vendor-ext-composability

One pad, three ways — a proof that the Alp SDK's vendor-extension layer (<alp/ext/<vendor>/...>) is additive, never exclusive.

Claiming a vendor-specific feature on a pad does not lock that pad out of the portable surfaces. The same physical pad stays usable as a plain digital GPIO and as its portable E1M peripheral.

Source: examples/peripheral-io/vendor-ext-composability/.

The three ways

All on the Arduino A1 analog input — EVK_ADC_ARDUINO_A1, which is ALP_E1M_ADC1.

Way	Surface	Call
1	Plain GPIO	alp_gpio_open(ALP_E1M_GPIO_ADC1) — the pad's universal GPIO secondary
2	Portable ADC	alp_adc_open(EVK_ADC_ARDUINO_A1) — the cross-vendor <alp/adc.h> surface, identical on every E1M SoM
3	Vendor extension	alp_alif_adc_set_trigger_source(handle) — an Alif-only knob from <alp/ext/alif/adc.h>, layered on the Way-2 handle

Way 3 augments the portable handle; it never replaces it. The header is pulled in defensively with __has_include, so the example compiles on every SoM. On non-Alif silicon the ALP_EXT_ALIF_ADC_AVAILABLE guard is false, the example says so, and the portable Way-2 read keeps working unchanged.

Why this matters

A pad is a shared resource. The SDK guarantees that reaching for a vendor knob never strands the pad — an app can still fall back to the portable peripheral or to a bare GPIO. That guarantee is what lets you prototype against the portable surface, opt into a vendor accelerator where it pays off, and stay portable everywhere else — on the same pins, in the same firmware.

board.yaml

som:
  sku: E1M-AEN801          # ships the <alp/ext/alif/adc.h> extension Way 3 uses

preset: e1m-evk

pins:
  - { e1m: E1M_ADC1, macro: EVK_ADC_ARDUINO_A1, doc: "Arduino A1 analog input; opened as portable ADC + (on Alif) the vendor-ext trigger knob, and digitally via E1M_GPIO_ADC1" }

cores:
  m55_hp:
    app: ./src
    peripherals:
      - adc
      - gpio

diagnostics:
  log_level: info

Only the routed ADC role is named in pins: — the GPIO secondary is the same pad and needs no separate route.

Expected output

Under native_sim the host overlay wires a GPIO-emul controller, so Way 1 actually toggles the pad. It wires no ADC controller, so Ways 2 and 3 take the graceful skip path — alp_adc_open returns NULL with ALP_ERR_NOT_READY, the same NULL-tolerant contract every SDK open() honours:

[flex] vendor-ext composability: one pad (ALP_E1M_ADC1), three ways
[flex] way 1: ALP_E1M_GPIO_ADC1 driven as a digital GPIO output -- ok
[flex] way 2: ADC open skipped (last_err=-2; native_sim has no ADC)
[flex] done

On a real Alif EVK all three ways run for real.

:::note Two emul controllers in the host overlay Reaching ALP_E1M_GPIO_ADC1 needs two GPIO-emul controllers, because a single emul controller caps at 32 pins. The shipped overlay shows the canonical two-controller split a full 52-slot board file uses. :::

Status

Runs on native_sim/native/64; the harness latches on [flex] done. Way 1 executes for real there; Ways 2/3 are proven on Alif silicon.

See also

• <alp/adc.h> reference
• <alp/peripheral.h> reference — GPIO
• adc-voltmeter — the portable ADC surface on its own
• Board portability
• Examples overview