src/amd/llvm/ac_llvm_cull.c

   1 /*
   2  * Copyright 2019 Advanced Micro Devices, Inc.
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the
   6  * "Software"), to deal in the Software without restriction, including
   7  * without limitation the rights to use, copy, modify, merge, publish,
   8  * distribute, sub license, and/or sell copies of the Software, and to
   9  * permit persons to whom the Software is furnished to do so, subject to
  10  * the following conditions:
  11  *
  12  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  13  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  14  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  15  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
  16  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  17  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  18  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  19  *
  20  * The above copyright notice and this permission notice (including the
  21  * next paragraph) shall be included in all copies or substantial portions
  22  * of the Software.
  23  *
  24  */
  25
  26 #include "ac_llvm_cull.h"
  27 #include <llvm-c/Core.h>
  28
  29 struct ac_position_w_info {
  30         /* If a primitive intersects the W=0 plane, it causes a reflection
  31          * of the determinant used for face culling. Every vertex behind
  32          * the W=0 plane negates the determinant, so having 2 vertices behind
  33          * the plane has no effect. This is i1 true if the determinant should be
  34          * negated.
  35          */
  36         LLVMValueRef w_reflection;
  37
  38         /* If we simplify the "-w <= p <= w" view culling equation, we get
  39          * "-w <= w", which can't be satisfied when w is negative.
  40          * In perspective projection, a negative W means that the primitive
  41          * is behind the viewer, but the equation is independent of the type
  42          * of projection.
  43          *
  44          * w_accepted is false when all W are negative and therefore
  45          * the primitive is invisible.
  46          */
  47         LLVMValueRef w_accepted;
  48
  49         LLVMValueRef all_w_positive;
  50         LLVMValueRef any_w_negative;
  51 };
  52
  53 static void ac_analyze_position_w(struct ac_llvm_context *ctx,
  54                                   LLVMValueRef pos[3][4],
  55                                   struct ac_position_w_info *w)
  56 {
  57         LLVMBuilderRef builder = ctx->builder;
  58         LLVMValueRef all_w_negative = ctx->i1true;
  59
  60         w->w_reflection = ctx->i1false;
  61         w->any_w_negative = ctx->i1false;
  62
  63         for (unsigned i = 0; i < 3; i++) {
  64                 LLVMValueRef neg_w;
  65
  66                 neg_w = LLVMBuildFCmp(builder, LLVMRealOLT, pos[i][3], ctx->f32_0, "");
  67                 /* If neg_w is true, negate w_reflection. */
  68                 w->w_reflection = LLVMBuildXor(builder, w->w_reflection, neg_w, "");
  69                 w->any_w_negative = LLVMBuildOr(builder, w->any_w_negative, neg_w, "");
  70                 all_w_negative = LLVMBuildAnd(builder, all_w_negative, neg_w, "");
  71         }
  72         w->all_w_positive = LLVMBuildNot(builder, w->any_w_negative, "");
  73         w->w_accepted = LLVMBuildNot(builder, all_w_negative, "");
  74 }
  75
  76 /* Perform front/back face culling and return true if the primitive is accepted. */
  77 static LLVMValueRef ac_cull_face(struct ac_llvm_context *ctx,
  78                                  LLVMValueRef pos[3][4],
  79                                  struct ac_position_w_info *w,
  80                                  bool cull_front,
  81                                  bool cull_back,
  82                                  bool cull_zero_area)
  83 {
  84         LLVMBuilderRef builder = ctx->builder;
  85
  86         if (cull_front && cull_back)
  87                 return ctx->i1false;
  88
  89         if (!cull_front && !cull_back && !cull_zero_area)
  90                 return ctx->i1true;
  91
  92         /* Front/back face culling. Also if the determinant == 0, the triangle
  93          * area is 0.
  94          */
  95         LLVMValueRef det_t0 = LLVMBuildFSub(builder, pos[2][0], pos[0][0], "");
  96         LLVMValueRef det_t1 = LLVMBuildFSub(builder, pos[1][1], pos[0][1], "");
  97         LLVMValueRef det_t2 = LLVMBuildFSub(builder, pos[0][0], pos[1][0], "");
  98         LLVMValueRef det_t3 = LLVMBuildFSub(builder, pos[0][1], pos[2][1], "");
  99         LLVMValueRef det_p0 = LLVMBuildFMul(builder, det_t0, det_t1, "");
 100         LLVMValueRef det_p1 = LLVMBuildFMul(builder, det_t2, det_t3, "");
 101         LLVMValueRef det = LLVMBuildFSub(builder, det_p0, det_p1, "");
 102
 103         /* Negative W negates the determinant. */
 104         det = LLVMBuildSelect(builder, w->w_reflection,
 105                               LLVMBuildFNeg(builder, det, ""),
 106                               det, "");
 107
 108         LLVMValueRef accepted = NULL;
 109         if (cull_front) {
 110                 LLVMRealPredicate cond = cull_zero_area ? LLVMRealOGT : LLVMRealOGE;
 111                 accepted = LLVMBuildFCmp(builder, cond, det, ctx->f32_0, "");
 112         } else if (cull_back) {
 113                 LLVMRealPredicate cond = cull_zero_area ? LLVMRealOLT : LLVMRealOLE;
 114                 accepted = LLVMBuildFCmp(builder, cond, det, ctx->f32_0, "");
 115         } else if (cull_zero_area) {
 116                 accepted = LLVMBuildFCmp(builder, LLVMRealONE, det, ctx->f32_0, "");
 117         }
 118         return accepted;
 119 }
 120
 121 /* Perform view culling and small primitive elimination and return true
 122  * if the primitive is accepted and initially_accepted == true. */
 123 static LLVMValueRef cull_bbox(struct ac_llvm_context *ctx,
 124                               LLVMValueRef pos[3][4],
 125                               LLVMValueRef initially_accepted,
 126                               struct ac_position_w_info *w,
 127                               LLVMValueRef vp_scale[2],
 128                               LLVMValueRef vp_translate[2],
 129                               LLVMValueRef small_prim_precision,
 130                               bool cull_view_xy,
 131                               bool cull_view_near_z,
 132                               bool cull_view_far_z,
 133                               bool cull_small_prims,
 134                               bool use_halfz_clip_space)
 135 {
 136         LLVMBuilderRef builder = ctx->builder;
 137
 138         if (!cull_view_xy && !cull_view_near_z && !cull_view_far_z && !cull_small_prims)
 139                 return initially_accepted;
 140
 141         /* Skip the culling if the primitive has already been rejected or
 142          * if any W is negative. The bounding box culling doesn't work when
 143          * W is negative.
 144          */
 145         LLVMValueRef cond = LLVMBuildAnd(builder, initially_accepted,
 146                                          w->all_w_positive, "");
 147         LLVMValueRef accepted_var = ac_build_alloca_undef(ctx, ctx->i1, "");
 148         LLVMBuildStore(builder, initially_accepted, accepted_var);
 149
 150         ac_build_ifcc(ctx, cond, 10000000 /* does this matter? */);
 151         {
 152                 LLVMValueRef bbox_min[3], bbox_max[3];
 153                 LLVMValueRef accepted = initially_accepted;
 154
 155                 /* Compute the primitive bounding box for easy culling. */
 156                 for (unsigned chan = 0; chan < (cull_view_near_z || cull_view_far_z ? 3 : 2); chan++) {
 157                         bbox_min[chan] = ac_build_fmin(ctx, pos[0][chan], pos[1][chan]);
 158                         bbox_min[chan] = ac_build_fmin(ctx, bbox_min[chan], pos[2][chan]);
 159
 160                         bbox_max[chan] = ac_build_fmax(ctx, pos[0][chan], pos[1][chan]);
 161                         bbox_max[chan] = ac_build_fmax(ctx, bbox_max[chan], pos[2][chan]);
 162                 }
 163
 164                 /* View culling. */
 165                 if (cull_view_xy || cull_view_near_z || cull_view_far_z) {
 166                         for (unsigned chan = 0; chan < 3; chan++) {
 167                                 LLVMValueRef visible;
 168
 169                                 if ((cull_view_xy && chan <= 1) ||
 170                                     (cull_view_near_z && chan == 2)) {
 171                                         float t = chan == 2 && use_halfz_clip_space ? 0 : -1;
 172                                         visible = LLVMBuildFCmp(builder, LLVMRealOGE, bbox_max[chan],
 173                                                                 LLVMConstReal(ctx->f32, t), "");
 174                                         accepted = LLVMBuildAnd(builder, accepted, visible, "");
 175                                 }
 176
 177                                 if ((cull_view_xy && chan <= 1) ||
 178                                     (cull_view_far_z && chan == 2)) {
 179                                         visible = LLVMBuildFCmp(builder, LLVMRealOLE, bbox_min[chan],
 180                                                                 ctx->f32_1, "");
 181                                         accepted = LLVMBuildAnd(builder, accepted, visible, "");
 182                                 }
 183                         }
 184                 }
 185
 186                 /* Small primitive elimination. */
 187                 if (cull_small_prims) {
 188                         /* Assuming a sample position at (0.5, 0.5), if we round
 189                          * the bounding box min/max extents and the results of
 190                          * the rounding are equal in either the X or Y direction,
 191                          * the bounding box does not intersect the sample.
 192                          *
 193                          * See these GDC slides for pictures:
 194                          * https://frostbite-wp-prd.s3.amazonaws.com/wp-content/uploads/2016/03/29204330/GDC_2016_Compute.pdf
 195                          */
 196                         LLVMValueRef min, max, not_equal[2], visible;
 197
 198                         for (unsigned chan = 0; chan < 2; chan++) {
 199                                 /* Convert the position to screen-space coordinates. */
 200                                 min = ac_build_fmad(ctx, bbox_min[chan],
 201                                                     vp_scale[chan], vp_translate[chan]);
 202                                 max = ac_build_fmad(ctx, bbox_max[chan],
 203                                                     vp_scale[chan], vp_translate[chan]);
 204                                 /* Scale the bounding box according to the precision of
 205                                  * the rasterizer and the number of MSAA samples. */
 206                                 min = LLVMBuildFSub(builder, min, small_prim_precision, "");
 207                                 max = LLVMBuildFAdd(builder, max, small_prim_precision, "");
 208
 209                                 /* Determine if the bbox intersects the sample point.
 210                                  * It also works for MSAA, but vp_scale, vp_translate,
 211                                  * and small_prim_precision are computed differently.
 212                                  */
 213                                 min = ac_build_round(ctx, min);
 214                                 max = ac_build_round(ctx, max);
 215                                 not_equal[chan] = LLVMBuildFCmp(builder, LLVMRealONE, min, max, "");
 216                         }
 217                         visible = LLVMBuildAnd(builder, not_equal[0], not_equal[1], "");
 218                         accepted = LLVMBuildAnd(builder, accepted, visible, "");
 219                 }
 220
 221                 LLVMBuildStore(builder, accepted, accepted_var);
 222         }
 223         ac_build_endif(ctx, 10000000);
 224
 225         return LLVMBuildLoad(builder, accepted_var, "");
 226 }
 227
 228 /**
 229  * Return i1 true if the primitive is accepted (not culled).
 230  *
 231  * \param pos                   Vertex positions 3x vec4
 232  * \param initially_accepted    AND'ed with the result. Some computations can be
 233  *                              skipped if this is false.
 234  * \param vp_scale              Viewport scale XY.
 235  *                              For MSAA, multiply them by the number of samples.
 236  * \param vp_translate          Viewport translation XY.
 237  *                              For MSAA, multiply them by the number of samples.
 238  * \param small_prim_precision  Precision of small primitive culling. This should
 239  *                              be the same as or greater than the precision of
 240  *                              the rasterizer. Set to num_samples / 2^subpixel_bits.
 241  *                              subpixel_bits are defined by the quantization mode.
 242  * \param options               See ac_cull_options.
 243  */
 244 LLVMValueRef ac_cull_triangle(struct ac_llvm_context *ctx,
 245                               LLVMValueRef pos[3][4],
 246                               LLVMValueRef initially_accepted,
 247                               LLVMValueRef vp_scale[2],
 248                               LLVMValueRef vp_translate[2],
 249                               LLVMValueRef small_prim_precision,
 250                               struct ac_cull_options *options)
 251 {
 252         struct ac_position_w_info w;
 253         ac_analyze_position_w(ctx, pos, &w);
 254
 255         /* W culling. */
 256         LLVMValueRef accepted = options->cull_w ? w.w_accepted : ctx->i1true;
 257         accepted = LLVMBuildAnd(ctx->builder, accepted, initially_accepted, "");
 258
 259         /* Face culling. */
 260         accepted = LLVMBuildAnd(ctx->builder, accepted,
 261                                 ac_cull_face(ctx, pos, &w,
 262                                              options->cull_front,
 263                                              options->cull_back,
 264                                              options->cull_zero_area), "");
 265
 266         /* View culling and small primitive elimination. */
 267         accepted = cull_bbox(ctx, pos, accepted, &w, vp_scale, vp_translate,
 268                              small_prim_precision,
 269                              options->cull_view_xy,
 270                              options->cull_view_near_z,
 271                              options->cull_view_far_z,
 272                              options->cull_small_prims,
 273                              options->use_halfz_clip_space);
 274         return accepted;
 275 }