1bf215546Sopenharmony_ci/*
2bf215546Sopenharmony_ci * Copyright © 2020 Mike Blumenkrantz
3bf215546Sopenharmony_ci *
4bf215546Sopenharmony_ci * Permission is hereby granted, free of charge, to any person obtaining a
5bf215546Sopenharmony_ci * copy of this software and associated documentation files (the "Software"),
6bf215546Sopenharmony_ci * to deal in the Software without restriction, including without limitation
7bf215546Sopenharmony_ci * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8bf215546Sopenharmony_ci * and/or sell copies of the Software, and to permit persons to whom the
9bf215546Sopenharmony_ci * Software is furnished to do so, subject to the following conditions:
10bf215546Sopenharmony_ci *
11bf215546Sopenharmony_ci * The above copyright notice and this permission notice (including the next
12bf215546Sopenharmony_ci * paragraph) shall be included in all copies or substantial portions of the
13bf215546Sopenharmony_ci * Software.
14bf215546Sopenharmony_ci *
15bf215546Sopenharmony_ci * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16bf215546Sopenharmony_ci * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17bf215546Sopenharmony_ci * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18bf215546Sopenharmony_ci * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19bf215546Sopenharmony_ci * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20bf215546Sopenharmony_ci * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21bf215546Sopenharmony_ci * IN THE SOFTWARE.
22bf215546Sopenharmony_ci *
23bf215546Sopenharmony_ci * Authors:
24bf215546Sopenharmony_ci *    Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
25bf215546Sopenharmony_ci */
26bf215546Sopenharmony_ci
27bf215546Sopenharmony_ci#include "nir.h"
28bf215546Sopenharmony_ci#include "nir_builder.h"
29bf215546Sopenharmony_ci
30bf215546Sopenharmony_ci
31bf215546Sopenharmony_ci/**
32bf215546Sopenharmony_ci * This pass uses the enabled clip planes from the rasterizer state to rewrite
33bf215546Sopenharmony_ci * vertex shader store operations and store a 0 to the corresponding gl_ClipDistance[n]
34bf215546Sopenharmony_ci * value if the plane is disabled
35bf215546Sopenharmony_ci */
36bf215546Sopenharmony_ci
37bf215546Sopenharmony_ci/* recursively nest if/else blocks until we get to an array index,
38bf215546Sopenharmony_ci * then overwrite it if that plane isn't enabled
39bf215546Sopenharmony_ci */
40bf215546Sopenharmony_cistatic void
41bf215546Sopenharmony_cirecursive_if_chain(nir_builder *b, nir_deref_instr *deref, nir_ssa_def *value, unsigned clip_plane_enable, nir_ssa_def *index, unsigned start, unsigned end)
42bf215546Sopenharmony_ci{
43bf215546Sopenharmony_ci   if (start == end - 1) {
44bf215546Sopenharmony_ci      /* store the original value again if the clip plane is enabled */
45bf215546Sopenharmony_ci      if (clip_plane_enable & (1 << start))
46bf215546Sopenharmony_ci         nir_store_deref(b, deref, value, 1 << start);
47bf215546Sopenharmony_ci      else
48bf215546Sopenharmony_ci         nir_store_deref(b, deref, nir_imm_int(b, 0), 1 << start);
49bf215546Sopenharmony_ci      return;
50bf215546Sopenharmony_ci   }
51bf215546Sopenharmony_ci
52bf215546Sopenharmony_ci   unsigned mid = start + (end - start) / 2;
53bf215546Sopenharmony_ci   nir_push_if(b, nir_ilt(b, index, nir_imm_int(b, mid)));
54bf215546Sopenharmony_ci   recursive_if_chain(b, deref, value, clip_plane_enable, index, start, mid);
55bf215546Sopenharmony_ci   nir_push_else(b, NULL);
56bf215546Sopenharmony_ci   recursive_if_chain(b, deref, value, clip_plane_enable, index, mid, end);
57bf215546Sopenharmony_ci   nir_pop_if(b, NULL);
58bf215546Sopenharmony_ci}
59bf215546Sopenharmony_ci
60bf215546Sopenharmony_ci/* vulkan (and some drivers) provides no concept of enabling clip planes through api,
61bf215546Sopenharmony_ci * so we rewrite disabled clip planes to a zero value in order to disable them
62bf215546Sopenharmony_ci */
63bf215546Sopenharmony_cistatic bool
64bf215546Sopenharmony_cilower_clip_plane_store(nir_intrinsic_instr *instr, unsigned clip_plane_enable, nir_builder *b)
65bf215546Sopenharmony_ci{
66bf215546Sopenharmony_ci   nir_variable *out;
67bf215546Sopenharmony_ci   unsigned plane;
68bf215546Sopenharmony_ci
69bf215546Sopenharmony_ci   if (instr->intrinsic != nir_intrinsic_store_deref)
70bf215546Sopenharmony_ci      return false;
71bf215546Sopenharmony_ci
72bf215546Sopenharmony_ci   nir_deref_instr *deref = nir_src_as_deref(instr->src[0]);
73bf215546Sopenharmony_ci
74bf215546Sopenharmony_ci   out = nir_deref_instr_get_variable(deref);
75bf215546Sopenharmony_ci   if ((out->data.location != VARYING_SLOT_CLIP_DIST0 &&
76bf215546Sopenharmony_ci        out->data.location != VARYING_SLOT_CLIP_DIST1) ||
77bf215546Sopenharmony_ci        out->data.mode != nir_var_shader_out)
78bf215546Sopenharmony_ci      return false;
79bf215546Sopenharmony_ci
80bf215546Sopenharmony_ci   b->cursor = nir_after_instr(&instr->instr);
81bf215546Sopenharmony_ci   if (deref->deref_type == nir_deref_type_var) {
82bf215546Sopenharmony_ci      int wrmask = nir_intrinsic_write_mask(instr);
83bf215546Sopenharmony_ci
84bf215546Sopenharmony_ci      nir_ssa_def *components[4];
85bf215546Sopenharmony_ci      int start = out->data.location == VARYING_SLOT_CLIP_DIST1 ? 4 : 0;
86bf215546Sopenharmony_ci      /* rewrite components as zeroes for planes that aren't enabled */
87bf215546Sopenharmony_ci      for (int i = 0; i < 4; i++) {
88bf215546Sopenharmony_ci         if (wrmask & (1 << i)) {
89bf215546Sopenharmony_ci            if (!(clip_plane_enable & (1 << (start + i))))
90bf215546Sopenharmony_ci               components[i] = nir_imm_int(b, 0);
91bf215546Sopenharmony_ci            else
92bf215546Sopenharmony_ci               components[i] = nir_channel(b, nir_ssa_for_src(b, instr->src[1], nir_src_num_components(instr->src[1])), i);
93bf215546Sopenharmony_ci         } else
94bf215546Sopenharmony_ci            components[i] = nir_ssa_undef(b, 1, 32);
95bf215546Sopenharmony_ci      }
96bf215546Sopenharmony_ci      nir_store_deref(b, deref, nir_vec(b, components, instr->num_components), wrmask);
97bf215546Sopenharmony_ci   } else  if (nir_src_is_const(deref->arr.index)) {
98bf215546Sopenharmony_ci      /* storing using a constant index */
99bf215546Sopenharmony_ci      plane = nir_src_as_uint(deref->arr.index);
100bf215546Sopenharmony_ci      /* no need to make changes if the clip plane is enabled */
101bf215546Sopenharmony_ci      if (clip_plane_enable & (1 << plane))
102bf215546Sopenharmony_ci         return false;
103bf215546Sopenharmony_ci
104bf215546Sopenharmony_ci      assert(nir_intrinsic_write_mask(instr) == 1);
105bf215546Sopenharmony_ci      nir_store_deref(b, deref, nir_imm_int(b, 0), 1);
106bf215546Sopenharmony_ci   } else {
107bf215546Sopenharmony_ci      /* storing using a variable index */
108bf215546Sopenharmony_ci      nir_ssa_def *index = nir_ssa_for_src(b, deref->arr.index, 1);
109bf215546Sopenharmony_ci      unsigned length = glsl_get_length(nir_deref_instr_parent(deref)->type);
110bf215546Sopenharmony_ci
111bf215546Sopenharmony_ci      recursive_if_chain(b, deref, instr->src[1].ssa, clip_plane_enable, index, 0, length);
112bf215546Sopenharmony_ci   }
113bf215546Sopenharmony_ci   nir_instr_remove(&instr->instr);
114bf215546Sopenharmony_ci   return true;
115bf215546Sopenharmony_ci}
116bf215546Sopenharmony_ci
117bf215546Sopenharmony_cibool
118bf215546Sopenharmony_cinir_lower_clip_disable(nir_shader *shader, unsigned clip_plane_enable)
119bf215546Sopenharmony_ci{
120bf215546Sopenharmony_ci   bool progress = false;
121bf215546Sopenharmony_ci
122bf215546Sopenharmony_ci   /* if all user planes are enabled in API that are written in the array, always ignore;
123bf215546Sopenharmony_ci    * this explicitly covers the 2x vec4 case
124bf215546Sopenharmony_ci    */
125bf215546Sopenharmony_ci   if (clip_plane_enable == u_bit_consecutive(0, shader->info.clip_distance_array_size))
126bf215546Sopenharmony_ci      return false;
127bf215546Sopenharmony_ci
128bf215546Sopenharmony_ci   nir_foreach_function(function, shader) {
129bf215546Sopenharmony_ci      if (function->impl) {
130bf215546Sopenharmony_ci         nir_builder builder;
131bf215546Sopenharmony_ci         nir_builder_init(&builder, function->impl);
132bf215546Sopenharmony_ci         nir_foreach_block(block, function->impl) {
133bf215546Sopenharmony_ci            nir_foreach_instr_safe(instr, block) {
134bf215546Sopenharmony_ci               if (instr->type == nir_instr_type_intrinsic)
135bf215546Sopenharmony_ci                  progress |= lower_clip_plane_store(nir_instr_as_intrinsic(instr),
136bf215546Sopenharmony_ci                                                     clip_plane_enable,
137bf215546Sopenharmony_ci                                                     &builder);
138bf215546Sopenharmony_ci            }
139bf215546Sopenharmony_ci         }
140bf215546Sopenharmony_ci
141bf215546Sopenharmony_ci         nir_metadata_preserve(function->impl, nir_metadata_block_index | nir_metadata_dominance);
142bf215546Sopenharmony_ci      }
143bf215546Sopenharmony_ci   }
144bf215546Sopenharmony_ci
145bf215546Sopenharmony_ci   return progress;
146bf215546Sopenharmony_ci}
147