AudioMixer: Multichannel handling of stereo volume
Assign left/right volume to 22.2 channels.
Test: mixerops_benchmark
Test: mixerops_tests
Bug: 193275879
Change-Id: Iea4dd08b34b2e1e5c17d7563702e1510d2f961e4
diff --git a/media/libaudioprocessing/AudioMixerOps.h b/media/libaudioprocessing/AudioMixerOps.h
index cd47dc6..2988c67 100644
--- a/media/libaudioprocessing/AudioMixerOps.h
+++ b/media/libaudioprocessing/AudioMixerOps.h
@@ -17,6 +17,8 @@
#ifndef ANDROID_AUDIO_MIXER_OPS_H
#define ANDROID_AUDIO_MIXER_OPS_H
+#include <audio_utils/channels.h>
+#include <audio_utils/primitives.h>
#include <system/audio.h>
namespace android {
@@ -229,15 +231,26 @@
* complexity of working on interleaved streams is now getting
* too high, and likely limits compiler optimization.
*/
-template <int MIXTYPE, int NCHAN,
+
+// compile-time function.
+constexpr inline bool usesCenterChannel(audio_channel_mask_t mask) {
+ using namespace audio_utils::channels;
+ for (size_t i = 0; i < std::size(kSideFromChannelIdx); ++i) {
+ if ((mask & (1 << i)) != 0 && kSideFromChannelIdx[i] == AUDIO_GEOMETRY_SIDE_CENTER) {
+ return true;
+ }
+ }
+ return false;
+}
+
+/*
+ * Applies stereo volume to the audio data based on proper left right channel affinity
+ * (templated channel MASK parameter).
+ */
+template <int MIXTYPE, audio_channel_mask_t MASK,
typename TO, typename TI, typename TV,
typename F>
-void stereoVolumeHelper(TO*& out, const TI*& in, const TV *vol, F f) {
- static_assert(NCHAN > 0 && NCHAN <= FCC_LIMIT);
- static_assert(MIXTYPE == MIXTYPE_MULTI_STEREOVOL
- || MIXTYPE == MIXTYPE_MULTI_SAVEONLY_STEREOVOL
- || MIXTYPE == MIXTYPE_STEREOEXPAND
- || MIXTYPE == MIXTYPE_MONOEXPAND);
+void stereoVolumeHelperWithChannelMask(TO*& out, const TI*& in, const TV *vol, F f) {
auto proc = [](auto& a, const auto& b) {
if constexpr (MIXTYPE == MIXTYPE_MULTI_STEREOVOL
|| MIXTYPE == MIXTYPE_STEREOEXPAND
@@ -250,59 +263,109 @@
auto inp = [&in]() -> const TI& {
if constexpr (MIXTYPE == MIXTYPE_STEREOEXPAND
|| MIXTYPE == MIXTYPE_MONOEXPAND) {
- return *in;
+ return *in; // note STEREOEXPAND assumes replicated L/R channels (see doc below).
} else {
return *in++;
}
};
- // HALs should only expose the canonical channel masks.
- proc(*out++, f(inp(), vol[0])); // front left
- if constexpr (NCHAN == 1) return;
- proc(*out++, f(inp(), vol[1])); // front right
- if constexpr (NCHAN == 2) return;
- if constexpr (NCHAN == 4) {
- proc(*out++, f(inp(), vol[0])); // back left
- proc(*out++, f(inp(), vol[1])); // back right
- return;
- }
-
- // TODO: Precompute center volume if not ramping.
std::decay_t<TV> center;
- if constexpr (std::is_floating_point_v<TV>) {
- center = (vol[0] + vol[1]) * 0.5; // do not use divide
- } else {
- center = (vol[0] >> 1) + (vol[1] >> 1); // rounds to 0.
- }
- proc(*out++, f(inp(), center)); // center (or 2.1 LFE)
- if constexpr (NCHAN == 3) return;
- if constexpr (NCHAN == 5) {
- proc(*out++, f(inp(), vol[0])); // back left
- proc(*out++, f(inp(), vol[1])); // back right
- return;
- }
-
- proc(*out++, f(inp(), center)); // lfe
- proc(*out++, f(inp(), vol[0])); // back left
- proc(*out++, f(inp(), vol[1])); // back right
- if constexpr (NCHAN == 6) return;
- if constexpr (NCHAN == 7) {
- proc(*out++, f(inp(), center)); // back center
- return;
- }
- // NCHAN == 8
- proc(*out++, f(inp(), vol[0])); // side left
- proc(*out++, f(inp(), vol[1])); // side right
- if constexpr (NCHAN > FCC_8) {
- // Mutes to zero extended surround channels.
- // 7.1.4 has the correct behavior.
- // 22.2 has the behavior that FLC and FRC will be mixed instead
- // of SL and SR and LFE will be center, not left.
- for (int i = 8; i < NCHAN; ++i) {
- // TODO: Consider using android::audio_utils::channels::kSideFromChannelIdx
- proc(*out++, f(inp(), 0.f));
+ constexpr bool USES_CENTER_CHANNEL = usesCenterChannel(MASK);
+ if constexpr (USES_CENTER_CHANNEL) {
+ if constexpr (std::is_floating_point_v<TV>) {
+ center = (vol[0] + vol[1]) * 0.5; // do not use divide
+ } else {
+ center = (vol[0] >> 1) + (vol[1] >> 1); // rounds to 0.
}
}
+
+ using namespace audio_utils::channels;
+
+ // if LFE and LFE2 are both present, they take left and right volume respectively.
+ constexpr unsigned LFE_LFE2 = \
+ AUDIO_CHANNEL_OUT_LOW_FREQUENCY | AUDIO_CHANNEL_OUT_LOW_FREQUENCY_2;
+ constexpr bool has_LFE_LFE2 = (MASK & LFE_LFE2) == LFE_LFE2;
+
+#pragma push_macro("DO_CHANNEL_POSITION")
+#undef DO_CHANNEL_POSITION
+#define DO_CHANNEL_POSITION(BIT_INDEX) \
+ if constexpr ((MASK & (1 << BIT_INDEX)) != 0) { \
+ constexpr auto side = kSideFromChannelIdx[BIT_INDEX]; \
+ if constexpr (side == AUDIO_GEOMETRY_SIDE_LEFT || \
+ has_LFE_LFE2 && (1 << BIT_INDEX) == AUDIO_CHANNEL_OUT_LOW_FREQUENCY) { \
+ proc(*out++, f(inp(), vol[0])); \
+ } else if constexpr (side == AUDIO_GEOMETRY_SIDE_RIGHT || \
+ has_LFE_LFE2 && (1 << BIT_INDEX) == AUDIO_CHANNEL_OUT_LOW_FREQUENCY_2) { \
+ proc(*out++, f(inp(), vol[1])); \
+ } else /* constexpr */ { \
+ proc(*out++, f(inp(), center)); \
+ } \
+ }
+
+ DO_CHANNEL_POSITION(0);
+ DO_CHANNEL_POSITION(1);
+ DO_CHANNEL_POSITION(2);
+ DO_CHANNEL_POSITION(3);
+ DO_CHANNEL_POSITION(4);
+ DO_CHANNEL_POSITION(5);
+ DO_CHANNEL_POSITION(6);
+ DO_CHANNEL_POSITION(7);
+
+ DO_CHANNEL_POSITION(8);
+ DO_CHANNEL_POSITION(9);
+ DO_CHANNEL_POSITION(10);
+ DO_CHANNEL_POSITION(11);
+ DO_CHANNEL_POSITION(12);
+ DO_CHANNEL_POSITION(13);
+ DO_CHANNEL_POSITION(14);
+ DO_CHANNEL_POSITION(15);
+
+ DO_CHANNEL_POSITION(16);
+ DO_CHANNEL_POSITION(17);
+ DO_CHANNEL_POSITION(18);
+ DO_CHANNEL_POSITION(19);
+ DO_CHANNEL_POSITION(20);
+ DO_CHANNEL_POSITION(21);
+ DO_CHANNEL_POSITION(22);
+ DO_CHANNEL_POSITION(23);
+ static_assert(FCC_LIMIT <= FCC_24); // Note: this may need to change.
+#pragma pop_macro("DO_CHANNEL_POSITION")
+}
+
+// These are the channel position masks we expect from the HAL.
+// See audio_channel_out_mask_from_count() but this is constexpr
+constexpr inline audio_channel_mask_t canonicalChannelMaskFromCount(size_t channelCount) {
+ constexpr audio_channel_mask_t canonical[] = {
+ [0] = AUDIO_CHANNEL_NONE,
+ [1] = AUDIO_CHANNEL_OUT_MONO,
+ [2] = AUDIO_CHANNEL_OUT_STEREO,
+ [3] = AUDIO_CHANNEL_OUT_2POINT1,
+ [4] = AUDIO_CHANNEL_OUT_QUAD,
+ [5] = AUDIO_CHANNEL_OUT_PENTA,
+ [6] = AUDIO_CHANNEL_OUT_5POINT1,
+ [7] = AUDIO_CHANNEL_OUT_6POINT1,
+ [8] = AUDIO_CHANNEL_OUT_7POINT1,
+ [12] = AUDIO_CHANNEL_OUT_7POINT1POINT4,
+ [24] = AUDIO_CHANNEL_OUT_22POINT2,
+ };
+ return channelCount < std::size(canonical) ? canonical[channelCount] : AUDIO_CHANNEL_NONE;
+}
+
+template <int MIXTYPE, int NCHAN,
+ typename TO, typename TI, typename TV,
+ typename F>
+void stereoVolumeHelper(TO*& out, const TI*& in, const TV *vol, F f) {
+ static_assert(NCHAN > 0 && NCHAN <= FCC_LIMIT);
+ static_assert(MIXTYPE == MIXTYPE_MULTI_STEREOVOL
+ || MIXTYPE == MIXTYPE_MULTI_SAVEONLY_STEREOVOL
+ || MIXTYPE == MIXTYPE_STEREOEXPAND
+ || MIXTYPE == MIXTYPE_MONOEXPAND);
+ constexpr audio_channel_mask_t MASK{canonicalChannelMaskFromCount(NCHAN)};
+ if constexpr (MASK == AUDIO_CHANNEL_NONE) {
+ ALOGE("%s: Invalid position count %d", __func__, NCHAN);
+ return; // not a valid system mask, ignore.
+ }
+ stereoVolumeHelperWithChannelMask<MIXTYPE, MASK, TO, TI, TV, F>(out, in, vol, f);
}
/*
diff --git a/media/libaudioprocessing/tests/Android.bp b/media/libaudioprocessing/tests/Android.bp
index 3856817..ad402db 100644
--- a/media/libaudioprocessing/tests/Android.bp
+++ b/media/libaudioprocessing/tests/Android.bp
@@ -76,6 +76,7 @@
//
cc_binary {
name: "mixerops_objdump",
+ header_libs: ["libaudioutils_headers"],
srcs: ["mixerops_objdump.cpp"],
}
@@ -84,6 +85,16 @@
//
cc_benchmark {
name: "mixerops_benchmark",
+ header_libs: ["libaudioutils_headers"],
srcs: ["mixerops_benchmark.cpp"],
static_libs: ["libgoogle-benchmark"],
}
+
+//
+// mixerops unit test
+//
+cc_test {
+ name: "mixerops_tests",
+ defaults: ["libaudioprocessing_test_defaults"],
+ srcs: ["mixerops_tests.cpp"],
+}
diff --git a/media/libaudioprocessing/tests/mixerops_benchmark.cpp b/media/libaudioprocessing/tests/mixerops_benchmark.cpp
index 7a4c5c7..f866b1a 100644
--- a/media/libaudioprocessing/tests/mixerops_benchmark.cpp
+++ b/media/libaudioprocessing/tests/mixerops_benchmark.cpp
@@ -16,11 +16,9 @@
#include <inttypes.h>
#include <type_traits>
-#include "../../../../system/media/audio_utils/include/audio_utils/primitives.h"
#define LOG_ALWAYS_FATAL(...)
#include <../AudioMixerOps.h>
-
#include <benchmark/benchmark.h>
using namespace android;
diff --git a/media/libaudioprocessing/tests/mixerops_tests.cpp b/media/libaudioprocessing/tests/mixerops_tests.cpp
new file mode 100644
index 0000000..2500ba9
--- /dev/null
+++ b/media/libaudioprocessing/tests/mixerops_tests.cpp
@@ -0,0 +1,175 @@
+/*
+ * Copyright (C) 2021 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+//#define LOG_NDEBUG 0
+#define LOG_TAG "mixerop_tests"
+#include <log/log.h>
+
+#include <inttypes.h>
+#include <type_traits>
+
+#include <../AudioMixerOps.h>
+#include <gtest/gtest.h>
+
+using namespace android;
+
+// Note: gtest templated tests require typenames, not integers.
+template <int MIXTYPE, int NCHAN>
+class MixerOpsBasicTest {
+public:
+ static void testStereoVolume() {
+ using namespace android::audio_utils::channels;
+
+ constexpr size_t FRAME_COUNT = 1000;
+ constexpr size_t SAMPLE_COUNT = FRAME_COUNT * NCHAN;
+
+ const float in[SAMPLE_COUNT] = {[0 ... (SAMPLE_COUNT - 1)] = 1.f};
+
+ AUDIO_GEOMETRY_SIDE sides[NCHAN];
+ size_t i = 0;
+ unsigned channel = canonicalChannelMaskFromCount(NCHAN);
+ constexpr unsigned LFE_LFE2 =
+ AUDIO_CHANNEL_OUT_LOW_FREQUENCY | AUDIO_CHANNEL_OUT_LOW_FREQUENCY_2;
+ bool has_LFE_LFE2 = (channel & LFE_LFE2) == LFE_LFE2;
+ while (channel != 0) {
+ const int index = __builtin_ctz(channel);
+ if (has_LFE_LFE2 && (1 << index) == AUDIO_CHANNEL_OUT_LOW_FREQUENCY) {
+ sides[i++] = AUDIO_GEOMETRY_SIDE_LEFT; // special case
+ } else if (has_LFE_LFE2 && (1 << index) == AUDIO_CHANNEL_OUT_LOW_FREQUENCY_2) {
+ sides[i++] = AUDIO_GEOMETRY_SIDE_RIGHT; // special case
+ } else {
+ sides[i++] = sideFromChannelIdx(index);
+ }
+ channel &= ~(1 << index);
+ }
+
+ float vola[2] = {1.f, 0.f}; // left volume at max.
+ float out[SAMPLE_COUNT]{};
+ float aux[FRAME_COUNT]{};
+ float volaux = 0.5;
+ {
+ volumeMulti<MIXTYPE, NCHAN>(out, FRAME_COUNT, in, aux, vola, volaux);
+ const float *outp = out;
+ const float *auxp = aux;
+ const float left = vola[0];
+ const float center = (vola[0] + vola[1]) * 0.5;
+ const float right = vola[1];
+ for (size_t i = 0; i < FRAME_COUNT; ++i) {
+ for (size_t j = 0; j < NCHAN; ++j) {
+ const float audio = *outp++;
+ if (sides[j] == AUDIO_GEOMETRY_SIDE_LEFT) {
+ EXPECT_EQ(left, audio);
+ } else if (sides[j] == AUDIO_GEOMETRY_SIDE_CENTER) {
+ EXPECT_EQ(center, audio);
+ } else {
+ EXPECT_EQ(right, audio);
+ }
+ }
+ EXPECT_EQ(volaux, *auxp++); // works if all channels contain 1.f
+ }
+ }
+ float volb[2] = {0.f, 0.5f}; // right volume at half max.
+ {
+ // this accumulates into out, aux.
+ // float out[SAMPLE_COUNT]{};
+ // float aux[FRAME_COUNT]{};
+ volumeMulti<MIXTYPE, NCHAN>(out, FRAME_COUNT, in, aux, volb, volaux);
+ const float *outp = out;
+ const float *auxp = aux;
+ const float left = vola[0] + volb[0];
+ const float center = (vola[0] + vola[1] + volb[0] + volb[1]) * 0.5;
+ const float right = vola[1] + volb[1];
+ for (size_t i = 0; i < FRAME_COUNT; ++i) {
+ for (size_t j = 0; j < NCHAN; ++j) {
+ const float audio = *outp++;
+ if (sides[j] == AUDIO_GEOMETRY_SIDE_LEFT) {
+ EXPECT_EQ(left, audio);
+ } else if (sides[j] == AUDIO_GEOMETRY_SIDE_CENTER) {
+ EXPECT_EQ(center, audio);
+ } else {
+ EXPECT_EQ(right, audio);
+ }
+ }
+ // aux is accumulated so 2x the amplitude
+ EXPECT_EQ(volaux * 2.f, *auxp++); // works if all channels contain 1.f
+ }
+ }
+
+ { // test aux as derived from out.
+ // AUX channel is the weighted sum of all of the output channels prior to volume
+ // adjustment. We must set L and R to the same volume to allow computation
+ // of AUX from the output values.
+ const float volmono = 0.25f;
+ const float vollr[2] = {volmono, volmono}; // all the same.
+ float out[SAMPLE_COUNT]{};
+ float aux[FRAME_COUNT]{};
+ volumeMulti<MIXTYPE, NCHAN>(out, FRAME_COUNT, in, aux, vollr, volaux);
+ const float *outp = out;
+ const float *auxp = aux;
+ for (size_t i = 0; i < FRAME_COUNT; ++i) {
+ float accum = 0.f;
+ for (size_t j = 0; j < NCHAN; ++j) {
+ accum += *outp++;
+ }
+ EXPECT_EQ(accum / NCHAN * volaux / volmono, *auxp++);
+ }
+ }
+ }
+};
+
+TEST(mixerops, stereovolume_1) { // Note: mono not used for output sinks yet.
+ MixerOpsBasicTest<MIXTYPE_MULTI_STEREOVOL, 1>::testStereoVolume();
+}
+TEST(mixerops, stereovolume_2) {
+ MixerOpsBasicTest<MIXTYPE_MULTI_STEREOVOL, 2>::testStereoVolume();
+}
+TEST(mixerops, stereovolume_3) {
+ MixerOpsBasicTest<MIXTYPE_MULTI_STEREOVOL, 3>::testStereoVolume();
+}
+TEST(mixerops, stereovolume_4) {
+ MixerOpsBasicTest<MIXTYPE_MULTI_STEREOVOL, 4>::testStereoVolume();
+}
+TEST(mixerops, stereovolume_5) {
+ MixerOpsBasicTest<MIXTYPE_MULTI_STEREOVOL, 5>::testStereoVolume();
+}
+TEST(mixerops, stereovolume_6) {
+ MixerOpsBasicTest<MIXTYPE_MULTI_STEREOVOL, 6>::testStereoVolume();
+}
+TEST(mixerops, stereovolume_7) {
+ MixerOpsBasicTest<MIXTYPE_MULTI_STEREOVOL, 7>::testStereoVolume();
+}
+TEST(mixerops, stereovolume_8) {
+ MixerOpsBasicTest<MIXTYPE_MULTI_STEREOVOL, 8>::testStereoVolume();
+}
+TEST(mixerops, stereovolume_12) {
+ if constexpr (FCC_LIMIT >= 12) { // NOTE: FCC_LIMIT is an enum, so can't #if
+ MixerOpsBasicTest<MIXTYPE_MULTI_STEREOVOL, 12>::testStereoVolume();
+ }
+}
+TEST(mixerops, stereovolume_24) {
+ if constexpr (FCC_LIMIT >= 24) {
+ MixerOpsBasicTest<MIXTYPE_MULTI_STEREOVOL, 24>::testStereoVolume();
+ }
+}
+TEST(mixerops, channel_equivalence) {
+ // we must match the constexpr function with the system determined channel mask from count.
+ for (size_t i = 0; i < FCC_LIMIT; ++i) {
+ const audio_channel_mask_t actual = canonicalChannelMaskFromCount(i);
+ const audio_channel_mask_t system = audio_channel_out_mask_from_count(i);
+ if (system == AUDIO_CHANNEL_INVALID) continue;
+ EXPECT_EQ(system, actual);
+ }
+}