ggml: update ggml submodule

danbev · Oct 8, 2024 · 9a6be60 · 9a6be60
1 parent eb8d7b7
commit 9a6be60
Show file tree

Hide file tree

Showing 4 changed files with 13 additions and 16 deletions.
diff --git a/fundamentals/ggml/Makefile b/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

diff --git a/fundamentals/ggml/ggml b/fundamentals/ggml/ggml
diff --git a/fundamentals/ggml/src/backend-schedular.c b/fundamentals/ggml/src/backend-schedular.c
@@ -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);

diff --git a/fundamentals/ggml/src/backend.c b/fundamentals/ggml/src/backend.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);
   }
+1 −1		Package.swift
+18 −86		README.md
+0 −160		build.zig
+4 −0		examples/CMakeLists.txt
+2 −7		examples/gpt-2/main-backend.cpp
+2 −6		examples/gpt-2/main-batched.cpp
+2 −3		examples/gpt-2/main-sched.cpp
+109 −77		examples/mnist/mnist-common.cpp
+83 −17		examples/mnist/mnist-common.h
+9 −14		examples/mnist/mnist-eval.cpp
+8 −9		examples/mnist/mnist-train.cpp
+7 −0		examples/perf-metal/CMakeLists.txt
+152 −0		examples/perf-metal/perf-metal.cpp
+2 −7		examples/simple/simple-backend.cpp
+1 −1		include/ggml-alloc.h
+144 −62		include/ggml-backend.h
+3 −3		include/ggml-blas.h
+9 −20		include/ggml-cann.h
+16 −16		include/ggml-cuda.h
+4 −9		include/ggml-metal.h
+5 −5		include/ggml-rpc.h
+8 −8		include/ggml-sycl.h
+7 −7		include/ggml-vulkan.h
+40 −39		include/ggml.h
+2 −2		scripts/sync-llama-am.sh
+1 −1		scripts/sync-llama.last
+1 −1		scripts/sync-llama.sh
+2 −2		scripts/sync-whisper-am.sh
+1 −1		scripts/sync-whisper.last
+1 −1		scripts/sync-whisper.sh
+18 −3		src/CMakeLists.txt
+152 −71		src/ggml-backend-impl.h
+537 −271		src/ggml-backend.cpp
+8 −10		src/ggml-blas.cpp
+67 −177		src/ggml-cann.cpp
+394 −255		src/ggml-cuda.cu
+79 −0		src/ggml-cuda/argmax.cu
+3 −0		src/ggml-cuda/argmax.cuh
+12 −0		src/ggml-cuda/common.cuh
+64 −0		src/ggml-cuda/count-equal.cu
+5 −0		src/ggml-cuda/count-equal.cuh
+1 −1		src/ggml-cuda/fattn-tile-f16.cu
+3 −3		src/ggml-cuda/fattn-vec-f16.cuh
+15 −0		src/ggml-impl.h
+2 −23		src/ggml-kompute.cpp
+1,960 −1,904		src/ggml-metal.m
+26 −27		src/ggml-rpc.cpp
+45 −62		src/ggml-sycl.cpp
+8 −8		src/ggml-sycl/dequantize.hpp
+215 −267		src/ggml-vulkan.cpp
+247 −78		src/ggml.c
+0 −10		tests/CMakeLists.txt
+0 −82		tests/test-backend-buffer.cpp
+91 −13		tests/test-backend-ops.cpp
+2 −11		tests/test-conv-transpose-1d.cpp
+2 −7		tests/test-conv1d.cpp
+2 −7		tests/test-conv2d.cpp
+0 −5		tests/test-mul-mat.cpp