Merge branch 'main' into UnifAB

Author: jrbyrnes

clang/test/Driver/linker-wrapper-image.c

@@ -1,6 +1,7 @@
 // REQUIRES: x86-registered-target
 // REQUIRES: nvptx-registered-target
 // REQUIRES: amdgpu-registered-target
+// REQUIRES: spirv-registered-target
 
 // RUN: %clang -cc1 %s -triple x86_64-unknown-linux-gnu -emit-obj -o %t.elf.o
 
@@ -263,3 +264,36 @@
 //      HIP: while.end:
 // HIP-NEXT:   ret void
 // HIP-NEXT: }
+
+// RUN: clang-offload-packager -o %t.out --image=file=%t.elf.o,kind=sycl,triple=spirv64-unknown-unknown,arch=generic
+// RUN: %clang -cc1 %s -triple x86_64-unknown-linux-gnu -emit-obj -o %t.o \
+// RUN:   -fembed-offload-object=%t.out
+// RUN: clang-linker-wrapper --print-wrapped-module --dry-run --host-triple=x86_64-unknown-linux-gnu \
+// RUN:   --linker-path=/usr/bin/ld %t.o -o a.out 2>&1 | FileCheck %s --check-prefixes=SYCL
+// RUN: clang-linker-wrapper --print-wrapped-module --dry-run --host-triple=x86_64-unknown-linux-gnu -r \
+// RUN:   --linker-path=/usr/bin/ld %t.o -o a.out 2>&1 | FileCheck %s --check-prefixes=SYCL
+
+//      SYCL: %__sycl.tgt_device_image = type { i16, i8, i8, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr }
+// SYCL-NEXT: %__sycl.tgt_bin_desc = type { i16, i16, ptr, ptr, ptr }
+
+//      SYCL: @.sycl_offloading.target.0 = internal unnamed_addr constant [1 x i8] zeroinitializer
+// SYCL-NEXT: @.sycl_offloading.opts.compile.0 = internal unnamed_addr constant [1 x i8] zeroinitializer
+// SYCL-NEXT: @.sycl_offloading.opts.link.0 = internal unnamed_addr constant [1 x i8] zeroinitializer
+// SYCL-NEXT: @.sycl_offloading.0.data = internal unnamed_addr constant [0 x i8] zeroinitializer, section ".llvm.offloading"
+// SYCL-NEXT: @.offloading.entry_name = internal unnamed_addr constant [5 x i8] c"stub\00", section ".llvm.rodata.offloading", align 1
+// SYCL-NEXT: @.offloading.entry.stub = weak constant %struct.__tgt_offload_entry { i64 0, i16 1, i16 8, i32 0, ptr null, ptr @.offloading.entry_name, i64 0, i64 0, ptr null }, section "llvm_offload_entries", align 8
+// SYCL-NEXT: @.sycl_offloading.entries_arr = internal constant [1 x %struct.__tgt_offload_entry] [%struct.__tgt_offload_entry { i64 0, i16 1, i16 8, i32 0, ptr null, ptr @.offloading.entry_name, i64 0, i64 0, ptr null }]
+// SYCL-NEXT: @.sycl_offloading.device_images = internal unnamed_addr constant [1 x %__sycl.tgt_device_image] [%__sycl.tgt_device_image { i16 3, i8 8, i8 0, ptr @.sycl_offloading.target.0, ptr @.sycl_offloading.opts.compile.0, ptr @.sycl_offloading.opts.link.0, ptr @.sycl_offloading.0.data, ptr @.sycl_offloading.0.data, ptr @.sycl_offloading.entries_arr, ptr getelementptr ([1 x %struct.__tgt_offload_entry], ptr @.sycl_offloading.entries_arr, i64 0, i64 1), ptr null, ptr null }]
+// SYCL-NEXT: @.sycl_offloading.descriptor = internal constant %__sycl.tgt_bin_desc { i16 1, i16 1, ptr @.sycl_offloading.device_images, ptr null, ptr null }
+
+//      SYCL: define internal void @sycl.descriptor_reg() section ".text.startup" {
+// SYCL-NEXT: entry:
+// SYCL-NEXT:   call void @__sycl_register_lib(ptr @.sycl_offloading.descriptor)
+// SYCL-NEXT:   ret void
+// SYCL-NEXT: }
+
+//      SYCL: define internal void @sycl.descriptor_unreg() section ".text.startup" {
+// SYCL-NEXT: entry:
+// SYCL-NEXT:   call void @__sycl_unregister_lib(ptr @.sycl_offloading.descriptor)
+// SYCL-NEXT:   ret void
+// SYCL-NEXT: }

clang/test/Driver/linker-wrapper.c

@@ -54,7 +54,7 @@ __attribute__((visibility("protected"), used)) int x;
 // RUN: clang-offload-packager -o %t.out \
 // RUN:   --image=file=%t.spirv.bc,kind=sycl,triple=spirv64-unknown-unknown,arch=generic
 // RUN: %clang -cc1 %s -triple x86_64-unknown-linux-gnu -emit-obj -o %t.o -fembed-offload-object=%t.out
-// RUN: not clang-linker-wrapper --host-triple=x86_64-unknown-linux-gnu --dry-run \
+// RUN: clang-linker-wrapper --host-triple=x86_64-unknown-linux-gnu --dry-run \
 // RUN:   --linker-path=/usr/bin/ld %t.o -o a.out 2>&1 | FileCheck %s --check-prefix=SPIRV-LINK
 
 // SPIRV-LINK: clang{{.*}} -o {{.*}}.img -dumpdir a.out.spirv64..img. --target=spirv64-unknown-unknown {{.*}}.o --sycl-link -Xlinker -triple=spirv64-unknown-unknown -Xlinker -arch=

clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp

@@ -717,6 +717,14 @@ wrapDeviceImages(ArrayRef<std::unique_ptr<MemoryBuffer>> Buffers,
             M, BuffersToWrap.front(), offloading::getOffloadEntryArray(M)))
       return std::move(Err);
     break;
+  case OFK_SYCL: {
+    // TODO: fill these options once the Driver supports them.
+    offloading::SYCLJITOptions Options;
+    if (Error Err =
+            offloading::wrapSYCLBinaries(M, BuffersToWrap.front(), Options))
+      return std::move(Err);
+    break;
+  }
   default:
     return createStringError(getOffloadKindName(Kind) +
                              " wrapping is not supported");
@@ -754,6 +762,32 @@ bundleOpenMP(ArrayRef<OffloadingImage> Images) {
   return std::move(Buffers);
 }
 
+Expected<SmallVector<std::unique_ptr<MemoryBuffer>>>
+bundleSYCL(ArrayRef<OffloadingImage> Images) {
+  SmallVector<std::unique_ptr<MemoryBuffer>> Buffers;
+  if (DryRun) {
+    // In dry-run mode there is an empty input which is insufficient for the
+    // testing. Therefore, we return here a stub image.
+    OffloadingImage Image;
+    Image.TheImageKind = IMG_None;
+    Image.TheOffloadKind = OffloadKind::OFK_SYCL;
+    Image.StringData["symbols"] = "stub";
+    Image.Image = MemoryBuffer::getMemBufferCopy("");
+    SmallString<0> SerializedImage = OffloadBinary::write(Image);
+    Buffers.emplace_back(MemoryBuffer::getMemBufferCopy(SerializedImage));
+    return std::move(Buffers);
+  }
+
+  for (const OffloadingImage &Image : Images) {
+    // clang-sycl-linker packs outputs into one binary blob. Therefore, it is
+    // passed to Offload Wrapper as is.
+    StringRef S(Image.Image->getBufferStart(), Image.Image->getBufferSize());
+    Buffers.emplace_back(MemoryBuffer::getMemBufferCopy(S));
+  }
+
+  return std::move(Buffers);
+}
+
 Expected<SmallVector<std::unique_ptr<MemoryBuffer>>>
 bundleCuda(ArrayRef<OffloadingImage> Images, const ArgList &Args) {
   SmallVector<std::pair<StringRef, StringRef>, 4> InputFiles;
@@ -806,8 +840,9 @@ bundleLinkedOutput(ArrayRef<OffloadingImage> Images, const ArgList &Args,
   llvm::TimeTraceScope TimeScope("Bundle linked output");
   switch (Kind) {
   case OFK_OpenMP:
-  case OFK_SYCL:
     return bundleOpenMP(Images);
+  case OFK_SYCL:
+    return bundleSYCL(Images);
   case OFK_Cuda:
     return bundleCuda(Images, Args);
   case OFK_HIP:

flang/lib/Lower/Bridge.cpp

@@ -2544,7 +2544,7 @@ class FirConverter : public Fortran::lower::AbstractConverter {
         auto loopOp = fir::DoLoopOp::create(
             *builder, loc, lowerValue, upperValue, stepValue,
             /*unordered=*/false,
-            /*finalCountValue=*/true,
+            /*finalCountValue=*/false,
             builder->createConvert(loc, loopVarType, lowerValue));
         info.loopOp = loopOp;
         builder->setInsertionPointToStart(loopOp.getBody());
@@ -2696,22 +2696,18 @@ class FirConverter : public Fortran::lower::AbstractConverter {
         // Decrement tripVariable.
         auto doLoopOp = mlir::cast<fir::DoLoopOp>(info.loopOp);
         builder->setInsertionPointToEnd(doLoopOp.getBody());
-        llvm::SmallVector<mlir::Value, 2> results;
-        results.push_back(mlir::arith::AddIOp::create(
-            *builder, loc, doLoopOp.getInductionVar(), doLoopOp.getStep(),
-            iofAttr));
         // Step loopVariable to help optimizations such as vectorization.
         // Induction variable elimination will clean up as necessary.
         mlir::Value step = builder->createConvert(
             loc, info.getLoopVariableType(), doLoopOp.getStep());
         mlir::Value loopVar =
             fir::LoadOp::create(*builder, loc, info.loopVariable);
-        results.push_back(
-            mlir::arith::AddIOp::create(*builder, loc, loopVar, step, iofAttr));
-        fir::ResultOp::create(*builder, loc, results);
+        mlir::Value loopVarInc =
+            mlir::arith::AddIOp::create(*builder, loc, loopVar, step, iofAttr);
+        fir::ResultOp::create(*builder, loc, loopVarInc);
         builder->setInsertionPointAfter(doLoopOp);
         // The loop control variable may be used after the loop.
-        fir::StoreOp::create(*builder, loc, doLoopOp.getResult(1),
+        fir::StoreOp::create(*builder, loc, doLoopOp.getResult(0),
                              info.loopVariable);
         continue;
       }

flang/lib/Lower/IO.cpp

@@ -977,9 +977,9 @@ static void genIoLoop(Fortran::lower::AbstractConverter &converter,
     fir::StoreOp::create(builder, loc, lcv, loopVar);
     genItemList(ioImpliedDo);
     builder.setInsertionPointToEnd(doLoopOp.getBody());
-    mlir::Value result = mlir::arith::AddIOp::create(
-        builder, loc, doLoopOp.getInductionVar(), doLoopOp.getStep(), iofAttr);
-    fir::ResultOp::create(builder, loc, result);
+    // fir.do_loop's induction variable's increment is implied,
+    // so we do not need to increment it explicitly.
+    fir::ResultOp::create(builder, loc, doLoopOp.getInductionVar());
     builder.setInsertionPointAfter(doLoopOp);
     // The loop control variable may be used after the loop.
     lcv = builder.createConvert(loc, fir::unwrapRefType(loopVar.getType()),

flang/test/Fir/convert-to-llvm-openmp-and-fir.fir

@@ -349,7 +349,7 @@ func.func @_QPopenmp_target_data_region() {
     %3 = fir.convert %c1024_i32 : (i32) -> index
     %c1 = arith.constant 1 : index
     %4 = fir.convert %2 : (index) -> i32
-    %5:2 = fir.do_loop %arg0 = %2 to %3 step %c1 iter_args(%arg1 = %4) -> (index, i32) {
+    %5 = fir.do_loop %arg0 = %2 to %3 step %c1 iter_args(%arg1 = %4) -> (i32) {
       fir.store %arg1 to %1 : !fir.ref<i32>
       %6 = fir.load %1 : !fir.ref<i32>
       %7 = fir.load %1 : !fir.ref<i32>
@@ -358,13 +358,12 @@ func.func @_QPopenmp_target_data_region() {
       %9 = arith.subi %8, %c1_i64 : i64
       %10 = fir.coordinate_of %0, %9 : (!fir.ref<!fir.array<1024xi32>>, i64) -> !fir.ref<i32>
       fir.store %6 to %10 : !fir.ref<i32>
-      %11 = arith.addi %arg0, %c1 overflow<nsw> : index
       %12 = fir.convert %c1 : (index) -> i32
       %13 = fir.load %1 : !fir.ref<i32>
       %14 = arith.addi %13, %12 overflow<nsw> : i32
-      fir.result %11, %14 : index, i32
+      fir.result %14 : i32
     }
-    fir.store %5#1 to %1 : !fir.ref<i32>
+    fir.store %5 to %1 : !fir.ref<i32>
     omp.terminator
   }
   return
@@ -404,7 +403,6 @@ func.func @_QPopenmp_target_data_region() {
 // CHECK:             %[[VAL_21:.*]] = llvm.sub %[[VAL_19]], %[[VAL_20]]  : i64
 // CHECK:             %[[VAL_22:.*]] = llvm.getelementptr %[[VAL_1]][0, %[[VAL_21]]] : (!llvm.ptr, i64) -> !llvm.ptr
 // CHECK:             llvm.store %[[VAL_17]], %[[VAL_22]] : i32, !llvm.ptr
-// CHECK:             %[[VAL_23:.*]] = llvm.add %[[VAL_12]], %[[VAL_8]] overflow<nsw> : i64
 // CHECK:             %[[VAL_24:.*]] = llvm.trunc %[[VAL_8]] : i64 to i32
 // CHECK:             %[[VAL_25:.*]] = llvm.load %[[VAL_3]] : !llvm.ptr -> i32
 // CHECK:             %[[VAL_26:.*]] = llvm.add %[[VAL_25]], %[[VAL_24]] overflow<nsw> : i32
@@ -653,18 +651,17 @@ func.func @_QPsb() {
   omp.sections   {
     omp.section {
       %2 = fir.convert %c1 : (index) -> i32
-      %3:2 = fir.do_loop %arg0 = %c1 to %c10 step %c1 iter_args(%arg1 = %2) -> (index, i32) {
+      %3 = fir.do_loop %arg0 = %c1 to %c10 step %c1 iter_args(%arg1 = %2) -> (i32) {
         fir.store %arg1 to %0 : !fir.ref<i32>
         %4 = fir.load %1 : !fir.ref<i32>
         %5 = arith.addi %4, %c1_i32 : i32
         fir.store %5 to %1 : !fir.ref<i32>
-        %6 = arith.addi %arg0, %c1 : index
         %7 = fir.convert %c1 : (index) -> i32
         %8 = fir.load %0 : !fir.ref<i32>
         %9 = arith.addi %8, %7 : i32
-        fir.result %6, %9 : index, i32
+        fir.result %9 : i32
       }
-      fir.store %3#1 to %0 : !fir.ref<i32>
+      fir.store %3 to %0 : !fir.ref<i32>
       omp.terminator
     }
     omp.terminator