ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-26 17:20:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix Makefile, add GGML_USE_IQK_MULMAT ifdefs to iqk-quantize

Browse Source
This commit is contained in:
Kawrakow
2024-07-17 16:51:34 +03:00
parent 6f0805a3c7
commit d556b1d809
2 changed files with 7 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
Makefile
View File

@@ -475,6 +475,7 @@ ifdef LLAMA_BLIS
OBJS += ggml-blas.o
endif # LLAMA_BLIS
OBJS += iqk-quantize.o
ifndef LLAMA_NO_IQK_MULMAT
MK_CPPFLAGS += -DGGML_USE_IQK_MULMAT
OBJS += iqk_mul_mat.o
@@ -707,6 +708,8 @@ OBJS += ggml-alloc.o ggml-backend.o ggml-quants.o unicode.o unicode-data.o
COMMON_H_DEPS = common/common.h common/sampling.h common/log.h llama.h
COMMON_DEPS = common.o sampling.o grammar-parser.o build-info.o json-schema-to-grammar.o
iqk-quantize.o: iqk-quantize.cpp iqk-quantize.h iqk_mul_mat.h ggml-quants.h ggml-common.h ggml.h ggml-impl.h
$(CXX) $(CXXFLAGS) -c $< -o $@
ifndef LLAMA_NO_IQK_MULMAT
iqk_mul_mat.o: iqk_mul_mat.cpp ggml-impl.h ggml.h ggml-quants.h ggml-common.h iqk_mul_mat.h
$(CXX) $(CXXFLAGS) -c $< -o $@

4
iqk-quantize.cpp
View File

@@ -14,7 +14,9 @@
// limitations under the License.
#include "iqk-quantize.h"
#if GGML_USE_IQK_MULMAT
#include "iqk_mul_mat.h"
#endif
#include "ggml-quants.h"
#include "ggml-impl.h"
#define GGML_COMMON_IMPL_C
@@ -243,9 +245,11 @@ void ggml_vec_dot_iq1_bn_q8_K64(int n, float * s, size_t bs, const void * vx, si
static_assert(QK_IQ1BN == 64, "This dot product implementation for iq1_bn requires a block size of 64");
#if GGML_USE_IQK_MULMAT
if (iqk_mul_mat(GGML_TASK_TYPE_COMPUTE, 1, 1, n, GGML_TYPE_IQ1_BN, vx, 0, GGML_TYPE_Q8_K64, vy, 0, s, 0, 0, 1)) {
return;
}
#endif
const block_iq1_bn * x = (const block_iq1_bn *)vx;