symm/*.S: Symbolic names for shuffles.

[catacomb] / symm / chacha-x86-sse2.S

diff --git a/symm/chacha-x86-sse2.S b/symm/chacha-x86-sse2.S
index 87fb096..ccdfa53 100644 (file)
--- a/symm/chacha-x86-sse2.S
+++ b/symm/chacha-x86-sse2.S
@@ -31,6 +31,14 @@
 #include "asm-common.h"
 
 ///--------------------------------------------------------------------------
+/// Local utilities.
+
+// Magic constants for shuffling.
+#define ROTL 0x93
+#define ROT2 0x4e
+#define ROTR 0x39
+
+///--------------------------------------------------------------------------
 /// Main code.
 
        .arch pentium4
@@ -101,9 +109,9 @@ loop:
 
        // c += d; b ^= c; b <<<=  7
        paddd   xmm2, xmm3
-       pshufd  xmm3, xmm3, 0x93
+       pshufd  xmm3, xmm3, ROTL
        pxor    xmm1, xmm2
-       pshufd  xmm2, xmm2, 0x4e
+       pshufd  xmm2, xmm2, ROT2
        movdqa  xmm4, xmm1
        pslld   xmm1, 7
        psrld   xmm4, 25
@@ -121,7 +129,7 @@ loop:
        //
        // The shuffles have quite high latency, so they've mostly been
        // pushed upwards.  The remaining one can't be moved, though.
-       pshufd  xmm1, xmm1, 0x39
+       pshufd  xmm1, xmm1, ROTR
 
        // Apply the diagonal quarterround to each of the columns
        // simultaneously.
@@ -152,9 +160,9 @@ loop:
 
        // c += d; b ^= c; b <<<=  7
        paddd   xmm2, xmm3
-       pshufd  xmm3, xmm3, 0x39
+       pshufd  xmm3, xmm3, ROTR
        pxor    xmm1, xmm2
-       pshufd  xmm2, xmm2, 0x4e
+       pshufd  xmm2, xmm2, ROT2
        movdqa  xmm4, xmm1
        pslld   xmm1, 7
        psrld   xmm4, 25
@@ -163,7 +171,7 @@ loop:
        // Finally, finish off undoing the transpose, and we're done for this
        // doubleround.  Again, most of this was done above so we don't have
        // to wait for the shuffles.
-       pshufd  xmm1, xmm1, 0x93
+       pshufd  xmm1, xmm1, ROTL
 
        // Decrement the loop counter and see if we should go round again.
        sub     ecx, 2