src/gallium/drivers/svga/svga_state_vs.c

   1 /**********************************************************
   2  * Copyright 2008-2009 VMware, Inc.  All rights reserved.
   3  *
   4  * Permission is hereby granted, free of charge, to any person
   5  * obtaining a copy of this software and associated documentation
   6  * files (the "Software"), to deal in the Software without
   7  * restriction, including without limitation the rights to use, copy,
   8  * modify, merge, publish, distribute, sublicense, and/or sell copies
   9  * of the Software, and to permit persons to whom the Software is
  10  * furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be
  13  * included in all copies or substantial portions of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  16  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  17  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  18  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  19  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  20  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  21  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  **********************************************************/
  25
  26 #include "util/u_inlines.h"
  27 #include "pipe/p_defines.h"
  28 #include "util/u_format.h"
  29 #include "util/u_math.h"
  30 #include "util/u_memory.h"
  31 #include "util/u_bitmask.h"
  32 #include "translate/translate.h"
  33 #include "tgsi/tgsi_ureg.h"
  34
  35 #include "svga_context.h"
  36 #include "svga_state.h"
  37 #include "svga_cmd.h"
  38 #include "svga_tgsi.h"
  39
  40 #include "svga_hw_reg.h"
  41
  42 /***********************************************************************
  43  */
  44
  45
  46 static INLINE int compare_vs_keys( const struct svga_vs_compile_key *a,
  47                                    const struct svga_vs_compile_key *b )
  48 {
  49    unsigned keysize = svga_vs_key_size( a );
  50    return memcmp( a, b, keysize );
  51 }
  52
  53
  54 static struct svga_shader_result *search_vs_key( struct svga_vertex_shader *vs,
  55                                                  const struct svga_vs_compile_key *key )
  56 {
  57    struct svga_shader_result *result = vs->base.results;
  58
  59    assert(key);
  60
  61    for ( ; result; result = result->next) {
  62       if (compare_vs_keys( key, &result->key.vkey ) == 0)
  63          return result;
  64    }
  65
  66    return NULL;
  67 }
  68
  69
  70 /**
  71  * If we fail to compile a vertex shader we'll use a dummy/fallback shader
  72  * that simply emits a (0,0,0,1) vertex position.
  73  */
  74 static const struct tgsi_token *
  75 get_dummy_vertex_shader(void)
  76 {
  77    static const float zero[4] = { 0.0, 0.0, 0.0, 1.0 };
  78    struct ureg_program *ureg;
  79    const struct tgsi_token *tokens;
  80    struct ureg_src src;
  81    struct ureg_dst dst;
  82    unsigned num_tokens;
  83
  84    ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
  85    if (!ureg)
  86       return NULL;
  87
  88    dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
  89    src = ureg_DECL_immediate(ureg, zero, 4);
  90    ureg_MOV(ureg, dst, src);
  91    ureg_END(ureg);
  92
  93    tokens = ureg_get_tokens(ureg, &num_tokens);
  94
  95    ureg_destroy(ureg);
  96
  97    return tokens;
  98 }
  99
 100
 101 static enum pipe_error compile_vs( struct svga_context *svga,
 102                                    struct svga_vertex_shader *vs,
 103                                    const struct svga_vs_compile_key *key,
 104                                    struct svga_shader_result **out_result )
 105 {
 106    struct svga_shader_result *result;
 107    enum pipe_error ret = PIPE_ERROR;
 108
 109    result = svga_translate_vertex_program( vs, key );
 110    if (result == NULL) {
 111       /* some problem during translation, try the dummy shader */
 112       const struct tgsi_token *dummy = get_dummy_vertex_shader();
 113       if (!dummy) {
 114          ret = PIPE_ERROR_OUT_OF_MEMORY;
 115          goto fail;
 116       }
 117       debug_printf("Failed to compile vertex shader, using dummy shader instead.\n");
 118       FREE((void *) vs->base.tokens);
 119       vs->base.tokens = dummy;
 120       result = svga_translate_vertex_program(vs, key);
 121       if (result == NULL) {
 122          ret = PIPE_ERROR;
 123          goto fail;
 124       }
 125    }
 126
 127    result->id = util_bitmask_add(svga->vs_bm);
 128    if(result->id == UTIL_BITMASK_INVALID_INDEX) {
 129       ret = PIPE_ERROR_OUT_OF_MEMORY;
 130       goto fail;
 131    }
 132
 133    ret = SVGA3D_DefineShader(svga->swc,
 134                              result->id,
 135                              SVGA3D_SHADERTYPE_VS,
 136                              result->tokens,
 137                              result->nr_tokens * sizeof result->tokens[0]);
 138    if (ret != PIPE_OK)
 139       goto fail;
 140
 141    *out_result = result;
 142    result->next = vs->base.results;
 143    vs->base.results = result;
 144    return PIPE_OK;
 145
 146 fail:
 147    if (result) {
 148       if (result->id != UTIL_BITMASK_INVALID_INDEX)
 149          util_bitmask_clear( svga->vs_bm, result->id );
 150       svga_destroy_shader_result( result );
 151    }
 152    return ret;
 153 }
 154
 155 /* SVGA_NEW_PRESCALE, SVGA_NEW_RAST, SVGA_NEW_ZERO_STRIDE, SVGA_NEW_FS
 156  */
 157 static void
 158 make_vs_key(struct svga_context *svga, struct svga_vs_compile_key *key)
 159 {
 160    memset(key, 0, sizeof *key);
 161    key->need_prescale = svga->state.hw_clear.prescale.enabled;
 162    key->allow_psiz = svga->curr.rast->templ.point_size_per_vertex;
 163    key->zero_stride_vertex_elements =
 164       svga->curr.zero_stride_vertex_elements;
 165    key->num_zero_stride_vertex_elements =
 166       svga->curr.num_zero_stride_vertex_elements;
 167
 168    /* SVGA_NEW_FS */
 169    key->fs_generic_inputs = svga->curr.fs->generic_inputs;
 170 }
 171
 172
 173
 174 static enum pipe_error
 175 emit_hw_vs(struct svga_context *svga, unsigned dirty)
 176 {
 177    struct svga_shader_result *result = NULL;
 178    unsigned id = SVGA3D_INVALID_ID;
 179    enum pipe_error ret = PIPE_OK;
 180
 181    /* SVGA_NEW_NEED_SWTNL */
 182    if (!svga->state.sw.need_swtnl) {
 183       struct svga_vertex_shader *vs = svga->curr.vs;
 184       struct svga_vs_compile_key key;
 185
 186       make_vs_key( svga, &key );
 187
 188       result = search_vs_key( vs, &key );
 189       if (!result) {
 190          ret = compile_vs( svga, vs, &key, &result );
 191          if (ret != PIPE_OK)
 192             return ret;
 193       }
 194
 195       assert (result);
 196       id = result->id;
 197    }
 198
 199    if (result != svga->state.hw_draw.vs) {
 200       ret = SVGA3D_SetShader(svga->swc,
 201                              SVGA3D_SHADERTYPE_VS,
 202                              id );
 203       if (ret != PIPE_OK)
 204          return ret;
 205
 206       svga->dirty |= SVGA_NEW_VS_RESULT;
 207       svga->state.hw_draw.vs = result;
 208    }
 209
 210    return PIPE_OK;
 211 }
 212
 213 struct svga_tracked_state svga_hw_vs =
 214 {
 215    "vertex shader (hwtnl)",
 216    (SVGA_NEW_VS |
 217     SVGA_NEW_FS |
 218     SVGA_NEW_PRESCALE |
 219     SVGA_NEW_NEED_SWTNL |
 220     SVGA_NEW_ZERO_STRIDE),
 221    emit_hw_vs
 222 };
 223
 224
 225 /**
 226  * This function handles the special case of vertex attributes
 227  * with stride=0.  Basically, copy those values into the constant
 228  * buffer and modify the vertex shader to get the values from the
 229  * constant buffer rather than a vertex array.
 230  */
 231 static enum pipe_error
 232 update_zero_stride( struct svga_context *svga,
 233                     unsigned dirty )
 234 {
 235    unsigned i;
 236
 237    svga->curr.zero_stride_vertex_elements = 0;
 238    svga->curr.num_zero_stride_vertex_elements = 0;
 239
 240    for (i = 0; i < svga->curr.velems->count; i++) {
 241       const struct pipe_vertex_element *vel = &svga->curr.velems->velem[i];
 242       const struct pipe_vertex_buffer *vbuffer = &svga->curr.vb[
 243          vel->vertex_buffer_index];
 244
 245       if (vbuffer->stride == 0) {
 246          unsigned const_idx =
 247             svga->curr.num_zero_stride_vertex_elements;
 248          struct pipe_transfer *transfer;
 249          struct translate *translate;
 250          struct translate_key key;
 251          void *mapped_buffer;
 252
 253          svga->curr.zero_stride_vertex_elements |= (1 << i);
 254          ++svga->curr.num_zero_stride_vertex_elements;
 255
 256          key.output_stride = 4 * sizeof(float);
 257          key.nr_elements = 1;
 258          key.element[0].type = TRANSLATE_ELEMENT_NORMAL;
 259          key.element[0].input_format = vel->src_format;
 260          key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 261          key.element[0].input_buffer = vel->vertex_buffer_index;
 262          key.element[0].input_offset = vel->src_offset;
 263          key.element[0].instance_divisor = vel->instance_divisor;
 264          key.element[0].output_offset = const_idx * 4 * sizeof(float);
 265
 266          translate_key_sanitize(&key);
 267          /* translate_generic_create is technically private but
 268           * we don't want to code-generate, just want generic
 269           * translation */
 270          translate = translate_generic_create(&key);
 271
 272          assert(vel->src_offset == 0);
 273
 274          mapped_buffer = pipe_buffer_map_range(&svga->pipe,
 275                                                vbuffer->buffer,
 276                                                vel->src_offset + vbuffer->buffer_offset,
 277                                                util_format_get_blocksize(vel->src_format),
 278                                                PIPE_TRANSFER_READ,
 279                                                &transfer);
 280          mapped_buffer = (uint8_t*)mapped_buffer - vel->src_offset;
 281
 282          translate->set_buffer(translate, vel->vertex_buffer_index,
 283                                mapped_buffer,
 284                                vbuffer->stride, ~0);
 285          translate->run(translate, 0, 1, 0,
 286                         svga->curr.zero_stride_constants);
 287
 288          pipe_buffer_unmap(&svga->pipe, transfer);
 289
 290          translate->release(translate);
 291       }
 292    }
 293
 294    if (svga->curr.num_zero_stride_vertex_elements)
 295       svga->dirty |= SVGA_NEW_ZERO_STRIDE;
 296
 297    return 0;
 298 }
 299
 300 struct svga_tracked_state svga_hw_update_zero_stride =
 301 {
 302    "update zero_stride",
 303    ( SVGA_NEW_VELEMENT |
 304      SVGA_NEW_VBUFFER ),
 305    update_zero_stride
 306 };