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
|
/*========================== begin_copyright_notice ============================
Copyright (C) 2025 Intel Corporation
SPDX-License-Identifier: MIT
============================= end_copyright_notice ===========================*/
#pragma once
#include "common/LLVMWarningsPush.hpp"
#include <llvm/IR/BasicBlock.h>
#include <llvm/IR/Dominators.h>
#include <llvm/IR/Function.h>
#include <llvm/IR/IRBuilder.h>
#include <llvm/Pass.h>
#include "common/LLVMWarningsPop.hpp"
#include "Compiler/CodeGenContextWrapper.hpp"
namespace IGC {
// This pass should be run after GVN. It searches for the patterns, where
// 1. There is a PHI instruction with the "same" incoming value from all
// incoming blocks
// - Which means there is common dominator for all predecessors which define
// the incoming values
// 2. Incoming value is limited for now to conversion operations, but can be
// extended in the future, if needed. Patterns like that are created by GVN's
// PRE and are exposed, when gid is used as an index for load and store. Source
// example:
// val = buffer[gid];
// outputBuffer[gid] = val;
// IR example:
// entry:
// %4 = add i32 %3, %payloadHeader.scalar
// br i1 %6, label %if.then, label %entry.if.end_crit_edge
// entry.if.end_crit_edge:
// %.pre = sext i32 %4 to i64
// br label %if.end
// if.then:
// %idxprom2 = sext i32 %4 to i64
// load
// br label %if.end
// if.end:
// %idxprom4.pre-phi = phi i64 [ %.pre, %entry.if.end_crit_edge ], [
// %idxprom2, %if.then ] store
//
// If such pattern is found, optimization:
// 1. Hoists the value '%.pre' to the common dominator
// 2. Replaces all uses of '%idxprom2' with '%.pre'
// 3. Replaces all uses of '%idxprom4.pre-phi' with '%.pre'
// 4. Removes '%idxprom2' and '%idxprom4.pre-phi'
// Result should be:
// entry:
// %4 = add i32 %3, %payloadHeader.scalar
// %.pre = sext i32 %4 to i64
// br i1 %6, label %if.then, label %entry.if.end_crit_edge
// entry.if.end_crit_edge:
// br label %if.end
// if.then:
// load
// br label %if.end
// if.end:
// store
// Then, SimplifyCFG is expected to run to remove `entry.if.end_crit_edge` block
// Then PromoteToPredicatedMemoryAccess pass can try to convert memory
// operations to predicated memory operations
class HoistConvOpToDom : public llvm::FunctionPass {
public:
static char ID;
explicit HoistConvOpToDom();
~HoistConvOpToDom() {}
llvm::StringRef getPassName() const override { return "HoistConvOpToDom"; }
virtual void getAnalysisUsage(llvm::AnalysisUsage &AU) const override {
AU.addRequired<CodeGenContextWrapper>();
AU.addRequired<llvm::DominatorTreeWrapperPass>();
}
bool runOnFunction(llvm::Function &F) override;
private:
llvm::BasicBlock *findNearestCommonDominator(const llvm::PHINode &PHI) const;
CodeGenContext *m_CGCtx = nullptr;
llvm::DominatorTree *m_DT = nullptr;
};
} // namespace IGC
|
