Merge branch 'ggerganov:master' into master

Author: apicalshark

ggml/src/ggml-cpu/ggml-cpu.cpp

@@ -534,9 +534,6 @@ static ggml_backend_feature * ggml_backend_cpu_get_features(ggml_backend_reg_t r
         if (ggml_cpu_has_dotprod()) {
             features.push_back({ "DOTPROD", "1" });
         }
-        if (ggml_cpu_has_matmul_int8()) {
-            features.push_back({ "MATMUL_INT8", "1" });
-        }
         if (ggml_cpu_get_sve_cnt() > 0) {
             static std::string sve_cnt = std::to_string(ggml_cpu_get_sve_cnt());
             features.push_back({ "SVE_CNT", sve_cnt.c_str() });

ggml/src/ggml-cuda/common.cuh

@@ -165,11 +165,11 @@ static const char * cu_get_error_str(CUresult err) {
 #define CU_CHECK(err) CUDA_CHECK_GEN(err, CUDA_SUCCESS, cu_get_error_str)
 #endif
 
-#if CUDART_VERSION >= 11100 || defined(GGML_USE_MUSA)
+#if CUDART_VERSION >= 11010 || defined(GGML_USE_MUSA)
 #define GGML_CUDA_ASSUME(x) __builtin_assume(x)
 #else
 #define GGML_CUDA_ASSUME(x)
-#endif // CUDART_VERSION >= 11100
+#endif // CUDART_VERSION >= 11010
 
 #ifdef GGML_CUDA_F16
 typedef half dfloat; // dequantize float

ggml/src/ggml-cuda/ggml-cuda.cu

@@ -2840,7 +2840,7 @@ bool ggml_backend_cuda_register_host_buffer(void * buffer, size_t size) {
         return false;
     }
 
-#if CUDART_VERSION >= 11100 || defined(GGML_USE_MUSA)
+#if CUDART_VERSION >= 11010 || defined(GGML_USE_MUSA)
     cudaError_t err = cudaHostRegister(buffer, size, cudaHostRegisterPortable | cudaHostRegisterReadOnly);
     if (err != cudaSuccess) {
         // clear the error
@@ -2852,8 +2852,10 @@ bool ggml_backend_cuda_register_host_buffer(void * buffer, size_t size) {
     }
     return true;
 #else
+    GGML_UNUSED(buffer);
+    GGML_UNUSED(size);
     return false;
-#endif
+#endif // CUDART_VERSION >= 11010 || defined(GGML_USE_MUSA)
 }
 
 void ggml_backend_cuda_unregister_host_buffer(void * buffer) {

ggml/src/ggml-cuda/sum.cu

@@ -1,6 +1,6 @@
-#if !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA) && CUDART_VERSION >= 11700
+#if !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA) && CUDART_VERSION >= 11070
 #define USE_CUB
-#endif // !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA) && CUDART_VERSION >= 11700
+#endif // !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA) && CUDART_VERSION >= 11070
 
 #ifdef USE_CUB
 #include <cub/cub.cuh>