[SYCL] Free functions and dynamic linking fixes for CUDA/HIP (intel#17899)

Jakub Chlanda · web-flow · commit fc328cc29b2d · 2025-04-11T10:35:29.000+01:00
diff --git a/llvm/lib/SYCLPostLink/ModuleSplitter.cpp b/llvm/lib/SYCLPostLink/ModuleSplitter.cpp
@@ -121,8 +121,9 @@ bool isGenericBuiltin(StringRef FName) {
 }
 
 bool isKernel(const Function &F) {
-  return F.getCallingConv() == CallingConv::SPIR_KERNEL ||
-         F.getCallingConv() == CallingConv::AMDGPU_KERNEL;
+  const auto CC = F.getCallingConv();
+  return CC == CallingConv::SPIR_KERNEL || CC == CallingConv::AMDGPU_KERNEL ||
+         CC == CallingConv::PTX_Kernel;
 }
 
 bool isEntryPoint(const Function &F, bool EmitOnlyKernelsAsEntryPoints) {
@@ -697,9 +698,13 @@ static bool mustPreserveGV(const GlobalValue &GV) {
     // kernels which are the entry points from host code to device code) that
     // cannot be imported which also means that there is no point of having it
     // visible outside of the current module.
-    if (AllowDeviceImageDependencies)
-      return F->getCallingConv() == CallingConv::SPIR_KERNEL ||
-             canBeImportedFunction(*F);
+    if (AllowDeviceImageDependencies) {
+      const auto CC = F->getCallingConv();
+      const bool SpirOrGPU = CC == CallingConv::SPIR_KERNEL ||
+                             CC == CallingConv::AMDGPU_KERNEL ||
+                             CC == CallingConv::PTX_Kernel;
+      return SpirOrGPU || canBeImportedFunction(*F);
+    }
 
     // Otherwise, we are being even more aggressive: SYCL modules are expected
     // to be self-contained, meaning that they have no external dependencies.
diff --git a/sycl/test-e2e/DeviceImageDependencies/NewOffloadDriver/free_function_kernels.cpp b/sycl/test-e2e/DeviceImageDependencies/NewOffloadDriver/free_function_kernels.cpp
@@ -1,16 +1,12 @@
+// UNSUPPORTED: true
+// UNSUPPORTED-TRACKER: https://github.com/intel/llvm/issues/17812
+
 // Ensure -fsycl-allow-device-dependencies can work with free function kernels.
 
 // REQUIRES: aspect-usm_shared_allocations
 // RUN: %{build} -o %t.out --offload-new-driver -fsycl-allow-device-image-dependencies
 // RUN: %{run} %t.out
 
-// The name mangling for free function kernels currently does not work with PTX.
-// UNSUPPORTED: target-nvidia
-// UNSUPPORTED-INTENDED: Not implemented yet for Nvidia/AMD backends.
-
-// XFAIL: target-amd
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/15742
-
 #include <iostream>
 #include <sycl/detail/core.hpp>
 #include <sycl/ext/oneapi/free_function_queries.hpp>
diff --git a/sycl/test-e2e/DeviceImageDependencies/free_function_kernels.cpp b/sycl/test-e2e/DeviceImageDependencies/free_function_kernels.cpp
@@ -4,12 +4,6 @@
 // RUN: %{build} -o %t.out -fsycl-allow-device-image-dependencies
 // RUN: %{run} %t.out
 
-// The name mangling for free function kernels currently does not work with PTX.
-// UNSUPPORTED: cuda
-
-// XFAIL: hip
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/15742
-
 #include <iostream>
 #include <sycl/detail/core.hpp>
 #include <sycl/ext/oneapi/free_function_queries.hpp>
diff --git a/sycl/test-e2e/Graph/Explicit/free_function_kernels.cpp b/sycl/test-e2e/Graph/Explicit/free_function_kernels.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 #define GRAPH_E2E_EXPLICIT
 
diff --git a/sycl/test-e2e/Graph/Explicit/work_group_memory_free_function.cpp b/sycl/test-e2e/Graph/Explicit/work_group_memory_free_function.cpp
@@ -5,9 +5,6 @@
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
-
 #define GRAPH_E2E_EXPLICIT
 
 #include "../Inputs/work_group_memory_free_function.cpp"
diff --git a/sycl/test-e2e/Graph/RecordReplay/free_function_kernels.cpp b/sycl/test-e2e/Graph/RecordReplay/free_function_kernels.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 #define GRAPH_E2E_RECORD_REPLAY
 
diff --git a/sycl/test-e2e/Graph/RecordReplay/work_group_memory_free_function.cpp b/sycl/test-e2e/Graph/RecordReplay/work_group_memory_free_function.cpp
@@ -5,9 +5,6 @@
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
-
 #define GRAPH_E2E_RECORD_REPLAY
 
 #include "../Inputs/work_group_memory_free_function.cpp"
diff --git a/sycl/test-e2e/Graph/Update/FreeFunctionKernels/dyn_work_group_memory_basic.cpp b/sycl/test-e2e/Graph/Update/FreeFunctionKernels/dyn_work_group_memory_basic.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 // Tests updating dynamic_work_group_memory with a new size.
 
diff --git a/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_before_finalize.cpp b/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_before_finalize.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 // Tests updating a graph node before finalization
 
diff --git a/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_multiple_exec_graphs.cpp b/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_multiple_exec_graphs.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 // Tests creating multiple executable graphs from the same modifiable graph and
 // only updating one of them.
diff --git a/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ordering.cpp b/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ordering.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 // Tests that updating a graph is ordered with respect to previous executions of
 // the graph which may be in flight.
diff --git a/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr.cpp b/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 // Tests updating a graph node using index-based explicit update
 
diff --git a/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr_3D.cpp b/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr_3D.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 // Tests updating a 3D ND-Range graph kernel node using index-based explicit
 // update
diff --git a/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr_double_update.cpp b/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr_double_update.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 // Tests updating a graph node using index-based explicit update
 
diff --git a/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr_multiple_nodes.cpp b/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr_multiple_nodes.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 // Tests updating a single dynamic parameter which is registered with multiple
 // graph nodes
diff --git a/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr_multiple_params.cpp b/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr_multiple_params.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 // Tests updating multiple parameters to a singlegraph node using index-based
 // explicit update
diff --git a/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr_subgraph.cpp b/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_ptr_subgraph.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 // Tests updating a graph node in an executable graph that was used as a
 // subgraph node in another executable graph is not reflected in the graph
diff --git a/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_scalar.cpp b/sycl/test-e2e/Graph/Update/FreeFunctionKernels/update_with_indices_scalar.cpp
@@ -4,9 +4,6 @@
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
 // Extra run to check for immediate-command-list in Level Zero
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}
-//
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/16004
 
 // Tests updating a graph node scalar argument using index-based explicit update
 
diff --git a/sycl/test-e2e/KernelAndProgram/free_function_apis.cpp b/sycl/test-e2e/KernelAndProgram/free_function_apis.cpp
@@ -2,9 +2,6 @@
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 
-// The name mangling for free function kernels currently does not work with PTX.
-// UNSUPPORTED: cuda
-
 #include <iostream>
 #include <sycl/detail/core.hpp>
 #include <sycl/ext/oneapi/experimental/free_function_traits.hpp>
diff --git a/sycl/test-e2e/KernelAndProgram/free_function_kernels.cpp b/sycl/test-e2e/KernelAndProgram/free_function_kernels.cpp
@@ -2,9 +2,6 @@
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 
-// The name mangling for free function kernels currently does not work with PTX.
-// UNSUPPORTED: cuda
-
 // This test tests free function kernel code generation and execution.
 
 #include <iostream>
diff --git a/sycl/test-e2e/WorkGroupMemory/reduction_free_function.cpp b/sycl/test-e2e/WorkGroupMemory/reduction_free_function.cpp
@@ -2,9 +2,6 @@
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 
-// UNSUPPORTED: cuda
-// UNSUPPORTED-TRACKER: https://github.com/intel/llvm/issues/16004
-
 #include "common_free_function.hpp"
 
 // Basic usage reduction test using free function kernels.
diff --git a/sycl/test/e2e_test_requirements/no-unsupported-without-info.cpp b/sycl/test/e2e_test_requirements/no-unsupported-without-info.cpp
@@ -54,7 +54,7 @@
 // tests to match the required format and in that case you should just update
 // (i.e. reduce) the number and the list below.
 //
-// NUMBER-OF-UNSUPPORTED-WITHOUT-INFO: 276
+// NUMBER-OF-UNSUPPORTED-WITHOUT-INFO: 273
 //
 // List of improperly UNSUPPORTED tests.
 // Remove the CHECK once the test has been properly UNSUPPORTED.
@@ -93,7 +93,6 @@
 // CHECK-NEXT: Basic/kernel_info_attr.cpp
 // CHECK-NEXT: Basic/submit_time.cpp
 // CHECK-NEXT: DeviceImageDependencies/dynamic.cpp
-// CHECK-NEXT: DeviceImageDependencies/free_function_kernels.cpp
 // CHECK-NEXT: DeviceImageDependencies/math_device_lib.cpp
 // CHECK-NEXT: DeviceImageDependencies/objects.cpp
 // CHECK-NEXT: DeviceImageDependencies/singleDynamicLibrary.cpp
@@ -205,8 +204,6 @@
 // CHECK-NEXT: InvokeSimd/Regression/ImplicitSubgroup/call_vadd_1d_spill.cpp
 // CHECK-NEXT: InvokeSimd/Regression/call_vadd_1d_spill.cpp
 // CHECK-NEXT: KernelAndProgram/cache-build-result.cpp
-// CHECK-NEXT: KernelAndProgram/free_function_apis.cpp
-// CHECK-NEXT: KernelAndProgram/free_function_kernels.cpp
 // CHECK-NEXT: KernelAndProgram/kernel-bundle-merge-options-env.cpp
 // CHECK-NEXT: KernelAndProgram/kernel-bundle-merge-options.cpp
 // CHECK-NEXT: KernelAndProgram/level-zero-static-link-flow.cpp
