[PATCH v2 2/3] msvcrt: Check for ERMS support and use rep stosb for large memset calls.

[Rémi Bernon]

Signed-off-by: Rémi Bernon <rbernon at codeweavers.com>
---
 dlls/msvcrt/math.c   | 13 +++++++++++++
 dlls/msvcrt/msvcrt.h |  1 +
 dlls/msvcrt/string.c |  8 ++++++++
 3 files changed, 22 insertions(+)

diff --git a/dlls/msvcrt/math.c b/dlls/msvcrt/math.c
index 7f59a4d20d4..9974e72d78f 100644
--- a/dlls/msvcrt/math.c
+++ b/dlls/msvcrt/math.c
@@ -43,6 +43,7 @@
 #include <limits.h>
 #include <locale.h>
 #include <math.h>
+#include <intrin.h>
 
 #include "msvcrt.h"
 #include "winternl.h"
@@ -64,11 +65,23 @@ typedef int (CDECL *MSVCRT_matherr_func)(struct _exception *);
 
 static MSVCRT_matherr_func MSVCRT_default_matherr_func = NULL;
 
+BOOL erms_supported;
 BOOL sse2_supported;
 static BOOL sse2_enabled;
 
 void msvcrt_init_math( void *module )
 {
+#if defined(__i386__) || defined(__x86_64__)
+    int regs[4];
+
+    __cpuid(regs, 0);
+    if (regs[0] >= 7)
+    {
+        __cpuidex(regs, 7, 0);
+        erms_supported = ((regs[1] >> 9) & 1);
+    }
+#endif
+
     sse2_supported = IsProcessorFeaturePresent( PF_XMMI64_INSTRUCTIONS_AVAILABLE );
 #if _MSVCR_VER <=71
     sse2_enabled = FALSE;
diff --git a/dlls/msvcrt/msvcrt.h b/dlls/msvcrt/msvcrt.h
index 60f8c2f5ef2..022eced35d9 100644
--- a/dlls/msvcrt/msvcrt.h
+++ b/dlls/msvcrt/msvcrt.h
@@ -33,6 +33,7 @@
 #undef strncpy
 #undef wcsncpy
 
+extern BOOL erms_supported DECLSPEC_HIDDEN;
 extern BOOL sse2_supported DECLSPEC_HIDDEN;
 
 #define DBL80_MAX_10_EXP 4932
diff --git a/dlls/msvcrt/string.c b/dlls/msvcrt/string.c
index b470a875607..26bb9cd8ba4 100644
--- a/dlls/msvcrt/string.c
+++ b/dlls/msvcrt/string.c
@@ -2857,6 +2857,14 @@ void * __cdecl memcpy(void *dst, const void *src, size_t n)
 
 static void memset_aligned_32(unsigned char *d, uint64_t v, size_t n)
 {
+#if defined(__i386__) || defined(__x86_64__)
+    if (n >= 2048 && erms_supported)
+    {
+        unsigned char c = v;
+        __asm__ __volatile__ ("cld; rep; stosb" : "+D"(d), "+c"(n) : "a"(c) : "memory", "cc");
+        return;
+    }
+#endif
     while (n >= 32)
     {
         *(uint64_t *)(d + n - 32) = v;
-- 
2.33.0