Skip to content

LLM inference in C/C++

C++ ggml

ggml: auto apply iGPU flag CUDA/HIP if integrated device (#23007)

fl0rianr30af6e2b98b0
diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index dc3e8fd626..18aaa09839 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -4994,8 +4994,14 @@ static void ggml_backend_cuda_device_get_memory(ggml_backend_dev_t dev, size_t *
}
static enum ggml_backend_dev_type ggml_backend_cuda_device_get_type(ggml_backend_dev_t dev) {
- GGML_UNUSED(dev);
- return GGML_BACKEND_DEVICE_TYPE_GPU;
+ ggml_backend_cuda_device_context * ctx = (ggml_backend_cuda_device_context *) dev->context;
+
+ cudaDeviceProp prop;
+ CUDA_CHECK(cudaGetDeviceProperties(&prop, ctx->device));
+
+ return prop.integrated
+ ? GGML_BACKEND_DEVICE_TYPE_IGPU
+ : GGML_BACKEND_DEVICE_TYPE_GPU;
}
static void ggml_backend_cuda_device_get_props(ggml_backend_dev_t dev, ggml_backend_dev_props * props) {