LLVM 20.0.0git
X86InstrFMA3Info.cpp
Go to the documentation of this file.
1//===-- X86InstrFMA3Info.cpp - X86 FMA3 Instruction Information -----------===//
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//
9// This file contains the implementation of the classes providing information
10// about existing X86 FMA3 opcodes, classifying and grouping them.
11//
12//===----------------------------------------------------------------------===//
13
14#include "X86InstrFMA3Info.h"
15#include "X86InstrInfo.h"
16#include "llvm/Support/Threading.h"
17#include <atomic>
18#include <cassert>
19#include <cstdint>
20
21using namespace llvm;
22
23#define FMA3GROUP(Name, Suf, Attrs) \
24 { { X86::Name##132##Suf, X86::Name##213##Suf, X86::Name##231##Suf }, Attrs },
25
26#define FMA3GROUP_MASKED(Name, Suf, Attrs) \
27 FMA3GROUP(Name, Suf, Attrs) \
28 FMA3GROUP(Name, Suf##k, Attrs | X86InstrFMA3Group::KMergeMasked) \
29 FMA3GROUP(Name, Suf##kz, Attrs | X86InstrFMA3Group::KZeroMasked)
30
31#define FMA3GROUP_PACKED_WIDTHS_Z(Name, Suf, Attrs) \
32 FMA3GROUP_MASKED(Name, Suf##Z128m, Attrs) \
33 FMA3GROUP_MASKED(Name, Suf##Z128r, Attrs) \
34 FMA3GROUP_MASKED(Name, Suf##Z256m, Attrs) \
35 FMA3GROUP_MASKED(Name, Suf##Z256r, Attrs) \
36 FMA3GROUP_MASKED(Name, Suf##Zm, Attrs) \
37 FMA3GROUP_MASKED(Name, Suf##Zr, Attrs) \
38
39#define FMA3GROUP_PACKED_WIDTHS_ALL(Name, Suf, Attrs) \
40 FMA3GROUP(Name, Suf##Ym, Attrs) \
41 FMA3GROUP(Name, Suf##Yr, Attrs) \
42 FMA3GROUP_PACKED_WIDTHS_Z(Name, Suf, Attrs) \
43 FMA3GROUP(Name, Suf##m, Attrs) \
44 FMA3GROUP(Name, Suf##r, Attrs)
45
46#define FMA3GROUP_PACKED(Name, Attrs) \
47 FMA3GROUP_PACKED_WIDTHS_ALL(Name, PD, Attrs) \
48 FMA3GROUP_PACKED_WIDTHS_Z(Name, PH, Attrs) \
49 FMA3GROUP_PACKED_WIDTHS_ALL(Name, PS, Attrs)
50
51#define FMA3GROUP_SCALAR_WIDTHS_Z(Name, Suf, Attrs) \
52 FMA3GROUP(Name, Suf##Zm, Attrs) \
53 FMA3GROUP_MASKED(Name, Suf##Zm_Int, Attrs | X86InstrFMA3Group::Intrinsic) \
54 FMA3GROUP(Name, Suf##Zr, Attrs) \
55 FMA3GROUP_MASKED(Name, Suf##Zr_Int, Attrs | X86InstrFMA3Group::Intrinsic) \
56
57#define FMA3GROUP_SCALAR_WIDTHS_ALL(Name, Suf, Attrs) \
58 FMA3GROUP_SCALAR_WIDTHS_Z(Name, Suf, Attrs) \
59 FMA3GROUP(Name, Suf##m, Attrs) \
60 FMA3GROUP(Name, Suf##m_Int, Attrs | X86InstrFMA3Group::Intrinsic) \
61 FMA3GROUP(Name, Suf##r, Attrs) \
62 FMA3GROUP(Name, Suf##r_Int, Attrs | X86InstrFMA3Group::Intrinsic)
63
64#define FMA3GROUP_SCALAR(Name, Attrs) \
65 FMA3GROUP_SCALAR_WIDTHS_ALL(Name, SD, Attrs) \
66 FMA3GROUP_SCALAR_WIDTHS_Z(Name, SH, Attrs) \
67 FMA3GROUP_SCALAR_WIDTHS_ALL(Name, SS, Attrs)
68
69#define FMA3GROUP_FULL(Name, Attrs) \
70 FMA3GROUP_PACKED(Name, Attrs) \
71 FMA3GROUP_SCALAR(Name, Attrs)
72
73static const X86InstrFMA3Group Groups[] = {
74 FMA3GROUP_FULL(VFMADD, 0)
75 FMA3GROUP_PACKED(VFMADDSUB, 0)
76 FMA3GROUP_FULL(VFMSUB, 0)
77 FMA3GROUP_PACKED(VFMSUBADD, 0)
78 FMA3GROUP_FULL(VFNMADD, 0)
79 FMA3GROUP_FULL(VFNMSUB, 0)
80};
81
82#define FMA3GROUP_PACKED_AVX512_WIDTHS(Name, Type, Suf, Attrs) \
83 FMA3GROUP_MASKED(Name, Type##Z128##Suf, Attrs) \
84 FMA3GROUP_MASKED(Name, Type##Z256##Suf, Attrs) \
85 FMA3GROUP_MASKED(Name, Type##Z##Suf, Attrs)
86
87#define FMA3GROUP_PACKED_AVX512(Name, Suf, Attrs) \
88 FMA3GROUP_PACKED_AVX512_WIDTHS(Name, PD, Suf, Attrs) \
89 FMA3GROUP_PACKED_AVX512_WIDTHS(Name, PH, Suf, Attrs) \
90 FMA3GROUP_PACKED_AVX512_WIDTHS(Name, PS, Suf, Attrs)
91
92#define FMA3GROUP_PACKED_AVX512_ROUND(Name, Suf, Attrs) \
93 FMA3GROUP_MASKED(Name, PDZ256##Suf, Attrs) \
94 FMA3GROUP_MASKED(Name, PDZ##Suf, Attrs) \
95 FMA3GROUP_MASKED(Name, PHZ256##Suf, Attrs) \
96 FMA3GROUP_MASKED(Name, PHZ##Suf, Attrs) \
97 FMA3GROUP_MASKED(Name, PSZ256##Suf, Attrs) \
98 FMA3GROUP_MASKED(Name, PSZ##Suf, Attrs)
99
100#define FMA3GROUP_SCALAR_AVX512_ROUND(Name, Suf, Attrs) \
101 FMA3GROUP(Name, SDZ##Suf, Attrs) \
102 FMA3GROUP_MASKED(Name, SDZ##Suf##_Int, Attrs) \
103 FMA3GROUP(Name, SHZ##Suf, Attrs) \
104 FMA3GROUP_MASKED(Name, SHZ##Suf##_Int, Attrs) \
105 FMA3GROUP(Name, SSZ##Suf, Attrs) \
106 FMA3GROUP_MASKED(Name, SSZ##Suf##_Int, Attrs)
107
108static const X86InstrFMA3Group BroadcastGroups[] = {
109 FMA3GROUP_PACKED_AVX512(VFMADD, mb, 0)
110 FMA3GROUP_PACKED_AVX512(VFMADDSUB, mb, 0)
111 FMA3GROUP_PACKED_AVX512(VFMSUB, mb, 0)
112 FMA3GROUP_PACKED_AVX512(VFMSUBADD, mb, 0)
113 FMA3GROUP_PACKED_AVX512(VFNMADD, mb, 0)
114 FMA3GROUP_PACKED_AVX512(VFNMSUB, mb, 0)
115};
116
117static const X86InstrFMA3Group RoundGroups[] = {
118 FMA3GROUP_PACKED_AVX512_ROUND(VFMADD, rb, 0)
119 FMA3GROUP_SCALAR_AVX512_ROUND(VFMADD, rb, X86InstrFMA3Group::Intrinsic)
120 FMA3GROUP_PACKED_AVX512_ROUND(VFMADDSUB, rb, 0)
121 FMA3GROUP_PACKED_AVX512_ROUND(VFMSUB, rb, 0)
122 FMA3GROUP_SCALAR_AVX512_ROUND(VFMSUB, rb, X86InstrFMA3Group::Intrinsic)
123 FMA3GROUP_PACKED_AVX512_ROUND(VFMSUBADD, rb, 0)
124 FMA3GROUP_PACKED_AVX512_ROUND(VFNMADD, rb, 0)
125 FMA3GROUP_SCALAR_AVX512_ROUND(VFNMADD, rb, X86InstrFMA3Group::Intrinsic)
126 FMA3GROUP_PACKED_AVX512_ROUND(VFNMSUB, rb, 0)
127 FMA3GROUP_SCALAR_AVX512_ROUND(VFNMSUB, rb, X86InstrFMA3Group::Intrinsic)
128};
129
130static void verifyTables() {
131#ifndef NDEBUG
132 static std::atomic<bool> TableChecked(false);
133 if (!TableChecked.load(std::memory_order_relaxed)) {
134 assert(llvm::is_sorted(Groups) && llvm::is_sorted(RoundGroups) &&
135 llvm::is_sorted(BroadcastGroups) && "FMA3 tables not sorted!");
136 TableChecked.store(true, std::memory_order_relaxed);
137 }
138#endif
139}
140
141/// Returns a reference to a group of FMA3 opcodes to where the given
142/// \p Opcode is included. If the given \p Opcode is not recognized as FMA3
143/// and not included into any FMA3 group, then nullptr is returned.
144const X86InstrFMA3Group *llvm::getFMA3Group(unsigned Opcode, uint64_t TSFlags) {
145
146 // FMA3 instructions have a well defined encoding pattern we can exploit.
147 uint8_t BaseOpcode = X86II::getBaseOpcodeFor(TSFlags);
148 bool IsFMA3Opcode = ((BaseOpcode >= 0x96 && BaseOpcode <= 0x9F) ||
149 (BaseOpcode >= 0xA6 && BaseOpcode <= 0xAF) ||
150 (BaseOpcode >= 0xB6 && BaseOpcode <= 0xBF));
151 bool IsFMA3Encoding = ((TSFlags & X86II::EncodingMask) == X86II::VEX &&
152 (TSFlags & X86II::OpMapMask) == X86II::T8) ||
153 ((TSFlags & X86II::EncodingMask) == X86II::EVEX &&
154 ((TSFlags & X86II::OpMapMask) == X86II::T8 ||
155 (TSFlags & X86II::OpMapMask) == X86II::T_MAP6));
156 bool IsFMA3Prefix = (TSFlags & X86II::OpPrefixMask) == X86II::PD;
157 if (!IsFMA3Opcode || !IsFMA3Encoding || !IsFMA3Prefix)
158 return nullptr;
159
160 verifyTables();
161
162 ArrayRef<X86InstrFMA3Group> Table;
163 if (TSFlags & X86II::EVEX_RC)
164 Table = ArrayRef(RoundGroups);
165 else if (TSFlags & X86II::EVEX_B)
166 Table = ArrayRef(BroadcastGroups);
167 else
168 Table = ArrayRef(Groups);
169
170 // FMA 132 instructions have an opcode of 0x96-0x9F
171 // FMA 213 instructions have an opcode of 0xA6-0xAF
172 // FMA 231 instructions have an opcode of 0xB6-0xBF
173 unsigned FormIndex = ((BaseOpcode - 0x90) >> 4) & 0x3;
174
175 auto I = partition_point(Table, [=](const X86InstrFMA3Group &Group) {
176 return Group.Opcodes[FormIndex] < Opcode;
177 });
178 assert(I != Table.end() && I->Opcodes[FormIndex] == Opcode &&
179 "Couldn't find FMA3 opcode!");
180 return I;
181}
I
#define I(x, y, z)
Definition: MD5.cpp:58
assert
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
verifyTables
static void verifyTables()
Definition: X86InstrFMA3Info.cpp:130
FMA3GROUP_PACKED_AVX512_ROUND
#define FMA3GROUP_PACKED_AVX512_ROUND(Name, Suf, Attrs)
Definition: X86InstrFMA3Info.cpp:92
FMA3GROUP_PACKED_AVX512
#define FMA3GROUP_PACKED_AVX512(Name, Suf, Attrs)
Definition: X86InstrFMA3Info.cpp:87
FMA3GROUP_SCALAR_AVX512_ROUND
#define FMA3GROUP_SCALAR_AVX512_ROUND(Name, Suf, Attrs)
Definition: X86InstrFMA3Info.cpp:100
RoundGroups
static const X86InstrFMA3Group RoundGroups[]
Definition: X86InstrFMA3Info.cpp:117
Groups
static const X86InstrFMA3Group Groups[]
Definition: X86InstrFMA3Info.cpp:73
FMA3GROUP_PACKED
#define FMA3GROUP_PACKED(Name, Attrs)
Definition: X86InstrFMA3Info.cpp:46
BroadcastGroups
static const X86InstrFMA3Group BroadcastGroups[]
Definition: X86InstrFMA3Info.cpp:108
FMA3GROUP_FULL
#define FMA3GROUP_FULL(Name, Attrs)
Definition: X86InstrFMA3Info.cpp:69
llvm::ArrayRef
ArrayRef - Represent a constant reference to an array (0 or more elements consecutively in memory),...
Definition: ArrayRef.h:41
llvm::ArrayRef::end
iterator end() const
Definition: ArrayRef.h:154
uint64_t
llvm::X86II::getBaseOpcodeFor
uint8_t getBaseOpcodeFor(uint64_t TSFlags)
Definition: X86BaseInfo.h:893
llvm::X86II::EVEX
@ EVEX
EVEX - Specifies that this instruction use EVEX form which provides syntax support up to 32 512-bit r...
Definition: X86BaseInfo.h:825
llvm::X86II::VEX
@ VEX
VEX - encoding using 0xC4/0xC5.
Definition: X86BaseInfo.h:818
llvm::X86II::PD
@ PD
PD - Prefix code for packed double precision vector floating point operations performed in the SSE re...
Definition: X86BaseInfo.h:721
llvm::X86II::T8
@ T8
T8, TA - Prefix after the 0x0F prefix.
Definition: X86BaseInfo.h:737
llvm
This is an optimization pass for GlobalISel generic memory operations.
Definition: AddressRanges.h:18
llvm::partition_point
auto partition_point(R &&Range, Predicate P)
Binary search for the first iterator in a range where a predicate is false.
Definition: STLExtras.h:2033
llvm::getFMA3Group
const X86InstrFMA3Group * getFMA3Group(unsigned Opcode, uint64_t TSFlags)
Returns a reference to a group of FMA3 opcodes to where the given Opcode is included.
Definition: X86InstrFMA3Info.cpp:144
llvm::is_sorted
bool is_sorted(R &&Range, Compare C)
Wrapper function around std::is_sorted to check if elements in a range R are sorted with respect to a...
Definition: STLExtras.h:1909
llvm::X86InstrFMA3Group
This class is used to group {132, 213, 231} forms of FMA opcodes together.
Definition: X86InstrFMA3Info.h:24
llvm::X86InstrFMA3Group::Intrinsic
@ Intrinsic
This bit must be set in the 'Attributes' field of FMA group if such group of FMA opcodes consists of ...
Definition: X86InstrFMA3Info.h:41
llvm::X86InstrFMA3Group::Opcodes
uint16_t Opcodes[3]
An array holding 3 forms of FMA opcodes.
Definition: X86InstrFMA3Info.h:26
Generated on Wed Nov 20 2024 09:19:24 for LLVM by doxygen 1.9.6