Add pure kernels for 34211

2025-07-31 10:58:49 +00:00 · 2025-07-25 14:01:55 +10:00 · 2025-07-25 14:01:55 +10:00 · 4d27a0a6d8
commit 4d27a0a6d8
parent f583bbf574
3 changed files with 501 additions and 0 deletions
--- a/OpenCL/m34211_a0-pure.cl
+++ b/OpenCL/m34211_a0-pure.cl
@ -0,0 +1,157 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include M2S(INCLUDE_PATH/inc_vendor.h)
+#include M2S(INCLUDE_PATH/inc_types.h)
+#include M2S(INCLUDE_PATH/inc_platform.cl)
+#include M2S(INCLUDE_PATH/inc_common.cl)
+#include M2S(INCLUDE_PATH/inc_rp.h)
+#include M2S(INCLUDE_PATH/inc_rp.cl)
+#include M2S(INCLUDE_PATH/inc_scalar.cl)
+#endif
+
+DECLSPEC u32 MurmurHash64A_truncated (PRIVATE_AS const u32 *data, const u32 len)
+{
+#define M 0xc6a4a7935bd1e995
+#define R 47
+
+  // Initialize hash
+  u64 hash = len * M;
+
+  // Twice the number of u64 blocks
+  const u32 num_u32_blocks = (len / 8) * 2;
+
+  // Loop over one u64 at a time
+  u32 i = 0;
+  while (i < num_u32_blocks)
+  {
+    // Reconstruct u64 from two u32s
+    u64 k = hl32_to_64 (data[i + 1], data[i]);
+
+    k *= M;
+    k ^= k >> R;
+    k *= M;
+
+    hash ^= k;
+    hash *= M;
+
+    i += 2;
+  }
+
+  // Up to 7 overflow bytes
+  const u32 overflow = len & 7;
+
+  if (overflow > 4)
+  {
+    hash ^= hl32_to_64 (data[i + 1], data[i]);
+    hash *= M;
+  }
+  else if (overflow > 0)
+  {
+    hash ^= hl32_to_64 (0, data[i]);
+    hash *= M;
+  }
+
+  hash ^= hash >> R;
+  hash *= M;
+  hash ^= hash >> R;
+
+#undef M
+#undef R
+
+  // Truncate to high 4 bytes
+  return (u32) (hash >> 32);
+}
+
+KERNEL_FQ KERNEL_FA void m34211_mxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  COPY_PW (pws[gid]);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < IL_CNT; il_pos++)
+  {
+    pw_t tmp = PASTE_PW;
+
+    tmp.pw_len = apply_rules (rules_buf[il_pos].cmds, tmp.i, tmp.pw_len);
+
+    u32 hash = MurmurHash64A_truncated (tmp.i, tmp.pw_len);
+
+    const u32 z = 0;
+
+    COMPARE_M_SCALAR (hash, z, z, z);
+  }
+}
+
+KERNEL_FQ KERNEL_FA void m34211_sxx (KERN_ATTR_RULES ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+
+  /**
+   * base
+   */
+
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R1],
+    0,
+    0
+  };
+
+  /**
+   * base
+   */
+
+  COPY_PW (pws[gid]);
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < IL_CNT; il_pos++)
+  {
+    pw_t tmp = PASTE_PW;
+
+    tmp.pw_len = apply_rules (rules_buf[il_pos].cmds, tmp.i, tmp.pw_len);
+
+    u32 hash = MurmurHash64A_truncated (tmp.i, tmp.pw_len);
+
+    const u32 z = 0;
+
+    COMPARE_S_SCALAR (hash, z, z, z);
+  }
+}
--- a/OpenCL/m34211_a1-pure.cl
+++ b/OpenCL/m34211_a1-pure.cl
@ -0,0 +1,173 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+//#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include M2S(INCLUDE_PATH/inc_vendor.h)
+#include M2S(INCLUDE_PATH/inc_types.h)
+#include M2S(INCLUDE_PATH/inc_platform.cl)
+#include M2S(INCLUDE_PATH/inc_common.cl)
+#include M2S(INCLUDE_PATH/inc_scalar.cl)
+#endif
+
+DECLSPEC u32 MurmurHash64A_truncated (PRIVATE_AS const u32 *data, const u32 len)
+{
+#define M 0xc6a4a7935bd1e995
+#define R 47
+
+  // Initialize hash
+  u64 hash = len * M;
+
+  // Twice the number of u64 blocks
+  const u32 num_u32_blocks = (len / 8) * 2;
+
+  // Loop over one u64 at a time
+  u32 i = 0;
+  while (i < num_u32_blocks)
+  {
+    // Reconstruct u64 from two u32s
+    u64 k = hl32_to_64 (data[i + 1], data[i]);
+
+    k *= M;
+    k ^= k >> R;
+    k *= M;
+
+    hash ^= k;
+    hash *= M;
+
+    i += 2;
+  }
+
+  // Up to 7 overflow bytes
+  const u32 overflow = len & 7;
+
+  if (overflow > 4)
+  {
+    hash ^= hl32_to_64 (data[i + 1], data[i]);
+    hash *= M;
+  }
+  else if (overflow > 0)
+  {
+    hash ^= hl32_to_64 (0, data[i]);
+    hash *= M;
+  }
+
+  hash ^= hash >> R;
+  hash *= M;
+  hash ^= hash >> R;
+
+#undef M
+#undef R
+
+  // Truncate to high 4 bytes
+  return (u32) (hash >> 32);
+}
+
+KERNEL_FQ KERNEL_FA void m34211_mxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * base
+   */
+
+  PRIVATE_AS u8 combined_buf[256] = {0};
+  const u32 *comb_ptr = (u32*) combined_buf;
+
+  // copy left buffer
+  GLOBAL_AS const u8 *left = (GLOBAL_AS const u8*) pws[gid].i;
+  // probably bad for performance
+  for (u32 i = 0; i < pws[gid].pw_len; i++)
+  {
+    combined_buf[i] = left[i];
+  }
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < IL_CNT; il_pos++)
+  {
+    // copy right buffer
+    GLOBAL_AS const u8 *right = (GLOBAL_AS const u8*) combs_buf[il_pos].i;
+    for (u32 i = 0; i < combs_buf[il_pos].pw_len; i++)
+    {
+      combined_buf[i + pws[gid].pw_len] = right[i];
+    }
+
+    u32 hash = MurmurHash64A_truncated (comb_ptr, pws[gid].pw_len + combs_buf[il_pos].pw_len);
+
+    const u32 z = 0;
+
+    COMPARE_M_SCALAR (hash, z, z, z);
+  }
+}
+
+KERNEL_FQ KERNEL_FA void m34211_sxx (KERN_ATTR_BASIC ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * base
+   */
+
+  PRIVATE_AS u8 combined_buf[256] = {0};
+  const u32 *comb_ptr = (u32*) combined_buf;
+
+  // copy left buffer
+  GLOBAL_AS const u8 *left = (GLOBAL_AS const u8*) pws[gid].i;
+  // probably bad for performance
+  for (u32 i = 0; i < pws[gid].pw_len; i++)
+  {
+    combined_buf[i] = left[i];
+  }
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R1],
+    0,
+    0
+  };
+
+  /**
+   * loop
+   */
+
+  for (u32 il_pos = 0; il_pos < IL_CNT; il_pos++)
+  {
+    // copy right buffer
+    GLOBAL_AS const u8 *right = (GLOBAL_AS const u8*) combs_buf[il_pos].i;
+    for (u32 i = 0; i < combs_buf[il_pos].pw_len; i++)
+    {
+      combined_buf[i + pws[gid].pw_len] = right[i];
+    }
+
+    u32 hash = MurmurHash64A_truncated (comb_ptr, pws[gid].pw_len + combs_buf[il_pos].pw_len);
+
+    const u32 z = 0;
+
+    COMPARE_S_SCALAR (hash, z, z, z);
+  }
+}
--- a/OpenCL/m34211_a3-pure.cl
+++ b/OpenCL/m34211_a3-pure.cl
@ -0,0 +1,171 @@
+/**
+ * Author......: See docs/credits.txt
+ * License.....: MIT
+ */
+
+#define NEW_SIMD_CODE
+
+#ifdef KERNEL_STATIC
+#include M2S(INCLUDE_PATH/inc_vendor.h)
+#include M2S(INCLUDE_PATH/inc_types.h)
+#include M2S(INCLUDE_PATH/inc_platform.cl)
+#include M2S(INCLUDE_PATH/inc_common.cl)
+#include M2S(INCLUDE_PATH/inc_simd.cl)
+#endif
+
+DECLSPEC u32 MurmurHash64A_truncated (PRIVATE_AS const u32 *data, const u32 len)
+{
+#define M 0xc6a4a7935bd1e995
+#define R 47
+
+  // Initialize hash
+  u64 hash = len * M;
+
+  // Twice the number of u64 blocks
+  const u32 num_u32_blocks = (len / 8) * 2;
+
+  // Loop over one u64 at a time
+  u32 i = 0;
+  while (i < num_u32_blocks)
+  {
+    // Reconstruct u64 from two u32s
+    u64 k = hl32_to_64 (data[i + 1], data[i]);
+
+    k *= M;
+    k ^= k >> R;
+    k *= M;
+
+    hash ^= k;
+    hash *= M;
+
+    i += 2;
+  }
+
+  // Up to 7 overflow bytes
+  const u32 overflow = len & 7;
+
+  if (overflow > 4)
+  {
+    hash ^= hl32_to_64 (data[i + 1], data[i]);
+    hash *= M;
+  }
+  else if (overflow > 0)
+  {
+    hash ^= hl32_to_64 (0, data[i]);
+    hash *= M;
+  }
+
+  hash ^= hash >> R;
+  hash *= M;
+  hash ^= hash >> R;
+
+#undef M
+#undef R
+
+  // Truncate to high 4 bytes
+  return (u32) (hash >> 32);
+}
+
+KERNEL_FQ KERNEL_FA void m34211_mxx (KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32x w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  /**
+   * loop
+   */
+
+  u32x w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < IL_CNT; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = words_buf_r[il_pos / VECT_SIZE];
+
+    const u32x w0 = w0l | w0r;
+
+    w[0] = w0;
+
+    const u32x hash = MurmurHash64A_truncated (w, pw_len);
+
+    const u32x z = 0;
+
+    COMPARE_M_SIMD (hash, z, z, z);
+  }
+}
+
+KERNEL_FQ KERNEL_FA void m34211_sxx (KERN_ATTR_VECTOR ())
+{
+  /**
+   * modifier
+   */
+
+  const u64 lid = get_local_id (0);
+  const u64 gid = get_global_id (0);
+
+  if (gid >= GID_CNT) return;
+
+  /**
+   * digest
+   */
+
+  const u32 search[4] =
+  {
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R0],
+    digests_buf[DIGESTS_OFFSET_HOST].digest_buf[DGST_R1],
+    0,
+    0
+  };
+
+  /**
+   * base
+   */
+
+  const u32 pw_len = pws[gid].pw_len;
+
+  u32x w[64] = { 0 };
+
+  for (u32 i = 0, idx = 0; i < pw_len; i += 4, idx += 1)
+  {
+    w[idx] = pws[gid].i[idx];
+  }
+
+  /**
+   * loop
+   */
+
+  u32x w0l = w[0];
+
+  for (u32 il_pos = 0; il_pos < IL_CNT; il_pos += VECT_SIZE)
+  {
+    const u32x w0r = words_buf_r[il_pos / VECT_SIZE];
+
+    const u32x w0 = w0l | w0r;
+
+    w[0] = w0;
+
+    const u32x hash = MurmurHash64A_truncated (w, pw_len);
+
+    const u32x z = 0;
+
+    COMPARE_S_SIMD (hash, z, z, z);
+  }
+}