nvcc

NVIDIA's CUDA compiler driver

TLDR

Compile CUDA program

$ nvcc [program.cu] -o [program]

Compile to object file

$ nvcc -c [kernel.cu] -o [kernel.o]

Compile with specific GPU architecture

$ nvcc -arch=sm_[75] [program.cu] -o [program]

Generate PTX code

$ nvcc -ptx [kernel.cu]

Compile with optimization

$ nvcc -O3 [program.cu] -o [program]

Compile with debug symbols

$ nvcc -g -G [program.cu] -o [program]

Link with external library

$ nvcc [program.cu] -o [program] -l[cublas]

Show compilation stages

$ nvcc --dryrun [program.cu]

SYNOPSIS

nvcc [-arch=arch] [-o output] [-c] [-g] [options] files

nvcc is NVIDIA's CUDA compiler driver. It compiles CUDA C/C++ code that runs on NVIDIA GPUs along with host code that runs on the CPU.Compilation separates device code (kernels running on GPU) from host code (CPU). Device code compiles to PTX intermediate representation or directly to SASS (GPU machine code).Architecture flags (-arch) target specific GPU generations. Use `-arch=native` to auto-detect visible GPUs, or `-arch=all` to compile for all supported architectures. Forward compatibility uses PTX that JIT-compiles at runtime.The compiler integrates with host compilers (gcc, clang, MSVC) for CPU code. Separate compilation allows mixing CUDA with regular C++ in large projects.Debug builds (-g -G) enable cuda-gdb debugging. Optimization levels affect both host and device code performance.CUDA libraries (cuBLAS, cuDNN, cuFFT) link like regular libraries. Header paths and library paths may need specification for non-standard installations.

PARAMETERS

-o FILE

Output file.

-c

Compile only, don't link.

-arch ARCH

GPU architecture (sm50, sm75, sm_86, etc.).

-code CODE

GPU code generation.

-gencode SPEC

Architecture/code pair (e.g., arch=compute75,code=sm75).

-ptx

Generate PTX assembly.

-g

Host debug symbols.

-G

Device debug symbols.

-O LEVEL

Optimization level (0-3).

-I DIR

Include directory.

-L DIR

Library directory.

-l LIB

Link library.

--dryrun

Show commands without executing.

-Xcompiler options

Pass options directly to the host compiler.

-std standard

C++ standard (e.g., c++14, c++17, c++20). Also accepted as `--std`.

-dc

Compile to relocatable device code (enables separate compilation).

-rdc true|false

Enable or disable relocatable device code.

-dlink

Link relocatable device code objects.

-ccbin PATH

Specify the host compiler binary (e.g., `/usr/bin/g++`).

-Xlinker options

Pass options directly to the host linker.

-lineinfo

Generate line-number information for device code (useful for profilers).

-use_fast_math

Enable fast math optimizations (implies `-ftz=true -prec-div=false -prec-sqrt=false`).

-keep

Retain intermediate compilation files.

-t N

Parallelize compilation across N threads.

-v, --verbose

Verbose output.

--version

Show version.

CAVEATS

Requires NVIDIA GPU and drivers. Architecture mismatch causes runtime errors. Debug builds much slower. Large register usage limits occupancy.

HISTORY

nvcc was introduced with CUDA by NVIDIA around 2007. It enabled general-purpose GPU computing by providing a C-like language for programming NVIDIA GPUs, transforming them from graphics-only to general computation.