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
|
/*========================== begin_copyright_notice ============================
Copyright (C) 2020-2021 Intel Corporation
SPDX-License-Identifier: MIT
============================= end_copyright_notice ===========================*/
#pragma once
#include "Compiler/MetaDataUtilsWrapper.h"
#include "Compiler/CodeGenContextWrapper.hpp"
#include "Compiler/ModuleAllocaAnalysis.hpp"
#include <llvm/Pass.h>
using namespace llvm;
namespace IGC
{
// It is convenient to represent the null pointer as the zero
// bit-pattern. However, SLM address 0 is legal, and we want to be able
// to use it.
// To go around this, we set all valid pointers to have a non-zero high
// nibble.
constexpr unsigned int VALID_LOCAL_HIGH_BITS = 0x10000000;
constexpr unsigned int LOW_BITS_MASK = VALID_LOCAL_HIGH_BITS - 1;
// Experimental pass to move private memory allocations to SLM where it's
// profitable. The pass is able to handle Compute and OpenCL shader types.
class PrivateMemoryToSLM : public ModulePass
{
public:
static char ID;
PrivateMemoryToSLM(bool enableOptReport = false);
PrivateMemoryToSLM(
std::string forcedBuffers,
bool enableOptReport);
~PrivateMemoryToSLM() {}
virtual StringRef getPassName() const override
{
return "PrivateMemoryToSLMPass";
}
virtual void getAnalysisUsage(AnalysisUsage& AU) const override
{
AU.setPreservesCFG();
AU.addRequired<MetaDataUtilsWrapper>();
AU.addRequired<CodeGenContextWrapper>();
AU.addRequired<ModuleAllocaAnalysis>();
}
virtual bool runOnModule(Module& M) override;
static const unsigned int SLM_LOCAL_VARIABLE_ALIGNMENT;
static const unsigned int SLM_LOCAL_SIZE_ALIGNMENT;
private:
bool m_EnableOptReport;
bool m_ForceAll;
std::vector<std::string> m_ForcedBuffers;
};
}
|
