Unroll by 2x

1 files changed, 45 insertions, 2 deletions

diff --git a/src/lib/block/shacal2/shacal2_armv8/shacal2_arvm8.cpp b/src/lib/block/shacal2/shacal2_armv8/shacal2_arvm8.cpp
index 656335b6c..be9b62cc6 100644
--- a/src/lib/block/shacal2/shacal2_armv8/shacal2_arvm8.cpp
+++ b/src/lib/block/shacal2/shacal2_armv8/shacal2_arvm8.cpp
@@ -19,6 +19,49 @@ void SHACAL2::armv8_encrypt_blocks(const uint8_t in[], uint8_t out[], size_t blo
    const uint32_t* input32 = reinterpret_cast<const uint32_t*>(in);
    uint32_t* output32 = reinterpret_cast<uint32_t*>(out);
 
+   while(blocks >= 2)
+      {
+      uint32x4_t B0_0 = vld1q_u32(input32 +  0);
+      uint32x4_t B0_1 = vld1q_u32(input32 +  4);
+      uint32x4_t B1_0 = vld1q_u32(input32 +  8);
+      uint32x4_t B1_1 = vld1q_u32(input32 + 12);
+
+      B0_0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B0_0)));
+      B0_1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B0_1)));
+      B1_0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B1_0)));
+      B1_1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B1_1)));
+
+      for(size_t i = 0; i != 8; ++i)
+         {
+         const auto RK0 = vld1q_u32(&m_RK[8*i]);
+         const auto RK1 = vld1q_u32(&m_RK[8*i+4]);
+
+         const auto T0_0 = vsha256hq_u32(B0_0, B0_1, RK0);
+         const auto T0_1 = vsha256h2q_u32(B0_1, B0_0, RK0);
+         const auto T1_0 = vsha256hq_u32(B1_0, B1_1, RK0);
+         const auto T1_1 = vsha256h2q_u32(B1_1, B1_0, RK0);
+
+         B0_0 = vsha256hq_u32(T0_0, T0_1, RK1);
+         B0_1 = vsha256h2q_u32(T0_1, T0_0, RK1);
+         B1_0 = vsha256hq_u32(T1_0, T1_1, RK1);
+         B1_1 = vsha256h2q_u32(T1_1, T1_0, RK1);
+         }
+
+      B0_0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B0_0)));
+      B0_1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B0_1)));
+      B1_0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B1_0)));
+      B1_1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B1_1)));
+
+      vst1q_u32(&output32[ 0], B0_0);
+      vst1q_u32(&output32[ 4], B0_1);
+      vst1q_u32(&output32[ 8], B1_0);
+      vst1q_u32(&output32[12], B1_1);
+
+      blocks -= 2;
+      input32 += 16;
+      output32 += 16;
+      }
+
    while(blocks > 0)
       {
       uint32x4_t B0 = vld1q_u32(input32 + 0);
@@ -46,8 +89,8 @@ void SHACAL2::armv8_encrypt_blocks(const uint8_t in[], uint8_t out[], size_t blo
       vst1q_u32(&output32[4], B1);
 
       blocks--;
-      input32 += 32/4;
-      output32 += 32/4;
+      input32 += 8;
+      output32 += 8;
       }
    }