Slightly increase blowfish based algorithms for NV

OpenCL/m03200.cl

@@ -294,11 +294,12 @@ __constant u32 c_sbox3[256] =
   0xb74e6132, 0xce77e25b, 0x578fdfe3, 0x3ac372e6
 };
 
+#ifdef IS_AMD
 #define BF_ROUND(L,R,N)       \
 {                             \
   uchar4 c = as_uchar4 ((L)); \
                               \
-  u32 tmp;                  \
+  u32 tmp;                    \
                               \
   tmp  = S0[c.s3];            \
   tmp += S1[c.s2];            \
@@ -307,6 +308,21 @@ __constant u32 c_sbox3[256] =
                               \
   (R) ^= tmp ^ P[(N)];        \
 }
+#endif
+
+#ifdef IS_NV
+#define BF_ROUND(L,R,N)           \
+{                                 \
+  u32 tmp;                        \
+                                  \
+  tmp  = S0[__bfe ((L), 24, 8)];  \
+  tmp += S1[__bfe ((L), 16, 8)];  \
+  tmp ^= S2[__bfe ((L),  8, 8)];  \
+  tmp += S3[__bfe ((L),  0, 8)];  \
+                                  \
+  (R) ^= tmp ^ P[(N)];            \
+}
+#endif
 
 #define BF_ENCRYPT(L,R) \
 {                       \

OpenCL/m09000.cl

@@ -294,7 +294,6 @@ __constant u32 c_sbox3[256] =
   0xb74e6132, 0xce77e25b, 0x578fdfe3, 0x3ac372e6
 };
 
-
 #ifdef IS_AMD
 #define BF_ROUND(L,R,N)       \
 {                             \
@@ -316,10 +315,10 @@ __constant u32 c_sbox3[256] =
 {                                 \
   u32 tmp;                        \
                                   \
-  tmp  = S0[((L) >> 24) & 0xff];  \
-  tmp += S1[((L) >> 16) & 0xff];  \
-  tmp ^= S2[((L) >>  8) & 0xff];  \
-  tmp += S3[((L) >>  0) & 0xff];  \
+  tmp  = S0[__bfe ((L), 24, 8)];  \
+  tmp += S1[__bfe ((L), 16, 8)];  \
+  tmp ^= S2[__bfe ((L),  8, 8)];  \
+  tmp += S3[__bfe ((L),  0, 8)];  \
                                   \
   (R) ^= tmp ^ P[(N)];            \
 }

OpenCL/types_ocl.c

@@ -22,15 +22,25 @@ static inline u64 swap64 (const u64 v)
 #endif
 
 #ifdef IS_NV
-static inline u32 __byte_perm (const u32 a, const u32 b, const u32 s)
+static inline u32 __byte_perm (const u32 a, const u32 b, const u32 c)
 {
   u32 r;
 
-  asm ("prmt.b32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(s));
+  asm ("prmt.b32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(c));
 
   return r;
 }
 
+static inline u32 __bfe (const u32 a, const u32 b, const u32 c)
+{
+  u32 r;
+
+  asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(c));
+
+  return r;
+}
+
+
 #if CUDA_ARCH >= 350
 
 static inline u32 amd_bytealign (const u32 a, const u32 b, const u32 c)