validate.sh

#!/bin/bash
# Validate the CUDA development environment (native, no Docker)

set -e

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
cd "$SCRIPT_DIR"

echo "=============================================="
echo "  CUDA Environment Validation"
echo "=============================================="
echo ""

echo "[1/4] GPU Detection..."
nvidia-smi --query-gpu=name,memory.total,compute_cap --format=csv
echo ""

echo "[2/4] CUDA Compiler..."
nvcc --version | head -4
echo ""

echo "[3/4] CUDA Tools..."
echo "  nvcc: $(which nvcc)"
echo "  cuda-gdb: $(which cuda-gdb 2>/dev/null || echo 'not found')"
echo "  compute-sanitizer: $(which compute-sanitizer 2>/dev/null || echo 'not found')"
echo "  ncu: $(which ncu 2>/dev/null || echo 'not found')"
echo ""

echo "[4/4] Compile & Run Test..."
cd "$SCRIPT_DIR/workspace"

# Create test file
cat > hello_test.cu << 'CUDA'
#include <stdio.h>
#include <cuda_runtime.h>

__global__ void hello() {
    if (threadIdx.x == 0) printf("Hello from GPU! Block %d\n", blockIdx.x);
}

int main() {
    int count;
    cudaGetDeviceCount(&count);
    if (count == 0) { printf("No GPU found!\n"); return 1; }
    
    cudaDeviceProp prop;
    cudaGetDeviceProperties(&prop, 0);
    printf("GPU: %s (sm_%d%d)\n", prop.name, prop.major, prop.minor);
    printf("Memory: %.1f GB\n", prop.totalGlobalMem / 1e9);
    printf("SMs: %d\n\n", prop.multiProcessorCount);
    
    hello<<<4, 32>>>();
    cudaDeviceSynchronize();
    return 0;
}
CUDA

# Detect architecture and compile
ARCH=$(nvidia-smi --query-gpu=compute_cap --format=csv,noheader | head -1 | tr -d ".")
echo "Compiling for sm_$ARCH..."
nvcc -O3 -arch=sm_${ARCH} -o hello_test hello_test.cu
./hello_test

# Cleanup
rm -f hello_test hello_test.cu

echo ""
echo "=============================================="
echo "  [PASS] All validations successful!"
echo "=============================================="