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 /* TODO: codegenerate the whole run function, skip this wrapper.
  80  * TODO: break dependency on tgsi_exec_machine struct
  81  * TODO: push Position calculation into the generated shader
  82  * TODO: process >1 quad at a time
  83  */
  84 static unsigned
  85 fs_sse_run( const struct sp_fragment_shader *base,
  86             struct tgsi_exec_machine *machine,
  87             struct quad_header *quad )
  88 {
  89    struct sp_sse_fragment_shader *shader = sp_sse_fragment_shader(base);
  90
  91    /* Compute X, Y, Z, W vals for this quad -- place in temp[0] for now */
  92    sp_setup_pos_vector(quad->posCoef,
  93                        (float)quad->input.x0, (float)quad->input.y0,
  94                        machine->Temps);
  95
  96    /* init kill mask */
  97    tgsi_set_kill_mask(machine, 0x0);
  98    tgsi_set_exec_mask(machine, 1, 1, 1, 1);
  99
 100    shader->func( machine,
 101                  machine->Consts,
 102                  (const float (*)[4])shader->immediates,
 103                  machine->InterpCoefs
 104                  //      , &machine->QuadPos
 105       );
 106
 107    quad->inout.mask &= ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]);
 108    if (quad->inout.mask == 0)
 109       return FALSE;
 110
 111
 112    /* store outputs */
 113    {
 114       const ubyte *sem_name = shader->base.info.output_semantic_name;
 115       const ubyte *sem_index = shader->base.info.output_semantic_index;
 116       const uint n = shader->base.info.num_outputs;
 117       uint i;
 118       for (i = 0; i < n; i++) {
 119          switch (sem_name[i]) {
 120          case TGSI_SEMANTIC_COLOR:
 121             {
 122                uint cbuf = sem_index[i];
 123                memcpy(quad->output.color[cbuf],
 124                       &machine->Outputs[i].xyzw[0].f[0],
 125                       sizeof(quad->output.color[0]) );
 126             }
 127             break;
 128          case TGSI_SEMANTIC_POSITION:
 129             {
 130                uint j;
 131                for (j = 0; j < 4; j++) {
 132                   quad->output.depth[j] = machine->Outputs[0].xyzw[2].f[j];
 133                }
 134             }
 135             break;
 136          }
 137       }
 138    }
 139
 140    return TRUE;
 141 }
 142
 143
 144 static void
 145 fs_sse_delete( struct sp_fragment_shader *base )
 146 {
 147    struct sp_sse_fragment_shader *shader = sp_sse_fragment_shader(base);
 148
 149    x86_release_func( &shader->sse2_program );
 150    FREE(shader);
 151 }
 152
 153
 154 struct sp_fragment_shader *
 155 softpipe_create_fs_sse(struct softpipe_context *softpipe,
 156                        const struct pipe_shader_state *templ)
 157 {
 158    struct sp_sse_fragment_shader *shader;
 159
 160    if (!softpipe->use_sse)
 161       return NULL;
 162
 163    shader = CALLOC_STRUCT(sp_sse_fragment_shader);
 164    if (!shader)
 165       return NULL;
 166
 167    x86_init_func( &shader->sse2_program );
 168
 169    if (!tgsi_emit_sse2( templ->tokens, &shader->sse2_program,
 170                         shader->immediates, FALSE )) {
 171       FREE(shader);
 172       return NULL;
 173    }
 174
 175    shader->func = (tgsi_sse2_fs_function) x86_get_func( &shader->sse2_program );
 176    if (!shader->func) {
 177       x86_release_func( &shader->sse2_program );
 178       FREE(shader);
 179       return NULL;
 180    }
 181
 182    shader->base.shader.tokens = NULL; /* don't hold reference to templ->tokens */
 183    shader->base.prepare = fs_sse_prepare;
 184    shader->base.run = fs_sse_run;
 185    shader->base.delete = fs_sse_delete;
 186
 187    return &shader->base;
 188 }
 189
 190
 191 #else
 192
 193 /* Maybe put this variant in the header file.
 194  */
 195 struct sp_fragment_shader *
 196 softpipe_create_fs_sse(struct softpipe_context *softpipe,
 197                        const struct pipe_shader_state *templ)
 198 {
 199    return NULL;
 200 }
 201
 202 #endif