v3.21.1

JayDDee · Feb 9, 2023 · 520d4d5 · 520d4d5
1 parent da7030f
commit 520d4d5
Show file tree

Hide file tree

Showing 19 changed files with 260 additions and 798 deletions.
diff --git a/RELEASE_NOTES b/RELEASE_NOTES
@@ -65,10 +65,15 @@ If not what makes it happen or not happen?
 Change Log
 ----------
 
+v3.21.1
+
+Fixed a segfault in some obsolete algos.
+Small optimizations to Hamsi & Shabal AVX2 & AVX512.
+
 v3.21.0
 
 Added minotaurx algo for stratum only.
-Blake256 & sha256 prehash optimised to ignore zero-padded data for AVX2 & AVX512.
+Blake256 & sha256 prehash optimized to ignore zero-padded data for AVX2 & AVX512.
 Other small improvements.
 
 v3.20.3

diff --git a/algo/hamsi/hamsi-hash-4way.c b/algo/hamsi/hamsi-hash-4way.c
@@ -585,9 +585,8 @@ do { \
   t = _mm512_xor_si512( t, c ); \
   d = mm512_xoror( a, b, t ); \
   t = mm512_xorand( t, a, b ); \
-  b = mm512_xor3( b, d, t ); \
   a = c; \
-  c = b; \
+  c = mm512_xor3( b, d, t ); \
   b = d; \
   d = mm512_not( t ); \
 } while (0)
@@ -635,7 +634,7 @@ do { \
 
 #define ROUND_BIG8( alpha ) \
 do { \
-   __m512i t0, t1, t2, t3; \
+   __m512i t0, t1, t2, t3, t4, t5; \
    s0 = _mm512_xor_si512( s0, alpha[ 0] ); /* m0 */ \
    s1 = _mm512_xor_si512( s1, alpha[ 1] ); /* c0 */ \
    s2 = _mm512_xor_si512( s2, alpha[ 2] ); /* m1 */ \
@@ -662,43 +661,35 @@ do { \
   s5 = mm512_swap64_32( s5 ); \
   sD = mm512_swap64_32( sD ); \
   sE = mm512_swap64_32( sE ); \
-  t1 = _mm512_mask_blend_epi32( 0xaaaa, s4, s5 ); \
-  t3 = _mm512_mask_blend_epi32( 0xaaaa, sD, sE ); \
-  L8( s0, t1, s9, t3 ); \
-  s4 = _mm512_mask_blend_epi32( 0x5555, s4, t1 ); \
-  s5 = _mm512_mask_blend_epi32( 0xaaaa, s5, t1 ); \
-  sD = _mm512_mask_blend_epi32( 0x5555, sD, t3 ); \
-  sE = _mm512_mask_blend_epi32( 0xaaaa, sE, t3 ); \
+  t0 = _mm512_mask_blend_epi32( 0xaaaa, s4, s5 ); \
+  t1 = _mm512_mask_blend_epi32( 0xaaaa, sD, sE ); \
+  L8( s0, t0, s9, t1 ); \
 \
   s6 = mm512_swap64_32( s6 ); \
   sF = mm512_swap64_32( sF ); \
-  t1 = _mm512_mask_blend_epi32( 0xaaaa, s5, s6 ); \
+  t2 = _mm512_mask_blend_epi32( 0xaaaa, s5, s6 ); \
   t3 = _mm512_mask_blend_epi32( 0xaaaa, sE, sF ); \
-  L8( s1, t1, sA, t3 ); \
-  s5 = _mm512_mask_blend_epi32( 0x5555, s5, t1 ); \
-  s6 = _mm512_mask_blend_epi32( 0xaaaa, s6, t1 ); \
-  sE = _mm512_mask_blend_epi32( 0x5555, sE, t3 ); \
-  sF = _mm512_mask_blend_epi32( 0xaaaa, sF, t3 ); \
+  L8( s1, t2, sA, t3 ); \
+  s5 = _mm512_mask_blend_epi32( 0x5555, t0, t2 ); \
+  sE = _mm512_mask_blend_epi32( 0x5555, t1, t3 ); \
 \
   s7 = mm512_swap64_32( s7 ); \
   sC = mm512_swap64_32( sC ); \
-  t1 = _mm512_mask_blend_epi32( 0xaaaa, s6, s7 ); \
-  t3 = _mm512_mask_blend_epi32( 0xaaaa, sF, sC ); \
-  L8( s2, t1, sB, t3 ); \
-  s6 = _mm512_mask_blend_epi32( 0x5555, s6, t1 ); \
-  s7 = _mm512_mask_blend_epi32( 0xaaaa, s7, t1 ); \
-  sF = _mm512_mask_blend_epi32( 0x5555, sF, t3 ); \
-  sC = _mm512_mask_blend_epi32( 0xaaaa, sC, t3 ); \
+  t4 = _mm512_mask_blend_epi32( 0xaaaa, s6, s7 ); \
+  t5 = _mm512_mask_blend_epi32( 0xaaaa, sF, sC ); \
+  L8( s2, t4, sB, t5 ); \
+  s6 = _mm512_mask_blend_epi32( 0x5555, t2, t4 ); \
+  sF = _mm512_mask_blend_epi32( 0x5555, t3, t5 ); \
   s6 = mm512_swap64_32( s6 ); \
   sF = mm512_swap64_32( sF ); \
 \
-  t1 = _mm512_mask_blend_epi32( 0xaaaa, s7, s4 ); \
+  t2 = _mm512_mask_blend_epi32( 0xaaaa, s7, s4 ); \
   t3 = _mm512_mask_blend_epi32( 0xaaaa, sC, sD ); \
-  L8( s3, t1, s8, t3 ); \
-  s7 = _mm512_mask_blend_epi32( 0x5555, s7, t1 ); \
-  s4 = _mm512_mask_blend_epi32( 0xaaaa, s4, t1 ); \
-  sC = _mm512_mask_blend_epi32( 0x5555, sC, t3 ); \
-  sD = _mm512_mask_blend_epi32( 0xaaaa, sD, t3 ); \
+  L8( s3, t2, s8, t3 ); \
+  s7 = _mm512_mask_blend_epi32( 0x5555, t4, t2 ); \
+  s4 = _mm512_mask_blend_epi32( 0xaaaa, t0, t2 ); \
+  sC = _mm512_mask_blend_epi32( 0x5555, t5, t3 ); \
+  sD = _mm512_mask_blend_epi32( 0xaaaa, t1, t3 ); \
   s7 = mm512_swap64_32( s7 ); \
   sC = mm512_swap64_32( sC ); \
 \
@@ -924,10 +915,9 @@ do { \
   d = _mm256_xor_si256( d, a ); \
   a = _mm256_and_si256( a, b ); \
   t = _mm256_xor_si256( t, a ); \
-  b = _mm256_xor_si256( b, d ); \
-  b = _mm256_xor_si256( b, t ); \
   a = c; \
-  c = b; \
+  c = _mm256_xor_si256( b, d ); \
+  c = _mm256_xor_si256( c, t ); \
   b = d; \
   d = mm256_not( t ); \
 } while (0)
@@ -977,7 +967,7 @@ do { \
 
 #define ROUND_BIG( alpha ) \
 do { \
-   __m256i t0, t1, t2, t3; \
+   __m256i t0, t1, t2, t3, t4, t5; \
    s0 = _mm256_xor_si256( s0, alpha[ 0] ); \
    s1 = _mm256_xor_si256( s1, alpha[ 1] ); \
    s2 = _mm256_xor_si256( s2, alpha[ 2] ); \
@@ -1004,43 +994,35 @@ do { \
   s5 = mm256_swap64_32( s5 ); \
   sD = mm256_swap64_32( sD ); \
   sE = mm256_swap64_32( sE ); \
-  t1 = _mm256_blend_epi32( s4, s5, 0xaa ); \
-  t3 = _mm256_blend_epi32( sD, sE, 0xaa ); \
-  L( s0, t1, s9, t3 ); \
-  s4 = _mm256_blend_epi32( s4, t1, 0x55 ); \
-  s5 = _mm256_blend_epi32( s5, t1, 0xaa ); \
-  sD = _mm256_blend_epi32( sD, t3, 0x55 ); \
-  sE = _mm256_blend_epi32( sE, t3, 0xaa ); \
+  t0 = _mm256_blend_epi32( s4, s5, 0xaa ); \
+  t1 = _mm256_blend_epi32( sD, sE, 0xaa ); \
+  L( s0, t0, s9, t1 ); \
 \
   s6 = mm256_swap64_32( s6 ); \
   sF = mm256_swap64_32( sF ); \
-  t1 = _mm256_blend_epi32( s5, s6, 0xaa ); \
+  t2 = _mm256_blend_epi32( s5, s6, 0xaa ); \
   t3 = _mm256_blend_epi32( sE, sF, 0xaa ); \
-  L( s1, t1, sA, t3 ); \
-  s5 = _mm256_blend_epi32( s5, t1, 0x55 ); \
-  s6 = _mm256_blend_epi32( s6, t1, 0xaa ); \
-  sE = _mm256_blend_epi32( sE, t3, 0x55 ); \
-  sF = _mm256_blend_epi32( sF, t3, 0xaa ); \
+  L( s1, t2, sA, t3 ); \
+  s5 = _mm256_blend_epi32( t0, t2, 0x55 ); \
+  sE = _mm256_blend_epi32( t1, t3, 0x55 ); \
 \
   s7 = mm256_swap64_32( s7 ); \
   sC = mm256_swap64_32( sC ); \
-  t1 = _mm256_blend_epi32( s6, s7, 0xaa ); \
-  t3 = _mm256_blend_epi32( sF, sC, 0xaa ); \
-  L( s2, t1, sB, t3 ); \
-  s6 = _mm256_blend_epi32( s6, t1, 0x55 ); \
-  s7 = _mm256_blend_epi32( s7, t1, 0xaa ); \
-  sF = _mm256_blend_epi32( sF, t3, 0x55 ); \
-  sC = _mm256_blend_epi32( sC, t3, 0xaa ); \
+  t4 = _mm256_blend_epi32( s6, s7, 0xaa ); \
+  t5 = _mm256_blend_epi32( sF, sC, 0xaa ); \
+  L( s2, t4, sB, t5 ); \
+  s6 = _mm256_blend_epi32( t2, t4, 0x55 ); \
+  sF = _mm256_blend_epi32( t3, t5, 0x55 ); \
   s6 = mm256_swap64_32( s6 ); \
   sF = mm256_swap64_32( sF ); \
 \
-  t1 = _mm256_blend_epi32( s7, s4, 0xaa ); \
+  t2 = _mm256_blend_epi32( s7, s4, 0xaa ); \
   t3 = _mm256_blend_epi32( sC, sD, 0xaa ); \
-  L( s3, t1, s8, t3 ); \
-  s7 = _mm256_blend_epi32( s7, t1, 0x55 ); \
-  s4 = _mm256_blend_epi32( s4, t1, 0xaa ); \
-  sC = _mm256_blend_epi32( sC, t3, 0x55 ); \
-  sD = _mm256_blend_epi32( sD, t3, 0xaa ); \
+  L( s3, t2, s8, t3 ); \
+  s7 = _mm256_blend_epi32( t4, t2, 0x55 ); \
+  s4 = _mm256_blend_epi32( t0, t2, 0xaa ); \
+  sC = _mm256_blend_epi32( t5, t3, 0x55 ); \
+  sD = _mm256_blend_epi32( t1, t3, 0xaa ); \
   s7 = mm256_swap64_32( s7 ); \
   sC = mm256_swap64_32( sC ); \
 \

diff --git a/algo/haval/haval-hash-4way.c b/algo/haval/haval-hash-4way.c
@@ -141,6 +141,13 @@ do { \
                        _mm_add_epi32( w, _mm_set1_epi32( c ) ) ); \
 } while (0)
 
+#define STEP1(n, p, x7, x6, x5, x4, x3, x2, x1, x0, w) \
+do { \
+   __m128i t = FP ## n ## _ ## p(x6, x5, x4, x3, x2, x1, x0); \
+   x7 = _mm_add_epi32( _mm_add_epi32( mm128_ror_32( t, 7 ), \
+                                      mm128_ror_32( x7, 11 ) ), w ); \
+} while (0)
+
 /*
  * PASSy(n, in) computes pass number "y", for a total of "n", using the
  * one-argument macro "in" to access input words. Current state is assumed
@@ -152,22 +159,22 @@ do { \
 #define PASS1(n, in)   do { \
 		unsigned pass_count; \
 		for (pass_count = 0; pass_count < 32; pass_count += 8) { \
-			STEP(n, 1, s7, s6, s5, s4, s3, s2, s1, s0, \
-				in(pass_count + 0), SPH_C32(0x00000000)); \
-			STEP(n, 1, s6, s5, s4, s3, s2, s1, s0, s7, \
-				in(pass_count + 1), SPH_C32(0x00000000)); \
-			STEP(n, 1, s5, s4, s3, s2, s1, s0, s7, s6, \
-				in(pass_count + 2), SPH_C32(0x00000000)); \
-			STEP(n, 1, s4, s3, s2, s1, s0, s7, s6, s5, \
-				in(pass_count + 3), SPH_C32(0x00000000)); \
-			STEP(n, 1, s3, s2, s1, s0, s7, s6, s5, s4, \
-				in(pass_count + 4), SPH_C32(0x00000000)); \
-			STEP(n, 1, s2, s1, s0, s7, s6, s5, s4, s3, \
-				in(pass_count + 5), SPH_C32(0x00000000)); \
-			STEP(n, 1, s1, s0, s7, s6, s5, s4, s3, s2, \
-				in(pass_count + 6), SPH_C32(0x00000000)); \
-			STEP(n, 1, s0, s7, s6, s5, s4, s3, s2, s1, \
-				in(pass_count + 7), SPH_C32(0x00000000)); \
+			STEP1(n, 1, s7, s6, s5, s4, s3, s2, s1, s0, \
+				in(pass_count + 0) ); \
+			STEP1(n, 1, s6, s5, s4, s3, s2, s1, s0, s7, \
+				in(pass_count + 1) ); \
+			STEP1(n, 1, s5, s4, s3, s2, s1, s0, s7, s6, \
+				in(pass_count + 2) ); \
+			STEP1(n, 1, s4, s3, s2, s1, s0, s7, s6, s5, \
+				in(pass_count + 3) ); \
+			STEP1(n, 1, s3, s2, s1, s0, s7, s6, s5, s4, \
+				in(pass_count + 4) ); \
+			STEP1(n, 1, s2, s1, s0, s7, s6, s5, s4, s3, \
+				in(pass_count + 5) ); \
+			STEP1(n, 1, s1, s0, s7, s6, s5, s4, s3, s2, \
+				in(pass_count + 6) ); \
+			STEP1(n, 1, s0, s7, s6, s5, s4, s3, s2, s1, \
+				in(pass_count + 7) ); \
    		} \
 	} while (0)
 
@@ -605,25 +612,32 @@ do { \
                        _mm256_add_epi32( w, _mm256_set1_epi32( c ) ) ); \
 } while (0)
 
+#define STEP1_8W(n, p, x7, x6, x5, x4, x3, x2, x1, x0, w) \
+do { \
+   __m256i t = FP ## n ## _ ## p ## _8W(x6, x5, x4, x3, x2, x1, x0); \
+   x7 = _mm256_add_epi32( _mm256_add_epi32( mm256_ror_32( t, 7 ), \
+                                      mm256_ror_32( x7, 11 ) ), w ); \
+} while (0)
+
 #define PASS1_8W(n, in)   do { \
       unsigned pass_count; \
       for (pass_count = 0; pass_count < 32; pass_count += 8) { \
-         STEP_8W(n, 1, s7, s6, s5, s4, s3, s2, s1, s0, \
-            in(pass_count + 0), SPH_C32(0x00000000)); \
-         STEP_8W(n, 1, s6, s5, s4, s3, s2, s1, s0, s7, \
-            in(pass_count + 1), SPH_C32(0x00000000)); \
-         STEP_8W(n, 1, s5, s4, s3, s2, s1, s0, s7, s6, \
-            in(pass_count + 2), SPH_C32(0x00000000)); \
-         STEP_8W(n, 1, s4, s3, s2, s1, s0, s7, s6, s5, \
-            in(pass_count + 3), SPH_C32(0x00000000)); \
-         STEP_8W(n, 1, s3, s2, s1, s0, s7, s6, s5, s4, \
-            in(pass_count + 4), SPH_C32(0x00000000)); \
-         STEP_8W(n, 1, s2, s1, s0, s7, s6, s5, s4, s3, \
-            in(pass_count + 5), SPH_C32(0x00000000)); \
-         STEP_8W(n, 1, s1, s0, s7, s6, s5, s4, s3, s2, \
-            in(pass_count + 6), SPH_C32(0x00000000)); \
-         STEP_8W(n, 1, s0, s7, s6, s5, s4, s3, s2, s1, \
-            in(pass_count + 7), SPH_C32(0x00000000)); \
+         STEP1_8W(n, 1, s7, s6, s5, s4, s3, s2, s1, s0, \
+            in(pass_count + 0) ); \
+         STEP1_8W(n, 1, s6, s5, s4, s3, s2, s1, s0, s7, \
+            in(pass_count + 1) ); \
+         STEP1_8W(n, 1, s5, s4, s3, s2, s1, s0, s7, s6, \
+            in(pass_count + 2) ); \
+         STEP1_8W(n, 1, s4, s3, s2, s1, s0, s7, s6, s5, \
+            in(pass_count + 3) ); \
+         STEP1_8W(n, 1, s3, s2, s1, s0, s7, s6, s5, s4, \
+            in(pass_count + 4) ); \
+         STEP1_8W(n, 1, s2, s1, s0, s7, s6, s5, s4, s3, \
+            in(pass_count + 5) ); \
+         STEP1_8W(n, 1, s1, s0, s7, s6, s5, s4, s3, s2, \
+            in(pass_count + 6) ); \
+         STEP1_8W(n, 1, s0, s7, s6, s5, s4, s3, s2, s1, \
+            in(pass_count + 7) ); \
          } \
    } while (0)
 

diff --git a/algo/keccak/keccak-hash-4way.c b/algo/keccak/keccak-hash-4way.c
@@ -72,11 +72,11 @@ static const uint64_t RC[] = {
 // Targetted macros, keccak-macros.h is included for each target.
 
 #define DECL64(x)          __m512i x
-#define XOR(d, a, b)     (d = _mm512_xor_si512(a,b))
-#define XOR64 XOR
+#define XOR(d, a, b)       (d = _mm512_xor_si512(a,b))
+#define XOR64              XOR
 #define AND64(d, a, b)     (d = _mm512_and_si512(a,b))
 #define OR64(d, a, b)      (d = _mm512_or_si512(a,b))
-#define NOT64(d, s)        (d = _mm512_xor_si512(s,m512_neg1))
+#define NOT64(d, s)        (d = mm512_not( s ) )
 #define ROL64(d, v, n)     (d = mm512_rol_64(v, n))
 #define XOROR(d, a, b, c)  (d = mm512_xoror(a, b, c))
 #define XORAND(d, a, b, c) (d = mm512_xorand(a, b, c))
@@ -257,14 +257,14 @@ keccak512_8way_close(void *cc, void *dst)
         kc->w[j ] = _mm256_xor_si256( kc->w[j], buf[j] ); \
 } while (0)
 
-#define DECL64(x)        __m256i x
-#define XOR(d, a, b)    (d = _mm256_xor_si256(a,b))
-#define XOR64 XOR
-#define AND64(d, a, b)   (d = _mm256_and_si256(a,b))
-#define OR64(d, a, b)    (d = _mm256_or_si256(a,b))
-#define NOT64(d, s)      (d = _mm256_xor_si256(s,m256_neg1))
-#define ROL64(d, v, n)   (d = mm256_rol_64(v, n))
-#define XOROR(d, a, b, c) (d = _mm256_xor_si256(a, _mm256_or_si256(b, c)))
+#define DECL64(x)          __m256i x
+#define XOR(d, a, b)       (d = _mm256_xor_si256(a,b))
+#define XOR64              XOR
+#define AND64(d, a, b)     (d = _mm256_and_si256(a,b))
+#define OR64(d, a, b)      (d = _mm256_or_si256(a,b))
+#define NOT64(d, s)        (d = mm256_not( s ) )
+#define ROL64(d, v, n)     (d = mm256_rol_64(v, n))
+#define XOROR(d, a, b, c)  (d = _mm256_xor_si256(a, _mm256_or_si256(b, c)))
 #define XORAND(d, a, b, c) (d = _mm256_xor_si256(a, _mm256_and_si256(b, c)))
 #define XOR3( d, a, b, c ) (d = mm256_xor3( a, b, c ))
 

diff --git a/algo/lyra2/lyra2z330.c b/algo/lyra2/lyra2z330.c
@@ -3,7 +3,7 @@
 #include "lyra2.h"
 #include "simd-utils.h"
 
-__thread uint64_t* lyra2z330_wholeMatrix;
+static __thread uint64_t* lyra2z330_wholeMatrix;
 
 void lyra2z330_hash(void *state, const void *input, uint32_t height)
 {