Merged main:dd577c05ad0d into amd-gfx:7e1b5aab3054

Local branch amd-gfx 7e1b5aa Merged main:d32509928ba6 into amd-gfx:de1faa086235
Remote branch main dd577c0 [clang] constexpr built-in reduce min/max function. (llvm#120866)

Author: SC llvm team

clang/docs/LanguageExtensions.rst

@@ -736,9 +736,10 @@ at the end to the next power of 2.
 
 These reductions support both fixed-sized and scalable vector types.
 
-The integer reduction intrinsics, including ``__builtin_reduce_add``,
-``__builtin_reduce_mul``, ``__builtin_reduce_and``, ``__builtin_reduce_or``,
-and ``__builtin_reduce_xor``, can be called in a ``constexpr`` context.
+The integer reduction intrinsics, including ``__builtin_reduce_max``,
+``__builtin_reduce_min``, ``__builtin_reduce_add``, ``__builtin_reduce_mul``,
+``__builtin_reduce_and``, ``__builtin_reduce_or``, and ``__builtin_reduce_xor``,
+can be called in a ``constexpr`` context.
 
 Example:
 

clang/docs/ReleaseNotes.rst

@@ -421,7 +421,8 @@ Non-comprehensive list of changes in this release
   ``__builtin_reduce_mul``, ``__builtin_reduce_and``, ``__builtin_reduce_or``,
   ``__builtin_reduce_xor``, ``__builtin_elementwise_popcount``,
   ``__builtin_elementwise_bitreverse``, ``__builtin_elementwise_add_sat``,
-  ``__builtin_elementwise_sub_sat``.
+  ``__builtin_elementwise_sub_sat``, ``__builtin_reduce_min`` (For integral element type),
+  ``__builtin_reduce_max`` (For integral element type).
 
 - Clang now rejects ``_BitInt`` matrix element types if the bit width is less than ``CHAR_WIDTH`` or
   not a power of two, matching preexisting behaviour for vector types.

clang/include/clang/Basic/Builtins.td

@@ -1462,13 +1462,13 @@ def ElementwiseSubSat : Builtin {
 
 def ReduceMax : Builtin {
   let Spellings = ["__builtin_reduce_max"];
-  let Attributes = [NoThrow, Const, CustomTypeChecking];
+  let Attributes = [NoThrow, Const, CustomTypeChecking, Constexpr];
   let Prototype = "void(...)";
 }
 
 def ReduceMin : Builtin {
   let Spellings = ["__builtin_reduce_min"];
-  let Attributes = [NoThrow, Const, CustomTypeChecking];
+  let Attributes = [NoThrow, Const, CustomTypeChecking, Constexpr];
   let Prototype = "void(...)";
 }
 

clang/lib/AST/ExprConstant.cpp

@@ -13604,7 +13604,9 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
   case Builtin::BI__builtin_reduce_mul:
   case Builtin::BI__builtin_reduce_and:
   case Builtin::BI__builtin_reduce_or:
-  case Builtin::BI__builtin_reduce_xor: {
+  case Builtin::BI__builtin_reduce_xor:
+  case Builtin::BI__builtin_reduce_min:
+  case Builtin::BI__builtin_reduce_max: {
     APValue Source;
     if (!EvaluateAsRValue(Info, E->getArg(0), Source))
       return false;
@@ -13641,6 +13643,14 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
         Reduced ^= Source.getVectorElt(EltNum).getInt();
         break;
       }
+      case Builtin::BI__builtin_reduce_min: {
+        Reduced = std::min(Reduced, Source.getVectorElt(EltNum).getInt());
+        break;
+      }
+      case Builtin::BI__builtin_reduce_max: {
+        Reduced = std::max(Reduced, Source.getVectorElt(EltNum).getInt());
+        break;
+      }
       }
     }
 

clang/test/Sema/constant_builtins_vector.cpp

@@ -798,6 +798,21 @@ static_assert(__builtin_reduce_xor((vector4long){(long long)0x1111111111111111L,
 static_assert(__builtin_reduce_xor((vector4uint){0x11111111U, 0x22222222U, 0x44444444U, 0x88888888U}) == 0xFFFFFFFFU);
 static_assert(__builtin_reduce_xor((vector4ulong){0x1111111111111111UL, 0x2222222222222222UL, 0x4444444444444444UL, 0x8888888888888888UL}) == 0xFFFFFFFFFFFFFFFFUL);
 
+static_assert(__builtin_reduce_min((vector4char){}) == 0);
+static_assert(__builtin_reduce_min((vector4char){(char)0x11, (char)0x22, (char)0x44, (char)0x88}) == (char)0x88);
+static_assert(__builtin_reduce_min((vector4short){(short)0x1111, (short)0x2222, (short)0x4444, (short)0x8888}) == (short)0x8888);
+static_assert(__builtin_reduce_min((vector4int){(int)0x11111111, (int)0x22222222, (int)0x44444444, (int)0x88888888}) == (int)0x88888888);
+static_assert(__builtin_reduce_min((vector4long){(long long)0x1111111111111111L, (long long)0x2222222222222222L, (long long)0x4444444444444444L, (long long)0x8888888888888888L}) == (long long)0x8888888888888888L);
+static_assert(__builtin_reduce_min((vector4uint){0x11111111U, 0x22222222U, 0x44444444U, 0x88888888U}) == 0x11111111U);
+static_assert(__builtin_reduce_min((vector4ulong){0x1111111111111111UL, 0x2222222222222222UL, 0x4444444444444444UL, 0x8888888888888888UL}) == 0x1111111111111111UL);
+static_assert(__builtin_reduce_max((vector4char){}) == 0);
+static_assert(__builtin_reduce_max((vector4char){(char)0x11, (char)0x22, (char)0x44, (char)0x88}) == (char)0x44);
+static_assert(__builtin_reduce_max((vector4short){(short)0x1111, (short)0x2222, (short)0x4444, (short)0x8888}) == (short)0x4444);
+static_assert(__builtin_reduce_max((vector4int){(int)0x11111111, (int)0x22222222, (int)0x44444444, (int)0x88888888}) == (int)0x44444444);
+static_assert(__builtin_reduce_max((vector4long){(long long)0x1111111111111111L, (long long)0x2222222222222222L, (long long)0x4444444444444444L, (long long)0x8888888888888888L}) == (long long)0x4444444444444444L);
+static_assert(__builtin_reduce_max((vector4uint){0x11111111U, 0x22222222U, 0x44444444U, 0x88888888U}) == 0x88888888U);
+static_assert(__builtin_reduce_max((vector4ulong){0x1111111111111111UL, 0x2222222222222222UL, 0x4444444444444444UL, 0x8888888888888888UL}) == 0x8888888888888888UL);
+
 static_assert(__builtin_bit_cast(unsigned, __builtin_elementwise_popcount((vector4char){1, 2, 3, 4})) == (LITTLE_END ? 0x01020101 : 0x01010201));
 static_assert(__builtin_bit_cast(unsigned long long, __builtin_elementwise_popcount((vector4short){0, 0x0F0F, ~0, ~0x0F0F})) == (LITTLE_END ? 0x0008001000080000 : 0x0000000800100008));
 static_assert(__builtin_reduce_add(__builtin_elementwise_popcount((vector4int){1, 2, 3, 4})) == 5);

compiler-rt/lib/tysan/tysan_interceptors.cpp

@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "interception/interception.h"
+#include "sanitizer_common/sanitizer_allocator_dlsym.h"
 #include "sanitizer_common/sanitizer_common.h"
 #include "tysan/tysan.h"
 
@@ -28,23 +29,11 @@ extern "C" int mallopt(int param, int value);
 using namespace __sanitizer;
 using namespace __tysan;
 
-static const uptr early_alloc_buf_size = 16384;
-static uptr allocated_bytes;
-static char early_alloc_buf[early_alloc_buf_size];
-
-static bool isInEarlyAllocBuf(const void *ptr) {
-  return ((uptr)ptr >= (uptr)early_alloc_buf &&
-          ((uptr)ptr - (uptr)early_alloc_buf) < sizeof(early_alloc_buf));
-}
-
-// Handle allocation requests early (before all interceptors are setup). dlsym,
-// for example, calls calloc.
-static void *handleEarlyAlloc(uptr size) {
-  void *mem = (void *)&early_alloc_buf[allocated_bytes];
-  allocated_bytes += size;
-  CHECK_LT(allocated_bytes, early_alloc_buf_size);
-  return mem;
-}
+namespace {
+struct DlsymAlloc : public DlSymAllocator<DlsymAlloc> {
+  static bool UseImpl() { return !tysan_inited; }
+};
+} // namespace
 
 INTERCEPTOR(void *, memset, void *dst, int v, uptr size) {
   if (!tysan_inited && REAL(memset) == nullptr)
@@ -111,16 +100,17 @@ INTERCEPTOR(char *, __strdup, const char *s) {
 #endif // TYSAN_INTERCEPT___STRDUP
 
 INTERCEPTOR(void *, malloc, uptr size) {
-  if (tysan_init_is_running && REAL(malloc) == nullptr)
-    return handleEarlyAlloc(size);
-
+  if (DlsymAlloc::Use())
+    return DlsymAlloc::Allocate(size);
   void *res = REAL(malloc)(size);
   if (res)
     tysan_set_type_unknown(res, size);
   return res;
 }
 
 INTERCEPTOR(void *, realloc, void *ptr, uptr size) {
+  if (DlsymAlloc::Use() || DlsymAlloc::PointerIsMine(ptr))
+    return DlsymAlloc::Realloc(ptr, size);
   void *res = REAL(realloc)(ptr, size);
   // We might want to copy the types from the original allocation (although
   // that would require that we knew its size).
@@ -130,21 +120,18 @@ INTERCEPTOR(void *, realloc, void *ptr, uptr size) {
 }
 
 INTERCEPTOR(void *, calloc, uptr nmemb, uptr size) {
-  if (tysan_init_is_running && REAL(calloc) == nullptr)
-    return handleEarlyAlloc(nmemb * size);
-
+  if (DlsymAlloc::Use())
+    return DlsymAlloc::Callocate(nmemb, size);
   void *res = REAL(calloc)(nmemb, size);
   if (res)
     tysan_set_type_unknown(res, nmemb * size);
   return res;
 }
 
-INTERCEPTOR(void, free, void *p) {
-  // There are only a few early allocation requests,
-  // so we simply skip the free.
-  if (isInEarlyAllocBuf(p))
-    return;
-  REAL(free)(p);
+INTERCEPTOR(void, free, void *ptr) {
+  if (DlsymAlloc::PointerIsMine(ptr))
+    return DlsymAlloc::Free(ptr);
+  REAL(free)(ptr);
 }
 
 INTERCEPTOR(void *, valloc, uptr size) {

libcxx/test/benchmarks/locale/num_get.bench.cpp

@@ -7,6 +7,8 @@
 //
 //===----------------------------------------------------------------------===//
 
+// UNSUPPORTED: c++03
+
 #include <ios>
 #include <locale>
 

llvm/include/llvm/Config/llvm-config.h.cmake

@@ -16,7 +16,7 @@
 
 /* Indicate that this is LLVM compiled from the amd-gfx branch. */
 #define LLVM_HAVE_BRANCH_AMD_GFX
-#define LLVM_MAIN_REVISION 522257
+#define LLVM_MAIN_REVISION 522261
 
 /* Define if LLVM_ENABLE_DUMP is enabled */
 #cmakedefine LLVM_ENABLE_DUMP