bitnet: CUDA, scalar, AVX2

1 files changed, 30 insertions, 0 deletions

diff --git a/ggml.c b/ggml.c
index 7ff1d0b7..e1439e2f 100644
--- a/ggml.c
+++ b/ggml.c
@@ -862,6 +862,18 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
         .vec_dot_type             = GGML_TYPE_Q8_K,
         .nrows                    = 1,
     },
+    [GGML_TYPE_IQ1_BN] = {
+        .type_name                = "iq1_bn",
+        .blck_size                = QK_IQ1BN,
+        .type_size                = sizeof(block_iq1_bn),
+        .is_quantized             = true,
+        .to_float                 = (ggml_to_float_t) dequantize_row_iq1_bn,
+        .from_float               = quantize_row_iq1_bn,
+        .from_float_reference     = (ggml_from_float_t)quantize_row_iq1_bn_reference,
+        .vec_dot                  = ggml_vec_dot_iq1_bn_q8_K64,
+        .vec_dot_type             = GGML_TYPE_Q8_K64,
+        .nrows                    = 1,
+    },
     [GGML_TYPE_IQ4_NL] = {
         .type_name                = "iq4_nl",
         .blck_size                = QK4_NL,
@@ -893,6 +905,13 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
         .is_quantized             = true,
         .from_float               = quantize_row_q8_K,
     },
+    [GGML_TYPE_Q8_K64] = {
+        .type_name                = "q8_K64",
+        .blck_size                = 64,
+        .type_size                = sizeof(block_q8_K64),
+        .is_quantized             = true,
+        .from_float               = quantize_row_q8_K64,
+    },
     [GGML_TYPE_BF16] = {
         .type_name                = "bf16",
         .blck_size                = 1,
@@ -3186,6 +3205,7 @@ enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype) {
         case GGML_FTYPE_MOSTLY_IQ3_XXS:       wtype = GGML_TYPE_IQ3_XXS;  break;
         case GGML_FTYPE_MOSTLY_IQ1_S:         wtype = GGML_TYPE_IQ1_S;    break;
         case GGML_FTYPE_MOSTLY_IQ1_M:         wtype = GGML_TYPE_IQ1_M;    break;
+        case GGML_FTYPE_MOSTLY_IQ1_BN:        wtype = GGML_TYPE_IQ1_BN;   break;
         case GGML_FTYPE_MOSTLY_IQ4_NL:        wtype = GGML_TYPE_IQ4_NL;   break;
         case GGML_FTYPE_MOSTLY_IQ4_XS:        wtype = GGML_TYPE_IQ4_XS;   break;
         case GGML_FTYPE_MOSTLY_IQ3_S:         wtype = GGML_TYPE_IQ3_S;    break;
@@ -9479,6 +9499,7 @@ static void ggml_compute_forward_add(
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
         case GGML_TYPE_IQ1_M:
+        case GGML_TYPE_IQ1_BN:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -9878,6 +9899,7 @@ static void ggml_compute_forward_add1(
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
         case GGML_TYPE_IQ1_M:
+        case GGML_TYPE_IQ1_BN:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -10007,6 +10029,7 @@ static void ggml_compute_forward_acc(
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
         case GGML_TYPE_IQ1_M:
+        case GGML_TYPE_IQ1_BN:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -12979,6 +13002,7 @@ static void ggml_compute_forward_out_prod(
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
         case GGML_TYPE_IQ1_M:
+        case GGML_TYPE_IQ1_BN:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -13172,6 +13196,7 @@ static void ggml_compute_forward_set(
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
         case GGML_TYPE_IQ1_M:
+        case GGML_TYPE_IQ1_BN:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -13447,6 +13472,7 @@ static void ggml_compute_forward_get_rows(
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
         case GGML_TYPE_IQ1_M:
+        case GGML_TYPE_IQ1_BN:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -14060,11 +14086,13 @@ static void ggml_compute_forward_clamp(
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
         case GGML_TYPE_IQ1_M:
+        case GGML_TYPE_IQ1_BN:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
         case GGML_TYPE_IQ2_S:
         case GGML_TYPE_Q8_K:
+        case GGML_TYPE_Q8_K64:
         case GGML_TYPE_I8:
         case GGML_TYPE_I16:
         case GGML_TYPE_I32:
@@ -21054,6 +21082,7 @@ void ggml_quantize_init(enum ggml_type type) {
         case GGML_TYPE_IQ1_M:   iq2xs_init_impl(type); break;
         case GGML_TYPE_IQ3_XXS: iq3xs_init_impl(256); break;
         case GGML_TYPE_IQ3_S:   iq3xs_init_impl(512); break;
+        case GGML_TYPE_IQ1_BN:  iq1bn_init_impl(); break;
         default: // nothing
             break;
     }
@@ -21122,6 +21151,7 @@ size_t ggml_quantize_chunk(
         case GGML_TYPE_IQ2_S:   result = quantize_iq2_s  (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
         case GGML_TYPE_IQ1_S:   result = quantize_iq1_s  (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
         case GGML_TYPE_IQ1_M:   result = quantize_iq1_m  (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_IQ1_BN:  result = quantize_iq1_bn (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
         case GGML_TYPE_IQ4_NL:  result = quantize_iq4_nl (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
         case GGML_TYPE_IQ4_XS:  result = quantize_iq4_xs (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
         case GGML_TYPE_F16: