4 files changed, 31 insertions, 1 deletions
diff --git a/ggml/src/ggml-cuda/common.cuh b/ggml/src/ggml-cuda/common.cuh
index c18e865a..07a53bcd 100644
--- a/ggml/src/ggml-cuda/common.cuh
+++ b/ggml/src/ggml-cuda/common.cuh
@@ -705,6 +705,13 @@ struct ggml_cuda_type_traits<GGML_TYPE_IQ5_K> {
 };
 
 template<>
+struct ggml_cuda_type_traits<GGML_TYPE_IQ6_K> {
+    static constexpr int qk = QK_K;
+    static constexpr int qr = QR6_XS;
+    static constexpr int qi = QI6_XS;
+};
+
+template<>
 struct ggml_cuda_type_traits<GGML_TYPE_IQ3_S> {
     static constexpr int qk = QK_K;
     static constexpr int qr = QR3_S;
diff --git a/ggml/src/ggml-cuda/iqk_mmvq.cu b/ggml/src/ggml-cuda/iqk_mmvq.cu
index 8def1547..ae5e6a3c 100644
--- a/ggml/src/ggml-cuda/iqk_mmvq.cu
+++ b/ggml/src/ggml-cuda/iqk_mmvq.cu
@@ -251,6 +251,16 @@ __device__ __forceinline__ float vec_dot_iq5_k_q8_1(
     return d5 * (__low2float(bq8_1[2*(i4/2)+0].ds) * sumi1 * ls1 + __low2float(bq8_1[2*(i4/2)+1].ds) * sumi2 * ls2);
 }
 
+#define VDR_IQ6_K_Q8_1_MMVQ 4
+#define VDR_IQ6_K_Q8_1_MMQ  4
+
+// TODO
+__device__ __forceinline__ float vec_dot_iq6_k_q8_1(
+    const void * __restrict__ vbq, const block_q8_1 * __restrict__ bq8_1, const int & kbx, const int & iqs) {
+
+    return 0;
+}
+
 static const __device__ uint32_t iq2k_table[512] = {
     0xe1e1e1e1, 0xe1e1e1f3, 0xe1e1e101, 0xe1e1e111, 0xe1e1f3e1, 0xe1e1f3f3, 0xe1e1f301, 0xe1e1f311,
     0xe1e101e1, 0xe1e101f3, 0xe1e10101, 0xe1e10111, 0xe1e111e1, 0xe1e111f3, 0xe1e11101, 0xe1e11111,
@@ -534,10 +544,16 @@ void mul_mat_vec_iq5_k_q8_1_cuda(
     iqk_mul_mat_vec_q_cuda<GGML_TYPE_IQ5_K, VDR_IQ5_K_Q8_1_MMVQ, vec_dot_iq5_k_q8_1>(vx, vy, dst, ncols_x, nrows_x, nrows_y, ncols_y, nrows_dst, stream);
 }
 
+void mul_mat_vec_iq6_k_q8_1_cuda(
+    const void * vx, const void * vy, float * dst,
+    const int ncols_x, const int nrows_x, const int nrows_y, const int ncols_y, const int nrows_dst, cudaStream_t stream) {
+
+    iqk_mul_mat_vec_q_cuda<GGML_TYPE_IQ6_K, VDR_IQ6_K_Q8_1_MMVQ, vec_dot_iq6_k_q8_1>(vx, vy, dst, ncols_x, nrows_x, nrows_y, ncols_y, nrows_dst, stream);
+}
+
 void mul_mat_vec_iq2_tn_q8_1_cuda(
     const void * vx, const void * vy, float * dst,
     const int ncols_x, const int nrows_x, const int nrows_y, const int ncols_y, const int nrows_dst, cudaStream_t stream) {
 
     iqk_mul_mat_vec_q_cuda<GGML_TYPE_IQ2_TN, VDR_IQ2_TN_Q8_1_MMVQ, vec_dot_iq2_tn_q8_1>(vx, vy, dst, ncols_x, nrows_x, nrows_y, ncols_y, nrows_dst, stream);
 }
-
diff --git a/ggml/src/ggml-cuda/iqk_mmvq.cuh b/ggml/src/ggml-cuda/iqk_mmvq.cuh
index 3dc5f41c..7af8e570 100644
--- a/ggml/src/ggml-cuda/iqk_mmvq.cuh
+++ b/ggml/src/ggml-cuda/iqk_mmvq.cuh
@@ -16,6 +16,10 @@ void mul_mat_vec_iq5_k_q8_1_cuda(
     const void * vx, const void * vy, float * dst,
     const int ncols_x, const int nrows_x, const int nrows_y, const int ncols_y, const int nrows_dst, cudaStream_t stream);
 
+void mul_mat_vec_iq6_k_q8_1_cuda(
+    const void * vx, const void * vy, float * dst,
+    const int ncols_x, const int nrows_x, const int nrows_y, const int ncols_y, const int nrows_dst, cudaStream_t stream);
+
 void mul_mat_vec_iq2_tn_q8_1_cuda(
     const void * vx, const void * vy, float * dst,
     const int ncols_x, const int nrows_x, const int nrows_y, const int ncols_y, const int nrows_dst, cudaStream_t stream);
diff --git a/ggml/src/ggml-cuda/mmvq.cu b/ggml/src/ggml-cuda/mmvq.cu
index 428d822f..9eb3fa4f 100644
--- a/ggml/src/ggml-cuda/mmvq.cu
+++ b/ggml/src/ggml-cuda/mmvq.cu
@@ -447,6 +447,9 @@ void ggml_cuda_op_mul_mat_vec_q(
         case GGML_TYPE_IQ5_K:
             mul_mat_vec_iq5_k_q8_1_cuda(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, src1_padded_row_size, src1_ncols, nrows_dst, stream);
             break;
+        case GGML_TYPE_IQ6_K:
+            mul_mat_vec_iq6_k_q8_1_cuda(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, src1_padded_row_size, src1_ncols, nrows_dst, stream);
+            break;
         case GGML_TYPE_IQ3_S:
             mul_mat_vec_iq3_s_q8_1_cuda(src0_dd_i, src1_ddq_i, dst_dd_i, ne00, row_diff, src1_padded_row_size, src1_ncols, nrows_dst, stream);
             break;