xchg.S: Another couple of (easy) exercises.

diff --git a/xchg.S b/xchg.S
index c99e0a612e5bbd608b73ac2d39136b5911dd97a7..40c5ba9e4e6834198cb020412d44690a5bce9979 100644 (file)
--- a/xchg.S
+++ b/xchg.S
@@ -2750,50 +2750,101 @@ endproc
 
 proc   x26
 
 
 proc   x26
 

+       // a bad way to rotate a right by 7 places
+

 #if defined(__x86_64__)
 
 #if defined(__x86_64__)
 

-       notimpl
+       mov     rbx, rax
+       ror     rbx, 7                  // better
+
+       mov     rdx, rax                // d' = a
+       shr     rax, 7                  // a' = a >> 7
+       shl     rdx, 0x39               // d' = a << 57
+       or      rax, rdx                // a' = a >>> 7

 
 #elif defined(__i386__)
 
 
 #elif defined(__i386__)
 

-       notimpl
+       mov     ebx, eax
+       ror     ebx, 7                  // better
+
+       mov     edx, eax                // d' = a
+       shr     eax, 7                  // a' = a >> 7
+       shl     edx, 0x39               // d' = a << 57
+       or      eax, edx                // a' = a >>> 7

 
 #elif defined(__arm__)
 
 
 #elif defined(__arm__)
 

-       notimpl
+       mov     r1, r0, ror #7          // easy way
+
+       // even the hard way is fairly easy on arm
+       mov     r3, r0, lsl #25
+       orr     r0, r3, r0, lsr #7      // hard way

 
 #elif defined(__aarch64__)
 
 
 #elif defined(__aarch64__)
 

-       notimpl
+       ror     x1, x0, #7              // easy way
+
+       // even the hard way is fairly easy on arm
+       lsl     x3, x0, #57
+       orr     x0, x3, x0, lsr #7      // hard way

 
 #else
        notimpl
 #endif
 
 
 #else
        notimpl
 #endif
 

+       ret
+

 endproc
 
 proc   x27
 
 endproc
 
 proc   x27
 

+       // shift a right by c places, in two halves
+

 #if defined(__x86_64__)
 
 #if defined(__x86_64__)
 

-       notimpl
+       mov     ch, cl                  // c' = [c, c]
+       inc     ch                      // c' = [c, c + 1]
+       shr     ch, 1
+       shr     cl, 1                   // c' = [floor(c/2), ceil(c/2)]
+       shr     rax, cl
+       xchg    ch, cl
+       shr     rax, cl

 
 #elif defined(__i386__)
 
 
 #elif defined(__i386__)
 

-       notimpl
+       mov     ch, cl                  // c' = [c, c]
+       inc     ch                      // c' = [c, c + 1]
+       shr     ch, 1
+       shr     cl, 1                   // c' = [floor(c/2), ceil(c/2)]
+       shr     eax, cl
+       xchg    ch, cl
+       shr     eax, cl

 
 #elif defined(__arm__)
 
 
 #elif defined(__arm__)
 

-       notimpl
+       // it would be clearer and more efficient to say: `mov r12, r2, lsr
+       // #1; sub r2, r2, r12', but that's not the lesson this exercise is
+       // trying to teach.
+       add     r12, r2, #1
+       mov     r2, r2, lsr #1
+       mov     r12, r12, lsr #1
+       mov     r0, r0, lsr r2
+       mov     r0, r0, lsr r12

 
 #elif defined(__aarch64__)
 
 
 #elif defined(__aarch64__)
 

-       notimpl
+       add     w16, w2, #1
+       lsr     w2, w2, #1
+       lsr     w16, w16, #1
+       lsr     x0, x0, x2
+       lsr     x0, x0, x16

 
 #else
        notimpl
 #endif
 
 
 #else
        notimpl
 #endif
 

+       ret
+

 endproc
 
 proc   x28
 endproc
 
 proc   x28