aboutsummaryrefslogtreecommitdiff
path: root/source/opt/block_merge_pass.cpp
blob: ad9de47ea660ca196dc6004372e081143a63aaf8 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
// Copyright (c) 2017 The Khronos Group Inc.
// Copyright (c) 2017 Valve Corporation
// Copyright (c) 2017 LunarG Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include "block_merge_pass.h"

#include "iterator.h"

namespace spvtools {
namespace opt {

namespace {

const int kEntryPointFunctionIdInIdx = 1;

} // anonymous namespace

bool BlockMergePass::IsLoopHeader(ir::BasicBlock* block_ptr) {
  auto iItr = block_ptr->tail();
  if (iItr == block_ptr->begin())
    return false;
  --iItr;
  return iItr->opcode() == SpvOpLoopMerge;
}

bool BlockMergePass::HasMultipleRefs(uint32_t labId) {
  const analysis::UseList* uses = def_use_mgr_->GetUses(labId);
  int rcnt = 0;
  for (const auto u : *uses) {
    // Don't count OpName
    if (u.inst->opcode() == SpvOpName)
      continue;
    if (rcnt == 1)
      return true;
    ++rcnt;
  }
  return false;
}

void BlockMergePass::KillInstAndName(ir::Instruction* inst) {
  const uint32_t id = inst->result_id();
  if (id != 0) {
    analysis::UseList* uses = def_use_mgr_->GetUses(id);
    if (uses != nullptr)
      for (auto u : *uses)
        if (u.inst->opcode() == SpvOpName) {
          def_use_mgr_->KillInst(u.inst);
          break;
        }
  }
  def_use_mgr_->KillInst(inst);
}

bool BlockMergePass::MergeBlocks(ir::Function* func) {
  bool modified = false;
  for (auto bi = func->begin(); bi != func->end(); ) {
    // Do not merge loop header blocks, at least for now.
    if (IsLoopHeader(&*bi)) {
      ++bi;
      continue;
    }
    // Find block with single successor which has no other predecessors.
    // Continue and Merge blocks are currently ruled out as second blocks.
    // Happily any such candidate blocks will have >1 uses due to their
    // LoopMerge instruction.
    // TODO(): Deal with phi instructions that reference the
    // second block. Happily, these references currently inhibit
    // the merge.
    auto ii = bi->end();
    --ii;
    ir::Instruction* br = &*ii;
    if (br->opcode() != SpvOpBranch) {
      ++bi;
      continue;
    }
    const uint32_t labId = br->GetSingleWordInOperand(0);
    if (HasMultipleRefs(labId)) {
      ++bi;
      continue;
    }
    // Merge blocks
    def_use_mgr_->KillInst(br);
    auto sbi = bi;
    for (; sbi != func->end(); ++sbi)
      if (sbi->id() == labId)
        break;
    // If bi is sbi's only predecessor, it dominates sbi and thus
    // sbi must follow bi in func's ordering.
    assert(sbi != func->end());
    bi->AddInstructions(&*sbi);
    KillInstAndName(sbi->GetLabelInst());
    (void) sbi.Erase();
    // reprocess block
    modified = true;
  }
  return modified;
}

void BlockMergePass::Initialize(ir::Module* module) {

  module_ = module;

  // Initialize function and block maps
  id2function_.clear();
  for (auto& fn : *module_) 
    id2function_[fn.result_id()] = &fn;

  // TODO(greg-lunarg): Reuse def/use from previous passes
  def_use_mgr_.reset(new analysis::DefUseManager(consumer(), module_));

  // Initialize extension whitelist
  InitExtensions();
};

bool BlockMergePass::AllExtensionsSupported() const {
  // If any extension not in whitelist, return false
  for (auto& ei : module_->extensions()) {
    const char* extName = reinterpret_cast<const char*>(
        &ei.GetInOperand(0).words[0]);
    if (extensions_whitelist_.find(extName) == extensions_whitelist_.end())
      return false;
  }
  return true;
}

Pass::Status BlockMergePass::ProcessImpl() {
  // Do not process if any disallowed extensions are enabled
  if (!AllExtensionsSupported())
    return Status::SuccessWithoutChange;
  bool modified = false;
  for (auto& e : module_->entry_points()) {
    ir::Function* fn =
        id2function_[e.GetSingleWordInOperand(kEntryPointFunctionIdInIdx)];
    modified = MergeBlocks(fn) || modified;
  }
  return modified ? Status::SuccessWithChange : Status::SuccessWithoutChange;
}

BlockMergePass::BlockMergePass()
    : module_(nullptr), def_use_mgr_(nullptr) {}

Pass::Status BlockMergePass::Process(ir::Module* module) {
  Initialize(module);
  return ProcessImpl();
}

void BlockMergePass::InitExtensions() {
  extensions_whitelist_.clear();
  extensions_whitelist_.insert({
    "SPV_AMD_shader_explicit_vertex_parameter",
    "SPV_AMD_shader_trinary_minmax",
    "SPV_AMD_gcn_shader",
    "SPV_KHR_shader_ballot",
    "SPV_AMD_shader_ballot",
    "SPV_AMD_gpu_shader_half_float",
    "SPV_KHR_shader_draw_parameters",
    "SPV_KHR_subgroup_vote",
    "SPV_KHR_16bit_storage",
    "SPV_KHR_device_group",
    "SPV_KHR_multiview",
    "SPV_NVX_multiview_per_view_attributes",
    "SPV_NV_viewport_array2",
    "SPV_NV_stereo_view_rendering",
    "SPV_NV_sample_mask_override_coverage",
    "SPV_NV_geometry_shader_passthrough",
    "SPV_AMD_texture_gather_bias_lod",
    "SPV_KHR_storage_buffer_storage_class",
    "SPV_KHR_variable_pointers",
    "SPV_AMD_gpu_shader_int16",
    "SPV_KHR_post_depth_coverage",
    "SPV_KHR_shader_atomic_counter_ops",
  });
}

}  // namespace opt
}  // namespace spvtools