Skip to content

Commit 1404710

Browse files
kunitokikunitoki
committed
More tweaks
1 parent 84f20f0 commit 1404710

File tree

3 files changed

+329
-30
lines changed

3 files changed

+329
-30
lines changed

‎modules/yup_audio_basics/buffers/yup_FloatVectorOperations.cpp‎

Lines changed: 156 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -1341,24 +1341,48 @@ double findMaximum (const double* src, Size num) noexcept
13411341
template <typename Size>
13421342
void convertFixedToFloat (float* dest, const int* src, float multiplier, Size num) noexcept
13431343
{
1344-
#if YUP_USE_ARM_NEON
1344+
#if YUP_USE_VDSP_FRAMEWORK
1345+
vDSP_vflt32 (reinterpret_cast<const int*> (src), 1, dest, 1, (vDSP_Length) num);
1346+
vDSP_vsmul (dest, 1, &multiplier, dest, 1, (vDSP_Length) num);
1347+
#elif YUP_USE_ARM_NEON
13451348
YUP_PERFORM_VEC_OP_SRC_DEST (dest[i] = (float) src[i] * multiplier,
13461349
vmulq_n_f32 (vcvtq_f32_s32 (vld1q_s32 (src)), multiplier),
13471350
YUP_LOAD_NONE,
13481351
YUP_INCREMENT_SRC_DEST, )
1349-
#else
1352+
#elif YUP_USE_SSE_INTRINSICS
13501353
YUP_PERFORM_VEC_OP_SRC_DEST (dest[i] = (float) src[i] * multiplier,
13511354
Mode::mul (mult, _mm_cvtepi32_ps (_mm_loadu_si128 (reinterpret_cast<const __m128i*> (src)))),
13521355
YUP_LOAD_NONE,
13531356
YUP_INCREMENT_SRC_DEST,
13541357
const Mode::ParallelType mult = Mode::load1 (multiplier);)
1358+
#else
1359+
for (Size i = 0; i < num; ++i)
1360+
dest[i] = (float) src[i] * multiplier;
13551361
#endif
13561362
}
13571363

13581364
template <typename Size>
13591365
void convertFloatToFixed (int* dest, const float* src, float multiplier, Size num) noexcept
13601366
{
1361-
#if YUP_USE_ARM_NEON
1367+
#if YUP_USE_VDSP_FRAMEWORK
1368+
constexpr Size kStackBufferSize = 256;
1369+
float stackBuffer[kStackBufferSize];
1370+
1371+
if (num <= kStackBufferSize)
1372+
{
1373+
vDSP_vsmul (src, 1, &multiplier, stackBuffer, 1, (vDSP_Length) num);
1374+
vDSP_vfix32 (stackBuffer, 1, reinterpret_cast<int*> (dest), 1, (vDSP_Length) num);
1375+
}
1376+
else
1377+
{
1378+
for (Size i = 0; i < num; i += kStackBufferSize)
1379+
{
1380+
const Size currentChunk = jmin (kStackBufferSize, num - i);
1381+
vDSP_vsmul (src + i, 1, &multiplier, stackBuffer, 1, (vDSP_Length) currentChunk);
1382+
vDSP_vfix32 (stackBuffer, 1, reinterpret_cast<int*> (dest + i), 1, (vDSP_Length) currentChunk);
1383+
}
1384+
}
1385+
#elif YUP_USE_ARM_NEON
13621386
const auto numLongs = num & ~3;
13631387

13641388
if (numLongs != 0)
@@ -1374,7 +1398,6 @@ void convertFloatToFixed (int* dest, const float* src, float multiplier, Size nu
13741398

13751399
for (Size i = numLongs; i < num; ++i)
13761400
dest[i] = (int) (src[i] * multiplier);
1377-
13781401
#elif YUP_USE_SSE_INTRINSICS
13791402
const auto numLongs = num & ~3;
13801403
const __m128 mult = _mm_set1_ps (multiplier);
@@ -1392,7 +1415,115 @@ void convertFloatToFixed (int* dest, const float* src, float multiplier, Size nu
13921415

13931416
for (Size i = numLongs; i < num; ++i)
13941417
dest[i] = (int) (src[i] * multiplier);
1418+
#else
1419+
for (Size i = 0; i < num; ++i)
1420+
dest[i] = (int) (src[i] * multiplier);
1421+
#endif
1422+
}
1423+
1424+
template <typename Size>
1425+
void convertFixedToFloat (double* dest, const int* src, double multiplier, Size num) noexcept
1426+
{
1427+
#if YUP_USE_VDSP_FRAMEWORK
1428+
vDSP_vflt32D (reinterpret_cast<const int*> (src), 1, dest, 1, (vDSP_Length) num);
1429+
vDSP_vsmulD (dest, 1, &multiplier, dest, 1, (vDSP_Length) num);
1430+
#elif YUP_USE_ARM_NEON
1431+
const auto numLongs = num & ~1;
1432+
1433+
if (numLongs != 0)
1434+
{
1435+
for (Size i = 0; i < numLongs; i += 2)
1436+
{
1437+
int32x2_t intVec = vld1_s32 (src + i);
1438+
int val0 = vget_lane_s32 (intVec, 0);
1439+
int val1 = vget_lane_s32 (intVec, 1);
1440+
dest[i] = (double) val0 * multiplier;
1441+
dest[i + 1] = (double) val1 * multiplier;
1442+
}
1443+
}
1444+
1445+
for (Size i = numLongs; i < num; ++i)
1446+
dest[i] = (double) src[i] * multiplier;
1447+
#elif YUP_USE_SSE_INTRINSICS
1448+
const auto numLongs = num & ~1;
1449+
1450+
if (numLongs != 0)
1451+
{
1452+
for (Size i = 0; i < numLongs; i += 2)
1453+
{
1454+
__m128i intVec = _mm_loadl_epi64 (reinterpret_cast<const __m128i*> (src + i));
1455+
int val0 = _mm_extract_epi32 (intVec, 0);
1456+
int val1 = _mm_extract_epi32 (intVec, 1);
1457+
dest[i] = (double) val0 * multiplier;
1458+
dest[i + 1] = (double) val1 * multiplier;
1459+
}
1460+
}
1461+
1462+
for (Size i = numLongs; i < num; ++i)
1463+
dest[i] = (double) src[i] * multiplier;
1464+
#else
1465+
for (Size i = 0; i < num; ++i)
1466+
dest[i] = (double) src[i] * multiplier;
1467+
1468+
#endif
1469+
}
1470+
1471+
template <typename Size>
1472+
void convertFloatToFixed (int* dest, const double* src, double multiplier, Size num) noexcept
1473+
{
1474+
#if YUP_USE_VDSP_FRAMEWORK
1475+
constexpr Size kStackBufferSize = 256;
1476+
double stackBuffer[kStackBufferSize];
1477+
1478+
if (num <= kStackBufferSize)
1479+
{
1480+
vDSP_vsmulD (src, 1, &multiplier, stackBuffer, 1, (vDSP_Length) num);
1481+
vDSP_vfix32D (stackBuffer, 1, reinterpret_cast<int*> (dest), 1, (vDSP_Length) num);
1482+
}
1483+
else
1484+
{
1485+
for (Size i = 0; i < num; i += kStackBufferSize)
1486+
{
1487+
const Size currentChunk = jmin (kStackBufferSize, num - i);
1488+
vDSP_vsmulD (src + i, 1, &multiplier, stackBuffer, 1, (vDSP_Length) currentChunk);
1489+
vDSP_vfix32D (stackBuffer, 1, reinterpret_cast<int*> (dest + i), 1, (vDSP_Length) currentChunk);
1490+
}
1491+
}
1492+
#elif YUP_USE_ARM_NEON
1493+
const auto numLongs = num & ~1;
1494+
1495+
if (numLongs != 0)
1496+
{
1497+
for (Size i = 0; i < numLongs; i += 2)
1498+
{
1499+
float64x2_t doubleVec = vld1q_f64 (src + i);
1500+
float64x2_t scaledVec = vmulq_n_f64 (doubleVec, multiplier);
1501+
double d0 = vgetq_lane_f64 (scaledVec, 0);
1502+
double d1 = vgetq_lane_f64 (scaledVec, 1);
1503+
dest[i] = (int) d0;
1504+
dest[i + 1] = (int) d1;
1505+
}
1506+
}
1507+
1508+
for (Size i = numLongs; i < num; ++i)
1509+
dest[i] = (int) (src[i] * multiplier);
1510+
#elif YUP_USE_SSE_INTRINSICS
1511+
const auto numLongs = num & ~1;
1512+
const __m128d mult = _mm_set1_pd (multiplier);
1513+
1514+
if (numLongs != 0)
1515+
{
1516+
for (Size i = 0; i < numLongs; i += 2)
1517+
{
1518+
__m128d doubleVec = _mm_loadu_pd (src + i);
1519+
__m128d scaledVec = _mm_mul_pd (doubleVec, mult);
1520+
__m128i intVec = _mm_cvtpd_epi32 (scaledVec);
1521+
_mm_storel_epi64 (reinterpret_cast<__m128i*> (dest + i), intVec);
1522+
}
1523+
}
13951524

1525+
for (Size i = numLongs; i < num; ++i)
1526+
dest[i] = (int) (src[i] * multiplier);
13961527
#else
13971528
for (Size i = 0; i < num; ++i)
13981529
dest[i] = (int) (src[i] * multiplier);
@@ -1639,30 +1770,32 @@ FloatType YUP_CALLTYPE FloatVectorOperationsBase<FloatType, CountType>::findMaxi
16391770
return FloatVectorHelpers::findMaximum (src, numValues);
16401771
}
16411772

1642-
template struct FloatVectorOperationsBase<float, int>;
1643-
template struct FloatVectorOperationsBase<float, size_t>;
1644-
template struct FloatVectorOperationsBase<double, int>;
1645-
template struct FloatVectorOperationsBase<double, size_t>;
1646-
1647-
void YUP_CALLTYPE FloatVectorOperations::convertFixedToFloat (float* dest, const int* src, float multiplier, size_t num) noexcept
1773+
template <typename FloatType, typename CountType>
1774+
void YUP_CALLTYPE FloatVectorOperationsBase<FloatType, CountType>::convertFixedToFloat (FloatType* dest,
1775+
const int* src,
1776+
FloatType multiplier,
1777+
CountType numValues) noexcept
16481778
{
1649-
FloatVectorHelpers::convertFixedToFloat (dest, src, multiplier, num);
1779+
FloatVectorHelpers::convertFixedToFloat (dest, src, multiplier, numValues);
16501780
}
16511781

1652-
void YUP_CALLTYPE FloatVectorOperations::convertFixedToFloat (float* dest, const int* src, float multiplier, int num) noexcept
1782+
template <typename FloatType, typename CountType>
1783+
void YUP_CALLTYPE FloatVectorOperationsBase<FloatType, CountType>::convertFloatToFixed (int* dest,
1784+
const FloatType* src,
1785+
FloatType multiplier,
1786+
CountType numValues) noexcept
16531787
{
1654-
FloatVectorHelpers::convertFixedToFloat (dest, src, multiplier, num);
1788+
FloatVectorHelpers::convertFloatToFixed (dest, src, multiplier, numValues);
16551789
}
16561790

1657-
void YUP_CALLTYPE FloatVectorOperations::convertFloatToFixed (int* dest, const float* src, float multiplier, size_t num) noexcept
1658-
{
1659-
FloatVectorHelpers::convertFloatToFixed (dest, src, multiplier, num);
1660-
}
1791+
//==============================================================================
16611792

1662-
void YUP_CALLTYPE FloatVectorOperations::convertFloatToFixed (int* dest, const float* src, float multiplier, int num) noexcept
1663-
{
1664-
FloatVectorHelpers::convertFloatToFixed (dest, src, multiplier, num);
1665-
}
1793+
template struct FloatVectorOperationsBase<float, int>;
1794+
template struct FloatVectorOperationsBase<float, size_t>;
1795+
template struct FloatVectorOperationsBase<double, int>;
1796+
template struct FloatVectorOperationsBase<double, size_t>;
1797+
1798+
//==============================================================================
16661799

16671800
intptr_t YUP_CALLTYPE FloatVectorOperations::getFpStatusRegister() noexcept
16681801
{
@@ -1770,6 +1903,8 @@ bool YUP_CALLTYPE FloatVectorOperations::areDenormalsDisabled() noexcept
17701903
#endif
17711904
}
17721905

1906+
//==============================================================================
1907+
17731908
ScopedNoDenormals::ScopedNoDenormals() noexcept
17741909
{
17751910
#if YUP_USE_SSE_INTRINSICS || (YUP_USE_ARM_NEON || (YUP_64BIT && YUP_ARM))

‎modules/yup_audio_basics/buffers/yup_FloatVectorOperations.h‎

Lines changed: 9 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -155,6 +155,12 @@ struct FloatVectorOperationsBase
155155

156156
/** Finds the maximum value in the given array. */
157157
static FloatType YUP_CALLTYPE findMaximum (const FloatType* src, CountType numValues) noexcept;
158+
159+
/** Converts an array of fixed-point integers to floating point values. */
160+
static void YUP_CALLTYPE convertFixedToFloat (FloatType* dest, const int* src, FloatType multiplier, CountType numValues) noexcept;
161+
162+
/** Converts an array of floating point values to fixed-point integers. */
163+
static void YUP_CALLTYPE convertFloatToFixed (int* dest, const FloatType* src, FloatType multiplier, CountType numValues) noexcept;
158164
};
159165

160166
#if ! DOXYGEN
@@ -180,7 +186,9 @@ struct NameForwarder : public Bases...
180186
Bases::clip...,
181187
Bases::findMinAndMax...,
182188
Bases::findMinimum...,
183-
Bases::findMaximum...;
189+
Bases::findMaximum...,
190+
Bases::convertFixedToFloat...,
191+
Bases::convertFloatToFixed...;
184192
};
185193

186194
} // namespace detail
@@ -199,13 +207,6 @@ struct NameForwarder : public Bases...
199207
class YUP_API FloatVectorOperations : public detail::NameForwarder<FloatVectorOperationsBase<float, int>, FloatVectorOperationsBase<float, size_t>, FloatVectorOperationsBase<double, int>, FloatVectorOperationsBase<double, size_t>>
200208
{
201209
public:
202-
static void YUP_CALLTYPE convertFixedToFloat (float* dest, const int* src, float multiplier, int num) noexcept;
203-
204-
static void YUP_CALLTYPE convertFixedToFloat (float* dest, const int* src, float multiplier, size_t num) noexcept;
205-
206-
static void YUP_CALLTYPE convertFloatToFixed (int* dest, const float* src, float multiplier, int num) noexcept;
207-
208-
static void YUP_CALLTYPE convertFloatToFixed (int* dest, const float* src, float multiplier, size_t num) noexcept;
209210

210211
/** This method enables or disables the SSE/NEON flush-to-zero mode. */
211212
static void YUP_CALLTYPE enableFlushToZeroMode (bool shouldEnable) noexcept;

0 commit comments

Comments
 (0)