Use SSE instructions for prefetch when __builtin_prefetch is unavailable

This notably gets prefetch working on MSVC

Implementation note:
https://docs.microsoft.com/en-us/windows/win32/api/winnt/nf-winnt-prefetchcacheline
MSVC does have PreFetchCacheLine, but that would require including <windows.h>
in a header

PiperOrigin-RevId: 449602543
Change-Id: I5e6ca4b7c3d287779aa03c2fd348b41fb65c3680

1 files changed, 27 insertions, 0 deletions

diff --git a/absl/base/internal/prefetch.h b/absl/base/internal/prefetch.h
index d4971a68..06419283 100644
--- a/absl/base/internal/prefetch.h
+++ b/absl/base/internal/prefetch.h
@@ -17,6 +17,15 @@
 
 #include "absl/base/config.h"
 
+#ifdef __SSE__
+#include <xmmintrin.h>
+#endif
+
+#if defined(_MSC_VER) && defined(ABSL_INTERNAL_HAVE_SSE)
+#include <intrin.h>
+#pragma intrinsic(_mm_prefetch)
+#endif
+
 // Compatibility wrappers around __builtin_prefetch, to prefetch data
 // for read if supported by the toolchain.
 
@@ -97,6 +106,24 @@ inline void PrefetchNta(const void* addr) {
   // Note: this uses prefetchtnta on Intel.
   __builtin_prefetch(addr, 0, 0);
 }
+
+#elif defined(ABSL_INTERNAL_HAVE_SSE)
+
+#define ABSL_INTERNAL_HAVE_PREFETCH 1
+
+inline void PrefetchT0(const void* addr) {
+  _mm_prefetch(reinterpret_cast<const char*>(addr), _MM_HINT_T0);
+}
+inline void PrefetchT1(const void* addr) {
+  _mm_prefetch(reinterpret_cast<const char*>(addr), _MM_HINT_T1);
+}
+inline void PrefetchT2(const void* addr) {
+  _mm_prefetch(reinterpret_cast<const char*>(addr), _MM_HINT_T2);
+}
+inline void PrefetchNta(const void* addr) {
+  _mm_prefetch(reinterpret_cast<const char*>(addr), _MM_HINT_NTA);
+}
+
 #else
 inline void PrefetchT0(const void*) {}
 inline void PrefetchT1(const void*) {}