src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 VMware, Inc.
   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 VMWARE 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 /**
  30  * @file
  31  * Blend LLVM IR generation -- AoS layout.
  32  *
  33  * AoS blending is in general much slower than SoA, but there are some cases
  34  * where it might be faster. In particular, if a pixel is rendered only once
  35  * then the overhead of tiling and untiling will dominate over the speedup that
  36  * SoA gives. So we might want to detect such cases and fallback to AoS in the
  37  * future, but for now this function is here for historical/benchmarking
  38  * purposes.
  39  *
  40  * Run lp_blend_test after any change to this file.
  41  *
  42  * @author Jose Fonseca <jfonseca@vmware.com>
  43  */
  44
  45
  46 #include "pipe/p_state.h"
  47 #include "util/u_debug.h"
  48 #include "util/u_format.h"
  49
  50 #include "gallivm/lp_bld_type.h"
  51 #include "gallivm/lp_bld_const.h"
  52 #include "gallivm/lp_bld_arit.h"
  53 #include "gallivm/lp_bld_logic.h"
  54 #include "gallivm/lp_bld_swizzle.h"
  55 #include "gallivm/lp_bld_bitarit.h"
  56 #include "gallivm/lp_bld_debug.h"
  57
  58 #include "lp_bld_blend.h"
  59
  60
  61 /**
  62  * We may the same values several times, so we keep them here to avoid
  63  * recomputing them. Also reusing the values allows us to do simplifications
  64  * that LLVM optimization passes wouldn't normally be able to do.
  65  */
  66 struct lp_build_blend_aos_context
  67 {
  68    struct lp_build_context base;
  69
  70    LLVMValueRef src;
  71    LLVMValueRef src_alpha;
  72    LLVMValueRef src1;
  73    LLVMValueRef src1_alpha;
  74    LLVMValueRef dst;
  75    LLVMValueRef const_;
  76    LLVMValueRef const_alpha;
  77    boolean has_dst_alpha;
  78
  79    LLVMValueRef inv_src;
  80    LLVMValueRef inv_src_alpha;
  81    LLVMValueRef inv_dst;
  82    LLVMValueRef inv_const;
  83    LLVMValueRef inv_const_alpha;
  84    LLVMValueRef saturate;
  85
  86    LLVMValueRef rgb_src_factor;
  87    LLVMValueRef alpha_src_factor;
  88    LLVMValueRef rgb_dst_factor;
  89    LLVMValueRef alpha_dst_factor;
  90 };
  91
  92
  93 static LLVMValueRef
  94 lp_build_blend_factor_unswizzled(struct lp_build_blend_aos_context *bld,
  95                                  unsigned factor,
  96                                  boolean alpha)
  97 {
  98    LLVMValueRef src_alpha = bld->src_alpha ? bld->src_alpha : bld->src;
  99    LLVMValueRef src1_alpha = bld->src1_alpha ? bld->src1_alpha : bld->src1;
 100    LLVMValueRef const_alpha = bld->const_alpha ? bld->const_alpha : bld->const_;
 101
 102    switch (factor) {
 103    case PIPE_BLENDFACTOR_ZERO:
 104       return bld->base.zero;
 105    case PIPE_BLENDFACTOR_ONE:
 106       return bld->base.one;
 107    case PIPE_BLENDFACTOR_SRC_COLOR:
 108       return bld->src;
 109    case PIPE_BLENDFACTOR_SRC_ALPHA:
 110       return src_alpha;
 111    case PIPE_BLENDFACTOR_DST_COLOR:
 112    case PIPE_BLENDFACTOR_DST_ALPHA:
 113       return bld->dst;
 114    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 115       if(alpha)
 116          return bld->base.one;
 117       else {
 118          /*
 119           * If there's no dst alpha the complement is zero but for unclamped
 120           * float inputs min can be non-zero (negative).
 121           */
 122          if (!bld->has_dst_alpha) {
 123             if (!bld->saturate)
 124                bld->saturate = lp_build_min(&bld->base, src_alpha, bld->base.zero);
 125          }
 126          else {
 127             if(!bld->inv_dst)
 128                bld->inv_dst = lp_build_comp(&bld->base, bld->dst);
 129             if(!bld->saturate)
 130                bld->saturate = lp_build_min(&bld->base, src_alpha, bld->inv_dst);
 131          }
 132          return bld->saturate;
 133       }
 134    case PIPE_BLENDFACTOR_CONST_COLOR:
 135       return bld->const_;
 136    case PIPE_BLENDFACTOR_CONST_ALPHA:
 137       return const_alpha;
 138    case PIPE_BLENDFACTOR_SRC1_COLOR:
 139       return bld->src1;
 140    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 141       return src1_alpha;
 142    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 143       if(!bld->inv_src)
 144          bld->inv_src = lp_build_comp(&bld->base, bld->src);
 145       return bld->inv_src;
 146    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 147       if(!bld->inv_src_alpha)
 148          bld->inv_src_alpha = lp_build_comp(&bld->base, src_alpha);
 149       return bld->inv_src_alpha;
 150    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 151    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 152       if(!bld->inv_dst)
 153          bld->inv_dst = lp_build_comp(&bld->base, bld->dst);
 154       return bld->inv_dst;
 155    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 156       if(!bld->inv_const)
 157          bld->inv_const = lp_build_comp(&bld->base, bld->const_);
 158       return bld->inv_const;
 159    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 160       if(!bld->inv_const_alpha)
 161          bld->inv_const_alpha = lp_build_comp(&bld->base, const_alpha);
 162       return bld->inv_const_alpha;
 163    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 164       return lp_build_comp(&bld->base, bld->src1);
 165    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 166       return lp_build_comp(&bld->base, src1_alpha);
 167    default:
 168       assert(0);
 169       return bld->base.zero;
 170    }
 171 }
 172
 173
 174 enum lp_build_blend_swizzle {
 175    LP_BUILD_BLEND_SWIZZLE_RGBA = 0,
 176    LP_BUILD_BLEND_SWIZZLE_AAAA = 1
 177 };
 178
 179
 180 /**
 181  * How should we shuffle the base factor.
 182  */
 183 static enum lp_build_blend_swizzle
 184 lp_build_blend_factor_swizzle(unsigned factor)
 185 {
 186    switch (factor) {
 187    case PIPE_BLENDFACTOR_ONE:
 188    case PIPE_BLENDFACTOR_ZERO:
 189    case PIPE_BLENDFACTOR_SRC_COLOR:
 190    case PIPE_BLENDFACTOR_DST_COLOR:
 191    case PIPE_BLENDFACTOR_CONST_COLOR:
 192    case PIPE_BLENDFACTOR_SRC1_COLOR:
 193    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 194    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 195    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 196    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 197       return LP_BUILD_BLEND_SWIZZLE_RGBA;
 198    case PIPE_BLENDFACTOR_SRC_ALPHA:
 199    case PIPE_BLENDFACTOR_DST_ALPHA:
 200    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 201    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 202    case PIPE_BLENDFACTOR_CONST_ALPHA:
 203    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 204    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 205    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 206    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 207       return LP_BUILD_BLEND_SWIZZLE_AAAA;
 208    default:
 209       assert(0);
 210       return LP_BUILD_BLEND_SWIZZLE_RGBA;
 211    }
 212 }
 213
 214
 215 static LLVMValueRef
 216 lp_build_blend_swizzle(struct lp_build_blend_aos_context *bld,
 217                        LLVMValueRef rgb,
 218                        LLVMValueRef alpha,
 219                        enum lp_build_blend_swizzle rgb_swizzle,
 220                        unsigned alpha_swizzle,
 221                        unsigned num_channels)
 222 {
 223    LLVMValueRef swizzled_rgb;
 224
 225    switch (rgb_swizzle) {
 226    case LP_BUILD_BLEND_SWIZZLE_RGBA:
 227       swizzled_rgb = rgb;
 228       break;
 229    case LP_BUILD_BLEND_SWIZZLE_AAAA:
 230       swizzled_rgb = lp_build_swizzle_scalar_aos(&bld->base, rgb, alpha_swizzle, num_channels);
 231       break;
 232    default:
 233       assert(0);
 234       swizzled_rgb = bld->base.undef;
 235    }
 236
 237    if (rgb != alpha) {
 238       swizzled_rgb = lp_build_select_aos(&bld->base, 1 << alpha_swizzle,
 239                                          alpha, swizzled_rgb,
 240                                          num_channels);
 241    }
 242
 243    return swizzled_rgb;
 244 }
 245
 246 /**
 247  * @sa http://www.opengl.org/sdk/docs/man/xhtml/glBlendFuncSeparate.xml
 248  */
 249 static LLVMValueRef
 250 lp_build_blend_factor(struct lp_build_blend_aos_context *bld,
 251                       unsigned rgb_factor,
 252                       unsigned alpha_factor,
 253                       unsigned alpha_swizzle,
 254                       unsigned num_channels)
 255 {
 256    LLVMValueRef rgb_factor_, alpha_factor_;
 257    enum lp_build_blend_swizzle rgb_swizzle;
 258
 259    if (alpha_swizzle == PIPE_SWIZZLE_X && num_channels == 1) {
 260       return lp_build_blend_factor_unswizzled(bld, alpha_factor, TRUE);
 261    }
 262
 263    rgb_factor_ = lp_build_blend_factor_unswizzled(bld, rgb_factor, FALSE);
 264
 265    if (alpha_swizzle != PIPE_SWIZZLE_NONE) {
 266       rgb_swizzle   = lp_build_blend_factor_swizzle(rgb_factor);
 267       alpha_factor_ = lp_build_blend_factor_unswizzled(bld, alpha_factor, TRUE);
 268       return lp_build_blend_swizzle(bld, rgb_factor_, alpha_factor_, rgb_swizzle,
 269                                     alpha_swizzle, num_channels);
 270    } else {
 271       return rgb_factor_;
 272    }
 273 }
 274
 275
 276 /**
 277  * Performs blending of src and dst pixels
 278  *
 279  * @param blend         the blend state of the shader variant
 280  * @param cbuf_format   format of the colour buffer
 281  * @param type          data type of the pixel vector
 282  * @param rt            render target index
 283  * @param src           blend src
 284  * @param src_alpha     blend src alpha (if not included in src)
 285  * @param src1          second blend src (for dual source blend)
 286  * @param src1_alpha    second blend src alpha (if not included in src1)
 287  * @param dst           blend dst
 288  * @param mask          optional mask to apply to the blending result
 289  * @param const_        const blend color
 290  * @param const_alpha   const blend color alpha (if not included in const_)
 291  * @param swizzle       swizzle values for RGBA
 292  *
 293  * @return the result of blending src and dst
 294  */
 295 LLVMValueRef
 296 lp_build_blend_aos(struct gallivm_state *gallivm,
 297                    const struct pipe_blend_state *blend,
 298                    enum pipe_format cbuf_format,
 299                    struct lp_type type,
 300                    unsigned rt,
 301                    LLVMValueRef src,
 302                    LLVMValueRef src_alpha,
 303                    LLVMValueRef src1,
 304                    LLVMValueRef src1_alpha,
 305                    LLVMValueRef dst,
 306                    LLVMValueRef mask,
 307                    LLVMValueRef const_,
 308                    LLVMValueRef const_alpha,
 309                    const unsigned char swizzle[4],
 310                    int nr_channels)
 311 {
 312    const struct pipe_rt_blend_state * state = &blend->rt[rt];
 313    const struct util_format_description * desc;
 314    struct lp_build_blend_aos_context bld;
 315    LLVMValueRef src_factor, dst_factor;
 316    LLVMValueRef result;
 317    unsigned alpha_swizzle = PIPE_SWIZZLE_NONE;
 318    unsigned i;
 319
 320    desc = util_format_description(cbuf_format);
 321
 322    /* Setup build context */
 323    memset(&bld, 0, sizeof bld);
 324    lp_build_context_init(&bld.base, gallivm, type);
 325    bld.src = src;
 326    bld.src1 = src1;
 327    bld.dst = dst;
 328    bld.const_ = const_;
 329    bld.src_alpha = src_alpha;
 330    bld.src1_alpha = src1_alpha;
 331    bld.const_alpha = const_alpha;
 332    bld.has_dst_alpha = FALSE;
 333
 334    /* Find the alpha channel if not provided seperately */
 335    if (!src_alpha) {
 336       for (i = 0; i < 4; ++i) {
 337          if (swizzle[i] == 3) {
 338             alpha_swizzle = i;
 339          }
 340       }
 341       /*
 342        * Note that we may get src_alpha included from source (and 4 channels)
 343        * even if the destination doesn't have an alpha channel (for rgbx
 344        * formats). Generally this shouldn't make much of a difference (we're
 345        * relying on blend factors being sanitized already if there's no
 346        * dst alpha).
 347        */
 348       bld.has_dst_alpha = desc->swizzle[3] <= PIPE_SWIZZLE_W;
 349    }
 350
 351    if (blend->logicop_enable) {
 352       if(!type.floating) {
 353          result = lp_build_logicop(gallivm->builder, blend->logicop_func, src, dst);
 354       }
 355       else {
 356          result = src;
 357       }
 358    } else if (!state->blend_enable) {
 359       result = src;
 360    } else {
 361       boolean rgb_alpha_same = (state->rgb_src_factor == state->rgb_dst_factor &&
 362                                 state->alpha_src_factor == state->alpha_dst_factor) ||
 363                                nr_channels == 1;
 364
 365       src_factor = lp_build_blend_factor(&bld, state->rgb_src_factor,
 366                                          state->alpha_src_factor,
 367                                          alpha_swizzle,
 368                                          nr_channels);
 369
 370       dst_factor = lp_build_blend_factor(&bld, state->rgb_dst_factor,
 371                                          state->alpha_dst_factor,
 372                                          alpha_swizzle,
 373                                          nr_channels);
 374
 375       result = lp_build_blend(&bld.base,
 376                               state->rgb_func,
 377                               state->rgb_src_factor,
 378                               state->rgb_dst_factor,
 379                               src,
 380                               dst,
 381                               src_factor,
 382                               dst_factor,
 383                               rgb_alpha_same,
 384                               false);
 385
 386       if(state->rgb_func != state->alpha_func && nr_channels > 1 &&
 387                             alpha_swizzle != PIPE_SWIZZLE_NONE) {
 388          LLVMValueRef alpha;
 389
 390          alpha = lp_build_blend(&bld.base,
 391                                 state->alpha_func,
 392                                 state->alpha_src_factor,
 393                                 state->alpha_dst_factor,
 394                                 src,
 395                                 dst,
 396                                 src_factor,
 397                                 dst_factor,
 398                                 rgb_alpha_same,
 399                                 false);
 400
 401          result = lp_build_blend_swizzle(&bld,
 402                                          result,
 403                                          alpha,
 404                                          LP_BUILD_BLEND_SWIZZLE_RGBA,
 405                                          alpha_swizzle,
 406                                          nr_channels);
 407       }
 408    }
 409
 410    /* Check if color mask is necessary */
 411    if (!util_format_colormask_full(desc, state->colormask)) {
 412       LLVMValueRef color_mask;
 413
 414       color_mask = lp_build_const_mask_aos_swizzled(gallivm, bld.base.type,
 415                                                     state->colormask, nr_channels, swizzle);
 416       lp_build_name(color_mask, "color_mask");
 417
 418       /* Combine with input mask if necessary */
 419       if (mask) {
 420          /* We can be blending floating values but masks are always integer... */
 421          unsigned floating = bld.base.type.floating;
 422          bld.base.type.floating = 0;
 423
 424          mask = lp_build_and(&bld.base, color_mask, mask);
 425
 426          bld.base.type.floating = floating;
 427       } else {
 428          mask = color_mask;
 429       }
 430    }
 431
 432    /* Apply mask, if one exists */
 433    if (mask) {
 434       result = lp_build_select(&bld.base, mask, result, dst);
 435    }
 436
 437    return result;
 438 }