merge main into amd-staging (llvm#3483)

Author: z1_cciauto

.ci/compute_projects.py

@@ -80,6 +80,7 @@
     "clang": {"compiler-rt"},
     "clang-tools-extra": {"libc"},
     "libc": {"libc"},
+    "compiler-rt": {"compiler-rt"},
     ".ci": {"compiler-rt", "libc"},
 }
 DEPENDENT_RUNTIMES_TO_TEST_NEEDS_RECONFIG = {

.ci/compute_projects_test.py

@@ -126,6 +126,32 @@ def test_clang_windows(self):
         )
         self.assertEqual(env_variables["enable_cir"], "OFF")
 
+    def test_compiler_rt(self):
+        env_variables = compute_projects.get_env_variables(
+            ["compiler-rt/lib/asan/asan_allocator.cpp"], "Linux"
+        )
+        self.assertEqual(
+            env_variables["projects_to_build"],
+            "clang;lld",
+        )
+        self.assertEqual(
+            env_variables["project_check_targets"],
+            "",
+        )
+        self.assertEqual(env_variables["runtimes_to_build"], "compiler-rt")
+        self.assertEqual(
+            env_variables["runtimes_check_targets"],
+            "check-compiler-rt",
+        )
+        self.assertEqual(
+            env_variables["runtimes_check_targets_needs_reconfig"],
+            "",
+        )
+        self.assertEqual(
+            env_variables["enable_cir"],
+            "OFF",
+        )
+
     def test_cir(self):
         env_variables = compute_projects.get_env_variables(
             ["clang/lib/CIR/CMakeLists.txt"], "Linux"

compiler-rt/lib/asan/asan_mapping.h

@@ -83,18 +83,29 @@
 // || `[0x0000000000, 0x0d5554ffff]` || LowMem     ||
 //
 // Default Linux/AArch64 (39-bit VMA) mapping:
-// || `[0x2000000000, 0x7fffffffff]` || highmem    ||
-// || `[0x1400000000, 0x1fffffffff]` || highshadow ||
-// || `[0x1200000000, 0x13ffffffff]` || shadowgap  ||
-// || `[0x1000000000, 0x11ffffffff]` || lowshadow  ||
-// || `[0x0000000000, 0x0fffffffff]` || lowmem     ||
+// TODO: this mapping is ok, but the allocator size is too large on non-Android
+//       AArch64 platforms (see asan_allocator.h)
+// || `[0x2000000000, 0x7fffffffff]` || highmem    || 384GB
+// || `[0x1400000000, 0x1fffffffff]` || highshadow || 48GB
+// || `[0x1200000000, 0x13ffffffff]` || shadowgap  || 8GB
+// || `[0x1000000000, 0x11ffffffff]` || lowshadow  || 8GB
+// || `[0x0000000000, 0x0fffffffff]` || lowmem     || 64GB
 //
 // Default Linux/AArch64 (42-bit VMA) mapping:
-// || `[0x10000000000, 0x3ffffffffff]` || highmem    ||
-// || `[0x0a000000000, 0x0ffffffffff]` || highshadow ||
-// || `[0x09000000000, 0x09fffffffff]` || shadowgap  ||
-// || `[0x08000000000, 0x08fffffffff]` || lowshadow  ||
-// || `[0x00000000000, 0x07fffffffff]` || lowmem     ||
+// TODO: this mapping is ok, but the allocator size is too large on non-Android
+//       AArch64 platforms (see asan_allocator.h)
+// || `[0x09000000000, 0x03ffffffffff]` || highmem    || 3520GB
+// || `[0x02200000000, 0x008fffffffff]` || highshadow || 440GB
+// || `[0x01200000000, 0x0021ffffffff]` || shadowgap  || 64GB
+// || `[0x01000000000, 0x0011ffffffff]` || lowshadow  || 8GB
+// || `[0x00000000000, 0x000fffffffff]` || lowmem     || 64GB
+//
+// Default Linux/AArch64 (48-bit VMA) mapping:
+// || `[0x201000000000, 0xffffffffffff]` || HighMem    || 229312GB
+// || `[0x041200000000, 0x200fffffffff]` || HighShadow || 28664GB
+// || `[0x001200000000, 0x0411ffffffff]` || ShadowGap  || 4096GB
+// || `[0x001000000000, 0x0011ffffffff]` || LowShadow  || 8GB
+// || `[0x000000000000, 0x000fffffffff]` || LowMem     || 64GB
 //
 // Default Linux/S390 mapping:
 // || `[0x30000000, 0x7fffffff]` || HighMem    ||

compiler-rt/lib/ubsan_minimal/ubsan_minimal_handlers.cpp

@@ -97,15 +97,19 @@ SANITIZER_INTERFACE_WEAK_DEF(void, __ubsan_report_error_fatal, const char *kind,
 
 #if defined(__ANDROID__)
 extern "C" __attribute__((weak)) void android_set_abort_message(const char *);
-static void abort_with_message(const char *kind, uintptr_t caller) {
+static void abort_with_message(const char *kind, uintptr_t caller,
+                               const uintptr_t *address) {
   char msg_buf[128];
-  format_msg(kind, caller, msg_buf, msg_buf + sizeof(msg_buf));
+  format_msg(kind, caller, address, msg_buf, msg_buf + sizeof(msg_buf));
   if (&android_set_abort_message)
     android_set_abort_message(msg_buf);
   abort();
 }
 #else
-static void abort_with_message(const char *kind, uintptr_t caller) { abort(); }
+static void abort_with_message(const char *kind, uintptr_t caller,
+                               const uintptr_t *address) {
+  abort();
+}
 #endif
 
 #if SANITIZER_DEBUG
@@ -132,7 +136,7 @@ void NORETURN CheckFailed(const char *file, int, const char *cond, u64, u64) {
   INTERFACE void __ubsan_handle_##name##_minimal_abort() {                     \
     uintptr_t caller = GET_CALLER_PC();                                        \
     __ubsan_report_error_fatal(kind, caller, nullptr);                         \
-    abort_with_message(kind, caller);                                          \
+    abort_with_message(kind, caller, nullptr);                                 \
   }
 
 #define HANDLER(name, kind)                                                    \
@@ -149,7 +153,7 @@ void NORETURN CheckFailed(const char *file, int, const char *cond, u64, u64) {
       const uintptr_t address) {                                               \
     uintptr_t caller = GET_CALLER_PC();                                        \
     __ubsan_report_error_fatal(kind, caller, &address);                        \
-    abort_with_message(kind, caller);                                          \
+    abort_with_message(kind, caller, &address);                                \
   }
 
 // A version of a handler that takes a pointer to a value.

flang/include/flang/Support/Fortran.h

@@ -95,8 +95,8 @@ static constexpr IgnoreTKRSet ignoreTKRAll{IgnoreTKR::Type, IgnoreTKR::Kind,
 std::string AsFortran(IgnoreTKRSet);
 
 bool AreCompatibleCUDADataAttrs(std::optional<CUDADataAttr>,
-    std::optional<CUDADataAttr>, IgnoreTKRSet, std::optional<std::string> *,
-    bool allowUnifiedMatchingRule, bool isHostDeviceProcedure,
+    std::optional<CUDADataAttr>, IgnoreTKRSet, bool allowUnifiedMatchingRule,
+    bool isHostDeviceProcedure,
     const LanguageFeatureControl *features = nullptr);
 
 static constexpr char blankCommonObjectName[] = "__BLNK__";

flang/lib/Evaluate/characteristics.cpp

@@ -400,7 +400,7 @@ bool DummyDataObject::IsCompatibleWith(const DummyDataObject &actual,
   }
   if (!attrs.test(Attr::Value) &&
       !common::AreCompatibleCUDADataAttrs(cudaDataAttr, actual.cudaDataAttr,
-          ignoreTKR, warning,
+          ignoreTKR,
           /*allowUnifiedMatchingRule=*/false,
           /*=isHostDeviceProcedure*/ false)) {
     if (whyNot) {
@@ -1816,7 +1816,7 @@ bool DistinguishUtils::Distinguishable(
       x.intent != common::Intent::In) {
     return true;
   } else if (!common::AreCompatibleCUDADataAttrs(x.cudaDataAttr, y.cudaDataAttr,
-                 x.ignoreTKR | y.ignoreTKR, nullptr,
+                 x.ignoreTKR | y.ignoreTKR,
                  /*allowUnifiedMatchingRule=*/false,
                  /*=isHostDeviceProcedure*/ false)) {
     return true;

flang/lib/Lower/Bridge.cpp

@@ -4902,7 +4902,6 @@ class FirConverter : public Fortran::lower::AbstractConverter {
 
     // device = device
     if (lhsIsDevice && rhsIsDevice) {
-      assert(rhs.isVariable() && "CUDA Fortran assignment rhs is not legal");
       auto transferKindAttr = cuf::DataTransferKindAttr::get(
           builder.getContext(), cuf::DataTransferKind::DeviceDevice);
       cuf::DataTransferOp::create(builder, loc, rhsVal, lhsVal, shape,

flang/lib/Semantics/check-call.cpp

@@ -1058,12 +1058,11 @@ static void CheckExplicitDataArg(const characteristics::DummyDataObject &dummy,
             dummyName);
       }
     }
-    std::optional<std::string> warning;
     bool isHostDeviceProc{procedure.cudaSubprogramAttrs &&
         *procedure.cudaSubprogramAttrs ==
             common::CUDASubprogramAttrs::HostDevice};
     if (!common::AreCompatibleCUDADataAttrs(dummyDataAttr, actualDataAttr,
-            dummy.ignoreTKR, &warning, /*allowUnifiedMatchingRule=*/true,
+            dummy.ignoreTKR, /*allowUnifiedMatchingRule=*/true,
             isHostDeviceProc, &context.languageFeatures())) {
       auto toStr{[](std::optional<common::CUDADataAttr> x) {
         return x ? "ATTRIBUTES("s +
@@ -1074,10 +1073,6 @@ static void CheckExplicitDataArg(const characteristics::DummyDataObject &dummy,
           "%s has %s but its associated actual argument has %s"_err_en_US,
           dummyName, toStr(dummyDataAttr), toStr(actualDataAttr));
     }
-    if (warning && context.ShouldWarn(common::UsageWarning::CUDAUsage)) {
-      messages.Say(common::UsageWarning::CUDAUsage, "%s"_warn_en_US,
-          std::move(*warning));
-    }
   }
 
   // Warning for breaking F'2023 change with character allocatables

flang/lib/Support/Fortran.cpp

@@ -103,8 +103,8 @@ std::string AsFortran(IgnoreTKRSet tkr) {
 /// dummy argument attribute while `y` represents the actual argument attribute.
 bool AreCompatibleCUDADataAttrs(std::optional<CUDADataAttr> x,
     std::optional<CUDADataAttr> y, IgnoreTKRSet ignoreTKR,
-    std::optional<std::string> *warning, bool allowUnifiedMatchingRule,
-    bool isHostDeviceProcedure, const LanguageFeatureControl *features) {
+    bool allowUnifiedMatchingRule, bool isHostDeviceProcedure,
+    const LanguageFeatureControl *features) {
   bool isCudaManaged{features
           ? features->IsEnabled(common::LanguageFeature::CudaManaged)
           : false};
@@ -145,9 +145,6 @@ bool AreCompatibleCUDADataAttrs(std::optional<CUDADataAttr> x,
                     *y == CUDADataAttr::Shared ||
                     *y == CUDADataAttr::Constant)) ||
             (!y && (isCudaUnified || isCudaManaged))) {
-          if (y && *y == CUDADataAttr::Shared && warning) {
-            *warning = "SHARED attribute ignored"s;
-          }
           return true;
         }
       } else if (*x == CUDADataAttr::Managed) {

flang/test/Lower/CUDA/cuda-data-transfer.cuf

@@ -444,3 +444,15 @@ subroutine sub23(n)
 end subroutine
 
 ! CHECK-LABEL: func.func @_QPsub23
+
+subroutine sub24()
+  real, managed :: m
+  real, device :: d(4)
+  m = d(1)
+end
+
+! CHECK-LABEL: func.func @_QPsub24()
+! CHECK: %[[D:.*]]:2 = hlfir.declare %1(%2) {data_attr = #cuf.cuda<device>, uniq_name = "_QFsub24Ed"} : (!fir.ref<!fir.array<4xf32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<4xf32>>, !fir.ref<!fir.array<4xf32>>)
+! CHECK: %[[M:.*]]:2 = hlfir.declare %4 {data_attr = #cuf.cuda<managed>, uniq_name = "_QFsub24Em"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
+! CHECK: %[[D1:.*]] = hlfir.designate %[[D]]#0 (%c1{{.*}})  : (!fir.ref<!fir.array<4xf32>>, index) -> !fir.ref<f32>
+! CHECK: cuf.data_transfer %[[D1]] to %[[M]]#0 {transfer_kind = #cuf.cuda_transfer<device_device>} : !fir.ref<f32>, !fir.ref<f32>