--- a/configure
+++ b/configure
@@ -1274,7 +1274,9 @@ echo "yes"
 if test -z "$_target" && x86 ; then
   cat > $TMPC << EOF
 int main(void) {
+#ifndef __x86_64__
     int test[(int)sizeof(char *)-7];
+#endif
     return 0;
 }
 EOF
@@ -1393,7 +1395,7 @@ fi
 def_fast_64bit='#define HAVE_FAST_64BIT 0'
 def_fast_unaligned='#define HAVE_FAST_UNALIGNED 0'
 arch_all='X86 IA64 SPARC ARM AVR32 SH4 PPC ALPHA MIPS PA_RISC S390 S390X VAX BFIN XTENSA TOMI GENERIC'
-subarch_all='X86_32 X86_64 PPC64'
+subarch_all='X86_32 X86_64 X86_64_X64 X86_64_X32 PPC64'
 case "$host_arch" in
   i[3-9]86|x86|x86pc|k5|k6|k6-2|k6-3|pentium*|athlon*|i586-i686)
   arch='x86'
@@ -1644,6 +1646,14 @@ case "$host_arch" in
   x86_64|amd64)
     arch='x86'
     subarch='x86_64'
+    cat > $TMPC << EOF
+int main(void) {
+    int test[(int)sizeof(char *)-7];
+    return 0;
+}
+EOF
+    cc_check && subarch="$subarch x86_64_x64" || subarch="$subarch x86_64_x32" CFLAGS="$CFLAGS -DMPG123_NO_LARGENAME"
+
     def_fast_unaligned='#define HAVE_FAST_UNALIGNED 1'
     def_fast_64bit='#define HAVE_FAST_64BIT 1'
     iproc='x86_64'
diff --git a/cpudetect.c b/cpudetect.c
index fb93116..edf4f0a 100644
--- a/cpudetect.c
+++ b/cpudetect.c
@@ -233,9 +233,9 @@ do_cpuid(unsigned int ax, unsigned int *p)
 {
 // code from libavcodec:
     __asm__ volatile
-        ("mov %%"REG_b", %%"REG_S"\n\t"
+        ("mov %%"REG_rb", %%"REG_rS"\n\t"
          "cpuid\n\t"
-         "xchg %%"REG_b", %%"REG_S
+         "xchg %%"REG_rb", %%"REG_rS
          : "=a" (p[0]), "=S" (p[1]),
            "=c" (p[2]), "=d" (p[3])
          : "0" (ax));
diff --git a/ffmpeg_files/x86_cpu.h b/ffmpeg_files/x86_cpu.h
index 49a5b28..46ca592 100644
--- a/ffmpeg_files/x86_cpu.h
+++ b/ffmpeg_files/x86_cpu.h
@@ -24,13 +24,15 @@
 #include <stdint.h>
 #include "config.h"
 
-#if ARCH_X86_64
+#if ARCH_X86_64_X64
+#    define REG_PTR_MOD "q"
 #    define REG_a "rax"
 #    define REG_b "rbx"
 #    define REG_c "rcx"
 #    define REG_d "rdx"
 #    define REG_D "rdi"
 #    define REG_S "rsi"
+#    define REG_8 "r8"
 #    define PTR_SIZE "8"
 typedef int64_t x86_reg;
 
@@ -42,9 +44,11 @@ typedef int64_t x86_reg;
 #    define REGc    rcx
 #    define REGd    rdx
 #    define REGSP   rsp
+#    define REG8    r8
 
-#elif ARCH_X86_32
+#elif ARCH_X86_64_X32 || ARCH_X86_32
 
+#    define REG_PTR_MOD "k"
 #    define REG_a "eax"
 #    define REG_b "ebx"
 #    define REG_c "ecx"
@@ -62,10 +66,31 @@ typedef int32_t x86_reg;
 #    define REGc    ecx
 #    define REGd    edx
 #    define REGSP   esp
+#    define REG8    r8d
 #else
 typedef int x86_reg;
 #endif
 
+#if ARCH_X86_64
+#    define REG_ra "rax"
+#    define REG_rb "rbx"
+#    define REG_rc "rcx"
+#    define REG_rd "rdx"
+#    define REG_rD "rdi"
+#    define REG_rS "rsi"
+#    define REG_rBP "rbp"
+typedef int64_t x86_native_reg;
+#elif ARCH_X86_32
+#    define REG_ra "eax"
+#    define REG_rb "ebx"
+#    define REG_rc "ecx"
+#    define REG_rd "edx"
+#    define REG_rD "edi"
+#    define REG_rS "esi"
+#    define REG_rBP "ebp"
+typedef int32_t x86_native_reg;
+#endif
+
 #define HAVE_7REGS (ARCH_X86_64 || (HAVE_EBX_AVAILABLE && HAVE_EBP_AVAILABLE))
 #define HAVE_6REGS (ARCH_X86_64 || (HAVE_EBX_AVAILABLE || HAVE_EBP_AVAILABLE))
 
diff --git a/libmpcodecs/vf_ilpack.c b/libmpcodecs/vf_ilpack.c
index db4a849..4bdc7c6 100644
--- a/libmpcodecs/vf_ilpack.c
+++ b/libmpcodecs/vf_ilpack.c
@@ -114,7 +114,7 @@ static void pack_li_0_MMX(unsigned char *dst, unsigned char *y,
     unsigned char *u, unsigned char *v, int w, int us, int vs)
 {
     __asm__ volatile (""
-        "push %%"REG_BP" \n\t"
+        "push %%"REG_rBP" \n\t"
 #if ARCH_X86_64
         "mov %6, %%"REG_BP" \n\t"
 #else
@@ -205,7 +205,7 @@ static void pack_li_0_MMX(unsigned char *dst, unsigned char *y,
         "decl %%ecx \n\t"
         "jnz .Lli0 \n\t"
         "emms \n\t"
-        "pop %%"REG_BP" \n\t"
+        "pop %%"REG_rBP" \n\t"
         :
         : "S" (y), "D" (dst), "a" (u), "b" (v), "c" (w/16),
 #if ARCH_X86_64
@@ -222,7 +222,7 @@ static void pack_li_1_MMX(unsigned char *dst, unsigned char *y,
     unsigned char *u, unsigned char *v, int w, int us, int vs)
 {
     __asm__ volatile (""
-        "push %%"REG_BP" \n\t"
+        "push %%"REG_rBP" \n\t"
 #if ARCH_X86_64
         "mov %6, %%"REG_BP" \n\t"
 #else
@@ -317,7 +317,7 @@ static void pack_li_1_MMX(unsigned char *dst, unsigned char *y,
         "decl %%ecx \n\t"
         "jnz .Lli1 \n\t"
         "emms \n\t"
-        "pop %%"REG_BP" \n\t"
+        "pop %%"REG_rBP" \n\t"
         :
         : "S" (y), "D" (dst), "a" (u), "b" (v), "c" (w/16),
 #if ARCH_X86_64