v24.2

RELEASE_NOTES

@@ -75,6 +75,12 @@ If not what makes it happen or not happen?
 Change Log
 ----------
 
+v24.2
+
+x86_64: Fixed blakes2s for AVX2 & AVX512, x25x for AVX512, broken in v3.23.4.
+x86_64: Initial support for CPUs with AVX10, needs GCC-14.
+ARM NEON: Various code optimisations.
+
 v24.1
 
 #414: fix bug in merkle error handling.

algo-gate-api.c

@@ -184,7 +184,7 @@ int scanhash_4way_64in_32out( struct work *work, uint32_t max_nonce,
 
 #endif
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 //int scanhash_8way_64_64( struct work *work, uint32_t max_nonce,
 //                      uint64_t *hashes_done, struct thr_info *mythr )

algo-gate-api.h

@@ -99,8 +99,11 @@ typedef  uint32_t set_t;
 #define AES_OPT          1 <<  7   // Intel Westmere, AArch64
 #define VAES_OPT         1 <<  8   // Icelake, Zen3
 #define SHA_OPT          1 <<  9   // Zen1, Icelake, AArch64 
+#define SHA256_OPT       1 <<  9   // Zen1, Icelake, AArch64 
 #define SHA512_OPT       1 << 10   // Intel Arrow Lake, AArch64 
 #define NEON_OPT         1 << 11   // AArch64 
+#define AVX10_256        1 << 12
+#define AVX10_512        1 << 13
 
 // AVX10 does not have explicit algo features:
 //  AVX10_512 is compatible with AVX512 + VAES
@@ -246,7 +249,7 @@ int scanhash_4way_64in_32out( struct work *work, uint32_t max_nonce,
 
 #endif
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 //int scanhash_8way_64in_64out( struct work *work, uint32_t max_nonce,
 //                      uint64_t *hashes_done, struct thr_info *mythr );

algo/argon2d/argon2d/opt.c

@@ -35,7 +35,7 @@
  * @pre all block pointers must be valid
  */
 
-#if defined(__AVX512F__)
+#if defined(SIMD512)
 
 static inline __m512i blamka( __m512i x, __m512i y )
 {
@@ -237,7 +237,7 @@ void fill_segment(const argon2_instance_t *instance,
     uint64_t pseudo_rand, ref_index, ref_lane;
     uint32_t prev_offset, curr_offset;
     uint32_t starting_index, i;
-#if defined(__AVX512F__)
+#if defined(SIMD512)
     __m512i state[ARGON2_512BIT_WORDS_IN_BLOCK];
 #elif defined(__AVX2__)
     __m256i state[ARGON2_HWORDS_IN_BLOCK];

algo/argon2d/blake2/blamka-round-opt.h

@@ -21,7 +21,7 @@
 #include "blake2-impl.h"
 #include "simd-utils.h"
 
-#if !defined(__AVX512F__)
+#if !defined(SIMD512)
 
 #if !defined(__AVX2__)
 

algo/blake/blake256-hash.c

@@ -1611,7 +1611,7 @@ void blake256_8way_final_rounds_le( void *final_hash, const void *midstate,
 
 #endif
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 ///////////////////////////////////////
 //
@@ -2617,7 +2617,7 @@ blake32_8way_close_le( blake_8way_small_context *sc, unsigned ub, unsigned n,
 #endif
 
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 //Blake-256 16 way AVX512
 

algo/blake/blake256-hash.h

@@ -147,7 +147,7 @@ void blake256r8_8way_close(void *cc, void *dst);
 #define blake256r8_8x32_update        blake256r14_8way_update
 #define blake256r8_8x32_close         blake256r14_8way_close
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 ///////////////////////////////////
 //

algo/blake/blake2b-hash.c

@@ -226,7 +226,7 @@ static const uint8_t sigma[12][16] =
 #define Mx_(n)      Mx__(n)
 #define Mx__(n)     M ## n
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 #define B2B8W_G(a, b, c, d, x, y) \
 { \

algo/blake/blake2b-hash.h

@@ -15,7 +15,7 @@
 #endif
 
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 typedef struct ALIGN( 64 ) {
    __m512i b[16]; // input buffer

algo/blake/blake2b.c

@@ -3,7 +3,7 @@
 #include <stdint.h>
 #include "blake2b-hash.h"
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
   #define BLAKE2B_8WAY
 #elif defined(__AVX2__)
   #define BLAKE2B_4WAY

algo/blake/blake2s-hash.c

@@ -497,7 +497,7 @@ int blake2s_8way_full_blocks( blake2s_8way_state *S, void *out,
 
 #endif // __AVX2__
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 // Blake2s-256 16 way
 

algo/blake/blake2s-hash.h

@@ -29,20 +29,20 @@
 #define ALIGN(x) __attribute__((aligned(x)))
 #endif
 
-   typedef struct __blake2s_nway_param
-   {
-      uint8_t  digest_length; // 1
-      uint8_t  key_length;    // 2
-      uint8_t  fanout;        // 3
-      uint8_t  depth;         // 4
-      uint32_t leaf_length;   // 8
-      uint8_t  node_offset[6];// 14
-      uint8_t  node_depth;    // 15
-      uint8_t  inner_length;  // 16
-      // uint8_t  reserved[0];
-      uint8_t  salt[8]; // 24
-      uint8_t  personal[8];  // 32
-   } blake2s_nway_param;
+typedef struct __blake2s_nway_param
+{
+   uint8_t  digest_length; // 1
+   uint8_t  key_length;    // 2
+   uint8_t  fanout;        // 3
+   uint8_t  depth;         // 4
+   uint32_t leaf_length;   // 8
+   uint8_t  node_offset[6];// 14
+   uint8_t  node_depth;    // 15
+   uint8_t  inner_length;  // 16
+   // uint8_t  reserved[0];
+   uint8_t  salt[8]; // 24
+   uint8_t  personal[8];  // 32
+} blake2s_nway_param;
 
 typedef struct ALIGN( 64 ) __blake2s_4way_state
 {
@@ -67,7 +67,7 @@ int blake2s_4way_full_blocks( blake2s_4way_state *S, void *out,
 typedef struct ALIGN( 64 ) __blake2s_8way_state
 {
    __m256i h[8];
-   uint8_t  buf[ 32 * 8 ];
+   uint8_t  buf[ 64 * 8 ];
    uint32_t t[2];
    uint32_t f[2];
    size_t   buflen;
@@ -83,12 +83,12 @@ int blake2s_8way_full_blocks( blake2s_8way_state *S, void *out,
 
 #endif
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 typedef struct ALIGN( 64 ) __blake2s_16way_state
 {
    __m512i h[8];
-   uint8_t  buf[ 32 * 16 ];
+   uint8_t  buf[ 64 * 16 ];
    uint32_t t[2];
    uint32_t f[2];
    size_t   buflen;

algo/blake/blake2s.c

@@ -3,7 +3,7 @@
 #include <string.h>
 #include <stdint.h>
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
   #define BLAKE2S_16WAY
 #elif defined(__AVX2__)
   #define BLAKE2S_8WAY

algo/blake/blake512-hash.c

@@ -349,16 +349,16 @@ void blake512_transform( uint64_t *H, const uint64_t *buf, const uint64_t T0,
    Va = v128_add64( Va, v128_add64( Vb, \
                             v128_set64( CBx( r, Sd ) ^ Mx( r, Sc ), \
                                         CBx( r, Sb ) ^ Mx( r, Sa ) ) ) ); \
-   Vd = v128_ror64( v128_xor( Vd, Va ), 32 ); \
+   Vd = v128_ror64xor( Vd, Va, 32 ); \
    Vc = v128_add64( Vc, Vd ); \
-   Vb = v128_ror64( v128_xor( Vb, Vc ), 25 ); \
+   Vb = v128_ror64xor( Vb, Vc, 25 ); \
 \
    Va = v128_add64( Va, v128_add64( Vb, \
                             v128_set64( CBx( r, Sc ) ^ Mx( r, Sd ), \
                                         CBx( r, Sa ) ^ Mx( r, Sb ) ) ) ); \
-   Vd = v128_ror64( v128_xor( Vd, Va ), 16 ); \
+   Vd = v128_ror64xor( Vd, Va, 16 ); \
    Vc = v128_add64( Vc, Vd ); \
-   Vb = v128_ror64( v128_xor( Vb, Vc ), 11 ); \
+   Vb = v128_ror64xor( Vb, Vc, 11 ); \
 }
 
 #define BLAKE512_ROUND( R ) \
@@ -559,7 +559,7 @@ void blake512_full( blake512_context *sc, void *dst, const void *data,
 
 #if defined(__AVX2__)
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 ////////////////////////////////////
 //
@@ -1887,13 +1887,13 @@ blake512_4x64_close(void *cc, void *dst)
 #define GB_2X64( m0, m1, c0, c1, a, b, c, d ) \
 { \
    a = v128_add64( v128_add64( v128_xor( v128_64( c1 ), m0 ), b ), a ); \
-   d = v128_ror64( v128_xor( d, a ), 32 ); \
+   d = v128_ror64xor( d, a, 32 ); \
    c = v128_add64( c, d ); \
-   b = v128_ror64( v128_xor( b, c ), 25 ); \
+   b = v128_ror64xor( b, c, 25 ); \
    a = v128_add64( v128_add64( v128_xor( v128_64( c0 ), m1 ), b ), a ); \
-   d = v128_ror64( v128_xor( d, a ), 16 ); \
+   d = v128_ror64xor( d, a, 16 ); \
    c = v128_add64( c, d ); \
-   b = v128_ror64( v128_xor( b, c ), 11 ); \
+   b = v128_ror64xor( b, c, 11 ); \
 }
 
 #define ROUND_B_2X64(r) \
@@ -2054,9 +2054,9 @@ void blake512_2x64_prehash_part1_le( blake_2x64_big_context *sc,
    // G4 skip nonce
    V0 = v128_add64( v128_add64( v128_xor( v128_64( CB9 ), sc->buf[ 8] ), V5 ),
                                           V0 );
-   VF = v128_ror64( v128_xor( VF, V0 ), 32 );
+   VF = v128_ror64xor( VF, V0, 32 );
    VA = v128_add64( VA, VF );
-   V5 = v128_ror64( v128_xor( V5, VA ), 25 );
+   V5 = v128_ror64xor( V5, VA, 25 );
    V0 = v128_add64( V0, V5 );
 
    GB_2X64( sc->buf[10], sc->buf[11], CBA, CBB, V1, V6, VB, VC );
@@ -2137,44 +2137,44 @@ void blake512_2x64_prehash_part2_le( blake_2x64_big_context *sc, void *hash,
 
    // finish round 0, with the nonce now available 
    V0 = v128_add64( V0, v128_xor( v128_64( CB8 ), M9 ) );
-   VF = v128_ror64( v128_xor( VF, V0 ), 16 );
+   VF = v128_ror64xor( VF, V0, 16 );
    VA = v128_add64( VA, VF );
-   V5 = v128_ror64( v128_xor( V5, VA ), 11 );
+   V5 = v128_ror64xor( V5, VA, 11 );
 
    // Round 1
    // G0
    GB_2X64(Mx(1, 0), Mx(1, 1), CBx(1, 0), CBx(1, 1), V0, V4, V8, VC);
 
    // G1
    V1 = v128_add64( V1, V5 );
-   VD = v128_ror64( v128_xor( VD, V1 ), 32 );
+   VD = v128_ror64xor( VD, V1, 32 );
    V9 = v128_add64( V9, VD );
-   V5 = v128_ror64( v128_xor( V5, V9 ), 25 );
+   V5 = v128_ror64xor( V5, V9, 25 );
    V1 = v128_add64( V1, v128_add64( v128_xor( v128_64( CBx(1,2) ), Mx(1,3) ),
                                               V5 ) );
-   VD = v128_ror64( v128_xor( VD, V1 ), 16 );
+   VD = v128_ror64xor( VD, V1, 16 );
    V9 = v128_add64( V9, VD );
-   V5 = v128_ror64( v128_xor( V5, V9 ), 11 );
+   V5 = v128_ror64xor( V5, V9, 11 );
 
    // G2
    V2 = v128_add64( V2, v128_xor( v128_64( CBF ), M9 ) );
-   VE = v128_ror64( v128_xor( VE, V2 ), 32 );
+   VE = v128_ror64xor( VE, V2, 32 );
    VA = v128_add64( VA, VE );
-   V6 = v128_ror64( v128_xor( V6, VA ), 25 );
+   V6 = v128_ror64xor( V6, VA, 25 );
    V2 = v128_add64( V2, v128_add64( v128_xor( v128_64( CB9 ), MF ), V6 ) );
-   VE = v128_ror64( v128_xor( VE, V2 ), 16 );
+   VE = v128_ror64xor( VE, V2, 16 );
    VA = v128_add64( VA, VE );
-   V6 = v128_ror64( v128_xor( V6, VA ), 11 );
+   V6 = v128_ror64xor( V6, VA, 11 );
 
    // G3
-   VF = v128_ror64( v128_xor( VF, V3 ), 32 );
+   VF = v128_ror64xor( VF, V3, 32 );
    VB = v128_add64( VB, VF );
-   V7 = v128_ror64( v128_xor( V7, VB ), 25 );
+   V7 = v128_ror64xor( V7, VB, 25 );
    V3 = v128_add64( V3, v128_add64( v128_xor( v128_64( CBx(1, 6) ), Mx(1, 7) ),
                                               V7 ) );
-   VF = v128_ror64( v128_xor( VF, V3 ), 16 );
+   VF = v128_ror64xor( VF, V3, 16 );
    VB = v128_add64( VB, VF );
-   V7 = v128_ror64( v128_xor( V7, VB ), 11 );
+   V7 = v128_ror64xor( V7, VB, 11 );
 
    // G4, G5, G6, G7
    GB_2X64(Mx(1, 8), Mx(1, 9), CBx(1, 8), CBx(1, 9), V0, V5, VA, VF);

algo/blake/blake512-hash.h

@@ -92,7 +92,7 @@ void blake512_4x64_final_le( blake_4x64_big_context *sc, void *hash,
 #define blake512_4way_prehash_le  blake512_4x64_prehash_le
 #define blake512_4way_final_le    blake512_4x64_final_le
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 ////////////////////////////
 //

algo/blake/blakecoin-gate.h

@@ -4,7 +4,7 @@
 #include "algo-gate-api.h"
 #include <stdint.h>
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
   #define BLAKECOIN_16WAY
 #elif defined(__AVX2__)
   #define BLAKECOIN_8WAY

algo/blake/sph_blake2b.c

@@ -101,15 +101,15 @@
 { \
    Va = v128_add64( Va, v128_add64( Vb, \
                  v128_set64( m[ sigmaR[ Sc ] ], m[ sigmaR[ Sa ] ] ) ) ); \
-   Vd = v128_ror64( v128_xor( Vd, Va ), 32 ); \
+   Vd = v128_ror64xor( Vd, Va, 32 ); \
    Vc = v128_add64( Vc, Vd ); \
-   Vb = v128_ror64( v128_xor( Vb, Vc ), 24 ); \
+   Vb = v128_ror64xor( Vb, Vc, 24 ); \
 \
    Va = v128_add64( Va, v128_add64( Vb, \
                  v128_set64( m[ sigmaR[ Sd ] ], m[ sigmaR[ Sb ] ] ) ) ); \
-   Vd = v128_ror64( v128_xor( Vd, Va ), 16 ); \
+   Vd = v128_ror64xor( Vd, Va, 16 ); \
    Vc = v128_add64( Vc, Vd ); \
-   Vb = v128_ror64( v128_xor( Vb, Vc ), 63 ); \
+   Vb = v128_ror64xor( Vb, Vc, 63 ); \
 }
 
 #define BLAKE2B_ROUND( R ) \

algo/bmw/bmw-hash-4way.h

@@ -87,7 +87,7 @@ void bmw256_8way_close( bmw256_8way_context *ctx, void *dst );
 
 #endif
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 // BMW-256 16 way 32
 
@@ -157,7 +157,7 @@ void bmw512_4way_addbits_and_close(
 
 #endif  // __AVX2__
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 // BMW-512 64 bit 8 way
 typedef struct

algo/bmw/bmw256-hash-4way.c

@@ -1057,7 +1057,7 @@ void bmw256_8way_close( bmw256_8way_context *ctx, void *dst )
 
 #endif // __AVX2__
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
 
 // BMW-256 16 way 32
 

algo/bmw/bmw512-gate.h

@@ -4,7 +4,7 @@
 #include "algo-gate-api.h"
 #include <stdint.h>
 
-#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
+#if defined(SIMD512)
   #define BMW512_8WAY 1
 #elif defined(__AVX2__)
   #define BMW512_4WAY 1