x86-64-asm: Fix mov im64,rax encoding

the avoidance of mov im32->reg64 wasn't working when reg64 was rax.
While fixing this also fix instructions which had the REX prefix
hardcoded in opcode and so didn't support extended registers which
would have added another REX prefix.

tests/asmtest.S

@@ -114,12 +114,21 @@ notl %r15d
      movzb 0x1000, %eax
      movzb 0x1000, %ax
                 
+     mov $0x12345678,%eax
+
 #ifdef __x86_64__
      movzb 0x1000, %rax
      movzbq 0x1000, %rbx
      movsbq 0x1000, %rdx
      movzwq 0x1000, %rdi
      movswq 0x1000, %rdx
+     movslq %eax, %rcx
+     mov $0x12345678,%rax
+     mov $0x12345678,%rdx
+     mov $0x12345678,%r10
+     mov $0x123456789abcdef0,%rax
+     mov $0x123456789abcdef0,%rcx
+     mov $0x123456789abcdef0,%r11
 #endif
         
 #ifdef __i386__
@@ -546,6 +555,7 @@ invlpg 0x1000
 cmpxchg8b 0x1002
 #ifdef __x86_64__
 cmpxchg16b (%rax)
+cmpxchg16b (%r10,%r11)
 #endif
 
 fcmovb %st(5), %st
@@ -569,6 +579,7 @@ fucomip %st(5), %st
  cmovne %ax, %si
 #ifdef __x86_64__
  bswapq %rsi
+ bswapq %r10
  cmovz %rdi,%rbx
 #endif
 
@@ -675,7 +686,9 @@ int $0x10
     prefetchw (%rdi)
     clflush 0x1000(%rax,%rcx)
     fxsaveq (%rdx)
+    fxsaveq (%r11)
     fxrstorq (%rcx)
+    fxrstorq (%r10)
 
 #endif
 
@@ -751,6 +764,9 @@ int $0x10
     sidtq 0x1000
 
     swapgs
+
+    str %rdx
+    str %r9
 #endif
 
     lmsw 0x1000
@@ -879,6 +895,7 @@ overrideme:
 #ifdef __x86_64__
     movq %rcx, %mm1
     movq %rdx, %xmm2
+    movq %r13, %xmm3
     /* movq mem64->xmm is encoded as f30f7e by GAS, but as
        660f6e by tcc (which really is a movd and would need 
        a REX.W prefix to be movq).  */