ggml: update ggml submodule

Author: danbev

fundamentals/ggml/Makefile

@@ -1,4 +1,4 @@
-CC = gcc
+CC = g++
 CXX = g++
 CFLAGS = -g -Wall -Iggml/include -Iggml/src/ -Wno-unused-variable
 LDFLAGS = -Lggml/build/src/
@@ -39,7 +39,7 @@ all: $(TARGETS_C)
 ggml-quants: CFLAGS += -Iggml/src -Iggml/include/ggml
 
 $(TARGETS_C): % : $(SRCDIR)/%.c | bindir
-	$(CC) $(CFLAGS) $(LDFLAGS)  -o ${BINDIR}/$@ $< \
+	$(CXX) $(CFLAGS) $(LDFLAGS)  -o ${BINDIR}/$@ $< \
 	$(WHOLE_ARCHIVE) \
 	-lm $(CUDA_FLAGS) $(VULKAN_FLAGS)
 
@@ -51,12 +51,12 @@ $(TARGETS_CPP): % : $(SRCDIR)/%.cpp | bindir
 .PHONY: pre-quants
 pre-quants: CFLAGS += -Iggml/src -Iggml/include/ggml
 pre-quants:
-	$(CC) -E $(CFLAGS) $(LDFLAGS) -o ${BINDIR}/quants.pre $(SRCDIR)/quants.c
+	$(CXX) -E $(CFLAGS) $(LDFLAGS) -o ${BINDIR}/quants.pre $(SRCDIR)/quants.c
 
 .PHONY: pre-ggml.c
 pre-ggml.c: CFLAGS += -Iggml/src -Iggml/include/ggml
 pre-ggml.c:
-	$(CC) -E $(CFLAGS) $(LDFLAGS) -o ${BINDIR}/ggml.c.pre ggml/src/ggml.c
+	$(CXX) -E $(CFLAGS) $(LDFLAGS) -o ${BINDIR}/ggml.c.pre ggml/src/ggml.c
 
 #.PHONY: backend-cuda
 backend-cuda: CUDA_FLAGS := -L${CUDA_DIR} -lcuda -lcublas -lculibos -lcudart -lcublasLt

fundamentals/ggml/ggml

@@ -38,20 +38,20 @@ int main(int argc, char **argv) {
   }
 
   // The following will call ggml_backend_registry_init
-  size_t count = ggml_backend_reg_get_count();
+  size_t count = ggml_backend_reg_count();
   printf("Number of backends registered: %ld\n", count);
   for (size_t i = 0; i < count; i++) {
-    printf("backend %ld name: %s\n", i, ggml_backend_reg_get_name(i));
+    printf("backend %ld name: %s\n", i, ggml_backend_dev_name(ggml_backend_dev_get(i)));
   }
 
-  ggml_backend_t cpu_backend = ggml_backend_reg_init_backend_from_str("CPU");
+  ggml_backend_t cpu_backend = ggml_backend_init_by_name("CPU", NULL);
   if (cpu_backend != NULL) {
       ggml_backend_buffer_t buffer = ggml_backend_alloc_buffer(cpu_backend, 10*4);
       print_backend_info(buffer, ctx);
       ggml_backend_free(cpu_backend);
   }
 
-  ggml_backend_t cuda_backend = ggml_backend_reg_init_backend_from_str("CUDA0");
+  ggml_backend_t cuda_backend = ggml_backend_init_by_name("CUDA0", NULL);
   if (cuda_backend != NULL) {
       ggml_backend_buffer_t buffer = ggml_backend_alloc_buffer(cuda_backend, 10*4);
       print_backend_info(buffer, ctx);

fundamentals/ggml/src/backend-schedular.c

@@ -30,32 +30,30 @@ int main(int argc, char **argv) {
   struct ggml_context* ctx = ggml_init(params);
   struct ggml_tensor* x = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 10);
   ggml_set_name(x, "x");
-  printf("x backend type (0=CPU, 10=GPU): %d\n", x->backend);
   if (x->buffer == NULL) {
     printf("x backend buffer is NULL\n");
   } else {
     printf("x backend buffer: %s\n", ggml_backend_buffer_name(x->buffer));
   }
 
   // The following will call ggml_backend_registry_init
-  size_t count = ggml_backend_reg_get_count();
+  size_t count = ggml_backend_reg_count();
   printf("Number of backends registered: %ld\n", count);
   for (size_t i = 0; i < count; i++) {
-    printf("backend %ld name: %s\n", i, ggml_backend_reg_get_name(i));
+    printf("backend %ld name: %s\n", i, ggml_backend_dev_name(ggml_backend_dev_get(i)));
   }
 
-  ggml_backend_t cpu_backend = ggml_backend_reg_init_backend_from_str("CPU");
+  ggml_backend_t cpu_backend = ggml_backend_init_by_name("CPU", NULL);
   if (cpu_backend != NULL) {
       ggml_backend_buffer_t buffer = ggml_backend_alloc_buffer(cpu_backend, 10*4);
       print_backend_info(buffer, ctx);
       ggml_backend_free(cpu_backend);
   }
 
-  ggml_backend_t cuda_backend = ggml_backend_reg_init_backend_from_str("CUDA0");
+  ggml_backend_t cuda_backend = ggml_backend_init_by_name("CUDA0", NULL);
   if (cuda_backend != NULL) {
       ggml_backend_buffer_t buffer = ggml_backend_alloc_buffer(cuda_backend, 10*4);
       print_backend_info(buffer, ctx);
-      printf("x backend type (%d=CPU, %d=GPU): %d\n", GGML_BACKEND_TYPE_CPU, GGML_BACKEND_TYPE_GPU, x->backend);
       if (x->buffer != NULL) {
           printf("x backend buffer: %s\n", ggml_backend_buffer_name(x->buffer));
       }
@@ -67,7 +65,6 @@ int main(int argc, char **argv) {
 
       // The following will copy the data from the host to the device.
       ggml_backend_tensor_set(x, data, 0, 10);
-      printf("x backend type (%d=CPU, %d=GPU): %d\n", GGML_BACKEND_TYPE_CPU, GGML_BACKEND_TYPE_GPU, x->backend);
       ggml_backend_free(cuda_backend);
   }