Removed GROUP/128/SSE. It wasn't compiling, and it needed an overhaul.

I'll do it someday when I'm bored.

5 files changed, 3 insertions, 303 deletions

diff --git a/include/gf_int.h b/include/gf_int.h
index ab9281d..9221569 100644
--- a/include/gf_int.h
+++ b/include/gf_int.h
@@ -144,7 +144,6 @@ typedef enum {GF_E_MDEFDIV, /* Dev != Default && Mult == Default */
               GF_E_GR_W_48, /* Mult == GROUP, w \in { 4, 8 } */
               GF_E_GR_W_16, /* Mult == GROUP, w == 16, arg1 != 4 || arg2 != 4 */
               GF_E_GR_128A, /* Mult == GROUP, w == 128, bad args */
-              GF_E_GR_SSE4, /* Mult == GROUP, w == 128, No SSE4  */
               GF_E_GR_A_27, /* Mult == GROUP, either arg > 27 */
               GF_E_GR_AR_W, /* Mult == GROUP, either arg > w  */
               GF_E_GR____J, /* Mult == GROUP, Reg == SSE|ALTMAP|NOSSE */
diff --git a/src/gf.c b/src/gf.c
index 09a668e..a443f17 100644
--- a/src/gf.c
+++ b/src/gf.c
@@ -75,7 +75,6 @@ void gf_error()
     case GF_E_GR_W_48: s = "With -m GROUP, w cannot be 4 or 8."; break;
     case GF_E_GR_W_16: s = "With -m GROUP, w == 16, arg1 and arg2 must be 4."; break;
     case GF_E_GR_128A: s = "With -m GROUP, w == 128, arg1 must be 4, and arg2 in { 4,8,16 }."; break;
-    case GF_E_GR_SSE4: s = "With -m GROUP, w == 128, you need SSE4."; break;
     case GF_E_GR_A_27: s = "With -m GROUP, arg1 and arg2 must be <= 27."; break;
     case GF_E_GR_AR_W: s = "With -m GROUP, arg1 and arg2 must be <= w."; break;
     case GF_E_GR____J: s = "Cannot use GROUP with -r ALTMAP|SSE|NOSSE."; break;
@@ -320,7 +319,6 @@ int gf_error_check(int w, int mult_type, int region_type, int divide_type,
     if (w == 16 && (arg1 != 4 || arg2 != 4))     { _gf_errno = GF_E_GR_W_16; return 0; }
     if (w == 128 && (arg1 != 4 || 
        (arg2 != 4 && arg2 != 8 && arg2 != 16))) { _gf_errno = GF_E_GR_128A; return 0; }
-    if (w == 128 && !sse4)                      { _gf_errno = GF_E_GR_SSE4; return 0; }
     if (arg1 > 27 || arg2 > 27)                 { _gf_errno = GF_E_GR_A_27; return 0; }
     if (arg1 > w || arg2 > w)                   { _gf_errno = GF_E_GR_AR_W; return 0; }
     if (raltmap || rsse || rnosse)              { _gf_errno = GF_E_GR____J; return 0; }
diff --git a/src/gf_int.h b/src/gf_int.h
index ab9281d..9221569 100644
--- a/src/gf_int.h
+++ b/src/gf_int.h
@@ -144,7 +144,6 @@ typedef enum {GF_E_MDEFDIV, /* Dev != Default && Mult == Default */
               GF_E_GR_W_48, /* Mult == GROUP, w \in { 4, 8 } */
               GF_E_GR_W_16, /* Mult == GROUP, w == 16, arg1 != 4 || arg2 != 4 */
               GF_E_GR_128A, /* Mult == GROUP, w == 128, bad args */
-              GF_E_GR_SSE4, /* Mult == GROUP, w == 128, No SSE4  */
               GF_E_GR_A_27, /* Mult == GROUP, either arg > 27 */
               GF_E_GR_AR_W, /* Mult == GROUP, either arg > w  */
               GF_E_GR____J, /* Mult == GROUP, Reg == SSE|ALTMAP|NOSSE */
diff --git a/src/gf_w128.c b/src/gf_w128.c
index c888f44..fae9f5c 100644
--- a/src/gf_w128.c
+++ b/src/gf_w128.c
@@ -984,45 +984,6 @@ void gf_w128_group_m_init(gf_t *gf, gf_val_128_t b128)
   return;
 }
 
-static
-void gf_w128_group_m_sse_init(gf_t *gf, gf_val_128_t b128)
-{
-#if defined(INTEL_SSE4)
-  int i, j;
-  int g_m;
-  uint64_t lbit, middlebit;
-  gf_internal_t *scratch;
-  gf_group_tables_t *gt;
-  scratch = (gf_internal_t *) gf->scratch;
-  gt = scratch->private;
-  g_m = scratch->arg1;
-
-  __m128i *table = (__m128i *)(gt->m_table), b, a, ubit, prim_poly;
-  prim_poly = _mm_insert_epi64(_mm_setzero_si128(), scratch->prim_poly, 0);
-  b = _mm_loadu_si128((__m128i *)(b128));
-
-  table[0] = _mm_setzero_si128();
-  table[1] = table[0];
-  table[1] = _mm_insert_epi64(table[1],b128[0],1);
-  table[1] = _mm_insert_epi64(table[1],b128[1],0);
-  lbit = 1;
-  lbit <<= 63;
-  ubit = _mm_set_epi32(0, 1, 0, 0);
-  for (i = 2; i < (1 << g_m); i <<= 1) {
-    a = table[(i >> 1)];
-    middlebit = (_mm_extract_epi64(a, 0x0) & lbit);
-    a = _mm_slli_epi64(a, 1);
-    if (middlebit) a = _mm_xor_si128(a, ubit);
-    table[i] = a;
-    if (_mm_extract_epi64(table[i >> 1], 0x1) & lbit) table[i] = _mm_xor_si128(table[i], prim_poly);
-    for (j = 0; j < i; j++) {
-      table[i + j] = _mm_xor_si128(table[i], table[j]);
-    }
-  }
-  return;
-#endif
-}
-
 void
 gf_w128_group_multiply(GFP gf, gf_val_128_t a128, gf_val_128_t b128, gf_val_128_t c128)
 {
@@ -1095,88 +1056,6 @@ gf_w128_group_multiply(GFP gf, gf_val_128_t a128, gf_val_128_t b128, gf_val_128_
   c128[1] = p_i[1];
 }
 
-void
-gf_w128_group_sse_multiply(GFP gf, gf_val_128_t a128, gf_val_128_t b128, gf_val_128_t c128)
-{
-#if defined(INTEL_SSE4) && defined(ARCH_64)
-  int i,j;
-  int i_r, i_m, t_m;
-  int mask_m, mask_r, mask_s;
-  int g_m, g_r;
-  uint32_t shiftbits;
-  uint64_t a[2], tbit = 1;
-  tbit <<= 63;
-  gf_internal_t *scratch;
-  gf_group_tables_t *gt;
-  __m128i p_i, *m_table, *r_table, zero;
-  
-  zero = _mm_setzero_si128();
-  scratch = (gf_internal_t *) gf->scratch;
-  gt = scratch->private;
-  m_table = (__m128i *)(gt->m_table);
-  r_table = (__m128i *)(gt->r_table);
-  g_m = scratch->arg1;
-  g_r = scratch->arg2;
-
-  mask_m = (1 << g_m) - 1;
-  mask_r = (1 << g_r) - 1;
-  mask_s = mask_m << (32-g_m); /*sets g_m leftmost bits to 1*/
-  if (b128[0] != _mm_extract_epi64(m_table[1], 1) || b128[1] != _mm_extract_epi64(m_table[1], 0)) {
-    gf_w128_group_m_sse_init(gf, b128);
-  } 
-
-  p_i = zero;
-  a[0] = a128[0];
-  a[1] = a128[1];
-
-  t_m = 0;
-  i_r = 0;
-
-  /* Top 64 bits */
-  for (i = ((GF_FIELD_WIDTH / 2) / g_m) - 1; i >= 0; i--) {
-    i_m = (a[0] >> (i * g_m)) & mask_m;
-    i_r ^= ((uint64_t)_mm_extract_epi64(p_i, 1) >> (64 - g_m)) & mask_r;
-     
-    shiftbits = _mm_extract_epi32(p_i, 1) & mask_s;
-    shiftbits >>= 32-g_m;
-    p_i = _mm_slli_epi64(p_i, g_m);
-    p_i = _mm_xor_si128(p_i, _mm_insert_epi32(zero, shiftbits, 2));
-
-    p_i = _mm_xor_si128(p_i, m_table[i_m]);
-    t_m += g_m;
-    if (t_m == g_r) {
-      p_i = _mm_xor_si128(p_i, r_table[i_r]);
-      t_m = 0;
-      i_r = 0;
-    } else {
-      i_r <<= g_m;
-    }
-  }
-
-  for (i = ((GF_FIELD_WIDTH / 2) / g_m) - 1; i >= 0; i--) {
-    i_m = (a[1] >> (i * g_m)) & mask_m;
-    i_r ^= (((uint64_t)_mm_extract_epi64(p_i,1)) >> (64 - g_m)) & mask_r;
-
-    shiftbits = _mm_extract_epi32(p_i, 1) & mask_s;
-    shiftbits >>= 32-g_m;
-    p_i = _mm_slli_epi64(p_i, g_m);
-    p_i = _mm_xor_si128(p_i, _mm_insert_epi32(zero, shiftbits, 2));
-    
-    p_i = _mm_xor_si128(p_i, m_table[i_m]);
-    t_m += g_m;
-    if (t_m == g_r) {
-      p_i = _mm_xor_si128(p_i, r_table[i_r]);
-      t_m = 0;
-      i_r = 0;
-    } else {
-      i_r <<= g_m;
-    }
-  }
-  c128[0] = _mm_extract_epi64(p_i, 1);
-  c128[1] = _mm_extract_epi64(p_i, 0);
-#endif
-}
-
 static
 void
 gf_w128_group_multiply_region(gf_t *gf, void *src, void *dest, gf_val_128_t val, int bytes, int xor)
@@ -1273,160 +1152,6 @@ gf_w128_group_multiply_region(gf_t *gf, void *src, void *dest, gf_val_128_t val,
   }
 }
 
-static
-void
-gf_w128_group_sse_multiply_region(gf_t *gf, void *src, void *dest, gf_val_128_t val, int bytes, int xor)
-{
-#if defined(INTEL_SSE4)
-  int i;
-  int i_r, i_m, t_m;
-  int mask_m, mask_r, mask_s;
-  int g_m, g_r;
-  uint32_t shiftbits;
-  uint64_t a[2];
-  gf_internal_t *scratch;
-  gf_group_tables_t *gt;
-  gf_region_data rd;
-  uint64_t *a128, *c128, *top;
-  __m128i *m_table, *r_table, p_i, zero;
-  zero = _mm_setzero_si128();
-  /* We only do this to check on alignment. */
-  gf_set_region_data(&rd, gf, src, dest, bytes, 0, xor, 8);
-      
-  if (val[0] == 0) {
-    if (val[1] == 0) { gf_multby_zero(dest, bytes, xor); return; }
-    if (val[1] == 1) { gf_multby_one(src, dest, bytes, xor); return; }
-  }
-    
-  scratch = (gf_internal_t *) gf->scratch;
-  gt = scratch->private;
-  m_table = (__m128i *)(gt->m_table);
-  r_table = (__m128i *)(gt->r_table);
-  g_m = scratch->arg1;
-  g_r = scratch->arg2;
-
-  mask_m = (1 << g_m) - 1;
-  mask_r = (1 << g_r) - 1;
-  mask_s = mask_m << (32-g_m);
-
-  if (val[0] != _mm_extract_epi64(m_table[1], 1) || val[1] != _mm_extract_epi64(m_table[1], 0)) {
-    gf_w128_group_m_sse_init(gf, val);
-  }
-
-  a128 = (uint64_t *) src;
-  c128 = (uint64_t *) dest;
-  top = (uint64_t *) rd.d_top;
-
-  if (xor){
-    while (c128 < top) {
-      p_i = zero;
-      a[0] = a128[0];
-      a[1] = a128[1];
-
-      t_m = 0;
-      i_r = 0;
-      /* Top 64 bits */
-      for (i = ((GF_FIELD_WIDTH / 2) / g_m) - 1; i >= 0; i--) {
-        i_m = (a[0] >> (i * g_m)) & mask_m;
-        i_r ^= ((uint64_t)_mm_extract_epi64(p_i, 1) >> (64 - g_m)) & mask_r;
-
-        shiftbits = _mm_extract_epi32(p_i, 1) & mask_s;
-        shiftbits >>= 32-g_m;
-        p_i = _mm_slli_epi64(p_i, g_m);
-        p_i = _mm_xor_si128(p_i, _mm_insert_epi32(zero, shiftbits, 2));
-        p_i = _mm_xor_si128(p_i, m_table[i_m]);
-        t_m += g_m;
-        if (t_m == g_r) {
-          p_i = _mm_xor_si128(p_i, r_table[i_r]);
-          t_m = 0;
-          i_r = 0;
-        } else {
-          i_r <<= g_m;
-        }
-      }
-
-      for (i = ((GF_FIELD_WIDTH / 2) / g_m) - 1; i >= 0; i--) {
-        i_m = (a[1] >> (i * g_m)) & mask_m;
-        i_r ^= ((uint64_t)_mm_extract_epi64(p_i, 1) >> (64 - g_m)) & mask_r;
-
-        shiftbits = _mm_extract_epi32(p_i, 1) & mask_s;
-        shiftbits >>= 32-g_m;
-        p_i = _mm_slli_epi64(p_i, g_m);
-        p_i = _mm_xor_si128(p_i, _mm_insert_epi32(zero, shiftbits, 2));
-
-        p_i = _mm_xor_si128(p_i, m_table[i_m]);
-        t_m += g_m;
-        if (t_m == g_r) {
-          p_i = _mm_xor_si128(p_i, r_table[i_r]);
-          t_m = 0;
-          i_r = 0;
-        } else {
-          i_r <<= g_m;
-        }
-      }
-
-      c128[0] ^= _mm_extract_epi64(p_i, 1);
-      c128[1] ^= _mm_extract_epi64(p_i, 0);
-      a128 += 2;
-      c128 += 2;
-    }
-  }else{
-    while (c128 < top) {
-      p_i = zero;
-      a[0] = a128[0];
-      a[1] = a128[1];
-
-      t_m = 0;
-      i_r = 0;
-      /* Top 64 bits */
-      for (i = ((GF_FIELD_WIDTH / 2) / g_m) - 1; i >= 0; i--) {
-        i_m = (a[0] >> (i * g_m)) & mask_m;
-        i_r ^= ((uint64_t)_mm_extract_epi64(p_i, 1) >> (64 - g_m)) & mask_r;
-
-        shiftbits = _mm_extract_epi32(p_i, 1) & mask_s;
-        shiftbits >>= 32-g_m;
-        p_i = _mm_slli_epi64(p_i, g_m);
-        p_i = _mm_xor_si128(p_i, _mm_insert_epi32(zero, shiftbits, 2));
-        p_i = _mm_xor_si128(p_i, m_table[i_m]);
-        t_m += g_m;
-        if (t_m == g_r) {
-          p_i = _mm_xor_si128(p_i, r_table[i_r]);
-          t_m = 0;
-          i_r = 0;
-        } else {
-          i_r <<= g_m;
-        }
-      }
-
-      for (i = ((GF_FIELD_WIDTH / 2) / g_m) - 1; i >= 0; i--) {
-        i_m = (a[1] >> (i * g_m)) & mask_m;
-        i_r ^= ((uint64_t)_mm_extract_epi64(p_i, 1) >> (64 - g_m)) & mask_r;
-
-        shiftbits = _mm_extract_epi32(p_i, 1) & mask_s;
-        shiftbits >>= 32-g_m;
-        p_i = _mm_slli_epi64(p_i, g_m);
-        p_i = _mm_xor_si128(p_i, _mm_insert_epi32(zero, shiftbits, 2));
-
-        p_i = _mm_xor_si128(p_i, m_table[i_m]);
-        t_m += g_m;
-        if (t_m == g_r) {
-          p_i = _mm_xor_si128(p_i, r_table[i_r]);
-          t_m = 0;
-          i_r = 0;
-        } else {
-          i_r <<= g_m;
-        }
-      }
-
-      c128[0] = _mm_extract_epi64(p_i, 1);
-      c128[1] = _mm_extract_epi64(p_i, 0);
-      a128 += 2;
-      c128 += 2;
-    }
-  }
-#endif
-}
-
 /* a^-1 -> b */
   void
 gf_w128_euclid(GFP gf, gf_val_128_t a128, gf_val_128_t b128)
@@ -1880,28 +1605,7 @@ int gf_w128_group_init(gf_t *gf)
   gf->inverse.w128 = gf_w128_euclid;
   gf->multiply_region.w128 = gf_w128_group_multiply_region;
 
-  /* JSP: I've got a problem compiling here -- something about "vmovq", and
-     I don't have the time to chase it down right now. */
-
-  #if defined(INTEL_SSE4) && defined(ARCH_64)
-    if(!(scratch->region_type & GF_REGION_NOSSE))
-    {
-      if ((g_m != 4) && ((g_r != 4) || (g_r != 8)))
-        return 0;
-      gt->r_table = (void *)(((uint64_t)gt->r_table + 15) & (~0xfULL)); /* aligns gt->r_table on a 16-bit boundary*/
-      gt->m_table = gt->r_table + 2*size_r;
-      gt->m_table[2] = 0;
-      gt->m_table[3] = 0;
-      gf->multiply.w128 = gf_w128_group_sse_multiply;
-      gf->multiply_region.w128 = gf_w128_group_sse_multiply_region;
-      gf_w128_group_r_sse_init(gf);
-    }
-    else
-      gf_w128_group_r_init(gf);
-  #else
-    if(scratch->region_type & GF_REGION_SSE) return 0;
-    else gf_w128_group_r_init(gf);
-  #endif
+  gf_w128_group_r_init(gf);
 
   return 1;
 }
diff --git a/tools/gf_methods.c b/tools/gf_methods.c
index 903d8b0..3afb438 100644
--- a/tools/gf_methods.c
+++ b/tools/gf_methods.c
@@ -20,8 +20,8 @@
 #define BNMULTS (8)
 static char *BMULTS[BNMULTS] = { "CARRY_FREE", "GROUP48", 
                                "TABLE", "LOG", "SPLIT4", "SPLIT8", "SPLIT88", "COMPOSITE" };
-#define NMULTS (15)
-static char *MULTS[NMULTS] = { "CARRY_FREE", "GROUP44", "GROUP48", "BYTWO_p", "BYTWO_b",
+#define NMULTS (16)
+static char *MULTS[NMULTS] = { "SHIFT", "CARRY_FREE", "GROUP44", "GROUP48", "BYTWO_p", "BYTWO_b",
                                "TABLE", "LOG", "LOG_ZERO", "LOG_ZERO_EXT", "SPLIT2",
                                "SPLIT4", "SPLIT8", "SPLIT16", "SPLIT88", "COMPOSITE" };