Shabal Optimisations

pongit · Sep 14, 2018 · 69419d4 · 69419d4
1 parent 21a176a
commit 69419d4
Show file tree

Hide file tree

Showing 6 changed files with 14 additions and 14 deletions.
diff --git a/Cargo.lock b/Cargo.lock
diff --git a/Cargo.toml b/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "scavenger"
-version = "1.2.0"
+version = "1.3.0"
 license = "GPL-3.0"
 authors = ["PoC Consortium <[email protected]>"]
 description = """

diff --git a/src/c/mshabal_128_avx.c b/src/c/mshabal_128_avx.c
@@ -578,9 +578,9 @@ static void simd128_avx_mshabal_compress_fast(mshabal_context_fast* sc, void* u1
     }
 
     // transfer results to ram
-    for (j = 0; j < 12; j++) _mm_storeu_si128((__m128i*)sc->state + j, A[j]);
-    for (j = 0; j < 16; j++) {
-        _mm_storeu_si128((__m128i*)sc->state + j + 12, B[j]);
+    //for (j = 0; j < 12; j++) _mm_storeu_si128((__m128i*)sc->state + j, A[j]);
+    for (j = 8; j < 10; j++) {
+       // _mm_storeu_si128((__m128i*)sc->state + j + 12, B[j]);
         _mm_storeu_si128((__m128i*)sc->state + j + 28, C[j]);
     }
 }

diff --git a/src/c/mshabal_128_sse2.c b/src/c/mshabal_128_sse2.c
@@ -588,9 +588,9 @@ static void simd128_sse2_mshabal_compress_fast(mshabal_context_fast* sc, void* u
     }
 
     // transfer results to ram
-    for (j = 0; j < 12; j++) _mm_storeu_si128((__m128i*)sc->state + j, A[j]);
-    for (j = 0; j < 16; j++) {
-        _mm_storeu_si128((__m128i*)sc->state + j + 12, B[j]);
+    //for (j = 0; j < 12; j++) _mm_storeu_si128((__m128i*)sc->state + j, A[j]);
+    for (j = 8; j < 10; j++) {
+       // _mm_storeu_si128((__m128i*)sc->state + j + 12, B[j]);
         _mm_storeu_si128((__m128i*)sc->state + j + 28, C[j]);
     }
 }

diff --git a/src/c/mshabal_256_avx2.c b/src/c/mshabal_256_avx2.c
@@ -638,9 +638,9 @@ static void simd256_mshabal_compress_fast(mshabal256_context_fast* sc, void* u1,
     }
 
     // transfer results to ram
-    for (j = 0; j < 12; j++) _mm256_storeu_si256((__m256i*)sc->state + j, A[j]);
-    for (j = 0; j < 16; j++) {
-        _mm256_storeu_si256((__m256i*)sc->state + j + 12, B[j]);
+    //for (j = 0; j < 12; j++) _mm256_storeu_si256((__m256i*)sc->state + j, A[j]);
+    for (j = 8; j < 10; j++) {
+        //_mm256_storeu_si256((__m256i*)sc->state + j + 12, B[j]);
         _mm256_storeu_si256((__m256i*)sc->state + j + 28, C[j]);
     }
 }

diff --git a/src/c/mshabal_512_avx512f.c b/src/c/mshabal_512_avx512f.c
@@ -737,9 +737,9 @@ static void simd512_mshabal_compress_fast(mshabal512_context_fast *sc, void *u1,
     }
 
     // transfer results to ram
-    for (j = 0; j < 12; j++) _mm512_storeu_si512((__m512i *)sc->state + j, A[j]);
-    for (j = 0; j < 16; j++) {
-        _mm512_storeu_si512((__m512i *)sc->state + j + 12, B[j]);
+    //for (j = 0; j < 12; j++) _mm512_storeu_si512((__m512i *)sc->state + j, A[j]);
+    for (j = 8 j < 10; j++) {
+        //_mm512_storeu_si512((__m512i *)sc->state + j + 12, B[j]);
         _mm512_storeu_si512((__m512i *)sc->state + j + 28, C[j]);
     }
 }