[User] Building with CUDA 11.0

[Microflame]

Prerequisites

• I am running the latest code.
• I carefully followed the README.md.
• I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
• I reviewed the Discussions, and have a new bug or useful enhancement to share.

Expected Behavior

I am trying to build llama.cpp with CUDA 11.0 using the following command:

$ make LLAMA_CUBLAS=ON CUDA_DOCKER_ARCH=sm_80 -j8

Current Behavior

Currently there are several issues with CUDA 11.0:

1. nvcc fails to properly forward -march=native -mtune=native to backend compiler which results in nvcc fatal : 'arch=native': expected a number

2. In CUDA 11.0 in cudaStreamWaitEvent(...) all three arguments are required. Since CUDA 11.1 third parameter is 0 by default.

3. CUDA 11.0 does not support __builtin_assume(). According to docs it is only available since CUDA 11.1

Environment and Context

I am building in nvidia/cuda:11.0.3-devel-ubuntu20.04 docker with just 'build-essential' installed.
Host driver is 536.23
Host OS is WSL

[Microflame]

Proposed solution:
#3132