Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 1 | //===- SSEDomainFix.cpp - Use proper int/float domain for SSE ---*- C++ -*-===// |
| 2 | // |
| 3 | // The LLVM Compiler Infrastructure |
| 4 | // |
| 5 | // This file is distributed under the University of Illinois Open Source |
| 6 | // License. See LICENSE.TXT for details. |
| 7 | // |
| 8 | //===----------------------------------------------------------------------===// |
| 9 | // |
| 10 | // This file contains the SSEDomainFix pass. |
| 11 | // |
| 12 | // Some SSE instructions like mov, and, or, xor are available in different |
| 13 | // variants for different operand types. These variant instructions are |
| 14 | // equivalent, but on Nehalem and newer cpus there is extra latency |
| 15 | // transferring data between integer and floating point domains. |
| 16 | // |
| 17 | // This pass changes the variant instructions to minimize domain crossings. |
| 18 | // |
| 19 | //===----------------------------------------------------------------------===// |
| 20 | |
| 21 | #define DEBUG_TYPE "sse-domain-fix" |
| 22 | #include "X86InstrInfo.h" |
| 23 | #include "llvm/CodeGen/MachineFunctionPass.h" |
| 24 | #include "llvm/CodeGen/MachineRegisterInfo.h" |
| 25 | #include "llvm/ADT/DepthFirstIterator.h" |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 26 | #include "llvm/Support/Allocator.h" |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 27 | #include "llvm/Support/Debug.h" |
| 28 | #include "llvm/Support/raw_ostream.h" |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 29 | using namespace llvm; |
| 30 | |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 31 | /// A DomainValue is a bit like LiveIntervals' ValNo, but it also keeps track |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 32 | /// of execution domains. |
| 33 | /// |
| 34 | /// An open DomainValue represents a set of instructions that can still switch |
| 35 | /// execution domain. Multiple registers may refer to the same open |
| 36 | /// DomainValue - they will eventually be collapsed to the same execution |
| 37 | /// domain. |
| 38 | /// |
| 39 | /// A collapsed DomainValue represents a single register that has been forced |
| 40 | /// into one of more execution domains. There is a separate collapsed |
| 41 | /// DomainValue for each register, but it may contain multiple execution |
| 42 | /// domains. A register value is initially created in a single execution |
| 43 | /// domain, but if we were forced to pay the penalty of a domain crossing, we |
| 44 | /// keep track of the fact the the register is now available in multiple |
| 45 | /// domains. |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 46 | namespace { |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 47 | struct DomainValue { |
| 48 | // Basic reference counting. |
| 49 | unsigned Refs; |
| 50 | |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 51 | // Bitmask of available domains. For an open DomainValue, it is the still |
| 52 | // possible domains for collapsing. For a collapsed DomainValue it is the |
| 53 | // domains where the register is available for free. |
| 54 | unsigned AvailableDomains; |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 55 | |
| 56 | // Position of the last defining instruction. |
| 57 | unsigned Dist; |
| 58 | |
| 59 | // Twiddleable instructions using or defining these registers. |
| 60 | SmallVector<MachineInstr*, 8> Instrs; |
| 61 | |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 62 | // A collapsed DomainValue has no instructions to twiddle - it simply keeps |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 63 | // track of the domains where the registers are already available. |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 64 | bool isCollapsed() const { return Instrs.empty(); } |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 65 | |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 66 | // Is domain available? |
| 67 | bool hasDomain(unsigned domain) const { |
| 68 | return AvailableDomains & (1u << domain); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 69 | } |
| 70 | |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 71 | // Mark domain as available. |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 72 | void addDomain(unsigned domain) { |
| 73 | AvailableDomains |= 1u << domain; |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 74 | } |
| 75 | |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 76 | // Restrict to a single domain available. |
| 77 | void setSingleDomain(unsigned domain) { |
| 78 | AvailableDomains = 1u << domain; |
| 79 | } |
| 80 | |
| 81 | // Return bitmask of domains that are available and in mask. |
| 82 | unsigned getCommonDomains(unsigned mask) const { |
| 83 | return AvailableDomains & mask; |
| 84 | } |
| 85 | |
| 86 | // First domain available. |
| 87 | unsigned getFirstDomain() const { |
| 88 | return CountTrailingZeros_32(AvailableDomains); |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 89 | } |
| 90 | |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 91 | DomainValue() { clear(); } |
| 92 | |
| 93 | void clear() { |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 94 | Refs = AvailableDomains = Dist = 0; |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 95 | Instrs.clear(); |
| 96 | } |
| 97 | }; |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 98 | } |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 99 | |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 100 | static const unsigned NumRegs = 16; |
| 101 | |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 102 | namespace { |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 103 | class SSEDomainFixPass : public MachineFunctionPass { |
| 104 | static char ID; |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 105 | SpecificBumpPtrAllocator<DomainValue> Allocator; |
| 106 | SmallVector<DomainValue*,16> Avail; |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 107 | |
| 108 | MachineFunction *MF; |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 109 | const X86InstrInfo *TII; |
| 110 | const TargetRegisterInfo *TRI; |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 111 | MachineBasicBlock *MBB; |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 112 | DomainValue **LiveRegs; |
| 113 | typedef DenseMap<MachineBasicBlock*,DomainValue**> LiveOutMap; |
| 114 | LiveOutMap LiveOuts; |
| 115 | unsigned Distance; |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 116 | |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 117 | public: |
Owen Anderson | 1f74590 | 2010-08-06 00:23:35 +0000 | [diff] [blame^] | 118 | SSEDomainFixPass() : MachineFunctionPass(&ID) {} |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 119 | |
| 120 | virtual void getAnalysisUsage(AnalysisUsage &AU) const { |
| 121 | AU.setPreservesAll(); |
| 122 | MachineFunctionPass::getAnalysisUsage(AU); |
| 123 | } |
| 124 | |
| 125 | virtual bool runOnMachineFunction(MachineFunction &MF); |
| 126 | |
| 127 | virtual const char *getPassName() const { |
| 128 | return "SSE execution domain fixup"; |
| 129 | } |
| 130 | |
| 131 | private: |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 132 | // Register mapping. |
| 133 | int RegIndex(unsigned Reg); |
| 134 | |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 135 | // DomainValue allocation. |
| 136 | DomainValue *Alloc(int domain = -1); |
| 137 | void Recycle(DomainValue*); |
| 138 | |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 139 | // LiveRegs manipulations. |
| 140 | void SetLiveReg(int rx, DomainValue *DV); |
| 141 | void Kill(int rx); |
| 142 | void Force(int rx, unsigned domain); |
| 143 | void Collapse(DomainValue *dv, unsigned domain); |
| 144 | bool Merge(DomainValue *A, DomainValue *B); |
| 145 | |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 146 | void enterBasicBlock(); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 147 | void visitGenericInstr(MachineInstr*); |
| 148 | void visitSoftInstr(MachineInstr*, unsigned mask); |
| 149 | void visitHardInstr(MachineInstr*, unsigned domain); |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 150 | }; |
| 151 | } |
| 152 | |
| 153 | char SSEDomainFixPass::ID = 0; |
| 154 | |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 155 | /// Translate TRI register number to an index into our smaller tables of |
| 156 | /// interesting registers. Return -1 for boring registers. |
| 157 | int SSEDomainFixPass::RegIndex(unsigned reg) { |
Jakob Stoklund Olesen | 160a3bf | 2010-05-26 21:47:28 +0000 | [diff] [blame] | 158 | assert(X86::XMM15 == X86::XMM0+NumRegs-1 && "Unexpected sort"); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 159 | reg -= X86::XMM0; |
John McCall | 795ee9d | 2010-04-06 23:35:53 +0000 | [diff] [blame] | 160 | return reg < NumRegs ? (int) reg : -1; |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 161 | } |
| 162 | |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 163 | DomainValue *SSEDomainFixPass::Alloc(int domain) { |
| 164 | DomainValue *dv = Avail.empty() ? |
| 165 | new(Allocator.Allocate()) DomainValue : |
| 166 | Avail.pop_back_val(); |
| 167 | dv->Dist = Distance; |
| 168 | if (domain >= 0) |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 169 | dv->addDomain(domain); |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 170 | return dv; |
| 171 | } |
| 172 | |
| 173 | void SSEDomainFixPass::Recycle(DomainValue *dv) { |
| 174 | assert(dv && "Cannot recycle NULL"); |
| 175 | dv->clear(); |
| 176 | Avail.push_back(dv); |
| 177 | } |
| 178 | |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 179 | /// Set LiveRegs[rx] = dv, updating reference counts. |
| 180 | void SSEDomainFixPass::SetLiveReg(int rx, DomainValue *dv) { |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 181 | assert(unsigned(rx) < NumRegs && "Invalid index"); |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 182 | if (!LiveRegs) { |
| 183 | LiveRegs = new DomainValue*[NumRegs]; |
| 184 | std::fill(LiveRegs, LiveRegs+NumRegs, (DomainValue*)0); |
| 185 | } |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 186 | |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 187 | if (LiveRegs[rx] == dv) |
| 188 | return; |
| 189 | if (LiveRegs[rx]) { |
| 190 | assert(LiveRegs[rx]->Refs && "Bad refcount"); |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 191 | if (--LiveRegs[rx]->Refs == 0) Recycle(LiveRegs[rx]); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 192 | } |
| 193 | LiveRegs[rx] = dv; |
| 194 | if (dv) ++dv->Refs; |
| 195 | } |
| 196 | |
| 197 | // Kill register rx, recycle or collapse any DomainValue. |
| 198 | void SSEDomainFixPass::Kill(int rx) { |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 199 | assert(unsigned(rx) < NumRegs && "Invalid index"); |
| 200 | if (!LiveRegs || !LiveRegs[rx]) return; |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 201 | |
| 202 | // Before killing the last reference to an open DomainValue, collapse it to |
| 203 | // the first available domain. |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 204 | if (LiveRegs[rx]->Refs == 1 && !LiveRegs[rx]->isCollapsed()) |
| 205 | Collapse(LiveRegs[rx], LiveRegs[rx]->getFirstDomain()); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 206 | else |
| 207 | SetLiveReg(rx, 0); |
| 208 | } |
| 209 | |
| 210 | /// Force register rx into domain. |
| 211 | void SSEDomainFixPass::Force(int rx, unsigned domain) { |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 212 | assert(unsigned(rx) < NumRegs && "Invalid index"); |
Jakob Stoklund Olesen | d77f818 | 2010-03-30 00:09:32 +0000 | [diff] [blame] | 213 | DomainValue *dv; |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 214 | if (LiveRegs && (dv = LiveRegs[rx])) { |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 215 | if (dv->isCollapsed()) |
| 216 | dv->addDomain(domain); |
Jakob Stoklund Olesen | 8ba1c6a | 2010-04-06 19:48:56 +0000 | [diff] [blame] | 217 | else if (dv->hasDomain(domain)) |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 218 | Collapse(dv, domain); |
Jakob Stoklund Olesen | 8ba1c6a | 2010-04-06 19:48:56 +0000 | [diff] [blame] | 219 | else { |
| 220 | // This is an incompatible open DomainValue. Collapse it to whatever and force |
| 221 | // the new value into domain. This costs a domain crossing. |
| 222 | Collapse(dv, dv->getFirstDomain()); |
| 223 | assert(LiveRegs[rx] && "Not live after collapse?"); |
| 224 | LiveRegs[rx]->addDomain(domain); |
| 225 | } |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 226 | } else { |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 227 | // Set up basic collapsed DomainValue. |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 228 | SetLiveReg(rx, Alloc(domain)); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 229 | } |
| 230 | } |
| 231 | |
| 232 | /// Collapse open DomainValue into given domain. If there are multiple |
| 233 | /// registers using dv, they each get a unique collapsed DomainValue. |
| 234 | void SSEDomainFixPass::Collapse(DomainValue *dv, unsigned domain) { |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 235 | assert(dv->hasDomain(domain) && "Cannot collapse"); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 236 | |
| 237 | // Collapse all the instructions. |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 238 | while (!dv->Instrs.empty()) |
| 239 | TII->SetSSEDomain(dv->Instrs.pop_back_val(), domain); |
| 240 | dv->setSingleDomain(domain); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 241 | |
| 242 | // If there are multiple users, give them new, unique DomainValues. |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 243 | if (LiveRegs && dv->Refs > 1) |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 244 | for (unsigned rx = 0; rx != NumRegs; ++rx) |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 245 | if (LiveRegs[rx] == dv) |
| 246 | SetLiveReg(rx, Alloc(domain)); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 247 | } |
| 248 | |
| 249 | /// Merge - All instructions and registers in B are moved to A, and B is |
| 250 | /// released. |
| 251 | bool SSEDomainFixPass::Merge(DomainValue *A, DomainValue *B) { |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 252 | assert(!A->isCollapsed() && "Cannot merge into collapsed"); |
| 253 | assert(!B->isCollapsed() && "Cannot merge from collapsed"); |
Jakob Stoklund Olesen | 85ffee2 | 2010-03-31 20:05:12 +0000 | [diff] [blame] | 254 | if (A == B) |
Jakob Stoklund Olesen | 5f282b5 | 2010-03-31 17:13:16 +0000 | [diff] [blame] | 255 | return true; |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 256 | // Restrict to the domains that A and B have in common. |
| 257 | unsigned common = A->getCommonDomains(B->AvailableDomains); |
| 258 | if (!common) |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 259 | return false; |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 260 | A->AvailableDomains = common; |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 261 | A->Dist = std::max(A->Dist, B->Dist); |
| 262 | A->Instrs.append(B->Instrs.begin(), B->Instrs.end()); |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 263 | for (unsigned rx = 0; rx != NumRegs; ++rx) |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 264 | if (LiveRegs[rx] == B) |
| 265 | SetLiveReg(rx, A); |
| 266 | return true; |
| 267 | } |
| 268 | |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 269 | void SSEDomainFixPass::enterBasicBlock() { |
| 270 | // Try to coalesce live-out registers from predecessors. |
Dan Gohman | 81bf03e | 2010-04-13 16:57:55 +0000 | [diff] [blame] | 271 | for (MachineBasicBlock::livein_iterator i = MBB->livein_begin(), |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 272 | e = MBB->livein_end(); i != e; ++i) { |
| 273 | int rx = RegIndex(*i); |
| 274 | if (rx < 0) continue; |
| 275 | for (MachineBasicBlock::const_pred_iterator pi = MBB->pred_begin(), |
| 276 | pe = MBB->pred_end(); pi != pe; ++pi) { |
Jakob Stoklund Olesen | b16df90 | 2010-03-31 00:40:08 +0000 | [diff] [blame] | 277 | LiveOutMap::const_iterator fi = LiveOuts.find(*pi); |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 278 | if (fi == LiveOuts.end()) continue; |
| 279 | DomainValue *pdv = fi->second[rx]; |
| 280 | if (!pdv) continue; |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 281 | if (!LiveRegs || !LiveRegs[rx]) { |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 282 | SetLiveReg(rx, pdv); |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 283 | continue; |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 284 | } |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 285 | |
| 286 | // We have a live DomainValue from more than one predecessor. |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 287 | if (LiveRegs[rx]->isCollapsed()) { |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 288 | // We are already collapsed, but predecessor is not. Force him. |
Jakob Stoklund Olesen | 8ba1c6a | 2010-04-06 19:48:56 +0000 | [diff] [blame] | 289 | unsigned domain = LiveRegs[rx]->getFirstDomain(); |
| 290 | if (!pdv->isCollapsed() && pdv->hasDomain(domain)) |
| 291 | Collapse(pdv, domain); |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 292 | continue; |
| 293 | } |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 294 | |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 295 | // Currently open, merge in predecessor. |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 296 | if (!pdv->isCollapsed()) |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 297 | Merge(LiveRegs[rx], pdv); |
| 298 | else |
Jakob Stoklund Olesen | 8ba1c6a | 2010-04-06 19:48:56 +0000 | [diff] [blame] | 299 | Force(rx, pdv->getFirstDomain()); |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 300 | } |
| 301 | } |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 302 | } |
| 303 | |
| 304 | // A hard instruction only works in one domain. All input registers will be |
| 305 | // forced into that domain. |
| 306 | void SSEDomainFixPass::visitHardInstr(MachineInstr *mi, unsigned domain) { |
| 307 | // Collapse all uses. |
| 308 | for (unsigned i = mi->getDesc().getNumDefs(), |
| 309 | e = mi->getDesc().getNumOperands(); i != e; ++i) { |
| 310 | MachineOperand &mo = mi->getOperand(i); |
| 311 | if (!mo.isReg()) continue; |
| 312 | int rx = RegIndex(mo.getReg()); |
| 313 | if (rx < 0) continue; |
| 314 | Force(rx, domain); |
| 315 | } |
| 316 | |
| 317 | // Kill all defs and force them. |
| 318 | for (unsigned i = 0, e = mi->getDesc().getNumDefs(); i != e; ++i) { |
| 319 | MachineOperand &mo = mi->getOperand(i); |
| 320 | if (!mo.isReg()) continue; |
| 321 | int rx = RegIndex(mo.getReg()); |
| 322 | if (rx < 0) continue; |
| 323 | Kill(rx); |
| 324 | Force(rx, domain); |
| 325 | } |
| 326 | } |
| 327 | |
| 328 | // A soft instruction can be changed to work in other domains given by mask. |
| 329 | void SSEDomainFixPass::visitSoftInstr(MachineInstr *mi, unsigned mask) { |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 330 | // Bitmask of available domains for this instruction after taking collapsed |
| 331 | // operands into account. |
| 332 | unsigned available = mask; |
| 333 | |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 334 | // Scan the explicit use operands for incoming domains. |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 335 | SmallVector<int, 4> used; |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 336 | if (LiveRegs) |
| 337 | for (unsigned i = mi->getDesc().getNumDefs(), |
| 338 | e = mi->getDesc().getNumOperands(); i != e; ++i) { |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 339 | MachineOperand &mo = mi->getOperand(i); |
| 340 | if (!mo.isReg()) continue; |
| 341 | int rx = RegIndex(mo.getReg()); |
| 342 | if (rx < 0) continue; |
| 343 | if (DomainValue *dv = LiveRegs[rx]) { |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 344 | // Bitmask of domains that dv and available have in common. |
| 345 | unsigned common = dv->getCommonDomains(available); |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 346 | // Is it possible to use this collapsed register for free? |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 347 | if (dv->isCollapsed()) { |
| 348 | // Restrict available domains to the ones in common with the operand. |
| 349 | // If there are no common domains, we must pay the cross-domain |
| 350 | // penalty for this operand. |
| 351 | if (common) available = common; |
| 352 | } else if (common) |
| 353 | // Open DomainValue is compatible, save it for merging. |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 354 | used.push_back(rx); |
| 355 | else |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 356 | // Open DomainValue is not compatible with instruction. It is useless |
| 357 | // now. |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 358 | Kill(rx); |
| 359 | } |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 360 | } |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 361 | |
| 362 | // If the collapsed operands force a single domain, propagate the collapse. |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 363 | if (isPowerOf2_32(available)) { |
| 364 | unsigned domain = CountTrailingZeros_32(available); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 365 | TII->SetSSEDomain(mi, domain); |
| 366 | visitHardInstr(mi, domain); |
| 367 | return; |
| 368 | } |
| 369 | |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 370 | // Kill off any remaining uses that don't match available, and build a list of |
| 371 | // incoming DomainValues that we want to merge. |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 372 | SmallVector<DomainValue*,4> doms; |
| 373 | for (SmallVector<int, 4>::iterator i=used.begin(), e=used.end(); i!=e; ++i) { |
| 374 | int rx = *i; |
| 375 | DomainValue *dv = LiveRegs[rx]; |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 376 | // This useless DomainValue could have been missed above. |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 377 | if (!dv->getCommonDomains(available)) { |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 378 | Kill(*i); |
| 379 | continue; |
| 380 | } |
| 381 | // sorted, uniqued insert. |
| 382 | bool inserted = false; |
| 383 | for (SmallVector<DomainValue*,4>::iterator i = doms.begin(), e = doms.end(); |
| 384 | i != e && !inserted; ++i) { |
| 385 | if (dv == *i) |
| 386 | inserted = true; |
| 387 | else if (dv->Dist < (*i)->Dist) { |
| 388 | inserted = true; |
| 389 | doms.insert(i, dv); |
| 390 | } |
| 391 | } |
| 392 | if (!inserted) |
| 393 | doms.push_back(dv); |
| 394 | } |
| 395 | |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 396 | // doms are now sorted in order of appearance. Try to merge them all, giving |
| 397 | // priority to the latest ones. |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 398 | DomainValue *dv = 0; |
| 399 | while (!doms.empty()) { |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 400 | if (!dv) { |
| 401 | dv = doms.pop_back_val(); |
| 402 | continue; |
| 403 | } |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 404 | |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 405 | DomainValue *latest = doms.pop_back_val(); |
| 406 | if (Merge(dv, latest)) continue; |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 407 | |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 408 | // If latest didn't merge, it is useless now. Kill all registers using it. |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 409 | for (SmallVector<int,4>::iterator i=used.begin(), e=used.end(); i != e; ++i) |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 410 | if (LiveRegs[*i] == latest) |
Chris Lattner | 563d83f | 2010-03-31 20:32:51 +0000 | [diff] [blame] | 411 | Kill(*i); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 412 | } |
| 413 | |
| 414 | // dv is the DomainValue we are going to use for this instruction. |
| 415 | if (!dv) |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 416 | dv = Alloc(); |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 417 | dv->Dist = Distance; |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 418 | dv->AvailableDomains = available; |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 419 | dv->Instrs.push_back(mi); |
| 420 | |
| 421 | // Finally set all defs and non-collapsed uses to dv. |
| 422 | for (unsigned i = 0, e = mi->getDesc().getNumOperands(); i != e; ++i) { |
| 423 | MachineOperand &mo = mi->getOperand(i); |
| 424 | if (!mo.isReg()) continue; |
| 425 | int rx = RegIndex(mo.getReg()); |
| 426 | if (rx < 0) continue; |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 427 | if (!LiveRegs || !LiveRegs[rx] || (mo.isDef() && LiveRegs[rx]!=dv)) { |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 428 | Kill(rx); |
| 429 | SetLiveReg(rx, dv); |
| 430 | } |
| 431 | } |
| 432 | } |
| 433 | |
| 434 | void SSEDomainFixPass::visitGenericInstr(MachineInstr *mi) { |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 435 | // Process explicit defs, kill any XMM registers redefined. |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 436 | for (unsigned i = 0, e = mi->getDesc().getNumDefs(); i != e; ++i) { |
| 437 | MachineOperand &mo = mi->getOperand(i); |
| 438 | if (!mo.isReg()) continue; |
| 439 | int rx = RegIndex(mo.getReg()); |
| 440 | if (rx < 0) continue; |
| 441 | Kill(rx); |
| 442 | } |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 443 | } |
| 444 | |
| 445 | bool SSEDomainFixPass::runOnMachineFunction(MachineFunction &mf) { |
| 446 | MF = &mf; |
| 447 | TII = static_cast<const X86InstrInfo*>(MF->getTarget().getInstrInfo()); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 448 | TRI = MF->getTarget().getRegisterInfo(); |
| 449 | MBB = 0; |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 450 | LiveRegs = 0; |
| 451 | Distance = 0; |
| 452 | assert(NumRegs == X86::VR128RegClass.getNumRegs() && "Bad regclass"); |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 453 | |
| 454 | // If no XMM registers are used in the function, we can skip it completely. |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 455 | bool anyregs = false; |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 456 | for (TargetRegisterClass::const_iterator I = X86::VR128RegClass.begin(), |
| 457 | E = X86::VR128RegClass.end(); I != E; ++I) |
| 458 | if (MF->getRegInfo().isPhysRegUsed(*I)) { |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 459 | anyregs = true; |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 460 | break; |
| 461 | } |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 462 | if (!anyregs) return false; |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 463 | |
| 464 | MachineBasicBlock *Entry = MF->begin(); |
| 465 | SmallPtrSet<MachineBasicBlock*, 16> Visited; |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 466 | for (df_ext_iterator<MachineBasicBlock*, SmallPtrSet<MachineBasicBlock*, 16> > |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 467 | DFI = df_ext_begin(Entry, Visited), DFE = df_ext_end(Entry, Visited); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 468 | DFI != DFE; ++DFI) { |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 469 | MBB = *DFI; |
| 470 | enterBasicBlock(); |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 471 | for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end(); I != E; |
| 472 | ++I) { |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 473 | MachineInstr *mi = I; |
| 474 | if (mi->isDebugValue()) continue; |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 475 | ++Distance; |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 476 | std::pair<uint16_t, uint16_t> domp = TII->GetSSEDomain(mi); |
| 477 | if (domp.first) |
| 478 | if (domp.second) |
| 479 | visitSoftInstr(mi, domp.second); |
| 480 | else |
| 481 | visitHardInstr(mi, domp.first); |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 482 | else if (LiveRegs) |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 483 | visitGenericInstr(mi); |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 484 | } |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 485 | |
| 486 | // Save live registers at end of MBB - used by enterBasicBlock(). |
| 487 | if (LiveRegs) |
| 488 | LiveOuts.insert(std::make_pair(MBB, LiveRegs)); |
| 489 | LiveRegs = 0; |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 490 | } |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 491 | |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 492 | // Clear the LiveOuts vectors. Should we also collapse any remaining |
| 493 | // DomainValues? |
| 494 | for (LiveOutMap::const_iterator i = LiveOuts.begin(), e = LiveOuts.end(); |
| 495 | i != e; ++i) |
Jakob Stoklund Olesen | e0103f0 | 2010-04-04 21:27:26 +0000 | [diff] [blame] | 496 | delete[] i->second; |
Jakob Stoklund Olesen | 1a5d2a8 | 2010-03-30 20:04:01 +0000 | [diff] [blame] | 497 | LiveOuts.clear(); |
Jakob Stoklund Olesen | bbef815 | 2010-04-04 18:00:21 +0000 | [diff] [blame] | 498 | Avail.clear(); |
| 499 | Allocator.DestroyAll(); |
Jakob Stoklund Olesen | e4b94b4 | 2010-03-29 23:24:21 +0000 | [diff] [blame] | 500 | |
Jakob Stoklund Olesen | 352aa50 | 2010-03-25 17:25:00 +0000 | [diff] [blame] | 501 | return false; |
| 502 | } |
| 503 | |
| 504 | FunctionPass *llvm::createSSEDomainFixPass() { |
| 505 | return new SSEDomainFixPass(); |
| 506 | } |