@@ -168,6 +168,10 @@ static void ggml_cuda_get_rows_switch_src0_type(
168
168
get_rows_cuda_float ((const float *) src0_d, src1_d, dst_d,
169
169
ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);
170
170
break ;
171
+ case GGML_TYPE_I32:
172
+ get_rows_cuda_float ((const int32_t *) src0_d, src1_d, dst_d,
173
+ ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);
174
+ break ;
171
175
case GGML_TYPE_BF16:
172
176
get_rows_cuda_float ((const nv_bfloat16 *) src0_d, src1_d, dst_d,
173
177
ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);
@@ -210,6 +214,10 @@ void get_rows_cuda(
210
214
ggml_cuda_get_rows_switch_src0_type (src0_d, src0_type, src1_d, (float *) dst_d,
211
215
ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);
212
216
break ;
217
+ case GGML_TYPE_I32:
218
+ ggml_cuda_get_rows_switch_src0_type (src0_d, src0_type, src1_d, (int32_t *) dst_d,
219
+ ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);
220
+ break ;
213
221
case GGML_TYPE_F16:
214
222
ggml_cuda_get_rows_switch_src0_type (src0_d, src0_type, src1_d, (half *) dst_d,
215
223
ne00, nb01, nb02, nb03, ne10, ne11, ne12, nb10, nb11, nb12, nb1, nb2, nb3, stream);
0 commit comments