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 /* Surely this should be defined somewhere in a tgsi header:
  49  */
  50 typedef void (PIPE_CDECL *codegen_function)(
  51    const struct tgsi_exec_vector *input,
  52    struct tgsi_exec_vector *output,
  53    const float (*constant)[4],
  54    struct tgsi_exec_vector *temporary,
  55    const struct tgsi_interp_coef *coef,
  56    float (*immediates)[4]
  57    //, const struct tgsi_exec_vector *quadPos
  58  );
  59
  60
  61 /**
  62  * Subclass of sp_fragment_shader
  63  */
  64 struct sp_sse_fragment_shader
  65 {
  66    struct sp_fragment_shader base;
  67    struct x86_function sse2_program;
  68    codegen_function func;
  69    float immediates[TGSI_EXEC_NUM_IMMEDIATES][4];
  70 };
  71
  72
  73 /** cast wrapper */
  74 static INLINE struct sp_sse_fragment_shader *
  75 sp_sse_fragment_shader(const struct sp_fragment_shader *base)
  76 {
  77    return (struct sp_sse_fragment_shader *) base;
  78 }
  79
  80
  81 static void
  82 fs_sse_prepare( const struct sp_fragment_shader *base,
  83                 struct tgsi_exec_machine *machine,
  84                 struct tgsi_sampler **samplers )
  85 {
  86 }
  87
  88
  89 /* TODO: codegenerate the whole run function, skip this wrapper.
  90  * TODO: break dependency on tgsi_exec_machine struct
  91  * TODO: push Position calculation into the generated shader
  92  * TODO: process >1 quad at a time
  93  */
  94 static unsigned
  95 fs_sse_run( const struct sp_fragment_shader *base,
  96             struct tgsi_exec_machine *machine,
  97             struct quad_header *quad )
  98 {
  99    struct sp_sse_fragment_shader *shader = sp_sse_fragment_shader(base);
 100
 101    /* Compute X, Y, Z, W vals for this quad -- place in temp[0] for now */
 102    sp_setup_pos_vector(quad->posCoef,
 103                        (float)quad->input.x0, (float)quad->input.y0,
 104                        machine->Temps);
 105
 106    /* init kill mask */
 107    tgsi_set_kill_mask(machine, 0x0);
 108    tgsi_set_exec_mask(machine, 1, 1, 1, 1);
 109
 110    shader->func( machine->Inputs,
 111                  machine->Outputs,
 112                  machine->Consts,
 113                  machine->Temps,
 114                  machine->InterpCoefs,
 115                  shader->immediates
 116                  //      , &machine->QuadPos
 117       );
 118
 119    return ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]);
 120 }
 121
 122
 123 static void
 124 fs_sse_delete( struct sp_fragment_shader *base )
 125 {
 126    struct sp_sse_fragment_shader *shader = sp_sse_fragment_shader(base);
 127
 128    x86_release_func( &shader->sse2_program );
 129    FREE(shader);
 130 }
 131
 132
 133 struct sp_fragment_shader *
 134 softpipe_create_fs_sse(struct softpipe_context *softpipe,
 135                        const struct pipe_shader_state *templ)
 136 {
 137    struct sp_sse_fragment_shader *shader;
 138
 139    if (!softpipe->use_sse)
 140       return NULL;
 141
 142    shader = CALLOC_STRUCT(sp_sse_fragment_shader);
 143    if (!shader)
 144       return NULL;
 145
 146    x86_init_func( &shader->sse2_program );
 147
 148    if (!tgsi_emit_sse2( templ->tokens, &shader->sse2_program,
 149                         shader->immediates, FALSE )) {
 150       FREE(shader);
 151       return NULL;
 152    }
 153
 154    shader->func = (codegen_function) x86_get_func( &shader->sse2_program );
 155    if (!shader->func) {
 156       x86_release_func( &shader->sse2_program );
 157       FREE(shader);
 158       return NULL;
 159    }
 160
 161    shader->base.shader.tokens = NULL; /* don't hold reference to templ->tokens */
 162    shader->base.prepare = fs_sse_prepare;
 163    shader->base.run = fs_sse_run;
 164    shader->base.delete = fs_sse_delete;
 165
 166    return &shader->base;
 167 }
 168
 169
 170 #else
 171
 172 /* Maybe put this variant in the header file.
 173  */
 174 struct sp_fragment_shader *
 175 softpipe_create_fs_sse(struct softpipe_context *softpipe,
 176                        const struct pipe_shader_state *templ)
 177 {
 178    return NULL;
 179 }
 180
 181 #endif