strix-halo-optimizations/lib/detect.sh

#!/usr/bin/env bash
# Hardware and configuration detection for Strix Halo

# Find the amdgpu DRM card path
find_gpu_card() {
    local card
    for card in /sys/class/drm/card*/device/vendor; do
        if [[ -f "$card" ]] && [[ "$(cat "$card")" == "0x1002" ]]; then
            echo "$(dirname "$card")"
            return 0
        fi
    done
    # Fallback: try any card with mem_info_vram_total (i.e., an amdgpu device)
    for card in /sys/class/drm/card*/device/mem_info_vram_total; do
        if [[ -f "$card" ]]; then
            echo "$(dirname "$card")"
            return 0
        fi
    done
    echo "/sys/class/drm/card1/device"  # last resort
}

GPU_SYSFS="$(find_gpu_card)"

# --- CPU ---
detect_cpu_model()   { grep -m1 'model name' /proc/cpuinfo | cut -d: -f2 | xargs; }
detect_cpu_cores()   { grep -c '^processor' /proc/cpuinfo; }
detect_cpu_physical() { grep 'cpu cores' /proc/cpuinfo | head -1 | cut -d: -f2 | xargs; }

# --- GPU ---
detect_gpu_name() {
    lspci | grep -i 'Display\|VGA' | grep -i 'AMD' | head -1 | sed 's/.*: //'
}

detect_gpu_device_id() {
    cat "$GPU_SYSFS/device" 2>/dev/null | sed 's/^0x//'
}

# --- Memory (bytes) ---
detect_vram_total() { cat "$GPU_SYSFS/mem_info_vram_total" 2>/dev/null || echo 0; }
detect_vram_used()  { cat "$GPU_SYSFS/mem_info_vram_used" 2>/dev/null || echo 0; }
detect_gtt_total()  { cat "$GPU_SYSFS/mem_info_gtt_total" 2>/dev/null || echo 0; }
detect_gtt_used()   { cat "$GPU_SYSFS/mem_info_gtt_used" 2>/dev/null || echo 0; }

detect_system_ram_kb() {
    local kb
    kb="$(grep MemTotal /proc/meminfo 2>/dev/null | awk '{print $2}')"
    echo "${kb:-0}"
}
detect_system_ram_bytes() { echo $(( $(detect_system_ram_kb) * 1024 )); }

# --- Kernel ---
detect_kernel_version() { uname -r; }

detect_kernel_param() {
    # Returns the value of a kernel param, or empty if not present
    local param="$1"
    local cmdline
    cmdline="$(cat /proc/cmdline)"
    # Escape dots for regex and anchor with word boundary (space or start-of-string)
    local pattern="${param//./\\.}"
    if [[ "$cmdline" =~ (^|[[:space:]])${pattern}=([^ ]+) ]]; then
        echo "${BASH_REMATCH[2]}"
    elif [[ "$cmdline" =~ (^|[[:space:]])${pattern}([[:space:]]|$) ]]; then
        echo "present"
    fi
}

detect_has_iommu_pt() {
    local val
    val="$(detect_kernel_param 'iommu')"
    [[ "$val" == "pt" ]]
}

detect_gttsize_param() { detect_kernel_param 'amdgpu.gttsize'; }
detect_pages_limit_param() { detect_kernel_param 'ttm.pages_limit'; }

# --- Tuned ---
detect_tuned_profile() {
    if is_cmd tuned-adm; then
        tuned-adm active 2>/dev/null | sed 's/Current active profile: //'
    else
        echo "tuned not installed"
    fi
}

# --- Firmware ---
detect_firmware_version() {
    rpm -q linux-firmware 2>/dev/null | sed 's/linux-firmware-//' | sed 's/\.fc.*//' || echo "unknown"
}

detect_firmware_bad() {
    # Returns 0 (true) if firmware is the known-bad version
    local fw
    fw="$(detect_firmware_version)"
    [[ "$fw" == *"20251125"* ]]
}

# --- ROCm ---
detect_rocm_version() {
    if [[ -f /opt/rocm/.info/version ]]; then
        cat /opt/rocm/.info/version
    else
        rpm -qa 2>/dev/null | grep '^rocm-core-' | head -1 | sed 's/rocm-core-//' | sed 's/-.*//' || echo "not installed"
    fi
}

detect_rocm_packages() {
    rpm -qa 2>/dev/null | grep -i rocm | sort
}

# --- Vulkan ---
detect_vulkan_driver() {
    if is_cmd vulkaninfo; then
        vulkaninfo --summary 2>/dev/null | grep 'driverName' | head -1 | awk '{print $NF}'
    else
        echo "vulkaninfo not available"
    fi
}

detect_vulkan_version() {
    if is_cmd vulkaninfo; then
        vulkaninfo --summary 2>/dev/null | grep 'apiVersion' | head -1 | awk '{print $NF}'
    fi
}

# --- Toolbox containers ---
detect_toolboxes() {
    if is_cmd toolbox; then
        toolbox list --containers 2>/dev/null | tail -n +2
    fi
}

detect_toolbox_names() {
    detect_toolboxes | awk '{print $2}' 2>/dev/null
}

# --- LLM stacks ---
detect_stack_ollama()   { is_cmd ollama && echo "installed" || echo "missing"; }
detect_stack_lmstudio() { is_cmd lms && echo "installed" || echo "missing"; }
detect_stack_llamacpp() { (is_cmd llama-cli || is_cmd llama-bench) && echo "installed" || echo "missing"; }
detect_stack_opencode() { is_cmd opencode && echo "installed" || echo "missing"; }

# --- Sensors ---
detect_gpu_temp() {
    # Returns temperature in millidegrees C
    local hwmon
    for hwmon in "$GPU_SYSFS"/hwmon/hwmon*/temp1_input; do
        if [[ -f "$hwmon" ]]; then
            cat "$hwmon"
            return
        fi
    done
    echo 0
}

detect_gpu_power() {
    # Returns power in microwatts
    local hwmon
    for hwmon in "$GPU_SYSFS"/hwmon/hwmon*/power1_average; do
        if [[ -f "$hwmon" ]]; then
            cat "$hwmon"
            return
        fi
    done
    echo 0
}

detect_gpu_busy() {
    cat "$GPU_SYSFS/gpu_busy_percent" 2>/dev/null || echo 0
}

# --- Total physical memory (visible + VRAM dedicated) ---
detect_total_physical_ram_kb() {
    local visible_kb vram_bytes vram_kb
    visible_kb="$(detect_system_ram_kb)"
    vram_bytes="$(detect_vram_total)"
    vram_kb=$(( vram_bytes / 1024 ))
    echo $(( visible_kb + vram_kb ))
}

# --- Recommended values for this system ---
recommended_gttsize_mib() {
    # Total physical RAM (including VRAM allocation) minus 4 GiB reserve, in MiB
    local total_kb
    total_kb="$(detect_total_physical_ram_kb)"
    local total_gib=$(( total_kb / 1024 / 1024 ))
    local gtt_gib=$(( total_gib - 4 ))
    echo $(( gtt_gib * 1024 ))
}

recommended_pages_limit() {
    # GTT GiB * 1024 MiB/GiB * 256 pages/MiB
    local gtt_mib
    gtt_mib="$(recommended_gttsize_mib)"
    echo $(( gtt_mib * 256 ))
}