Optimize Skein

- Eliminate modulus math (It runs slow inside microcode).
- Convert whatever the hell was going on into a rotate op. Like... Someone kinda reinvented the wheel in order to do a simple rotate.

1 files changed, 48 insertions, 30 deletions

diff --git a/xmrstak/backend/amd/amd_gpu/opencl/wolf-skein.cl b/xmrstak/backend/amd/amd_gpu/opencl/wolf-skein.cl
index 868757b..bebc2ab 100644
--- a/xmrstak/backend/amd/amd_gpu/opencl/wolf-skein.cl
+++ b/xmrstak/backend/amd/amd_gpu/opencl/wolf-skein.cl
@@ -22,68 +22,59 @@ static const __constant ulong SKEIN512_256_IV[8] =
 	0xC36FBAF9393AD185UL, 0x3EEDBA1833EDFC13UL
 };
 
-#define SKEIN_INJECT_KEY(p, s)	do { \
+#define SKEIN_INJECT_KEY(p, s, q)	do { \
 	p += h; \
-	p.s5 += t[s % 3]; \
-	p.s6 += t[(s + 1) % 3]; \
-	p.s7 += s; \
+	p.s5 += t[s]; \
+	p.s6 += t[select(s + 1U, 0U, s == 2U)]; \
+	p.s7 += q; \
 } while(0)
 
-ulong SKEIN_ROT(const uint2 x, const uint y)
-{
-	if(y < 32) return(as_ulong(amd_bitalign(x, x.s10, 32 - y)));
-	else return(as_ulong(amd_bitalign(x.s10, x, 32 - (y - 32))));
-}
-
-void SkeinMix8(ulong4 *pv0, ulong4 *pv1, const uint rc0, const uint rc1, const uint rc2, const uint rc3)
+void SkeinMix8(ulong4 *pv0, ulong4 *pv1, const ulong4 rc)
 {
 	*pv0 += *pv1;
-	(*pv1).s0 = SKEIN_ROT(as_uint2((*pv1).s0), rc0);
-	(*pv1).s1 = SKEIN_ROT(as_uint2((*pv1).s1), rc1);
-	(*pv1).s2 = SKEIN_ROT(as_uint2((*pv1).s2), rc2);
-	(*pv1).s3 = SKEIN_ROT(as_uint2((*pv1).s3), rc3);
+	*pv1 = rotate(*pv1, (ulong4)rc);
 	*pv1 ^= *pv0;
 }
 
-ulong8 SkeinEvenRound(ulong8 p, const ulong8 h, const ulong *t, const uint s)
+ulong8 SkeinEvenRound(ulong8 p, const ulong8 h, const ulong *t, const uint s, const uint q)
 {
-	SKEIN_INJECT_KEY(p, s);
+	SKEIN_INJECT_KEY(p, s, q);
 	ulong4 pv0 = p.even, pv1 = p.odd;
 	
-	SkeinMix8(&pv0, &pv1, 46, 36, 19, 37);
+	SkeinMix8(&pv0, &pv1, (ulong4)(46, 36, 19, 37));
 	pv0 = shuffle(pv0, (ulong4)(1, 2, 3, 0));
 	pv1 = shuffle(pv1, (ulong4)(0, 3, 2, 1));
 	
-	SkeinMix8(&pv0, &pv1, 33, 27, 14, 42);
+	SkeinMix8(&pv0, &pv1, (ulong4)(33, 27, 14, 42));
 	pv0 = shuffle(pv0, (ulong4)(1, 2, 3, 0));
 	pv1 = shuffle(pv1, (ulong4)(0, 3, 2, 1));
 	
-	SkeinMix8(&pv0, &pv1, 17, 49, 36, 39);
+	SkeinMix8(&pv0, &pv1, (ulong4)(17, 49, 36, 39));
 	pv0 = shuffle(pv0, (ulong4)(1, 2, 3, 0));
 	pv1 = shuffle(pv1, (ulong4)(0, 3, 2, 1));
 	
-	SkeinMix8(&pv0, &pv1, 44, 9, 54, 56);
+	SkeinMix8(&pv0, &pv1, (ulong4)(44, 9, 54, 56));
 	return(shuffle2(pv0, pv1, (ulong8)(1, 4, 2, 7, 3, 6, 0, 5)));
 }
 
-ulong8 SkeinOddRound(ulong8 p, const ulong8 h, const ulong *t, const uint s)
+ulong8 SkeinOddRound(ulong8 p, const ulong8 h, const ulong *t, const uint s, const uint q)
 {
-	SKEIN_INJECT_KEY(p, s);
+	SKEIN_INJECT_KEY(p, s, q);
     ulong4 pv0 = p.even, pv1 = p.odd;
     
-	SkeinMix8(&pv0, &pv1, 39, 30, 34, 24);
+	SkeinMix8(&pv0, &pv1, (ulong4)(39, 30, 34, 24));
 	pv0 = shuffle(pv0, (ulong4)(1, 2, 3, 0));
 	pv1 = shuffle(pv1, (ulong4)(0, 3, 2, 1));
 	
-	SkeinMix8(&pv0, &pv1, 13, 50, 10, 17);
+	SkeinMix8(&pv0, &pv1, (ulong4)(13, 50, 10, 17));
 	pv0 = shuffle(pv0, (ulong4)(1, 2, 3, 0));
 	pv1 = shuffle(pv1, (ulong4)(0, 3, 2, 1));
 	
-	SkeinMix8(&pv0, &pv1, 25, 29, 39, 43);
+	SkeinMix8(&pv0, &pv1, (ulong4)(25, 29, 39, 43));
 	pv0 = shuffle(pv0, (ulong4)(1, 2, 3, 0));
 	pv1 = shuffle(pv1, (ulong4)(0, 3, 2, 1));
 	
-	SkeinMix8(&pv0, &pv1, 8, 35, 56, 22);
+	SkeinMix8(&pv0, &pv1, (ulong4)(8, 35, 56, 22));
 	return(shuffle2(pv0, pv1, (ulong8)(1, 4, 2, 7, 3, 6, 0, 5)));
 }
 
@@ -92,20 +83,47 @@ ulong8 Skein512Block(ulong8 p, ulong8 h, ulong h8, const ulong *t)
 	#pragma unroll
 	for(int i = 0; i < 18; ++i)
 	{
-		p = SkeinEvenRound(p, h, t, i);
+		p = SkeinEvenRound(p, h, t, 0U, i);
 		++i;
 		ulong tmp = h.s0;
 		h = shuffle(h, (ulong8)(1, 2, 3, 4, 5, 6, 7, 0));
 		h.s7 = h8;
 		h8 = tmp;
-		p = SkeinOddRound(p, h, t, i);
+		p = SkeinOddRound(p, h, t, 1U, i);
+		++i;
+		tmp = h.s0;
+		h = shuffle(h, (ulong8)(1, 2, 3, 4, 5, 6, 7, 0));
+		h.s7 = h8;
+		h8 = tmp;
+		p = SkeinEvenRound(p, h, t, 2U, i);
+		++i;
+		tmp = h.s0;
+		h = shuffle(h, (ulong8)(1, 2, 3, 4, 5, 6, 7, 0));
+		h.s7 = h8;
+		h8 = tmp;
+		p = SkeinOddRound(p, h, t, 0U, i);
+		++i;
+		tmp = h.s0;
+		h = shuffle(h, (ulong8)(1, 2, 3, 4, 5, 6, 7, 0));
+		h.s7 = h8;
+		h8 = tmp;
+		p = SkeinEvenRound(p, h, t, 1U, i);
+		++i;
+		tmp = h.s0;
+		h = shuffle(h, (ulong8)(1, 2, 3, 4, 5, 6, 7, 0));
+		h.s7 = h8;
+		h8 = tmp;
+		p = SkeinOddRound(p, h, t, 2U, i);
 		tmp = h.s0;
 		h = shuffle(h, (ulong8)(1, 2, 3, 4, 5, 6, 7, 0));
 		h.s7 = h8;
 		h8 = tmp;
 	}
 	
-	SKEIN_INJECT_KEY(p, 18);
+	p += h;
+	p.s5 += t[0];
+	p.s6 += t[1];
+	p.s7 += 18;
 	return(p);
 }