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 static void get_src_ptr( struct aos_compilation *cp,
  99                          struct x86_reg src,
 100                          struct x86_reg elt,
 101                          unsigned a )
 102 {
 103    struct x86_reg attrib = x86_make_disp(aos_get_x86( cp, 0, X86_ATTRIBS ),
 104                                          a * sizeof(struct aos_attrib));
 105
 106    struct x86_reg input_ptr = x86_make_disp(attrib,
 107                                             Offset(struct aos_attrib, input_ptr));
 108
 109    struct x86_reg input_stride = x86_make_disp(attrib,
 110                                                Offset(struct aos_attrib, input_stride));
 111
 112    /* Calculate pointer to current attrib:
 113     */
 114    x86_mov(cp->func, src, input_stride);
 115    x86_imul(cp->func, src, elt);
 116    x86_add(cp->func, src, input_ptr);
 117 }
 118
 119
 120 /* Extended swizzles?  Maybe later.
 121  */
 122 static void emit_swizzle( struct aos_compilation *cp,
 123                           struct x86_reg dest,
 124                           struct x86_reg src,
 125                           ubyte shuffle )
 126 {
 127    sse_shufps(cp->func, dest, src, shuffle);
 128 }
 129
 130
 131 static boolean load_input( struct aos_compilation *cp,
 132                            unsigned idx,
 133                            boolean linear )
 134 {
 135    unsigned format = cp->vaos->base.key.element[idx].in.format;
 136    struct x86_reg src = cp->tmp_EAX;
 137    struct x86_reg dataXMM = aos_get_xmm_reg(cp);
 138
 139    /* Figure out source pointer address:
 140     */
 141    get_src_ptr(cp,
 142                src,
 143                linear ? cp->idx_EBX : x86_deref(cp->idx_EBX),
 144                idx);
 145
 146    src = x86_deref(src);
 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
 183 boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear )
 184 {
 185    unsigned i;
 186
 187    for (i = 0; i < cp->vaos->base.key.nr_inputs; i++) {
 188       if (!load_input( cp, i, linear ))
 189          return FALSE;
 190       cp->insn_counter++;
 191    }
 192
 193    return TRUE;
 194 }
 195
 196
 197
 198
 199
 200
 201
 202 static void emit_store_R32G32B32A32( struct aos_compilation *cp,
 203                                      struct x86_reg dst_ptr,
 204                                      struct x86_reg dataXMM )
 205 {
 206    sse_movups(cp->func, dst_ptr, dataXMM);
 207 }
 208
 209 static void emit_store_R32G32B32( struct aos_compilation *cp,
 210                                   struct x86_reg dst_ptr,
 211                                   struct x86_reg dataXMM )
 212 {
 213    sse_movlps(cp->func, dst_ptr, dataXMM);
 214    sse_shufps(cp->func, dataXMM, dataXMM, SHUF(Z,Z,Z,Z) ); /* NOTE! destructive */
 215    sse_movss(cp->func, x86_make_disp(dst_ptr,8), dataXMM);
 216 }
 217
 218 static void emit_store_R32G32( struct aos_compilation *cp,
 219                                struct x86_reg dst_ptr,
 220                                struct x86_reg dataXMM )
 221 {
 222    sse_movlps(cp->func, dst_ptr, dataXMM);
 223 }
 224
 225 static void emit_store_R32( struct aos_compilation *cp,
 226                             struct x86_reg dst_ptr,
 227                             struct x86_reg dataXMM )
 228 {
 229    sse_movss(cp->func, dst_ptr, dataXMM);
 230 }
 231
 232
 233
 234 static void emit_store_R8G8B8A8_UNORM( struct aos_compilation *cp,
 235                                        struct x86_reg dst_ptr,
 236                                        struct x86_reg dataXMM )
 237 {
 238    sse_mulps(cp->func, dataXMM, aos_get_internal(cp, IMM_255));
 239    sse2_cvtps2dq(cp->func, dataXMM, dataXMM);
 240    sse2_packssdw(cp->func, dataXMM, dataXMM);
 241    sse2_packuswb(cp->func, dataXMM, dataXMM);
 242    sse_movss(cp->func, dst_ptr, dataXMM);
 243 }
 244
 245
 246
 247
 248
 249 static boolean emit_output( struct aos_compilation *cp,
 250                             struct x86_reg ptr,
 251                             struct x86_reg dataXMM,
 252                             unsigned format )
 253 {
 254    switch (format) {
 255    case EMIT_1F:
 256    case EMIT_1F_PSIZE:
 257       emit_store_R32(cp, ptr, dataXMM);
 258       break;
 259    case EMIT_2F:
 260       emit_store_R32G32(cp, ptr, dataXMM);
 261       break;
 262    case EMIT_3F:
 263       emit_store_R32G32B32(cp, ptr, dataXMM);
 264       break;
 265    case EMIT_4F:
 266       emit_store_R32G32B32A32(cp, ptr, dataXMM);
 267       break;
 268    case EMIT_4UB:
 269       if (1) {
 270          emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W));
 271          emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
 272       }
 273       else {
 274          emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
 275       }
 276       break;
 277    default:
 278       ERROR(cp, "unhandled output format");
 279       return FALSE;
 280    }
 281
 282    return TRUE;
 283 }
 284
 285
 286
 287 boolean aos_emit_outputs( struct aos_compilation *cp )
 288 {
 289    unsigned i;
 290
 291    for (i = 0; i < cp->vaos->base.key.nr_outputs; i++) {
 292       unsigned format = cp->vaos->base.key.element[i].out.format;
 293       unsigned offset = cp->vaos->base.key.element[i].out.offset;
 294       unsigned vs_output = cp->vaos->base.key.element[i].out.vs_output;
 295
 296       struct x86_reg data;
 297
 298       if (format == EMIT_1F_PSIZE) {
 299          data = aos_get_internal_xmm( cp, IMM_PSIZE );
 300       }
 301       else {
 302          data = aos_get_shader_reg( cp,
 303                                     TGSI_FILE_OUTPUT,
 304                                     vs_output );
 305       }
 306
 307       if (data.file != file_XMM) {
 308          struct x86_reg tmp = aos_get_xmm_reg( cp );
 309          sse_movups(cp->func, tmp, data);
 310          data = tmp;
 311       }
 312
 313       if (!emit_output( cp,
 314                         x86_make_disp( cp->outbuf_ECX, offset ),
 315                         data,
 316                         format ))
 317          return FALSE;
 318
 319       aos_release_xmm_reg( cp, data.idx );
 320
 321       cp->insn_counter++;
 322    }
 323
 324    return TRUE;
 325 }
 326
 327 #endif