Aligned malloc using Intel's _mm_malloc(), cleanup on the t1 memory allocation, getting rid of some leftover debug code

ChangeLog

@@ -10,6 +10,7 @@ August 21, 2007
 * [FOD] Robustified MJ2 codecs
 * [Parvatha] Solved problems with codec reading from image file directory when filename had more than one "." in name
 * [Callum Lerwick] Minor cleanup patch, that gets rid of a bunch of "old style declaration" warnings from Intel's compiler
+* [Callum Lerwick] Aligned malloc using Intel's _mm_malloc(). Cleanup on the t1 memory allocation, getting rid of some leftover debug code
 
 August 20, 2007
 + [FOD] Added support for the TGA file format in the codec

libopenjpeg/j2k_lib.h

@@ -32,6 +32,10 @@
 The functions in J2K_LIB.C are internal utilities mainly used for memory management.
 */
 
+#ifndef __GCC__
+#define __attribute__(x) /* */
+#endif
+
 /** @defgroup MISC MISC - Miscellaneous internal functions */
 /*@{*/
 
@@ -50,7 +54,16 @@ Allocate a memory block with elements initialized to 0
 @param size Bytes to allocate
 @return Returns a void pointer to the allocated space, or NULL if there is insufficient memory available
 */
-void* opj_malloc( size_t size );
+void* __attribute__ ((malloc)) opj_malloc( size_t size );
+
+/**
+Allocate memory aligned to a 16 byte boundry
+@param size Bytes to allocate
+@return Returns a void pointer to the allocated space, or NULL if there is insufficient memory available
+*/
+#include <xmmintrin.h>
+#define opj_aligned_malloc(size)	_mm_malloc(size, 16)
+#define opj_aligned_free(m) _mm_free(m)
 
 /**
 Reallocate memory blocks.
@@ -58,7 +71,7 @@ Reallocate memory blocks.
 @param size New size in bytes
 @return Returns a void pointer to the reallocated (and possibly moved) memory block
 */
-void* opj_realloc( void *memblock, size_t size );
+void* __attribute__ ((malloc)) opj_realloc( void *memblock, size_t size );
 
 /**
 Deallocates or frees a memory block.

libopenjpeg/t1.c

@@ -758,47 +758,41 @@ static double t1_getwmsedec(
 	return wmsedec;
 }
 
-static void allocate_buffers(
+static bool allocate_buffers(
 		opj_t1_t *t1,
 		int w,
 		int h)
 {
-	int datasize;
+	int datasize=w * h;
 	int flagssize;
 
-	datasize=w * h;
-	//fprintf(stderr,"w=%i h=%i datasize=%i flagssize=%i\n",w,h,datasize,flagssize);
-
 	if(datasize > t1->datasize){
-		//fprintf(stderr,"Allocating t1->data: datasize=%i\n",datasize);
+		opj_aligned_free(t1->data);
-		free(t1->data);
+		t1->data=opj_aligned_malloc(datasize * sizeof(int));
-		t1->data=malloc(datasize * sizeof(int));
 		if(!t1->data){
-			return;
+			return false;
 		}
 		t1->datasize=datasize;
 	}
-	//memset(t1->data,0xff,t1->datasize);
 	memset(t1->data,0,datasize * sizeof(int));
 
 	t1->flags_stride=w+2;
-	flagssize=t1->flags_stride * (h+2);
+	flagssize = t1->flags_stride * (h+2);
 
 	if(flagssize > t1->flagssize){
-		//fprintf(stderr,"Allocating t1->flags: flagssize=%i\n",flagssize);
+		opj_aligned_free(t1->flags);
-		free(t1->flags);
+		t1->flags=opj_aligned_malloc(flagssize * sizeof(flag_t));
-		t1->flags=malloc(flagssize * sizeof(flag_t));
 		if(!t1->flags){
-			fprintf(stderr,"Allocating t1->flags FAILED!\n");
+			return false;
-			return;
 		}
 		t1->flagssize=flagssize;
 	}
-	//memset(t1->flags,0xff,t1->flagssize);
 	memset(t1->flags,0,flagssize * sizeof(flag_t));
 
 	t1->w=w;
 	t1->h=h;
+
+	return true;
 }
 
 /** mod fixed_quality */
@@ -948,12 +942,15 @@ static void t1_decode_cblk(
 	
 	opj_raw_t *raw = t1->raw;	/* RAW component */
 	opj_mqc_t *mqc = t1->mqc;	/* MQC component */
-	
+
-	allocate_buffers(
+	if(!allocate_buffers(
-			t1,
+				t1,
-			cblk->x1 - cblk->x0,
+				cblk->x1 - cblk->x0,
-			cblk->y1 - cblk->y0);
+				cblk->y1 - cblk->y0))
-	
+	{
+		return;
+	}
+
 	bpno = roishift + cblk->numbps - 1;
 	passtype = 2;
 	
@@ -1025,8 +1022,8 @@ void t1_destroy(opj_t1_t *t1) {
 		/* destroy MQC and RAW handles */
 		mqc_destroy(t1->mqc);
 		raw_destroy(t1->raw);
-		free(t1->data);
+		opj_aligned_free(t1->data);
-		free(t1->flags);
+		opj_aligned_free(t1->flags);
 		free(t1);
 	}
 }
@@ -1067,10 +1064,13 @@ void t1_encode_cblks(
 							y += pres->y1 - pres->y0;
 						}
 
-						allocate_buffers(
+						if(!allocate_buffers(
-								t1,
+									t1,
-								cblk->x1 - cblk->x0,
+									cblk->x1 - cblk->x0,
-								cblk->y1 - cblk->y0);
+									cblk->y1 - cblk->y0))
+						{
+							return;
+						}
 
 						w = tilec->x1 - tilec->x0;
 						if (tcp->tccps[compno].qmfbid == 1) {

libopenjpeg/tcd.c

@@ -192,7 +192,7 @@ void tcd_malloc_encode(opj_tcd_t *tcd, opj_image_t * image, opj_cp_t * cp, int c
 			tilec->x1 = int_ceildiv(tile->x1, image->comps[compno].dx);
 			tilec->y1 = int_ceildiv(tile->y1, image->comps[compno].dy);
 			
-			tilec->data = (int *) opj_malloc((tilec->x1 - tilec->x0) * (tilec->y1 - tilec->y0) * sizeof(int));
+			tilec->data = (int *) opj_aligned_malloc((tilec->x1 - tilec->x0) * (tilec->y1 - tilec->y0) * sizeof(int));
 			tilec->numresolutions = tccp->numresolutions;
 
 			tilec->resolutions = (opj_tcd_resolution_t *) opj_malloc(tilec->numresolutions * sizeof(opj_tcd_resolution_t));
@@ -437,7 +437,7 @@ void tcd_init_encode(opj_tcd_t *tcd, opj_image_t * image, opj_cp_t * cp, int cur
 			tilec->x1 = int_ceildiv(tile->x1, image->comps[compno].dx);
 			tilec->y1 = int_ceildiv(tile->y1, image->comps[compno].dy);
 			
-			tilec->data = (int *) opj_malloc((tilec->x1 - tilec->x0) * (tilec->y1 - tilec->y0) * sizeof(int));
+			tilec->data = (int *) opj_aligned_malloc((tilec->x1 - tilec->x0) * (tilec->y1 - tilec->y0) * sizeof(int));
 			tilec->numresolutions = tccp->numresolutions;
 			/* tilec->resolutions=(opj_tcd_resolution_t*)opj_realloc(tilec->resolutions,tilec->numresolutions*sizeof(opj_tcd_resolution_t)); */
 			for (resno = 0; resno < tilec->numresolutions; resno++) {
@@ -676,7 +676,7 @@ void tcd_malloc_decode_tile(opj_tcd_t *tcd, opj_image_t * image, opj_cp_t * cp,
 		tilec->x1 = int_ceildiv(tile->x1, image->comps[compno].dx);
 		tilec->y1 = int_ceildiv(tile->y1, image->comps[compno].dy);
 		
-		tilec->data = (int *) opj_malloc((tilec->x1 - tilec->x0) * (tilec->y1 - tilec->y0) * sizeof(int));
+		tilec->data = (int*) opj_aligned_malloc((tilec->x1 - tilec->x0) * (tilec->y1 - tilec->y0) * sizeof(int));
 		tilec->numresolutions = tccp->numresolutions;
 		tilec->resolutions = (opj_tcd_resolution_t *) opj_malloc(tilec->numresolutions * sizeof(opj_tcd_resolution_t));
 		
@@ -1279,14 +1279,14 @@ int tcd_encode_tile(opj_tcd_t *tcd, int tileno, unsigned char *dest, int len, op
 	if(tcd->cur_tp_num == tcd->cur_totnum_tp - 1){
 		tcd->encoding_time = opj_clock() - tcd->encoding_time;
 		opj_event_msg(tcd->cinfo, EVT_INFO, "- tile encoded in %f s\n", tcd->encoding_time);
-	
+
 		/* cleaning memory */
 		for (compno = 0; compno < tile->numcomps; compno++) {
 			opj_tcd_tilecomp_t *tilec = &tile->comps[compno];
-			opj_free(tilec->data);
+			opj_aligned_free(tilec->data);
 		}
 	}
-	
+
 	return l;
 }
 
@@ -1401,12 +1401,12 @@ bool tcd_decode_tile(opj_tcd_t *tcd, unsigned char *src, int len, int tileno) {
 	
 	tile_time = opj_clock() - tile_time;	/* time needed to decode a tile */
 	opj_event_msg(tcd->cinfo, EVT_INFO, "- tile decoded in %f s\n", tile_time);
-		
+
 	for (compno = 0; compno < tile->numcomps; compno++) {
-		opj_free(tcd->tcd_image->tiles[tileno].comps[compno].data);
+		opj_aligned_free(tcd->tcd_image->tiles[tileno].comps[compno].data);
 		tcd->tcd_image->tiles[tileno].comps[compno].data = NULL;
 	}
-	
+
 	if (eof) {
 		return false;
 	}