[libc] Change the GPU to use builtin memory functions

Summary:
The GPU build is special in the sense that we always know that
up-to-date `clang` is always going to be the compiler. This allows us to
rely directly on builtins, which allow us to push a lot of this
complexity into the backend. Backend implementations are favored on
the GPU because it allows us to do a lot more target specific
optimizations. This patch changes over the common memory functions to
use builtin versions when building for AMDGPU or NVPTX.

Author: jhuber6

libc/src/string/memory_utils/generic/builtin.h

@@ -0,0 +1,47 @@
+//===-- Trivial builtin implementations  ----------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIBC_SRC_STRING_MEMORY_UTILS_GENERIC_BUILTIN_H
+#define LLVM_LIBC_SRC_STRING_MEMORY_UTILS_GENERIC_BUILTIN_H
+
+#include "src/__support/macros/attributes.h" // LIBC_INLINE
+#include "src/__support/macros/config.h"     // LIBC_HAS_BUILTIN
+#include "src/string/memory_utils/utils.h"   // Ptr, CPtr
+
+#include <stddef.h> // size_t
+
+namespace LIBC_NAMESPACE {
+
+static_assert(LIBC_HAS_BUILTIN(__builtin_memcpy), "Builtin not defined");
+static_assert(LIBC_HAS_BUILTIN(__builtin_memset), "Builtin not defined");
+static_assert(LIBC_HAS_BUILTIN(__builtin_memmove), "Builtin not defined");
+static_assert(LIBC_HAS_BUILTIN(__builtin_bcmp), "Builtin not defined");
+
+[[maybe_unused]] LIBC_INLINE void
+inline_memcpy_builtin(Ptr dst, CPtr src, size_t count, size_t offset = 0) {
+  __builtin_memcpy(dst + offset, src + offset, count);
+}
+
+[[maybe_unused]] LIBC_INLINE void inline_memmove_builtin(Ptr dst, CPtr src,
+                                                         size_t count) {
+  __builtin_memmove(dst, src, count);
+}
+
+[[maybe_unused]] LIBC_INLINE static void
+inline_memset_builtin(Ptr dst, uint8_t value, size_t count, size_t offset = 0) {
+  __builtin_memset(dst + offset, value, count);
+}
+
+[[maybe_unused]] LIBC_INLINE int
+inline_bcmp_builtin(CPtr p1, CPtr p2, size_t count, size_t offset = 0) {
+  return __builtin_bcmp(p1 + offset, p2 + offset, count);
+}
+
+} // namespace LIBC_NAMESPACE
+
+#endif // LLVM_LIBC_SRC_STRING_MEMORY_UTILS_GENERIC_BUILTIN_H

libc/src/string/memory_utils/inline_bcmp.h

@@ -23,9 +23,13 @@
 #elif defined(LIBC_TARGET_ARCH_IS_ANY_RISCV)
 #include "src/string/memory_utils/riscv/inline_bcmp.h"
 #define LIBC_SRC_STRING_MEMORY_UTILS_BCMP inline_bcmp_riscv
-#elif defined(LIBC_TARGET_ARCH_IS_ARM) || defined(LIBC_TARGET_ARCH_IS_GPU)
+// FIXME: The NVPTX builtin for `bcmp` currently does not work.
+#elif defined(LIBC_TARGET_ARCH_IS_ARM) || defined(LIBC_TARGET_ARCH_IS_NVPTX)
 #include "src/string/memory_utils/generic/byte_per_byte.h"
 #define LIBC_SRC_STRING_MEMORY_UTILS_BCMP inline_bcmp_byte_per_byte
+#elif defined(LIBC_TARGET_ARCH_IS_GPU)
+#include "src/string/memory_utils/generic/builtin.h"
+#define LIBC_SRC_STRING_MEMORY_UTILS_BCMP inline_bcmp_builtin
 #else
 #error "Unsupported architecture"
 #endif

libc/src/string/memory_utils/inline_memcpy.h

@@ -28,9 +28,12 @@
 #elif defined(LIBC_TARGET_ARCH_IS_ANY_RISCV)
 #include "src/string/memory_utils/riscv/inline_memcpy.h"
 #define LIBC_SRC_STRING_MEMORY_UTILS_MEMCPY inline_memcpy_riscv
-#elif defined(LIBC_TARGET_ARCH_IS_ARM) || defined(LIBC_TARGET_ARCH_IS_GPU)
+#elif defined(LIBC_TARGET_ARCH_IS_ARM)
 #include "src/string/memory_utils/generic/byte_per_byte.h"
 #define LIBC_SRC_STRING_MEMORY_UTILS_MEMCPY inline_memcpy_byte_per_byte
+#elif defined(LIBC_TARGET_ARCH_IS_GPU)
+#include "src/string/memory_utils/generic/builtin.h"
+#define LIBC_SRC_STRING_MEMORY_UTILS_MEMCPY inline_memcpy_builtin
 #else
 #error "Unsupported architecture"
 #endif

libc/src/string/memory_utils/inline_memmove.h

@@ -20,9 +20,12 @@
 #elif defined(LIBC_TARGET_ARCH_IS_ANY_RISCV)
 #include "src/string/memory_utils/riscv/inline_memmove.h"
 #define LIBC_SRC_STRING_MEMORY_UTILS_MEMMOVE inline_memmove_riscv
-#elif defined(LIBC_TARGET_ARCH_IS_ARM) || defined(LIBC_TARGET_ARCH_IS_GPU)
+#elif defined(LIBC_TARGET_ARCH_IS_ARM)
 #include "src/string/memory_utils/generic/byte_per_byte.h"
 #define LIBC_SRC_STRING_MEMORY_UTILS_MEMMOVE inline_memmove_byte_per_byte
+#elif defined(LIBC_TARGET_ARCH_IS_GPU)
+#include "src/string/memory_utils/generic/builtin.h"
+#define LIBC_SRC_STRING_MEMORY_UTILS_MEMMOVE inline_memmove_builtin
 #else
 #error "Unsupported architecture"
 #endif

libc/src/string/memory_utils/inline_memset.h

@@ -24,9 +24,12 @@
 #elif defined(LIBC_TARGET_ARCH_IS_ANY_RISCV)
 #include "src/string/memory_utils/riscv/inline_memset.h"
 #define LIBC_SRC_STRING_MEMORY_UTILS_MEMSET inline_memset_riscv
-#elif defined(LIBC_TARGET_ARCH_IS_ARM) || defined(LIBC_TARGET_ARCH_IS_GPU)
+#elif defined(LIBC_TARGET_ARCH_IS_ARM)
 #include "src/string/memory_utils/generic/byte_per_byte.h"
 #define LIBC_SRC_STRING_MEMORY_UTILS_MEMSET inline_memset_byte_per_byte
+#elif defined(LIBC_TARGET_ARCH_IS_GPU)
+#include "src/string/memory_utils/generic/builtin.h"
+#define LIBC_SRC_STRING_MEMORY_UTILS_MEMSET inline_memset_builtin
 #else
 #error "Unsupported architecture"
 #endif