.nvptx64.sm_52.img. --target=nvptx64-nvidia-cuda -march=sm_52
// CUDA: fatbinary{{.*}}-64 --create {{.*}}.fatbin --image=profile=sm_70,file=[[IMG_SM70]] --image=profile=sm_52,file=[[IMG_SM52]]
// CUDA: fatbinary{{.*}}-64 --create {{.*}}.fatbin --image3=kind=elf,sm=70,file=[[IMG_SM70]] --image3=kind=elf,sm=52,file=[[IMG_SM52]]
// CUDA: usr/bin/ld{{.*}} {{.*}}.openmp.image.{{.*}}.o {{.*}}.cuda.image.{{.*}}.o

// RUN: llvm-offload-binary -o %t.out \
Expand Down Expand Up @@ -240,7 +240,7 @@ __attribute__((visibility("protected"), used)) int x;
// RUN: %t.o -o a.out 2>&1 | FileCheck %s --check-prefix=RELOCATABLE-LINK-CUDA

// RELOCATABLE-LINK-CUDA: clang{{.*}} -o {{.*}}.img -dumpdir a.out.nvptx64.sm_89.img. --target=nvptx64-nvidia-cuda
// RELOCATABLE-LINK-CUDA: fatbinary{{.*}} -64 --create {{.*}}.fatbin --image=profile=sm_89,file={{.*}}.img
// RELOCATABLE-LINK-CUDA: fatbinary{{.*}} -64 --create {{.*}}.fatbin --image3=kind=elf,sm=89,file={{.*}}.img
// RELOCATABLE-LINK-CUDA: /usr/bin/ld.lld{{.*}}-r
// RELOCATABLE-LINK-CUDA: llvm-objcopy{{.*}}a.out --remove-section .llvm.offloading

Expand Down
19 changes: 16 additions & 3 deletions clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -481,9 +481,22 @@ fatbinary(ArrayRef<std::pair<StringRef, StringRef>> InputFiles,
CmdArgs.push_back(Triple.isArch64Bit() ? "-64" : "-32");
CmdArgs.push_back("--create");
CmdArgs.push_back(*TempFileOrErr);
for (const auto &[File, Arch] : InputFiles)
CmdArgs.push_back(
Args.MakeArgString("--image=profile=" + Arch + ",file=" + File));

for (const auto &[File, Arch] : InputFiles) {
// When bundling SYCL NVPTX, we may get both a cubin (sm_*) and a PTX
// assembly image (compute_*). `fatbinary` needs the right kind for each.
StringRef Kind = "elf";
StringRef ArchId = Arch;
if (Arch.starts_with("sm_")) {
ArchId = Arch.drop_front(3);
} else if (Arch.starts_with("compute_")) {
Kind = "ptx";
ArchId = Arch.drop_front(8);
}

CmdArgs.push_back(Args.MakeArgString("--image3=kind=" + Kind +
",sm=" + ArchId + ",file=" + File));
}

if (Error Err = executeCommands(*FatBinaryPath, CmdArgs))
return std::move(Err);
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/AtomicRef/atomic_memory_order_acq_rel.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -O3 -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_70 %}
// RUN: %{build} -O3 -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

// NOTE: Tests fetch_add for acquire and release memory ordering.
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/AtomicRef/atomic_memory_order_seq_cst.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -O3 -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_70 %}
// RUN: %{build} -O3 -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

#include "atomic_memory_order.h"
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/GroupAlgorithm/root_group.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@
// XFAIL: (opencl && !cpu)
// XFAIL-TRACKER: https://github.com/intel/llvm/issues/14641

// RUN: %{build} -I . -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_70 %}
// RUN: %{build} -I . -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

// Disabled temporarily while investigation into the failure is ongoing.
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Matrix/joint_matrix_tensorcores_sm70.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
//===----------------------------------------------------------------------===//

// REQUIRES: target-nvidia
// RUN: %{build} -Xsycl-target-backend --cuda-gpu-arch=sm_70 -o %t.out
// RUN: %{build} -Xsycl-target-backend --cuda-gpu-arch=sm_75 -o %t.out
// RUN: %{run} %t.out
//
// This tests the unified matrix extension interfaces for the cuda backend.
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Matrix/joint_matrix_tensorcores_sm72.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
//===----------------------------------------------------------------------===//

// REQUIRES: target-nvidia
// RUN: %{build} -Xsycl-target-backend --cuda-gpu-arch=sm_72 -o %t.out
// RUN: %{build} -Xsycl-target-backend --cuda-gpu-arch=sm_75 -o %t.out
// RUN: %{run} %t.out
//
// This tests the unified matrix extension interfaces for the cuda backend.
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Reduction/reduction_range_1d_dw.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -DENABLE_64_BIT=false -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60 %}
// RUN: %{build} -DENABLE_64_BIT=false -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

#include "reduction_utils.hpp"
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Reduction/reduction_range_1d_dw_64bit.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -DENABLE_64_BIT=true -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60 %}
// RUN: %{build} -DENABLE_64_BIT=true -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

#include "reduction_range_1d_dw.cpp"
2 changes: 1 addition & 1 deletion sycl/test-e2e/Reduction/reduction_range_1d_reducer_skip.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60 %}
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

// This test performs basic checks of parallel_for(range<1>, reduction, func)
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Reduction/reduction_range_1d_rw.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60 %}
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

// This test performs basic checks of parallel_for(range<1>, reduction, func)
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Reduction/reduction_range_2d_dw.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60 %}
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

// This test performs basic checks of parallel_for(range<2>, reduction, func)
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Reduction/reduction_range_2d_dw_reducer_skip.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60 %}
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

// This test performs basic checks of parallel_for(range<2>, reduction, func)
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Reduction/reduction_range_2d_rw.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60 %}
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

// This test performs basic checks of parallel_for(range<2>, reduction, func)
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Reduction/reduction_range_3d_dw.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60 %}
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

// This test performs basic checks of parallel_for(range<3>, reduction, func)
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Reduction/reduction_range_3d_rw.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60 %}
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

// This test performs basic checks of parallel_for(range<3>, reduction, func)
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Reduction/reduction_range_3d_rw_reducer_skip.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60 %}
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

// This test performs basic checks of parallel_for(range<3>, reduction, func)
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/Reduction/reduction_range_usm_dw.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60 %}
// RUN: %{build} -o %t.out %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %}
// RUN: %{run} %t.out

#include "reduction_utils.hpp"
Expand Down
2 changes: 1 addition & 1 deletion sycl/test-e2e/USM/P2P/p2p_atomics.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
// REQUIRES: cuda || hip || level_zero
// RUN: %{build} %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_61 %} -o %t.out
// RUN: %{build} %if target-nvidia %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75 %} -o %t.out
// RUN: %{run} %t.out

#include <cassert>
Expand Down