[Driver][SYCL] Enable Dead Parameter Elimination Optimization

clang/include/clang/Driver/Options.td

@@ -3551,7 +3551,12 @@ def fsycl_esimd : Flag<["-"], "fsycl-explicit-simd">, Group<sycl_Group>, Flags<[
 def fno_sycl_esimd : Flag<["-"], "fno-sycl-explicit-simd">, Group<sycl_Group>,
   HelpText<"Disable SYCL explicit SIMD extension">, Flags<[NoArgumentUnused, CoreOption]>;
 defm sycl_early_optimizations : OptOutFFlag<"sycl-early-optimizations", "Enable", "Disable", " standard optimization pipeline for SYCL device compiler", [CoreOption]>;
-
+def fsycl_dead_args_optimization : Flag<["-"], "fsycl-dead-args-optimization">,
+  Group<sycl_Group>, Flags<[NoArgumentUnused, CoreOption]>, HelpText<"Enables "
+  "elimination of DPC++ dead kernel arguments">;
+def fno_sycl_dead_args_optimization : Flag<["-"], "fno-sycl-dead-args-optimization">,
+  Group<sycl_Group>, Flags<[NoArgumentUnused, CoreOption]>, HelpText<"Disables "
+  "elimination of DPC++ dead kernel arguments">;
 //===----------------------------------------------------------------------===//
 // CC1 Options
 //===----------------------------------------------------------------------===//

clang/lib/Driver/Driver.cpp

@@ -3514,6 +3514,9 @@ class OffloadingActionBuilder final {
     /// Flag to signal if the user requested device code split.
     bool DeviceCodeSplit = false;
 
+    /// Flag to signal if DAE optimization is turned on.
+    bool EnableDAE = false;
+
     /// The SYCL actions for the current input.
     ActionList SYCLDeviceActions;
 
@@ -3951,7 +3954,7 @@ class OffloadingActionBuilder final {
         ActionList WrapperInputs;
         // post link is not optional - even if not splitting, always need to
         // process specialization constants
-        bool MultiFileActionDeps = !isSpirvAOT || DeviceCodeSplit;
+        bool MultiFileActionDeps = !isSpirvAOT || DeviceCodeSplit || EnableDAE;
         types::ID PostLinkOutType = isNVPTX || !MultiFileActionDeps
                                         ? types::TY_LLVM_BC
                                         : types::TY_Tempfiletable;
@@ -4108,6 +4111,9 @@ class OffloadingActionBuilder final {
       WrapDeviceOnlyBinary = Args.hasArg(options::OPT_fsycl_link_EQ);
       auto *DeviceCodeSplitArg =
           Args.getLastArg(options::OPT_fsycl_device_code_split_EQ);
+      EnableDAE =
+          Args.hasFlag(options::OPT_fsycl_dead_args_optimization,
+                       options::OPT_fno_sycl_dead_args_optimization, false);
       // -fsycl-device-code-split is an alias to
       // -fsycl-device-code-split=per_source
       DeviceCodeSplit = DeviceCodeSplitArg &&

clang/lib/Driver/ToolChains/Clang.cpp

@@ -4123,6 +4123,11 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       CmdArgs.push_back("-mllvm");
       CmdArgs.push_back("-sycl-opt");
     }
+    // Turn on Dead Parameter Elimination Optimization with early optimizations
+    if (!RawTriple.isNVPTX() &&
+        Args.hasFlag(options::OPT_fsycl_dead_args_optimization,
+                     options::OPT_fno_sycl_dead_args_optimization, false))
+      CmdArgs.push_back("-fenable-sycl-dae");
 
     // Pass the triple of host when doing SYCL
     auto AuxT = llvm::Triple(llvm::sys::getProcessTriple());
@@ -7807,6 +7812,11 @@ void SYCLPostLink::ConstructJob(Compilation &C, const JobAction &JA,
   // OPT_fsycl_device_code_split is not checked as it is an alias to
   // -fsycl-device-code-split=per_source
 
+  // Turn on Dead Parameter Elimination Optimization with early optimizations
+  if (!getToolChain().getTriple().isNVPTX() &&
+      TCArgs.hasFlag(options::OPT_fsycl_dead_args_optimization,
+                     options::OPT_fno_sycl_dead_args_optimization, false))
+    addArgs(CmdArgs, TCArgs, {"-emit-param-info"});
   if (JA.getType() == types::TY_LLVM_BC) {
     // single file output requested - this means only perform necessary IR
     // transformations (like specialization constant intrinsic lowering) and

clang/test/Driver/sycl-device-optimizations.cpp

@@ -28,3 +28,11 @@
 // RUN:   %clang_cl -### -fsycl -fintelfpga %s 2>&1 \
 // RUN:   | FileCheck -check-prefix=CHECK-NO-SYCL-EARLY-OPTS %s
 // CHECK-NO-SYCL-EARLY-OPTS: "-fno-sycl-early-optimizations"
+
+/// Check that Dead Parameter Elimination Optimization is enabled
+// RUN:   %clang -### -fsycl -fsycl-dead-args-optimization %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHECK-DAE %s
+// RUN:   %clang_cl -### -fsycl -fsycl-dead-args-optimization %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHECK-DAE %s
+// CHECK-DAE: clang{{.*}} "-fenable-sycl-dae"
+// CHECK-DAE: sycl-post-link{{.*}} "-emit-param-info"

sycl/test/basic_tests/sampler/sampler.cpp

@@ -1,4 +1,4 @@
-// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out -L %opencl_libs_dir -lOpenCL
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -fsycl-dead-args-optimization %s -o %t.out -L %opencl_libs_dir -lOpenCL
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out

sycl/test/kernel_from_file/hw.cpp

@@ -2,7 +2,9 @@
 // CUDA does not support SPIR-V.
 
 //-fsycl-targets=%sycl_triple
-// RUN: %clangxx -fsycl-device-only -fno-sycl-use-bitcode -Xclang -fsycl-int-header=%t.h -c %s -o %t.spv -I %sycl_include -Xclang -verify-ignore-unexpected=note,warning -Wno-sycl-strict
+// Runtime assumes that the image passed with SYCL_USE_KERNEL_SPV has no
+// eliminated arguments, compile without early optimizations.
+// RUN: %clangxx -fsycl-device-only -fno-sycl-early-optimizations -fno-sycl-use-bitcode -Xclang -fsycl-int-header=%t.h -c %s -o %t.spv -I %sycl_include -Xclang -verify-ignore-unexpected=note,warning -Wno-sycl-strict
 // RUN: %clangxx -include %t.h -g %s -o %t.out -lsycl -I %sycl_include -Xclang -verify-ignore-unexpected=note,warning
 // RUN: env SYCL_BE=%sycl_be SYCL_USE_KERNEL_SPV=%t.spv %t.out | FileCheck %s
 // CHECK: Passed

sycl/test/multi_ptr/multi_ptr.cpp

@@ -1,9 +1,9 @@
-// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -fsycl-dead-args-optimization %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
-// RUN: %clangxx -DRESTRICT_WRITE_ACCESS_TO_CONSTANT_PTR -fsycl -fsycl-targets=%sycl_triple %s -o %t1.out
+// RUN: %clangxx -DRESTRICT_WRITE_ACCESS_TO_CONSTANT_PTR -fsycl -fsycl-targets=%sycl_triple -fsycl-dead-args-optimization %s -o %t1.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t1.out
 // RUN: %CPU_RUN_PLACEHOLDER %t1.out
 // RUN: %GPU_RUN_PLACEHOLDER %t1.out

sycl/test/scheduler/HandleException.cpp

@@ -1,4 +1,4 @@
-// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -I %sycl_source_dir %s -o %t.out
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -fsycl-dead-args-optimization -I %sycl_source_dir %s -o %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 #include <CL/sycl.hpp>
 #include <array>

sycl/test/scheduler/HostAccDestruction.cpp

@@ -1,4 +1,4 @@
-// RUN: %clangxx -fsycl -I %sycl_source_dir %s -o %t.out
+// RUN: %clangxx -fsycl -fsycl-dead-args-optimization -I %sycl_source_dir %s -o %t.out
 // RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
 //==---------------------- HostAccDestruction.cpp --------------------------==//
 //

sycl/test/scheduler/ReleaseResourcesTest.cpp

@@ -1,4 +1,4 @@
-// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -I %sycl_source_dir %s -o %t.out
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -fsycl-dead-args-optimization -I %sycl_source_dir %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
 // RUN: env SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER

sycl/test/separate-compile/test.cpp

@@ -3,13 +3,13 @@
 //
 // >> ---- compile src1
 // >> device compilation...
-// RUN: %clangxx -fsycl-device-only -Xclang -fsycl-int-header=sycl_ihdr_a.h %s -c -o a_kernel.bc -I %sycl_include -Wno-sycl-strict
+// RUN: %clangxx -fsycl-device-only -fno-sycl-early-optimizations -Xclang -fsycl-int-header=sycl_ihdr_a.h %s -c -o a_kernel.bc -I %sycl_include -Wno-sycl-strict
 // >> host compilation...
 // RUN: %clangxx -include sycl_ihdr_a.h -g -c %s -o a.o -I %sycl_include -Wno-sycl-strict
 //
 // >> ---- compile src2
 // >> device compilation...
-// RUN: %clangxx -DB_CPP=1 -fsycl-device-only -Xclang -fsycl-int-header=sycl_ihdr_b.h %s -c -o b_kernel.bc -I %sycl_include -Wno-sycl-strict
+// RUN: %clangxx -DB_CPP=1 -fsycl-device-only -fno-sycl-early-optimizations -Xclang -fsycl-int-header=sycl_ihdr_b.h %s -c -o b_kernel.bc -I %sycl_include -Wno-sycl-strict
 // >> host compilation...
 // RUN: %clangxx -DB_CPP=1 -include sycl_ihdr_b.h -g -c %s -o b.o -I %sycl_include -Wno-sycl-strict
 //

sycl/test/usm/pfor_flatten.cpp

@@ -1,7 +1,7 @@
 // UNSUPPORTED: cuda
 // CUDA does not support the unnamed lambda extension.
 //
-// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple  -fsycl-unnamed-lambda %s -o %t1.out
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -fsycl-unnamed-lambda -fsycl-dead-args-optimization %s -o %t1.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t1.out
 // RUN: %CPU_RUN_PLACEHOLDER %t1.out
 // RUN: %GPU_RUN_PLACEHOLDER %t1.out