@@@ i386 wip

diff --git a/symm/chacha-x86ish-sse2.S b/symm/chacha-x86ish-sse2.S
index 974ec5b..d21a624 100644 (file)
--- a/symm/chacha-x86ish-sse2.S
+++ b/symm/chacha-x86ish-sse2.S
@@ -259,4 +259,138 @@ FUNC(chacha_core_x86ish_sse2)
 
 ENDFUNC
 
+FUNC(chacha_multi_i386_sse2)
+       // Arguments are on the stack:
+       //
+       // [sp +  4]    pointer to state
+       // [sp +  8]    input pointer (or null)
+       // [sp + 12]    output pointer
+       // [sp + 16]    number of blocks to process
+       // [sp + 20]    number of rounds per block
+
+       pushreg SI
+       pushreg DI
+       pushreg BX
+       stalloc 4*64
+  endprologue
+
+       // Load the arguments.
+       mov     BX, [SP + 272]          // = state pointer
+       mov     SI, [SP + 276]          // = source pointer
+       mov     DI, [SP + 280]          // = destination pointer
+       mov     CX, [SP + 284]          // = block count
+       mov     DX, [SP + 288]          // = (initial) round count
+
+       // Do chunks of four blocks at a time.
+       sub     CX, 4
+       jb      8f
+
+       // Inhale the initial state.
+       movdqu  xmm1, [BX +  0]
+       movdqu  xmm3, [BX + 16]
+       movdqu  xmm5, [BX + 32]
+       movdqu  xmm0, [BX + 48]
+
+       // Set the counters and initialize the working blocks.
+       pxor    xmm2, xmm2
+       pxor    xmm4, xmm4
+       pxor    xmm6, xmm6
+       pxor    xmm7, xmm7
+
+       xor     eax, eax
+       mov     al, 1
+       pinsrw  xmm2, eax, 4
+       mov     al, 2
+       pinsrw  xmm4, eax, 4
+       mov     al, 3
+       pinsrw  xmm6, eax, 4
+       mov     al, 4
+       pinsrw  xmm7, eax, 4
+
+       movdqa  [SP +  16], xmm3
+       movdqa  [SP +  32], xmm5
+       movdqa  [SP +  48], xmm0
+
+       paddq   xmm2, xmm3
+       paddq   xmm4, xmm3
+       paddq   xmm6, xmm3
+       paddq   xmm7, xmm3
+
+       movdqu  [BX + 48], xmm7
+
+       // a += b; d ^= a; d <<<= 16
+       paddd   xmm1, xmm3              // a += b
+
+       movdqa  [SP +   0], xmm1
+
+       pxor    xmm0, xmm1              // d ^= a
+       pxor    xmm2, xmm1
+       pxor    xmm4, xmm1
+       pxor    xmm6, xmm1
+
+       movdqa  xmm1, xmm0
+       movdqa  xmm3, xmm2
+       movdqa  xmm5, xmm4
+       movdqa  xmm7, xmm6
+
+       pslld   xmm0, 16                // d << 16
+       pslld   xmm2, 16
+       pslld   xmm4, 16
+       pslld   xmm6, 16
+
+       pslrd   xmm1, 16                // d >> 16
+       pslrd   xmm3, 16
+       pslrd   xmm5, 16
+       pslrd   xmm7, 16
+
+       por     xmm0, xmm1              // d <<<= 16
+        movdqa xmm1, [SP + 32]
+       por     xmm2, xmm3
+        movdqa xmm3, [SP + 48]
+       por     xmm4, xmm5
+       por     xmm6, xmm7
+
+       movdqa  [SP +  48], xmm0
+       movdqa  [SP + 112], xmm2
+       movdqa  [SP + 176], xmm4
+       movdqa  [SP + 240], xmm6
+
+       // c += d; b ^= c; c <<<= 12
+       paddd   xmm0, xmm1              // c += d
+       paddd   xmm2, xmm1
+       paddd   xmm4, xmm1
+       paddd   xmm6, xmm1
+
+       movdqa  [SP +  32], xmm0
+       movdqa  [SP +  96], xmm0
+       movdqa  [SP + 160], xmm0
+       movdqa  [SP + 224], xmm0
+
+       pxor    xmm0, xmm3              // b ^= c
+       pxor    xmm2, xmm3
+       pxor    xmm4, xmm3
+       pxor    xmm6, xmm3
+
+       movdqa  xmm1, xmm0
+       movdqa  xmm3, xmm2
+       movdqa  xmm5, xmm4
+       movdqa  xmm7, xmm6
+
+       pslld   xmm0, 16                // d << 16
+       pslld   xmm2, 16
+       pslld   xmm4, 16
+       pslld   xmm6, 16
+
+       pslrd   xmm1, 16                // d >> 16
+       pslrd   xmm3, 16
+       pslrd   xmm5, 16
+       pslrd   xmm7, 16
+
+       por     xmm0, xmm1              // d <<<= 16
+       por     xmm2, xmm3
+       por     xmm4, xmm5
+       por     xmm6, xmm7
+
+ENDFUNC
+
 ///----- That's all, folks --------------------------------------------------