* License.....: MIT
*/
-#define DEVICE_TYPE_CPU 2
-#define DEVICE_TYPE_GPU 4
-
typedef uchar u8;
typedef ushort u16;
typedef uint u32;
#define VECT_SIZE 1
#endif
-#define CONCAT(a, b) a##b
+#define CONCAT(a, b) a##b
#define VTYPE(type, width) CONCAT(type, width)
#if VECT_SIZE == 1
typedef VTYPE(ulong, VECT_SIZE) u64x;
#endif
-// this one needs to die
-#define allx(r) r
-
-static inline u32 l32_from_64_S (u64 a)
+inline u32 l32_from_64_S (u64 a)
{
const u32 r = (u32) (a);
return r;
}
-static inline u32 h32_from_64_S (u64 a)
+inline u32 h32_from_64_S (u64 a)
{
a >>= 32;
return r;
}
-static inline u64 hl32_to_64_S (const u32 a, const u32 b)
+inline u64 hl32_to_64_S (const u32 a, const u32 b)
{
return as_ulong ((uint2) (b, a));
}
-static inline u32x l32_from_64 (u64x a)
+inline u32x l32_from_64 (u64x a)
{
u32x r;
return r;
}
-static inline u32x h32_from_64 (u64x a)
+inline u32x h32_from_64 (u64x a)
{
a >>= 32;
return r;
}
-static inline u64x hl32_to_64 (const u32x a, const u32x b)
+inline u64x hl32_to_64 (const u32x a, const u32x b)
{
u64x r;
}
#ifdef IS_AMD
-static inline u32 swap32_S (const u32 v)
+inline u32 swap32_S (const u32 v)
{
return (as_uint (as_uchar4 (v).s3210));
}
-static inline u64 swap64_S (const u64 v)
+inline u64 swap64_S (const u64 v)
{
return (as_ulong (as_uchar8 (v).s76543210));
}
-static inline u32 rotr32_S (const u32 a, const u32 n)
+inline u32 rotr32_S (const u32 a, const u32 n)
{
return rotate (a, 32 - n);
}
-static inline u32 rotl32_S (const u32 a, const u32 n)
+inline u32 rotl32_S (const u32 a, const u32 n)
{
return rotate (a, n);
}
-static inline u64 rotr64_S (const u64 a, const u32 n)
+inline u64 rotr64_S (const u64 a, const u32 n)
{
- #if (DEVICE_TYPE == DEVICE_TYPE_GPU)
-
- #ifdef amd_bitalign
-
const u32 a0 = h32_from_64_S (a);
const u32 a1 = l32_from_64_S (a);
const u64 r = hl32_to_64_S (t0, t1);
- #else
-
- const u64 r = rotate (a, (u64) 64 - n);
-
- #endif
-
- #else
-
- const u64 r = rotate (a, (u64) 64 - n);
-
- #endif
-
return r;
}
-static inline u64 rotl64_S (const u64 a, const u32 n)
+inline u64 rotl64_S (const u64 a, const u32 n)
{
return rotr64_S (a, 64 - n);
}
-static inline u32x swap32 (const u32x v)
+inline u32x swap32 (const u32x v)
{
return ((v >> 24) & 0x000000ff)
| ((v >> 8) & 0x0000ff00)
| ((v << 24) & 0xff000000);
}
-static inline u64x swap64 (const u64x v)
+inline u64x swap64 (const u64x v)
{
return ((v >> 56) & 0x00000000000000ff)
| ((v >> 40) & 0x000000000000ff00)
| ((v << 56) & 0xff00000000000000);
}
-static inline u32x rotr32 (const u32x a, const u32 n)
+inline u32x rotr32 (const u32x a, const u32 n)
{
return rotate (a, 32 - n);
}
-static inline u32x rotl32 (const u32x a, const u32 n)
+inline u32x rotl32 (const u32x a, const u32 n)
{
return rotate (a, n);
}
-static inline u64x rotr64 (const u64x a, const u32 n)
+inline u64x rotr64 (const u64x a, const u32 n)
{
- #if (DEVICE_TYPE == DEVICE_TYPE_GPU)
-
- #ifdef amd_bitalign
const u32x a0 = h32_from_64 (a);
const u32x a1 = l32_from_64 (a);
const u64x r = hl32_to_64 (t0, t1);
- #else
-
- const u64x r = rotate (a, (u64) 64 - n);
-
- #endif
-
- #else
-
- const u64x r = rotate (a, (u64) 64 - n);
-
- #endif
-
return r;
}
-static inline u64x rotl64 (const u64x a, const u32 n)
+inline u64x rotl64 (const u64x a, const u32 n)
{
return rotr64 (a, 64 - n);
}
-static inline u32 __bfe (const u32 a, const u32 b, const u32 c)
+inline u32x __bfe (const u32x a, const u32x b, const u32x c)
{
-#ifdef amd_bfe
return amd_bfe (a, b, c);
-#else
- #define BIT(x) (1 << (x))
- #define BIT_MASK(x) (BIT (x) - 1)
- #define BFE(x,y,z) (((x) >> (y)) & BIT_MASK (z))
+}
- return BFE (a, b, c);
-#endif
+inline u32 __bfe_S (const u32 a, const u32 b, const u32 c)
+{
+ return amd_bfe (a, b, c);
}
-static inline u32 amd_bytealign_S (const u32 a, const u32 b, const u32 c)
+inline u32 amd_bytealign_S (const u32 a, const u32 b, const u32 c)
{
-#ifdef amd_bytealign
return amd_bytealign (a, b, c);
-#else
- const u64 tmp = ((((u64) a) << 32) | ((u64) b)) >> ((c & 3) * 8);
-
- return (u32) (tmp);
-#endif
}
#endif
#ifdef IS_NV
-static inline u32 swap32_S (const u32 v)
+inline u32 swap32_S (const u32 v)
{
u32 r;
return r;
}
-static inline u64 swap64_S (const u64 v)
+inline u64 swap64_S (const u64 v)
{
u32 il;
u32 ir;
return r;
}
-static inline u32 rotr32_S (const u32 a, const u32 n)
+inline u32 rotr32_S (const u32 a, const u32 n)
{
return rotate (a, 32 - n);
}
-static inline u32 rotl32_S (const u32 a, const u32 n)
+inline u32 rotl32_S (const u32 a, const u32 n)
{
return rotate (a, n);
}
-#if CUDA_ARCH >= 350
-static inline u64 rotr64_S (const u64 a, const u32 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;
-}
-#else
-static inline u64 rotr64_S (const u64 a, const u32 n)
+inline u64 rotr64_S (const u64 a, const u32 n)
{
return rotate (a, (u64) 64 - n);
}
-#endif
-static inline u64 rotl64_S (const u64 a, const u32 n)
+inline u64 rotl64_S (const u64 a, const u32 n)
{
return rotr64_S (a, 64 - n);
}
-#if CUDA_ARCH >= 500
-static inline u32 lut3_2d_S (const u32 a, const u32 b, const u32 c)
-{
- u32 r;
-
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
-
- return r;
-}
-
-static inline u32 lut3_39_S (const u32 a, const u32 b, const u32 c)
-{
- u32 r;
-
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
-
- return r;
-}
-
-static inline u32 lut3_59_S (const u32 a, const u32 b, const u32 c)
-{
- u32 r;
-
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
-
- return r;
-}
-
-static inline u32 lut3_96_S (const u32 a, const u32 b, const u32 c)
-{
- u32 r;
-
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
-
- return r;
-}
-
-static inline u32 lut3_e4_S (const u32 a, const u32 b, const u32 c)
-{
- u32 r;
-
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
-
- return r;
-}
-
-static inline u32 lut3_e8_S (const u32 a, const u32 b, const u32 c)
-{
- u32 r;
-
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
-
- return r;
-}
-
-static inline u32 lut3_ca_S (const u32 a, const u32 b, const u32 c)
-{
- u32 r;
-
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
-
- return r;
-}
-#endif
-
-static inline u32 __byte_perm_S (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"(c));
-
- return r;
-}
-
-static inline u32x swap32 (const u32x v)
+inline u32x swap32 (const u32x v)
{
return ((v >> 24) & 0x000000ff)
| ((v >> 8) & 0x0000ff00)
| ((v << 24) & 0xff000000);
}
-static inline u64x swap64 (const u64x v)
+inline u64x swap64 (const u64x v)
{
return ((v >> 56) & 0x00000000000000ff)
| ((v >> 40) & 0x000000000000ff00)
| ((v << 56) & 0xff00000000000000);
}
-static inline u32x rotr32 (const u32x a, const u32 n)
+inline u32x rotr32 (const u32x a, const u32 n)
{
return rotate (a, 32 - n);
}
-static inline u32x rotl32 (const u32x a, const u32 n)
+inline u32x rotl32 (const u32x a, const u32 n)
{
return rotate (a, n);
}
-#if CUDA_ARCH >= 350
-static inline u64x rotr64 (const u64x a, const u32 n)
-{
- u64x r;
-
- u32 il;
- u32 ir;
- u32 tl;
- u32 tr;
-
- #if VECT_SIZE == 1
-
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r) : "r"(tl), "r"(tr));
-
- #endif
-
- #if VECT_SIZE >= 2
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.s0));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.s0) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.s1));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.s1) : "r"(tl), "r"(tr));
- }
-
- #endif
-
- #if VECT_SIZE >= 4
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.s2));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.s2) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.s3));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.s3) : "r"(tl), "r"(tr));
- }
-
- #endif
-
- #if VECT_SIZE >= 8
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.s4));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.s4) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.s5));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.s5) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.s6));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.s6) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.s7));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.s7) : "r"(tl), "r"(tr));
- }
-
- #endif
-
- #if VECT_SIZE >= 16
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.s8));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.s8) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.s9));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.s9) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.sa));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.sa) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.sb));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.sb) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.sc));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.sc) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.sd));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.sd) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.se));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.se) : "r"(tl), "r"(tr));
- }
-
- {
- asm ("mov.b64 {%0, %1}, %2;" : "=r"(il), "=r"(ir) : "l"(a.sf));
-
- 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));
- }
-
- asm ("mov.b64 %0, {%1, %2};" : "=l"(r.sf) : "r"(tl), "r"(tr));
- }
-
- #endif
-
- return r;
-}
-#else
-static inline u64x rotr64 (const u64x a, const u32 n)
+inline u64x rotr64 (const u64x a, const u32 n)
{
return rotate (a, (u64) 64 - n);
}
-#endif
-static inline u64x rotl64 (const u64x a, const u32 n)
+inline u64x rotl64 (const u64x a, const u32 n)
{
- return rotr64 (a, (u64) 64 - n);
+ return rotate (a, (u64) n);
}
-static inline u32x __byte_perm (const u32x a, const u32x b, const u32x c)
+inline u32x __byte_perm (const u32x a, const u32x b, const u32x c)
{
u32x r;
return r;
}
-static inline u32 __bfe (const u32 a, const u32 b, const u32 c)
+inline u32 __byte_perm_S (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)
-{
- u32 r;
-
- asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r) : "r"(b), "r"(a), "r"((c & 3) * 8));
+ asm ("prmt.b32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(c));
return r;
}
-#else
-static inline u32 amd_bytealign (const u32 a, const u32 b, const u32 c)
-{
- return __byte_perm_S (b, a, (0x76543210 >> ((c & 3) * 4)) & 0xffff);
-}
-#endif
-#if CUDA_ARCH >= 500
-static inline u32x lut3_2d (const u32x a, const u32x b, const u32x c)
+inline u32x __bfe (const u32x a, const u32x b, const u32x c)
{
u32x r;
#if VECT_SIZE == 1
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(c));
#endif
#if VECT_SIZE >= 2
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.s0) : "r" (a.s0), "r" (b.s0), "r" (c.s0));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.s1) : "r" (a.s1), "r" (b.s1), "r" (c.s1));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.s0) : "r"(a.s0), "r"(b.s0), "r"(c.s0));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.s1) : "r"(a.s1), "r"(b.s1), "r"(c.s1));
#endif
#if VECT_SIZE >= 4
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.s2) : "r" (a.s2), "r" (b.s2), "r" (c.s2));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.s3) : "r" (a.s3), "r" (b.s3), "r" (c.s3));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.s2) : "r"(a.s2), "r"(b.s2), "r"(c.s2));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.s3) : "r"(a.s3), "r"(b.s3), "r"(c.s3));
#endif
#if VECT_SIZE >= 8
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.s4) : "r" (a.s4), "r" (b.s4), "r" (c.s4));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.s5) : "r" (a.s5), "r" (b.s5), "r" (c.s5));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.s6) : "r" (a.s6), "r" (b.s6), "r" (c.s6));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.s7) : "r" (a.s7), "r" (b.s7), "r" (c.s7));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.s4) : "r"(a.s4), "r"(b.s4), "r"(c.s4));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.s5) : "r"(a.s5), "r"(b.s5), "r"(c.s5));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.s6) : "r"(a.s6), "r"(b.s6), "r"(c.s6));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.s7) : "r"(a.s7), "r"(b.s7), "r"(c.s7));
#endif
#if VECT_SIZE >= 16
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.s8) : "r" (a.s8), "r" (b.s8), "r" (c.s8));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.s9) : "r" (a.s9), "r" (b.s9), "r" (c.s9));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.sa) : "r" (a.sa), "r" (b.sa), "r" (c.sa));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.sb) : "r" (a.sb), "r" (b.sb), "r" (c.sb));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.sc) : "r" (a.sc), "r" (b.sc), "r" (c.sc));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.sd) : "r" (a.sd), "r" (b.sd), "r" (c.sd));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.se) : "r" (a.se), "r" (b.se), "r" (c.se));
- asm ("lop3.b32 %0, %1, %2, %3, 0x2d;" : "=r" (r.sf) : "r" (a.sf), "r" (b.sf), "r" (c.sf));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.s8) : "r"(a.s8), "r"(b.s8), "r"(c.s8));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.s9) : "r"(a.s9), "r"(b.s9), "r"(c.s9));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.sa) : "r"(a.sa), "r"(b.sa), "r"(c.sa));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.sb) : "r"(a.sb), "r"(b.sb), "r"(c.sb));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.sc) : "r"(a.sc), "r"(b.sc), "r"(c.sc));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.sd) : "r"(a.sd), "r"(b.sd), "r"(c.sd));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.se) : "r"(a.se), "r"(b.se), "r"(c.se));
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r.sf) : "r"(a.sf), "r"(b.sf), "r"(c.sf));
#endif
return r;
}
-static inline u32x lut3_39 (const u32x a, const u32x b, const u32x c)
+inline u32 __bfe_S (const u32 a, const u32 b, const u32 c)
{
- u32x r;
-
- #if VECT_SIZE == 1
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
- #endif
-
- #if VECT_SIZE >= 2
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.s0) : "r" (a.s0), "r" (b.s0), "r" (c.s0));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.s1) : "r" (a.s1), "r" (b.s1), "r" (c.s1));
- #endif
-
- #if VECT_SIZE >= 4
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.s2) : "r" (a.s2), "r" (b.s2), "r" (c.s2));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.s3) : "r" (a.s3), "r" (b.s3), "r" (c.s3));
- #endif
-
- #if VECT_SIZE >= 8
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.s4) : "r" (a.s4), "r" (b.s4), "r" (c.s4));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.s5) : "r" (a.s5), "r" (b.s5), "r" (c.s5));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.s6) : "r" (a.s6), "r" (b.s6), "r" (c.s6));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.s7) : "r" (a.s7), "r" (b.s7), "r" (c.s7));
- #endif
+ u32 r;
- #if VECT_SIZE >= 16
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.s8) : "r" (a.s8), "r" (b.s8), "r" (c.s8));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.s9) : "r" (a.s9), "r" (b.s9), "r" (c.s9));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.sa) : "r" (a.sa), "r" (b.sa), "r" (c.sa));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.sb) : "r" (a.sb), "r" (b.sb), "r" (c.sb));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.sc) : "r" (a.sc), "r" (b.sc), "r" (c.sc));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.sd) : "r" (a.sd), "r" (b.sd), "r" (c.sd));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.se) : "r" (a.se), "r" (b.se), "r" (c.se));
- asm ("lop3.b32 %0, %1, %2, %3, 0x39;" : "=r" (r.sf) : "r" (a.sf), "r" (b.sf), "r" (c.sf));
- #endif
+ asm ("bfe.u32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(c));
return r;
}
-static inline u32x lut3_59 (const u32x a, const u32x b, const u32x c)
+inline u32x amd_bytealign (const u32x a, const u32x b, const u32x c)
{
u32x r;
- #if VECT_SIZE == 1
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
- #endif
-
- #if VECT_SIZE >= 2
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.s0) : "r" (a.s0), "r" (b.s0), "r" (c.s0));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.s1) : "r" (a.s1), "r" (b.s1), "r" (c.s1));
- #endif
-
- #if VECT_SIZE >= 4
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.s2) : "r" (a.s2), "r" (b.s2), "r" (c.s2));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.s3) : "r" (a.s3), "r" (b.s3), "r" (c.s3));
- #endif
-
- #if VECT_SIZE >= 8
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.s4) : "r" (a.s4), "r" (b.s4), "r" (c.s4));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.s5) : "r" (a.s5), "r" (b.s5), "r" (c.s5));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.s6) : "r" (a.s6), "r" (b.s6), "r" (c.s6));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.s7) : "r" (a.s7), "r" (b.s7), "r" (c.s7));
- #endif
-
- #if VECT_SIZE >= 16
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.s8) : "r" (a.s8), "r" (b.s8), "r" (c.s8));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.s9) : "r" (a.s9), "r" (b.s9), "r" (c.s9));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.sa) : "r" (a.sa), "r" (b.sa), "r" (c.sa));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.sb) : "r" (a.sb), "r" (b.sb), "r" (c.sb));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.sc) : "r" (a.sc), "r" (b.sc), "r" (c.sc));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.sd) : "r" (a.sd), "r" (b.sd), "r" (c.sd));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.se) : "r" (a.se), "r" (b.se), "r" (c.se));
- asm ("lop3.b32 %0, %1, %2, %3, 0x59;" : "=r" (r.sf) : "r" (a.sf), "r" (b.sf), "r" (c.sf));
- #endif
-
- return r;
-}
-
-static inline u32x lut3_96 (const u32x a, const u32x b, const u32x c)
-{
- u32x r;
+ #if CUDA_ARCH >= 350
#if VECT_SIZE == 1
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r) : "r"(b), "r"(a), "r"((c & 3) * 8));
#endif
#if VECT_SIZE >= 2
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.s0) : "r" (a.s0), "r" (b.s0), "r" (c.s0));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.s1) : "r" (a.s1), "r" (b.s1), "r" (c.s1));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.s0) : "r"(b.s0), "r"(a.s0), "r"((c.s0 & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.s1) : "r"(b.s1), "r"(a.s1), "r"((c.s1 & 3) * 8));
#endif
#if VECT_SIZE >= 4
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.s2) : "r" (a.s2), "r" (b.s2), "r" (c.s2));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.s3) : "r" (a.s3), "r" (b.s3), "r" (c.s3));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.s2) : "r"(b.s2), "r"(a.s2), "r"((c.s2 & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.s3) : "r"(b.s3), "r"(a.s3), "r"((c.s3 & 3) * 8));
#endif
#if VECT_SIZE >= 8
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.s4) : "r" (a.s4), "r" (b.s4), "r" (c.s4));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.s5) : "r" (a.s5), "r" (b.s5), "r" (c.s5));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.s6) : "r" (a.s6), "r" (b.s6), "r" (c.s6));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.s7) : "r" (a.s7), "r" (b.s7), "r" (c.s7));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.s4) : "r"(b.s4), "r"(a.s4), "r"((c.s4 & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.s5) : "r"(b.s5), "r"(a.s5), "r"((c.s5 & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.s6) : "r"(b.s6), "r"(a.s6), "r"((c.s6 & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.s7) : "r"(b.s7), "r"(a.s7), "r"((c.s7 & 3) * 8));
#endif
#if VECT_SIZE >= 16
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.s8) : "r" (a.s8), "r" (b.s8), "r" (c.s8));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.s9) : "r" (a.s9), "r" (b.s9), "r" (c.s9));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.sa) : "r" (a.sa), "r" (b.sa), "r" (c.sa));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.sb) : "r" (a.sb), "r" (b.sb), "r" (c.sb));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.sc) : "r" (a.sc), "r" (b.sc), "r" (c.sc));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.sd) : "r" (a.sd), "r" (b.sd), "r" (c.sd));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.se) : "r" (a.se), "r" (b.se), "r" (c.se));
- asm ("lop3.b32 %0, %1, %2, %3, 0x96;" : "=r" (r.sf) : "r" (a.sf), "r" (b.sf), "r" (c.sf));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.s8) : "r"(b.s8), "r"(a.s8), "r"((c.s8 & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.s9) : "r"(b.s9), "r"(a.s9), "r"((c.s9 & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.sa) : "r"(b.sa), "r"(a.sa), "r"((c.sa & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.sb) : "r"(b.sb), "r"(a.sb), "r"((c.sb & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.sc) : "r"(b.sc), "r"(a.sc), "r"((c.sc & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.sd) : "r"(b.sd), "r"(a.sd), "r"((c.sd & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.se) : "r"(b.se), "r"(a.se), "r"((c.se & 3) * 8));
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r.sf) : "r"(b.sf), "r"(a.sf), "r"((c.sf & 3) * 8));
#endif
- return r;
-}
-
-static inline u32x lut3_e4 (const u32x a, const u32x b, const u32x c)
-{
- u32x r;
-
- #if VECT_SIZE == 1
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
- #endif
-
- #if VECT_SIZE >= 2
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.s0) : "r" (a.s0), "r" (b.s0), "r" (c.s0));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.s1) : "r" (a.s1), "r" (b.s1), "r" (c.s1));
- #endif
+ #else
- #if VECT_SIZE >= 4
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.s2) : "r" (a.s2), "r" (b.s2), "r" (c.s2));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.s3) : "r" (a.s3), "r" (b.s3), "r" (c.s3));
- #endif
+ r = __byte_perm (b, a, ((u32x) (0x76543210) >> ((c & 3) * 4)) & 0xffff);
- #if VECT_SIZE >= 8
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.s4) : "r" (a.s4), "r" (b.s4), "r" (c.s4));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.s5) : "r" (a.s5), "r" (b.s5), "r" (c.s5));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.s6) : "r" (a.s6), "r" (b.s6), "r" (c.s6));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.s7) : "r" (a.s7), "r" (b.s7), "r" (c.s7));
- #endif
-
- #if VECT_SIZE >= 16
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.s8) : "r" (a.s8), "r" (b.s8), "r" (c.s8));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.s9) : "r" (a.s9), "r" (b.s9), "r" (c.s9));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.sa) : "r" (a.sa), "r" (b.sa), "r" (c.sa));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.sb) : "r" (a.sb), "r" (b.sb), "r" (c.sb));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.sc) : "r" (a.sc), "r" (b.sc), "r" (c.sc));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.sd) : "r" (a.sd), "r" (b.sd), "r" (c.sd));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.se) : "r" (a.se), "r" (b.se), "r" (c.se));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe4;" : "=r" (r.sf) : "r" (a.sf), "r" (b.sf), "r" (c.sf));
#endif
return r;
}
-static inline u32x lut3_e8 (const u32x a, const u32x b, const u32x c)
+inline u32 amd_bytealign_S (const u32 a, const u32 b, const u32 c)
{
- u32x r;
+ u32 r;
- #if VECT_SIZE == 1
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
- #endif
+ #if CUDA_ARCH >= 350
- #if VECT_SIZE >= 2
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.s0) : "r" (a.s0), "r" (b.s0), "r" (c.s0));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.s1) : "r" (a.s1), "r" (b.s1), "r" (c.s1));
- #endif
+ asm ("shf.r.wrap.b32 %0, %1, %2, %3;" : "=r"(r) : "r"(b), "r"(a), "r"((c & 3) * 8));
- #if VECT_SIZE >= 4
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.s2) : "r" (a.s2), "r" (b.s2), "r" (c.s2));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.s3) : "r" (a.s3), "r" (b.s3), "r" (c.s3));
- #endif
+ #else
- #if VECT_SIZE >= 8
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.s4) : "r" (a.s4), "r" (b.s4), "r" (c.s4));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.s5) : "r" (a.s5), "r" (b.s5), "r" (c.s5));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.s6) : "r" (a.s6), "r" (b.s6), "r" (c.s6));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.s7) : "r" (a.s7), "r" (b.s7), "r" (c.s7));
- #endif
+ r = __byte_perm_S (b, a, (0x76543210 >> ((c & 3) * 4)) & 0xffff);
- #if VECT_SIZE >= 16
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.s8) : "r" (a.s8), "r" (b.s8), "r" (c.s8));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.s9) : "r" (a.s9), "r" (b.s9), "r" (c.s9));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.sa) : "r" (a.sa), "r" (b.sa), "r" (c.sa));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.sb) : "r" (a.sb), "r" (b.sb), "r" (c.sb));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.sc) : "r" (a.sc), "r" (b.sc), "r" (c.sc));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.sd) : "r" (a.sd), "r" (b.sd), "r" (c.sd));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.se) : "r" (a.se), "r" (b.se), "r" (c.se));
- asm ("lop3.b32 %0, %1, %2, %3, 0xe8;" : "=r" (r.sf) : "r" (a.sf), "r" (b.sf), "r" (c.sf));
#endif
return r;
}
-
-static inline u32x lut3_ca (const u32x a, const u32x b, const u32x c)
-{
- u32x r;
-
- #if VECT_SIZE == 1
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r) : "r" (a), "r" (b), "r" (c));
- #endif
-
- #if VECT_SIZE >= 2
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.s0) : "r" (a.s0), "r" (b.s0), "r" (c.s0));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.s1) : "r" (a.s1), "r" (b.s1), "r" (c.s1));
- #endif
-
- #if VECT_SIZE >= 4
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.s2) : "r" (a.s2), "r" (b.s2), "r" (c.s2));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.s3) : "r" (a.s3), "r" (b.s3), "r" (c.s3));
- #endif
-
- #if VECT_SIZE >= 8
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.s4) : "r" (a.s4), "r" (b.s4), "r" (c.s4));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.s5) : "r" (a.s5), "r" (b.s5), "r" (c.s5));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.s6) : "r" (a.s6), "r" (b.s6), "r" (c.s6));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.s7) : "r" (a.s7), "r" (b.s7), "r" (c.s7));
- #endif
-
- #if VECT_SIZE >= 16
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.s8) : "r" (a.s8), "r" (b.s8), "r" (c.s8));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.s9) : "r" (a.s9), "r" (b.s9), "r" (c.s9));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.sa) : "r" (a.sa), "r" (b.sa), "r" (c.sa));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.sb) : "r" (a.sb), "r" (b.sb), "r" (c.sb));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.sc) : "r" (a.sc), "r" (b.sc), "r" (c.sc));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.sd) : "r" (a.sd), "r" (b.sd), "r" (c.sd));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.se) : "r" (a.se), "r" (b.se), "r" (c.se));
- asm ("lop3.b32 %0, %1, %2, %3, 0xca;" : "=r" (r.sf) : "r" (a.sf), "r" (b.sf), "r" (c.sf));
- #endif
-
- return r;
-}
-
-#endif
#endif
#ifdef IS_GENERIC
-static inline u32 swap32_S (const u32 v)
+inline u32 swap32_S (const u32 v)
{
return (as_uint (as_uchar4 (v).s3210));
}
-static inline u64 swap64_S (const u64 v)
+inline u64 swap64_S (const u64 v)
{
return (as_ulong (as_uchar8 (v).s76543210));
}
-static inline u32 rotr32_S (const u32 a, const u32 n)
+inline u32 rotr32_S (const u32 a, const u32 n)
{
return rotate (a, 32 - n);
}
-static inline u32 rotl32_S (const u32 a, const u32 n)
+inline u32 rotl32_S (const u32 a, const u32 n)
{
return rotate (a, n);
}
-static inline u64 rotr64_S (const u64 a, const u32 n)
+inline u64 rotr64_S (const u64 a, const u32 n)
{
return rotate (a, (u64) 64 - n);
}
-static inline u64 rotl64_S (const u64 a, const u32 n)
+inline u64 rotl64_S (const u64 a, const u32 n)
{
return rotate (a, (u64) n);
}
-static inline u32 amd_bytealign_S (const u32 a, const u32 b, const u32 c)
-{
- const u64 tmp = ((((u64) a) << 32) | ((u64) b)) >> ((c & 3) * 8);
-
- return (u32) (tmp);
-}
-
-static inline u32x swap32 (const u32x v)
+inline u32x swap32 (const u32x v)
{
return ((v >> 24) & 0x000000ff)
| ((v >> 8) & 0x0000ff00)
| ((v << 24) & 0xff000000);
}
-static inline u64x swap64 (const u64x v)
+inline u64x swap64 (const u64x v)
{
return ((v >> 56) & 0x00000000000000ff)
| ((v >> 40) & 0x000000000000ff00)
| ((v << 56) & 0xff00000000000000);
}
-static inline u32x rotr32 (const u32x a, const u32 n)
+inline u32x rotr32 (const u32x a, const u32 n)
{
return rotate (a, 32 - n);
}
-static inline u32x rotl32 (const u32x a, const u32 n)
+inline u32x rotl32 (const u32x a, const u32 n)
{
return rotate (a, n);
}
-static inline u64x rotr64 (const u64x a, const u32 n)
+inline u64x rotr64 (const u64x a, const u32 n)
{
return rotate (a, (u64) 64 - n);
}
-static inline u64x rotl64 (const u64x a, const u32 n)
+inline u64x rotl64 (const u64x a, const u32 n)
{
return rotate (a, (u64) n);
}
-static inline u32 __bfe (const u32 a, const u32 b, const u32 c)
+inline u32x __bfe (const u32x a, const u32x b, const u32x c)
+{
+ #define BIT(x) ((u32x) (1u) << (x))
+ #define BIT_MASK(x) (BIT (x) - 1)
+ #define BFE(x,y,z) (((x) >> (y)) & BIT_MASK (z))
+
+ return BFE (a, b, c);
+
+ #undef BIT
+ #undef BIT_MASK
+ #undef BFE
+}
+
+inline u32 __bfe_S (const u32 a, const u32 b, const u32 c)
{
- #define BIT(x) (1 << (x))
+ #define BIT(x) (1u << (x))
#define BIT_MASK(x) (BIT (x) - 1)
#define BFE(x,y,z) (((x) >> (y)) & BIT_MASK (z))
return BFE (a, b, c);
+
+ #undef BIT
+ #undef BIT_MASK
+ #undef BFE
}
-static inline u32x amd_bytealign (const u32x a, const u32x b, const u32 c)
+inline u32x amd_bytealign (const u32x a, const u32x b, const u32 c)
{
#if VECT_SIZE == 1
const u64x tmp = ((((u64x) (a)) << 32) | ((u64x) (b))) >> ((c & 3) * 8);
return (u32x) (tmp.s0, tmp.s1, tmp.s2, tmp.s3, tmp.s4, tmp.s5, tmp.s6, tmp.s7, tmp.s8, tmp.s9, tmp.sa, tmp.sb, tmp.sc, tmp.sd, tmp.se, tmp.sf);
#endif
}
+
+inline u32 amd_bytealign_S (const u32 a, const u32 b, const u32 c)
+{
+ const u64 tmp = ((((u64) a) << 32) | ((u64) b)) >> ((c & 3) * 8);
+
+ return (u32) (tmp);
+}
+
#endif
typedef struct
u32 digest_buf[4];
#elif defined _KEEPASS_
u32 digest_buf[4];
+ #elif defined _ZIP2_
+ u32 digest_buf[4];
#endif
} digest_t;
u32 salt_buf[16];
u32 data_buf[112];
u32 keyfile_buf[16];
+ u32 signature;
} tc_t;
} oldoffice34_t;
+typedef struct
+{
+ u32 salt_buf[128];
+ u32 salt_len;
+
+ u32 pc_digest[5];
+ u32 pc_offset;
+
+} pstoken_t;
+
+typedef struct
+{
+ u32 type;
+ u32 mode;
+ u32 magic;
+ u32 salt_len;
+ u32 salt_buf[4];
+ u32 verify_bytes;
+ u32 compress_length;
+ u32 data_len;
+ u32 data_buf[2048];
+ u32 auth_len;
+ u32 auth_buf[4];
+
+} zip2_t;
+
typedef struct
{
u32 version;
typedef struct
{
+ u32 salt_pos;
+ u32 digest_pos;
+ u32 hash_pos;
u32 gidvid;
u32 il_pos;