Fix asm for dynamic linking

R15 and CX are both clobbered under dynamic linking; avoid them.

Fixes #54

xxhash_amd64.s

diff --git a/xxhash_amd64.s b/xxhash_amd64.s
index d580e32..be8db5b 100644
--- a/xxhash_amd64.s
+++ b/xxhash_amd64.s
@@ -6,7 +6,7 @@
 
 // Register allocation:
 // AX	h
-// CX	pointer to advance through b
+// SI	pointer to advance through b
 // DX	n
 // BX	loop end
 // R8	v1, k1
@@ -16,39 +16,39 @@
 // R12	tmp
 // R13	prime1v
 // R14	prime2v
-// R15	prime4v
+// DI	prime4v
 
-// round reads from and advances the buffer pointer in CX.
+// round reads from and advances the buffer pointer in SI.
 // It assumes that R13 has prime1v and R14 has prime2v.
 #define round(r) \
-	MOVQ  (CX), R12 \
-	ADDQ  $8, CX    \
+	MOVQ  (SI), R12 \
+	ADDQ  $8, SI    \
 	IMULQ R14, R12  \
 	ADDQ  R12, r    \
 	ROLQ  $31, r    \
 	IMULQ R13, r
 
 // mergeRound applies a merge round on the two registers acc and val.
-// It assumes that R13 has prime1v, R14 has prime2v, and R15 has prime4v.
+// It assumes that R13 has prime1v, R14 has prime2v, and DI has prime4v.
 #define mergeRound(acc, val) \
 	IMULQ R14, val \
 	ROLQ  $31, val \
 	IMULQ R13, val \
 	XORQ  val, acc \
 	IMULQ R13, acc \
-	ADDQ  R15, acc
+	ADDQ  DI, acc
 
 // func Sum64(b []byte) uint64
 TEXT ·Sum64(SB), NOSPLIT, $0-32
 	// Load fixed primes.
 	MOVQ ·prime1v(SB), R13
 	MOVQ ·prime2v(SB), R14
-	MOVQ ·prime4v(SB), R15
+	MOVQ ·prime4v(SB), DI
 
 	// Load slice.
-	MOVQ b_base+0(FP), CX
+	MOVQ b_base+0(FP), SI
 	MOVQ b_len+8(FP), DX
-	LEAQ (CX)(DX*1), BX
+	LEAQ (SI)(DX*1), BX
 
 	// The first loop limit will be len(b)-32.
 	SUBQ $32, BX
@@ -65,14 +65,14 @@
 	XORQ R11, R11
 	SUBQ R13, R11
 
-	// Loop until CX > BX.
+	// Loop until SI > BX.
 blockLoop:
 	round(R8)
 	round(R9)
 	round(R10)
 	round(R11)
 
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JLE  blockLoop
 
 	MOVQ R8, AX
@@ -100,16 +100,16 @@
 afterBlocks:
 	ADDQ DX, AX
 
-	// Right now BX has len(b)-32, and we want to loop until CX > len(b)-8.
+	// Right now BX has len(b)-32, and we want to loop until SI > len(b)-8.
 	ADDQ $24, BX
 
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JG   fourByte
 
 wordLoop:
 	// Calculate k1.
-	MOVQ  (CX), R8
-	ADDQ  $8, CX
+	MOVQ  (SI), R8
+	ADDQ  $8, SI
 	IMULQ R14, R8
 	ROLQ  $31, R8
 	IMULQ R13, R8
@@ -117,18 +117,18 @@
 	XORQ  R8, AX
 	ROLQ  $27, AX
 	IMULQ R13, AX
-	ADDQ  R15, AX
+	ADDQ  DI, AX
 
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JLE  wordLoop
 
 fourByte:
 	ADDQ $4, BX
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JG   singles
 
-	MOVL  (CX), R8
-	ADDQ  $4, CX
+	MOVL  (SI), R8
+	ADDQ  $4, SI
 	IMULQ R13, R8
 	XORQ  R8, AX
 
@@ -138,19 +138,19 @@
 
 singles:
 	ADDQ $4, BX
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JGE  finalize
 
 singlesLoop:
-	MOVBQZX (CX), R12
-	ADDQ    $1, CX
+	MOVBQZX (SI), R12
+	ADDQ    $1, SI
 	IMULQ   ·prime5v(SB), R12
 	XORQ    R12, AX
 
 	ROLQ  $11, AX
 	IMULQ R13, AX
 
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JL   singlesLoop
 
 finalize:
@@ -179,9 +179,9 @@
 	MOVQ ·prime2v(SB), R14
 
 	// Load slice.
-	MOVQ b_base+8(FP), CX
+	MOVQ b_base+8(FP), SI
 	MOVQ b_len+16(FP), DX
-	LEAQ (CX)(DX*1), BX
+	LEAQ (SI)(DX*1), BX
 	SUBQ $32, BX
 
 	// Load vN from d.
@@ -199,7 +199,7 @@
 	round(R10)
 	round(R11)
 
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JLE  blockLoop
 
 	// Copy vN back to d.
@@ -208,8 +208,8 @@
 	MOVQ R10, 16(AX)
 	MOVQ R11, 24(AX)
 
-	// The number of bytes written is CX minus the old base pointer.
-	SUBQ b_base+8(FP), CX
-	MOVQ CX, ret+32(FP)
+	// The number of bytes written is SI minus the old base pointer.
+	SUBQ b_base+8(FP), SI
+	MOVQ SI, ret+32(FP)
 
 	RET