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          /*
 118           * if there's separate src_alpha there's no dst alpha hence the complement
 119           * is zero but for unclamped float inputs min can be non-zero (negative).
 120           */
 121          if (bld->src_alpha) {
 122             if (!bld->saturate)
 123                bld->saturate = lp_build_min(&bld->base, src_alpha, bld->base.zero);
 124          }
 125          else {
 126             if(!bld->inv_dst)
 127                bld->inv_dst = lp_build_comp(&bld->base, bld->dst);
 128             if(!bld->saturate)
 129                bld->saturate = lp_build_min(&bld->base, src_alpha, bld->inv_dst);
 130          }
 131          return bld->saturate;
 132       }
 133    case PIPE_BLENDFACTOR_CONST_COLOR:
 134       return bld->const_;
 135    case PIPE_BLENDFACTOR_CONST_ALPHA:
 136       return const_alpha;
 137    case PIPE_BLENDFACTOR_SRC1_COLOR:
 138       return bld->src1;
 139    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 140       return src1_alpha;
 141    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 142       if(!bld->inv_src)
 143          bld->inv_src = lp_build_comp(&bld->base, bld->src);
 144       return bld->inv_src;
 145    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 146       if(!bld->inv_src_alpha)
 147          bld->inv_src_alpha = lp_build_comp(&bld->base, src_alpha);
 148       return bld->inv_src_alpha;
 149    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 150    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 151       if(!bld->inv_dst)
 152          bld->inv_dst = lp_build_comp(&bld->base, bld->dst);
 153       return bld->inv_dst;
 154    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 155       if(!bld->inv_const)
 156          bld->inv_const = lp_build_comp(&bld->base, bld->const_);
 157       return bld->inv_const;
 158    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 159       if(!bld->inv_const_alpha)
 160          bld->inv_const_alpha = lp_build_comp(&bld->base, const_alpha);
 161       return bld->inv_const_alpha;
 162    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 163       return lp_build_comp(&bld->base, bld->src1);
 164    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 165       return lp_build_comp(&bld->base, src1_alpha);
 166    default:
 167       assert(0);
 168       return bld->base.zero;
 169    }
 170 }
 171
 172
 173 enum lp_build_blend_swizzle {
 174    LP_BUILD_BLEND_SWIZZLE_RGBA = 0,
 175    LP_BUILD_BLEND_SWIZZLE_AAAA = 1
 176 };
 177
 178
 179 /**
 180  * How should we shuffle the base factor.
 181  */
 182 static enum lp_build_blend_swizzle
 183 lp_build_blend_factor_swizzle(unsigned factor)
 184 {
 185    switch (factor) {
 186    case PIPE_BLENDFACTOR_ONE:
 187    case PIPE_BLENDFACTOR_ZERO:
 188    case PIPE_BLENDFACTOR_SRC_COLOR:
 189    case PIPE_BLENDFACTOR_DST_COLOR:
 190    case PIPE_BLENDFACTOR_CONST_COLOR:
 191    case PIPE_BLENDFACTOR_SRC1_COLOR:
 192    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
 193    case PIPE_BLENDFACTOR_INV_DST_COLOR:
 194    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
 195    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
 196       return LP_BUILD_BLEND_SWIZZLE_RGBA;
 197    case PIPE_BLENDFACTOR_SRC_ALPHA:
 198    case PIPE_BLENDFACTOR_DST_ALPHA:
 199    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
 200    case PIPE_BLENDFACTOR_SRC1_ALPHA:
 201    case PIPE_BLENDFACTOR_CONST_ALPHA:
 202    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
 203    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
 204    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
 205    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
 206       return LP_BUILD_BLEND_SWIZZLE_AAAA;
 207    default:
 208       assert(0);
 209       return LP_BUILD_BLEND_SWIZZLE_RGBA;
 210    }
 211 }
 212
 213
 214 static LLVMValueRef
 215 lp_build_blend_swizzle(struct lp_build_blend_aos_context *bld,
 216                        LLVMValueRef rgb,
 217                        LLVMValueRef alpha,
 218                        enum lp_build_blend_swizzle rgb_swizzle,
 219                        unsigned alpha_swizzle,
 220                        unsigned num_channels)
 221 {
 222    LLVMValueRef swizzled_rgb;
 223
 224    switch (rgb_swizzle) {
 225    case LP_BUILD_BLEND_SWIZZLE_RGBA:
 226       swizzled_rgb = rgb;
 227       break;
 228    case LP_BUILD_BLEND_SWIZZLE_AAAA:
 229       swizzled_rgb = lp_build_swizzle_scalar_aos(&bld->base, rgb, alpha_swizzle, num_channels);
 230       break;
 231    default:
 232       assert(0);
 233       swizzled_rgb = bld->base.undef;
 234    }
 235
 236    if (rgb != alpha) {
 237       swizzled_rgb = lp_build_select_aos(&bld->base, 1 << alpha_swizzle,
 238                                          alpha, swizzled_rgb,
 239                                          num_channels);
 240    }
 241
 242    return swizzled_rgb;
 243 }
 244
 245 /**
 246  * @sa http://www.opengl.org/sdk/docs/man/xhtml/glBlendFuncSeparate.xml
 247  */
 248 static LLVMValueRef
 249 lp_build_blend_factor(struct lp_build_blend_aos_context *bld,
 250                       unsigned rgb_factor,
 251                       unsigned alpha_factor,
 252                       unsigned alpha_swizzle,
 253                       unsigned num_channels)
 254 {
 255    LLVMValueRef rgb_factor_, alpha_factor_;
 256    enum lp_build_blend_swizzle rgb_swizzle;
 257
 258    if (alpha_swizzle == UTIL_FORMAT_SWIZZLE_X && num_channels == 1) {
 259       return lp_build_blend_factor_unswizzled(bld, alpha_factor, TRUE);
 260    }
 261
 262    rgb_factor_ = lp_build_blend_factor_unswizzled(bld, rgb_factor, FALSE);
 263
 264    if (alpha_swizzle != UTIL_FORMAT_SWIZZLE_NONE) {
 265       rgb_swizzle   = lp_build_blend_factor_swizzle(rgb_factor);
 266       alpha_factor_ = lp_build_blend_factor_unswizzled(bld, alpha_factor, TRUE);
 267       return lp_build_blend_swizzle(bld, rgb_factor_, alpha_factor_, rgb_swizzle, alpha_swizzle, num_channels);
 268    } else {
 269       return rgb_factor_;
 270    }
 271 }
 272
 273
 274 /**
 275  * Performs blending of src and dst pixels
 276  *
 277  * @param blend         the blend state of the shader variant
 278  * @param cbuf_format   format of the colour buffer
 279  * @param type          data type of the pixel vector
 280  * @param rt            render target index
 281  * @param src           blend src
 282  * @param src_alpha     blend src alpha (if not included in src)
 283  * @param src1          second blend src (for dual source blend)
 284  * @param src1_alpha    second blend src alpha (if not included in src1)
 285  * @param dst           blend dst
 286  * @param mask          optional mask to apply to the blending result
 287  * @param const_        const blend color
 288  * @param const_alpha   const blend color alpha (if not included in const_)
 289  * @param swizzle       swizzle values for RGBA
 290  *
 291  * @return the result of blending src and dst
 292  */
 293 LLVMValueRef
 294 lp_build_blend_aos(struct gallivm_state *gallivm,
 295                    const struct pipe_blend_state *blend,
 296                    enum pipe_format cbuf_format,
 297                    struct lp_type type,
 298                    unsigned rt,
 299                    LLVMValueRef src,
 300                    LLVMValueRef src_alpha,
 301                    LLVMValueRef src1,
 302                    LLVMValueRef src1_alpha,
 303                    LLVMValueRef dst,
 304                    LLVMValueRef mask,
 305                    LLVMValueRef const_,
 306                    LLVMValueRef const_alpha,
 307                    const unsigned char swizzle[4],
 308                    int nr_channels)
 309 {
 310    const struct pipe_rt_blend_state * state = &blend->rt[rt];
 311    const struct util_format_description * desc;
 312    struct lp_build_blend_aos_context bld;
 313    LLVMValueRef src_factor, dst_factor;
 314    LLVMValueRef result;
 315    unsigned alpha_swizzle = UTIL_FORMAT_SWIZZLE_NONE;
 316    unsigned i;
 317
 318    desc = util_format_description(cbuf_format);
 319
 320    /* Setup build context */
 321    memset(&bld, 0, sizeof bld);
 322    lp_build_context_init(&bld.base, gallivm, type);
 323    bld.src = src;
 324    bld.src1 = src1;
 325    bld.dst = dst;
 326    bld.const_ = const_;
 327    bld.src_alpha = src_alpha;
 328    bld.src1_alpha = src1_alpha;
 329    bld.const_alpha = const_alpha;
 330
 331    /* Find the alpha channel if not provided seperately */
 332    if (!src_alpha) {
 333       for (i = 0; i < 4; ++i) {
 334          if (swizzle[i] == 3) {
 335             alpha_swizzle = i;
 336          }
 337       }
 338    }
 339
 340    if (blend->logicop_enable) {
 341       if(!type.floating) {
 342          result = lp_build_logicop(gallivm->builder, blend->logicop_func, src, dst);
 343       }
 344       else {
 345          result = src;
 346       }
 347    } else if (!state->blend_enable) {
 348       result = src;
 349    } else {
 350       boolean rgb_alpha_same = (state->rgb_src_factor == state->rgb_dst_factor && state->alpha_src_factor == state->alpha_dst_factor) || nr_channels == 1;
 351
 352       src_factor = lp_build_blend_factor(&bld, state->rgb_src_factor,
 353                                          state->alpha_src_factor,
 354                                          alpha_swizzle,
 355                                          nr_channels);
 356
 357       dst_factor = lp_build_blend_factor(&bld, state->rgb_dst_factor,
 358                                          state->alpha_dst_factor,
 359                                          alpha_swizzle,
 360                                          nr_channels);
 361
 362       result = lp_build_blend(&bld.base,
 363                               state->rgb_func,
 364                               state->rgb_src_factor,
 365                               state->rgb_dst_factor,
 366                               src,
 367                               dst,
 368                               src_factor,
 369                               dst_factor,
 370                               rgb_alpha_same,
 371                               false);
 372
 373       if(state->rgb_func != state->alpha_func && nr_channels > 1 && alpha_swizzle != UTIL_FORMAT_SWIZZLE_NONE) {
 374          LLVMValueRef alpha;
 375
 376          alpha = lp_build_blend(&bld.base,
 377                                 state->alpha_func,
 378                                 state->alpha_src_factor,
 379                                 state->alpha_dst_factor,
 380                                 src,
 381                                 dst,
 382                                 src_factor,
 383                                 dst_factor,
 384                                 rgb_alpha_same,
 385                                 false);
 386
 387          result = lp_build_blend_swizzle(&bld,
 388                                          result,
 389                                          alpha,
 390                                          LP_BUILD_BLEND_SWIZZLE_RGBA,
 391                                          alpha_swizzle,
 392                                          nr_channels);
 393       }
 394    }
 395
 396    /* Check if color mask is necessary */
 397    if (!util_format_colormask_full(desc, state->colormask)) {
 398       LLVMValueRef color_mask;
 399
 400       color_mask = lp_build_const_mask_aos_swizzled(gallivm, bld.base.type, state->colormask, nr_channels, swizzle);
 401       lp_build_name(color_mask, "color_mask");
 402
 403       /* Combine with input mask if necessary */
 404       if (mask) {
 405          /* We can be blending floating values but masks are always integer... */
 406          unsigned floating = bld.base.type.floating;
 407          bld.base.type.floating = 0;
 408
 409          mask = lp_build_and(&bld.base, color_mask, mask);
 410
 411          bld.base.type.floating = floating;
 412       } else {
 413          mask = color_mask;
 414       }
 415    }
 416
 417    /* Apply mask, if one exists */
 418    if (mask) {
 419       result = lp_build_select(&bld.base, mask, result, dst);
 420    }
 421
 422    return result;
 423 }