libfame/patches/libfame-0.9.1-pic.patch

diff -Nurp libfame-0.9.1/src/dct_mmx.h libfame-0.9.1-pic/src/dct_mmx.h
--- libfame-0.9.1/src/dct_mmx.h	2002-04-14 12:22:05.000000000 +0100
+++ libfame-0.9.1-pic/src/dct_mmx.h	2005-04-24 00:48:52.000000000 +0100
@@ -22,6 +22,9 @@
 
 #define precision
 
+extern FAME_ALIGNED short const _mmx_1[];
+extern FAME_ALIGNED short const _mmx_cos[];
+
 static void inline dct_aan_pass(dct_t *cache)
 {
   //  register unsigned short const *mmx_cos = _mmx_cos;
@@ -66,42 +69,42 @@ static void inline dct_aan_pass(dct_t *c
 #ifdef precision
       "psllw $0x01, %%mm5\n"                 /* precision(va0) += 1 bit */
 #endif
-    "paddw " ASMSYM "_mmx_1, %%mm4\n"   /* + 1 */
+    "paddw (%2), %%mm4\n"                    /* + 1 */
 //      "pmulhw 16(%1), %%mm5\n"               /* (v14+v16)*COS6 -> mm5 (va0) */
-      "pmulhw " ASMSYM "_mmx_cos+16, %%mm5\n" /* (v14+v16)*COS6 -> mm5 (va0) */
+      "pmulhw 16(%3), %%mm5\n"               /* (v14+v16)*COS6 -> mm5 (va0) */
       ""                                     /* STEP 4 */
 #ifdef precision
       "psllw $0x02, %%mm6\n"                 /* precision(v22) += 1 bit */
 #else
       "psllw $0x01, %%mm6\n"                 /* */
 #endif
-    "paddw " ASMSYM "_mmx_1, %%mm4\n"   /* + 1 */
+    "paddw (%2), %%mm4\n"                    /* + 1 */
       //      "pmulhw  8(%1), %%mm6\n"               /* 2*v22*COS4/2 -> mm6 (v32)*/
-      "pmulhw " ASMSYM "_mmx_cos+8, %%mm6\n" /* 2*v22*COS4/2 -> mm6 (v32)*/
+      "pmulhw 8(%3), %%mm6\n"                /* 2*v22*COS4/2 -> mm6 (v32)*/
 #ifdef precision
       "psllw $0x02, %%mm2\n"                 /* precision(v15) += 1 bit */
 #else
       "psllw $0x01, %%mm2\n"                 /* */
 #endif
-    "paddw " ASMSYM "_mmx_1, %%mm4\n"   /* + 1 */
+    "paddw (%2), %%mm4\n"                    /* + 1 */
       //      "pmulhw  8(%1), %%mm2\n"               /* 2*v15*COS4/2 -> mm2 (v35) */
-      "pmulhw " ASMSYM "_mmx_cos+8, %%mm2\n" /* 2*v15*COS4/2 -> mm2 (v35) */
+      "pmulhw 8(%3), %%mm2\n"                /* 2*v15*COS4/2 -> mm2 (v35) */
 #ifdef precision
       "psllw $0x02, %%mm4\n"                 /* precision(v14) += 1 bit */
 #else
       "psllw $0x01, %%mm4\n"                 /* */
 #endif
-    "paddw " ASMSYM "_mmx_1, %%mm4\n"   /* + 1 */
+    "paddw (%2), %%mm4\n"                    /* + 1 */
       //      "pmulhw  0(%1), %%mm4\n"               /* 2 * v14 * -COS2/2 -> mm4 */
-      "pmulhw " ASMSYM "_mmx_cos, %%mm4\n"   /* 2 * v14 * -COS2/2 -> mm4 */
+      "pmulhw (%3), %%mm4\n"                 /* 2 * v14 * -COS2/2 -> mm4 */
       "psubsw %%mm5, %%mm4\n"                /* v14*-COS2 - va0 -> mm4 (v34) */
 #ifdef precision
       "psllw $0x01, %%mm1\n"                 /* precision(v16) += 1 bit */
 #endif
       "psubsw %%mm1, %%mm5\n"                /* va0 - v16 -> mm5 */
-    "paddw " ASMSYM "_mmx_1, %%mm4\n"   /* + 1 */
+    "paddw (%2), %%mm4\n"                    /* + 1 */
       //      "pmulhw 24(%1), %%mm1\n"               /* v16 * (COS8 - 1) -> mm1 */
-      "pmulhw " ASMSYM "_mmx_cos+24, %%mm1\n" /* v16 * (COS8 - 1) -> mm1 */
+      "pmulhw 24(%3), %%mm1\n"               /* v16 * (COS8 - 1) -> mm1 */
       "psubsw %%mm5, %%mm1\n"                /* v16 * COS8 - va0 -> mm1 (v36)*/
       ""                                     /* STEP 5 */
       "movq 0x70(%0), %%mm0\n"               /* retrieve v07 -> mm0 */
@@ -138,8 +141,8 @@ static void inline dct_aan_pass(dct_t *c
       "movq %%mm0, 0x30(%0)\n"               /* store line 3 */
       "movq %%mm4, 0x50(%0)\n"               /* store line 5 */
       "movq %%mm2, 0x70(%0)\n"               /* store line 7 */
-      : "=r"(cache)/*, "=r"(mmx_cos)*/
-      : "0"(cache)/*, "1"(mmx_cos)*/
+      : "=r"(cache)
+      : "0"(cache), "r"(_mmx_1), "r"(_mmx_cos)
       : "memory");
 }
 
diff -Nurp libfame-0.9.1/src/dequantize_mmx.h libfame-0.9.1-pic/src/dequantize_mmx.h
--- libfame-0.9.1/src/dequantize_mmx.h	2002-04-23 22:40:56.000000000 +0100
+++ libfame-0.9.1-pic/src/dequantize_mmx.h	2005-04-24 00:44:26.000000000 +0100
@@ -27,8 +27,8 @@
     "pmullw 0x" #x "8(%3), %%mm5\n"   /* premultiply for iDCT */	 \
     "psrlw $0x0b, %%mm4\n"            /* keep 5 bits */ \
     "psrlw $0x0b, %%mm5\n"            /* keep 5 bits */ \
-    "paddw " ASMSYM "_mmx_1, %%mm4\n"   /* + 1 */               \
-    "paddw " ASMSYM "_mmx_1, %%mm5\n"   /* + 1 */               \
+    "paddw (%8), %%mm4\n"             /* + 1 */               \
+    "paddw (%8), %%mm5\n"             /* + 1 */               \
     "psrlw $0x01, %%mm4\n"            /* keep 4 bits rounded */ \
     "psrlw $0x01, %%mm5\n"            /* keep 4 bits rounded */ \
     "psllw $0x04, %%mm0\n"            /* multiply by 16 for iDCT */	 \
@@ -107,7 +107,7 @@ static void inline dequantize_intra_glob
 		DEQUANTIZE_GLOBAL_MISMATCH_CONTROL()
 		DEQUANTIZE_PRESCALE_STEP(7)
 	        : "=r"(block), "=r"(dqmatrix), "=r"(cache), "=r"(psmatrix)
-		: "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix)
+		: "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix), "r"(_mmx_1)
 		: "memory");
 
   asm volatile("movd %%mm6, %0\n"           /* export mismatch */
@@ -160,8 +160,8 @@ static void inline dequantize_intra_loca
     "pcmpeqw %%mm7, %%mm3\n"          /* invert sign */                  \
     "paddw %%mm2, %%mm0\n"            /* sub 1 if >0 */                  \
     "paddw %%mm3, %%mm1\n"            /* sub 1 if >0 */                  \
-    "por " ASMSYM "_mmx_1, %%mm0\n"   /* or 1 */               \
-    "por " ASMSYM "_mmx_1, %%mm1\n"   /* or 1 */              \
+    "por (%8), %%mm0\n"               /* or 1 */               \
+    "por (%8), %%mm1\n"               /* or 1 */              \
     "pand %%mm4, %%mm0\n"             /* [0-3]=0 if [0-3] was zero */	 \
     "pand %%mm5, %%mm1\n"             /* [4-7]=0 if [4-7] was zero */
 
@@ -184,7 +184,7 @@ static void inline dequantize_intra_loca
 		DEQUANTIZE_INTRA_LOCAL_STEP(7)
 		DEQUANTIZE_PRESCALE_STEP(7)
 	        : "=r"(block), "=r"(dqmatrix), "=r"(cache), "=r"(psmatrix)
-		: "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix)
+		: "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix), "r"(_mmx_1)
 		: "memory");
 }
 
@@ -256,7 +256,7 @@ static void inline dequantize_inter_glob
 		/* resetting the accumulator when the block is coded intra */
 		DEQUANTIZE_PRESCALE_STEP(7) 
 	        : "=r"(block), "=r"(dqmatrix), "=r"(cache), "=r"(psmatrix)
-		: "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix)
+		: "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix), "r"(_mmx_1)
 		: "memory");
 
   asm volatile("movd %%mm6, %0\n"           /* export mismatch */
@@ -324,8 +324,8 @@ static void inline dequantize_inter_loca
     "pcmpeqw %%mm7, %%mm3\n"          /* invert sign */                  \
     "paddw %%mm2, %%mm0\n"            /* sub 1 if >0 */                  \
     "paddw %%mm3, %%mm1\n"            /* sub 1 if >0 */                  \
-    "por " ASMSYM "_mmx_1, %%mm0\n"   /* or 1 */               \
-    "por " ASMSYM "_mmx_1, %%mm1\n"   /* or 1 */               \
+    "por (%8), %%mm0\n"               /* or 1 */               \
+    "por (%8), %%mm1\n"               /* or 1 */               \
     "pand %%mm4, %%mm0\n"             /* [0-3]=0 if [0-3] was zero */	 \
     "pand %%mm5, %%mm1\n"             /* [4-7]=0 if [4-7] was zero */
 
@@ -348,6 +348,6 @@ static void inline dequantize_inter_loca
 		DEQUANTIZE_INTER_LOCAL_STEP(7)
 		DEQUANTIZE_PRESCALE_STEP(7)
 	        : "=r"(block), "=r"(dqmatrix), "=r"(cache), "=r"(psmatrix)
-		: "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix)
+		: "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix), "r"(_mmx_1)
 		: "memory");
 }
diff -Nurp libfame-0.9.1/src/fame_syntax_mpeg1.c libfame-0.9.1-pic/src/fame_syntax_mpeg1.c
--- libfame-0.9.1/src/fame_syntax_mpeg1.c	2002-10-05 13:44:47.000000000 +0100
+++ libfame-0.9.1-pic/src/fame_syntax_mpeg1.c	2005-04-24 00:19:09.000000000 +0100
@@ -469,89 +469,6 @@ static void mpeg1_block_intra(fame_synta
   fast_bitbuffer_write(data, shift, table[v+255].code, table[v+255].length);
 
   /* encode AC coefficients */
-#if defined(HAS_BSWAP)
-  {
-    unsigned long dummy1, dummy2;
-
-    /* Note:
-        movsx mpeg1_table_clip+4096(, %%eax ,2), %%eax
-       has been replaced by
-        movw mpeg1_table_clip+4096(, %%eax ,2), %%ax
-        movsx %%ax, %%eax
-       because the first instruction failed on a PIII!! (wrong sign extension)
-       whereas it worked well on my P75 :)
-    */
-    /* Ok, a bit of explanations for a couple of tricks:
-         The DC value of block is already coded and stored in v so we can use it to store something.
-	 We add one index to the zigzag table so that after coding block[63] we go to index 0. There
-	 we need to escape the zero counting loop (1), what we ensure by putting a non-zero value in
-	 the DC coefficient. Then we can test for index == 0 to exit.
-	 Now this non-zero value is a bit special :)
-	 In order to have one more 'half' register, we store sp value (16 less significant bit of the
-	 32 bit register esp) *plus one* in the DC coefficient. Since the stack is aligned at an 
-	 address multiple of 4 bytes (at least), we are sure that sp != 0xffff and thus sp+1 will
-	 never be zero. We then retrieve sp at the end for it is needed by 'pop' instructions.
-     */
-    /* TODO : echange the role of edx and esp */
-    __asm__ __volatile__ ("pushl %%ebx\n"             /* save ebx            */
-			  "pushl %%ebp\n"             /* save stack pointer  */
-			  "inc %%sp\n"                /* make sure sp != 0   */
-			  "movw %%sp, (%%edx)\n"      /* store sp+1 in DC ;) */
-			  "movl %%esi, %%ebp\n"       /* ebp = vlc_table     */
-			  "xorl %%eax, %%eax\n"       /* eax = 0             */
-			  "movl $" ASMSYM "mpeg1_zigzag_table+1, %%esi\n" /*esi = zigzag*/
-			  "lea 1(%%esi), %%ebx\n"     /* ebx = zigzag_table+1*/
-			  "neg %%ebx\n"               /* ebx = -(esi+1)      */
-			  ".p2align 4,,7\n"           /* align for jump      */
-			  "0: xorw %%sp, %%sp\n"      /* sp = 0              */
-			  "1: movb (%%esi), %%al\n"   /* eax = index in block*/
-			  "incl %%esi\n"              /* (faster than lodsb) */
-			  "addw (%%edx, %%eax, 2), %%sp\n" /* sp = unzig     */
-			  "jz 1b\n"                   /* coeff == 0 then loop*/
-			  "orl %%eax, %%eax\n"        /* index == 0 then quit*/
-			  "jz 2f\n"                   /* (faster than jcxz)  */
-			  "movsx %%sp, %%eax\n"       /* extend sign         */
-			  "movw " ASMSYM "mpeg1_table_clip_data+4096(, %%eax ,2), %%ax\n" /*clip*/
-			  "movsx %%ax, %%eax\n"       /* extend sign         */
-			  "addl %%esi, %%ebx\n"       /* ebx = run           */
-			  "shll $7, %%eax\n"          /* eax *= 128(indexing)*/
-			  "lea (%%eax, %%ebx, 2), %%eax\n" /*eax = 2 * offset*/
-			  "lea (%%ebp, %%eax, 4), %%ebx\n" /* ebx = &vlc     */
-			  "movl (%%ebx), %%eax\n"     /* eax = code          */
-			  "addl 4(%%ebx), %%ecx\n"    /* ecx = shift+=length */
-			  "xorl %%ebx, %%ebx\n"       /* ebx = 0             */
-			  "shrd %%cl, %%eax, %%ebx\n" /* adjust code to fit  */
-			  "shr %%cl, %%eax\n"         /* adjust code to fit  */
-			  "bswap %%eax\n"      /* reverse byte order of code */
-			  "bswap %%ebx\n"      /* reverse byte order of code */
-			  "or %%eax, (%%edi)\n"       /* put first 32 bits   */
-			  "movl %%ecx, %%eax\n"       /* eax = shift + length*/
-			  "shrl $5, %%eax\n"          /* get dword increment */
-			  "andl $31, %%ecx\n"         /* mask shift          */
-			  "lea   (%%edi, %%eax, 4), %%edi\n"/* data+=(ecx>32)*/
-			  "orl %%ebx, (%%edi)\n"      /* put last 32 bits    */
-			  "xorl %%eax, %%eax\n"       /* eax = 0             */
-			  "lea 1(%%esi), %%ebx\n"     /* ebx = esi + 1 (last)*/
-			  "neg %%ebx\n"               /* ebx = -(esi + 1)    */
-			  "jmp 0b\n"                  /* loop                */
-			  "2:\n"
-			  "movw (%%edx), %%sp\n"      /* retrieve sp+1       */
-			  "dec %%sp\n"                /* restore esp */
-			  "popl %%ebp\n"              /* reload stack pointer*/
-			  "popl %%ebx\n"              /* reload ebx          */
-			  : "=c"(shift),
-			    "=a"(dummy1),
-			    "=d"(block),
-			    "=D"(data),
-			    "=S"(dummy2)
-			  : "d"(block),
-			    "c"(shift),
-			    "D"(data),
-                            "S"(syntax_mpeg1->vlc_table)
-			  : "memory");
-    block[0] = v; /* restore DC value */
-  }
-#else
   {
     short i;
     unsigned long last;
@@ -573,7 +490,6 @@ static void mpeg1_block_intra(fame_synta
       }
     }
   }
-#endif /* HAS_BSWAP */
 
   /* mark end of block */
   fast_bitbuffer_write(data, shift, 2, 2);
diff -Nurp libfame-0.9.1/src/half_mmx.h libfame-0.9.1-pic/src/half_mmx.h
--- libfame-0.9.1/src/half_mmx.h	2002-04-30 19:04:02.000000000 +0100
+++ libfame-0.9.1-pic/src/half_mmx.h	2005-04-24 00:44:49.000000000 +0100
@@ -68,8 +68,8 @@ static void inline mmx_interpolate(unsig
 		"paddw %%mm5, %%mm6\n"     /* mm6 = ref00+ref10+ref11+1-r 4-7*/
 		"psrlw $1, %%mm4\n"        /* divide by 2 */
 		"psrlw $1, %%mm5\n"        /* divide by 2 */
-		"paddw " ASMSYM "_mmx_one, %%mm3\n"  /* add 1 */
-		"paddw " ASMSYM "_mmx_one, %%mm6\n"  /* add 1 */
+		"paddw (%8), %%mm3\n"      /* add 1 */
+		"paddw (%8), %%mm6\n"      /* add 1 */
 		"packuswb %%mm5, %%mm4\n"  /* pack to byte and saturate */
 		"movq 1(%3), %%mm1\n"      /* mm1 = [ref+1] */
 		"movq %%mm1, %%mm2\n"      /* mm2 = mm1 */
@@ -87,7 +87,7 @@ static void inline mmx_interpolate(unsig
 		"movl 12(%0), %3\n"        /* %3 = ref[3] */
 		"movq %%mm3, (%3)\n"       /* store in frame */
 		: "=r"(ref), "=r"(pitch), "=r"(rc), "=r"(dummy)
-		: "0"(ref), "1"(pitch), "2"(rc), "3"(dummy)
+		: "0"(ref), "1"(pitch), "2"(rc), "3"(dummy), "r"(_mmx_one)
 		: "memory");
 }
 
diff -Nurp libfame-0.9.1/src/half_sse.h libfame-0.9.1-pic/src/half_sse.h
--- libfame-0.9.1/src/half_sse.h	2002-01-27 02:24:56.000000000 +0000
+++ libfame-0.9.1-pic/src/half_sse.h	2005-04-24 00:57:08.000000000 +0100
@@ -71,8 +71,8 @@ static void inline mmx_interpolate_signe
 		"paddw %%mm5, %%mm6\n"     /* mm6 = ref00+ref10+ref11+1-r 4-7*/
 		"psrlw $1, %%mm4\n"        /* divide by 2 */
 		"psrlw $1, %%mm5\n"        /* divide by 2 */
-		"paddw " ASMSYM "_mmx_one, %%mm3\n"  /* add 1 */
-		"paddw " ASMSYM "_mmx_one, %%mm6\n"  /* add 1 */
+		"paddw (%8), %%mm3\n"      /* add 1 */
+		"paddw (%8), %%mm6\n"      /* add 1 */
 		"packuswb %%mm5, %%mm4\n"  /* pack to byte and saturate */
 		"movq 1(%3), %%mm1\n"      /* mm1 = [ref+1] */
 		"movq %%mm1, %%mm2\n"      /* mm2 = mm1 */
@@ -90,7 +90,7 @@ static void inline mmx_interpolate_signe
 		"movl 12(%0), %3\n"        /* %3 = ref[3] */
 		"movq %%mm3, (%3)\n"       /* store in frame */
 		: "=r"(ref), "=r"(pitch), "=r"(rc), "=r"(dummy)
-		: "0"(ref), "1"(pitch), "2"(rc), "3"(dummy)
+		: "0"(ref), "1"(pitch), "2"(rc), "3"(dummy), "r"(_mmx_one)
 		: "memory");
 }
 
diff -Nurp libfame-0.9.1/src/idct_mmx.h libfame-0.9.1-pic/src/idct_mmx.h
--- libfame-0.9.1/src/idct_mmx.h	2002-04-14 12:22:05.000000000 +0100
+++ libfame-0.9.1-pic/src/idct_mmx.h	2005-04-24 00:51:00.000000000 +0100
@@ -18,6 +18,10 @@
 */
 /*************************** MMX accelerated iDCT ****************************/
 
+extern FAME_ALIGNED short const _mmx_1[];
+extern FAME_ALIGNED short const _mmx_cos[];
+extern FAME_ALIGNED short const _mmx_icos[];
+
 static void inline idct_aan_pass(dct_t * block)
 {
   //  register unsigned short const *mmx_icos = _mmx_icos;
@@ -65,9 +69,9 @@ static void inline idct_aan_pass(dct_t *
 	block[row*8+6] = v45;           - v71, v11, v44, v65, v24 -
       */
       "psllw $0x02, %%mm5\n"               /* adjust v22 for multiply      */
-      "paddw " ASMSYM "_mmx_1, %%mm5\n"   /* + 1 for rounding */
+      "paddw (%2), %%mm5\n"                /* + 1 for rounding */
       //      "pmulhw 8(%1), %%mm5\n"              /* 4*v15*ICOS4/4 -> mm5 (v23)   */
-      "pmulhw " ASMSYM "_mmx_icos+8, %%mm5\n" /* 4*v15*ICOS4/4 -> mm5 (v23)*/
+      "pmulhw 8(%3), %%mm5\n"              /* 4*v15*ICOS4/4 -> mm5 (v23)*/
       "psubsw %%mm4, %%mm5\n"              /* v23 - v62 -> mm5 (v24)       */
       "movq %%mm3, %%mm6\n"                /* v44 -> mm6                   */
       "paddsw %%mm5, %%mm6\n"              /* v44 + v24 -> mm6 (v45)       */
@@ -125,25 +129,25 @@ static void inline idct_aan_pass(dct_t *
 	block[row*8+4] += v55;          - -
       */
       "psllw $0x02, %%mm0\n"               /* adjust v12 for multiply      */
-      "paddw " ASMSYM "_mmx_1, %%mm0\n"   /* + 1 for rounding */
+      "paddw (%2), %%mm0\n"                /* + 1 for rounding */
       //      "pmulhw 8(%1), %%mm0\n"              /* 4*v12*ICOS4/4 -> mm0 (v13)   */
-      "pmulhw " ASMSYM "_mmx_icos+8, %%mm0\n" /* 4*v12*ICOS4/4 -> mm0 (v13)  */
+      "pmulhw 8(%3), %%mm0\n"              /* 4*v12*ICOS4/4 -> mm0 (v13)  */
       "movq   %%mm2, %%mm6\n"              /* v51 -> mm6                   */
       "psubsw %%mm1, %%mm6\n"              /* v51 - v71 -> mm6 (va2)       */
       "psllw $0x03, %%mm2\n"               /* adjust v51 for multiply      */
-      "paddw " ASMSYM "_mmx_1, %%mm2\n"   /* + 1 for rounding */
+      "paddw (%2), %%mm2\n"                /* + 1 for rounding */
       /* should add another one here but it seems to look better without */
       //      "pmulhw 16(%1), %%mm2\n"             /* 8*v51*ICOS6/8 -> mm2 (v53)   */
-      "pmulhw " ASMSYM "_mmx_icos+16, %%mm2\n" /* 8*v51*ICOS6/8 -> mm2 (v53) */
+      "pmulhw 16(%3), %%mm2\n"             /* 8*v51*ICOS6/8 -> mm2 (v53) */
       "psllw $0x02, %%mm1\n"               /* adjust v71 for multiply      */
-      "paddw " ASMSYM "_mmx_1, %%mm1\n"   /* + 1 for rounding */
+      "paddw (%2), %%mm1\n"                /* + 1 for rounding */
       /* should add another one here but it seems to look better without */
       //      "pmulhw 0(%1), %%mm1\n"              /* 4*v71*ICOS2/4 -> mm1 (v73)   */
-      "pmulhw " ASMSYM "_mmx_icos, %%mm1\n" /* 4*v71*ICOS2/4 -> mm1 (v73)   */
+      "pmulhw (%3), %%mm1\n"               /* 4*v71*ICOS2/4 -> mm1 (v73)   */
       "psllw $0x01, %%mm6\n"               /* adjust va2 for multiply      */
-      "paddw " ASMSYM "_mmx_1, %%mm6\n"   /* + 1 for rounding */
+      "paddw (%2), %%mm6\n"                /* + 1 for rounding */
       //      "pmulhw 24(%1), %%mm6\n"             /* 2*v12*ICOS8/2 -> mm6 (va3)   */
-      "pmulhw " ASMSYM "_mmx_icos+24, %%mm6\n" /* 2*v12*ICOS8/2 -> mm6 (va3) */
+      "pmulhw 24(%3), %%mm6\n"             /* 2*v12*ICOS8/2 -> mm6 (va3) */
       "psubsw %%mm6, %%mm2\n"              /* v53 - va3 -> mm2 (v54)       */
       "psubsw %%mm6, %%mm1\n"              /* v73 - va3 -> mm1 (v74)       */
       "psubsw %%mm3, %%mm1\n"              /* v74 - v32 -> mm3 (v75)       */
@@ -167,8 +171,8 @@ static void inline idct_aan_pass(dct_t *
       "paddsw %%mm0, %%mm7\n"              /* v65 + v55 -> mm7             */
       "movq %%mm6, 0x30(%0)\n"             /* mm6 -> line 3                */
       "movq %%mm7, 0x40(%0)\n"             /* mm7 -> line 4                */
-      : "=r"(block)/*, "=r"(mmx_icos)*/
-      : "0"(block)/*, "1"(mmx_icos)*/
+      : "=r"(block)
+      : "0"(block), "r"(_mmx_1), "r"(_mmx_icos)
       : "memory");
 }
 
1	niro	1115	diff -Nurp libfame-0.9.1/src/dct_mmx.h libfame-0.9.1-pic/src/dct_mmx.h
2			--- libfame-0.9.1/src/dct_mmx.h 2002-04-14 12:22:05.000000000 +0100
3			+++ libfame-0.9.1-pic/src/dct_mmx.h 2005-04-24 00:48:52.000000000 +0100
4			@@ -22,6 +22,9 @@
5
6			#define precision
7
8			+extern FAME_ALIGNED short const _mmx_1[];
9			+extern FAME_ALIGNED short const _mmx_cos[];
10			+
11			static void inline dct_aan_pass(dct_t *cache)
12			{
13			// register unsigned short const *mmx_cos = _mmx_cos;
14			@@ -66,42 +69,42 @@ static void inline dct_aan_pass(dct_t *c
15			#ifdef precision
16			"psllw $0x01, %%mm5\n" /* precision(va0) += 1 bit */
17			#endif
18			- "paddw " ASMSYM "_mmx_1, %%mm4\n" /* + 1 */
19			+ "paddw (%2), %%mm4\n" /* + 1 */
20			// "pmulhw 16(%1), %%mm5\n" /* (v14+v16)COS6 -> mm5 (va0) /
21			- "pmulhw " ASMSYM "_mmx_cos+16, %%mm5\n" /* (v14+v16)COS6 -> mm5 (va0) /
22			+ "pmulhw 16(%3), %%mm5\n" /* (v14+v16)COS6 -> mm5 (va0) /
23			"" /* STEP 4 */
24			#ifdef precision
25			"psllw $0x02, %%mm6\n" /* precision(v22) += 1 bit */
26			#else
27			"psllw $0x01, %%mm6\n" /* */
28			#endif
29			- "paddw " ASMSYM "_mmx_1, %%mm4\n" /* + 1 */
30			+ "paddw (%2), %%mm4\n" /* + 1 */
31			// "pmulhw 8(%1), %%mm6\n" /* 2v22COS4/2 -> mm6 (v32)*/
32			- "pmulhw " ASMSYM "_mmx_cos+8, %%mm6\n" /* 2v22COS4/2 -> mm6 (v32)*/
33			+ "pmulhw 8(%3), %%mm6\n" /* 2v22COS4/2 -> mm6 (v32)*/
34			#ifdef precision
35			"psllw $0x02, %%mm2\n" /* precision(v15) += 1 bit */
36			#else
37			"psllw $0x01, %%mm2\n" /* */
38			#endif
39			- "paddw " ASMSYM "_mmx_1, %%mm4\n" /* + 1 */
40			+ "paddw (%2), %%mm4\n" /* + 1 */
41			// "pmulhw 8(%1), %%mm2\n" /* 2v15COS4/2 -> mm2 (v35) */
42			- "pmulhw " ASMSYM "_mmx_cos+8, %%mm2\n" /* 2v15COS4/2 -> mm2 (v35) */
43			+ "pmulhw 8(%3), %%mm2\n" /* 2v15COS4/2 -> mm2 (v35) */
44			#ifdef precision
45			"psllw $0x02, %%mm4\n" /* precision(v14) += 1 bit */
46			#else
47			"psllw $0x01, %%mm4\n" /* */
48			#endif
49			- "paddw " ASMSYM "_mmx_1, %%mm4\n" /* + 1 */
50			+ "paddw (%2), %%mm4\n" /* + 1 */
51			// "pmulhw 0(%1), %%mm4\n" /* 2 * v14 * -COS2/2 -> mm4 */
52			- "pmulhw " ASMSYM "_mmx_cos, %%mm4\n" /* 2 * v14 * -COS2/2 -> mm4 */
53			+ "pmulhw (%3), %%mm4\n" /* 2 * v14 * -COS2/2 -> mm4 */
54			"psubsw %%mm5, %%mm4\n" /* v14-COS2 - va0 -> mm4 (v34) /
55			#ifdef precision
56			"psllw $0x01, %%mm1\n" /* precision(v16) += 1 bit */
57			#endif
58			"psubsw %%mm1, %%mm5\n" /* va0 - v16 -> mm5 */
59			- "paddw " ASMSYM "_mmx_1, %%mm4\n" /* + 1 */
60			+ "paddw (%2), %%mm4\n" /* + 1 */
61			// "pmulhw 24(%1), %%mm1\n" /* v16 * (COS8 - 1) -> mm1 */
62			- "pmulhw " ASMSYM "_mmx_cos+24, %%mm1\n" /* v16 * (COS8 - 1) -> mm1 */
63			+ "pmulhw 24(%3), %%mm1\n" /* v16 * (COS8 - 1) -> mm1 */
64			"psubsw %%mm5, %%mm1\n" /* v16 * COS8 - va0 -> mm1 (v36)*/
65			"" /* STEP 5 */
66			"movq 0x70(%0), %%mm0\n" /* retrieve v07 -> mm0 */
67			@@ -138,8 +141,8 @@ static void inline dct_aan_pass(dct_t *c
68			"movq %%mm0, 0x30(%0)\n" /* store line 3 */
69			"movq %%mm4, 0x50(%0)\n" /* store line 5 */
70			"movq %%mm2, 0x70(%0)\n" /* store line 7 */
71			- : "=r"(cache)/, "=r"(mmx_cos)/
72			- : "0"(cache)/, "1"(mmx_cos)/
73			+ : "=r"(cache)
74			+ : "0"(cache), "r"(_mmx_1), "r"(_mmx_cos)
75			: "memory");
76			}
77
78			diff -Nurp libfame-0.9.1/src/dequantize_mmx.h libfame-0.9.1-pic/src/dequantize_mmx.h
79			--- libfame-0.9.1/src/dequantize_mmx.h 2002-04-23 22:40:56.000000000 +0100
80			+++ libfame-0.9.1-pic/src/dequantize_mmx.h 2005-04-24 00:44:26.000000000 +0100
81			@@ -27,8 +27,8 @@
82			"pmullw 0x" #x "8(%3), %%mm5\n" /* premultiply for iDCT */ \
83			"psrlw $0x0b, %%mm4\n" /* keep 5 bits */ \
84			"psrlw $0x0b, %%mm5\n" /* keep 5 bits */ \
85			- "paddw " ASMSYM "_mmx_1, %%mm4\n" /* + 1 */ \
86			- "paddw " ASMSYM "_mmx_1, %%mm5\n" /* + 1 */ \
87			+ "paddw (%8), %%mm4\n" /* + 1 */ \
88			+ "paddw (%8), %%mm5\n" /* + 1 */ \
89			"psrlw $0x01, %%mm4\n" /* keep 4 bits rounded */ \
90			"psrlw $0x01, %%mm5\n" /* keep 4 bits rounded */ \
91			"psllw $0x04, %%mm0\n" /* multiply by 16 for iDCT */ \
92			@@ -107,7 +107,7 @@ static void inline dequantize_intra_glob
93			DEQUANTIZE_GLOBAL_MISMATCH_CONTROL()
94			DEQUANTIZE_PRESCALE_STEP(7)
95			: "=r"(block), "=r"(dqmatrix), "=r"(cache), "=r"(psmatrix)
96			- : "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix)
97			+ : "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix), "r"(_mmx_1)
98			: "memory");
99
100			asm volatile("movd %%mm6, %0\n" /* export mismatch */
101			@@ -160,8 +160,8 @@ static void inline dequantize_intra_loca
102			"pcmpeqw %%mm7, %%mm3\n" /* invert sign */ \
103			"paddw %%mm2, %%mm0\n" /* sub 1 if >0 */ \
104			"paddw %%mm3, %%mm1\n" /* sub 1 if >0 */ \
105			- "por " ASMSYM "_mmx_1, %%mm0\n" /* or 1 */ \
106			- "por " ASMSYM "_mmx_1, %%mm1\n" /* or 1 */ \
107			+ "por (%8), %%mm0\n" /* or 1 */ \
108			+ "por (%8), %%mm1\n" /* or 1 */ \
109			"pand %%mm4, %%mm0\n" /* [0-3]=0 if [0-3] was zero */ \
110			"pand %%mm5, %%mm1\n" /* [4-7]=0 if [4-7] was zero */
111
112			@@ -184,7 +184,7 @@ static void inline dequantize_intra_loca
113			DEQUANTIZE_INTRA_LOCAL_STEP(7)
114			DEQUANTIZE_PRESCALE_STEP(7)
115			: "=r"(block), "=r"(dqmatrix), "=r"(cache), "=r"(psmatrix)
116			- : "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix)
117			+ : "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix), "r"(_mmx_1)
118			: "memory");
119			}
120
121			@@ -256,7 +256,7 @@ static void inline dequantize_inter_glob
122			/* resetting the accumulator when the block is coded intra */
123			DEQUANTIZE_PRESCALE_STEP(7)
124			: "=r"(block), "=r"(dqmatrix), "=r"(cache), "=r"(psmatrix)
125			- : "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix)
126			+ : "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix), "r"(_mmx_1)
127			: "memory");
128
129			asm volatile("movd %%mm6, %0\n" /* export mismatch */
130			@@ -324,8 +324,8 @@ static void inline dequantize_inter_loca
131			"pcmpeqw %%mm7, %%mm3\n" /* invert sign */ \
132			"paddw %%mm2, %%mm0\n" /* sub 1 if >0 */ \
133			"paddw %%mm3, %%mm1\n" /* sub 1 if >0 */ \
134			- "por " ASMSYM "_mmx_1, %%mm0\n" /* or 1 */ \
135			- "por " ASMSYM "_mmx_1, %%mm1\n" /* or 1 */ \
136			+ "por (%8), %%mm0\n" /* or 1 */ \
137			+ "por (%8), %%mm1\n" /* or 1 */ \
138			"pand %%mm4, %%mm0\n" /* [0-3]=0 if [0-3] was zero */ \
139			"pand %%mm5, %%mm1\n" /* [4-7]=0 if [4-7] was zero */
140
141			@@ -348,6 +348,6 @@ static void inline dequantize_inter_loca
142			DEQUANTIZE_INTER_LOCAL_STEP(7)
143			DEQUANTIZE_PRESCALE_STEP(7)
144			: "=r"(block), "=r"(dqmatrix), "=r"(cache), "=r"(psmatrix)
145			- : "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix)
146			+ : "0"(block), "1"(dqmatrix), "2"(cache), "3"(psmatrix), "r"(_mmx_1)
147			: "memory");
148			}
149			diff -Nurp libfame-0.9.1/src/fame_syntax_mpeg1.c libfame-0.9.1-pic/src/fame_syntax_mpeg1.c
150			--- libfame-0.9.1/src/fame_syntax_mpeg1.c 2002-10-05 13:44:47.000000000 +0100
151			+++ libfame-0.9.1-pic/src/fame_syntax_mpeg1.c 2005-04-24 00:19:09.000000000 +0100
152			@@ -469,89 +469,6 @@ static void mpeg1_block_intra(fame_synta
153			fast_bitbuffer_write(data, shift, table[v+255].code, table[v+255].length);
154
155			/* encode AC coefficients */
156			-#if defined(HAS_BSWAP)
157			- {
158			- unsigned long dummy1, dummy2;
159			-
160			- /* Note:
161			- movsx mpeg1_table_clip+4096(, %%eax ,2), %%eax
162			- has been replaced by
163			- movw mpeg1_table_clip+4096(, %%eax ,2), %%ax
164			- movsx %%ax, %%eax
165			- because the first instruction failed on a PIII!! (wrong sign extension)
166			- whereas it worked well on my P75 :)
167			- */
168			- /* Ok, a bit of explanations for a couple of tricks:
169			- The DC value of block is already coded and stored in v so we can use it to store something.
170			- We add one index to the zigzag table so that after coding block[63] we go to index 0. There
171			- we need to escape the zero counting loop (1), what we ensure by putting a non-zero value in
172			- the DC coefficient. Then we can test for index == 0 to exit.
173			- Now this non-zero value is a bit special :)
174			- In order to have one more 'half' register, we store sp value (16 less significant bit of the
175			- 32 bit register esp) plus one in the DC coefficient. Since the stack is aligned at an
176			- address multiple of 4 bytes (at least), we are sure that sp != 0xffff and thus sp+1 will
177			- never be zero. We then retrieve sp at the end for it is needed by 'pop' instructions.
178			- */
179			- /* TODO : echange the role of edx and esp */
180			- __asm__ __volatile__ ("pushl %%ebx\n" /* save ebx */
181			- "pushl %%ebp\n" /* save stack pointer */
182			- "inc %%sp\n" /* make sure sp != 0 */
183			- "movw %%sp, (%%edx)\n" /* store sp+1 in DC ;) */
184			- "movl %%esi, %%ebp\n" /* ebp = vlc_table */
185			- "xorl %%eax, %%eax\n" /* eax = 0 */
186			- "movl $" ASMSYM "mpeg1_zigzag_table+1, %%esi\n" /esi = zigzag/
187			- "lea 1(%%esi), %%ebx\n" /* ebx = zigzag_table+1*/
188			- "neg %%ebx\n" /* ebx = -(esi+1) */
189			- ".p2align 4,,7\n" /* align for jump */
190			- "0: xorw %%sp, %%sp\n" /* sp = 0 */
191			- "1: movb (%%esi), %%al\n" /* eax = index in block*/
192			- "incl %%esi\n" /* (faster than lodsb) */
193			- "addw (%%edx, %%eax, 2), %%sp\n" /* sp = unzig */
194			- "jz 1b\n" /* coeff == 0 then loop*/
195			- "orl %%eax, %%eax\n" /* index == 0 then quit*/
196			- "jz 2f\n" /* (faster than jcxz) */
197			- "movsx %%sp, %%eax\n" /* extend sign */
198			- "movw " ASMSYM "mpeg1_table_clip_data+4096(, %%eax ,2), %%ax\n" /clip/
199			- "movsx %%ax, %%eax\n" /* extend sign */
200			- "addl %%esi, %%ebx\n" /* ebx = run */
201			- "shll $7, %%eax\n" /* eax = 128(indexing)/
202			- "lea (%%eax, %%ebx, 2), %%eax\n" /eax = 2 offset*/
203			- "lea (%%ebp, %%eax, 4), %%ebx\n" /* ebx = &vlc */
204			- "movl (%%ebx), %%eax\n" /* eax = code */
205			- "addl 4(%%ebx), %%ecx\n" /* ecx = shift+=length */
206			- "xorl %%ebx, %%ebx\n" /* ebx = 0 */
207			- "shrd %%cl, %%eax, %%ebx\n" /* adjust code to fit */
208			- "shr %%cl, %%eax\n" /* adjust code to fit */
209			- "bswap %%eax\n" /* reverse byte order of code */
210			- "bswap %%ebx\n" /* reverse byte order of code */
211			- "or %%eax, (%%edi)\n" /* put first 32 bits */
212			- "movl %%ecx, %%eax\n" /* eax = shift + length*/
213			- "shrl $5, %%eax\n" /* get dword increment */
214			- "andl $31, %%ecx\n" /* mask shift */
215			- "lea (%%edi, %%eax, 4), %%edi\n"/* data+=(ecx>32)*/
216			- "orl %%ebx, (%%edi)\n" /* put last 32 bits */
217			- "xorl %%eax, %%eax\n" /* eax = 0 */
218			- "lea 1(%%esi), %%ebx\n" /* ebx = esi + 1 (last)*/
219			- "neg %%ebx\n" /* ebx = -(esi + 1) */
220			- "jmp 0b\n" /* loop */
221			- "2:\n"
222			- "movw (%%edx), %%sp\n" /* retrieve sp+1 */
223			- "dec %%sp\n" /* restore esp */
224			- "popl %%ebp\n" /* reload stack pointer*/
225			- "popl %%ebx\n" /* reload ebx */
226			- : "=c"(shift),
227			- "=a"(dummy1),
228			- "=d"(block),
229			- "=D"(data),
230			- "=S"(dummy2)
231			- : "d"(block),
232			- "c"(shift),
233			- "D"(data),
234			- "S"(syntax_mpeg1->vlc_table)
235			- : "memory");
236			- block[0] = v; /* restore DC value */
237			- }
238			-#else
239			{
240			short i;
241			unsigned long last;
242			@@ -573,7 +490,6 @@ static void mpeg1_block_intra(fame_synta
243			}
244			}
245			}
246			-#endif /* HAS_BSWAP */
247
248			/* mark end of block */
249			fast_bitbuffer_write(data, shift, 2, 2);
250			diff -Nurp libfame-0.9.1/src/half_mmx.h libfame-0.9.1-pic/src/half_mmx.h
251			--- libfame-0.9.1/src/half_mmx.h 2002-04-30 19:04:02.000000000 +0100
252			+++ libfame-0.9.1-pic/src/half_mmx.h 2005-04-24 00:44:49.000000000 +0100
253			@@ -68,8 +68,8 @@ static void inline mmx_interpolate(unsig
254			"paddw %%mm5, %%mm6\n" /* mm6 = ref00+ref10+ref11+1-r 4-7*/
255			"psrlw $1, %%mm4\n" /* divide by 2 */
256			"psrlw $1, %%mm5\n" /* divide by 2 */
257			- "paddw " ASMSYM "_mmx_one, %%mm3\n" /* add 1 */
258			- "paddw " ASMSYM "_mmx_one, %%mm6\n" /* add 1 */
259			+ "paddw (%8), %%mm3\n" /* add 1 */
260			+ "paddw (%8), %%mm6\n" /* add 1 */
261			"packuswb %%mm5, %%mm4\n" /* pack to byte and saturate */
262			"movq 1(%3), %%mm1\n" /* mm1 = [ref+1] */
263			"movq %%mm1, %%mm2\n" /* mm2 = mm1 */
264			@@ -87,7 +87,7 @@ static void inline mmx_interpolate(unsig
265			"movl 12(%0), %3\n" /* %3 = ref[3] */
266			"movq %%mm3, (%3)\n" /* store in frame */
267			: "=r"(ref), "=r"(pitch), "=r"(rc), "=r"(dummy)
268			- : "0"(ref), "1"(pitch), "2"(rc), "3"(dummy)
269			+ : "0"(ref), "1"(pitch), "2"(rc), "3"(dummy), "r"(_mmx_one)
270			: "memory");
271			}
272
273			diff -Nurp libfame-0.9.1/src/half_sse.h libfame-0.9.1-pic/src/half_sse.h
274			--- libfame-0.9.1/src/half_sse.h 2002-01-27 02:24:56.000000000 +0000
275			+++ libfame-0.9.1-pic/src/half_sse.h 2005-04-24 00:57:08.000000000 +0100
276			@@ -71,8 +71,8 @@ static void inline mmx_interpolate_signe
277			"paddw %%mm5, %%mm6\n" /* mm6 = ref00+ref10+ref11+1-r 4-7*/
278			"psrlw $1, %%mm4\n" /* divide by 2 */
279			"psrlw $1, %%mm5\n" /* divide by 2 */
280			- "paddw " ASMSYM "_mmx_one, %%mm3\n" /* add 1 */
281			- "paddw " ASMSYM "_mmx_one, %%mm6\n" /* add 1 */
282			+ "paddw (%8), %%mm3\n" /* add 1 */
283			+ "paddw (%8), %%mm6\n" /* add 1 */
284			"packuswb %%mm5, %%mm4\n" /* pack to byte and saturate */
285			"movq 1(%3), %%mm1\n" /* mm1 = [ref+1] */
286			"movq %%mm1, %%mm2\n" /* mm2 = mm1 */
287			@@ -90,7 +90,7 @@ static void inline mmx_interpolate_signe
288			"movl 12(%0), %3\n" /* %3 = ref[3] */
289			"movq %%mm3, (%3)\n" /* store in frame */
290			: "=r"(ref), "=r"(pitch), "=r"(rc), "=r"(dummy)
291			- : "0"(ref), "1"(pitch), "2"(rc), "3"(dummy)
292			+ : "0"(ref), "1"(pitch), "2"(rc), "3"(dummy), "r"(_mmx_one)
293			: "memory");
294			}
295
296			diff -Nurp libfame-0.9.1/src/idct_mmx.h libfame-0.9.1-pic/src/idct_mmx.h
297			--- libfame-0.9.1/src/idct_mmx.h 2002-04-14 12:22:05.000000000 +0100
298			+++ libfame-0.9.1-pic/src/idct_mmx.h 2005-04-24 00:51:00.000000000 +0100
299			@@ -18,6 +18,10 @@
300			*/
301			/************************* MMX accelerated iDCT **************************/
302
303			+extern FAME_ALIGNED short const _mmx_1[];
304			+extern FAME_ALIGNED short const _mmx_cos[];
305			+extern FAME_ALIGNED short const _mmx_icos[];
306			+
307			static void inline idct_aan_pass(dct_t * block)
308			{
309			// register unsigned short const *mmx_icos = _mmx_icos;
310			@@ -65,9 +69,9 @@ static void inline idct_aan_pass(dct_t *
311			block[row*8+6] = v45; - v71, v11, v44, v65, v24 -
312			*/
313			"psllw $0x02, %%mm5\n" /* adjust v22 for multiply */
314			- "paddw " ASMSYM "_mmx_1, %%mm5\n" /* + 1 for rounding */
315			+ "paddw (%2), %%mm5\n" /* + 1 for rounding */
316			// "pmulhw 8(%1), %%mm5\n" /* 4v15ICOS4/4 -> mm5 (v23) */
317			- "pmulhw " ASMSYM "_mmx_icos+8, %%mm5\n" /* 4v15ICOS4/4 -> mm5 (v23)*/
318			+ "pmulhw 8(%3), %%mm5\n" /* 4v15ICOS4/4 -> mm5 (v23)*/
319			"psubsw %%mm4, %%mm5\n" /* v23 - v62 -> mm5 (v24) */
320			"movq %%mm3, %%mm6\n" /* v44 -> mm6 */
321			"paddsw %%mm5, %%mm6\n" /* v44 + v24 -> mm6 (v45) */
322			@@ -125,25 +129,25 @@ static void inline idct_aan_pass(dct_t *
323			block[row*8+4] += v55; - -
324			*/
325			"psllw $0x02, %%mm0\n" /* adjust v12 for multiply */
326			- "paddw " ASMSYM "_mmx_1, %%mm0\n" /* + 1 for rounding */
327			+ "paddw (%2), %%mm0\n" /* + 1 for rounding */
328			// "pmulhw 8(%1), %%mm0\n" /* 4v12ICOS4/4 -> mm0 (v13) */
329			- "pmulhw " ASMSYM "_mmx_icos+8, %%mm0\n" /* 4v12ICOS4/4 -> mm0 (v13) */
330			+ "pmulhw 8(%3), %%mm0\n" /* 4v12ICOS4/4 -> mm0 (v13) */
331			"movq %%mm2, %%mm6\n" /* v51 -> mm6 */
332			"psubsw %%mm1, %%mm6\n" /* v51 - v71 -> mm6 (va2) */
333			"psllw $0x03, %%mm2\n" /* adjust v51 for multiply */
334			- "paddw " ASMSYM "_mmx_1, %%mm2\n" /* + 1 for rounding */
335			+ "paddw (%2), %%mm2\n" /* + 1 for rounding */
336			/* should add another one here but it seems to look better without */
337			// "pmulhw 16(%1), %%mm2\n" /* 8v51ICOS6/8 -> mm2 (v53) */
338			- "pmulhw " ASMSYM "_mmx_icos+16, %%mm2\n" /* 8v51ICOS6/8 -> mm2 (v53) */
339			+ "pmulhw 16(%3), %%mm2\n" /* 8v51ICOS6/8 -> mm2 (v53) */
340			"psllw $0x02, %%mm1\n" /* adjust v71 for multiply */
341			- "paddw " ASMSYM "_mmx_1, %%mm1\n" /* + 1 for rounding */
342			+ "paddw (%2), %%mm1\n" /* + 1 for rounding */
343			/* should add another one here but it seems to look better without */
344			// "pmulhw 0(%1), %%mm1\n" /* 4v71ICOS2/4 -> mm1 (v73) */
345			- "pmulhw " ASMSYM "_mmx_icos, %%mm1\n" /* 4v71ICOS2/4 -> mm1 (v73) */
346			+ "pmulhw (%3), %%mm1\n" /* 4v71ICOS2/4 -> mm1 (v73) */
347			"psllw $0x01, %%mm6\n" /* adjust va2 for multiply */
348			- "paddw " ASMSYM "_mmx_1, %%mm6\n" /* + 1 for rounding */
349			+ "paddw (%2), %%mm6\n" /* + 1 for rounding */
350			// "pmulhw 24(%1), %%mm6\n" /* 2v12ICOS8/2 -> mm6 (va3) */
351			- "pmulhw " ASMSYM "_mmx_icos+24, %%mm6\n" /* 2v12ICOS8/2 -> mm6 (va3) */
352			+ "pmulhw 24(%3), %%mm6\n" /* 2v12ICOS8/2 -> mm6 (va3) */
353			"psubsw %%mm6, %%mm2\n" /* v53 - va3 -> mm2 (v54) */
354			"psubsw %%mm6, %%mm1\n" /* v73 - va3 -> mm1 (v74) */
355			"psubsw %%mm3, %%mm1\n" /* v74 - v32 -> mm3 (v75) */
356			@@ -167,8 +171,8 @@ static void inline idct_aan_pass(dct_t *
357			"paddsw %%mm0, %%mm7\n" /* v65 + v55 -> mm7 */
358			"movq %%mm6, 0x30(%0)\n" /* mm6 -> line 3 */
359			"movq %%mm7, 0x40(%0)\n" /* mm7 -> line 4 */
360			- : "=r"(block)/, "=r"(mmx_icos)/
361			- : "0"(block)/, "1"(mmx_icos)/
362			+ : "=r"(block)
363			+ : "0"(block), "r"(_mmx_1), "r"(_mmx_icos)
364			: "memory");
365			}
366