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    return ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]);
 108 }
 109
 110
 111 static void
 112 fs_sse_delete( struct sp_fragment_shader *base )
 113 {
 114    struct sp_sse_fragment_shader *shader = sp_sse_fragment_shader(base);
 115
 116    x86_release_func( &shader->sse2_program );
 117    FREE(shader);
 118 }
 119
 120
 121 struct sp_fragment_shader *
 122 softpipe_create_fs_sse(struct softpipe_context *softpipe,
 123                        const struct pipe_shader_state *templ)
 124 {
 125    struct sp_sse_fragment_shader *shader;
 126
 127    if (!softpipe->use_sse)
 128       return NULL;
 129
 130    shader = CALLOC_STRUCT(sp_sse_fragment_shader);
 131    if (!shader)
 132       return NULL;
 133
 134    x86_init_func( &shader->sse2_program );
 135
 136    if (!tgsi_emit_sse2( templ->tokens, &shader->sse2_program,
 137                         shader->immediates, FALSE )) {
 138       FREE(shader);
 139       return NULL;
 140    }
 141
 142    shader->func = (tgsi_sse2_fs_function) x86_get_func( &shader->sse2_program );
 143    if (!shader->func) {
 144       x86_release_func( &shader->sse2_program );
 145       FREE(shader);
 146       return NULL;
 147    }
 148
 149    shader->base.shader.tokens = NULL; /* don't hold reference to templ->tokens */
 150    shader->base.prepare = fs_sse_prepare;
 151    shader->base.run = fs_sse_run;
 152    shader->base.delete = fs_sse_delete;
 153
 154    return &shader->base;
 155 }
 156
 157
 158 #else
 159
 160 /* Maybe put this variant in the header file.
 161  */
 162 struct sp_fragment_shader *
 163 softpipe_create_fs_sse(struct softpipe_context *softpipe,
 164                        const struct pipe_shader_state *templ)
 165 {
 166    return NULL;
 167 }
 168
 169 #endif