Fixup previous memcpy optimization.

Different gcc versions handle __builtin_memcpy differently.
Add -minline-all-string to force inlining of memcpy on old gcc.
Always use __builtin_memcpy for all memcpy calls.
Use memcpy4() for the option rom case where 4-byte accesses is important.

3 files changed, 7 insertions, 20 deletions

diff --git a/src/optionroms.c b/src/optionroms.c
index def3feb..32fb44b 100644
--- a/src/optionroms.c
+++ b/src/optionroms.c
@@ -171,7 +171,7 @@ copy_rom(struct rom_header *rom)
     }
     dprintf(4, "Copying option rom (size %d) from %p to %x\n"
             , romsize, rom, next_rom);
-    memcpy((void*)next_rom, rom, romsize);
+    memcpy4((void*)next_rom, rom, romsize);
     return (struct rom_header *)next_rom;
 }
 
diff --git a/src/util.c b/src/util.c
index a39668b..3c7117c 100644
--- a/src/util.c
+++ b/src/util.c
@@ -181,25 +181,15 @@ memcpy_far(u16 d_seg, void *d_far, u16 s_seg, const void *s_far, size_t len)
         : "cc", "memory");
 }
 
-noinline void *
-__memcpy(void *d1, const void *s1, size_t len)
+// Memcpy that uses 4-byte accesses
+void
+memcpy4(void *d1, const void *s1, size_t len)
 {
-    void *d = d1;
-    if (((u32)d1 | (u32)s1 | len) & 3) {
-        // non-aligned memcpy
-        asm volatile(
-            "rep movsb (%%esi),%%es:(%%edi)\n"
-            : "+c"(len), "+S"(s1), "+D"(d)
-            : : "cc", "memory");
-        return d1;
-    }
-    // Common case - use 4-byte copy
     len /= 4;
     asm volatile(
         "rep movsl (%%esi),%%es:(%%edi)\n"
-        : "+c"(len), "+S"(s1), "+D"(d)
+        : "+c"(len), "+S"(s1), "+D"(d1)
         : : "cc", "memory");
-    return d1;
 }
 
 void *
diff --git a/src/util.h b/src/util.h
index ceb775e..bb1a445 100644
--- a/src/util.h
+++ b/src/util.h
@@ -73,11 +73,8 @@ int memcmp(const void *s1, const void *s2, size_t n);
 size_t strlen(const char *s);
 int strcmp(const char *s1, const char *s2);
 void *memset(void *s, int c, size_t n);
-void *__memcpy(void *d1, const void *s1, size_t len);
-#define memcpy(d1, s1, len) (                           \
-        (__builtin_constant_p(len) && (len) <= 20)      \
-        ? __builtin_memcpy((d1), (s1), (len))           \
-        :  __memcpy((d1), (s1), (len)))
+void memcpy4(void *d1, const void *s1, size_t len);
+#define memcpy(d1, s1, len) __builtin_memcpy((d1), (s1), (len))
 inline void memcpy_far(u16 d_seg, void *d_far
                        , u16 s_seg, const void *s_far, size_t len);
 void *memmove(void *d, const void *s, size_t len);