Modified: trunk/Source/WebCore/platform/audio/StereoPanner.cpp (268538 => 268539)
--- trunk/Source/WebCore/platform/audio/StereoPanner.cpp 2020-10-15 18:19:42 UTC (rev 268538)
+++ trunk/Source/WebCore/platform/audio/StereoPanner.cpp 2020-10-15 18:21:25 UTC (rev 268539)
@@ -29,6 +29,7 @@
#if ENABLE(WEB_AUDIO)
+#include "VectorMath.h"
#include <wtf/MathExtras.h>
namespace WebCore {
@@ -116,11 +117,9 @@
if (!sourceL || !sourceR || !destinationL || !destinationR)
return;
-
+
float targetPan = clampTo(panValue, -1.0, 1.0);
- int n = framesToProcess;
-
if (numberOfInputChannels == 1) {
double panRadian = (targetPan * 0.5 + 0.5) * piOverTwoDouble;
@@ -127,27 +126,20 @@
double gainL = cos(panRadian);
double gainR = sin(panRadian);
- while (n--) {
- float inputL = *sourceL++;
- *destinationL++ = static_cast<float>(inputL * gainL);
- *destinationR++ = static_cast<float>(inputL * gainR);
- }
+ VectorMath::multiplyByScalar(sourceL, gainL, destinationL, framesToProcess);
+ VectorMath::multiplyByScalar(sourceL, gainR, destinationR, framesToProcess);
} else {
double panRadian = (targetPan <= 0 ? targetPan + 1 : targetPan) * piOverTwoDouble;
double gainL = cos(panRadian);
double gainR = sin(panRadian);
-
- while (n--) {
- float inputL = *sourceL++;
- float inputR = *sourceR++;
- if (targetPan <= 0) {
- *destinationL++ = static_cast<float>(inputL + inputR * gainL);
- *destinationR++ = static_cast<float>(inputR * gainR);
- } else {
- *destinationL++ = static_cast<float>(inputL * gainL);
- *destinationR++ = static_cast<float>(inputR + inputL * gainR);
- }
+
+ if (targetPan <= 0) {
+ VectorMath::multiplyByScalarThenAddToVector(sourceR, gainL, sourceL, destinationL, framesToProcess);
+ VectorMath::multiplyByScalar(sourceR, gainR, destinationR, framesToProcess);
+ } else {
+ VectorMath::multiplyByScalar(sourceL, gainL, destinationL, framesToProcess);
+ VectorMath::multiplyByScalarThenAddToVector(sourceL, gainR, sourceR, destinationR, framesToProcess);
}
}
}
Modified: trunk/Source/WebCore/platform/audio/VectorMath.cpp (268538 => 268539)
--- trunk/Source/WebCore/platform/audio/VectorMath.cpp 2020-10-15 18:19:42 UTC (rev 268538)
+++ trunk/Source/WebCore/platform/audio/VectorMath.cpp 2020-10-15 18:21:25 UTC (rev 268539)
@@ -52,9 +52,9 @@
#if USE(ACCELERATE)
// On the Mac we use the highly optimized versions in Accelerate.framework
-void multiplyByScalar(const float* inputVector, float scale, float* outputVector, size_t numberOfElementsToProcess)
+void multiplyByScalar(const float* inputVector, float scalar, float* outputVector, size_t numberOfElementsToProcess)
{
- vDSP_vsmul(inputVector, 1, &scale, outputVector, 1, numberOfElementsToProcess);
+ vDSP_vsmul(inputVector, 1, &scalar, outputVector, 1, numberOfElementsToProcess);
}
void add(const float* inputVector1, const float* inputVector2, float* outputVector, size_t numberOfElementsToProcess)
@@ -86,11 +86,16 @@
vDSP_zvmul(&sc1, 1, &sc2, 1, &dest, 1, numberOfElementsToProcess, 1);
}
-void multiplyByScalarThenAddToOutput(const float* inputVector, float scale, float* outputVector, size_t numberOfElementsToProcess)
+void multiplyByScalarThenAddToOutput(const float* inputVector, float scalar, float* outputVector, size_t numberOfElementsToProcess)
{
- vDSP_vsma(inputVector, 1, &scale, outputVector, 1, outputVector, 1, numberOfElementsToProcess);
+ vDSP_vsma(inputVector, 1, &scalar, outputVector, 1, outputVector, 1, numberOfElementsToProcess);
}
+void multiplyByScalarThenAddToVector(const float* inputVector1, float scalar, const float* inputVector2, float* outputVector, size_t numberOfElementsToProcess)
+{
+ vDSP_vsma(inputVector1, 1, &scalar, inputVector2, 1, outputVector, 1, numberOfElementsToProcess);
+}
+
void addVectorsThenMultiplyByScalar(const float* inputVector1, const float* inputVector2, float scalar, float* outputVector, size_t numberOfElementsToProcess)
{
vDSP_vasm(inputVector1, 1, inputVector2, 1, &scalar, outputVector, 1, numberOfElementsToProcess);
@@ -128,14 +133,20 @@
return !(reinterpret_cast<uintptr_t>(vector) & 0x0F);
}
-void multiplyByScalarThenAddToOutput(const float* inputVector, float scale, float* outputVector, size_t numberOfElementsToProcess)
+void multiplyByScalarThenAddToVector(const float* inputVector1, float scalar, const float* inputVector2, float* outputVector, size_t numberOfElementsToProcess)
{
+ multiplyByScalarThenAddToOutput(inputVector1, scalar, outputVector, numberOfElementsToProcess);
+ add(outputVector, inputVector2, outputVector, numberOfElementsToProcess);
+}
+
+void multiplyByScalarThenAddToOutput(const float* inputVector, float scalar, float* outputVector, size_t numberOfElementsToProcess)
+{
size_t n = numberOfElementsToProcess;
#if CPU(X86_SSE2)
// If the inputVector address is not 16-byte aligned, the first several frames (at most three) should be processed separately.
while (!is16ByteAligned(inputVector) && n) {
- *outputVector += scale * *inputVector;
+ *outputVector += scalar * *inputVector;
inputVector++;
outputVector++;
n--;
@@ -148,7 +159,7 @@
__m128 pSource;
__m128 dest;
__m128 temp;
- __m128 mScale = _mm_set_ps1(scale);
+ __m128 mScale = _mm_set_ps1(scalar);
bool destAligned = is16ByteAligned(outputVector);
@@ -174,7 +185,7 @@
size_t tailFrames = n % 4;
const float* endP = outputVector + n - tailFrames;
- float32x4_t k = vdupq_n_f32(scale);
+ float32x4_t k = vdupq_n_f32(scalar);
while (outputVector < endP) {
float32x4_t source = vld1q_f32(inputVector);
float32x4_t dest = vld1q_f32(outputVector);
@@ -188,13 +199,13 @@
n = tailFrames;
#endif
while (n--) {
- *outputVector += *inputVector * scale;
+ *outputVector += *inputVector * scalar;
++inputVector;
++outputVector;
}
}
-void multiplyByScalar(const float* inputVector, float scale, float* outputVector, size_t numberOfElementsToProcess)
+void multiplyByScalar(const float* inputVector, float scalar, float* outputVector, size_t numberOfElementsToProcess)
{
size_t n = numberOfElementsToProcess;
@@ -201,7 +212,7 @@
#if CPU(X86_SSE2)
// If the inputVector address is not 16-byte aligned, the first several frames (at most three) should be processed separately.
while (!is16ByteAligned(inputVector) && n) {
- *outputVector = scale * *inputVector;
+ *outputVector = scalar * *inputVector;
inputVector++;
outputVector++;
n--;
@@ -209,7 +220,7 @@
// Now the inputVector address is aligned and start to apply SSE.
size_t group = n / 4;
- __m128 mScale = _mm_set_ps1(scale);
+ __m128 mScale = _mm_set_ps1(scalar);
__m128* pSource;
__m128* pDest;
__m128 dest;
@@ -243,7 +254,7 @@
while (outputVector < endP) {
float32x4_t source = vld1q_f32(inputVector);
- vst1q_f32(outputVector, vmulq_n_f32(source, scale));
+ vst1q_f32(outputVector, vmulq_n_f32(source, scalar));
inputVector += 4;
outputVector += 4;
@@ -251,7 +262,7 @@
n = tailFrames;
#endif
while (n--) {
- *outputVector = scale * *inputVector;
+ *outputVector = scalar * *inputVector;
++inputVector;
++outputVector;
}
Modified: trunk/Source/WebCore/platform/audio/VectorMath.h (268538 => 268539)
--- trunk/Source/WebCore/platform/audio/VectorMath.h 2020-10-15 18:19:42 UTC (rev 268538)
+++ trunk/Source/WebCore/platform/audio/VectorMath.h 2020-10-15 18:21:25 UTC (rev 268539)
@@ -31,15 +31,20 @@
namespace VectorMath {
-// Multiples inputVector by scalar then adds the result to outputVector (vsma).
+// Multiples inputVector by scalar then adds the result to outputVector (simplified vsma).
// for (n = 0; n < numberOfElementsToProcess; ++n)
-// outputVector[n] += inputVector[n] * scale;
-void multiplyByScalarThenAddToOutput(const float* inputVector, float scale, float* outputVector, size_t numberOfElementsToProcess);
+// outputVector[n] += inputVector[n] * scalar;
+void multiplyByScalarThenAddToOutput(const float* inputVector, float scalar, float* outputVector, size_t numberOfElementsToProcess);
+// Adds a vector inputVector2 to the product of a scalar value and a single-precision vector inputVector1 (vsma).
+// for (n = 0; n < numberOfElementsToProcess; ++n)
+// outputVector[n] = inputVector1[n] * scalar + inputVector2[n];
+void multiplyByScalarThenAddToVector(const float* inputVector1, float scalar, const float* inputVector2, float* outputVector, size_t numberOfElementsToProcess);
+
// Multiplies the sum of two vectors by a scalar value (vasm).
void addVectorsThenMultiplyByScalar(const float* inputVector1, const float* inputVector2, float scalar, float* outputVector, size_t numberOfElementsToProcess);
-void multiplyByScalar(const float* inputVector, float scale, float* outputVector, size_t numberOfElementsToProcess);
+void multiplyByScalar(const float* inputVector, float scalar, float* outputVector, size_t numberOfElementsToProcess);
void addScalar(const float* inputVector, float scalar, float* outputVector, size_t numberOfElementsToProcess);
void add(const float* inputVector1, const float* inputVector2, float* outputVector, size_t numberOfElementsToProcess);
