micropython.cmake

# MicroPython CMake glue for audiodsp (esp32, rp2, …).
# For Make-based ports (unix, windows), see micropython.mk in this dir.
#
# Point USER_C_MODULES at this repo (or this file) directly, e.g.:
#   idf.py build -DUSER_C_MODULES=<path to audiodsp>
# Or let the parent workspace's own aggregator discover it alongside
# other usermods.

set(MPAUDIO_MOD_DIR ${CMAKE_CURRENT_LIST_DIR})
set(MPAUDIO_SRC_DIR ${MPAUDIO_MOD_DIR}/src)

add_library(usermod_mpaudio INTERFACE)

# --- cp_compat: see micropython.mk for provenance/rationale (identical
#     source list, kept in sync by hand -- both build flavors compile the
#     same C files).
target_sources(usermod_mpaudio INTERFACE
    ${MPAUDIO_SRC_DIR}/cp_compat/audiodsp_build.c
    ${MPAUDIO_SRC_DIR}/cp_compat/argcheck.c
    ${MPAUDIO_SRC_DIR}/cp_compat/enum.c
    ${MPAUDIO_SRC_DIR}/cp_compat/context_manager_helpers.c
    ${MPAUDIO_SRC_DIR}/cp_compat/proto.c
    ${MPAUDIO_SRC_DIR}/cp_compat/util.c
    ${MPAUDIO_SRC_DIR}/cp_compat/objproperty.c
    ${MPAUDIO_SRC_DIR}/cp_compat/namedtuple.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_sample.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_port.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_pump_lock.c
    ${MPAUDIO_SRC_DIR}/audiopump/audiopump.c
    ${MPAUDIO_SRC_DIR}/audiopump/audiopump_ring.c
    ${MPAUDIO_SRC_DIR}/audiopump/audiopump_events.c
    ${MPAUDIO_SRC_DIR}/audiopump/audiopump_tap.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_rawsample.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_synth_dsp.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_envelope.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_distortion.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_biquad.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_echo.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_phaser.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_chorus.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_flanger.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_granular_pitch_shift.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_multitap.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_pitchshift.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_freeverb.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_dynamics.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_splitter.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_midside.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_remix.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_cast.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_multiply.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_suboctave.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_feedback_delay.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_filter_f32.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_shaper.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_samplehold.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_ladder.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_trig.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_fft.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_convolve.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_tank.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_modal.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_oscillator.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_sampler.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_meter.c
    ${MPAUDIO_SRC_DIR}/audiometer/module.c
    ${MPAUDIO_SRC_DIR}/mpaudio_modules.c
    ${MPAUDIO_SRC_DIR}/audiocore/__init__.c
    ${MPAUDIO_SRC_DIR}/audiocore/RawSample.c
    ${MPAUDIO_SRC_DIR}/audiocore/WaveFile.c
    ${MPAUDIO_SRC_DIR}/audiocore/module.c
    ${MPAUDIO_SRC_DIR}/synthio/__init__.c
    ${MPAUDIO_SRC_DIR}/synthio/Math.c
    ${MPAUDIO_SRC_DIR}/synthio/LFO.c
    ${MPAUDIO_SRC_DIR}/synthio/Biquad.c
    ${MPAUDIO_SRC_DIR}/synthio/Note.c
    ${MPAUDIO_SRC_DIR}/synthio/Synthesizer.c
    ${MPAUDIO_SRC_DIR}/synthio/MidiTrack.c
    ${MPAUDIO_SRC_DIR}/audiomixer/Mixer.c
    ${MPAUDIO_SRC_DIR}/audiomixer/MixerVoice.c
    ${MPAUDIO_SRC_DIR}/audiomixer/module.c
    ${MPAUDIO_SRC_DIR}/audiospeed/Resampler.c
    ${MPAUDIO_SRC_DIR}/audiospeed/SpeedChanger.c
    ${MPAUDIO_SRC_DIR}/audiospeed/module.c
    ${MPAUDIO_SRC_DIR}/audiofreeverb/Freeverb.c
    ${MPAUDIO_SRC_DIR}/audiofreeverb/module.c
    ${MPAUDIO_SRC_DIR}/audiofilters/__init__.c
    ${MPAUDIO_SRC_DIR}/audiofilters/Filter.c
    ${MPAUDIO_SRC_DIR}/audiofilters/Distortion.c
    ${MPAUDIO_SRC_DIR}/audiofilters/Phaser.c
    ${MPAUDIO_SRC_DIR}/audiofilters/module.c
    ${MPAUDIO_SRC_DIR}/audiodelays/Chorus.c
    ${MPAUDIO_SRC_DIR}/audiodelays/Echo.c
    ${MPAUDIO_SRC_DIR}/audiodelays/Flanger.c
    ${MPAUDIO_SRC_DIR}/audiodelays/GranularPitchShift.c
    ${MPAUDIO_SRC_DIR}/audiodelays/MultiTapDelay.c
    ${MPAUDIO_SRC_DIR}/audiodelays/PitchShift.c
    ${MPAUDIO_SRC_DIR}/audiodelays/module.c
    ${MPAUDIO_SRC_DIR}/audiodynamics/Dynamics.c
    ${MPAUDIO_SRC_DIR}/audiodynamics/module.c
    ${MPAUDIO_SRC_DIR}/audioroute/MidSide.c
    ${MPAUDIO_SRC_DIR}/audioroute/Port.c
    ${MPAUDIO_SRC_DIR}/audioroute/Splitter.c
    ${MPAUDIO_SRC_DIR}/audioroute/SplitterTap.c
    ${MPAUDIO_SRC_DIR}/audioroute/module.c
    ${MPAUDIO_SRC_DIR}/audiomath/Multiply.c
    ${MPAUDIO_SRC_DIR}/audiomath/SubOctave.c
    ${MPAUDIO_SRC_DIR}/audiomath/module.c
    ${MPAUDIO_SRC_DIR}/audioecho/FeedbackDelay.c
    ${MPAUDIO_SRC_DIR}/audioecho/module.c
    ${MPAUDIO_SRC_DIR}/audioshaper/Waveshaper.c
    ${MPAUDIO_SRC_DIR}/audioshaper/SampleHold.c
    ${MPAUDIO_SRC_DIR}/audioshaper/module.c
    ${MPAUDIO_SRC_DIR}/audioladder/Ladder.c
    ${MPAUDIO_SRC_DIR}/audioladder/module.c
    ${MPAUDIO_SRC_DIR}/audiooscillator/Oscillator.c
    ${MPAUDIO_SRC_DIR}/audiooscillator/module.c
    ${MPAUDIO_SRC_DIR}/audioconvolve/Convolver.c
    ${MPAUDIO_SRC_DIR}/audioconvolve/module.c
    ${MPAUDIO_SRC_DIR}/audiobiquad/Biquad.c
    ${MPAUDIO_SRC_DIR}/audiobiquad/AllPass.c
    ${MPAUDIO_SRC_DIR}/audiobiquad/module.c
    ${MPAUDIO_SRC_DIR}/audioverb/Tank.c
    ${MPAUDIO_SRC_DIR}/audioverb/module.c
    ${MPAUDIO_SRC_DIR}/audiomodal/Bank.c
    ${MPAUDIO_SRC_DIR}/audiomodal/module.c
    ${MPAUDIO_SRC_DIR}/audiosampler/SampleBuffer.c
    ${MPAUDIO_SRC_DIR}/audiosampler/Sampler.c
    ${MPAUDIO_SRC_DIR}/audiosampler/SamplerVoice.c
    ${MPAUDIO_SRC_DIR}/audiosampler/module.c
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_graph.c
    ${MPAUDIO_SRC_DIR}/audiograph/Graph.c
    ${MPAUDIO_SRC_DIR}/audiograph/module.c
)

target_include_directories(usermod_mpaudio INTERFACE ${MPAUDIO_SRC_DIR})

# --- which pump lock ------------------------------------------------------
#
# Nothing to say any more, and that is the change. This used to define
# AUDIODSP_PUMP_LOCK_FREERTOS here because the macro that would look obvious --
# ESP_PLATFORM -- is NOT defined for a user C module, and the POSIX branch
# compiled and LINKED on esp32 anyway (IDF's newlib has pthread.h). The spike
# lost a whole firmware to exactly that. There is no branch to choose now: the
# lock takes whatever mutex the bound driver hands it, and which driver bound
# is a run-time answer -- audiopump.driver().

# --- tier 5: audiomp3 (MP3Decoder) ---
#
# See micropython.mk for the full rationale (license, the cloned mp3
# sibling in the parent workspace, the allocator wiring, the mp3dec.h
# platform-detection escape hatch) -- only the CMake-specific parts
# (target-arch detection, per-file compile flags) are re-explained here.
#
# mp3dec.h's closed (__GNUC__, arch) list covers ARM (both `ARM` and
# `__ARMEL__`) and x86 (`__i386__`/`__amd64__`) but nothing else -- crucially,
# neither Xtensa nor RISC-V is in it, so *every* esp32 target this workspace
# builds (Xtensa: esp32/esp32s2/esp32s3; RISC-V: esp32c2/c3/c5/c6/p4) needs
# the `MP3DEC_GENERIC` escape hatch, while rp2 (RP2040/RP2350, both ARM
# Cortex-M, matching the `__GNUC__ && __ARMEL__` branch natively) does not.
# `CONFIG_IDF_TARGET_ARCH_RISCV`/`_ARCH_XTENSA` are the sdkconfig-derived
# CMake variables ESP-IDF's own esp32_common.cmake uses for this exact same
# arch branch (its `MICROPY_CROSS_FLAGS` selection) -- reused here instead
# of re-deriving "which target is this" from scratch; both are reliably set
# by the time usermod.cmake runs. rp2 has no such variable (native ARM
# match, no define needed). If a future non-ARM, non-x86 CMake port shows
# up, add its own arch-detection variable to this same `if()`.
if(CONFIG_IDF_TARGET_ARCH_RISCV OR CONFIG_IDF_TARGET_ARCH_XTENSA)
    target_compile_definitions(usermod_mpaudio INTERFACE MP3DEC_GENERIC)
    # The real compile picks this up fine via the usermod_mpaudio ->
    # usermod -> MICROPY_TARGET INTERFACE chain (ninja resolves transitive
    # usage requirements) -- but the QSTR-extraction preprocessing pass
    # (py/mkrules.cmake's qstr.i.last rule, ~line 63) builds its own flag
    # list via a raw `get_target_property(... MICROPY_TARGET
    # COMPILE_DEFINITIONS)`, which does NOT walk linked INTERFACE libraries
    # the way an actual compile does -- confirmed the hard way: that pass's
    # actual `-E` invocation was missing MP3DEC_GENERIC even though the
    # object-compile rule's DEFINES had it, and hit mp3dec.h's `#error`
    # directly. py.cmake's own `micropy_gather_target_properties()` exists
    # to bridge exactly this INTERFACE-library gap, but esp32_common.cmake
    # only calls it over `__COMPONENT_NAMES_RESOLVED` (registered ESP-IDF
    # components) -- "usermod" is a plain CMake target inside the main
    # component, not a component itself, so that loop never reaches it.
    # `MICROPY_CPP_DEF_EXTRA` is the accumulator both that gather-loop and
    # mkrules.cmake's own qstr flag-building read from (mkrules.cmake:68);
    # appending to it directly, here, is a normal (non-cache) variable
    # write in the same include()-preserved scope mkrules.cmake later reads
    # from -- the portable fix, not an esp32-specific one.
    list(APPEND MICROPY_CPP_DEF_EXTRA MP3DEC_GENERIC)
endif()

set(MPAUDIO_MP3_SRC_DIR ${MPAUDIO_MOD_DIR}/.deps/mp3/src)
if(NOT EXISTS ${MPAUDIO_MP3_SRC_DIR}/mp3dec.c)
    set(MPAUDIO_MP3_SRC_DIR ${MPAUDIO_MOD_DIR}/../mp3/src)
endif()
# Neither there: fetch our own pinned copy (DEPENDENCIES.lock, plus our
# patches) into .deps/, the way a fresh clone outside any workspace builds.
if(NOT EXISTS ${MPAUDIO_MP3_SRC_DIR}/mp3dec.c AND NOT DEFINED ENV{AUDIODSP_OPTIONAL_DEPS})
    message(STATUS "audiodsp: fetching Adafruit_MP3 into .deps/ (scripts/fetch_deps.sh)")
    execute_process(COMMAND bash ${MPAUDIO_MOD_DIR}/scripts/fetch_deps.sh RESULT_VARIABLE _mpaudio_fetch_rc)
    set(MPAUDIO_MP3_SRC_DIR ${MPAUDIO_MOD_DIR}/.deps/mp3/src)
    if(NOT EXISTS ${MPAUDIO_MP3_SRC_DIR}/mp3dec.c)
        message(FATAL_ERROR "audiodsp: fetching Adafruit_MP3 failed (scripts/fetch_deps.sh exit ${_mpaudio_fetch_rc}). Set AUDIODSP_OPTIONAL_DEPS=1 to build without audiomp3.")
    endif()
endif()
if(EXISTS ${MPAUDIO_MP3_SRC_DIR}/mp3dec.c)
    target_include_directories(usermod_mpaudio INTERFACE ${MPAUDIO_MP3_SRC_DIR})
    target_sources(usermod_mpaudio INTERFACE
        ${MPAUDIO_MP3_SRC_DIR}/bitstream.c
        ${MPAUDIO_MP3_SRC_DIR}/dct32.c
        ${MPAUDIO_MP3_SRC_DIR}/dequant.c
        ${MPAUDIO_MP3_SRC_DIR}/dqchan.c
        ${MPAUDIO_MP3_SRC_DIR}/huffman.c
        ${MPAUDIO_MP3_SRC_DIR}/hufftabs.c
        ${MPAUDIO_MP3_SRC_DIR}/imdct.c
        ${MPAUDIO_MP3_SRC_DIR}/mp3dec.c
        ${MPAUDIO_MP3_SRC_DIR}/mp3tabs.c
        ${MPAUDIO_MP3_SRC_DIR}/polyphase.c
        ${MPAUDIO_MP3_SRC_DIR}/scalfact.c
        ${MPAUDIO_MP3_SRC_DIR}/stproc.c
        ${MPAUDIO_MP3_SRC_DIR}/subband.c
        ${MPAUDIO_MP3_SRC_DIR}/trigtabs.c
        ${MPAUDIO_SRC_DIR}/audiomp3/mp3_alloc.c
        ${MPAUDIO_SRC_DIR}/audiomp3/mp3_buffers.c
        ${MPAUDIO_SRC_DIR}/audiomp3/MP3Decoder.c
        ${MPAUDIO_SRC_DIR}/audiomp3/module.c
    )

    # buffers.c's MPDEC_ALLOCATOR(x)/MPDEC_FREE(x) (used nowhere else in
    # lib/mp3) route through this port's own mp3_alloc()/mp3_free(): the
    # compile is src/audiomp3/mp3_buffers.c, which defines them and includes
    # buffers.c. Not a set_source_files_properties() on buffers.c: that is
    # scoped to this directory, and the ESP32 build compiles the linked copy of
    # every usermod source from the project's directory, where it was lost
    # (the buffers stayed static .bss, 23.8 kB of internal RAM).
endif()

# --- audiovorbis (VorbisDecoder) ---
#
# See micropython.mk: Xiph's Tremor from third_party/tremor, compiled as one
# translation unit by src/shared/audiodsp_vorbis.c, which renames its symbols
# and silences its warnings for its own code only, so IDF's -Werror=all sees
# nothing. Its memory comes through the driver's hooks (PSRAM on a board with
# it); see docs/vorbis.md.
target_sources(usermod_mpaudio INTERFACE
    ${MPAUDIO_SRC_DIR}/shared/audiodsp_vorbis.c
    ${MPAUDIO_SRC_DIR}/audiovorbis/VorbisDecoder.c
    ${MPAUDIO_SRC_DIR}/audiovorbis/module.c
)


# CIRCUITPY_SYNTHIO_MAX_CHANNELS (phase 8e decision, made while actually
# building the ESP32-P4 firmware -- see docs/porting-plan.md "8e"): the
# header default of 2 is not just conservative, it is *broken* for
# essentially any real patch -- confirmed in phase 7/8d, two notes held on
# a plain 2-oscillator/detuned voice is already 4 concurrent Notes, and the
# excess is silently REFUSED rather than erroring. Corrected 2026-09-02
# (audiodsp#14): this said "silently steals/truncates", which is the wrong
# failure. find_channel_with_note (src/synthio/__init__.c:361) reclaims only
# from RELEASED channels; when every channel is held the press is dropped
# outright (:425). A note that never sounds is the one you cannot hear go
# wrong. CircuitPython's
# own more-capable boards all raise it well past 2 (raspberrypi: 24,
# nordic/mimxrt10xx: 12) -- 2 is really only survivable on the most
# memory-starved boards, not a sane default for anything with real RAM.
# Raised 8 -> 14, then 14 -> 64 the same day (#31): 14 fits the drum kits,
# 64 fits a ten-finger chord on the melodic library, whose instruments press
# up to 9 Notes per key. The RAM cost is ~+900 bytes per Synthesizer on ARM.
# Nobody has measured RP2040 or ESP32-P4 voice headroom under a full 64-voice
# load; the ceiling used to be an accidental CPU governor, and a board that
# cannot render 64 voices will now try and drop buffers rather than silently
# thin. Measure before shipping this to hardware.
#
# The original 8 -> 14 note follows. 8 was chosen as a
# modest middle ground before the instrument library existed to measure
# against. It does now, and 14 is not a preference but a fact about it:
# counted live, the resident permanent Note objects per drum kit are
# cr78 14, tr808 13, tr909 13, drumtraks 13, linndrum 13, sp1200 12,
# simmons_sdsv 12, dmx 11, tr707 11, tr606 8. At 8 -- and at 12 -- half the
# kits refuse notes, and by the paragraph above that refusal is silent. A
# ceiling below 14 is therefore a bug in this workspace, not a conservative
# choice; cr78 needs exactly 14.
#
# It also collapses the three-numbers problem: header, Make and CMake now
# all read 14, so a patch behaves the same wherever it is built, and the
# desktop value keeps matching the parity oracle's own build (14) as
# micropython.mk:116-122 requires.
#
# The cost is 20 bytes of GC heap per channel -- +120 bytes per Synthesizer
# against a 1024/2048-byte render buffer the same object already allocates
# unconditionally (src/synthio/__init__.c:327-329). Idle channels cost one
# pointer compare per block (:244-246). No gate anywhere exercises this
# constant: nothing in .github/ or tools/ invokes CMake.
#
# A board with less headroom can still override via CFLAGS_EXTRA or
# board-specific target_compile_definitions -- on CMake ports CFLAGS_EXTRA
# must be an ENVIRONMENT variable, which is how the shared infrastructure
# reads it (cmods/micropython/py/mkrules.cmake:79-86), not a make variable
# as on the unix Make path. Per-board tuning is the fuller phase 10 (port
# matrix) job, not this one.
# --- which audiodsp this firmware was built from -----------------------------
# See micropython.mk for why this is computed from the checkout and how it is
# described. It is computed when the firmware BUILDS, not when CMake
# configures: a configure-time value stays in CMakeCache.txt, so an
# incremental build in a reused build directory, after the checkout moved,
# flashed the old revision over the new code (#156). The target below
# rewrites a header on every build, only when the strings change, and the
# compiler's dependency tracking recompiles the one file that includes it.
set(MPAUDIO_GENHDR_DIR ${CMAKE_BINARY_DIR}/audiodsp_genhdr)
set(MPAUDIO_REVISION_HEADER ${MPAUDIO_GENHDR_DIR}/audiodsp_revision.h)
set(MPAUDIO_REVISION_COMMAND ${CMAKE_COMMAND}
    -DMOD_DIR=${MPAUDIO_MOD_DIR} -DOUT=${MPAUDIO_REVISION_HEADER}
    -P ${MPAUDIO_SRC_DIR}/cp_compat/audiodsp_revision.cmake)
# Once now, so the header exists before anything compiles.
execute_process(COMMAND ${MPAUDIO_REVISION_COMMAND})
if(NOT TARGET audiodsp_revision)
    add_custom_target(audiodsp_revision
        COMMAND ${MPAUDIO_REVISION_COMMAND}
        BYPRODUCTS ${MPAUDIO_REVISION_HEADER}
        VERBATIM)
endif()
# An interface library's dependencies are followed by whatever links it.
add_dependencies(usermod_mpaudio audiodsp_revision)
target_include_directories(usermod_mpaudio INTERFACE ${MPAUDIO_GENHDR_DIR})
target_compile_definitions(usermod_mpaudio INTERFACE AUDIODSP_REVISION_HEADER=1)

target_compile_definitions(usermod_mpaudio INTERFACE CIRCUITPY_SYNTHIO_MAX_CHANNELS=64)

# --- AUDIODSP_PUMP_IRAM: the pump's block loop out of flash ----------------
#
# audiodsp#142. The block loop runs from flash through the cache, and an
# ESP-IDF flash write disables that cache -- measured on an ESP32-P4 at 48 kHz
# stereo, one 64 ms open/write/flush/close of 4 kB cost the speaker ~6 ms of
# audio, permanently, because the wire paces the pump and it can never catch
# the time back.
#
# The expansion is passed IN rather than written in the C, because this
# repository's sources are not allowed to know what platform they are on and
# `tools/check_portable.py` bans `IRAM_ATTR` by name -- the header it comes
# from is an IDF header. A plain section attribute needs no header and is what
# `IRAM_ATTR` expands to, minus the `__COUNTER__` suffix the IDF uses to give
# each function its own section; the IDF's linker fragments glob `.iram1*`, so
# a fixed name lands in the same place.
#
# OFF BY DEFAULT and opt-in per build:
#
#     idf.py -DAUDIODSP_PUMP_IRAM=1 build      (or set it in the board cmake)
#
# because IRAM is a few hundred kB shared with every driver that wants some,
# and nobody has paid for the answer yet. What this covers is the pump's OWN
# per-block path -- the loop, the ring, the tap, the event apply. It does NOT
# cover the graph the loop pulls, which is every DSP node in the tree and far
# too much to place; so the honest expectation is a REDUCTION in the 6 ms
# rather than a zero, and the board measurement is what says which.
if(AUDIODSP_PUMP_IRAM AND (CONFIG_IDF_TARGET_ARCH_RISCV OR CONFIG_IDF_TARGET_ARCH_XTENSA))
    target_compile_definitions(usermod_mpaudio INTERFACE
        "AUDIODSP_HOT=__attribute__((section(\".iram1.audiodsp\")))")
    message(STATUS "audiodsp: pump block loop in IRAM (AUDIODSP_PUMP_IRAM=1)")
endif()

# audiometer's feed and analysis, and the FFT they share with audioconvolve,
# in IRAM: ON wherever there is IRAM, unlike the pump loop above, because the
# meter runs inside usbif's sound-card pump and from flash it cost that core
# 8.9 % against the spike's 4.8 % from IRAM (media modules roadmap, Phase 6).
# A few kB; -DAUDIODSP_METER_IRAM=0 leaves it in flash.
if(NOT DEFINED AUDIODSP_METER_IRAM)
    set(AUDIODSP_METER_IRAM 1)
endif()
if(AUDIODSP_METER_IRAM AND (CONFIG_IDF_TARGET_ARCH_RISCV OR CONFIG_IDF_TARGET_ARCH_XTENSA))
    target_compile_definitions(usermod_mpaudio INTERFACE
        "AUDIODSP_METER_HOT=__attribute__((section(\".iram1.audiometer\")))")
endif()

target_link_libraries(usermod INTERFACE usermod_mpaudio)