[Alignment][NFC] Use Align with TargetLowering::setMinFunctionAlignment
[llvm-core.git] / include / llvm / MCA / Support.h
blob1da097c909227e66abe283cde0974f2ee1f7ccde
1 //===--------------------- Support.h ----------------------------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 /// \file
9 ///
10 /// Helper functions used by various pipeline components.
11 ///
12 //===----------------------------------------------------------------------===//
14 #ifndef LLVM_MCA_SUPPORT_H
15 #define LLVM_MCA_SUPPORT_H
17 #include "llvm/ADT/ArrayRef.h"
18 #include "llvm/ADT/SmallVector.h"
19 #include "llvm/MC/MCSchedule.h"
20 #include "llvm/Support/Error.h"
22 namespace llvm {
23 namespace mca {
25 template <typename T>
26 class InstructionError : public ErrorInfo<InstructionError<T>> {
27 public:
28 static char ID;
29 std::string Message;
30 const T &Inst;
32 InstructionError(std::string M, const T &MCI)
33 : Message(std::move(M)), Inst(MCI) {}
35 void log(raw_ostream &OS) const override { OS << Message; }
37 std::error_code convertToErrorCode() const override {
38 return inconvertibleErrorCode();
42 template <typename T> char InstructionError<T>::ID;
44 /// This class represents the number of cycles per resource (fractions of
45 /// cycles). That quantity is managed here as a ratio, and accessed via the
46 /// double cast-operator below. The two quantities, number of cycles and
47 /// number of resources, are kept separate. This is used by the
48 /// ResourcePressureView to calculate the average resource cycles
49 /// per instruction/iteration.
50 class ResourceCycles {
51 unsigned Numerator, Denominator;
53 public:
54 ResourceCycles() : Numerator(0), Denominator(1) {}
55 ResourceCycles(unsigned Cycles, unsigned ResourceUnits = 1)
56 : Numerator(Cycles), Denominator(ResourceUnits) {}
58 operator double() const {
59 assert(Denominator && "Invalid denominator (must be non-zero).");
60 return (Denominator == 1) ? Numerator : (double)Numerator / Denominator;
63 unsigned getNumerator() const { return Numerator; }
64 unsigned getDenominator() const { return Denominator; }
66 // Add the components of RHS to this instance. Instead of calculating
67 // the final value here, we keep track of the numerator and denominator
68 // separately, to reduce floating point error.
69 ResourceCycles &operator+=(const ResourceCycles &RHS);
72 /// Populates vector Masks with processor resource masks.
73 ///
74 /// The number of bits set in a mask depends on the processor resource type.
75 /// Each processor resource mask has at least one bit set. For groups, the
76 /// number of bits set in the mask is equal to the cardinality of the group plus
77 /// one. Excluding the most significant bit, the remaining bits in the mask
78 /// identify processor resources that are part of the group.
79 ///
80 /// Example:
81 ///
82 /// ResourceA -- Mask: 0b001
83 /// ResourceB -- Mask: 0b010
84 /// ResourceAB -- Mask: 0b100 U (ResourceA::Mask | ResourceB::Mask) == 0b111
85 ///
86 /// ResourceAB is a processor resource group containing ResourceA and ResourceB.
87 /// Each resource mask uniquely identifies a resource; both ResourceA and
88 /// ResourceB only have one bit set.
89 /// ResourceAB is a group; excluding the most significant bit in the mask, the
90 /// remaining bits identify the composition of the group.
91 ///
92 /// Resource masks are used by the ResourceManager to solve set membership
93 /// problems with simple bit manipulation operations.
94 void computeProcResourceMasks(const MCSchedModel &SM,
95 MutableArrayRef<uint64_t> Masks);
97 // Returns the index of the highest bit set. For resource masks, the position of
98 // the highest bit set can be used to construct a resource mask identifier.
99 inline unsigned getResourceStateIndex(uint64_t Mask) {
100 assert(Mask && "Processor Resource Mask cannot be zero!");
101 return (std::numeric_limits<uint64_t>::digits - countLeadingZeros(Mask)) - 1;
104 /// Compute the reciprocal block throughput from a set of processor resource
105 /// cycles. The reciprocal block throughput is computed as the MAX between:
106 /// - NumMicroOps / DispatchWidth
107 /// - ProcResourceCycles / #ProcResourceUnits (for every consumed resource).
108 double computeBlockRThroughput(const MCSchedModel &SM, unsigned DispatchWidth,
109 unsigned NumMicroOps,
110 ArrayRef<unsigned> ProcResourceUsage);
111 } // namespace mca
112 } // namespace llvm
114 #endif // LLVM_MCA_SUPPORT_H