Merge "EffectDownmix: Use full 22.2 matrix" into sc-v2-dev (9dc68b99) · Commits · e / os / android_frameworks_av

media/libeffects/downmix/benchmark/downmix_benchmark.cpp

+24 −22

Original line number	Diff line number	Diff line
		@@ -61,31 +61,33 @@ static constexpr size_t kFrameCount = 1000;

		/*
		Pixel 4XL
		$ adb shell /data/benchmarktest/downmix_benchmark/vendor/downmix_benchmark

		--------------------------------------------------------
		Benchmark Time CPU Iterations
		--------------------------------------------------------
		BM_Downmix/0 2845 ns 2839 ns 246585 AUDIO_CHANNEL_OUT_MONO
		BM_Downmix/1 2844 ns 2838 ns 246599
		BM_Downmix/2 3727 ns 3719 ns 188227 AUDIO_CHANNEL_OUT_STEREO
		BM_Downmix/3 4609 ns 4600 ns 152148 AUDIO_CHANNEL_OUT_2POINT1
		BM_Downmix/4 3727 ns 3719 ns 188228 AUDIO_CHANNEL_OUT_2POINT0POINT2
		BM_Downmix/5 1787 ns 1784 ns 392384 AUDIO_CHANNEL_OUT_QUAD
		BM_Downmix/6 1787 ns 1783 ns 392527 AUDIO_CHANNEL_OUT_QUAD_SIDE
		BM_Downmix/7 5493 ns 5481 ns 127740 AUDIO_CHANNEL_OUT_SURROUND
		BM_Downmix/8 4610 ns 4600 ns 152168 AUDIO_CHANNEL_OUT_2POINT1POINT2
		BM_Downmix/9 4610 ns 4600 ns 152162 AUDIO_CHANNEL_OUT_3POINT0POINT2
		BM_Downmix/10 6377 ns 6362 ns 110042 AUDIO_CHANNEL_OUT_PENTA
		BM_Downmix/11 5493 ns 5481 ns 127683 AUDIO_CHANNEL_OUT_3POINT1POINT2
		BM_Downmix/12 2758 ns 2752 ns 251488 AUDIO_CHANNEL_OUT_5POINT1
		BM_Downmix/13 2683 ns 2677 ns 261421 AUDIO_CHANNEL_OUT_5POINT1_SIDE
		BM_Downmix/14 8141 ns 8124 ns 86157 AUDIO_CHANNEL_OUT_6POINT1
		BM_Downmix/15 7265 ns 7249 ns 96554 AUDIO_CHANNEL_OUT_5POINT1POINT2
		BM_Downmix/16 3158 ns 3151 ns 222188 AUDIO_CHANNEL_OUT_7POINT1
		BM_Downmix/17 7291 ns 7276 ns 96226 AUDIO_CHANNEL_OUT_5POINT1POINT4
		BM_Downmix/18 9050 ns 9031 ns 77512 AUDIO_CHANNEL_OUT_7POINT1POINT2
		BM_Downmix/19 9056 ns 9036 ns 77467 AUDIO_CHANNEL_OUT_7POINT1POINT4
		BM_Downmix/20 6426 ns 6412 ns 109164 AUDIO_CHANNEL_OUT_13POINT_360RA
		BM_Downmix/21 11743 ns 11716 ns 59762 AUDIO_CHANNEL_OUT_22POINT2
		BM_Downmix/0 3638 ns 3624 ns 197517 AUDIO_CHANNEL_OUT_MONO
		BM_Downmix/1 4040 ns 4024 ns 178766
		BM_Downmix/2 4759 ns 4740 ns 134741 AUDIO_CHANNEL_OUT_STEREO
		BM_Downmix/3 6042 ns 6017 ns 129546 AUDIO_CHANNEL_OUT_2POINT1
		BM_Downmix/4 6897 ns 6868 ns 96316 AUDIO_CHANNEL_OUT_2POINT0POINT2
		BM_Downmix/5 2117 ns 2109 ns 331705 AUDIO_CHANNEL_OUT_QUAD
		BM_Downmix/6 2097 ns 2088 ns 335421 AUDIO_CHANNEL_OUT_QUAD_SIDE
		BM_Downmix/7 7291 ns 7263 ns 96256 AUDIO_CHANNEL_OUT_SURROUND
		BM_Downmix/8 8246 ns 8206 ns 84318 AUDIO_CHANNEL_OUT_2POINT1POINT2
		BM_Downmix/9 8341 ns 8303 ns 84298 AUDIO_CHANNEL_OUT_3POINT0POINT2
		BM_Downmix/10 7549 ns 7517 ns 84293 AUDIO_CHANNEL_OUT_PENTA
		BM_Downmix/11 9395 ns 9354 ns 75209 AUDIO_CHANNEL_OUT_3POINT1POINT2
		BM_Downmix/12 3267 ns 3253 ns 215596 AUDIO_CHANNEL_OUT_5POINT1
		BM_Downmix/13 3178 ns 3163 ns 220132 AUDIO_CHANNEL_OUT_5POINT1_SIDE
		BM_Downmix/14 10245 ns 10199 ns 67486 AUDIO_CHANNEL_OUT_6POINT1
		BM_Downmix/15 10975 ns 10929 ns 61359 AUDIO_CHANNEL_OUT_5POINT1POINT2
		BM_Downmix/16 3796 ns 3780 ns 184728 AUDIO_CHANNEL_OUT_7POINT1
		BM_Downmix/17 13562 ns 13503 ns 51823 AUDIO_CHANNEL_OUT_5POINT1POINT4
		BM_Downmix/18 13573 ns 13516 ns 51800 AUDIO_CHANNEL_OUT_7POINT1POINT2
		BM_Downmix/19 15502 ns 15435 ns 47147 AUDIO_CHANNEL_OUT_7POINT1POINT4
		BM_Downmix/20 16693 ns 16624 ns 42109 AUDIO_CHANNEL_OUT_13POINT_360RA
		BM_Downmix/21 28267 ns 28116 ns 24982 AUDIO_CHANNEL_OUT_22POINT2
		*/

		static void BM_Downmix(benchmark::State& state) {

media/libeffects/downmix/tests/downmix_tests.cpp

+76 −37

Original line number	Diff line number	Diff line
		@@ -52,34 +52,63 @@ static constexpr audio_channel_mask_t kChannelPositionMasks[] = {
		AUDIO_CHANNEL_OUT_22POINT2,
		};

		static constexpr audio_channel_mask_t kConsideredChannels =
		(audio_channel_mask_t)(AUDIO_CHANNEL_OUT_7POINT1 \| AUDIO_CHANNEL_OUT_BACK_CENTER);
		constexpr float COEF_25 = 0.2508909536f;
		constexpr float COEF_35 = 0.3543928915f;
		constexpr float COEF_36 = 0.3552343859f;
		constexpr float COEF_61 = 0.6057043428f;

		constexpr inline float kScaleFromChannelIdx[] = {
		constexpr inline float kScaleFromChannelIdxLeft[] = {
		1.f, // AUDIO_CHANNEL_OUT_FRONT_LEFT = 0x1u,
		1.f, // AUDIO_CHANNEL_OUT_FRONT_RIGHT = 0x2u,
		0.f, // AUDIO_CHANNEL_OUT_FRONT_RIGHT = 0x2u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_FRONT_CENTER = 0x4u,
		0.5f, // AUDIO_CHANNEL_OUT_LOW_FREQUENCY = 0x8u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_BACK_LEFT = 0x10u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_BACK_RIGHT = 0x20u,
		0, // AUDIO_CHANNEL_OUT_FRONT_LEFT_OF_CENTER = 0x40u,
		0, // AUDIO_CHANNEL_OUT_FRONT_RIGHT_OF_CENTER = 0x80u,
		0.f, // AUDIO_CHANNEL_OUT_BACK_RIGHT = 0x20u,
		COEF_61, // AUDIO_CHANNEL_OUT_FRONT_LEFT_OF_CENTER = 0x40u,
		COEF_25, // AUDIO_CHANNEL_OUT_FRONT_RIGHT_OF_CENTER = 0x80u,
		0.5f, // AUDIO_CHANNEL_OUT_BACK_CENTER = 0x100u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_SIDE_LEFT = 0x200u,
		0.f, // AUDIO_CHANNEL_OUT_SIDE_RIGHT = 0x400u,
		COEF_36, // AUDIO_CHANNEL_OUT_TOP_CENTER = 0x800u,
		1.f, // AUDIO_CHANNEL_OUT_TOP_FRONT_LEFT = 0x1000u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_TOP_FRONT_CENTER = 0x2000u,
		0.f, // AUDIO_CHANNEL_OUT_TOP_FRONT_RIGHT = 0x4000u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_TOP_BACK_LEFT = 0x8000u,
		COEF_35, // AUDIO_CHANNEL_OUT_TOP_BACK_CENTER = 0x10000u,
		0.f, // AUDIO_CHANNEL_OUT_TOP_BACK_RIGHT = 0x20000u,
		COEF_61, // AUDIO_CHANNEL_OUT_TOP_SIDE_LEFT = 0x40000u,
		0.f, // AUDIO_CHANNEL_OUT_TOP_SIDE_RIGHT = 0x80000u,
		1.f, // AUDIO_CHANNEL_OUT_BOTTOM_FRONT_LEFT = 0x100000u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_BOTTOM_FRONT_CENTER = 0x200000u,
		0.f, // AUDIO_CHANNEL_OUT_BOTTOM_FRONT_RIGHT = 0x400000u,
		0.f, // AUDIO_CHANNEL_OUT_LOW_FREQUENCY_2 = 0x800000u,
		};

		constexpr inline float kScaleFromChannelIdxRight[] = {
		0.f, // AUDIO_CHANNEL_OUT_FRONT_LEFT = 0x1u,
		1.f, // AUDIO_CHANNEL_OUT_FRONT_RIGHT = 0x2u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_FRONT_CENTER = 0x4u,
		0.5f, // AUDIO_CHANNEL_OUT_LOW_FREQUENCY = 0x8u,
		0.f, // AUDIO_CHANNEL_OUT_BACK_LEFT = 0x10u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_BACK_RIGHT = 0x20u,
		COEF_25, // AUDIO_CHANNEL_OUT_FRONT_LEFT_OF_CENTER = 0x40u,
		COEF_61, // AUDIO_CHANNEL_OUT_FRONT_RIGHT_OF_CENTER = 0x80u,
		0.5f, // AUDIO_CHANNEL_OUT_BACK_CENTER = 0x100u,
		0.f, // AUDIO_CHANNEL_OUT_SIDE_LEFT = 0x200u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_SIDE_RIGHT = 0x400u,
		0, // AUDIO_CHANNEL_OUT_TOP_CENTER = 0x800u,
		0, // AUDIO_CHANNEL_OUT_TOP_FRONT_LEFT = 0x1000u,
		0, // AUDIO_CHANNEL_OUT_TOP_FRONT_CENTER = 0x2000u,
		0, // AUDIO_CHANNEL_OUT_TOP_FRONT_RIGHT = 0x4000u,
		0, // AUDIO_CHANNEL_OUT_TOP_BACK_LEFT = 0x8000u,
		0, // AUDIO_CHANNEL_OUT_TOP_BACK_CENTER = 0x10000u,
		0, // AUDIO_CHANNEL_OUT_TOP_BACK_RIGHT = 0x20000u,
		0, // AUDIO_CHANNEL_OUT_TOP_SIDE_LEFT = 0x40000u,
		0, // AUDIO_CHANNEL_OUT_TOP_SIDE_RIGHT = 0x80000u,
		0, // AUDIO_CHANNEL_OUT_BOTTOM_FRONT_LEFT = 0x100000u,
		0, // AUDIO_CHANNEL_OUT_BOTTOM_FRONT_CENTER = 0x200000u,
		0, // AUDIO_CHANNEL_OUT_BOTTOM_FRONT_RIGHT = 0x400000u,
		0, // AUDIO_CHANNEL_OUT_LOW_FREQUENCY_2 = 0x800000u,
		COEF_36, // AUDIO_CHANNEL_OUT_TOP_CENTER = 0x800u,
		0.f, // AUDIO_CHANNEL_OUT_TOP_FRONT_LEFT = 0x1000u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_TOP_FRONT_CENTER = 0x2000u,
		1.f, // AUDIO_CHANNEL_OUT_TOP_FRONT_RIGHT = 0x4000u,
		0.f, // AUDIO_CHANNEL_OUT_TOP_BACK_LEFT = 0x8000u,
		COEF_35, // AUDIO_CHANNEL_OUT_TOP_BACK_CENTER = 0x10000u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_TOP_BACK_RIGHT = 0x20000u,
		0.f, // AUDIO_CHANNEL_OUT_TOP_SIDE_LEFT = 0x40000u,
		COEF_61, // AUDIO_CHANNEL_OUT_TOP_SIDE_RIGHT = 0x80000u,
		0.f, // AUDIO_CHANNEL_OUT_BOTTOM_FRONT_LEFT = 0x100000u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_BOTTOM_FRONT_CENTER = 0x200000u,
		1.f, // AUDIO_CHANNEL_OUT_BOTTOM_FRONT_RIGHT = 0x400000u,
		M_SQRT1_2, // AUDIO_CHANNEL_OUT_LOW_FREQUENCY_2 = 0x800000u,
		};

		// Downmix doesn't change with sample rate
		@@ -155,30 +184,40 @@ public:
		savedPower[index][0] = power[0];
		savedPower[index][1] = power[1];

		// Confirm exactly the mix amount prescribed by the existing downmix effect.
		// For future changes to the downmix effect, the nearness needs to be relaxed
		// to compare behavior S or earlier.
		if ((channelBit & kConsideredChannels) == 0) {
		// for channels not considered, expect 0 power for legacy downmix
		EXPECT_EQ(0.f, power[0]);
		EXPECT_EQ(0.f, power[1]);
		continue;
		}

		constexpr float POWER_TOLERANCE = 0.001;
		const float expectedPower = kScaleFromChannelIdx[index] * kScaleFromChannelIdx[index];
		const float expectedPower =
		kScaleFromChannelIdxLeft[index] * kScaleFromChannelIdxLeft[index]
		+ kScaleFromChannelIdxRight[index] * kScaleFromChannelIdxRight[index];
		EXPECT_NEAR(expectedPower, power[0] + power[1], POWER_TOLERANCE);
		switch (side) {
		case AUDIO_GEOMETRY_SIDE_LEFT:
		EXPECT_EQ(0.f, power[1]); // always true
		EXPECT_NEAR(expectedPower, power[0], POWER_TOLERANCE);
		if (channelBit == AUDIO_CHANNEL_OUT_FRONT_LEFT_OF_CENTER) {
		break;
		}
		EXPECT_EQ(0.f, power[1]);
		break;
		case AUDIO_GEOMETRY_SIDE_RIGHT:
		EXPECT_EQ(0.f, power[0]); // always true
		EXPECT_NEAR(expectedPower, power[1], POWER_TOLERANCE);
		if (channelBit == AUDIO_CHANNEL_OUT_FRONT_RIGHT_OF_CENTER) {
		break;
		}
		EXPECT_EQ(0.f, power[0]);
		break;
		case AUDIO_GEOMETRY_SIDE_CENTER:
		if (channelBit == AUDIO_CHANNEL_OUT_LOW_FREQUENCY) {
		if (channelMask & AUDIO_CHANNEL_OUT_LOW_FREQUENCY_2) {
		EXPECT_EQ(0.f, power[1]);
		break;
		} else {
		EXPECT_NEAR_EPSILON(power[0], power[1]); // always true
		EXPECT_NEAR(expectedPower, power[0], POWER_TOLERANCE);
		EXPECT_NEAR(expectedPower, power[0] + power[1], POWER_TOLERANCE);
		break;
		}
		} else if (channelBit == AUDIO_CHANNEL_OUT_LOW_FREQUENCY_2) {
		EXPECT_EQ(0.f, power[0]);
		EXPECT_NEAR(expectedPower, power[1], POWER_TOLERANCE);
		break;
		}
		EXPECT_NEAR_EPSILON(power[0], power[1]);
		break;
		}
		}