NEON fast path for box blur

src/opts/SkBlurImage_opts_neon.cpp

 /*
  * Copyright 2013 The Android Open Source Project
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 
 #include "SkBitmap.h"
 #include "SkColorPriv.h"
 #include "SkBlurImage_opts.h"
 #include "SkRect.h"
 
 #include <arm_neon.h>
 
 namespace {
 
 enum BlurDirection {
     kX, kY
 };
 
 /**
+ * Helper function to load 2 pixels from diffent rows to a 8x8 NEON register
+ * and also pre-load pixels for future read
+ */
+template<BlurDirection srcDirection>
+inline uint8x8_t load_2_pixels(const SkPMColor* src, int srcStride) {
+    if (srcDirection == kX) {
+        uint32x2_t temp = vdup_n_u32(0);
+        // 10% faster by add these 2 prefetches
+        SK_PREFETCH(src + 16);
+        SK_PREFETCH(src + srcStride + 16);
+        return vreinterpret_u8_u32(vld1_lane_u32(src + srcStride, vld1_lane_u32(src, temp, 0), 1));
+    } else {
+        // this prefetch doesn't impact the performance
+        SK_PREFETCH(src + srcStride);
+        return vld1_u8((uint8_t*)src);
+    }
+}
+
+/**
+ * Helper function to store the low 8-bits from a 16x8 NEON register to 2 rows
+ */
+template<BlurDirection dstDirection>
+inline void store_2_pixels(uint16x8_t result16x8, SkPMColor* dst, int dstStride) {
+    if (dstDirection == kX) {
+        uint32x2_t temp = vreinterpret_u32_u8(vmovn_u16(result16x8));
+        vst1_lane_u32(dst, temp, 0);
+        vst1_lane_u32(dst + dstStride, temp, 1);
+    } else {
+        uint8x8_t temp = vmovn_u16(result16x8);
+        vst1_u8((uint8_t*)dst, temp);
+    }
+}
+
+/**
+ * fast path for kernel size less than 128
+ */
+template<BlurDirection srcDirection, BlurDirection dstDirection>
+void SkDoubleRowBoxBlur_NEON(const SkPMColor** src, int srcStride, SkPMColor** dst, int kernelSize,
+                        int leftOffset, int rightOffset, int width, int* height)
+{
+    const int rightBorder = SkMin32(rightOffset + 1, width);
+    const int srcStrideX = srcDirection == kX ? 1 : srcStride;
+    const int dstStrideX = dstDirection == kX ? 1 : *height;
+    const int srcStrideY = srcDirection == kX ? srcStride : 1;
+    const int dstStrideY = dstDirection == kX ? width : 1;
+    const uint16x8_t scale = vdupq_n_u16((1 << 15) / kernelSize);
+
+    for (; *height >= 2; *height -= 2) {
+        uint16x8_t sum = vdupq_n_u16(0);
+        const SkPMColor* p = *src;
+        for (int i = 0; i < rightBorder; i++) {
+            sum = vaddw_u8(sum,
+                load_2_pixels<srcDirection>(p, srcStride));
+            p += srcStrideX;
+        }
+
+        const SkPMColor* sptr = *src;
+        SkPMColor* dptr = *dst;
+        for (int x = 0; x < width; x++) {
+            // val = (sum * scale * 2 + 0x8000) >> 16
+            uint16x8_t resultPixles = vreinterpretq_u16_s16(vqrdmulhq_s16(

[kevin.petit.not.used.account, 2013/12/12 15:29:32]

typo: "Pixles" => "Pixels". Otherwise, nice use of vqrdmulh :-).

[zheng.xu, 2013/12/13 04:31:59]

Done.

+                vreinterpretq_s16_u16(sum), vreinterpretq_s16_u16(scale)));
+            store_2_pixels<dstDirection>(resultPixles, dptr, width);
+
+            if (x >= leftOffset) {
+                sum = vsubw_u8(sum, 

[kevin.petit.not.used.account, 2013/12/12 15:29:32]

stray space character at the end of the line

[zheng.xu, 2013/12/13 04:31:59]

Done.

+                    load_2_pixels<srcDirection>(sptr - leftOffset * srcStrideX, srcStride));
+            }
+            if (x + rightOffset + 1 < width) {
+                sum = vaddw_u8(sum, 

[kevin.petit.not.used.account, 2013/12/12 15:29:32]

stray space character at the end of the line

[zheng.xu, 2013/12/13 04:31:59]

Done.

+                    load_2_pixels<srcDirection>(sptr + (rightOffset + 1) * srcStrideX, srcStride));
+            }
+            sptr += srcStrideX;
+            dptr += dstStrideX;
+        }
+        *src += srcStrideY * 2;
+        *dst += dstStrideY * 2;
+    }
+}
+
+
+/**
  * Helper function to spread the components of a 32-bit integer into the
  * lower 8 bits of each 16-bit element of a NEON register.
  */
 
 static inline uint16x4_t expand(uint32_t a) {
     // ( ARGB ) -> ( ARGB ARGB ) -> ( A R G B A R G B )
     uint8x8_t v8 = vreinterpret_u8_u32(vdup_n_u32(a));
     // ( A R G B A R G B ) -> ( 0A 0R 0G 0B 0A 0R 0G 0B ) -> ( 0A 0R 0G 0B )
     return vget_low_u16(vmovl_u8(v8));
 }
 
 template<BlurDirection srcDirection, BlurDirection dstDirection>
 void SkBoxBlur_NEON(const SkPMColor* src, int srcStride, SkPMColor* dst, int kernelSize,
                     int leftOffset, int rightOffset, int width, int height)
 {
     const int rightBorder = SkMin32(rightOffset + 1, width);
     const int srcStrideX = srcDirection == kX ? 1 : srcStride;
     const int dstStrideX = dstDirection == kX ? 1 : height;
     const int srcStrideY = srcDirection == kX ? srcStride : 1;
     const int dstStrideY = dstDirection == kX ? width : 1;
     const uint32x4_t scale = vdupq_n_u32((1 << 24) / kernelSize);
     const uint32x4_t half = vdupq_n_u32(1 << 23);
-    for (int y = 0; y < height; ++y) {
+
+    if (kernelSize < 128)
+    {
+        SkDoubleRowBoxBlur_NEON<srcDirection, dstDirection>(&src, srcStride, &dst, kernelSize,
+            leftOffset, rightOffset, width, &height);
+    }
+
+    for (; height > 0; height--) {

[kevin.petit.not.used.account, 2013/12/12 15:29:32]

That means the last row in the case of an odd count will be processed in a
different way with different precision/rounding. Is this a problem?

[zheng.xu, 2013/12/13 04:31:59]

I think it is not. Because this algorithm is already an approximation for
Gausscian Blur, the small differences should be acceptable.

         uint32x4_t sum = vdupq_n_u32(0);
         const SkPMColor* p = src;
         for (int i = 0; i < rightBorder; ++i) {
             sum = vaddw_u16(sum, expand(*p));
             p += srcStrideX;
         }
 
         const SkPMColor* sptr = src;
         SkPMColor* dptr = dst;
         for (int x = 0; x < width; ++x) {
@@ skipping 14 matching lines @@
 
             if (x >= leftOffset) {
                 const SkPMColor* l = sptr - leftOffset * srcStrideX;
                 sum = vsubw_u16(sum, expand(*l));
             }
             if (x + rightOffset + 1 < width) {
                 const SkPMColor* r = sptr + (rightOffset + 1) * srcStrideX;
                 sum = vaddw_u16(sum, expand(*r));
             }
             sptr += srcStrideX;
-            if (srcDirection == kY) {
+            if (srcDirection == kX) {
+                SK_PREFETCH(sptr + (rightOffset + 16) * srcStrideX);
+            } else {
                 SK_PREFETCH(sptr + (rightOffset + 1) * srcStrideX);
             }
             dptr += dstStrideX;
         }
         src += srcStrideY;
         dst += dstStrideY;
     }
 }
 
 } // namespace
 
 bool SkBoxBlurGetPlatformProcs_NEON(SkBoxBlurProc* boxBlurX,
                                     SkBoxBlurProc* boxBlurY,
                                     SkBoxBlurProc* boxBlurXY,
                                     SkBoxBlurProc* boxBlurYX) {
     *boxBlurX = SkBoxBlur_NEON<kX, kX>;
     *boxBlurY = SkBoxBlur_NEON<kY, kY>;
     *boxBlurXY = SkBoxBlur_NEON<kX, kY>;
     *boxBlurYX = SkBoxBlur_NEON<kY, kX>;
     return true;
 }