Merge branch 'sycl' of https://github.com/intel/llvm into benchmarkin…

…g-workflow

.github/workflows/sycl-linux-precommit.yml

@@ -244,7 +244,6 @@ jobs:
 #
 #      env: '{"LIT_FILTER":"PerformanceTests/"}'
 #      extra_lit_opts: -a -j 1 --param enable-perf-tests=True
-#      extra_cmake_args: ${{ matrix.extra_cmake_args }}
 #
 #      ref: ${{ github.sha }}
 #      merge_ref: ''
@@ -324,4 +323,4 @@ jobs:
 
       sycl_toolchain_artifact: sycl_linux_default
       sycl_toolchain_archive: ${{ needs.build.outputs.artifact_archive_name }}
-      sycl_toolchain_decompress_command: ${{ needs.build.outputs.artifact_decompress_command }}
+      sycl_toolchain_decompress_command: ${{ needs.build.outputs.artifact_decompress_command }}

devops/dependencies-igc-dev.json

@@ -1,10 +1,10 @@
 {
   "linux": {
     "igc_dev": {
-      "github_tag": "igc-dev-97b3d8f",
-      "version": "97b3d8f",
-      "updated_at": "2025-01-08T17:43:30Z",
-      "url": "https://api.github.com/repos/intel/intel-graphics-compiler/actions/artifacts/2403247641/zip",
+      "github_tag": "igc-dev-61b96b3",
+      "version": "61b96b3",
+      "updated_at": "2025-01-15T17:43:30Z",
+      "url": "https://api.github.com/repos/intel/intel-graphics-compiler/actions/artifacts/2435370337/zip",
       "root": "{DEPS_ROOT}/opencl/runtime/linux/oclgpu"
     }
   }

devops/scripts/install_build_tools.sh

@@ -24,5 +24,5 @@ apt update && apt install -yqq \
       jq \
       curl \
       libhwloc-dev \
-      libzstd-dev
-
+      libzstd-dev \
+      time

libclc/clc/include/clc/common/clc_degrees.h

@@ -0,0 +1,12 @@
+#ifndef __CLC_COMMON_CLC_DEGREES_H__
+#define __CLC_COMMON_CLC_DEGREES_H__
+
+#define __CLC_BODY <clc/math/unary_decl.inc>
+#define __CLC_FUNCTION __clc_degrees
+
+#include <clc/math/gentype.inc>
+
+#undef __CLC_BODY
+#undef __CLC_FUNCTION
+
+#endif // __CLC_COMMON_CLC_DEGREES_H__

libclc/clc/include/clc/common/unary_def.inc

@@ -0,0 +1,9 @@
+#include <clc/utils.h>
+
+#ifndef __CLC_FUNCTION
+#define __CLC_FUNCTION(x) __CLC_CONCAT(__clc_, x)
+#endif
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE FUNCTION(__CLC_GENTYPE a) {
+  return __CLC_FUNCTION(FUNCTION)(a);
+}

libclc/clc/lib/generic/SOURCES

@@ -1,3 +1,4 @@
+common/clc_degrees.cl
 geometric/clc_dot.cl
 integer/clc_abs.cl
 integer/clc_abs_diff.cl

libclc/clc/lib/generic/common/clc_degrees.cl

@@ -0,0 +1,31 @@
+#include <clc/clcmacro.h>
+
+_CLC_OVERLOAD _CLC_DEF float __clc_degrees(float radians) {
+  // 180/pi = ~57.29577951308232087685 or 0x1.ca5dc1a63c1f8p+5 or 0x1.ca5dc2p+5F
+  return 0x1.ca5dc2p+5F * radians;
+}
+
+_CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, float, __clc_degrees, float);
+
+#ifdef cl_khr_fp64
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+
+_CLC_OVERLOAD _CLC_DEF double __clc_degrees(double radians) {
+  // 180/pi = ~57.29577951308232087685 or 0x1.ca5dc1a63c1f8p+5 or 0x1.ca5dc2p+5F
+  return 0x1.ca5dc1a63c1f8p+5 * radians;
+}
+
+_CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, double, __clc_degrees, double);
+
+#endif
+
+#ifdef cl_khr_fp16
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+
+_CLC_OVERLOAD _CLC_DEF half __clc_degrees(half radians) {
+  return __clc_degrees((float)radians);
+}
+
+_CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, half, __clc_degrees, half);
+
+#endif

libclc/generic/lib/common/degrees.cl

@@ -21,23 +21,8 @@
  */
 
 #include <clc/clc.h>
-#include <clc/clcmacro.h>
-#include <libspirv/spirv.h>
+#include <clc/common/clc_degrees.h>
 
-_CLC_OVERLOAD _CLC_DEF float degrees(float radians) {
-  return __spirv_ocl_degrees(radians);
-}
-
-_CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, float, degrees, float);
-
-
-#ifdef cl_khr_fp64
-#pragma OPENCL EXTENSION cl_khr_fp64 : enable
-
-_CLC_OVERLOAD _CLC_DEF double degrees(double radians) {
-  return __spirv_ocl_degrees(radians);
-}
-
-_CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, double, degrees, double);
-
-#endif
+#define FUNCTION degrees
+#define __CLC_BODY <clc/common/unary_def.inc>
+#include <clc/math/gentype.inc>

libclc/libspirv/lib/generic/common/degrees.cl

@@ -6,38 +6,11 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include <clc/clc.h>
+#include <clc/common/clc_degrees.h>
 #include <libspirv/spirv.h>
 
-#include <clc/clcmacro.h>
-
-_CLC_OVERLOAD _CLC_DEF float __spirv_ocl_degrees(float radians) {
-  // 180/pi = ~57.29577951308232087685 or 0x1.ca5dc1a63c1f8p+5 or 0x1.ca5dc2p+5F
-  return 0x1.ca5dc2p+5F * radians;
-}
-
-_CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, float, __spirv_ocl_degrees, float);
-
-
-#ifdef cl_khr_fp64
-#pragma OPENCL EXTENSION cl_khr_fp64 : enable
-
-_CLC_OVERLOAD _CLC_DEF double __spirv_ocl_degrees(double radians) {
-  // 180/pi = ~57.29577951308232087685 or 0x1.ca5dc1a63c1f8p+5 or 0x1.ca5dc2p+5F
-  return 0x1.ca5dc1a63c1f8p+5 * radians;
-}
-
-_CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, double, __spirv_ocl_degrees,
-                     double);
-
-#endif
-
-#ifdef cl_khr_fp16
-#pragma OPENCL EXTENSION cl_khr_fp16 : enable
-
-_CLC_OVERLOAD _CLC_DEF half __spirv_ocl_degrees(half radians) {
-  return __spirv_ocl_degrees((float)radians);
-}
-
-_CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, half, __spirv_ocl_degrees, half);
-
-#endif
+#define FUNCTION __spirv_ocl_degrees
+#define __CLC_FUNCTION(x) __clc_degrees
+#define __CLC_BODY <clc/common/unary_def.inc>
+#include <clc/math/gentype.inc>

llvm-spirv/test/extensions/INTEL/SPV_INTEL_joint_matrix/array_of_matrices.ll

@@ -13,12 +13,8 @@
 ; RUN: llvm-spirv %t.spv -to-text -o %t.spt
 ; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
 
-; TODO: enable back once the reverse translation with untyped pointers is fixed
-; R/UN: llvm-spirv -r %t.spv -o %t.rev.bc
-; R/UN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM-GENERAL
-
-; R/UN: llvm-spirv -r %t.spv -o %t.rev.bc
-; R/UN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM-INTEL
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
 
 ; CHECK-SPIRV-DAG: Capability JointMatrixINTEL
 ; CHECK-SPIRV-DAG: Extension "SPV_INTEL_joint_matrix"

llvm-spirv/test/extensions/INTEL/SPV_INTEL_joint_matrix/cooperative_matrix_apply.ll

@@ -12,11 +12,8 @@
 ; RUN: llvm-spirv %t.spv -to-text -o %t.spt
 ; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
 
-; TODO: enable back once the reverse translation with untyped pointers is fixed
-; R/UN: llvm-spirv -r %t.spv -o %t.rev.bc
-; R/UN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM-MATRIX
-; R/UN: llvm-spirv -r %t.spv -o %t.rev.bc
-; R/UN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM-GENERAL
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
 
 ; CHECK-SPIRV-DAG: Capability CooperativeMatrixKHR
 ; CHECK-SPIRV-DAG: Capability CooperativeMatrixInvocationInstructionsINTEL

llvm-spirv/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_bfloat16.ll

@@ -13,12 +13,8 @@
 ; RUN: llvm-spirv %t.spv -to-text -o %t.spt
 ; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
 
-; TODO: enable back once the reverse translation with untyped pointers is fixed
-; R/UN: llvm-spirv -r %t.spv -o %t.rev.bc
-; R/UN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM-GENERAL
-
-; R/UN: llvm-spirv -r %t.spv -o %t.rev.bc
-; R/UN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM-INTEL
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
 
 ; CHECK-SPIRV-DAG: TypeInt [[#SHORT:]] 16
 ; CHECK-SPIRV-DAG: TypeInt [[#INT:]] 32

llvm-spirv/test/extensions/INTEL/SPV_INTEL_joint_matrix/joint_matrix_half.ll

@@ -13,13 +13,8 @@
 ; RUN: llvm-spirv %t.spv -to-text -o %t.spt
 ; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
 
-; TODO: enable back once the reverse translation with untyped pointers is fixed
-; R/UN: llvm-spirv -r %t.spv -o %t.rev.bc
-; R/UN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM-GENERAL
-
-; R/UN: llvm-spirv -r %t.spv -o %t.rev.bc
-; R/UN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM-INTEL
-
+; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
 
 ; CHECK-SPIRV-DAG: TypeInt [[#INT:]] 32
 ; CHECK-SPIRV-DAG: TypeFloat [[#Half:]] 16

llvm-spirv/test/extensions/KHR/SPV_KHR_cooperative_matrix/array_of_matrices.ll

@@ -14,9 +14,8 @@
 ; RUN: llvm-spirv %t.spv -to-text -o %t.spt
 ; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
 
-; TODO: enable back once the reverse translation with untyped pointers is fixed
-; R/UN: llvm-spirv -r --spirv-target-env=SPV-IR %t.spv -o %t.rev.bc
-; R/UN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
+; RUN: llvm-spirv -r --spirv-target-env=SPV-IR %t.spv -o %t.rev.bc
+; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
 
 ; CHECK-SPIRV-DAG: Capability CooperativeMatrixKHR
 ; CHECK-SPIRV-DAG: Extension "SPV_KHR_cooperative_matrix"

llvm/lib/SYCLLowerIR/ESIMD/ESIMDOptimizeVecArgCallConv.cpp

@@ -354,9 +354,8 @@ optimizeFunction(Function *OldF,
       // preserve data flow equality to the original.
       unsigned OldArgNo = PI.getFormalParam().getArgNo();
       unsigned NewArgNo = oldArgNo2NewArgNo(OldArgNo, SretInd);
-      Instruction *At = nullptr;
       Value *Val = NewF->getArg(NewArgNo);
-      StoreInst *St = new StoreInst(Val, Alloca, false, Al, At);
+      StoreInst *St = new StoreInst(Val, Alloca, false, Al);
       NewInsts.push_back(St);
     }
   }

llvm/lib/SYCLLowerIR/ESIMD/LowerESIMD.cpp

@@ -1239,7 +1239,7 @@ static Instruction *addCastInstIfNeeded(Instruction *OldI, Instruction *NewI,
   if (OITy != NITy) {
     auto CastOpcode = CastInst::getCastOpcode(NewI, false, OITy, false);
     NewI = CastInst::Create(CastOpcode, NewI, OITy,
-                            NewI->getName() + ".cast.ty", OldI);
+                            NewI->getName() + ".cast.ty", OldI->getIterator());
     NewI->setDebugLoc(OldI->getDebugLoc());
   }
   return NewI;
@@ -1565,7 +1565,7 @@ static void translateESIMDIntrinsicCall(CallInst &CI) {
   CallInst *NewCI = IntrinsicInst::Create(
       NewFDecl, GenXArgs,
       NewFDecl->getReturnType()->isVoidTy() ? "" : CI.getName() + ".esimd",
-      &CI);
+      CI.getIterator());
   NewCI->setDebugLoc(CI.getDebugLoc());
   if (DoesFunctionReturnStructure) {
     IRBuilder<> Builder(&CI);

llvm/lib/SYCLLowerIR/GlobalOffset.cpp

@@ -250,7 +250,7 @@ void GlobalOffsetPass::addImplicitParameterToCallers(
           /* Func= */ CalleeWithImplicitParam,
           /* Args= */ ImplicitOffsets,
           /* NameStr= */ Twine(),
-          /* InsertBefore= */ CallToOld);
+          /* InsertBefore= */ CallToOld->getIterator());
       NewCallInst->setTailCallKind(CallToOld->getTailCallKind());
       NewCallInst->copyMetadata(*CallToOld);
       CallToOld->replaceAllUsesWith(NewCallInst);

llvm/lib/SYCLLowerIR/LocalAccessorToSharedMemory.cpp

@@ -177,7 +177,7 @@ Function *LocalAccessorToSharedMemoryPass::processKernel(Module &M,
               ConstantInt::get(Type::getInt32Ty(M.getContext()), 0, false),
               NFA,
           },
-          /* NameStr= */ Twine{NFA->getName()}, InsertBefore);
+          /* NameStr= */ Twine{NFA->getName()}, InsertBefore->getIterator());
       // Then create a bitcast to make sure the new pointer is the same type
       // as the old one. This will only ever be a `i8 addrspace(3)*` to `i32
       // addrspace(3)*` type of cast.

llvm/lib/SYCLLowerIR/LowerInvokeSimd.cpp

@@ -461,8 +461,8 @@ bool processInvokeSimdCall(CallInst *InvokeSimd,
     NewInvokeSimdArgs.push_back(NewHelper);
     auto ThirdArg = std::next(InvokeSimd->arg_begin(), 2);
     NewInvokeSimdArgs.append(ThirdArg, InvokeSimd->arg_end());
-    CallInst *NewInvokeSimd =
-        CallInst::Create(NewInvokeSimdF, NewInvokeSimdArgs, "", InvokeSimd);
+    CallInst *NewInvokeSimd = CallInst::Create(
+        NewInvokeSimdF, NewInvokeSimdArgs, "", InvokeSimd->getIterator());
     // - transfer flags, attributes (with shrinking), calling convention:
     NewInvokeSimd->copyIRFlags(InvokeSimd);
     NewInvokeSimd->setCallingConv(InvokeSimd->getCallingConv());

llvm/lib/SYCLLowerIR/LowerWGScope.cpp

@@ -981,7 +981,7 @@ Value *spirv::genPseudoLocalID(Instruction &Before, const Triple &TT) {
       Align Alignment = M.getDataLayout().getPreferredAlign(G);
       G->setAlignment(MaybeAlign(Alignment));
     }
-    Value *Res = new LoadInst(G->getValueType(), G, "", &Before);
+    Value *Res = new LoadInst(G->getValueType(), G, "", Before.getIterator());
     return Res;
   }
 }

llvm/lib/SYCLLowerIR/SYCLConditionalCallOnDevice.cpp

@@ -53,7 +53,7 @@ SYCLConditionalCallOnDevicePass::run(Module &M, ModuleAnalysisManager &) {
     // (FAction). FAction should be a literal (i.e. not a pointer). The
     // structure of the header file ensures that there is exactly one such
     // instruction.
-    bool CallFound = false;
+    [[maybe_unused]] bool CallFound = false;
     for (Instruction &I : instructions(FCaller)) {
       if (auto *CI = dyn_cast<CallInst>(&I);
           CI && (Intrinsic::IndependentIntrinsics::not_intrinsic ==
@@ -121,8 +121,8 @@ SYCLConditionalCallOnDevicePass::run(Module &M, ModuleAnalysisManager &) {
         Args.push_back(Call->getArgOperand(I));
 
       // Create the new call instruction
-      auto *NewCall =
-          CallInst::Create(NewFCaller, Args, /*	NameStr = */ "", Call);
+      auto *NewCall = CallInst::Create(NewFCaller, Args, /*	NameStr = */ "",
+                                       Call->getIterator());
       NewCall->setCallingConv(Call->getCallingConv());
       NewCall->setDebugLoc(Call->getDebugLoc());
 

llvm/lib/SYCLLowerIR/SpecConstants.cpp

@@ -514,15 +514,15 @@ Instruction *emitCall(Type *RetTy, StringRef BaseFunctionName,
       auto *NewFT = FunctionType::get(NewRetTy, ArgTys, false /*isVarArg*/);
       auto NewFC = M->getOrInsertFunction(FunctionName, NewFT);
 
-      auto *Call =
-          CallInst::Create(NewFT, NewFC.getCallee(), Args, "", InsertBefore);
+      auto *Call = CallInst::Create(NewFT, NewFC.getCallee(), Args, "",
+                                    InsertBefore->getIterator());
       if (IsSPIROrSPIRV) {
         cast<Function>(NewFC.getCallee())
             ->setCallingConv(CallingConv::SPIR_FUNC);
         Call->setCallingConv(CallingConv::SPIR_FUNC);
       }
       return CastInst::CreateTruncOrBitCast(Call, RetTy, "tobool",
-                                            InsertBefore);
+                                            InsertBefore->getIterator());
     }
   }
 
@@ -711,22 +711,22 @@ Value *createLoadFromBuffer(CallInst *InsertBefore, Value *Buffer,
   Type *Int32Ty = Type::getInt32Ty(C);
   GetElementPtrInst *GEP = GetElementPtrInst::Create(
       Int8Ty, Buffer, {ConstantInt::get(Int32Ty, Offset, false)}, "gep",
-      InsertBefore);
+      InsertBefore->getIterator());
 
   Instruction *BitCast = nullptr;
   if (SCType->isIntegerTy(1)) // No bitcast to i1 before load
     BitCast = GEP;
   else
     BitCast =
         new BitCastInst(GEP, PointerType::get(SCType, GEP->getAddressSpace()),
-                        "bc", InsertBefore);
+                        "bc", InsertBefore->getIterator());
 
   // When we encounter i1 spec constant, we still load the whole byte
   Value *Load = new LoadInst(SCType->isIntegerTy(1) ? Int8Ty : SCType, BitCast,
-                             "load", InsertBefore);
+                             "load", InsertBefore->getIterator());
   if (SCType->isIntegerTy(1)) // trunc back to i1 if necessary
     Load = CastInst::CreateIntegerCast(Load, SCType, /* IsSigned */ false,
-                                       "tobool", InsertBefore);
+                                       "tobool", InsertBefore->getIterator());
 
   return Load;
 }
@@ -993,8 +993,8 @@ PreservedAnalyses SpecConstantsPass::run(Module &M,
         if (SCTy->isIntegerTy(1)) {
           assert(DefaultValue->getType()->isIntegerTy(8) &&
                  "For bool spec constant default value is expected to be i8");
-          Replacement =
-              new TruncInst(DefaultValue, Type::getInt1Ty(Ctx), "bool", CI);
+          Replacement = new TruncInst(DefaultValue, Type::getInt1Ty(Ctx),
+                                      "bool", CI->getIterator());
         } else
           Replacement = DefaultValue;
       }
@@ -1021,9 +1021,9 @@ PreservedAnalyses SpecConstantsPass::run(Module &M,
         Value *ArraySize =
             Mode == HandlingMode::emulation ? DefaultValue : Replacement;
         assert(ArraySize->getType()->isIntegerTy() && "Expecting integer type");
-        Replacement =
-            new AllocaInst(Intr->getAllocatedType(), Intr->getAddressSpace(),
-                           ArraySize, Intr->getAlign(), "alloca", CI);
+        Replacement = new AllocaInst(
+            Intr->getAllocatedType(), Intr->getAddressSpace(), ArraySize,
+            Intr->getAlign(), "alloca", CI->getIterator());
       }
 
       if (HasSretParameter)