Add support for ARMv8 SM4 instructions

Tested in qemu

4 files changed, 221 insertions, 0 deletions

diff --git a/src/lib/block/sm4/sm4.cpp b/src/lib/block/sm4/sm4.cpp
index 7c409d40f..7a370a67b 100644
--- a/src/lib/block/sm4/sm4.cpp
+++ b/src/lib/block/sm4/sm4.cpp
@@ -1,12 +1,14 @@
 /*
 * SM4
 * (C) 2017 Ribose Inc
+* (C) 2018 Jack Lloyd
 *
 * Botan is released under the Simplified BSD License (see license.txt)
 */
 
 #include <botan/sm4.h>
 #include <botan/loadstor.h>
+#include <botan/cpuid.h>
 
 namespace Botan {
 
@@ -126,6 +128,11 @@ void SM4::encrypt_n(const uint8_t in[], uint8_t out[], size_t blocks) const
    {
    verify_key_set(m_RK.empty() == false);
 
+#if defined(BOTAN_HAS_SM4_ARMV8)
+   if(CPUID::has_arm_sm4())
+      return sm4_armv8_encrypt(in, out, blocks);
+#endif
+
    for(size_t i = 0; i != blocks; ++i)
       {
       uint32_t B0 = load_be<uint32_t>(in, 0);
@@ -156,6 +163,11 @@ void SM4::decrypt_n(const uint8_t in[], uint8_t out[], size_t blocks) const
    {
    verify_key_set(m_RK.empty() == false);
 
+#if defined(BOTAN_HAS_SM4_ARMV8)
+   if(CPUID::has_arm_sm4())
+      return sm4_armv8_decrypt(in, out, blocks);
+#endif
+
    for(size_t i = 0; i != blocks; ++i)
       {
       uint32_t B0 = load_be<uint32_t>(in, 0);
@@ -219,4 +231,28 @@ void SM4::clear()
    zap(m_RK);
    }
 
+size_t SM4::parallelism() const
+   {
+#if defined(BOTAN_HAS_SM4_ARMV8)
+   if(CPUID::has_arm_sm4())
+      {
+      return 4;
+      }
+#endif
+
+   return 1;
+   }
+
+std::string SM4::provider() const
+   {
+#if defined(BOTAN_HAS_SM4_ARMV8)
+   if(CPUID::has_arm_sm4())
+      {
+      return "armv8";
+      }
+#endif
+
+   return "base";
+   }
+
 }
diff --git a/src/lib/block/sm4/sm4.h b/src/lib/block/sm4/sm4.h
index bf3fa59dc..7d1d8ba98 100644
--- a/src/lib/block/sm4/sm4.h
+++ b/src/lib/block/sm4/sm4.h
@@ -24,9 +24,17 @@ class BOTAN_PUBLIC_API(2,2) SM4 final : public Block_Cipher_Fixed_Params<16, 16>
       void clear() override;
       std::string name() const override { return "SM4"; }
       BlockCipher* clone() const override { return new SM4; }
+
+      std::string provider() const override;
+      size_t parallelism() const override;
    private:
       void key_schedule(const uint8_t[], size_t) override;
 
+#if defined(BOTAN_HAS_SM4_ARMV8)
+      void sm4_armv8_encrypt(const uint8_t in[], uint8_t out[], size_t blocks) const;
+      void sm4_armv8_decrypt(const uint8_t in[], uint8_t out[], size_t blocks) const;
+#endif
+
       secure_vector<uint32_t> m_RK;
    };
 
diff --git a/src/lib/block/sm4/sm4_armv8/info.txt b/src/lib/block/sm4/sm4_armv8/info.txt
new file mode 100644
index 000000000..7f503d361
--- /dev/null
+++ b/src/lib/block/sm4/sm4_armv8/info.txt
@@ -0,0 +1,9 @@
+<defines>
+SM4_ARMV8 -> 20180709
+</defines>
+
+need_isa armv8sm4
+
+<cc>
+gcc:8
+</cc>
diff --git a/src/lib/block/sm4/sm4_armv8/sm4_armv8.cpp b/src/lib/block/sm4/sm4_armv8/sm4_armv8.cpp
new file mode 100644
index 000000000..b91004aea
--- /dev/null
+++ b/src/lib/block/sm4/sm4_armv8/sm4_armv8.cpp
@@ -0,0 +1,168 @@
+/*
+* (C) 2018 Jack Lloyd
+*
+* Botan is released under the Simplified BSD License (see license.txt)
+*/
+
+#include <botan/sm4.h>
+#include <arm_neon.h>
+
+namespace Botan {
+
+namespace {
+
+inline uint32x4_t qswap_32(uint32x4_t B)
+   {
+   static const uint8x16_t tbl = (uint8x16_t){12, 13, 14, 15, 8, 9, 10, 11, 4, 5, 6, 7, 0, 1, 2, 3 };
+   return vreinterpretq_u32_u8(vqtbl1q_u8(vreinterpretq_u8_u32(B), tbl));
+   }
+
+inline uint32x4_t bswap_32(uint32x4_t B)
+   {
+   return vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B)));
+   }
+
+/*
+ Swap both the quad-words and bytes within each word
+ equivalent to return bswap_32(qswap_32(B))
+*/
+inline uint32x4_t bqswap_32(uint32x4_t B)
+   {
+   static const uint8x16_t tbl = (uint8x16_t){15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0 };
+   return vreinterpretq_u32_u8(vqtbl1q_u8(vreinterpretq_u8_u32(B), tbl));
+   }
+
+#define SM4_E(B0, B1, B2, B3, K) do { \
+   B0 = vsm4eq_u32(B0, K);            \
+   B1 = vsm4eq_u32(B1, K);            \
+   B2 = vsm4eq_u32(B2, K);            \
+   B3 = vsm4eq_u32(B3, K);            \
+   } while(0)
+
+}
+
+BOTAN_FUNC_ISA("+sm4")
+void SM4::sm4_armv8_encrypt(const uint8_t input8[], uint8_t output8[], size_t blocks) const
+   {
+   const uint32x4_t K0 = vld1q_u32(&m_RK[ 0]);
+   const uint32x4_t K1 = vld1q_u32(&m_RK[ 4]);
+   const uint32x4_t K2 = vld1q_u32(&m_RK[ 8]);
+   const uint32x4_t K3 = vld1q_u32(&m_RK[12]);
+   const uint32x4_t K4 = vld1q_u32(&m_RK[16]);
+   const uint32x4_t K5 = vld1q_u32(&m_RK[20]);
+   const uint32x4_t K6 = vld1q_u32(&m_RK[24]);
+   const uint32x4_t K7 = vld1q_u32(&m_RK[28]);
+
+   const uint32_t* input32 = reinterpret_cast<const uint32_t*>(reinterpret_cast<const void*>(input8));
+   uint32_t* output32 = reinterpret_cast<uint32_t*>(reinterpret_cast<void*>(output8));
+
+   while(blocks >= 4)
+      {
+      uint32x4_t B0 = bswap_32(vld1q_u32(input32));
+      uint32x4_t B1 = bswap_32(vld1q_u32(input32+4));
+      uint32x4_t B2 = bswap_32(vld1q_u32(input32+8));
+      uint32x4_t B3 = bswap_32(vld1q_u32(input32+12));
+
+      SM4_E(B0, B1, B2, B3, K0);
+      SM4_E(B0, B1, B2, B3, K1);
+      SM4_E(B0, B1, B2, B3, K2);
+      SM4_E(B0, B1, B2, B3, K3);
+      SM4_E(B0, B1, B2, B3, K4);
+      SM4_E(B0, B1, B2, B3, K5);
+      SM4_E(B0, B1, B2, B3, K6);
+      SM4_E(B0, B1, B2, B3, K7);
+
+      vst1q_u32(output32   , bqswap_32(B0));
+      vst1q_u32(output32+ 4, bqswap_32(B1));
+      vst1q_u32(output32+ 8, bqswap_32(B2));
+      vst1q_u32(output32+12, bqswap_32(B3));
+
+      input32 += 4*4;
+      output32 += 4*4;
+      blocks -= 4;
+      }
+
+   for(size_t i = 0; i != blocks; ++i)
+      {
+      uint32x4_t B = bswap_32(vld1q_u32(input32));
+
+      B = vsm4eq_u32(B, K0);
+      B = vsm4eq_u32(B, K1);
+      B = vsm4eq_u32(B, K2);
+      B = vsm4eq_u32(B, K3);
+      B = vsm4eq_u32(B, K4);
+      B = vsm4eq_u32(B, K5);
+      B = vsm4eq_u32(B, K6);
+      B = vsm4eq_u32(B, K7);
+
+      vst1q_u32(output32, bqswap_32(B));
+
+      input32 += 4;
+      output32 += 4;
+      }
+   }
+
+BOTAN_FUNC_ISA("+sm4")
+void SM4::sm4_armv8_decrypt(const uint8_t input8[], uint8_t output8[], size_t blocks) const
+   {
+   const uint32x4_t K0 = qswap_32(vld1q_u32(&m_RK[ 0]));
+   const uint32x4_t K1 = qswap_32(vld1q_u32(&m_RK[ 4]));
+   const uint32x4_t K2 = qswap_32(vld1q_u32(&m_RK[ 8]));
+   const uint32x4_t K3 = qswap_32(vld1q_u32(&m_RK[12]));
+   const uint32x4_t K4 = qswap_32(vld1q_u32(&m_RK[16]));
+   const uint32x4_t K5 = qswap_32(vld1q_u32(&m_RK[20]));
+   const uint32x4_t K6 = qswap_32(vld1q_u32(&m_RK[24]));
+   const uint32x4_t K7 = qswap_32(vld1q_u32(&m_RK[28]));
+
+   const uint32_t* input32 = reinterpret_cast<const uint32_t*>(reinterpret_cast<const void*>(input8));
+   uint32_t* output32 = reinterpret_cast<uint32_t*>(reinterpret_cast<void*>(output8));
+
+   while(blocks >= 4)
+      {
+      uint32x4_t B0 = bswap_32(vld1q_u32(input32));
+      uint32x4_t B1 = bswap_32(vld1q_u32(input32+4));
+      uint32x4_t B2 = bswap_32(vld1q_u32(input32+8));
+      uint32x4_t B3 = bswap_32(vld1q_u32(input32+12));
+
+      SM4_E(B0, B1, B2, B3, K7);
+      SM4_E(B0, B1, B2, B3, K6);
+      SM4_E(B0, B1, B2, B3, K5);
+      SM4_E(B0, B1, B2, B3, K4);
+      SM4_E(B0, B1, B2, B3, K3);
+      SM4_E(B0, B1, B2, B3, K2);
+      SM4_E(B0, B1, B2, B3, K1);
+      SM4_E(B0, B1, B2, B3, K0);
+
+      vst1q_u32(output32   , bqswap_32(B0));
+      vst1q_u32(output32+ 4, bqswap_32(B1));
+      vst1q_u32(output32+ 8, bqswap_32(B2));
+      vst1q_u32(output32+12, bqswap_32(B3));
+
+      input32 += 4*4;
+      output32 += 4*4;
+      blocks -= 4;
+      }
+
+   for(size_t i = 0; i != blocks; ++i)
+      {
+      uint32x4_t B = bswap_32(vld1q_u32(input32));
+
+      B = vsm4eq_u32(B, K7);
+      B = vsm4eq_u32(B, K6);
+      B = vsm4eq_u32(B, K5);
+      B = vsm4eq_u32(B, K4);
+      B = vsm4eq_u32(B, K3);
+      B = vsm4eq_u32(B, K2);
+      B = vsm4eq_u32(B, K1);
+      B = vsm4eq_u32(B, K0);
+
+      vst1q_u32(output32, bqswap_32(B));
+
+      input32 += 4;
+      output32 += 4;
+      }
+   }
+
+#undef SM4_E
+
+}