src/gallium/state_trackers/vega/asm_fill.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 VMware, Inc.  All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the
   7  * "Software"), to deal in the Software without restriction, including
   8  * without limitation the rights to use, copy, modify, merge, publish,
   9  * distribute, sub license, and/or sell copies of the Software, and to
  10  * permit persons to whom the Software is furnished to do so, subject to
  11  * the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the
  14  * next paragraph) shall be included in all copies or substantial portions
  15  * of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  19  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  20  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  21  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  22  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  23  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  24  *
  25  **************************************************************************/
  26
  27 #ifndef ASM_FILL_H
  28 #define ASM_FILL_H
  29
  30 static const char solid_fill_asm[] =
  31    "MOV %s, CONST[0]\n";
  32
  33
  34 static const char linear_grad_asm[] =
  35    "MOV TEMP[0].xy, IN[0]\n"
  36    "MOV TEMP[0].z, CONST[1].yyyy\n"
  37    "DP3 TEMP[1], CONST[2], TEMP[0]\n"
  38    "DP3 TEMP[2], CONST[3], TEMP[0]\n"
  39    "DP3 TEMP[3], CONST[4], TEMP[0]\n"
  40    "RCP TEMP[3], TEMP[3]\n"
  41    "MUL TEMP[1], TEMP[1], TEMP[3]\n"
  42    "MUL TEMP[2], TEMP[2], TEMP[3]\n"
  43    "MOV TEMP[4].x, TEMP[1]\n"
  44    "MOV TEMP[4].y, TEMP[2]\n"
  45    "MUL TEMP[0], CONST[0].yyyy, TEMP[4].yyyy\n"
  46    "MAD TEMP[1], CONST[0].xxxx, TEMP[4].xxxx, TEMP[0]\n"
  47    "MUL TEMP[2], TEMP[1], CONST[0].zzzz\n"
  48    "TEX %s, TEMP[2], SAMP[0], 1D\n";
  49
  50 static const char radial_grad_asm[] =
  51    "MOV TEMP[0].xy, IN[0]\n"
  52    "MOV TEMP[0].z, CONST[1].yyyy\n"
  53    "DP3 TEMP[1], CONST[2], TEMP[0]\n"
  54    "DP3 TEMP[2], CONST[3], TEMP[0]\n"
  55    "DP3 TEMP[3], CONST[4], TEMP[0]\n"
  56    "RCP TEMP[3], TEMP[3]\n"
  57    "MUL TEMP[1], TEMP[1], TEMP[3]\n"
  58    "MUL TEMP[2], TEMP[2], TEMP[3]\n"
  59    "MOV TEMP[5].x, TEMP[1]\n"
  60    "MOV TEMP[5].y, TEMP[2]\n"
  61    "MUL TEMP[0], CONST[0].yyyy, TEMP[5].yyyy\n"
  62    "MAD TEMP[1], CONST[0].xxxx, TEMP[5].xxxx, TEMP[0]\n"
  63    "ADD TEMP[1], TEMP[1], TEMP[1]\n"
  64    "MUL TEMP[3], TEMP[5].yyyy, TEMP[5].yyyy\n"
  65    "MAD TEMP[4], TEMP[5].xxxx, TEMP[5].xxxx, TEMP[3]\n"
  66    "MOV TEMP[4], -TEMP[4]\n"
  67    "MUL TEMP[2], CONST[0].zzzz, TEMP[4]\n"
  68    "MUL TEMP[0], CONST[1].wwww, TEMP[2]\n"
  69    "MUL TEMP[3], TEMP[1], TEMP[1]\n"
  70    "SUB TEMP[2], TEMP[3], TEMP[0]\n"
  71    "RSQ TEMP[2], |TEMP[2]|\n"
  72    "RCP TEMP[2], TEMP[2]\n"
  73    "SUB TEMP[1], TEMP[2], TEMP[1]\n"
  74    "ADD TEMP[0], CONST[0].zzzz, CONST[0].zzzz\n"
  75    "RCP TEMP[0], TEMP[0]\n"
  76    "MUL TEMP[2], TEMP[1], TEMP[0]\n"
  77    "TEX %s, TEMP[2], SAMP[0], 1D\n";
  78
  79 static const char pattern_asm[] =
  80    "MOV TEMP[0].xy, IN[0]\n"
  81    "MOV TEMP[0].z, CONST[1].yyyy\n"
  82    "DP3 TEMP[1], CONST[2], TEMP[0]\n"
  83    "DP3 TEMP[2], CONST[3], TEMP[0]\n"
  84    "DP3 TEMP[3], CONST[4], TEMP[0]\n"
  85    "RCP TEMP[3], TEMP[3]\n"
  86    "MUL TEMP[1], TEMP[1], TEMP[3]\n"
  87    "MUL TEMP[2], TEMP[2], TEMP[3]\n"
  88    "MOV TEMP[4].x, TEMP[1]\n"
  89    "MOV TEMP[4].y, TEMP[2]\n"
  90    "RCP TEMP[0], CONST[1].zwzw\n"
  91    "MOV TEMP[1], TEMP[4]\n"
  92    "MUL TEMP[1].x, TEMP[1], TEMP[0]\n"
  93    "MUL TEMP[1].y, TEMP[1], TEMP[0]\n"
  94    "TEX %s, TEMP[1], SAMP[0], 2D\n";
  95
  96
  97 static const char mask_asm[] =
  98    "TEX TEMP[1], IN[0], SAMP[1], 2D\n"
  99    "MUL TEMP[0].w, TEMP[0].wwww, TEMP[1].wwww\n"
 100    "MOV %s, TEMP[0]\n";
 101
 102
 103 static const char image_normal_asm[] =
 104    "TEX %s, IN[1], SAMP[3], 2D\n";
 105
 106 static const char image_multiply_asm[] =
 107    "TEX TEMP[1], IN[1], SAMP[3], 2D\n"
 108    "MUL %s, TEMP[0], TEMP[1]\n";
 109
 110 static const char image_stencil_asm[] =
 111    "TEX TEMP[1], IN[1], SAMP[3], 2D\n"
 112    "MUL %s, TEMP[0], TEMP[1]\n";
 113
 114
 115 #define EXTENDED_BLEND_OVER                     \
 116    "SUB TEMP[3], CONST[1].yyyy, TEMP[1].wwww\n" \
 117    "SUB TEMP[4], CONST[1].yyyy, TEMP[0].wwww\n" \
 118    "MUL TEMP[3], TEMP[0], TEMP[3]\n"            \
 119    "MUL TEMP[4], TEMP[1], TEMP[4]\n"            \
 120    "ADD TEMP[3], TEMP[3], TEMP[4]\n"
 121
 122 static const char blend_multiply_asm[] =
 123    "TEX TEMP[1], IN[0], SAMP[2], 2D\n"
 124    EXTENDED_BLEND_OVER
 125    "MUL TEMP[4], TEMP[0], TEMP[1]\n"
 126    "ADD TEMP[1], TEMP[4], TEMP[3]\n"/*result.rgb*/
 127    "MUL TEMP[2], TEMP[0].wwww, TEMP[1].wwww\n"
 128    "ADD TEMP[3], TEMP[0].wwww, TEMP[1].wwww\n"
 129    "SUB TEMP[1].w, TEMP[3], TEMP[2]\n"
 130    "MOV %s, TEMP[1]\n";
 131 #if 1
 132 static const char blend_screen_asm[] =
 133    "TEX TEMP[1], IN[0], SAMP[2], 2D\n"
 134    "ADD TEMP[3], TEMP[0], TEMP[1]\n"
 135    "MUL TEMP[2], TEMP[0], TEMP[1]\n"
 136    "SUB %s, TEMP[3], TEMP[2]\n";
 137 #else
 138 static const char blend_screen_asm[] =
 139    "TEX TEMP[1], IN[0], SAMP[2], 2D\n"
 140    "MOV %s, TEMP[1]\n";
 141 #endif
 142
 143 static const char blend_darken_asm[] =
 144    "TEX TEMP[1], IN[0], SAMP[2], 2D\n"
 145    EXTENDED_BLEND_OVER
 146    "MUL TEMP[4], TEMP[0], TEMP[1].wwww\n"
 147    "MUL TEMP[5], TEMP[1], TEMP[0].wwww\n"
 148    "MIN TEMP[4], TEMP[4], TEMP[5]\n"
 149    "ADD TEMP[1], TEMP[3], TEMP[4]\n"
 150    "MUL TEMP[2], TEMP[0].wwww, TEMP[1].wwww\n"
 151    "ADD TEMP[3], TEMP[0].wwww, TEMP[1].wwww\n"
 152    "SUB TEMP[1].w, TEMP[3], TEMP[2]\n"
 153    "MOV %s, TEMP[1]\n";
 154
 155 static const char blend_lighten_asm[] =
 156    "TEX TEMP[1], IN[0], SAMP[2], 2D\n"
 157    EXTENDED_BLEND_OVER
 158    "MUL TEMP[4], TEMP[0], TEMP[1].wwww\n"
 159    "MUL TEMP[5], TEMP[1], TEMP[0].wwww\n"
 160    "MAX TEMP[4], TEMP[4], TEMP[5]\n"
 161    "ADD TEMP[1], TEMP[3], TEMP[4]\n"
 162    "MUL TEMP[2], TEMP[0].wwww, TEMP[1].wwww\n"
 163    "ADD TEMP[3], TEMP[0].wwww, TEMP[1].wwww\n"
 164    "SUB TEMP[1].w, TEMP[3], TEMP[2]\n"
 165    "MOV %s, TEMP[1]\n";
 166
 167
 168 static const char premultiply_asm[] =
 169    "MUL TEMP[0].xyz, TEMP[0], TEMP[0].wwww\n";
 170
 171 static const char unpremultiply_asm[] =
 172    "TEX TEMP[0], IN[0], SAMP[1], 2D\n";
 173
 174
 175 static const char color_bw_asm[] =
 176    "ADD TEMP[1], CONST[1].yyyy, CONST[1].yyyy\n"
 177    "RCP TEMP[2], TEMP[1]\n"
 178    "ADD TEMP[1], CONST[1].yyyy, TEMP[2]\n"
 179    "ADD TEMP[2].x, TEMP[0].xxxx, TEMP[0].yyyy\n"
 180    "ADD TEMP[2].x, TEMP[0].zzzz, TEMP[0].xxxx\n"
 181    "SGE TEMP[0].xyz, TEMP[2].xxxx, TEMP[1]\n"
 182    "SGE TEMP[0].w, TEMP[0].wwww, TEMP[2].yyyy\n"
 183    "MOV %s, TEMP[0]\n";
 184
 185
 186 struct shader_asm_info {
 187    VGint id;
 188    VGint num_tokens;
 189    const char * txt;
 190
 191    VGboolean needs_position;
 192
 193    VGint start_const;
 194    VGint num_consts;
 195
 196    VGint start_sampler;
 197    VGint num_samplers;
 198
 199    VGint start_temp;
 200    VGint num_temps;
 201 };
 202
 203
 204 static const struct shader_asm_info shaders_asm[] = {
 205    /* fills */
 206    {VEGA_SOLID_FILL_SHADER,       40,  solid_fill_asm,
 207     VG_FALSE, 0, 1, 0, 0, 0, 0},
 208    {VEGA_LINEAR_GRADIENT_SHADER, 200,  linear_grad_asm,
 209     VG_TRUE,  0, 5, 0, 1, 0, 5},
 210    {VEGA_RADIAL_GRADIENT_SHADER, 200,  radial_grad_asm,
 211     VG_TRUE,  0, 5, 0, 1, 0, 6},
 212    {VEGA_PATTERN_SHADER,         100,      pattern_asm,
 213     VG_TRUE,  1, 4, 0, 1, 0, 5},
 214
 215    /* image draw modes */
 216    {VEGA_IMAGE_NORMAL_SHADER,    200, image_normal_asm,
 217     VG_TRUE,  0, 0, 3, 1, 0, 0},
 218    {VEGA_IMAGE_MULTIPLY_SHADER,  200, image_multiply_asm,
 219     VG_TRUE,  0, 0, 3, 1, 0, 2},
 220    {VEGA_IMAGE_STENCIL_SHADER,   200, image_stencil_asm,
 221     VG_TRUE,  0, 0, 3, 1, 0, 2},
 222
 223    {VEGA_MASK_SHADER,            100,         mask_asm,
 224     VG_TRUE,  0, 0, 1, 1, 0, 2},
 225
 226    /* extra blend modes */
 227    {VEGA_BLEND_MULTIPLY_SHADER,  200, blend_multiply_asm,
 228     VG_TRUE,  1, 1, 2, 1, 0, 5},
 229    {VEGA_BLEND_SCREEN_SHADER,    200, blend_screen_asm,
 230     VG_TRUE,  0, 0, 2, 1, 0, 4},
 231    {VEGA_BLEND_DARKEN_SHADER,    200, blend_darken_asm,
 232     VG_TRUE,  1, 1, 2, 1, 0, 6},
 233    {VEGA_BLEND_LIGHTEN_SHADER,   200, blend_lighten_asm,
 234     VG_TRUE,  1, 1, 2, 1, 0, 6},
 235
 236    /* premultiply */
 237    {VEGA_PREMULTIPLY_SHADER,   100, premultiply_asm,
 238     VG_FALSE,  0, 0, 0, 0, 0, 1},
 239    {VEGA_UNPREMULTIPLY_SHADER,   100, unpremultiply_asm,
 240     VG_FALSE,  0, 0, 0, 0, 0, 1},
 241
 242    /* color transform to black and white */
 243    {VEGA_BW_SHADER,   150, color_bw_asm,
 244     VG_FALSE,  1, 1, 0, 0, 0, 3},
 245 };
 246 #endif