--- xvid_bench.c	2006/10/30 11:21:42	1.33
+++ xvid_bench.c	2008/11/26 23:37:28	1.38
@@ -19,7 +19,7 @@
  *  along with this program; if not, write to the Free Software
  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  *
- * $Id: xvid_bench.c,v 1.33 2006/10/30 11:21:42 Skal Exp $
+ * $Id: xvid_bench.c,v 1.38 2008/11/26 23:37:28 Isibaar Exp $
  *
  ****************************************************************************/
 
@@ -58,6 +58,7 @@
 #include "utils/timer.h"
 #include "quant/quant_matrix.c"
 #include "bitstream/cbp.h"
+#include "bitstream/bitstream.h"
 
 #include <math.h>
 
@@ -114,19 +115,18 @@
 
 CPU cpu_list[] = {
 	{ "PLAINC ", 0 },
-#ifdef ARCH_IS_IA32
+#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)
 	{ "MMX    ", XVID_CPU_MMX },
 	{ "MMXEXT ", XVID_CPU_MMXEXT | XVID_CPU_MMX },
 	{ "SSE2   ", XVID_CPU_SSE2 | XVID_CPU_MMX },
+	{ "SSE3   ", XVID_CPU_SSE3 | XVID_CPU_SSE2 | XVID_CPU_MMX },
+	{ "SSE41  ", XVID_CPU_SSE41| XVID_CPU_SSE3 | XVID_CPU_SSE2 | XVID_CPU_MMX },
 	{ "3DNOW  ", XVID_CPU_3DNOW },
 	{ "3DNOWE ", XVID_CPU_3DNOW | XVID_CPU_3DNOWEXT },
 #endif
 #ifdef ARCH_IS_PPC
 	{ "ALTIVEC", XVID_CPU_ALTIVEC },
 #endif
-#ifdef ARCH_IS_X86_64
-	{ "X86_64 ", XVID_CPU_ASM},
-#endif
 #ifdef ARCH_IS_IA64
 //	{ "IA64   ", XVID_CPU_IA64 },
 #endif
@@ -411,6 +411,15 @@
 				cpu->name, t, s,
 				(s!=26274)?"| ERROR": "" );
 
+                t = gettime_usec();
+                emms();
+                for(tst=0; tst<nb_tests; ++tst) s = sad8bi(Cur, Ref1, Ref2, 8);
+                emms();
+                t = (gettime_usec() - t) / nb_tests;
+                printf( "%s - sad8bi %.3f usec       sad=%d %s\n",
+                                cpu->name, t, s,
+                                (s!=4002)?"| ERROR": "" );
+
 		t = gettime_usec();
 		emms();
 		for(tst=0; tst<nb_tests; ++tst) s = dev16(Cur, 16);
@@ -730,6 +739,7 @@
   set_intra_matrix( mpeg_quant_matrices, Quant );                \
   emms();                                   \
   for(q=1; q<=max_Q; ++q) {                 \
+	init_intra_matrix( mpeg_quant_matrices, q ); \
 	for(tst=0; tst<nb_tests; ++tst)         \
 	  (FUNC)((DST), (SRC), q, q, mpeg_quant_matrices);           \
 	byte_swap((uint8_t*)(DST), 64*sizeof((DST)[0]), sizeof((DST)[0]));  \
@@ -740,7 +750,7 @@
 t = (gettime_usec()-t-overhead)/nb_tests/qm
 
 #define TEST_INTRA(REFFUNC, NEWFUNC, RANGE)              \
-{ int i,q,s;\
+{ int32_t i,q,s;\
 	DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16); \
   DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16); \
   DECLARE_ALIGNED_MATRIX(Dst2,8, 8, int16_t, 16); \
@@ -772,11 +782,11 @@
 
 void test_quant()
 {
-	const int nb_tests = 1*speed_ref;
-	const int max_Q = 31;
+	const int32_t nb_tests = 1*speed_ref;
+	const int32_t max_Q = 31;
 	DECLARE_ALIGNED_MATRIX(mpeg_quant_matrices, 8, 64, uint16_t, 16);
 
-	int i, qm;
+	int32_t i, qm;
 	CPU *cpu;
 	DECLARE_ALIGNED_MATRIX(Src, 8, 8, int16_t, 16);
 	DECLARE_ALIGNED_MATRIX(Dst, 8, 8, int16_t, 16);
@@ -795,7 +805,7 @@
 	for(cpu = cpu_list; cpu->name!=0; ++cpu)
 	{
 		double t, overhead;
-		int tst, q;
+		int32_t tst, q;
 		uint32_t s;
 
 		if (!init_cpu(cpu))
@@ -819,7 +829,7 @@
 		TEST_QUANT2(quant_mpeg_intra, Dst, Src);
 		printf("%s -   quant_mpeg_intra %.3f usec       crc32=0x%08x %s\n",
 			   cpu->name, t, s,
-			   (s!=0xfd6a21a4)? "| ERROR": "");
+			   (s!=0x3b999af6)? "| ERROR": "");
 
 		TEST_QUANT(quant_mpeg_inter, Dst, Src);
 		printf("%s -   quant_mpeg_inter %.3f usec       crc32=0x%08x %s\n",
@@ -1720,11 +1730,20 @@
 t = (gettime_usec() - t) / nb_tests;  \
 	iCrc = calc_crc((uint8_t*)Dst0, sizeof(Dst0), CRC32_INITIAL)
 
-#define TEST_YUYV(FUNC, S)                \
+#define TEST_YUYV(FUNC, S, FLIP)                \
 ENTER                               \
-for(tst=0; tst<nb_tests; ++tst) (FUNC)(Dst0[0], S*WIDTH, Src0[0], Src0[1], Src0[2], WIDTH, WIDTH/2, WIDTH, HEIGHT, 0); \
+for(tst=0; tst<nb_tests; ++tst) (FUNC)(Dst0[0], S*WIDTH, Src0[0], Src0[1], Src0[2], WIDTH, WIDTH/2, WIDTH, HEIGHT, (FLIP)); \
 LEAVE
 
+static const int yuv_CRCs[6][2] = {
+	{0x0f4fb96b,0x780b6a68}
+,	{0xa986b289,0x65e49b76}
+,	{0x7f19c152,0xd539b86e}
+,	{0x0f4fb96b,0x780b6a68}
+,	{0xa986b289,0x65e49b76}
+,	{0x36ab8b57,0x1cd92fee}
+};
+
 #define WIDTH 128
 #define HEIGHT 32
 void test_yuv()
@@ -1733,7 +1752,7 @@
 	CPU *cpu;
 	uint8_t Src0[3][WIDTH*HEIGHT];
 	uint8_t Dst0[4][WIDTH*HEIGHT];
-	int i, j;
+	int i, j, with_flip;
 	double t;
 	int tst, iCrc;
 
@@ -1741,38 +1760,54 @@
 	ieee_reseed(1);
 	for(i=0; i<(int)sizeof(Src0); ++i) Src0[0][i] = ieee_rand(0,255);
 	for(i=0; i<(int)sizeof(Dst0); ++i) Dst0[0][i] = 0x5a;
+	
+        printf( "\n ===  test YUV ===\n" );
 
-	printf( "\n ===  test YUV ===\n" );
+        for(with_flip=0; with_flip<=1; ++with_flip) {
+
+		init_cpu(&cpu_list[0]);
+		TEST_YUYV(yv12_to_yuyv_c, 4, with_flip);
+		printf(" yv12_to_yuyv_c %.3f usec       crc32=0x%08x %s\n",
+			   t, iCrc, (iCrc!=yuv_CRCs[0][with_flip])?"| ERROR": "" );
+		TEST_YUYV(yv12_to_uyvy_c, 4, with_flip);
+		printf(" yv12_to_uyvy_c %.3f usec       crc32=0x%08x %s\n",
+		   	t, iCrc, (iCrc!=yuv_CRCs[1][with_flip])?"| ERROR": "" );
+ 
+        	TEST_YUYV(yv12_to_bgra_c, 4, with_flip);
+        	printf(" yv12_to_bgra_c %.3f usec       crc32=0x%08x %s\n",
+               		t, iCrc, (iCrc!=yuv_CRCs[2][with_flip])?"| ERROR": "" );
+
+#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)
+		init_cpu(&cpu_list[1]);
+		TEST_YUYV(yv12_to_yuyv_mmx, 4, with_flip);
+		printf(" yv12_to_yuyv_mmx %.3f usec       crc32=0x%08x %s\n",
+			t, iCrc, (iCrc!=yuv_CRCs[3][with_flip])?"| ERROR": "" );
+
+		TEST_YUYV(yv12_to_uyvy_mmx, 4, with_flip);
+		printf(" yv12_to_uyvy_mmx %.3f usec       crc32=0x%08x %s\n",
+			t, iCrc, (iCrc!=yuv_CRCs[4][with_flip])?"| ERROR": "" );
+
+        	TEST_YUYV(yv12_to_bgra_mmx, 4, with_flip);
+        	printf(" yv12_to_bgra_mmx %.3f usec       crc32=0x%08x %s\n",
+                	t, iCrc, (iCrc!=yuv_CRCs[5][with_flip])?"| ERROR": "" );
 
-	init_cpu(&cpu_list[0]);
-	TEST_YUYV(yv12_to_yuyv_c, 4);
-	printf(" yv12_to_yuyv_c %.3f usec       crc32=0x%08x %s\n",
-		   t, iCrc, (iCrc!=0xeb1a0b0a)?"| ERROR": "" );
-	TEST_YUYV(yv12_to_uyvy_c, 4);
-	printf(" yv12_to_uyvy_c %.3f usec       crc32=0x%08x %s\n",
-		   t, iCrc, (iCrc!=0x6e82f55b)?"| ERROR": "" );
-
-#ifdef ARCH_IS_IA32
-	init_cpu(&cpu_list[1]);
-	TEST_YUYV(yv12_to_yuyv_mmx, 4);
-	printf(" yv12_to_yuyv_mmx %.3f usec       crc32=0x%08x %s\n",
-		t, iCrc, (iCrc!=0xeb1a0b0a)?"| ERROR": "" );
-
-	TEST_YUYV(yv12_to_uyvy_mmx, 4);
-	printf(" yv12_to_uyvy_mmx %.3f usec       crc32=0x%08x %s\n",
-		t, iCrc, (iCrc!=0x6e82f55b)?"| ERROR": "" );
 #endif
 
 #ifdef ARCH_IS_PPC
-	init_cpu(&cpu_list[1]);
-	TEST_YUYV(yv12_to_yuyv_altivec_c, 4);
-	printf(" yv12_to_yuyv_altivec_c %.3f usec       crc32=0x%08x %s\n",
-		t, iCrc, (iCrc!=0xeb1a0b0a)?"| ERROR": "" );
-
-	TEST_YUYV(yv12_to_uyvy_altivec_c, 4);
-	printf(" yv12_to_uyvy_altivec_c %.3f usec       crc32=0x%08x %s\n",
-		t, iCrc, (iCrc!=0x6e82f55b)?"| ERROR": "" );
+		init_cpu(&cpu_list[1]);
+		TEST_YUYV(yv12_to_yuyv_altivec_c, 4, with_flip);
+		printf(" yv12_to_yuyv_altivec_c %.3f usec       crc32=0x%08x %s\n",
+			t, iCrc, (iCrc!=yuv_CRCs[3][with_flip])?"| ERROR": "" );
+
+		TEST_YUYV(yv12_to_uyvy_altivec_c, 4, with_flip);
+		printf(" yv12_to_uyvy_altivec_c %.3f usec       crc32=0x%08x %s\n",
+			t, iCrc, (iCrc!=yuv_CRCs[4][with_flip])?"| ERROR": "" );
+
+         	TEST_YUYV(yv12_to_bgra_altivec_c, 4, with_flip);
+		printf(" yv12_to_bgra_altivec_c %.3f usec       crc32=0x%08x %s\n",
+                        t, iCrc, (iCrc!=yuv_CRCs[5][with_flip])?"| ERROR": "" );
 #endif
+	}
 	printf( " --- \n" );
 }
 
@@ -1821,7 +1856,7 @@
 				t, iCrc, (iCrc!=yv12_CRCs[with_flip][with_uv])?"| ERROR": "" );
 			/* if (!with_uv) PRINT_NxN(Dst0[1], WIDTH/2, HEIGHT/2, WIDTH ); */
 
-#ifdef ARCH_IS_IA32
+#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)
 			init_cpu(&cpu_list[1]);
 			TEST_YV2(yv12_to_yv12_mmx, with_uv, with_flip);
 			printf(" yv12_to_yv12_mmx %.3f usec     \tcrc32=0x%08x %s\n",
@@ -1932,7 +1967,7 @@
 
 void test_gcd()
 {
-	const int nb_tests = 10*speed_ref;
+  const int nb_tests = 10*speed_ref;
   int i;
   uint32_t crc1=0, crc2=0;
   uint32_t n0, n, d0, d;
@@ -2009,6 +2044,7 @@
     printf( "ERROR! please post your platform/compiler specs to xvid-devel@xvid.org !\n" );
   }
 }
+
 /*********************************************************************
  * test SSIM functions
  *********************************************************************/
@@ -2058,6 +2094,7 @@
 		lum8x8 = lum_8x8_c;
 		lum2x8 = lum_2x8_c;
 		csim   = consim_c;
+#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)
 		if (cpu->cpu & XVID_CPU_MMX){
 			lum8x8 = lum_8x8_mmx;
 			csim = consim_mmx;
@@ -2065,7 +2102,7 @@
 		if (cpu->cpu & XVID_CPU_MMX){
 			csim = consim_sse2;
 		}
-
+#endif
 		t = gettime_usec();
 		emms();
 		for(tst=0; tst<nb_tests; ++tst) m = lum8x8(Ref1, 16);
@@ -2097,6 +2134,62 @@
 }
 
 /*********************************************************************
+ * test bitstream functions
+ *********************************************************************/
+
+#define BIT_BUF_SIZE 2000
+
+static void test_bits()
+{
+  const int nb_tests = 50*speed_ref;
+  int tst;
+  uint32_t Crc;
+  uint8_t Buf[BIT_BUF_SIZE];
+  uint32_t Extracted[BIT_BUF_SIZE*8]; /* worst case: bits read 1 by 1 */
+  int Lens[BIT_BUF_SIZE*8];     
+  double t1;
+
+
+  printf( "\n ===  test bitstream ===\n" );
+  ieee_reseed(1);
+  Crc = 0;
+
+  t1 = gettime_usec();
+  for(tst=0; tst<nb_tests; ++tst) {
+  	Bitstream bs;
+  	int m, m2, l, l2;
+
+	for(l=0; l<BIT_BUF_SIZE; ++l)
+		Buf[l] = (uint8_t)ieee_rand(0,255);
+
+  	l = BIT_BUF_SIZE - ieee_rand(1,BIT_BUF_SIZE/10);
+  	BitstreamInit(&bs, (void*)(Buf+BIT_BUF_SIZE-l), l);
+
+
+	BitstreamReset(&bs);
+	for(l2=l*8, m=0; l2>0; m++) {
+		const int b = ieee_rand(1,32);
+		Lens[m] = b;
+		l2 -= b;
+		if (l2<0) break;
+		Extracted[m] = BitstreamShowBits(&bs, b);
+		BitstreamSkip(&bs, b);
+//		printf( "<= %d: %d 0x%x\n", m, b, Extracted[m]);
+	}
+
+	BitstreamReset(&bs);
+	for(m2=0; m2<m; ++m2) {
+		const int b = Lens[m2];
+		const uint32_t v = BitstreamGetBits(&bs, b);
+		Crc |= (v!=Extracted[m2]);
+//		printf( "=> %d: %d 0x%x %c\n", m2, b, v, " *"[Crc]);
+	}
+  }
+  t1 = (gettime_usec() - t1) / nb_tests;
+  printf(" test_bits   %.3f usec   %s\n", t1, (Crc!=0)?"| ERROR": "" );
+}
+
+/*********************************************************************
  * main
  *********************************************************************/
 
@@ -2111,7 +2204,7 @@
 	int c, what = 0;
 	int width, height;
 	uint32_t chksum = 0;
-  const char * test_bitstream = 0;
+	const char * test_bitstream = 0;
 
 	cpu_mask = 0;  // default => will use autodectect
 	for(c=1; c<argc; ++c)
@@ -2121,7 +2214,9 @@
 	  else if (!strcmp(argv[c], "-mmx"))    cpu_mask = XVID_CPU_MMX    | XVID_CPU_FORCE;
 	  else if (!strcmp(argv[c], "-mmxext")) cpu_mask = XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE;
 	  else if (!strcmp(argv[c], "-sse2"))   cpu_mask = XVID_CPU_SSE2   | XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE;
-	  else if (!strcmp(argv[c], "-3dnow"))  cpu_mask = XVID_CPU_3DNOW  | XVID_CPU_FORCE;
+	  else if (!strcmp(argv[c], "-sse3"))   cpu_mask = XVID_CPU_SSE3   | XVID_CPU_SSE2 | XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE;
+	  else if (!strcmp(argv[c], "-sse4"))   cpu_mask = XVID_CPU_SSE41  | XVID_CPU_SSE3 | XVID_CPU_SSE2 | XVID_CPU_MMXEXT | XVID_CPU_MMX | XVID_CPU_FORCE;
+      else if (!strcmp(argv[c], "-3dnow"))  cpu_mask = XVID_CPU_3DNOW  | XVID_CPU_FORCE;
 	  else if (!strcmp(argv[c], "-3dnowe")) cpu_mask = XVID_CPU_3DNOW  | XVID_CPU_3DNOWEXT | XVID_CPU_FORCE;
 	  else if (!strcmp(argv[c], "-altivec")) cpu_mask = XVID_CPU_ALTIVEC | XVID_CPU_FORCE;
 	  else if (!strcmp(argv[c], "-spd")) {
@@ -2168,7 +2263,7 @@
 	if (what==0 || what==14) test_yuv();
 	if (what==0 || what==15) test_SSIM();
 	if (what==0 || what==16) test_yuv2();
-
+	if (what==0 || what==17) test_bits();
 
 	if (what==7) {
 		test_IEEE1180_compliance(-256, 255, 1);
@@ -2196,7 +2291,7 @@
 			   "    may appear to be slow.\n");
 	}
 
-#ifdef ARCH_IS_IA32
+#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)
 	if (what == 0 || what == 5) {
 		printf("\n"
 			   "NB: MMX mpeg4 quantization is known to have very small errors (+/-1 magnitude)\n"