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
  28 #include <llvm-c/Core.h>
  29
  30 struct ac_position_w_info {
  31    /* If a primitive intersects the W=0 plane, it causes a reflection
  32     * of the determinant used for face culling. Every vertex behind
  33     * the W=0 plane negates the determinant, so having 2 vertices behind
  34     * the plane has no effect. This is i1 true if the determinant should be
  35     * negated.
  36     */
  37    LLVMValueRef w_reflection;
  38
  39    /* If we simplify the "-w <= p <= w" view culling equation, we get
  40     * "-w <= w", which can't be satisfied when w is negative.
  41     * In perspective projection, a negative W means that the primitive
  42     * is behind the viewer, but the equation is independent of the type
  43     * of projection.
  44     *
  45     * w_accepted is false when all W are negative and therefore
  46     * the primitive is invisible.
  47     */
  48    LLVMValueRef w_accepted;
  49
  50    LLVMValueRef all_w_positive;
  51    LLVMValueRef any_w_negative;
  52 };
  53
  54 static void ac_analyze_position_w(struct ac_llvm_context *ctx, 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, LLVMValueRef pos[3][4],
  78                                  struct ac_position_w_info *w, bool cull_front, bool cull_back,
  79                                  bool cull_zero_area)
  80 {
  81    LLVMBuilderRef builder = ctx->builder;
  82
  83    if (cull_front && cull_back)
  84       return ctx->i1false;
  85
  86    if (!cull_front && !cull_back && !cull_zero_area)
  87       return ctx->i1true;
  88
  89    /* Front/back face culling. Also if the determinant == 0, the triangle
  90     * area is 0.
  91     */
  92    LLVMValueRef det_t0 = LLVMBuildFSub(builder, pos[2][0], pos[0][0], "");
  93    LLVMValueRef det_t1 = LLVMBuildFSub(builder, pos[1][1], pos[0][1], "");
  94    LLVMValueRef det_t2 = LLVMBuildFSub(builder, pos[0][0], pos[1][0], "");
  95    LLVMValueRef det_t3 = LLVMBuildFSub(builder, pos[0][1], pos[2][1], "");
  96    LLVMValueRef det_p0 = LLVMBuildFMul(builder, det_t0, det_t1, "");
  97    LLVMValueRef det_p1 = LLVMBuildFMul(builder, det_t2, det_t3, "");
  98    LLVMValueRef det = LLVMBuildFSub(builder, det_p0, det_p1, "");
  99
 100    /* Negative W negates the determinant. */
 101    det = LLVMBuildSelect(builder, w->w_reflection, LLVMBuildFNeg(builder, det, ""), det, "");
 102
 103    LLVMValueRef accepted = NULL;
 104    if (cull_front) {
 105       LLVMRealPredicate cond = cull_zero_area ? LLVMRealOGT : LLVMRealOGE;
 106       accepted = LLVMBuildFCmp(builder, cond, det, ctx->f32_0, "");
 107    } else if (cull_back) {
 108       LLVMRealPredicate cond = cull_zero_area ? LLVMRealOLT : LLVMRealOLE;
 109       accepted = LLVMBuildFCmp(builder, cond, det, ctx->f32_0, "");
 110    } else if (cull_zero_area) {
 111       accepted = LLVMBuildFCmp(builder, LLVMRealONE, det, ctx->f32_0, "");
 112    }
 113    return accepted;
 114 }
 115
 116 /* Perform view culling and small primitive elimination and return true
 117  * if the primitive is accepted and initially_accepted == true. */
 118 static LLVMValueRef cull_bbox(struct ac_llvm_context *ctx, LLVMValueRef pos[3][4],
 119                               LLVMValueRef initially_accepted, struct ac_position_w_info *w,
 120                               LLVMValueRef vp_scale[2], LLVMValueRef vp_translate[2],
 121                               LLVMValueRef small_prim_precision, bool cull_view_xy,
 122                               bool cull_view_near_z, bool cull_view_far_z, bool cull_small_prims,
 123                               bool use_halfz_clip_space)
 124 {
 125    LLVMBuilderRef builder = ctx->builder;
 126
 127    if (!cull_view_xy && !cull_view_near_z && !cull_view_far_z && !cull_small_prims)
 128       return initially_accepted;
 129
 130    /* Skip the culling if the primitive has already been rejected or
 131     * if any W is negative. The bounding box culling doesn't work when
 132     * W is negative.
 133     */
 134    LLVMValueRef cond = LLVMBuildAnd(builder, initially_accepted, w->all_w_positive, "");
 135    LLVMValueRef accepted_var = ac_build_alloca_undef(ctx, ctx->i1, "");
 136    LLVMBuildStore(builder, initially_accepted, accepted_var);
 137
 138    ac_build_ifcc(ctx, cond, 10000000 /* does this matter? */);
 139    {
 140       LLVMValueRef bbox_min[3], bbox_max[3];
 141       LLVMValueRef accepted = initially_accepted;
 142
 143       /* Compute the primitive bounding box for easy culling. */
 144       for (unsigned chan = 0; chan < (cull_view_near_z || cull_view_far_z ? 3 : 2); chan++) {
 145          bbox_min[chan] = ac_build_fmin(ctx, pos[0][chan], pos[1][chan]);
 146          bbox_min[chan] = ac_build_fmin(ctx, bbox_min[chan], pos[2][chan]);
 147
 148          bbox_max[chan] = ac_build_fmax(ctx, pos[0][chan], pos[1][chan]);
 149          bbox_max[chan] = ac_build_fmax(ctx, bbox_max[chan], pos[2][chan]);
 150       }
 151
 152       /* View culling. */
 153       if (cull_view_xy || cull_view_near_z || cull_view_far_z) {
 154          for (unsigned chan = 0; chan < 3; chan++) {
 155             LLVMValueRef visible;
 156
 157             if ((cull_view_xy && chan <= 1) || (cull_view_near_z && chan == 2)) {
 158                float t = chan == 2 && use_halfz_clip_space ? 0 : -1;
 159                visible = LLVMBuildFCmp(builder, LLVMRealOGE, bbox_max[chan],
 160                                        LLVMConstReal(ctx->f32, t), "");
 161                accepted = LLVMBuildAnd(builder, accepted, visible, "");
 162             }
 163
 164             if ((cull_view_xy && chan <= 1) || (cull_view_far_z && chan == 2)) {
 165                visible = LLVMBuildFCmp(builder, LLVMRealOLE, bbox_min[chan], ctx->f32_1, "");
 166                accepted = LLVMBuildAnd(builder, accepted, visible, "");
 167             }
 168          }
 169       }
 170
 171       /* Small primitive elimination. */
 172       if (cull_small_prims) {
 173          /* Assuming a sample position at (0.5, 0.5), if we round
 174           * the bounding box min/max extents and the results of
 175           * the rounding are equal in either the X or Y direction,
 176           * the bounding box does not intersect the sample.
 177           *
 178           * See these GDC slides for pictures:
 179           * https://frostbite-wp-prd.s3.amazonaws.com/wp-content/uploads/2016/03/29204330/GDC_2016_Compute.pdf
 180           */
 181          LLVMValueRef min, max, not_equal[2], visible;
 182
 183          for (unsigned chan = 0; chan < 2; chan++) {
 184             /* Convert the position to screen-space coordinates. */
 185             min = ac_build_fmad(ctx, bbox_min[chan], vp_scale[chan], vp_translate[chan]);
 186             max = ac_build_fmad(ctx, bbox_max[chan], vp_scale[chan], vp_translate[chan]);
 187             /* Scale the bounding box according to the precision of
 188              * the rasterizer and the number of MSAA samples. */
 189             min = LLVMBuildFSub(builder, min, small_prim_precision, "");
 190             max = LLVMBuildFAdd(builder, max, small_prim_precision, "");
 191
 192             /* Determine if the bbox intersects the sample point.
 193              * It also works for MSAA, but vp_scale, vp_translate,
 194              * and small_prim_precision are computed differently.
 195              */
 196             min = ac_build_round(ctx, min);
 197             max = ac_build_round(ctx, max);
 198             not_equal[chan] = LLVMBuildFCmp(builder, LLVMRealONE, min, max, "");
 199          }
 200          visible = LLVMBuildAnd(builder, not_equal[0], not_equal[1], "");
 201          accepted = LLVMBuildAnd(builder, accepted, visible, "");
 202       }
 203
 204       LLVMBuildStore(builder, accepted, accepted_var);
 205    }
 206    ac_build_endif(ctx, 10000000);
 207
 208    return LLVMBuildLoad(builder, accepted_var, "");
 209 }
 210
 211 /**
 212  * Return i1 true if the primitive is accepted (not culled).
 213  *
 214  * \param pos                   Vertex positions 3x vec4
 215  * \param initially_accepted    AND'ed with the result. Some computations can be
 216  *                              skipped if this is false.
 217  * \param vp_scale              Viewport scale XY.
 218  *                              For MSAA, multiply them by the number of samples.
 219  * \param vp_translate          Viewport translation XY.
 220  *                              For MSAA, multiply them by the number of samples.
 221  * \param small_prim_precision  Precision of small primitive culling. This should
 222  *                              be the same as or greater than the precision of
 223  *                              the rasterizer. Set to num_samples / 2^subpixel_bits.
 224  *                              subpixel_bits are defined by the quantization mode.
 225  * \param options               See ac_cull_options.
 226  */
 227 LLVMValueRef ac_cull_triangle(struct ac_llvm_context *ctx, LLVMValueRef pos[3][4],
 228                               LLVMValueRef initially_accepted, LLVMValueRef vp_scale[2],
 229                               LLVMValueRef vp_translate[2], LLVMValueRef small_prim_precision,
 230                               struct ac_cull_options *options)
 231 {
 232    struct ac_position_w_info w;
 233    ac_analyze_position_w(ctx, pos, &w);
 234
 235    /* W culling. */
 236    LLVMValueRef accepted = options->cull_w ? w.w_accepted : ctx->i1true;
 237    accepted = LLVMBuildAnd(ctx->builder, accepted, initially_accepted, "");
 238
 239    /* Face culling. */
 240    accepted = LLVMBuildAnd(
 241       ctx->builder, accepted,
 242       ac_cull_face(ctx, pos, &w, options->cull_front, options->cull_back, options->cull_zero_area),
 243       "");
 244
 245    /* View culling and small primitive elimination. */
 246    accepted = cull_bbox(ctx, pos, accepted, &w, vp_scale, vp_translate, small_prim_precision,
 247                         options->cull_view_xy, options->cull_view_near_z, options->cull_view_far_z,
 248                         options->cull_small_prims, options->use_halfz_clip_space);
 249    return accepted;
 250 }