code review 7547050: crypto/rc4: faster amd64, 386 implementations

src/pkg/crypto/rc4/rc4_386.s

 // Copyright 2013 The Go Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
 // func xorKeyStream(dst, src *byte, n int, state *[256]byte, i, j *uint8)
 TEXT ·xorKeyStream(SB),7,$0
-        MOVQ dst+0(FP), DI
-        MOVQ src+8(FP), SI
-        MOVQ n+16(FP), CX
-        MOVQ state+24(FP), R8
+        MOVL dst+0(FP), DI
+        MOVL src+4(FP), SI
+        MOVL state+12(FP), BP
 
-        MOVQ xPtr+32(FP), AX
-        MOVBQZX (AX), AX
-        MOVQ yPtr+40(FP), BX
-        MOVBQZX (BX), BX
+        MOVL xPtr+16(FP), AX
+        MOVBLZX (AX), AX
+        MOVL yPtr+20(FP), BX
+        MOVBLZX (BX), BX
+        CMPL n+8(FP), $0
+        JEQ done
 
 loop:
-        CMPQ CX, $0
-        JE done
-
-        // c.i += 1
+        // i += 1
         INCB AX
 
-        // c.j += c.s[c.i]
-        MOVB (R8)(AX*1), R9
-        ADDB R9, BX
+        // j += c.s[i]
+        MOVBLZX (BP)(AX*1), DX
+        ADDB DX, BX
+        MOVBLZX BX, BX
 
-        MOVBQZX (R8)(BX*1), R10
+        // c.s[i], c.s[j] = c.s[j], c.s[i]
+        MOVBLZX (BP)(BX*1), CX
+        MOVB CX, (BP)(AX*1)
+        MOVB DX, (BP)(BX*1)
 
-        MOVB R10, (R8)(AX*1)
-        MOVB R9, (R8)(BX*1)
+        // *dst = *src ^ c.s[c.s[i]+c.s[j]]
+        ADDB DX, CX
+        MOVBLZX CX, CX
+        MOVB (BP)(CX*1), CX
+        XORB (SI), CX
+        MOVBLZX CX, CX
+        MOVB CX, (DI)
 
-        // R11 = c.s[c.i]+c.s[c.j]
-        MOVQ R10, R11
-        ADDB R9, R11
+        INCL SI
+        INCL DI
+        DECL n+8(FP)
+        JNE loop
 
-        MOVB (R8)(R11*1), R11
-        MOVB (SI), R12
-        XORB R11, R12
-        MOVB R12, (DI)
-
-        INCQ SI
-        INCQ DI
-        DECQ CX
-
-        JMP loop
 done:
-        MOVQ xPtr+32(FP), R8
-        MOVB AX, (R8)
-        MOVQ yPtr+40(FP), R8
-        MOVB BX, (R8)
+        MOVL xPtr+16(FP), CX
+        MOVB AX, (CX)
+        MOVL yPtr+20(FP), CX
+        MOVB BX, (CX)
 
         RET