Update CMakeLists.txt

Author: Nexesenex

CMakeLists.txt

@@ -156,6 +156,8 @@ if (LLAMA_CUBLAS)
             list(APPEND GGML_SOURCES_CUDA ${SRCS})
             file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q5_1.cu")
             list(APPEND GGML_SOURCES_CUDA ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q6_0.cu")
+            list(APPEND GGML_SOURCES_CUDA ${SRCS})
             file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q2_k.cu")
             list(APPEND GGML_SOURCES_CUDA ${SRCS})
             file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q3_k.cu")
@@ -332,13 +334,60 @@ if (LLAMA_HIPBLAS)
         list(APPEND GGML_SOURCES_ROCM "ggml/src/ggml-cuda/ggml-cuda.cu")
         file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/fattn-wmma*.cu")
         list(APPEND GGML_SOURCES_ROCM ${SRCS})
-        file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq*.cu")
-        list(APPEND GGML_SOURCES_ROCM ${SRCS})
         add_compile_definitions(GGML_USE_HIP GGML_USE_CUDA SD_USE_CUBLAS)
         add_library(ggml-rocm ${GGML_SOURCES_CUDA})
         if (LLAMA_CUDA_FORCE_DMMV)
             target_compile_definitions(ggml-rocm PUBLIC GGML_CUDA_FORCE_DMMV)
         endif()
+		
+        if (LLAMA_CUDA_DISABLE_MMQ_IQ1_S_Q4_1)
+            # all quants necessary for Kobold CPP Frankenstein are compiled
+            # the other are ignored but not deleted from the ggml_cuda templates directory
+            # file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-iq1_s.cu")
+            # list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-iq1_m.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_xxs.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_xs.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_s.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-iq3_xxs.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-iq3_s.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-iq4_xs.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-iq4_nl.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q4_0.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            # file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q4_1.cu")
+            # list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q5_0.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q5_1.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q6_0.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q2_k.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q3_k.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q4_k.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q5_k.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q6_k.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq-instance-q8_0.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+        else ()
+            # Build All MMQ Kernels
+            file(GLOB SRCS "ggml/src/ggml-cuda/template-instances/mmq*.cu")
+            list(APPEND GGML_SOURCES_ROCM ${SRCS})
+        endif()
 
         if (LLAMA_CUDA_FA_ALL_QUANTS)
             # all quants necessary for Kobold CPP Frankenstein are compiled