src/audio/SDL_mixer_MMX.c
changeset 539 a9e38f3b8e4d
child 887 b4b64bb88f2f
equal deleted inserted replaced
538:d3abe873e3f7 539:a9e38f3b8e4d
       
     1 // MMX assembler version of SDL_MixAudio for signed little endian 16 bit samples and signed 8 bit samples
       
     2 // Copyright 2002 Stephane Marchesin (stephane.marchesin@wanadoo.fr)
       
     3 // This code is licensed under the LGPL (see COPYING for details)
       
     4 // 
       
     5 // Assumes buffer size in bytes is a multiple of 16
       
     6 // Assumes SDL_MIX_MAXVOLUME = 128
       
     7 
       
     8 
       
     9 ////////////////////////////////////////////////
       
    10 // Mixing for 16 bit signed buffers
       
    11 ////////////////////////////////////////////////
       
    12 
       
    13 #if defined(i386) && defined(__GNUC__) && defined(USE_ASMBLIT)
       
    14 void SDL_MixAudio_MMX_S16(char* dst,char* src,unsigned int size,int volume)
       
    15 {
       
    16     __asm__ __volatile__ (
       
    17 
       
    18 "	movl %0,%%edi\n"	// edi = dst
       
    19 "	movl %1,%%esi\n"	// esi = src
       
    20 "	movl %3,%%eax\n"	// eax = volume
       
    21 
       
    22 "	movl %2,%%ebx\n"	// ebx = size
       
    23 
       
    24 "	shrl $4,%%ebx\n"	// process 16 bytes per iteration = 8 samples
       
    25 
       
    26 "	jz .endS16\n"
       
    27 
       
    28 "	pxor %%mm0,%%mm0\n"
       
    29 
       
    30 "	movd %%eax,%%mm0\n"
       
    31 "	movq %%mm0,%%mm1\n"
       
    32 "	psllq $16,%%mm0\n"
       
    33 "	por %%mm1,%%mm0\n"
       
    34 "	psllq $16,%%mm0\n"
       
    35 "	por %%mm1,%%mm0\n"
       
    36 "	psllq $16,%%mm0\n"
       
    37 "	por %%mm1,%%mm0\n"		// mm0 = vol|vol|vol|vol
       
    38 
       
    39 ".align 16\n"
       
    40 "	.mixloopS16:\n"
       
    41 
       
    42 "	movq (%%esi),%%mm1\n" // mm1 = a|b|c|d
       
    43 
       
    44 "	movq %%mm1,%%mm2\n" // mm2 = a|b|c|d
       
    45 
       
    46 "	movq 8(%%esi),%%mm4\n" // mm4 = e|f|g|h
       
    47 
       
    48 	// pré charger le buffer dst dans mm7
       
    49 "	movq (%%edi),%%mm7\n" // mm7 = dst[0]"
       
    50 
       
    51 	// multiplier par le volume
       
    52 "	pmullw %%mm0,%%mm1\n" // mm1 = l(a*v)|l(b*v)|l(c*v)|l(d*v)
       
    53 
       
    54 "	pmulhw %%mm0,%%mm2\n" // mm2 = h(a*v)|h(b*v)|h(c*v)|h(d*v)
       
    55 "	movq %%mm4,%%mm5\n" // mm5 = e|f|g|h
       
    56 
       
    57 "	pmullw %%mm0,%%mm4\n" // mm4 = l(e*v)|l(f*v)|l(g*v)|l(h*v)
       
    58 
       
    59 "	pmulhw %%mm0,%%mm5\n" // mm5 = h(e*v)|h(f*v)|h(g*v)|h(h*v)
       
    60 "	movq %%mm1,%%mm3\n" // mm3 = l(a*v)|l(b*v)|l(c*v)|l(d*v)
       
    61 
       
    62 "	punpckhwd %%mm2,%%mm1\n" // mm1 = a*v|b*v
       
    63 
       
    64 "	movq %%mm4,%%mm6\n" // mm6 = l(e*v)|l(f*v)|l(g*v)|l(h*v)
       
    65 "	punpcklwd %%mm2,%%mm3\n" // mm3 = c*v|d*v
       
    66 
       
    67 "	punpckhwd %%mm5,%%mm4\n" // mm4 = e*f|f*v
       
    68 
       
    69 "	punpcklwd %%mm5,%%mm6\n" // mm6 = g*v|h*v
       
    70 
       
    71 	// pré charger le buffer dst dans mm5
       
    72 "	movq 8(%%edi),%%mm5\n" // mm5 = dst[1]
       
    73 
       
    74 	// diviser par 128
       
    75 "	psrad $7,%%mm1\n" // mm1 = a*v/128|b*v/128 , 128 = SDL_MIX_MAXVOLUME
       
    76 "	addl $16,%%esi\n"
       
    77 
       
    78 "	psrad $7,%%mm3\n" // mm3 = c*v/128|d*v/128
       
    79 
       
    80 "	psrad $7,%%mm4\n" // mm4 = e*v/128|f*v/128
       
    81 
       
    82 	// mm1 = le sample avec le volume modifié
       
    83 "	packssdw %%mm1,%%mm3\n" // mm3 = s(a*v|b*v|c*v|d*v)
       
    84 
       
    85 "	psrad $7,%%mm6\n" // mm6= g*v/128|h*v/128
       
    86 "	paddsw %%mm7,%%mm3\n" // mm3 = adjust_volume(src)+dst
       
    87 
       
    88 	// mm4 = le sample avec le volume modifié
       
    89 "	packssdw %%mm4,%%mm6\n" // mm6 = s(e*v|f*v|g*v|h*v)
       
    90 "	movq %%mm3,(%%edi)\n"
       
    91 
       
    92 "	paddsw %%mm5,%%mm6\n" // mm6 = adjust_volume(src)+dst
       
    93 
       
    94 "	movq %%mm6,8(%%edi)\n"
       
    95 
       
    96 "	addl $16,%%edi\n"
       
    97 
       
    98 "	dec %%ebx\n"
       
    99 
       
   100 "	jnz .mixloopS16\n"
       
   101 
       
   102 "	emms\n"
       
   103 
       
   104 ".endS16:\n"
       
   105 	 :
       
   106 	 : "m" (dst), "m"(src),"m"(size),
       
   107 	 "m"(volume)
       
   108 	 : "eax","ebx", "esi", "edi","memory"
       
   109 	 );
       
   110 }
       
   111 
       
   112 
       
   113 
       
   114 ////////////////////////////////////////////////
       
   115 // Mixing for 8 bit signed buffers
       
   116 ////////////////////////////////////////////////
       
   117 
       
   118 void SDL_MixAudio_MMX_S8(char* dst,char* src,unsigned int size,int volume)
       
   119 {
       
   120     __asm__ __volatile__ (
       
   121 
       
   122 "	movl %0,%%edi\n"	// edi = dst
       
   123 "	movl %1,%%esi\n"	// esi = src
       
   124 "	movl %3,%%eax\n"	// eax = volume
       
   125 
       
   126 "	movd %%ebx,%%mm0\n"
       
   127 "	movq %%mm0,%%mm1\n"
       
   128 "	psllq $16,%%mm0\n"
       
   129 "	por %%mm1,%%mm0\n"
       
   130 "	psllq $16,%%mm0\n"
       
   131 "	por %%mm1,%%mm0\n"
       
   132 "	psllq $16,%%mm0\n"
       
   133 "	por %%mm1,%%mm0\n"
       
   134 
       
   135 "	movl %2,%%ebx\n"	// ebx = size
       
   136 "	shr $3,%%ebx\n"	// process 8 bytes per iteration = 8 samples
       
   137 
       
   138 "	cmp $0,%%ebx\n"
       
   139 "	je .endS8\n"
       
   140 
       
   141 ".align 16\n"
       
   142 "	.mixloopS8:\n"
       
   143 
       
   144 "	pxor %%mm2,%%mm2\n"		// mm2 = 0
       
   145 "	movq (%%esi),%%mm1\n"	// mm1 = a|b|c|d|e|f|g|h
       
   146 
       
   147 "	movq %%mm1,%%mm3\n" 	// mm3 = a|b|c|d|e|f|g|h
       
   148 
       
   149 	// on va faire le "sign extension" en faisant un cmp avec 0 qui retourne 1 si <0, 0 si >0
       
   150 "	pcmpgtb %%mm1,%%mm2\n"	// mm2 = 11111111|00000000|00000000....
       
   151 
       
   152 "	punpckhbw %%mm2,%%mm1\n"	// mm1 = 0|a|0|b|0|c|0|d
       
   153 
       
   154 "	punpcklbw %%mm2,%%mm3\n"	// mm3 = 0|e|0|f|0|g|0|h
       
   155 "	movq (%%edi),%%mm2\n"	// mm2 = destination
       
   156 
       
   157 "	pmullw %%mm0,%%mm1\n"	// mm1 = v*a|v*b|v*c|v*d
       
   158 "	addl $8,%%esi\n"
       
   159 
       
   160 "	pmullw %%mm0,%%mm3\n"	// mm3 = v*e|v*f|v*g|v*h
       
   161 "	psraw $7,%%mm1\n"		// mm1 = v*a/128|v*b/128|v*c/128|v*d/128 
       
   162 
       
   163 "	psraw $7,%%mm3\n"		// mm3 = v*e/128|v*f/128|v*g/128|v*h/128
       
   164 
       
   165 "	packsswb %%mm1,%%mm3\n"	// mm1 = v*a/128|v*b/128|v*c/128|v*d/128|v*e/128|v*f/128|v*g/128|v*h/128
       
   166 
       
   167 "	paddsb %%mm2,%%mm3\n"	// add to destination buffer
       
   168 
       
   169 "	movq %%mm3,(%%edi)\n"	// store back to ram
       
   170 "	addl $8,%%edi\n"
       
   171 
       
   172 "	dec %%ebx\n"
       
   173 
       
   174 "	jnz .mixloopS8\n"
       
   175 
       
   176 ".endS8:\n"
       
   177 "	emms\n"
       
   178 	 :
       
   179 	 : "m" (dst), "m"(src),"m"(size),
       
   180 	 "m"(volume)
       
   181 	 : "eax","ebx", "esi", "edi","memory"
       
   182 	 );
       
   183 }
       
   184 #endif
       
   185