Skip to content

LLM inference in C/C++

C++ ggml

opencl: do not use `clCreateBufferWithProperties` when targeting CL 2.x (#25673)

Hongqiang Wang12127defda4f
diff --git a/ggml/src/ggml-opencl/ggml-opencl.cpp b/ggml/src/ggml-opencl/ggml-opencl.cpp
index 7b69175921..b14ea8133b 100644
--- a/ggml/src/ggml-opencl/ggml-opencl.cpp
+++ b/ggml/src/ggml-opencl/ggml-opencl.cpp
@@ -10525,10 +10525,16 @@ static ggml_backend_buffer_t ggml_backend_opencl_buffer_type_alloc_buffer(ggml_b
cl_int err;
cl_mem mem = clCreateBuffer(backend_ctx->context, CL_MEM_READ_WRITE, size, NULL, &err);
+#if GGML_OPENCL_TARGET_VERSION >= 300
+ // clCreateBufferWithProperties and cl_mem_properties are OpenCL 3.0. Drivers older than
+ // that do not export the symbol, so a build targeting them fails to link. The large
+ // buffer extension is only ever enabled on drivers that are well past 3.0, so this path
+ // is dead there anyway.
if (err != CL_SUCCESS && backend_ctx->adreno_use_large_buffer) {
cl_mem_properties props[] = { 0x41A6 /* CL_LARGE_BUFFER_QCOM */, 1, 0 };
mem = clCreateBufferWithProperties(backend_ctx->context, props, CL_MEM_READ_WRITE, size, NULL, &err);
}
+#endif
if (err != CL_SUCCESS) {
GGML_LOG_INFO("%s: failed to allocate %.2f MiB\n", __func__, size / 1024.0 / 1024.0);