KXX_DECL u32 k36, k37, k38, k39, k40, k41;
KXX_DECL u32 k42, k43, k44, k45, k46, k47;
- #ifdef IS_NV
- #if CUDA_ARCH >= 500
- #else
+ #ifdef _unroll
#pragma unroll
#endif
- #endif
-
- #ifdef IS_AMD
- #pragma unroll
- #endif
-
- #ifdef IS_GENERIC
- #pragma unroll 1
- #endif
-
for (u32 i = 0; i < 2; i++)
{
if (i) KEYSET10 else KEYSET00
u32 tmpResult = 0;
+ #ifdef _unroll
#pragma unroll
+ #endif
for (int i = 0; i < 32; i++)
{
const u32 b0 = -((search[0] >> i) & 1);
u32 out0[32];
u32 out1[32];
+ #ifdef _unroll
#pragma unroll
+ #endif
for (int i = 0; i < 32; i++)
{
out0[i] = out[ 0 + 31 - i];
transpose32c (out0);
transpose32c (out1);
+ #ifdef _unroll
#pragma unroll
+ #endif
for (int slice = 0; slice < 32; slice++)
{
const u32 r0 = out0[31 - slice];