SkScaledCodec class

src/codec/SkJpegCodec.cpp

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include "SkCodec.h"
 #include "SkJpegCodec.h"
 #include "SkJpegDecoderMgr.h"
 #include "SkJpegUtility_codec.h"
 #include "SkCodecPriv.h"
 #include "SkColorPriv.h"
+#include "SkScaledCodec.h"
 #include "SkScanlineDecoder.h"
 #include "SkStream.h"
 #include "SkTemplates.h"
 #include "SkTypes.h"
 
 // stdio is needed for libjpeg-turbo
 #include <stdio.h>
 
 extern "C" {
     #include "jpeglibmangler.h"
@@ skipping 158 matching lines @@
     dinfo.global_state = DSTATE_READY;
     dinfo.num_components = 0;
     dinfo.scale_num = num;
     dinfo.scale_denom = denom;
     turbo_jpeg_calc_output_dimensions(&dinfo);
 
     // Return the calculated output dimensions for the given scale
     return SkISize::Make(dinfo.output_width, dinfo.output_height);
 }
 
+// The swizzler is only used for sampling in the x direction
+// when sampling a scanlineDecoder is used. The swizzler is not used in onGetPixels
+SkCodec::Result SkJpegCodec::initializeSwizzler(const SkImageInfo& requestedInfo,
+                                               void* dst, size_t rowBytes,
+                                               const Options& options,
+                                               SkPMColor ctable[],
+                                               int* ctableCount) {
+
+    const SkColorType srcColorType = requestedInfo.colorType();
+    SkSwizzler::SrcConfig srcConfig;
+    switch (srcColorType) {
+        case kGray_8_SkColorType:
+            srcConfig = SkSwizzler::kGray;
+            break;
+        case kRGBA_8888_SkColorType:
+            srcConfig = SkSwizzler::kRGBX;
+            break;
+        case kBGRA_8888_SkColorType:
+            srcConfig = SkSwizzler::kBGRX;
+            break;
+        case kRGB_565_SkColorType:
+            srcConfig = SkSwizzler::kRGB_565;
+            break;
+        default:
+            //would have exited before now if the colorType was supported by jpeg
+            SkASSERT(false);
+    }
+
+    fSwizzler.reset(SkSwizzler::CreateSwizzler(srcConfig, NULL, requestedInfo,
+                                               options.fZeroInitialized, this->getInfo().width()));
+    if (!fSwizzler) {
+        // FIXME: CreateSwizzler could fail for another reason.
+        return kUnimplemented;
+    }
+    return kSuccess;
+}
+
 /*
  * Handles rewinding the input stream if it is necessary
  */
 bool SkJpegCodec::handleRewind() {
     switch(this->rewindIfNeeded()) {
         case kCouldNotRewind_RewindState:
             return fDecoderMgr->returnFalse("could not rewind");
         case kRewound_RewindState: {
             JpegDecoderMgr* decoderMgr = NULL;
             if (!ReadHeader(this->stream(), NULL, &decoderMgr)) {
@@ skipping 63 matching lines @@
                 // much faster than decoding to color and then converting
                 fDecoderMgr->dinfo()->out_color_space = JCS_GRAYSCALE;
             }
             return true;
         default:
             return false;
     }
 }
 
 /*
- * Checks if we can scale to the requested dimensions and scales the dimensions
- * if possible
+ * Checks if we can natively scale to the requested dimensions and natively scales the 
+ * dimensions if possible
  */
-bool SkJpegCodec::scaleToDimensions(uint32_t dstWidth, uint32_t dstHeight) {
+bool SkJpegCodec::nativelyScaleToDimensions(uint32_t dstWidth, uint32_t dstHeight) {
     // libjpeg-turbo can scale to 1/8, 1/4, 3/8, 1/2, 5/8, 3/4, 7/8, and 1/1
     fDecoderMgr->dinfo()->scale_denom = 8;
     fDecoderMgr->dinfo()->scale_num = 8;
     turbo_jpeg_calc_output_dimensions(fDecoderMgr->dinfo());
     while (fDecoderMgr->dinfo()->output_width != dstWidth ||
             fDecoderMgr->dinfo()->output_height != dstHeight) {
 
         // Return a failure if we have tried all of the possible scales
         if (1 == fDecoderMgr->dinfo()->scale_num ||
                 dstWidth > fDecoderMgr->dinfo()->output_width ||
                 dstHeight > fDecoderMgr->dinfo()->output_height) {
-            return fDecoderMgr->returnFalse("could not scale to requested dimensions");
+            // reset native scale settings on failure because this may be supported by the swizzler 
+            this->fDecoderMgr->dinfo()->scale_num = 8;
+            turbo_jpeg_calc_output_dimensions(this->fDecoderMgr->dinfo());
+            return false;
         }
 
         // Try the next scale
         fDecoderMgr->dinfo()->scale_num -= 1;
         turbo_jpeg_calc_output_dimensions(fDecoderMgr->dinfo());
     }
     return true;
 }
 
 /*
@@ skipping 19 matching lines @@
     if (setjmp(fDecoderMgr->getJmpBuf())) {
         return fDecoderMgr->returnFailure("setjmp", kInvalidInput);
     }
 
     // Check if we can decode to the requested destination and set the output color space
     if (!this->setOutputColorSpace(dstInfo)) {
         return fDecoderMgr->returnFailure("conversion_possible", kInvalidConversion);
     }
 
     // Perform the necessary scaling
-    if (!this->scaleToDimensions(dstInfo.width(), dstInfo.height())) {
+    if (!this->nativelyScaleToDimensions(dstInfo.width(), dstInfo.height())) {
         return fDecoderMgr->returnFailure("cannot scale to requested dims", kInvalidScale);
     }
 
     // Now, given valid output dimensions, we can start the decompress
     if (!turbo_jpeg_start_decompress(dinfo)) {
         return fDecoderMgr->returnFailure("startDecompress", kInvalidInput);
     }
 
     // The recommended output buffer height should always be 1 in high quality modes.
     // If it's not, we want to know because it means our strategy is not optimal.
@@ skipping 46 matching lines @@
 /*
  * Enable scanline decoding for jpegs
  */
 class SkJpegScanlineDecoder : public SkScanlineDecoder {
 public:
     SkJpegScanlineDecoder(const SkImageInfo& dstInfo, SkJpegCodec* codec,
             const SkCodec::Options& opts)
         : INHERITED(dstInfo)
         , fCodec(codec)
         , fOpts(opts)
-    {}
+    {
+        if(fCodec->fSwizzler) {
+            fStorage.reset(fCodec->getInfo().width() * dstInfo.minRowBytes());

[scroggo, 2015/08/03 19:16:20]

This seems like more than we need to allocate. (Maybe this is why it was such a
large part of the slowdown?)

In SkPngScanlineDecoder we set this to <src width> * <src bytes per pixel>.

I assume it should be the same here? Instead, we are multiplying <src width> *
<dst width> * <dst bytes per pixel>

[emmaleer, 2015/08/04 18:56:53]

Acknowledged.

+            fSrcRow = static_cast<uint8_t*>(fStorage.get());
+        }

[scroggo, 2015/08/03 19:16:20]

else {
  fSrcRow = NULL;
}

[emmaleer, 2015/08/04 18:56:53]

Acknowledged.

+    }
 
     virtual ~SkJpegScanlineDecoder() {
         if (setjmp(fCodec->fDecoderMgr->getJmpBuf())) {
             SkCodecPrintf("setjmp: Error in libjpeg finish_decompress\n");
             return;
         }
 
         // We may not have decoded the entire image.  Prevent libjpeg-turbo from failing on a
         // partial decode.
         fCodec->fDecoderMgr->dinfo()->output_scanline = fCodec->getInfo().height();
         turbo_jpeg_finish_decompress(fCodec->fDecoderMgr->dinfo());
     }
 
     SkCodec::Result onGetScanlines(void* dst, int count, size_t rowBytes) override {
         // Set the jump location for libjpeg errors
         if (setjmp(fCodec->fDecoderMgr->getJmpBuf())) {
             return fCodec->fDecoderMgr->returnFailure("setjmp", SkCodec::kInvalidInput);
         }
+        // Read rows one at a time
+        JSAMPLE* dstRow;
+        if (fCodec->fSwizzler) {
+            // write data to storage row, then sample using swizzler         
+            dstRow = fSrcRow;
+        } else {
+            // write data directly to dst
+            dstRow = (JSAMPLE*) dst;
+        }
 
-        // Read rows one at a time
-        JSAMPLE* dstRow = (JSAMPLE*) dst;
         for (int y = 0; y < count; y++) {
             // Read row of the image
             uint32_t rowsDecoded =
                     turbo_jpeg_read_scanlines(fCodec->fDecoderMgr->dinfo(), &dstRow, 1);
             if (rowsDecoded != 1) {
                 if (SkCodec::kNo_ZeroInitialized == fOpts.fZeroInitialized ||
                         kN32_SkColorType == this->dstInfo().colorType()) {
                     SkSwizzler::Fill(dstRow, this->dstInfo(), rowBytes,
                             count - y, SK_ColorBLACK, NULL);
                 }
                 fCodec->fDecoderMgr->dinfo()->output_scanline = this->dstInfo().height();
                 return SkCodec::kIncompleteInput;
             }
 
             // Convert to RGBA if necessary
             if (JCS_CMYK == fCodec->fDecoderMgr->dinfo()->out_color_space) {
                 convert_CMYK_to_RGBA(dstRow, this->dstInfo().width());
             }
 
-            // Move to the next row
-            dstRow = SkTAddOffset<JSAMPLE>(dstRow, rowBytes);
+            if(fCodec->fSwizzler) {
+                // use swizzler to sample row
+                fCodec->fSwizzler->swizzle(dst, dstRow);
+                dst = SkTAddOffset<JSAMPLE>(dst, rowBytes);
+            } else {
+                dstRow = SkTAddOffset<JSAMPLE>(dstRow, rowBytes);
+            }
         }
-
         return SkCodec::kSuccess;
     }
 
 #ifndef TURBO_HAS_SKIP
 #define turbo_jpeg_skip_scanlines(dinfo, count)                              \
     SkAutoMalloc storage(dinfo->output_width * dinfo->out_color_components); \
     uint8_t* storagePtr = static_cast<uint8_t*>(storage.get());              \
     for (int y = 0; y < count; y++) {                                        \
         turbo_jpeg_read_scanlines(dinfo, &storagePtr, 1);                    \
     }
 #endif
 
     SkCodec::Result onSkipScanlines(int count) override {
         // Set the jump location for libjpeg errors
         if (setjmp(fCodec->fDecoderMgr->getJmpBuf())) {
             return fCodec->fDecoderMgr->returnFailure("setjmp", SkCodec::kInvalidInput);
         }
 
         turbo_jpeg_skip_scanlines(fCodec->fDecoderMgr->dinfo(), count);
 
         return SkCodec::kSuccess;
     }
 
 private:
     SkAutoTDelete<SkJpegCodec> fCodec;
+    SkAutoMalloc               fStorage;    // Only used if sampling is needed
+    uint8_t*                   fSrcRow;     // Only used if sampling is needed
     const SkCodec::Options&    fOpts;
 
     typedef SkScanlineDecoder INHERITED;
 };
 
 SkScanlineDecoder* SkJpegCodec::onGetScanlineDecoder(const SkImageInfo& dstInfo,
         const Options& options, SkPMColor ctable[], int* ctableCount) {
 
     // Rewind the stream if needed
     if (!this->handleRewind()) {
@@ skipping 16 matching lines @@
         return NULL;
     }
 
     // Check if we can decode to the requested destination and set the output color space
     if (!codec->setOutputColorSpace(dstInfo)) {
         SkCodecPrintf("Cannot convert to output type\n");
         return NULL;
     }
 
     // Perform the necessary scaling
-    if (!codec->scaleToDimensions(dstInfo.width(), dstInfo.height())) {
-        SkCodecPrintf("Cannot scale to output dimensions\n");
-        return NULL;
+    if (!codec->nativelyScaleToDimensions(dstInfo.width(), dstInfo.height())) {
+        // native scaling to dstInfo dimensions not supported
+
+        if (!SkScaledCodec::DimensionsSupportedForSampling(this->getInfo(), dstInfo)) {
+            return NULL;
+        }
+        // create swizzler for sampling
+        if (codec->initializeSwizzler(dstInfo, NULL, dstInfo.minRowBytes(), options, ctable,
+                                      ctableCount) != kSuccess) {
+            SkCodecPrintf("failed to initialize the swizzler.\n");
+            return NULL;
+        }
     }
 
     // Now, given valid output dimensions, we can start the decompress
     if (!turbo_jpeg_start_decompress(codec->fDecoderMgr->dinfo())) {
         SkCodecPrintf("start decompress failed\n");
         return NULL;
     }
 
     // Return the new scanline decoder
     return SkNEW_ARGS(SkJpegScanlineDecoder, (dstInfo, codec.detach(), options));
 }