[clang][SYCL] Do not decompose SYCL functors unless necessary

clang/include/clang/Basic/LangOptions.def

@@ -312,7 +312,7 @@ ENUM_LANGOPT(SYCLVersion  , SYCLMajorVersion, 2, SYCL_None, "Version of the SYCL
 LANGOPT(DeclareSPIRVBuiltins, 1, 0, "Declare SPIR-V builtin functions")
 LANGOPT(SYCLExplicitSIMD  , 1, 0, "SYCL compilation with explicit SIMD extension")
 LANGOPT(EnableDAEInSpirKernels , 1, 0, "Enable Dead Argument Elimination in SPIR kernels")
-LANGOPT(SYCLDecomposeStruct, 1, 1, "Force top level decomposition of SYCL functor")
+LANGOPT(SYCLDecomposeStruct, 1, 0, "Force top level decomposition of SYCL functor")
 LANGOPT(
     SYCLValueFitInMaxInt, 1, 1,
     "SYCL compiler assumes value fits within MAX_INT for member function of "

clang/include/clang/Driver/Options.td

@@ -7106,11 +7106,11 @@ defm sycl_instrument_device_code
                     "(experimental)">>;
 defm sycl_decompose_functor
     : BoolFOption<"sycl-decompose-functor",
-          LangOpts<"SYCLDecomposeStruct">, DefaultTrue,
+          LangOpts<"SYCLDecomposeStruct">, DefaultFalse,
           PosFlag<SetTrue, [], [ClangOption, CLOption], "Do">,
           NegFlag<SetFalse, [], [ClangOption, CLOption], "Do not">,
           BothFlags<[], [ClangOption, CLOption, CC1Option],
-           " decompose SYCL functor if possible (experimental, CUDA only)">>;
+           " decompose SYCL functor if possible (default is false)">>;
 defm sycl_cuda_compat
     : BoolFOption<"sycl-cuda-compatibility", LangOpts<"SYCLCUDACompat">, DefaultFalse,
           PosFlag<SetTrue, [], [ClangOption, CLOption, CC1Option], "Enable CUDA compatibility mode (experimental). "

clang/lib/Sema/SemaSYCL.cpp

@@ -319,10 +319,7 @@ ExprResult SemaSYCL::BuildSYCLBuiltinBaseTypeExpr(SourceLocation Loc,
 
 /// Returns true if the target requires a new type.
 /// This happens if a pointer to generic cannot be passed
-static bool targetRequiresNewType(ASTContext &Context) {
-  llvm::Triple T = Context.getTargetInfo().getTriple();
-  return !T.isNVPTX();
-}
+static bool targetRequiresNewType(ASTContext &Context) { return false; }
 
 // This information is from Section 4.13 of the SYCL spec
 // https://www.khronos.org/registry/SYCL/specs/sycl-1.2.1.pdf

clang/test/SemaSYCL/no-decomp.cpp

@@ -1,4 +1,5 @@
-// RUN: %clang_cc1 -fsycl-is-device -triple nvptx64-nvidia-cuda -ast-dump %s | FileCheck %s -check-prefix=ALL -check-prefix=DECOMP
+// RUN: %clang_cc1 -fsycl-is-device -triple nvptx64-nvidia-cuda -ast-dump %s | FileCheck %s -check-prefix=ALL -check-prefix=NODECOMP
+// RUN: %clang_cc1 -fsycl-is-device -triple spir64 -ast-dump %s | FileCheck %s -check-prefix=ALL -check-prefix=NODECOMP
 // RUN: %clang_cc1 -fsycl-is-device -fno-sycl-decompose-functor -triple nvptx64-nvidia-cuda -ast-dump %s | FileCheck %s -check-prefix=ALL -check-prefix=NODECOMP
 // RUN: %clang_cc1 -fsycl-is-device -fsycl-decompose-functor -triple nvptx64-nvidia-cuda -ast-dump %s | FileCheck %s -check-prefix=ALL -check-prefix=DECOMP
 

sycl/test-e2e/ESIMD/lsc/lsc_load_store_2d_smoke.cpp

@@ -7,7 +7,9 @@
 //===----------------------------------------------------------------------===//
 // REQUIRES: arch-intel_gpu_pvc
 // REQUIRES-INTEL-DRIVER: lin: 30508
-// RUN: %{build} -o %t.out
+// Shouldn't have to use -fsycl-decompose-functor,
+// See https://github.com/intel/llvm-test-suite/issues/18317
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: %{run} %t.out
 
 // Basic test for new lsc_load_2d/lsc_store_2d API.

sycl/test-e2e/ESIMD/unified_memory_api/scatter_usm.cpp

@@ -5,7 +5,9 @@
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===------------------------------------------------------------------===//
-// RUN: %{build} -fsycl-device-code-split=per_kernel -D__ESIMD_GATHER_SCATTER_LLVM_IR -o %t.out
+// Shouldn't have to use -fsycl-decompose-functor,
+// See https://github.com/intel/llvm-test-suite/issues/18317
+// RUN: %{build} -fsycl-device-code-split=per_kernel -fsycl-decompose-functor -D__ESIMD_GATHER_SCATTER_LLVM_IR -o %t.out
 // RUN: %{run} %t.out
 
 // The test verifies esimd::scatter() functions accepting USM pointer

sycl/test-e2e/ESIMD/usm_gather_scatter_rgba.cpp

@@ -5,7 +5,9 @@
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
-// RUN: %{build} -o %t.out
+// Shouldn't have to use -fsycl-decompose-functor,
+// See https://github.com/intel/llvm-test-suite/issues/18317
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: %{run} %t.out
 //
 // The test checks functionality of the gather_rgba/scatter_rgba USM-based ESIMD

sycl/test-e2e/ESIMD/usm_gather_scatter_rgba_64.cpp

@@ -5,7 +5,9 @@
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
-// RUN: %{build} -o %t.out
+// Shouldn't have to use -fsycl-decompose-functor,
+// See https://github.com/intel/llvm-test-suite/issues/18317
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: %{run} %t.out
 //
 // The test checks functionality of the gather_rgba/scatter_rgba USM-based ESIMD

sycl/test-e2e/ESIMD/vadd_raw_send_gen12.cpp

@@ -6,7 +6,9 @@
 //
 //===---------------------------------------------------------===//
 // REQUIRES: gpu-intel-gen12
-// RUN: %{build} -o %t.out
+// Shouldn't have to use -fsycl-decompose-functor,
+// See https://github.com/intel/llvm-test-suite/issues/18317
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: %{run} %t.out
 
 #include "esimd_test_utils.hpp"

sycl/test-e2e/Graph/Update/dyn_cgf_different_arg_nums.cpp

@@ -1,4 +1,7 @@
-// RUN: %{build} -o %t.out
+// By default functors are no longer decomposed preventing the use of set_arg in
+// this test, -fsycl-decompose-functor is used to force the old behavior
+//
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: env SYCL_UR_TRACE=2 %{run} %t.out | FileCheck %s
 // Extra run to check for leaks in Level Zero using UR_L0_LEAKS_DEBUG
 // RUN: %if level_zero %{env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 env SYCL_UR_TRACE=2 %{l0_leak_check} %{run} %t.out 2>&1 | FileCheck %s --implicit-check-not=LEAK %}

sycl/test-e2e/Graph/Update/dyn_cgf_with_different_type_dyn_params.cpp

@@ -1,4 +1,7 @@
-// RUN: %{build} -o %t.out
+// By default functors are no longer decomposed preventing the use of set_arg in
+// this test, -fsycl-decompose-functor is used to force the old behavior
+//
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using UR_L0_LEAKS_DEBUG
 // 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 %}

sycl/test-e2e/Graph/Update/update_nullptr.cpp

@@ -1,4 +1,7 @@
-// RUN: %{build} -o %t.out
+// By default functors are no longer decomposed preventing the use of set_arg in
+// this test, -fsycl-decompose-functor is used to force the old behavior
+//
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using UR_L0_LEAKS_DEBUG
 // 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 %}

sycl/test-e2e/Graph/Update/update_with_indices_ptr_multiple_nodes_different_indices.cpp

@@ -1,4 +1,7 @@
-// RUN: %{build} -o %t.out
+// By default functors are no longer decomposed preventing the use of set_arg in
+// this test, -fsycl-decompose-functor is used to force the old behavior
+//
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using UR_L0_LEAKS_DEBUG
 // 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 %}

sycl/test-e2e/Graph/Update/update_with_indices_ptr_multiple_params.cpp

@@ -1,4 +1,7 @@
-// RUN: %{build} -o %t.out
+// By default functors are no longer decomposed preventing the use of set_arg in
+// this test, -fsycl-decompose-functor is used to force the old behavior
+//
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using UR_L0_LEAKS_DEBUG
 // 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 %}

sycl/test-e2e/Graph/Update/update_with_indices_scalar.cpp

@@ -1,4 +1,7 @@
-// RUN: %{build} -o %t.out
+// By default functors are no longer decomposed preventing the use of set_arg in
+// this test, -fsycl-decompose-functor is used to force the old behavior
+//
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using UR_L0_LEAKS_DEBUG
 // 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 %}

sycl/test-e2e/Graph/Update/whole_update_dynamic_param.cpp

@@ -1,4 +1,7 @@
-// RUN: %{build} -o %t.out
+// By default functors are no longer decomposed preventing the use of set_arg in
+// this test, -fsycl-decompose-functor is used to force the old behavior
+//
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: %{run} %t.out
 // Extra run to check for leaks in Level Zero using UR_L0_LEAKS_DEBUG
 // 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 %}

sycl/test-e2e/Tracing/usm/queue_single_task_nullptr.cpp

@@ -1,5 +1,5 @@
 // UNSUPPORTED: windows || target-amd
-// RUN: %{build} -o %t.out
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: not --crash env SYCL_TRACE_TERMINATE_ON_WARNING=1 %{run} sycl-trace --verify %t.out | FileCheck %s
 
 // Test parameter analysis of USM usage

sycl/test-e2e/Tracing/usm/queue_single_task_released_pointer.cpp

@@ -1,5 +1,5 @@
 // UNSUPPORTED: windows || hip
-// RUN: %{build} -o %t.out
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: not --crash env SYCL_TRACE_TERMINATE_ON_WARNING=1 %{run} sycl-trace --verify %t.out | FileCheck %s
 
 // Test parameter analysis of USM usage

sycl/test-e2e/XPTI/basic_event_collection_linux.cpp

@@ -1,6 +1,6 @@
 // REQUIRES: xptifw, opencl, cpu, linux
 // RUN: %build_collector
-// RUN: %{build} -o %t.out
+// RUN: %{build} -fsycl-decompose-functor -o %t.out
 // RUN: env UR_ENABLE_LAYERS=UR_LAYER_TRACING env XPTI_TRACE_ENABLE=1 env XPTI_FRAMEWORK_DISPATCHER=%xptifw_dispatcher env XPTI_SUBSCRIBERS=%t_collector.dll %{run} %t.out | FileCheck %s
 
 #include "basic_event_collection.inc"

sycl/test/check_device_code/extensions/properties/properties_cache_control.cpp

@@ -82,19 +82,19 @@ void cache_control_read_write_func() {
 }
 
 // CHECK-IR: spir_kernel{{.*}}cache_control_read_hint_func
-// CHECK-IR: {{.*}}addrspacecast ptr addrspace(1){{.*}}!spirv.Decorations [[RHINT:.*]]
+// CHECK-IR: {{.*}}load ptr addrspace(4), ptr{{.*}}!spirv.Decorations [[RHINT:.*]]
 // CHECK-IR: ret void
 
 // CHECK-IR: spir_kernel{{.*}}cache_control_read_assertion_func
-// CHECK-IR: {{.*}}addrspacecast ptr addrspace(1){{.*}}!spirv.Decorations [[RASSERT:.*]]
+// CHECK-IR: {{.*}}load ptr addrspace(4), ptr{{.*}}!spirv.Decorations [[RASSERT:.*]]
 // CHECK-IR: ret void
 
 // CHECK-IR: spir_kernel{{.*}}cache_control_write_hint_func
-// CHECK-IR: {{.*}}addrspacecast ptr addrspace(1){{.*}}!spirv.Decorations [[WHINT:.*]]
+// CHECK-IR: {{.*}}load ptr addrspace(4), ptr{{.*}}!spirv.Decorations [[WHINT:.*]]
 // CHECK-IR: ret void
 
 // CHECK-IR: spir_kernel{{.*}}cache_control_read_write_func
-// CHECK-IR: {{.*}}addrspacecast ptr addrspace(1){{.*}}!spirv.Decorations [[RWHINT:.*]]
+// CHECK-IR: {{.*}}load ptr addrspace(4), ptr{{.*}}!spirv.Decorations [[RWHINT:.*]]
 // CHECK-IR: ret void
 
 // CHECK-IR: [[RHINT]] = !{[[RHINT1:.*]], [[RHINT2:.*]], [[RHINT3:.*]]}