src/mesa/tnl/t_vtx_x86_gcc.S

   1 /**************************************************************************
   2
   3 Copyright 2004 Tungsten Graphics Inc., Cedar Park, Texas.
   4
   5 All Rights Reserved.
   6
   7 Permission is hereby granted, free of charge, to any person obtaining a
   8 copy of this software and associated documentation files (the "Software"),
   9 to deal in the Software without restriction, including without limitation
  10 on the rights to use, copy, modify, merge, publish, distribute, sub
  11 license, and/or sell copies of the Software, and to permit persons to whom
  12 the Software is furnished to do so, subject to the following conditions:
  13
  14 The above copyright notice and this permission notice (including the next
  15 paragraph) shall be included in all copies or substantial portions of the
  16 Software.
  17
  18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  19 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  20 FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  21 ATI, TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  22 DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  23 OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  24 USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26 **************************************************************************/
  27
  28 /*
  29  * Authors:
  30  *   Keith Whitwell <keith@tungstengraphics.com>
  31  */
  32
  33
  34 #define GLOBL( x )      \
  35 .globl x;               \
  36 x:
  37
  38 .data
  39 .align 4
  40
  41 // Someone who knew a lot about this sort of thing would use this
  42 // macro to note current offsets, etc in a special region of the
  43 // object file & just make everything work out neat.  I don't know
  44 // enough to do that...
  45
  46 #define SUBST( x ) (0x10101010 + x)
  47
  48
  49
  50 GLOBL ( _x86_Vertex1fv )
  51         movl    4(%esp), %ecx
  52         push    %edi
  53         push    %esi
  54         movl    SUBST(0), %edi  // 0x0 --> tnl->vtx.vbptr
  55         movl    (%ecx), %edx    // load v[0]
  56         movl    %edx, (%edi)    // tnl->vtx.vbptr[0] = v[0]
  57         addl    $4, %edi        // tnl->vtx.vbptr += 1
  58         movl    $SUBST(1), %ecx // 0x1 --> (tnl->vtx.vertex_size - 1)
  59         movl    $SUBST(2), %esi // 0x2 --> (tnl->vtx.vertex + 1)
  60         repz
  61         movsl   %ds:(%esi), %es:(%edi)
  62         movl    %edi, SUBST(0)  // 0x0 --> tnl->vtx.vbptr
  63         movl    SUBST(3), %edx  // 0x3 --> counter
  64         pop     %esi
  65         pop     %edi
  66         dec     %edx            // counter--
  67         movl    %edx, SUBST(3)  // 0x3 --> counter
  68         je      .5              // if (counter != 0)
  69         ret                     //    return
  70 .5:     mov     $SUBST(4), %eax // else notify()
  71         jmp     *%eax           // jmp $0x10101014 doesn't seem to work
  72 GLOBL ( _x86_Vertex1fv_end )
  73
  74
  75 .align 4
  76 GLOBL ( _x86_Vertex2fv )
  77         movl    4(%esp), %ecx
  78         push    %edi
  79         push    %esi
  80         movl    SUBST(0), %edi  // load tnl->vtx.vbptr
  81         movl    (%ecx), %edx    // load v[0]
  82         movl    4(%ecx), %eax   // load v[1]
  83         movl    %edx, (%edi)    // tnl->vtx.vbptr[0] = v[0]
  84         movl    %eax, 4(%edi)   // tnl->vtx.vbptr[1] = v[1]
  85         addl    $8, %edi        // tnl->vtx.vbptr += 2
  86         movl    $SUBST(1), %ecx // vertex_size - 2
  87         movl    $SUBST(2), %esi // tnl->vtx.vertex + 2
  88         repz
  89         movsl %ds:(%esi), %es:(%edi)
  90         movl    %edi, SUBST(0)  // save tnl->vtx.vbptr
  91         movl    SUBST(3), %edx  // load counter
  92         pop     %esi
  93         pop     %edi
  94         dec     %edx            // counter--
  95         movl    %edx, SUBST(3)  // save counter
  96         je      .6              // if (counter != 0)
  97         ret                     //    return
  98 .6:     mov     $SUBST(4), %eax // else notify()
  99         jmp     *%eax           // jmp $0x10101014 doesn't seem to work
 100 GLOBL ( _x86_Vertex2fv_end )
 101
 102 .align 4
 103 GLOBL ( _x86_Vertex3fv )
 104         movl    4(%esp), %ecx
 105         push    %edi
 106         push    %esi
 107         movl    SUBST(0), %edi  // load tnl->vtx.vbptr
 108         movl    (%ecx), %edx    // load v[0]
 109         movl    4(%ecx), %eax   // load v[1]
 110         movl    8(%ecx), %esi   // load v[2]
 111         movl    %edx, (%edi)    // tnl->vtx.vbptr[0] = v[0]
 112         movl    %eax, 4(%edi)   // tnl->vtx.vbptr[1] = v[1]
 113         movl    %esi, 8(%edi)   // tnl->vtx.vbptr[2] = v[2]
 114         addl    $12, %edi       // tnl->vtx.vbptr += 3
 115         movl    $SUBST(1), %ecx // vertex_size - 3
 116         movl    $SUBST(2), %esi // tnl->vtx.vertex + 3
 117         repz
 118         movsl %ds:(%esi), %es:(%edi)
 119         movl    %edi, SUBST(0)  // save tnl->vtx.vbptr
 120         movl    SUBST(3), %edx  // load counter
 121         pop     %esi
 122         pop     %edi
 123         dec     %edx            // counter--
 124         movl    %edx, SUBST(3)  // save counter
 125         je      .7              // if (counter != 0)
 126         ret                     //    return
 127 .7:     mov     $SUBST(4), %eax // else notify()
 128         jmp     *%eax           // jmp $0x10101014 doesn't seem to work
 129 GLOBL ( _x86_Vertex3fv_end )
 130
 131
 132 .align 4
 133 GLOBL ( _x86_Vertex4fv )
 134         movl    4(%esp), %ecx
 135         push    %edi
 136         push    %esi
 137         movl    SUBST(0), %edi  // load tnl->vtx.vbptr
 138         movl    (%ecx), %edx    // load v[0]
 139         movl    4(%ecx), %eax   // load v[1]
 140         movl    8(%ecx), %esi   // load v[2]
 141         movl    12(%ecx), %ecx  // load v[3]
 142         movl    %edx, (%edi)    // tnl->vtx.vbptr[0] = v[0]
 143         movl    %eax, 4(%edi)   // tnl->vtx.vbptr[1] = v[1]
 144         movl    %esi, 8(%edi)   // tnl->vtx.vbptr[2] = v[2]
 145         movl    %ecx, 12(%edi)  // tnl->vtx.vbptr[3] = v[3]
 146         addl    $16, %edi       // tnl->vtx.vbptr += 4
 147         movl    $SUBST(1), %ecx // vertex_size - 4
 148         movl    $SUBST(2), %esi // tnl->vtx.vertex + 3
 149         repz
 150         movsl   %ds:(%esi), %es:(%edi)
 151         movl    %edi, SUBST(0)  // save tnl->vtx.vbptr
 152         movl    SUBST(3), %edx  // load counter
 153         pop     %esi
 154         pop     %edi
 155         dec     %edx            // counter--
 156         movl    %edx, SUBST(3)  // save counter
 157         je      .6              // if (counter != 0)
 158         ret                     //    return
 159 .8:     mov     $SUBST(4), %eax // else notify()
 160         jmp     *%eax           // jmp $0x10101014 doesn't seem to work
 161 GLOBL ( _x86_Vertex4fv_end )
 162
 163
 164
 165 /**
 166  * Generic handlers for vector format data.
 167  */
 168
 169 GLOBL( _x86_Attribute1fv)
 170         movl 4(%esp), %ecx
 171         movl (%ecx), %eax       /* load v[0] */
 172         movl %eax, SUBST(0)     /* store v[0] to current vertex */
 173         ret
 174 GLOBL ( _x86_Attribute1fv_end )
 175
 176 GLOBL( _x86_Attribute2fv)
 177         movl 4(%esp), %ecx
 178         movl (%ecx), %eax       /* load v[0] */
 179         movl 4(%ecx), %edx      /* load v[1] */
 180         movl %eax, SUBST(0)     /* store v[0] to current vertex */
 181         movl %edx, SUBST(1)     /* store v[1] to current vertex */
 182         ret
 183 GLOBL ( _x86_Attribute2fv_end )
 184
 185
 186 GLOBL( _x86_Attribute3fv)
 187         movl 4(%esp), %ecx
 188         movl (%ecx), %eax       /* load v[0] */
 189         movl 4(%ecx), %edx      /* load v[1] */
 190         movl 8(%ecx), %ecx      /* load v[2] */
 191         movl %eax, SUBST(0)     /* store v[0] to current vertex */
 192         movl %edx, SUBST(1)     /* store v[1] to current vertex */
 193         movl %ecx, SUBST(2)     /* store v[2] to current vertex */
 194         ret
 195 GLOBL ( _x86_Attribute3fv_end )
 196
 197 GLOBL( _x86_Attribute4fv)
 198         movl 4(%esp), %ecx
 199         movl (%ecx), %eax       /* load v[0] */
 200         movl 4(%ecx), %edx      /* load v[1] */
 201         movl %eax, SUBST(0)     /* store v[0] to current vertex */
 202         movl %edx, SUBST(1)     /* store v[1] to current vertex */
 203         movl 8(%ecx), %eax      /* load v[2] */
 204         movl 12(%ecx), %edx     /* load v[3] */
 205         movl %eax, SUBST(2)     /* store v[2] to current vertex */
 206         movl %edx, SUBST(3)     /* store v[3] to current vertex */
 207         ret
 208 GLOBL ( _x86_Attribute4fv_end )
 209
 210
 211 // Choosers:
 212
 213 // Must generate all of these ahead of first usage.  Generate at
 214 // compile-time?
 215
 216 // NOT CURRENTLY USED
 217
 218
 219 GLOBL( _x86_choose_fv)
 220         subl    $12, %esp       // gcc does 16 byte alignment of stack frames?
 221         movl    $SUBST(0), (%esp)       // arg 0 - attrib
 222         movl    $SUBST(1), 4(%esp)      // arg 1 - N
 223         call    _do_choose      // new function returned in %eax
 224         add     $12, %esp       // tear down stack frame
 225         jmp     *%eax           // jump to new func
 226 GLOBL ( _x86_choosefv_end )
 227
 228
 229
 230
 231 // FIRST LEVEL FUNCTIONS -- these are plugged directly into GL dispatch.
 232
 233
 234 // NOT CURRENTLY USED
 235
 236
 237
 238 // In the 1st level dispatch functions, switch to a different
 239 // calling convention -- (const GLfloat *v) in %ecx.
 240 //
 241 // As with regular (x86) dispatch, don't create a new stack frame -
 242 // just let the 'ret' in the dispatched function return straight
 243 // back to the original caller.
 244
 245
 246
 247 // Vertex/Normal/Color, etc: the address of the function pointer
 248 // is known at codegen time.
 249
 250
 251 // Unfortunately, have to play with the stack in the non-fv case:
 252 //
 253 GLOBL( _x86_dispatch_attrf )
 254         subl    $12, %esp       // gcc does 16 byte alignment of stack frames?
 255         leal    16(%esp), %edx  // address of first float on stack
 256         movl    %edx, (%esp)    // save as 'v'
 257         call    SUBST(0)        // 0x0 --> tabfv[attr][n]
 258         addl    $12, %esp       // tear down frame
 259         ret                     // return
 260 GLOBL( _x86_dispatch_attrf_end )
 261
 262 // The fv case is simpler:
 263 //
 264 GLOBL( _x86_dispatch_attrfv )
 265         jmp     SUBST(0)        // 0x0 --> tabfv[attr][n]
 266 GLOBL( _x86_dispatch_attrfv_end )
 267
 268
 269 // MultiTexcoord: the address of the function pointer must be
 270 // calculated, but can use the index argument slot to hold 'v', and
 271 // avoid setting up a new stack frame.
 272
 273 // Also, will only need a maximum of four of each of these per context:
 274 //
 275 GLOBL( _x86_dispatch_multitexcoordf )
 276         movl    4(%esp), %ecx
 277         leal    8(%esp), %edx
 278         andl    $7, %ecx
 279         movl    %edx, 4(%esp)
 280         sall    $4, %ecx
 281         jmp     *SUBST(0)(%ecx) // 0x0 - tabfv[tex0][n]
 282 GLOBL( _x86_dispatch_multitexcoordf_end )
 283
 284 GLOBL( _x86_dispatch_multitexcoordfv )
 285         movl    4(%esp), %ecx
 286         movl    8(%esp), %edx
 287         andl    $7, %ecx
 288         movl    %edx, 4(%esp)
 289         sall    $4, %ecx
 290         jmp     *SUBST(0)(%ecx) // 0x0 - tabfv[tex0][n]
 291 GLOBL( _x86_dispatch_multitexcoordfv_end )
 292
 293 // VertexAttrib: the address of the function pointer must be
 294 // calculated.
 295
 296 GLOBL( _x86_dispatch_vertexattribf )
 297         movl    $16, %ecx
 298         movl    4(%esp), %eax
 299         cmpl    $16, %eax
 300         cmovge  %ecx, %eax
 301         leal    8(%esp), %ecx   // calculate 'v'
 302         movl    %ecx, 4(%esp)   // save in 1st arg slot
 303         sall    $4, %eax
 304         jmp     *SUBST(0)(%eax) // 0x0 - tabfv[0][n]
 305 GLOBL( _x86_dispatch_vertexattribf_end )
 306
 307 GLOBL( _x86_dispatch_vertexattribfv )
 308         movl    $16, %ecx
 309         movl    4(%esp), %eax
 310         cmpl    $16, %eax
 311         cmovge  %ecx, %eax
 312         movl    8(%esp), %ecx   // load 'v'
 313         movl    %ecx, 4(%esp)   // save in 1st arg slot
 314         sall    $4, %eax
 315         jmp     *SUBST(0)(%eax) // 0x0 - tabfv[0][n]
 316 GLOBL( _x86_dispatch_vertexattribfv_end )
 317