moved ALIGN_CST into util/utils.h and renamed WEBP_ALIGN_xxx

Note that ALIGN_CST is still kept different in dec/frame.c for now,
because the values is 31 there, not 15. We might re-unite these two
later.

Change-Id: Ibbee607fac4eef02f175b56f0bb0ba359fda3b87

src/enc/analysis.c

@@ -405,8 +405,8 @@ typedef struct {
 static int DoSegmentsJob(SegmentJob* const job, VP8EncIterator* const it) {
   int ok = 1;
   if (!VP8IteratorIsDone(it)) {
-    uint8_t tmp[32 + ALIGN_CST];
-    uint8_t* const scratch = (uint8_t*)DO_ALIGN(tmp);
+    uint8_t tmp[32 + WEBP_ALIGN_CST];
+    uint8_t* const scratch = (uint8_t*)WEBP_ALIGN(tmp);
     do {
       // Let's pretend we have perfect lossless reconstruction.
       VP8IteratorImport(it, scratch);

src/enc/histogram.c

@@ -20,9 +20,6 @@
 #include "../dsp/lossless.h"
 #include "../utils/utils.h"
 
-#define ALIGN_CST 15
-#define DO_ALIGN(PTR) ((uintptr_t)((PTR) + ALIGN_CST) & ~ALIGN_CST)
-
 #define MAX_COST 1.e38
 
 // Number of partitions for the three dominant (literal, red and blue) symbol
@@ -115,7 +112,8 @@ VP8LHistogramSet* VP8LAllocateHistogramSet(int size, int cache_bits) {
   VP8LHistogramSet* set;
   const int histo_size = VP8LGetHistogramSize(cache_bits);
   const size_t total_size =
-      sizeof(*set) + size * (sizeof(*set->histograms) + histo_size + ALIGN_CST);
+      sizeof(*set) + size * (sizeof(*set->histograms) +
+      histo_size + WEBP_ALIGN_CST);
   uint8_t* memory = (uint8_t*)WebPSafeMalloc(total_size, sizeof(*memory));
   if (memory == NULL) return NULL;
 
@@ -126,7 +124,7 @@ VP8LHistogramSet* VP8LAllocateHistogramSet(int size, int cache_bits) {
   set->max_size = size;
   set->size = size;
   for (i = 0; i < size; ++i) {
-    memory = (uint8_t*)DO_ALIGN(memory);
+    memory = (uint8_t*)WEBP_ALIGN(memory);
     set->histograms[i] = (VP8LHistogram*)memory;
     // literal_ won't necessary be aligned.
     set->histograms[i]->literal_ = (uint32_t*)(memory + sizeof(VP8LHistogram));

src/enc/iterator.c

@@ -70,13 +70,13 @@ void VP8IteratorInit(VP8Encoder* const enc, VP8EncIterator* const it) {
   it->enc_ = enc;
   it->y_stride_  = enc->pic_->y_stride;
   it->uv_stride_ = enc->pic_->uv_stride;
-  it->yuv_in_   = (uint8_t*)DO_ALIGN(it->yuv_mem_);
+  it->yuv_in_   = (uint8_t*)WEBP_ALIGN(it->yuv_mem_);
   it->yuv_out_  = it->yuv_in_ + YUV_SIZE_ENC;
   it->yuv_out2_ = it->yuv_out_ + YUV_SIZE_ENC;
   it->yuv_p_    = it->yuv_out2_ + YUV_SIZE_ENC;
   it->lf_stats_ = enc->lf_stats_;
   it->percent0_ = enc->percent_;
-  it->y_left_ = (uint8_t*)DO_ALIGN(it->yuv_left_mem_ + 1);
+  it->y_left_ = (uint8_t*)WEBP_ALIGN(it->yuv_left_mem_ + 1);
   it->u_left_ = it->y_left_ + 16 + 16;
   it->v_left_ = it->u_left_ + 16;
   VP8IteratorReset(it);

src/enc/vp8enci.h

@@ -19,6 +19,7 @@
 #include "../dsp/dsp.h"
 #include "../utils/bit_writer.h"
 #include "../utils/thread.h"
+#include "../utils/utils.h"
 #include "../webp/encode.h"
 
 #ifdef WEBP_EXPERIMENTAL_FEATURES
@@ -78,9 +79,6 @@ typedef enum {   // Rate-distortion optimization levels
 #define U_OFF_ENC    (16)
 #define V_OFF_ENC    (16 + 8)
 
-#define ALIGN_CST 15
-#define DO_ALIGN(PTR) ((uintptr_t)((PTR) + ALIGN_CST) & ~ALIGN_CST)
-
 extern const int VP8Scan[16];           // in quant.c
 extern const int VP8UVModeOffsets[4];   // in analyze.c
 extern const int VP8I16ModeOffsets[4];
@@ -254,9 +252,9 @@ typedef struct {
   uint8_t* uv_top_;    // top u/v samples at position 'x_', packed as 16 bytes
 
   // memory for storing y/u/v_left_
-  uint8_t yuv_left_mem_[17 + 16 + 16 + 8 + ALIGN_CST];
+  uint8_t yuv_left_mem_[17 + 16 + 16 + 8 + WEBP_ALIGN_CST];
   // memory for yuv_*
-  uint8_t yuv_mem_[3 * YUV_SIZE_ENC + PRED_SIZE_ENC + ALIGN_CST];
+  uint8_t yuv_mem_[3 * YUV_SIZE_ENC + PRED_SIZE_ENC + WEBP_ALIGN_CST];
 } VP8EncIterator;
 
   // in iterator.c

src/enc/webpenc.c

@@ -145,16 +145,16 @@ static VP8Encoder* InitVP8Encoder(const WebPConfig* const config,
   const int preds_h = 4 * mb_h + 1;
   const size_t preds_size = preds_w * preds_h * sizeof(uint8_t);
   const int top_stride = mb_w * 16;
-  const size_t nz_size = (mb_w + 1) * sizeof(uint32_t) + ALIGN_CST;
+  const size_t nz_size = (mb_w + 1) * sizeof(uint32_t) + WEBP_ALIGN_CST;
   const size_t info_size = mb_w * mb_h * sizeof(VP8MBInfo);
   const size_t samples_size = 2 * top_stride * sizeof(uint8_t)  // top-luma/u/v
-                            + ALIGN_CST;                        // align all
+                            + WEBP_ALIGN_CST;                   // align all
   const size_t lf_stats_size =
-      config->autofilter ? sizeof(LFStats) + ALIGN_CST : 0;
+      config->autofilter ? sizeof(LFStats) + WEBP_ALIGN_CST : 0;
   VP8Encoder* enc;
   uint8_t* mem;
   const uint64_t size = (uint64_t)sizeof(VP8Encoder)   // main struct
-                      + ALIGN_CST                      // cache alignment
+                      + WEBP_ALIGN_CST                 // cache alignment
                       + info_size                      // modes info
                       + preds_size                     // prediction modes
                       + samples_size                   // top/left samples
@@ -171,7 +171,7 @@ static VP8Encoder* InitVP8Encoder(const WebPConfig* const config,
          "            non-zero: %ld\n"
          "            lf-stats: %ld\n"
          "               total: %ld\n",
-         sizeof(VP8Encoder) + ALIGN_CST, info_size,
+         sizeof(VP8Encoder) + WEBP_ALIGN_CST, info_size,
          preds_size, samples_size, nz_size, lf_stats_size, size);
   printf("Transient object sizes:\n"
          "      VP8EncIterator: %ld\n"
@@ -192,7 +192,7 @@ static VP8Encoder* InitVP8Encoder(const WebPConfig* const config,
     return NULL;
   }
   enc = (VP8Encoder*)mem;
-  mem = (uint8_t*)DO_ALIGN(mem + sizeof(*enc));
+  mem = (uint8_t*)WEBP_ALIGN(mem + sizeof(*enc));
   memset(enc, 0, sizeof(*enc));
   enc->num_parts_ = 1 << config->partitions;
   enc->mb_w_ = mb_w;
@@ -202,13 +202,13 @@ static VP8Encoder* InitVP8Encoder(const WebPConfig* const config,
   mem += info_size;
   enc->preds_ = ((uint8_t*)mem) + 1 + enc->preds_w_;
   mem += preds_w * preds_h * sizeof(uint8_t);
-  enc->nz_ = 1 + (uint32_t*)DO_ALIGN(mem);
+  enc->nz_ = 1 + (uint32_t*)WEBP_ALIGN(mem);
   mem += nz_size;
-  enc->lf_stats_ = lf_stats_size ? (LFStats*)DO_ALIGN(mem) : NULL;
+  enc->lf_stats_ = lf_stats_size ? (LFStats*)WEBP_ALIGN(mem) : NULL;
   mem += lf_stats_size;
 
   // top samples (all 16-aligned)
-  mem = (uint8_t*)DO_ALIGN(mem);
+  mem = (uint8_t*)WEBP_ALIGN(mem);
   enc->y_top_ = (uint8_t*)mem;
   enc->uv_top_ = enc->y_top_ + top_stride;
   mem += 2 * top_stride;

src/utils/utils.h

@@ -43,6 +43,12 @@ WEBP_EXTERN(void*) WebPSafeCalloc(uint64_t nmemb, size_t size);
 // Companion deallocation function to the above allocations.
 WEBP_EXTERN(void) WebPSafeFree(void* const ptr);
 
+//------------------------------------------------------------------------------
+// Alignment
+
+#define WEBP_ALIGN_CST 15
+#define WEBP_ALIGN(PTR) ((uintptr_t)((PTR) + WEBP_ALIGN_CST) & ~WEBP_ALIGN_CST)
+
 //------------------------------------------------------------------------------
 // Reading/writing data.