Fixed some speeds:

1800
3200
7100
7200
7400
8200
9300
9000
9600
11300
11600

OpenCL/m01800.cl

@@ -141,6 +141,7 @@ static void sha512_transform (const u64 w[16], u64 digest[8])
   digest[7] += h;
 }
 
+#ifdef IS_AMD
 static void sha512_transform_workaround (const u64 w[16], u64 digest[8])
 {
   u64 w0_t = w[ 0];
@@ -205,6 +206,7 @@ static void sha512_transform_workaround (const u64 w[16], u64 digest[8])
   digest[6] += g;
   digest[7] += h;
 }
+#endif
 
 static void sha512_init (sha512_ctx_t *sha512_ctx)
 {
@@ -243,7 +245,13 @@ static void sha512_update (sha512_ctx_t *sha512_ctx, const u64 *buf, int len)
     PUTCHAR64_BE (sha512_ctx->buf, pos++, GETCHAR64_BE (buf, i));
   }
 
+  #ifdef IS_AMD
   sha512_transform_workaround (sha512_ctx->buf, sha512_ctx->state);
+  #endif
+
+  #ifdef IS_NV
+  sha512_transform (sha512_ctx->buf, sha512_ctx->state);
+  #endif
 
   len -= cnt;
 

OpenCL/m09000.cl

@@ -303,6 +303,7 @@ __constant u32 c_pbox[18] =
   0x9216d5d9, 0x8979fb1b
 };
 
+/*
 #define BF_ROUND(L,R,N)       \
 {                             \
   uchar4 c = as_uchar4 ((L)); \
@@ -316,6 +317,20 @@ __constant u32 c_pbox[18] =
                               \
   (R) ^= tmp ^ P[(N)];        \
 }
+*/
+
+
+#define BF_ROUND(L,R,N)       \
+{                             \
+  u32 tmp;                  \
+                              \
+  tmp  = S0[((L) >> 24) & 0xff];            \
+  tmp += S1[((L) >> 16) & 0xff];            \
+  tmp ^= S2[((L) >>  8) & 0xff];            \
+  tmp += S3[((L) >>  0) & 0xff];            \
+                              \
+  (R) ^= tmp ^ P[(N)];        \
+}
 
 #define BF_ENCRYPT(L,R) \
 {                       \

OpenCL/m09600.cl

@@ -1001,7 +1001,7 @@ __constant u64 k_sha512[80] =
   SHA512C4c, SHA512C4d, SHA512C4e, SHA512C4f,
 };
 
-static void sha512_transform (volatile const u64 w0[4], volatile const u64 w1[4], volatile const u64 w2[4], volatile const u64 w3[4], volatile u64 dgst[8])
+static void sha512_transform (const u64 w0[4], const u64 w1[4], const u64 w2[4], const u64 w3[4], u64 dgst[8])
 {
   u64 a = dgst[0];
   u64 b = dgst[1];
@@ -1071,7 +1071,7 @@ static void sha512_transform (volatile const u64 w0[4], volatile const u64 w1[4]
 
   ROUND_STEP (0);
 
-  #pragma unroll
+  //#pragma unroll
   for (int i = 16; i < 80; i += 16)
   {
     ROUND_EXPAND (); ROUND_STEP (i);

OpenCL/m11300.cl

@@ -925,22 +925,22 @@ static void sha512_transform (const u64 w[16], u64 dgst[8])
   u64 g = dgst[6];
   u64 h = dgst[7];
 
-  volatile u64 w0_t = w[ 0];
-  volatile u64 w1_t = w[ 1];
-  volatile u64 w2_t = w[ 2];
-  volatile u64 w3_t = w[ 3];
-  volatile u64 w4_t = w[ 4];
-  volatile u64 w5_t = w[ 5];
-  volatile u64 w6_t = w[ 6];
-  volatile u64 w7_t = w[ 7];
-  volatile u64 w8_t = w[ 8];
-  volatile u64 w9_t = w[ 9];
-  volatile u64 wa_t = w[10];
-  volatile u64 wb_t = w[11];
-  volatile u64 wc_t = w[12];
-  volatile u64 wd_t = w[13];
-  volatile u64 we_t = w[14];
-  volatile u64 wf_t = w[15];
+  u64 w0_t = w[ 0];
+  u64 w1_t = w[ 1];
+  u64 w2_t = w[ 2];
+  u64 w3_t = w[ 3];
+  u64 w4_t = w[ 4];
+  u64 w5_t = w[ 5];
+  u64 w6_t = w[ 6];
+  u64 w7_t = w[ 7];
+  u64 w8_t = w[ 8];
+  u64 w9_t = w[ 9];
+  u64 wa_t = w[10];
+  u64 wb_t = w[11];
+  u64 wc_t = w[12];
+  u64 wd_t = w[13];
+  u64 we_t = w[14];
+  u64 wf_t = w[15];
 
   #define ROUND_EXPAND()                            \
   {                                                 \
@@ -984,7 +984,7 @@ static void sha512_transform (const u64 w[16], u64 dgst[8])
 
   ROUND_STEP (0);
 
-  #pragma unroll
+//  #pragma unroll
   for (int i = 16; i < 80; i += 16)
   {
     ROUND_EXPAND (); ROUND_STEP (i);

OpenCL/types_ocl.c

@@ -181,78 +181,26 @@ static u64 rotl64 (const u64 a, const u32 n)
 
 #ifdef IS_NV
 
-#if CUDA_ARCH >= 350
-
 static u32 rotr32 (const u32 a, const u32 n)
 {
-  u32 r;
-
-  asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(a), "r"(n));
-
-  return r;
+  return rotate (a, 32 - n);
 }
 
 static u32 rotl32 (const u32 a, const u32 n)
 {
-  return rotr32 (a, 32 - n);
+  return rotate (a, n);
 }
 
-static u64 rotr64 (const u64 a, const u32 n)
+static u64 rotr64 (const u64 a, const u64 n)
 {
-  u32 il;
-  u32 ir;
-
-  asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a));
-
-  u32 tl;
-  u32 tr;
-
-  if (n >= 32)
-  {
-    asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(tl) : "r"(ir), "r"(il), "r"(n - 32));
-    asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(tr) : "r"(il), "r"(ir), "r"(n - 32));
-  }
-  else
-  {
-    asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(tl) : "r"(il), "r"(ir), "r"(n));
-    asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(tr) : "r"(ir), "r"(il), "r"(n));
-  }
-
-  u64 r;
-
-  asm ("mov.b64 %0, {%1, %2};" : "=l"(r) : "r"(tl), "r"(tr));
-
-  return r;
+  return rotate (a, 64 - n);
 }
 
-static u64 rotl64 (const u64 a, const u32 n)
+static u64 rotl64 (const u64 a, const u64 n)
 {
-  return rotr64 (a, 64 - n);
+  return rotate (a, n);
 }
 
-#else
-
-static u32 rotr32 (const u32 a, const u32 n)
-{
-  return (((a) >> (n)) | ((a) << (32 - (n))));
-}
-
-static u32 rotl32 (const u32 a, const u32 n)
-{
-  return rotr32 (a, 32 - n);
-}
-
-static u64 rotr64 (const u64 a, const u32 n)
-{
-  return (((a) >> (n)) | ((a) << (64 - (n))));
-}
-
-static u64 rotl64 (const u64 a, const u32 n)
-{
-  return rotr64 (a, 64 - n);
-}
-
-#endif
 #endif
 
 typedef struct