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
  78    LLVMValueRef inv_src;
  79    LLVMValueRef inv_src_alpha;
  80    LLVMValueRef inv_dst;
  81    LLVMValueRef inv_const;
  82    LLVMValueRef inv_const_alpha;
  83    LLVMValueRef saturate;
  84
  85    LLVMValueRef rgb_src_factor;
  86    LLVMValueRef alpha_src_factor;
  87    LLVMValueRef rgb_dst_factor;
  88    LLVMValueRef alpha_dst_factor;
  89 };
  90
  91
  92 static LLVMValueRef
  93 lp_build_blend_factor_unswizzled(struct lp_build_blend_aos_context *bld,
  94                                  unsigned factor,
  95                                  boolean alpha)
  96 {
  97    LLVMValueRef src_alpha = bld->src_alpha ? bld->src_alpha : bld->src;
  98    LLVMValueRef src1_alpha = bld->src1_alpha ? bld->src1_alpha : bld->src1;
  99    LLVMValueRef const_alpha = bld->const_alpha ? bld->const_alpha : bld->const_;
 100
 101    switch (factor) {
 102    case PIPE_BLENDFACTOR_ZERO:
 103       return bld->base.zero;
 104    case PIPE_BLENDFACTOR_ONE:
 105       return bld->base.one;
 106    case PIPE_BLENDFACTOR_SRC_COLOR:
 107       return bld->src;
 108    case PIPE_BLENDFACTOR_SRC_ALPHA:
 109       return src_alpha;
 110    case PIPE_BLENDFACTOR_DST_COLOR:
 111    case PIPE_BLENDFACTOR_DST_ALPHA:
 112       return bld->dst;
 113    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 114       if(alpha)
 115          return bld->base.one;
 116       else {
 117          if(!bld->inv_dst)
 118             bld->inv_dst = lp_build_comp(&bld->base, bld->dst);
 119          if(!bld->saturate)
 120             bld->saturate = lp_build_min(&bld->base, src_alpha, bld->inv_dst);
 121          return bld->saturate;
 122       }
 123    case PIPE_BLENDFACTOR_CONST_COLOR:
 124       return bld->const_;
 125    case PIPE_BLENDFACTOR_CONST_ALPHA:
 126       return const_alpha;
 127    case PIPE_BLENDFACTOR_SRC1_COLOR:
 128       return bld->src1;
 129    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 130       return src1_alpha;
 131    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 132       if(!bld->inv_src)
 133          bld->inv_src = lp_build_comp(&bld->base, bld->src);
 134       return bld->inv_src;
 135    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 136       if(!bld->inv_src_alpha)
 137          bld->inv_src_alpha = lp_build_comp(&bld->base, src_alpha);
 138       return bld->inv_src_alpha;
 139    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 140    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 141       if(!bld->inv_dst)
 142          bld->inv_dst = lp_build_comp(&bld->base, bld->dst);
 143       return bld->inv_dst;
 144    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 145       if(!bld->inv_const)
 146          bld->inv_const = lp_build_comp(&bld->base, bld->const_);
 147       return bld->inv_const;
 148    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 149       if(!bld->inv_const_alpha)
 150          bld->inv_const_alpha = lp_build_comp(&bld->base, const_alpha);
 151       return bld->inv_const_alpha;
 152    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 153       return lp_build_comp(&bld->base, bld->src1);
 154    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 155       return lp_build_comp(&bld->base, src1_alpha);
 156    default:
 157       assert(0);
 158       return bld->base.zero;
 159    }
 160 }
 161
 162
 163 enum lp_build_blend_swizzle {
 164    LP_BUILD_BLEND_SWIZZLE_RGBA = 0,
 165    LP_BUILD_BLEND_SWIZZLE_AAAA = 1
 166 };
 167
 168
 169 /**
 170  * How should we shuffle the base factor.
 171  */
 172 static enum lp_build_blend_swizzle
 173 lp_build_blend_factor_swizzle(unsigned factor)
 174 {
 175    switch (factor) {
 176    case PIPE_BLENDFACTOR_ONE:
 177    case PIPE_BLENDFACTOR_ZERO:
 178    case PIPE_BLENDFACTOR_SRC_COLOR:
 179    case PIPE_BLENDFACTOR_DST_COLOR:
 180    case PIPE_BLENDFACTOR_CONST_COLOR:
 181    case PIPE_BLENDFACTOR_SRC1_COLOR:
 182    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 183    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 184    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 185    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 186       return LP_BUILD_BLEND_SWIZZLE_RGBA;
 187    case PIPE_BLENDFACTOR_SRC_ALPHA:
 188    case PIPE_BLENDFACTOR_DST_ALPHA:
 189    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 190    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 191    case PIPE_BLENDFACTOR_CONST_ALPHA:
 192    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 193    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 194    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 195    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 196       return LP_BUILD_BLEND_SWIZZLE_AAAA;
 197    default:
 198       assert(0);
 199       return LP_BUILD_BLEND_SWIZZLE_RGBA;
 200    }
 201 }
 202
 203
 204 static LLVMValueRef
 205 lp_build_blend_swizzle(struct lp_build_blend_aos_context *bld,
 206                        LLVMValueRef rgb,
 207                        LLVMValueRef alpha,
 208                        enum lp_build_blend_swizzle rgb_swizzle,
 209                        unsigned alpha_swizzle,
 210                        unsigned num_channels)
 211 {
 212    LLVMValueRef swizzled_rgb;
 213
 214    switch (rgb_swizzle) {
 215    case LP_BUILD_BLEND_SWIZZLE_RGBA:
 216       swizzled_rgb = rgb;
 217       break;
 218    case LP_BUILD_BLEND_SWIZZLE_AAAA:
 219       swizzled_rgb = lp_build_swizzle_scalar_aos(&bld->base, rgb, alpha_swizzle, num_channels);
 220       break;
 221    default:
 222       assert(0);
 223       swizzled_rgb = bld->base.undef;
 224    }
 225
 226    if (rgb != alpha) {
 227       swizzled_rgb = lp_build_select_aos(&bld->base, 1 << alpha_swizzle,
 228                                          alpha, swizzled_rgb,
 229                                          num_channels);
 230    }
 231
 232    return swizzled_rgb;
 233 }
 234
 235 /**
 236  * @sa http://www.opengl.org/sdk/docs/man/xhtml/glBlendFuncSeparate.xml
 237  */
 238 static LLVMValueRef
 239 lp_build_blend_factor(struct lp_build_blend_aos_context *bld,
 240                       unsigned rgb_factor,
 241                       unsigned alpha_factor,
 242                       unsigned alpha_swizzle,
 243                       unsigned num_channels)
 244 {
 245    LLVMValueRef rgb_factor_, alpha_factor_;
 246    enum lp_build_blend_swizzle rgb_swizzle;
 247
 248    if (alpha_swizzle == 0) {
 249       return lp_build_blend_factor_unswizzled(bld, alpha_factor, TRUE);
 250    }
 251
 252    rgb_factor_ = lp_build_blend_factor_unswizzled(bld, rgb_factor, FALSE);
 253
 254    if (alpha_swizzle != UTIL_FORMAT_SWIZZLE_NONE) {
 255       rgb_swizzle   = lp_build_blend_factor_swizzle(rgb_factor);
 256       alpha_factor_ = lp_build_blend_factor_unswizzled(bld, alpha_factor, TRUE);
 257       return lp_build_blend_swizzle(bld, rgb_factor_, alpha_factor_, rgb_swizzle, alpha_swizzle, num_channels);
 258    } else {
 259       return rgb_factor_;
 260    }
 261 }
 262
 263
 264 /**
 265  * Performs blending of src and dst pixels
 266  *
 267  * @param blend         the blend state of the shader variant
 268  * @param cbuf_format   format of the colour buffer
 269  * @param type          data type of the pixel vector
 270  * @param rt            render target index
 271  * @param src           blend src
 272  * @param src_alpha     blend src alpha (if not included in src)
 273  * @param src1          second blend src (for dual source blend)
 274  * @param src1_alpha    second blend src alpha (if not included in src1)
 275  * @param dst           blend dst
 276  * @param mask          optional mask to apply to the blending result
 277  * @param const_        const blend color
 278  * @param const_alpha   const blend color alpha (if not included in const_)
 279  * @param swizzle       swizzle values for RGBA
 280  *
 281  * @return the result of blending src and dst
 282  */
 283 LLVMValueRef
 284 lp_build_blend_aos(struct gallivm_state *gallivm,
 285                    const struct pipe_blend_state *blend,
 286                    enum pipe_format cbuf_format,
 287                    struct lp_type type,
 288                    unsigned rt,
 289                    LLVMValueRef src,
 290                    LLVMValueRef src_alpha,
 291                    LLVMValueRef src1,
 292                    LLVMValueRef src1_alpha,
 293                    LLVMValueRef dst,
 294                    LLVMValueRef mask,
 295                    LLVMValueRef const_,
 296                    LLVMValueRef const_alpha,
 297                    const unsigned char swizzle[4],
 298                    int nr_channels)
 299 {
 300    const struct pipe_rt_blend_state * state = &blend->rt[rt];
 301    const struct util_format_description * desc;
 302    struct lp_build_blend_aos_context bld;
 303    LLVMValueRef src_factor, dst_factor;
 304    LLVMValueRef result;
 305    unsigned alpha_swizzle = UTIL_FORMAT_SWIZZLE_NONE;
 306    unsigned i;
 307
 308    desc = util_format_description(cbuf_format);
 309
 310    /* Setup build context */
 311    memset(&bld, 0, sizeof bld);
 312    lp_build_context_init(&bld.base, gallivm, type);
 313    bld.src = src;
 314    bld.src1 = src1;
 315    bld.dst = dst;
 316    bld.const_ = const_;
 317    bld.src_alpha = src_alpha;
 318    bld.src1_alpha = src1_alpha;
 319    bld.const_alpha = const_alpha;
 320
 321    /* Find the alpha channel if not provided seperately */
 322    if (!src_alpha) {
 323       for (i = 0; i < 4; ++i) {
 324          if (swizzle[i] == 3) {
 325             alpha_swizzle = i;
 326          }
 327       }
 328    }
 329
 330    if (blend->logicop_enable) {
 331       if(!type.floating) {
 332          result = lp_build_logicop(gallivm->builder, blend->logicop_func, src, dst);
 333       }
 334       else {
 335          result = src;
 336       }
 337    } else if (!state->blend_enable) {
 338       result = src;
 339    } else {
 340       boolean rgb_alpha_same = (state->rgb_src_factor == state->rgb_dst_factor && state->alpha_src_factor == state->alpha_dst_factor) || nr_channels == 1;
 341
 342       src_factor = lp_build_blend_factor(&bld, state->rgb_src_factor,
 343                                          state->alpha_src_factor,
 344                                          alpha_swizzle,
 345                                          nr_channels);
 346
 347       dst_factor = lp_build_blend_factor(&bld, state->rgb_dst_factor,
 348                                          state->alpha_dst_factor,
 349                                          alpha_swizzle,
 350                                          nr_channels);
 351
 352       result = lp_build_blend(&bld.base,
 353                               state->rgb_func,
 354                               state->rgb_src_factor,
 355                               state->rgb_dst_factor,
 356                               src,
 357                               dst,
 358                               src_factor,
 359                               dst_factor,
 360                               rgb_alpha_same,
 361                               false);
 362
 363       if(state->rgb_func != state->alpha_func && nr_channels > 1 && alpha_swizzle != UTIL_FORMAT_SWIZZLE_NONE) {
 364          LLVMValueRef alpha;
 365
 366          alpha = lp_build_blend(&bld.base,
 367                                 state->alpha_func,
 368                                 state->alpha_src_factor,
 369                                 state->alpha_dst_factor,
 370                                 src,
 371                                 dst,
 372                                 src_factor,
 373                                 dst_factor,
 374                                 rgb_alpha_same,
 375                                 false);
 376
 377          result = lp_build_blend_swizzle(&bld,
 378                                          result,
 379                                          alpha,
 380                                          LP_BUILD_BLEND_SWIZZLE_RGBA,
 381                                          alpha_swizzle,
 382                                          nr_channels);
 383       }
 384    }
 385
 386    /* Check if color mask is necessary */
 387    if (!util_format_colormask_full(desc, state->colormask)) {
 388       LLVMValueRef color_mask;
 389
 390       color_mask = lp_build_const_mask_aos_swizzled(gallivm, bld.base.type, state->colormask, nr_channels, swizzle);
 391       lp_build_name(color_mask, "color_mask");
 392
 393       /* Combine with input mask if necessary */
 394       if (mask) {
 395          /* We can be blending floating values but masks are always integer... */
 396          unsigned floating = bld.base.type.floating;
 397          bld.base.type.floating = 0;
 398
 399          mask = lp_build_and(&bld.base, color_mask, mask);
 400
 401          bld.base.type.floating = floating;
 402       } else {
 403          mask = color_mask;
 404       }
 405    }
 406
 407    /* Apply mask, if one exists */
 408    if (mask) {
 409       result = lp_build_select(&bld.base, mask, result, dst);
 410    }
 411
 412    return result;
 413 }