1/*
2 * Copyright © 2016 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24#include "compiler/nir/nir_builder.h"
25#include "brw_nir.h"
26
27/**
28 * Implements the WaPreventHSTessLevelsInterference workaround (for Gen7-8).
29 *
30 * From the Broadwell PRM, Volume 7 (3D-Media-GPGPU), Page 494 (below the
31 * definition of the patch header layouts):
32 *
33 *    "HW Bug: The Tessellation stage will incorrectly add domain points
34 *     along patch edges under the following conditions, which may result
35 *     in conformance failures and/or cracking artifacts:
36 *
37 *       * QUAD domain
38 *       * INTEGER partitioning
39 *       * All three TessFactors in a given U or V direction (e.g., V
40 *         direction: UEQ0, InsideV, UEQ1) are all exactly 1.0
41 *       * All three TessFactors in the other direction are > 1.0 and all
42 *         round up to the same integer value (e.g, U direction:
43 *         VEQ0 = 3.1, InsideU = 3.7, VEQ1 = 3.4)
44 *
45 *     The suggested workaround (to be implemented as part of the postamble
46 *     to the HS shader in the HS kernel) is:
47 *
48 *     if (
49 *        (TF[UEQ0] > 1.0) ||
50 *        (TF[VEQ0] > 1.0) ||
51 *        (TF[UEQ1] > 1.0) ||
52 *        (TF[VEQ1] > 1.0) ||
53 *        (TF[INSIDE_U] > 1.0) ||
54 *        (TF[INSIDE_V] > 1.0) )
55 *     {
56 *        TF[INSIDE_U] = (TF[INSIDE_U] == 1.0) ? 2.0 : TF[INSIDE_U];
57 *        TF[INSIDE_V] = (TF[INSIDE_V] == 1.0) ? 2.0 : TF[INSIDE_V];
58 *     }"
59 *
60 * There's a subtlety here.  Intel internal HSD-ES bug 1208668495 notes
61 * that the above workaround fails to fix certain GL/ES CTS tests which
62 * have inside tessellation factors of -1.0.  This can be explained by
63 * a quote from the ARB_tessellation_shader specification:
64 *
65 *    "If "equal_spacing" is used, the floating-point tessellation level is
66 *     first clamped to the range [1,<max>], where <max> is implementation-
67 *     dependent maximum tessellation level (MAX_TESS_GEN_LEVEL)."
68 *
69 * In other words, the actual inner tessellation factor used is
70 * clamp(TF[INSIDE_*], 1.0, 64.0).  So we want to compare the clamped
71 * value against 1.0.  To accomplish this, we change the comparison from
72 * (TF[INSIDE_*] == 1.0) to (TF[INSIDE_*] <= 1.0).
73 */
74
75static inline nir_ssa_def *
76load_output(nir_builder *b, int num_components, int offset, int component)
77{
78   nir_intrinsic_instr *load =
79      nir_intrinsic_instr_create(b->shader, nir_intrinsic_load_output);
80   nir_ssa_dest_init(&load->instr, &load->dest, num_components, 32, NULL);
81   load->num_components = num_components;
82   load->src[0] = nir_src_for_ssa(nir_imm_int(b, 0));
83   nir_intrinsic_set_base(load, offset);
84   nir_intrinsic_set_component(load, component);
85
86   nir_builder_instr_insert(b, &load->instr);
87
88   return &load->dest.ssa;
89}
90
91static void
92emit_quads_workaround(nir_builder *b, nir_block *block)
93{
94   b->cursor = nir_after_block_before_jump(block);
95
96   nir_ssa_def *inner = load_output(b, 2, 0, 2);
97   nir_ssa_def *outer = load_output(b, 4, 1, 0);
98
99   nir_ssa_def *any_greater_than_1 =
100       nir_ior(b, nir_bany(b, nir_flt(b, nir_imm_float(b, 1.0f), outer)),
101                  nir_bany(b, nir_flt(b, nir_imm_float(b, 1.0f), inner)));
102
103   nir_if *if_stmt = nir_if_create(b->shader);
104   if_stmt->condition = nir_src_for_ssa(any_greater_than_1);
105   nir_builder_cf_insert(b, &if_stmt->cf_node);
106
107   /* Fill out the new then-block */
108   b->cursor = nir_after_cf_list(&if_stmt->then_list);
109
110   inner = nir_bcsel(b, nir_fge(b, nir_imm_float(b, 1.0f), inner),
111                        nir_imm_float(b, 2.0f), inner);
112
113   nir_intrinsic_instr *store =
114      nir_intrinsic_instr_create(b->shader, nir_intrinsic_store_output);
115   store->num_components = 2;
116   nir_intrinsic_set_write_mask(store, WRITEMASK_XY);
117   nir_intrinsic_set_component(store, 2);
118   store->src[0] = nir_src_for_ssa(inner);
119   store->src[1] = nir_src_for_ssa(nir_imm_int(b, 0));
120   nir_builder_instr_insert(b, &store->instr);
121}
122
123void
124brw_nir_apply_tcs_quads_workaround(nir_shader *nir)
125{
126   assert(nir->info.stage == MESA_SHADER_TESS_CTRL);
127
128   nir_function_impl *impl = nir_shader_get_entrypoint(nir);
129
130   nir_builder b;
131   nir_builder_init(&b, impl);
132
133   /* emit_quads_workaround() inserts an if statement into each block,
134    * which splits it in two.  This changes the set of predecessors of
135    * the end block.  We want to process the original set, so to be safe,
136    * save it off to an array first.
137    */
138   const unsigned num_end_preds = impl->end_block->predecessors->entries;
139   nir_block *end_preds[num_end_preds];
140   unsigned i = 0;
141
142   set_foreach(impl->end_block->predecessors, entry) {
143      end_preds[i++] = (nir_block *) entry->key;
144   }
145
146   for (i = 0; i < num_end_preds; i++) {
147      emit_quads_workaround(&b, end_preds[i]);
148   }
149
150   nir_metadata_preserve(impl, 0);
151}
152