Transparent hugepage support

[x264] / common / common.h

diff --git a/common/common.h b/common/common.h
index 46344b0c842835887b7263fd49db4ff81e6a0857..12e5763bb1bd6593e1672e65af97ec46074dfe59 100644 (file)
--- a/common/common.h
+++ b/common/common.h
@@ -54,6 +54,31 @@ do {\
     memset( var, 0, size );\
 } while( 0 )
 
+/* Macros for merging multiple allocations into a single large malloc, for improved
+ * use with huge pages. */
+
+/* Needs to be enough to contain any set of buffers that use combined allocations */
+#define PREALLOC_BUF_SIZE 1024
+
+#define PREALLOC_INIT\
+    int    prealloc_idx = 0;\
+    size_t prealloc_size = 0;\
+    uint8_t **preallocs[PREALLOC_BUF_SIZE];
+
+#define PREALLOC( var, size )\
+do {\
+    var = (void*)prealloc_size;\
+    preallocs[prealloc_idx++] = (uint8_t**)&var;\
+    prealloc_size += ALIGN(size, NATIVE_ALIGN);\
+} while(0)
+
+#define PREALLOC_END( ptr )\
+do {\
+    CHECKED_MALLOC( ptr, prealloc_size );\
+    while( prealloc_idx-- )\
+        *preallocs[prealloc_idx] += (intptr_t)ptr;\
+} while(0)
+
 #define ARRAY_SIZE(array)  (sizeof(array)/sizeof(array[0]))
 
 #define X264_BFRAME_MAX 16
@@ -96,10 +121,6 @@ do {\
 #include <assert.h>
 #include <limits.h>
 
-#if HAVE_OPENCL
-#include "opencl.h"
-#endif
-
 #if HAVE_INTERLACED
 #   define MB_INTERLACED h->mb.b_interlaced
 #   define SLICE_MBAFF h->sh.b_mbaff
@@ -209,6 +230,10 @@ static const uint8_t x264_scan8[16*3 + 3] =
 };
 
 #include "x264.h"
+#if HAVE_OPENCL
+#include "opencl.h"
+#endif
+#include "cabac.h"
 #include "bitstream.h"
 #include "set.h"
 #include "predict.h"
@@ -216,7 +241,6 @@ static const uint8_t x264_scan8[16*3 + 3] =
 #include "mc.h"
 #include "frame.h"
 #include "dct.h"
-#include "cabac.h"
 #include "quant.h"
 #include "cpu.h"
 #include "threadpool.h"
@@ -610,11 +634,11 @@ struct x264_t
     /* Current MB DCT coeffs */
     struct
     {
-        ALIGNED_16( dctcoef luma16x16_dc[3][16] );
+        ALIGNED_N( dctcoef luma16x16_dc[3][16] );
         ALIGNED_16( dctcoef chroma_dc[2][8] );
         // FIXME share memory?
-        ALIGNED_16( dctcoef luma8x8[12][64] );
-        ALIGNED_16( dctcoef luma4x4[16*3][16] );
+        ALIGNED_N( dctcoef luma8x8[12][64] );
+        ALIGNED_N( dctcoef luma4x4[16*3][16] );
     } dct;
 
     /* MB table and cache for current frame/mb */
@@ -700,6 +724,7 @@ struct x264_t
          * and won't be copied from one thread to another */
 
         /* mb table */
+        uint8_t *base;                      /* base pointer for all malloced data in this mb */
         int8_t  *type;                      /* mb type */
         uint8_t *partition;                 /* mb partition */
         int8_t  *qp;                        /* mb qp */
@@ -753,7 +778,7 @@ struct x264_t
 #define FENC_STRIDE 16
 #define FDEC_STRIDE 32
             ALIGNED_16( pixel fenc_buf[48*FENC_STRIDE] );
-            ALIGNED_16( pixel fdec_buf[52*FDEC_STRIDE] );
+            ALIGNED_N( pixel fdec_buf[52*FDEC_STRIDE] );
 
             /* i4x4 and i8x8 backup data, for skipping the encode stage when possible */
             ALIGNED_16( pixel i4x4_fdec_buf[16*16] );
@@ -770,8 +795,8 @@ struct x264_t
             ALIGNED_16( dctcoef fenc_dct4[16][16] );
 
             /* Psy RD SATD/SA8D scores cache */
-            ALIGNED_16( uint64_t fenc_hadamard_cache[9] );
-            ALIGNED_16( uint32_t fenc_satd_cache[32] );
+            ALIGNED_N( uint64_t fenc_hadamard_cache[9] );
+            ALIGNED_N( uint32_t fenc_satd_cache[32] );
 
             /* pointer over mb of the frame to be compressed */
             pixel *p_fenc[3]; /* y,u,v */
@@ -905,8 +930,8 @@ struct x264_t
     uint32_t (*nr_residual_sum)[64];
     uint32_t *nr_count;
 
-    ALIGNED_16( udctcoef nr_offset_denoise[4][64] );
-    ALIGNED_16( uint32_t nr_residual_sum_buf[2][4][64] );
+    ALIGNED_N( udctcoef nr_offset_denoise[4][64] );
+    ALIGNED_N( uint32_t nr_residual_sum_buf[2][4][64] );
     uint32_t nr_count_buf[2][4];
 
     uint8_t luma2chroma_pixel[7]; /* Subsampled pixel size */