src/gallium/drivers/softpipe/sp_fs_sse.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  * Execute fragment shader using runtime SSE code generation.
  30  */
  31
  32 #include "sp_context.h"
  33 #include "sp_state.h"
  34 #include "sp_fs.h"
  35 #include "sp_quad.h"
  36
  37 #include "pipe/p_state.h"
  38 #include "pipe/p_defines.h"
  39 #include "util/u_memory.h"
  40 #include "tgsi/tgsi_exec.h"
  41 #include "tgsi/tgsi_sse2.h"
  42
  43
  44 #if defined(PIPE_ARCH_X86)
  45
  46 #include "rtasm/rtasm_x86sse.h"
  47
  48
  49
  50 /**
  51  * Subclass of sp_fragment_shader
  52  */
  53 struct sp_sse_fragment_shader
  54 {
  55    struct sp_fragment_shader base;
  56    struct x86_function sse2_program;
  57    tgsi_sse2_fs_function func;
  58    float immediates[TGSI_EXEC_NUM_IMMEDIATES][4];
  59 };
  60
  61
  62 /** cast wrapper */
  63 static INLINE struct sp_sse_fragment_shader *
  64 sp_sse_fragment_shader(const struct sp_fragment_shader *base)
  65 {
  66    return (struct sp_sse_fragment_shader *) base;
  67 }
  68
  69
  70 static void
  71 fs_sse_prepare( const struct sp_fragment_shader *base,
  72                 struct tgsi_exec_machine *machine,
  73                 struct tgsi_sampler **samplers )
  74 {
  75    machine->Samplers = samplers;
  76 }
  77
  78
  79
  80 /**
  81  * Compute quad X,Y,Z,W for the four fragments in a quad.
  82  *
  83  * This should really be part of the compiled shader.
  84  */
  85 static void
  86 setup_pos_vector(const struct tgsi_interp_coef *coef,
  87                     float x, float y,
  88                     struct tgsi_exec_vector *quadpos)
  89 {
  90    uint chan;
  91    /* do X */
  92    quadpos->xyzw[0].f[0] = x;
  93    quadpos->xyzw[0].f[1] = x + 1;
  94    quadpos->xyzw[0].f[2] = x;
  95    quadpos->xyzw[0].f[3] = x + 1;
  96
  97    /* do Y */
  98    quadpos->xyzw[1].f[0] = y;
  99    quadpos->xyzw[1].f[1] = y;
 100    quadpos->xyzw[1].f[2] = y + 1;
 101    quadpos->xyzw[1].f[3] = y + 1;
 102
 103    /* do Z and W for all fragments in the quad */
 104    for (chan = 2; chan < 4; chan++) {
 105       const float dadx = coef->dadx[chan];
 106       const float dady = coef->dady[chan];
 107       const float a0 = coef->a0[chan] + dadx * x + dady * y;
 108       quadpos->xyzw[chan].f[0] = a0;
 109       quadpos->xyzw[chan].f[1] = a0 + dadx;
 110       quadpos->xyzw[chan].f[2] = a0 + dady;
 111       quadpos->xyzw[chan].f[3] = a0 + dadx + dady;
 112    }
 113 }
 114
 115
 116 /* TODO: codegenerate the whole run function, skip this wrapper.
 117  * TODO: break dependency on tgsi_exec_machine struct
 118  * TODO: push Position calculation into the generated shader
 119  * TODO: process >1 quad at a time
 120  */
 121 static unsigned
 122 fs_sse_run( const struct sp_fragment_shader *base,
 123             struct tgsi_exec_machine *machine,
 124             struct quad_header *quad )
 125 {
 126    struct sp_sse_fragment_shader *shader = sp_sse_fragment_shader(base);
 127
 128    /* Compute X, Y, Z, W vals for this quad -- place in temp[0] for now */
 129    setup_pos_vector(quad->posCoef,
 130                     (float)quad->input.x0, (float)quad->input.y0,
 131                     machine->Temps);
 132
 133    /* init kill mask */
 134    tgsi_set_kill_mask(machine, 0x0);
 135    tgsi_set_exec_mask(machine, 1, 1, 1, 1);
 136
 137    shader->func( machine,
 138                  (const float (*)[4])machine->Consts[0],
 139                  (const float (*)[4])shader->immediates,
 140                  machine->InterpCoefs
 141                  /*, &machine->QuadPos*/
 142       );
 143
 144    quad->inout.mask &= ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]);
 145    if (quad->inout.mask == 0)
 146       return FALSE;
 147
 148    /* store outputs */
 149    {
 150       const ubyte *sem_name = base->info.output_semantic_name;
 151       const ubyte *sem_index = base->info.output_semantic_index;
 152       const uint n = base->info.num_outputs;
 153       uint i;
 154       for (i = 0; i < n; i++) {
 155          switch (sem_name[i]) {
 156          case TGSI_SEMANTIC_COLOR:
 157             {
 158                uint cbuf = sem_index[i];
 159
 160                assert(sizeof(quad->output.color[cbuf]) ==
 161                       sizeof(machine->Outputs[i]));
 162
 163                /* copy float[4][4] result */
 164                memcpy(quad->output.color[cbuf],
 165                       &machine->Outputs[i],
 166                       sizeof(quad->output.color[0]) );
 167             }
 168             break;
 169          case TGSI_SEMANTIC_POSITION:
 170             {
 171                uint j;
 172                for (j = 0; j < 4; j++)
 173                   quad->output.depth[j] = machine->Outputs[i].xyzw[2].f[j];
 174             }
 175             break;
 176          case TGSI_SEMANTIC_STENCIL:
 177             {
 178                uint j;
 179                for (j = 0; j < 4; j++)
 180                   quad->output.stencil[j] = machine->Outputs[i].xyzw[1].f[j];
 181             }
 182             break;
 183          }
 184       }
 185    }
 186
 187    return TRUE;
 188 }
 189
 190
 191 static void
 192 fs_sse_delete( struct sp_fragment_shader *base )
 193 {
 194    struct sp_sse_fragment_shader *shader = sp_sse_fragment_shader(base);
 195
 196    x86_release_func( &shader->sse2_program );
 197    FREE(shader);
 198 }
 199
 200
 201 struct sp_fragment_shader *
 202 softpipe_create_fs_sse(struct softpipe_context *softpipe,
 203                        const struct pipe_shader_state *templ)
 204 {
 205    struct sp_sse_fragment_shader *shader;
 206
 207    if (!softpipe->use_sse)
 208       return NULL;
 209
 210    shader = CALLOC_STRUCT(sp_sse_fragment_shader);
 211    if (!shader)
 212       return NULL;
 213
 214    x86_init_func( &shader->sse2_program );
 215
 216    if (!tgsi_emit_sse2( templ->tokens, &shader->sse2_program,
 217                         shader->immediates, FALSE )) {
 218       FREE(shader);
 219       return NULL;
 220    }
 221
 222    shader->func = (tgsi_sse2_fs_function) x86_get_func( &shader->sse2_program );
 223    if (!shader->func) {
 224       x86_release_func( &shader->sse2_program );
 225       FREE(shader);
 226       return NULL;
 227    }
 228
 229    shader->base.shader.tokens = NULL; /* don't hold reference to templ->tokens */
 230    shader->base.prepare = fs_sse_prepare;
 231    shader->base.run = fs_sse_run;
 232    shader->base.delete = fs_sse_delete;
 233
 234    return &shader->base;
 235 }
 236
 237
 238 #else
 239
 240 /* Maybe put this variant in the header file.
 241  */
 242 struct sp_fragment_shader *
 243 softpipe_create_fs_sse(struct softpipe_context *softpipe,
 244                        const struct pipe_shader_state *templ)
 245 {
 246    return NULL;
 247 }
 248
 249 #endif