src/gallium/auxiliary/draw/draw_vs_aos_io.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28
  29 #include "util/u_memory.h"
  30 #include "pipe/p_shader_tokens.h"
  31 #include "tgsi/tgsi_parse.h"
  32 #include "tgsi/tgsi_util.h"
  33 #include "tgsi/tgsi_exec.h"
  34 #include "draw_vs.h"
  35 #include "draw_vs_aos.h"
  36 #include "draw_vertex.h"
  37
  38 #include "rtasm/rtasm_x86sse.h"
  39
  40 #ifdef PIPE_ARCH_X86
  41
  42 /* Note - don't yet have to worry about interacting with the code in
  43  * draw_vs_aos.c as there is no intermingling of generated code...
  44  * That may have to change, we'll see.
  45  */
  46 static void emit_load_R32G32B32A32( struct aos_compilation *cp,
  47                                     struct x86_reg data,
  48                                     struct x86_reg src_ptr )
  49 {
  50    sse_movups(cp->func, data, src_ptr);
  51 }
  52
  53 static void emit_load_R32G32B32( struct aos_compilation *cp,
  54                                  struct x86_reg data,
  55                                  struct x86_reg src_ptr )
  56 {
  57    sse_movss(cp->func, data, x86_make_disp(src_ptr, 8));
  58    /* data = z ? ? ? */
  59    sse_shufps(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ), SHUF(X,Y,Z,W) );
  60    /* data = z ? 0 1 */
  61    sse_shufps(cp->func, data, data, SHUF(Y,Z,X,W) );
  62    /* data = ? 0 z 1 */
  63    sse_movlps(cp->func, data, src_ptr);
  64    /* data = x y z 1 */
  65 }
  66
  67 static void emit_load_R32G32( struct aos_compilation *cp,
  68                            struct x86_reg data,
  69                            struct x86_reg src_ptr )
  70 {
  71    sse_movups(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ) );
  72    sse_movlps(cp->func, data, src_ptr);
  73 }
  74
  75
  76 static void emit_load_R32( struct aos_compilation *cp,
  77                            struct x86_reg data,
  78                            struct x86_reg src_ptr )
  79 {
  80    sse_movss(cp->func, data, src_ptr);
  81    sse_orps(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ) );
  82 }
  83
  84
  85 static void emit_load_R8G8B8A8_UNORM( struct aos_compilation *cp,
  86                                        struct x86_reg data,
  87                                        struct x86_reg src_ptr )
  88 {
  89    sse_movss(cp->func, data, src_ptr);
  90    sse2_punpcklbw(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ));
  91    sse2_punpcklbw(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ));
  92    sse2_cvtdq2ps(cp->func, data, data);
  93    sse_mulps(cp->func, data, aos_get_internal(cp, IMM_INV_255));
  94 }
  95
  96
  97
  98 /* Extended swizzles?  Maybe later.
  99  */
 100 static void emit_swizzle( struct aos_compilation *cp,
 101                           struct x86_reg dest,
 102                           struct x86_reg src,
 103                           ubyte shuffle )
 104 {
 105    sse_shufps(cp->func, dest, src, shuffle);
 106 }
 107
 108
 109
 110 static boolean get_buffer_ptr( struct aos_compilation *cp,
 111                             unsigned buf_idx,
 112                             struct x86_reg elt,
 113                             struct x86_reg ptr)
 114 {
 115    struct x86_reg buf = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ),
 116                                       buf_idx * sizeof(struct aos_buffer));
 117
 118    struct x86_reg buf_base_ptr = x86_make_disp(buf,
 119                                                Offset(struct aos_buffer, base_ptr));
 120
 121    struct x86_reg buf_stride = x86_make_disp(buf,
 122                                              Offset(struct aos_buffer, stride));
 123
 124    /* Calculate pointer to current attrib:
 125     */
 126    x86_mov(cp->func, ptr, buf_stride);
 127    x86_imul(cp->func, ptr, elt);
 128    x86_add(cp->func, ptr, buf_base_ptr);
 129
 130    return TRUE;
 131 }
 132
 133
 134
 135
 136 static boolean load_input( struct aos_compilation *cp,
 137                            unsigned idx,
 138                            struct x86_reg bufptr )
 139 {
 140    unsigned format = cp->vaos->base.key.element[idx].in.format;
 141    unsigned offset = cp->vaos->base.key.element[idx].in.offset;
 142    struct x86_reg dataXMM = aos_get_xmm_reg(cp);
 143
 144    /* Figure out source pointer address:
 145     */
 146    struct x86_reg src = x86_make_disp(bufptr, offset);
 147
 148    aos_adopt_xmm_reg( cp,
 149                       dataXMM,
 150                       TGSI_FILE_INPUT,
 151                       idx,
 152                       TRUE );
 153
 154    switch (format) {
 155    case PIPE_FORMAT_R32_FLOAT:
 156       emit_load_R32(cp, dataXMM, src);
 157       break;
 158    case PIPE_FORMAT_R32G32_FLOAT:
 159       emit_load_R32G32(cp, dataXMM, src);
 160       break;
 161    case PIPE_FORMAT_R32G32B32_FLOAT:
 162       emit_load_R32G32B32(cp, dataXMM, src);
 163       break;
 164    case PIPE_FORMAT_R32G32B32A32_FLOAT:
 165       emit_load_R32G32B32A32(cp, dataXMM, src);
 166       break;
 167    case PIPE_FORMAT_B8G8R8A8_UNORM:
 168       emit_load_R8G8B8A8_UNORM(cp, dataXMM, src);
 169       emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W));
 170       break;
 171    case PIPE_FORMAT_R8G8B8A8_UNORM:
 172       emit_load_R8G8B8A8_UNORM(cp, dataXMM, src);
 173       break;
 174    default:
 175       ERROR(cp, "unhandled input format");
 176       return FALSE;
 177    }
 178
 179    return TRUE;
 180 }
 181
 182 static boolean load_inputs( struct aos_compilation *cp,
 183                             unsigned buffer,
 184                             struct x86_reg ptr )
 185 {
 186    unsigned i;
 187
 188    for (i = 0; i < cp->vaos->base.key.nr_inputs; i++) {
 189       if (cp->vaos->base.key.element[i].in.buffer == buffer) {
 190
 191          if (!load_input( cp, i, ptr ))
 192             return FALSE;
 193
 194          cp->insn_counter++;
 195       }
 196    }
 197
 198    return TRUE;
 199 }
 200
 201 boolean aos_init_inputs( struct aos_compilation *cp, boolean linear )
 202 {
 203    if (linear && cp->vaos->nr_vb == 1) {
 204
 205       struct x86_reg elt = cp->idx_EBX;
 206       struct x86_reg ptr = cp->tmp_EAX;
 207
 208       if (!get_buffer_ptr( cp, 0, elt, ptr ))
 209          return FALSE;
 210
 211       /* In the linear, single buffer case, keep the buffer pointer
 212        * instead of the index number.
 213        */
 214       x86_mov( cp->func, elt, ptr );
 215    }
 216
 217    return TRUE;
 218 }
 219
 220 boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear )
 221 {
 222    if (linear && cp->vaos->nr_vb == 1) {
 223
 224       load_inputs( cp, 0, cp->idx_EBX );
 225
 226    }
 227    else {
 228       struct x86_reg elt = linear ? cp->idx_EBX : x86_deref(cp->idx_EBX);
 229       unsigned j;
 230
 231       for (j = 0; j < cp->vaos->nr_vb; j++) {
 232          struct x86_reg ptr = cp->tmp_EAX;
 233
 234          if (!get_buffer_ptr( cp, j, elt, ptr ))
 235             return FALSE;
 236
 237          cp->insn_counter++;
 238
 239          if (!load_inputs( cp, j, ptr ))
 240             return FALSE;
 241       }
 242    }
 243
 244    return TRUE;
 245 }
 246
 247 boolean aos_incr_inputs( struct aos_compilation *cp, boolean linear )
 248 {
 249    if (linear && cp->vaos->nr_vb == 1) {
 250       struct x86_reg stride = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ),
 251                                             (0 * sizeof(struct aos_buffer) +
 252                                              Offset(struct aos_buffer, stride)));
 253
 254       x86_add(cp->func, cp->idx_EBX, stride);
 255    }
 256    else if (linear) {
 257       x86_inc(cp->func, cp->idx_EBX);
 258    }
 259    else {
 260       x86_lea(cp->func, cp->idx_EBX, x86_make_disp(cp->idx_EBX, 4));
 261    }
 262 }
 263
 264
 265
 266
 267
 268
 269 static void emit_store_R32G32B32A32( struct aos_compilation *cp,
 270                                      struct x86_reg dst_ptr,
 271                                      struct x86_reg dataXMM )
 272 {
 273    sse_movups(cp->func, dst_ptr, dataXMM);
 274 }
 275
 276 static void emit_store_R32G32B32( struct aos_compilation *cp,
 277                                   struct x86_reg dst_ptr,
 278                                   struct x86_reg dataXMM )
 279 {
 280    sse_movlps(cp->func, dst_ptr, dataXMM);
 281    sse_shufps(cp->func, dataXMM, dataXMM, SHUF(Z,Z,Z,Z) ); /* NOTE! destructive */
 282    sse_movss(cp->func, x86_make_disp(dst_ptr,8), dataXMM);
 283 }
 284
 285 static void emit_store_R32G32( struct aos_compilation *cp,
 286                                struct x86_reg dst_ptr,
 287                                struct x86_reg dataXMM )
 288 {
 289    sse_movlps(cp->func, dst_ptr, dataXMM);
 290 }
 291
 292 static void emit_store_R32( struct aos_compilation *cp,
 293                             struct x86_reg dst_ptr,
 294                             struct x86_reg dataXMM )
 295 {
 296    sse_movss(cp->func, dst_ptr, dataXMM);
 297 }
 298
 299
 300
 301 static void emit_store_R8G8B8A8_UNORM( struct aos_compilation *cp,
 302                                        struct x86_reg dst_ptr,
 303                                        struct x86_reg dataXMM )
 304 {
 305    sse_mulps(cp->func, dataXMM, aos_get_internal(cp, IMM_255));
 306    sse2_cvtps2dq(cp->func, dataXMM, dataXMM);
 307    sse2_packssdw(cp->func, dataXMM, dataXMM);
 308    sse2_packuswb(cp->func, dataXMM, dataXMM);
 309    sse_movss(cp->func, dst_ptr, dataXMM);
 310 }
 311
 312
 313
 314
 315
 316 static boolean emit_output( struct aos_compilation *cp,
 317                             struct x86_reg ptr,
 318                             struct x86_reg dataXMM,
 319                             unsigned format )
 320 {
 321    switch (format) {
 322    case EMIT_1F:
 323    case EMIT_1F_PSIZE:
 324       emit_store_R32(cp, ptr, dataXMM);
 325       break;
 326    case EMIT_2F:
 327       emit_store_R32G32(cp, ptr, dataXMM);
 328       break;
 329    case EMIT_3F:
 330       emit_store_R32G32B32(cp, ptr, dataXMM);
 331       break;
 332    case EMIT_4F:
 333       emit_store_R32G32B32A32(cp, ptr, dataXMM);
 334       break;
 335    case EMIT_4UB:
 336       if (1) {
 337          emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W));
 338          emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
 339       }
 340       else {
 341          emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
 342       }
 343       break;
 344    default:
 345       ERROR(cp, "unhandled output format");
 346       return FALSE;
 347    }
 348
 349    return TRUE;
 350 }
 351
 352
 353
 354 boolean aos_emit_outputs( struct aos_compilation *cp )
 355 {
 356    unsigned i;
 357
 358    for (i = 0; i < cp->vaos->base.key.nr_outputs; i++) {
 359       unsigned format = cp->vaos->base.key.element[i].out.format;
 360       unsigned offset = cp->vaos->base.key.element[i].out.offset;
 361       unsigned vs_output = cp->vaos->base.key.element[i].out.vs_output;
 362
 363       struct x86_reg data;
 364
 365       if (format == EMIT_1F_PSIZE) {
 366          data = aos_get_internal_xmm( cp, IMM_PSIZE );
 367       }
 368       else {
 369          data = aos_get_shader_reg( cp,
 370                                     TGSI_FILE_OUTPUT,
 371                                     vs_output );
 372       }
 373
 374       if (data.file != file_XMM) {
 375          struct x86_reg tmp = aos_get_xmm_reg( cp );
 376          sse_movups(cp->func, tmp, data);
 377          data = tmp;
 378       }
 379
 380       if (!emit_output( cp,
 381                         x86_make_disp( cp->outbuf_ECX, offset ),
 382                         data,
 383                         format ))
 384          return FALSE;
 385
 386       aos_release_xmm_reg( cp, data.idx );
 387
 388       cp->insn_counter++;
 389    }
 390
 391    return TRUE;
 392 }
 393
 394 #endif