[WIP] support arm_neon_optional flag in blink.

Source/platform/audio/cpu/arm/VectorMathNEON.cpp

Index: Source/platform/audio/cpu/arm/VectorMathNEON.cpp
diff --git a/Source/platform/audio/cpu/arm/VectorMathNEON.cpp b/Source/platform/audio/cpu/arm/VectorMathNEON.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..643283debb62a1b55aae1aefebb031de20ecc246
--- /dev/null
+++ b/Source/platform/audio/cpu/arm/VectorMathNEON.cpp
@@ -0,0 +1,300 @@
+/*
+ * Copyright (C) 2010, Google Inc. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1.  Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2.  Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY APPLE INC. AND ITS CONTRIBUTORS ``AS IS'' AND ANY
+ * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR ITS CONTRIBUTORS BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "config.h"
+
+#if ENABLE(WEB_AUDIO)
+
+#include "platform/audio/cpu/arm/VectorMathNEON.h"
+
+#include "wtf/Assertions.h"
+#include "wtf/CPU.h"
+#include <algorithm>
+#include <arm_neon.h>
+
+namespace blink {
+
+namespace VectorMath {
+
+void vsma(const float* sourceP, int sourceStride, const float* scale, float* destP, int destStride, size_t framesToProcess)
+{
+    int n = framesToProcess;
+
+    if (WTF_CPU_ARM_HAS_NEON()) {
+        if ((sourceStride == 1) && (destStride == 1)) {
+            int tailFrames = n % 4;
+            const float* endP = destP + n - tailFrames;
+
+            float32x4_t k = vdupq_n_f32(*scale);
+            while (destP < endP) {
+                float32x4_t source = vld1q_f32(sourceP);
+                float32x4_t dest = vld1q_f32(destP);
+
+                dest = vmlaq_f32(dest, source, k);
+                vst1q_f32(destP, dest);
+
+                sourceP += 4;
+                destP += 4;
+            }
+            n = tailFrames;
+        }
+    }
+
+    while (n) {
+        *destP += *sourceP * *scale;
+        sourceP += sourceStride;
+        destP += destStride;
+        n--;
+    }
+}
+
+void vsmul(const float* sourceP, int sourceStride, const float* scale, float* destP, int destStride, size_t framesToProcess)
+{
+    int n = framesToProcess;
+
+    if (WTF_CPU_ARM_HAS_NEON()) {
+        if ((sourceStride == 1) && (destStride == 1)) {
+            float k = *scale;
+            int tailFrames = n % 4;
+            const float* endP = destP + n - tailFrames;
+
+            while (destP < endP) {
+                float32x4_t source = vld1q_f32(sourceP);
+                vst1q_f32(destP, vmulq_n_f32(source, k));
+
+                sourceP += 4;
+                destP += 4;
+            }
+            n = tailFrames;
+        }
+    }
+
+    float k = *scale;
+    while (n--) {
+        *destP = k * *sourceP;
+        sourceP += sourceStride;
+        destP += destStride;
+    }
+}
+
+void vadd(const float* source1P, int sourceStride1, const float* source2P, int sourceStride2, float* destP, int destStride, size_t framesToProcess)
+{
+    int n = framesToProcess;
+
+    if (WTF_CPU_ARM_HAS_NEON()) {
+        if ((sourceStride1 ==1) && (sourceStride2 == 1) && (destStride == 1)) {
+            int tailFrames = n % 4;
+            const float* endP = destP + n - tailFrames;
+
+            while (destP < endP) {
+                float32x4_t source1 = vld1q_f32(source1P);
+                float32x4_t source2 = vld1q_f32(source2P);
+                vst1q_f32(destP, vaddq_f32(source1, source2));
+
+                source1P += 4;
+                source2P += 4;
+                destP += 4;
+            }
+            n = tailFrames;
+        }
+    }
+
+    while (n--) {
+        *destP = *source1P + *source2P;
+        source1P += sourceStride1;
+        source2P += sourceStride2;
+        destP += destStride;
+    }
+}
+
+void vmul(const float* source1P, int sourceStride1, const float* source2P, int sourceStride2, float* destP, int destStride, size_t framesToProcess)
+{
+    int n = framesToProcess;
+
+    if (WTF_CPU_ARM_HAS_NEON()) {
+        if ((sourceStride1 ==1) && (sourceStride2 == 1) && (destStride == 1)) {
+            int tailFrames = n % 4;
+            const float* endP = destP + n - tailFrames;
+
+            while (destP < endP) {
+                float32x4_t source1 = vld1q_f32(source1P);
+                float32x4_t source2 = vld1q_f32(source2P);
+                vst1q_f32(destP, vmulq_f32(source1, source2));
+
+                source1P += 4;
+                source2P += 4;
+                destP += 4;
+            }
+            n = tailFrames;
+        }
+    }
+
+    while (n) {
+        *destP = *source1P * *source2P;
+        source1P += sourceStride1;
+        source2P += sourceStride2;
+        destP += destStride;
+        n--;
+    }
+}
+
+void zvmul(const float* real1P, const float* imag1P, const float* real2P, const float* imag2P, float* realDestP, float* imagDestP, size_t framesToProcess)
+{
+    unsigned i = 0;
+
+    unsigned endSize = framesToProcess - framesToProcess % 4;
+
+    if (WTF_CPU_ARM_HAS_NEON()) {
+        while (i < endSize) {
+            float32x4_t real1 = vld1q_f32(real1P + i);
+            float32x4_t real2 = vld1q_f32(real2P + i);
+            float32x4_t imag1 = vld1q_f32(imag1P + i);
+            float32x4_t imag2 = vld1q_f32(imag2P + i);
+
+            float32x4_t realResult = vmlsq_f32(vmulq_f32(real1, real2), imag1, imag2);
+            float32x4_t imagResult = vmlaq_f32(vmulq_f32(real1, imag2), imag1, real2);
+
+            vst1q_f32(realDestP + i, realResult);
+            vst1q_f32(imagDestP + i, imagResult);
+
+            i += 4;
+        }
+    }
+    for (; i < framesToProcess; ++i) {
+        // Read and compute result before storing them, in case the
+        // destination is the same as one of the sources.
+        float realResult = real1P[i] * real2P[i] - imag1P[i] * imag2P[i];
+        float imagResult = real1P[i] * imag2P[i] + imag1P[i] * real2P[i];
+
+        realDestP[i] = realResult;
+        imagDestP[i] = imagResult;
+    }
+}
+
+void vsvesq(const float* sourceP, int sourceStride, float* sumP, size_t framesToProcess)
+{
+    int n = framesToProcess;
+    float sum = 0;
+
+    if (WTF_CPU_ARM_HAS_NEON()) {
+        if (sourceStride == 1) {
+            int tailFrames = n % 4;
+            const float* endP = sourceP + n - tailFrames;
+
+            float32x4_t fourSum = vdupq_n_f32(0);
+            while (sourceP < endP) {
+                float32x4_t source = vld1q_f32(sourceP);
+                fourSum = vmlaq_f32(fourSum, source, source);
+                sourceP += 4;
+            }
+            float32x2_t twoSum = vadd_f32(vget_low_f32(fourSum), vget_high_f32(fourSum));
+
+            float groupSum[2];
+            vst1_f32(groupSum, twoSum);
+            sum += groupSum[0] + groupSum[1];
+
+            n = tailFrames;
+        }
+    }
+
+    while (n--) {
+        float sample = *sourceP;
+        sum += sample * sample;
+        sourceP += sourceStride;
+    }
+
+    ASSERT(sumP);
+    *sumP = sum;
+}
+
+void vmaxmgv(const float* sourceP, int sourceStride, float* maxP, size_t framesToProcess)
+{
+    int n = framesToProcess;
+    float max = 0;
+
+    if (WTF_CPU_ARM_HAS_NEON()) {
+        if (sourceStride == 1) {
+            int tailFrames = n % 4;
+            const float* endP = sourceP + n - tailFrames;
+
+            float32x4_t fourMax = vdupq_n_f32(0);
+            while (sourceP < endP) {
+                float32x4_t source = vld1q_f32(sourceP);
+                fourMax = vmaxq_f32(fourMax, vabsq_f32(source));
+                sourceP += 4;
+            }
+            float32x2_t twoMax = vmax_f32(vget_low_f32(fourMax), vget_high_f32(fourMax));
+
+            float groupMax[2];
+            vst1_f32(groupMax, twoMax);
+            max = std::max(groupMax[0], groupMax[1]);
+
+            n = tailFrames;
+        }
+    }
+
+    while (n--) {
+        max = std::max(max, fabsf(*sourceP));
+        sourceP += sourceStride;
+    }
+
+    ASSERT(maxP);
+    *maxP = max;
+}
+
+void vclip(const float* sourceP, int sourceStride, const float* lowThresholdP, const float* highThresholdP, float* destP, int destStride, size_t framesToProcess)
+{
+    int n = framesToProcess;
+    float lowThreshold = *lowThresholdP;
+    float highThreshold = *highThresholdP;
+
+    if (WTF_CPU_ARM_HAS_NEON()) {
+        if ((sourceStride == 1) && (destStride == 1)) {
+            int tailFrames = n % 4;
+            const float* endP = destP + n - tailFrames;
+
+            float32x4_t low = vdupq_n_f32(lowThreshold);
+            float32x4_t high = vdupq_n_f32(highThreshold);
+            while (destP < endP) {
+                float32x4_t source = vld1q_f32(sourceP);
+                vst1q_f32(destP, vmaxq_f32(vminq_f32(source, high), low));
+                sourceP += 4;
+                destP += 4;
+            }
+            n = tailFrames;
+        }
+    }
+
+    while (n--) {
+        *destP = std::max(std::min(*sourceP, highThreshold), lowThreshold);
+        sourceP += sourceStride;
+        destP += destStride;
+    }
+}
+
+} // namespace VectorMath
+
+} // namespace blink
+
+#endif // ENABLE(WEB_AUDIO)