blob: 9218bed5d882a9981e92b4e300af3728bc93a235 [file] [log] [blame]
Chandler Carruth7132e002007-08-04 01:51:18 +00001//===-- AutoUpgrade.cpp - Implement auto-upgrade helper functions ---------===//
2//
3// The LLVM Compiler Infrastructure
4//
Chris Lattnerf3ebc3f2007-12-29 20:36:04 +00005// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
Chandler Carruth7132e002007-08-04 01:51:18 +00007//
8//===----------------------------------------------------------------------===//
9//
Sanjay Patel19792fb2015-03-10 16:08:36 +000010// This file implements the auto-upgrade helper functions.
11// This is where deprecated IR intrinsics and other IR features are updated to
12// current specifications.
Chandler Carruth7132e002007-08-04 01:51:18 +000013//
14//===----------------------------------------------------------------------===//
15
Chandler Carruth91065212014-03-05 10:34:14 +000016#include "llvm/IR/AutoUpgrade.h"
Chandler Carruth1305dc32014-03-04 11:45:46 +000017#include "llvm/IR/CFG.h"
Chandler Carruth219b89b2014-03-04 11:01:28 +000018#include "llvm/IR/CallSite.h"
Chandler Carruth9fb823b2013-01-02 11:36:10 +000019#include "llvm/IR/Constants.h"
Chandler Carruthd9903882015-01-14 11:23:27 +000020#include "llvm/IR/DIBuilder.h"
Chandler Carruth9a4c9e52014-03-06 00:46:21 +000021#include "llvm/IR/DebugInfo.h"
Manman Ren2ebfb422014-01-16 01:51:12 +000022#include "llvm/IR/DiagnosticInfo.h"
Chandler Carruth9fb823b2013-01-02 11:36:10 +000023#include "llvm/IR/Function.h"
24#include "llvm/IR/IRBuilder.h"
25#include "llvm/IR/Instruction.h"
26#include "llvm/IR/IntrinsicInst.h"
27#include "llvm/IR/LLVMContext.h"
28#include "llvm/IR/Module.h"
Torok Edwin56d06592009-07-11 20:10:48 +000029#include "llvm/Support/ErrorHandling.h"
Jeroen Ketemaab99b592015-09-30 10:56:37 +000030#include "llvm/Support/Regex.h"
Anton Korobeynikov579f0712008-02-20 11:08:44 +000031#include <cstring>
Chandler Carruth7132e002007-08-04 01:51:18 +000032using namespace llvm;
33
Rafael Espindolad7325ee2016-10-03 15:51:42 +000034static void rename(GlobalValue *GV) { GV->setName(GV->getName() + ".old"); }
35
Nadav Rotem17ee58a2012-06-10 18:42:51 +000036// Upgrade the declarations of the SSE4.1 functions whose arguments have
37// changed their type from v4f32 to v2i64.
38static bool UpgradeSSE41Function(Function* F, Intrinsic::ID IID,
39 Function *&NewFn) {
40 // Check whether this is an old version of the function, which received
41 // v4f32 arguments.
42 Type *Arg0Type = F->getFunctionType()->getParamType(0);
43 if (Arg0Type != VectorType::get(Type::getFloatTy(F->getContext()), 4))
44 return false;
45
46 // Yes, it's old, replace it with new version.
Rafael Espindolad7325ee2016-10-03 15:51:42 +000047 rename(F);
Nadav Rotem17ee58a2012-06-10 18:42:51 +000048 NewFn = Intrinsic::getDeclaration(F->getParent(), IID);
49 return true;
50}
Chandler Carruth7132e002007-08-04 01:51:18 +000051
Chandler Carruth373b2b12014-09-06 10:00:01 +000052// Upgrade the declarations of intrinsic functions whose 8-bit immediate mask
53// arguments have changed their type from i32 to i8.
54static bool UpgradeX86IntrinsicsWith8BitMask(Function *F, Intrinsic::ID IID,
55 Function *&NewFn) {
56 // Check that the last argument is an i32.
57 Type *LastArgType = F->getFunctionType()->getParamType(
58 F->getFunctionType()->getNumParams() - 1);
59 if (!LastArgType->isIntegerTy(32))
60 return false;
61
62 // Move this function aside and map down.
Rafael Espindolad7325ee2016-10-03 15:51:42 +000063 rename(F);
Chandler Carruth373b2b12014-09-06 10:00:01 +000064 NewFn = Intrinsic::getDeclaration(F->getParent(), IID);
65 return true;
66}
67
Evan Cheng0e179d02007-12-17 22:33:23 +000068static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
Chandler Carruth7132e002007-08-04 01:51:18 +000069 assert(F && "Illegal to upgrade a non-existent Function.");
70
Chandler Carruth7132e002007-08-04 01:51:18 +000071 // Quickly eliminate it, if it's not a candidate.
Eric Liuc9c68172016-07-08 16:09:51 +000072 StringRef Name = F->getName();
Chris Lattnerb372f662011-06-18 18:56:39 +000073 if (Name.size() <= 8 || !Name.startswith("llvm."))
Evan Cheng0e179d02007-12-17 22:33:23 +000074 return false;
Chris Lattnerb372f662011-06-18 18:56:39 +000075 Name = Name.substr(5); // Strip off "llvm."
Chris Lattner0bcbde42011-11-27 08:42:07 +000076
Chris Lattnerb372f662011-06-18 18:56:39 +000077 switch (Name[0]) {
Chandler Carruth7132e002007-08-04 01:51:18 +000078 default: break;
Joel Jones43cb8782012-07-13 23:25:25 +000079 case 'a': {
80 if (Name.startswith("arm.neon.vclz")) {
81 Type* args[2] = {
Matt Arsenaultc4c92262013-07-20 17:46:00 +000082 F->arg_begin()->getType(),
Joel Jones43cb8782012-07-13 23:25:25 +000083 Type::getInt1Ty(F->getContext())
84 };
85 // Can't use Intrinsic::getDeclaration here as it adds a ".i1" to
86 // the end of the name. Change name from llvm.arm.neon.vclz.* to
87 // llvm.ctlz.*
88 FunctionType* fType = FunctionType::get(F->getReturnType(), args, false);
Matt Arsenaultc4c92262013-07-20 17:46:00 +000089 NewFn = Function::Create(fType, F->getLinkage(),
Joel Jones43cb8782012-07-13 23:25:25 +000090 "llvm.ctlz." + Name.substr(14), F->getParent());
91 return true;
92 }
Joel Jonesb84f7be2012-07-18 00:02:16 +000093 if (Name.startswith("arm.neon.vcnt")) {
94 NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::ctpop,
95 F->arg_begin()->getType());
96 return true;
97 }
Jeroen Ketemaab99b592015-09-30 10:56:37 +000098 Regex vldRegex("^arm\\.neon\\.vld([1234]|[234]lane)\\.v[a-z0-9]*$");
99 if (vldRegex.match(Name)) {
100 auto fArgs = F->getFunctionType()->params();
101 SmallVector<Type *, 4> Tys(fArgs.begin(), fArgs.end());
102 // Can't use Intrinsic::getDeclaration here as the return types might
103 // then only be structurally equal.
104 FunctionType* fType = FunctionType::get(F->getReturnType(), Tys, false);
105 NewFn = Function::Create(fType, F->getLinkage(),
106 "llvm." + Name + ".p0i8", F->getParent());
107 return true;
108 }
109 Regex vstRegex("^arm\\.neon\\.vst([1234]|[234]lane)\\.v[a-z0-9]*$");
110 if (vstRegex.match(Name)) {
Craig Topper26260942015-10-18 05:15:34 +0000111 static const Intrinsic::ID StoreInts[] = {Intrinsic::arm_neon_vst1,
112 Intrinsic::arm_neon_vst2,
113 Intrinsic::arm_neon_vst3,
114 Intrinsic::arm_neon_vst4};
Jeroen Ketemaab99b592015-09-30 10:56:37 +0000115
Craig Topper26260942015-10-18 05:15:34 +0000116 static const Intrinsic::ID StoreLaneInts[] = {
117 Intrinsic::arm_neon_vst2lane, Intrinsic::arm_neon_vst3lane,
118 Intrinsic::arm_neon_vst4lane
119 };
Jeroen Ketemaab99b592015-09-30 10:56:37 +0000120
121 auto fArgs = F->getFunctionType()->params();
122 Type *Tys[] = {fArgs[0], fArgs[1]};
123 if (Name.find("lane") == StringRef::npos)
124 NewFn = Intrinsic::getDeclaration(F->getParent(),
125 StoreInts[fArgs.size() - 3], Tys);
126 else
127 NewFn = Intrinsic::getDeclaration(F->getParent(),
128 StoreLaneInts[fArgs.size() - 5], Tys);
129 return true;
130 }
Marcin Koscielnicki3fdc2572016-04-19 20:51:05 +0000131 if (Name == "aarch64.thread.pointer" || Name == "arm.thread.pointer") {
132 NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::thread_pointer);
133 return true;
134 }
Joel Jones43cb8782012-07-13 23:25:25 +0000135 break;
136 }
Jeroen Ketemaab99b592015-09-30 10:56:37 +0000137
Chandler Carruth58a71ed2011-12-12 04:26:04 +0000138 case 'c': {
Chandler Carruth58a71ed2011-12-12 04:26:04 +0000139 if (Name.startswith("ctlz.") && F->arg_size() == 1) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000140 rename(F);
Chandler Carruthd4a02402011-12-12 10:57:20 +0000141 NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::ctlz,
142 F->arg_begin()->getType());
Chandler Carruth58a71ed2011-12-12 04:26:04 +0000143 return true;
144 }
145 if (Name.startswith("cttz.") && F->arg_size() == 1) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000146 rename(F);
Chandler Carruthd4a02402011-12-12 10:57:20 +0000147 NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::cttz,
148 F->arg_begin()->getType());
Chandler Carruth58a71ed2011-12-12 04:26:04 +0000149 return true;
150 }
151 break;
152 }
Mehdi Amini8c629ec2016-08-13 23:31:24 +0000153 case 'i': {
154 if (Name.startswith("invariant.start")) {
155 auto Args = F->getFunctionType()->params();
156 Type* ObjectPtr[1] = {Args[1]};
157 if (F->getName() !=
158 Intrinsic::getName(Intrinsic::invariant_start, ObjectPtr)) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000159 rename(F);
Mehdi Amini8c629ec2016-08-13 23:31:24 +0000160 NewFn = Intrinsic::getDeclaration(
161 F->getParent(), Intrinsic::invariant_start, ObjectPtr);
162 return true;
163 }
164 }
165 if (Name.startswith("invariant.end")) {
166 auto Args = F->getFunctionType()->params();
167 Type* ObjectPtr[1] = {Args[2]};
168 if (F->getName() !=
169 Intrinsic::getName(Intrinsic::invariant_end, ObjectPtr)) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000170 rename(F);
Mehdi Amini8c629ec2016-08-13 23:31:24 +0000171 NewFn = Intrinsic::getDeclaration(F->getParent(),
172 Intrinsic::invariant_end, ObjectPtr);
173 return true;
174 }
175 }
176 break;
177 }
Artur Pilipenko7ad95ec2016-06-28 18:27:25 +0000178 case 'm': {
179 if (Name.startswith("masked.load.")) {
180 Type *Tys[] = { F->getReturnType(), F->arg_begin()->getType() };
181 if (F->getName() != Intrinsic::getName(Intrinsic::masked_load, Tys)) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000182 rename(F);
Artur Pilipenko7ad95ec2016-06-28 18:27:25 +0000183 NewFn = Intrinsic::getDeclaration(F->getParent(),
184 Intrinsic::masked_load,
185 Tys);
186 return true;
187 }
188 }
189 if (Name.startswith("masked.store.")) {
190 auto Args = F->getFunctionType()->params();
191 Type *Tys[] = { Args[0], Args[1] };
192 if (F->getName() != Intrinsic::getName(Intrinsic::masked_store, Tys)) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000193 rename(F);
Artur Pilipenko7ad95ec2016-06-28 18:27:25 +0000194 NewFn = Intrinsic::getDeclaration(F->getParent(),
195 Intrinsic::masked_store,
196 Tys);
197 return true;
198 }
199 }
200 break;
201 }
202
Matt Arsenaultfbcbce42013-10-07 18:06:48 +0000203 case 'o':
204 // We only need to change the name to match the mangling including the
205 // address space.
206 if (F->arg_size() == 2 && Name.startswith("objectsize.")) {
207 Type *Tys[2] = { F->getReturnType(), F->arg_begin()->getType() };
208 if (F->getName() != Intrinsic::getName(Intrinsic::objectsize, Tys)) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000209 rename(F);
Matt Arsenaultfbcbce42013-10-07 18:06:48 +0000210 NewFn = Intrinsic::getDeclaration(F->getParent(),
211 Intrinsic::objectsize, Tys);
212 return true;
213 }
214 }
215 break;
216
Tim Shen00127562016-04-08 21:26:31 +0000217 case 's':
218 if (Name == "stackprotectorcheck") {
219 NewFn = nullptr;
220 return true;
221 }
George Burgess IV63b06e12016-11-08 04:01:50 +0000222 break;
Tim Shen00127562016-04-08 21:26:31 +0000223
Craig Topper3b1817d2012-02-03 06:10:55 +0000224 case 'x': {
Craig Topper5aebb862016-07-04 20:56:38 +0000225 bool IsX86 = Name.startswith("x86.");
226 if (IsX86)
227 Name = Name.substr(4);
228
Craig Toppere6915b82016-11-15 05:04:51 +0000229 // All of the intrinsics matches below should be marked with which llvm
230 // version started autoupgrading them. At some point in the future we would
231 // like to use this information to remove upgrade code for some older
232 // intrinsics. It is currently undecided how we will determine that future
233 // point.
Craig Topper5aebb862016-07-04 20:56:38 +0000234 if (IsX86 &&
Craig Toppere6915b82016-11-15 05:04:51 +0000235 (Name.startswith("sse2.pcmpeq.") || // Added in 3.1
236 Name.startswith("sse2.pcmpgt.") || // Added in 3.1
237 Name.startswith("avx2.pcmpeq.") || // Added in 3.1
238 Name.startswith("avx2.pcmpgt.") || // Added in 3.1
239 Name.startswith("avx512.mask.pcmpeq.") || // Added in 3.9
240 Name.startswith("avx512.mask.pcmpgt.") || // Added in 3.9
Craig Topper6910fa02016-11-16 05:24:10 +0000241 Name == "sse.add.ss" || // Added in 4.0
242 Name == "sse2.add.sd" || // Added in 4.0
243 Name == "sse.sub.ss" || // Added in 4.0
244 Name == "sse2.sub.sd" || // Added in 4.0
245 Name == "sse.mul.ss" || // Added in 4.0
246 Name == "sse2.mul.sd" || // Added in 4.0
247 Name == "sse.div.ss" || // Added in 4.0
248 Name == "sse2.div.sd" || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000249 Name == "sse41.pmaxsb" || // Added in 3.9
250 Name == "sse2.pmaxs.w" || // Added in 3.9
251 Name == "sse41.pmaxsd" || // Added in 3.9
252 Name == "sse2.pmaxu.b" || // Added in 3.9
253 Name == "sse41.pmaxuw" || // Added in 3.9
254 Name == "sse41.pmaxud" || // Added in 3.9
255 Name == "sse41.pminsb" || // Added in 3.9
256 Name == "sse2.pmins.w" || // Added in 3.9
257 Name == "sse41.pminsd" || // Added in 3.9
258 Name == "sse2.pminu.b" || // Added in 3.9
259 Name == "sse41.pminuw" || // Added in 3.9
260 Name == "sse41.pminud" || // Added in 3.9
Craig Topperedab02b2016-12-10 23:09:43 +0000261 Name.startswith("avx512.mask.pshuf.b.") || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000262 Name.startswith("avx2.pmax") || // Added in 3.9
263 Name.startswith("avx2.pmin") || // Added in 3.9
264 Name.startswith("avx512.mask.pmax") || // Added in 4.0
265 Name.startswith("avx512.mask.pmin") || // Added in 4.0
266 Name.startswith("avx2.vbroadcast") || // Added in 3.8
267 Name.startswith("avx2.pbroadcast") || // Added in 3.8
268 Name.startswith("avx.vpermil.") || // Added in 3.1
269 Name.startswith("sse2.pshuf") || // Added in 3.9
270 Name.startswith("avx512.pbroadcast") || // Added in 3.9
271 Name.startswith("avx512.mask.broadcast.s") || // Added in 3.9
272 Name.startswith("avx512.mask.movddup") || // Added in 3.9
273 Name.startswith("avx512.mask.movshdup") || // Added in 3.9
274 Name.startswith("avx512.mask.movsldup") || // Added in 3.9
275 Name.startswith("avx512.mask.pshuf.d.") || // Added in 3.9
276 Name.startswith("avx512.mask.pshufl.w.") || // Added in 3.9
277 Name.startswith("avx512.mask.pshufh.w.") || // Added in 3.9
278 Name.startswith("avx512.mask.shuf.p") || // Added in 4.0
279 Name.startswith("avx512.mask.vpermil.p") || // Added in 3.9
280 Name.startswith("avx512.mask.perm.df.") || // Added in 3.9
281 Name.startswith("avx512.mask.perm.di.") || // Added in 3.9
282 Name.startswith("avx512.mask.punpckl") || // Added in 3.9
283 Name.startswith("avx512.mask.punpckh") || // Added in 3.9
284 Name.startswith("avx512.mask.unpckl.") || // Added in 3.9
285 Name.startswith("avx512.mask.unpckh.") || // Added in 3.9
286 Name.startswith("avx512.mask.pand.") || // Added in 3.9
287 Name.startswith("avx512.mask.pandn.") || // Added in 3.9
288 Name.startswith("avx512.mask.por.") || // Added in 3.9
289 Name.startswith("avx512.mask.pxor.") || // Added in 3.9
290 Name.startswith("avx512.mask.and.") || // Added in 3.9
291 Name.startswith("avx512.mask.andn.") || // Added in 3.9
292 Name.startswith("avx512.mask.or.") || // Added in 3.9
293 Name.startswith("avx512.mask.xor.") || // Added in 3.9
294 Name.startswith("avx512.mask.padd.") || // Added in 4.0
295 Name.startswith("avx512.mask.psub.") || // Added in 4.0
296 Name.startswith("avx512.mask.pmull.") || // Added in 4.0
Simon Pilgrimb57dd172016-11-16 14:48:32 +0000297 Name.startswith("avx512.mask.cvtdq2pd.") || // Added in 4.0
298 Name.startswith("avx512.mask.cvtudq2pd.") || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000299 Name == "avx512.mask.add.pd.128" || // Added in 4.0
300 Name == "avx512.mask.add.pd.256" || // Added in 4.0
301 Name == "avx512.mask.add.ps.128" || // Added in 4.0
302 Name == "avx512.mask.add.ps.256" || // Added in 4.0
303 Name == "avx512.mask.div.pd.128" || // Added in 4.0
304 Name == "avx512.mask.div.pd.256" || // Added in 4.0
305 Name == "avx512.mask.div.ps.128" || // Added in 4.0
306 Name == "avx512.mask.div.ps.256" || // Added in 4.0
307 Name == "avx512.mask.mul.pd.128" || // Added in 4.0
308 Name == "avx512.mask.mul.pd.256" || // Added in 4.0
309 Name == "avx512.mask.mul.ps.128" || // Added in 4.0
310 Name == "avx512.mask.mul.ps.256" || // Added in 4.0
311 Name == "avx512.mask.sub.pd.128" || // Added in 4.0
312 Name == "avx512.mask.sub.pd.256" || // Added in 4.0
313 Name == "avx512.mask.sub.ps.128" || // Added in 4.0
314 Name == "avx512.mask.sub.ps.256" || // Added in 4.0
Craig Topperabe7c5b2016-12-10 21:15:52 +0000315 Name == "avx512.mask.vpermilvar.ps.128" || // Added in 4.0
316 Name == "avx512.mask.vpermilvar.ps.256" || // Added in 4.0
317 Name == "avx512.mask.vpermilvar.pd.128" || // Added in 4.0
318 Name == "avx512.mask.vpermilvar.pd.256" || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000319 Name.startswith("avx512.mask.psll.d") || // Added in 4.0
320 Name.startswith("avx512.mask.psll.q") || // Added in 4.0
321 Name.startswith("avx512.mask.psll.w") || // Added in 4.0
322 Name.startswith("avx512.mask.psra.d") || // Added in 4.0
323 Name.startswith("avx512.mask.psra.q") || // Added in 4.0
324 Name.startswith("avx512.mask.psra.w") || // Added in 4.0
325 Name.startswith("avx512.mask.psrl.d") || // Added in 4.0
326 Name.startswith("avx512.mask.psrl.q") || // Added in 4.0
327 Name.startswith("avx512.mask.psrl.w") || // Added in 4.0
328 Name.startswith("avx512.mask.pslli") || // Added in 4.0
329 Name.startswith("avx512.mask.psrai") || // Added in 4.0
330 Name.startswith("avx512.mask.psrli") || // Added in 4.0
Craig Topper02b5a1b2016-11-18 05:04:44 +0000331 Name.startswith("avx512.mask.psllv") || // Added in 4.0
332 Name.startswith("avx512.mask.psrav") || // Added in 4.0
333 Name.startswith("avx512.mask.psrlv") || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000334 Name.startswith("sse41.pmovsx") || // Added in 3.8
335 Name.startswith("sse41.pmovzx") || // Added in 3.9
336 Name.startswith("avx2.pmovsx") || // Added in 3.9
337 Name.startswith("avx2.pmovzx") || // Added in 3.9
338 Name.startswith("avx512.mask.pmovsx") || // Added in 4.0
339 Name.startswith("avx512.mask.pmovzx") || // Added in 4.0
340 Name == "sse2.cvtdq2pd" || // Added in 3.9
341 Name == "sse2.cvtps2pd" || // Added in 3.9
342 Name == "avx.cvtdq2.pd.256" || // Added in 3.9
343 Name == "avx.cvt.ps2.pd.256" || // Added in 3.9
344 Name.startswith("avx.vinsertf128.") || // Added in 3.7
345 Name == "avx2.vinserti128" || // Added in 3.7
346 Name.startswith("avx.vextractf128.") || // Added in 3.7
347 Name == "avx2.vextracti128" || // Added in 3.7
348 Name.startswith("sse4a.movnt.") || // Added in 3.9
349 Name.startswith("avx.movnt.") || // Added in 3.2
350 Name.startswith("avx512.storent.") || // Added in 3.9
351 Name == "sse2.storel.dq" || // Added in 3.9
352 Name.startswith("sse.storeu.") || // Added in 3.9
353 Name.startswith("sse2.storeu.") || // Added in 3.9
354 Name.startswith("avx.storeu.") || // Added in 3.9
355 Name.startswith("avx512.mask.storeu.") || // Added in 3.9
356 Name.startswith("avx512.mask.store.p") || // Added in 3.9
357 Name.startswith("avx512.mask.store.b.") || // Added in 3.9
358 Name.startswith("avx512.mask.store.w.") || // Added in 3.9
359 Name.startswith("avx512.mask.store.d.") || // Added in 3.9
360 Name.startswith("avx512.mask.store.q.") || // Added in 3.9
361 Name.startswith("avx512.mask.loadu.") || // Added in 3.9
362 Name.startswith("avx512.mask.load.") || // Added in 3.9
363 Name == "sse42.crc32.64.8" || // Added in 3.4
364 Name.startswith("avx.vbroadcast.s") || // Added in 3.5
365 Name.startswith("avx512.mask.palignr.") || // Added in 3.9
Craig Topperf57e17d2016-11-23 06:54:55 +0000366 Name.startswith("avx512.mask.valign.") || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000367 Name.startswith("sse2.psll.dq") || // Added in 3.7
368 Name.startswith("sse2.psrl.dq") || // Added in 3.7
369 Name.startswith("avx2.psll.dq") || // Added in 3.7
370 Name.startswith("avx2.psrl.dq") || // Added in 3.7
371 Name.startswith("avx512.psll.dq") || // Added in 3.9
372 Name.startswith("avx512.psrl.dq") || // Added in 3.9
373 Name == "sse41.pblendw" || // Added in 3.7
374 Name.startswith("sse41.blendp") || // Added in 3.7
375 Name.startswith("avx.blend.p") || // Added in 3.7
376 Name == "avx2.pblendw" || // Added in 3.7
377 Name.startswith("avx2.pblendd.") || // Added in 3.7
378 Name.startswith("avx.vbroadcastf128") || // Added in 4.0
379 Name == "avx2.vbroadcasti128" || // Added in 3.7
380 Name == "xop.vpcmov" || // Added in 3.8
Ayman Musa4d602432016-11-16 09:00:28 +0000381 Name.startswith("avx512.mask.move.s") || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000382 (Name.startswith("xop.vpcom") && // Added in 3.2
383 F->arg_size() == 2))) {
Craig Topperc6207612014-04-09 06:08:46 +0000384 NewFn = nullptr;
Craig Topper3b1817d2012-02-03 06:10:55 +0000385 return true;
386 }
Nadav Rotem17ee58a2012-06-10 18:42:51 +0000387 // SSE4.1 ptest functions may have an old signature.
Craig Toppere6915b82016-11-15 05:04:51 +0000388 if (IsX86 && Name.startswith("sse41.ptest")) { // Added in 3.2
Craig Topper5aebb862016-07-04 20:56:38 +0000389 if (Name.substr(11) == "c")
Nadav Rotem17ee58a2012-06-10 18:42:51 +0000390 return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestc, NewFn);
Craig Topper5aebb862016-07-04 20:56:38 +0000391 if (Name.substr(11) == "z")
Nadav Rotem17ee58a2012-06-10 18:42:51 +0000392 return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestz, NewFn);
Craig Topper5aebb862016-07-04 20:56:38 +0000393 if (Name.substr(11) == "nzc")
Nadav Rotem17ee58a2012-06-10 18:42:51 +0000394 return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestnzc, NewFn);
395 }
Sanjay Patel1c3eaec2015-02-28 22:25:06 +0000396 // Several blend and other instructions with masks used the wrong number of
Chandler Carruth373b2b12014-09-06 10:00:01 +0000397 // bits.
Craig Toppere6915b82016-11-15 05:04:51 +0000398 if (IsX86 && Name == "sse41.insertps") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000399 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_insertps,
400 NewFn);
Craig Toppere6915b82016-11-15 05:04:51 +0000401 if (IsX86 && Name == "sse41.dppd") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000402 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_dppd,
403 NewFn);
Craig Toppere6915b82016-11-15 05:04:51 +0000404 if (IsX86 && Name == "sse41.dpps") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000405 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_dpps,
406 NewFn);
Craig Toppere6915b82016-11-15 05:04:51 +0000407 if (IsX86 && Name == "sse41.mpsadbw") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000408 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_mpsadbw,
409 NewFn);
Craig Toppere6915b82016-11-15 05:04:51 +0000410 if (IsX86 && Name == "avx.dp.ps.256") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000411 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_avx_dp_ps_256,
412 NewFn);
Craig Toppere6915b82016-11-15 05:04:51 +0000413 if (IsX86 && Name == "avx2.mpsadbw") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000414 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_avx2_mpsadbw,
415 NewFn);
Craig Topper29f2e952015-01-25 23:26:02 +0000416
Craig Toppere6915b82016-11-15 05:04:51 +0000417 // frcz.ss/sd may need to have an argument dropped. Added in 3.2
Craig Topper5aebb862016-07-04 20:56:38 +0000418 if (IsX86 && Name.startswith("xop.vfrcz.ss") && F->arg_size() == 2) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000419 rename(F);
Craig Topper71dc02d2012-06-13 07:18:53 +0000420 NewFn = Intrinsic::getDeclaration(F->getParent(),
421 Intrinsic::x86_xop_vfrcz_ss);
422 return true;
423 }
Craig Topper5aebb862016-07-04 20:56:38 +0000424 if (IsX86 && Name.startswith("xop.vfrcz.sd") && F->arg_size() == 2) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000425 rename(F);
Craig Topper71dc02d2012-06-13 07:18:53 +0000426 NewFn = Intrinsic::getDeclaration(F->getParent(),
427 Intrinsic::x86_xop_vfrcz_sd);
428 return true;
429 }
Simon Pilgrime85506b2016-06-03 08:06:03 +0000430 // Upgrade any XOP PERMIL2 index operand still using a float/double vector.
Craig Toppere6915b82016-11-15 05:04:51 +0000431 if (IsX86 && Name.startswith("xop.vpermil2")) { // Added in 3.9
Simon Pilgrime85506b2016-06-03 08:06:03 +0000432 auto Params = F->getFunctionType()->params();
433 auto Idx = Params[2];
434 if (Idx->getScalarType()->isFloatingPointTy()) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000435 rename(F);
Simon Pilgrime85506b2016-06-03 08:06:03 +0000436 unsigned IdxSize = Idx->getPrimitiveSizeInBits();
437 unsigned EltSize = Idx->getScalarSizeInBits();
438 Intrinsic::ID Permil2ID;
439 if (EltSize == 64 && IdxSize == 128)
440 Permil2ID = Intrinsic::x86_xop_vpermil2pd;
441 else if (EltSize == 32 && IdxSize == 128)
442 Permil2ID = Intrinsic::x86_xop_vpermil2ps;
443 else if (EltSize == 64 && IdxSize == 256)
444 Permil2ID = Intrinsic::x86_xop_vpermil2pd_256;
445 else
446 Permil2ID = Intrinsic::x86_xop_vpermil2ps_256;
447 NewFn = Intrinsic::getDeclaration(F->getParent(), Permil2ID);
448 return true;
449 }
450 }
Craig Topper3b1817d2012-02-03 06:10:55 +0000451 break;
452 }
Chris Lattnerb372f662011-06-18 18:56:39 +0000453 }
Chandler Carruth7132e002007-08-04 01:51:18 +0000454
Nadav Rotem17ee58a2012-06-10 18:42:51 +0000455 // This may not belong here. This function is effectively being overloaded
456 // to both detect an intrinsic which needs upgrading, and to provide the
457 // upgraded form of the intrinsic. We should perhaps have two separate
Chandler Carruth7132e002007-08-04 01:51:18 +0000458 // functions for this.
Evan Cheng0e179d02007-12-17 22:33:23 +0000459 return false;
Chandler Carruth7132e002007-08-04 01:51:18 +0000460}
461
Evan Cheng0e179d02007-12-17 22:33:23 +0000462bool llvm::UpgradeIntrinsicFunction(Function *F, Function *&NewFn) {
Craig Topperc6207612014-04-09 06:08:46 +0000463 NewFn = nullptr;
Evan Cheng0e179d02007-12-17 22:33:23 +0000464 bool Upgraded = UpgradeIntrinsicFunction1(F, NewFn);
Filipe Cabecinhas0011c582015-07-03 20:12:01 +0000465 assert(F != NewFn && "Intrinsic function upgraded to the same function");
Duncan Sands38ef3a82007-12-03 20:06:50 +0000466
467 // Upgrade intrinsic attributes. This does not change the function.
Evan Cheng0e179d02007-12-17 22:33:23 +0000468 if (NewFn)
469 F = NewFn;
Pete Cooper9e1d3352015-05-20 17:16:39 +0000470 if (Intrinsic::ID id = F->getIntrinsicID())
471 F->setAttributes(Intrinsic::getAttributes(F->getContext(), id));
Duncan Sands38ef3a82007-12-03 20:06:50 +0000472 return Upgraded;
473}
474
Bill Wendlinge26fffc2010-09-10 18:51:56 +0000475bool llvm::UpgradeGlobalVariable(GlobalVariable *GV) {
Chris Lattner80ed9dc2011-06-18 06:05:24 +0000476 // Nothing to do yet.
Bill Wendlinge26fffc2010-09-10 18:51:56 +0000477 return false;
478}
479
Simon Pilgrimf7186822016-06-09 21:09:03 +0000480// Handles upgrading SSE2/AVX2/AVX512BW PSLLDQ intrinsics by converting them
Craig Topperb324e432015-02-18 06:24:44 +0000481// to byte shuffles.
Craig Topper46b34fe2016-07-12 01:42:33 +0000482static Value *UpgradeX86PSLLDQIntrinsics(IRBuilder<> &Builder,
Craig Topper7355ac32016-05-29 06:37:33 +0000483 Value *Op, unsigned Shift) {
484 Type *ResultTy = Op->getType();
485 unsigned NumElts = ResultTy->getVectorNumElements() * 8;
Craig Topperb324e432015-02-18 06:24:44 +0000486
487 // Bitcast from a 64-bit element type to a byte element type.
Craig Topper46b34fe2016-07-12 01:42:33 +0000488 Type *VecTy = VectorType::get(Builder.getInt8Ty(), NumElts);
Craig Topper7355ac32016-05-29 06:37:33 +0000489 Op = Builder.CreateBitCast(Op, VecTy, "cast");
490
Craig Topperb324e432015-02-18 06:24:44 +0000491 // We'll be shuffling in zeroes.
Craig Topper7355ac32016-05-29 06:37:33 +0000492 Value *Res = Constant::getNullValue(VecTy);
Craig Topperb324e432015-02-18 06:24:44 +0000493
494 // If shift is less than 16, emit a shuffle to move the bytes. Otherwise,
495 // we'll just return the zero vector.
496 if (Shift < 16) {
Craig Topper99d1eab2016-06-12 00:41:19 +0000497 uint32_t Idxs[64];
Simon Pilgrimf7186822016-06-09 21:09:03 +0000498 // 256/512-bit version is split into 2/4 16-byte lanes.
Craig Topperb324e432015-02-18 06:24:44 +0000499 for (unsigned l = 0; l != NumElts; l += 16)
500 for (unsigned i = 0; i != 16; ++i) {
501 unsigned Idx = NumElts + i - Shift;
502 if (Idx < NumElts)
503 Idx -= NumElts - 16; // end of lane, switch operand.
Craig Topper7355ac32016-05-29 06:37:33 +0000504 Idxs[l + i] = Idx + l;
Craig Topperb324e432015-02-18 06:24:44 +0000505 }
506
Craig Topper7355ac32016-05-29 06:37:33 +0000507 Res = Builder.CreateShuffleVector(Res, Op, makeArrayRef(Idxs, NumElts));
Craig Topperb324e432015-02-18 06:24:44 +0000508 }
509
510 // Bitcast back to a 64-bit element type.
Craig Topper7355ac32016-05-29 06:37:33 +0000511 return Builder.CreateBitCast(Res, ResultTy, "cast");
Craig Topperb324e432015-02-18 06:24:44 +0000512}
513
Craig Topperea703ae2016-06-13 02:36:42 +0000514// Handles upgrading SSE2/AVX2/AVX512BW PSRLDQ intrinsics by converting them
515// to byte shuffles.
Craig Topper46b34fe2016-07-12 01:42:33 +0000516static Value *UpgradeX86PSRLDQIntrinsics(IRBuilder<> &Builder, Value *Op,
Craig Topperea703ae2016-06-13 02:36:42 +0000517 unsigned Shift) {
518 Type *ResultTy = Op->getType();
519 unsigned NumElts = ResultTy->getVectorNumElements() * 8;
520
521 // Bitcast from a 64-bit element type to a byte element type.
Craig Topper46b34fe2016-07-12 01:42:33 +0000522 Type *VecTy = VectorType::get(Builder.getInt8Ty(), NumElts);
Craig Topperea703ae2016-06-13 02:36:42 +0000523 Op = Builder.CreateBitCast(Op, VecTy, "cast");
524
525 // We'll be shuffling in zeroes.
526 Value *Res = Constant::getNullValue(VecTy);
527
528 // If shift is less than 16, emit a shuffle to move the bytes. Otherwise,
529 // we'll just return the zero vector.
530 if (Shift < 16) {
531 uint32_t Idxs[64];
532 // 256/512-bit version is split into 2/4 16-byte lanes.
533 for (unsigned l = 0; l != NumElts; l += 16)
534 for (unsigned i = 0; i != 16; ++i) {
535 unsigned Idx = i + Shift;
536 if (Idx >= 16)
537 Idx += NumElts - 16; // end of lane, switch operand.
538 Idxs[l + i] = Idx + l;
539 }
540
541 Res = Builder.CreateShuffleVector(Op, Res, makeArrayRef(Idxs, NumElts));
542 }
543
544 // Bitcast back to a 64-bit element type.
545 return Builder.CreateBitCast(Res, ResultTy, "cast");
546}
547
548static Value *getX86MaskVec(IRBuilder<> &Builder, Value *Mask,
549 unsigned NumElts) {
550 llvm::VectorType *MaskTy = llvm::VectorType::get(Builder.getInt1Ty(),
551 cast<IntegerType>(Mask->getType())->getBitWidth());
552 Mask = Builder.CreateBitCast(Mask, MaskTy);
553
554 // If we have less than 8 elements, then the starting mask was an i8 and
555 // we need to extract down to the right number of elements.
556 if (NumElts < 8) {
557 uint32_t Indices[4];
558 for (unsigned i = 0; i != NumElts; ++i)
559 Indices[i] = i;
560 Mask = Builder.CreateShuffleVector(Mask, Mask,
561 makeArrayRef(Indices, NumElts),
562 "extract");
563 }
564
565 return Mask;
566}
567
568static Value *EmitX86Select(IRBuilder<> &Builder, Value *Mask,
569 Value *Op0, Value *Op1) {
570 // If the mask is all ones just emit the align operation.
571 if (const auto *C = dyn_cast<Constant>(Mask))
572 if (C->isAllOnesValue())
573 return Op0;
574
575 Mask = getX86MaskVec(Builder, Mask, Op0->getType()->getVectorNumElements());
576 return Builder.CreateSelect(Mask, Op0, Op1);
577}
578
Craig Topperf57e17d2016-11-23 06:54:55 +0000579// Handle autoupgrade for masked PALIGNR and VALIGND/Q intrinsics.
580// PALIGNR handles large immediates by shifting while VALIGN masks the immediate
581// so we need to handle both cases. VALIGN also doesn't have 128-bit lanes.
582static Value *UpgradeX86ALIGNIntrinsics(IRBuilder<> &Builder, Value *Op0,
583 Value *Op1, Value *Shift,
584 Value *Passthru, Value *Mask,
585 bool IsVALIGN) {
Craig Topper33350cc2016-06-06 06:12:54 +0000586 unsigned ShiftVal = cast<llvm::ConstantInt>(Shift)->getZExtValue();
587
588 unsigned NumElts = Op0->getType()->getVectorNumElements();
Craig Topperf57e17d2016-11-23 06:54:55 +0000589 assert((IsVALIGN || NumElts % 16 == 0) && "Illegal NumElts for PALIGNR!");
590 assert((!IsVALIGN || NumElts <= 16) && "NumElts too large for VALIGN!");
591 assert(isPowerOf2_32(NumElts) && "NumElts not a power of 2!");
592
593 // Mask the immediate for VALIGN.
594 if (IsVALIGN)
595 ShiftVal &= (NumElts - 1);
Craig Topper33350cc2016-06-06 06:12:54 +0000596
597 // If palignr is shifting the pair of vectors more than the size of two
598 // lanes, emit zero.
599 if (ShiftVal >= 32)
600 return llvm::Constant::getNullValue(Op0->getType());
601
602 // If palignr is shifting the pair of input vectors more than one lane,
603 // but less than two lanes, convert to shifting in zeroes.
604 if (ShiftVal > 16) {
605 ShiftVal -= 16;
606 Op1 = Op0;
607 Op0 = llvm::Constant::getNullValue(Op0->getType());
608 }
609
Craig Topper99d1eab2016-06-12 00:41:19 +0000610 uint32_t Indices[64];
Craig Topper33350cc2016-06-06 06:12:54 +0000611 // 256-bit palignr operates on 128-bit lanes so we need to handle that
Craig Topperf57e17d2016-11-23 06:54:55 +0000612 for (unsigned l = 0; l < NumElts; l += 16) {
Craig Topper33350cc2016-06-06 06:12:54 +0000613 for (unsigned i = 0; i != 16; ++i) {
614 unsigned Idx = ShiftVal + i;
Craig Topperf57e17d2016-11-23 06:54:55 +0000615 if (!IsVALIGN && Idx >= 16) // Disable wrap for VALIGN.
Craig Topper33350cc2016-06-06 06:12:54 +0000616 Idx += NumElts - 16; // End of lane, switch operand.
617 Indices[l + i] = Idx + l;
618 }
619 }
620
621 Value *Align = Builder.CreateShuffleVector(Op1, Op0,
622 makeArrayRef(Indices, NumElts),
623 "palignr");
624
Craig Topperea703ae2016-06-13 02:36:42 +0000625 return EmitX86Select(Builder, Mask, Align, Passthru);
Craig Topperb324e432015-02-18 06:24:44 +0000626}
627
Craig Topper46b34fe2016-07-12 01:42:33 +0000628static Value *UpgradeMaskedStore(IRBuilder<> &Builder,
Craig Topper50f85c22016-05-31 01:50:02 +0000629 Value *Ptr, Value *Data, Value *Mask,
630 bool Aligned) {
631 // Cast the pointer to the right type.
632 Ptr = Builder.CreateBitCast(Ptr,
633 llvm::PointerType::getUnqual(Data->getType()));
634 unsigned Align =
635 Aligned ? cast<VectorType>(Data->getType())->getBitWidth() / 8 : 1;
636
637 // If the mask is all ones just emit a regular store.
638 if (const auto *C = dyn_cast<Constant>(Mask))
639 if (C->isAllOnesValue())
640 return Builder.CreateAlignedStore(Data, Ptr, Align);
641
642 // Convert the mask from an integer type to a vector of i1.
643 unsigned NumElts = Data->getType()->getVectorNumElements();
Craig Topperea703ae2016-06-13 02:36:42 +0000644 Mask = getX86MaskVec(Builder, Mask, NumElts);
Craig Topper50f85c22016-05-31 01:50:02 +0000645 return Builder.CreateMaskedStore(Data, Ptr, Align, Mask);
646}
647
Craig Topper46b34fe2016-07-12 01:42:33 +0000648static Value *UpgradeMaskedLoad(IRBuilder<> &Builder,
Craig Topperf10fbfa2016-06-02 04:19:36 +0000649 Value *Ptr, Value *Passthru, Value *Mask,
650 bool Aligned) {
651 // Cast the pointer to the right type.
652 Ptr = Builder.CreateBitCast(Ptr,
653 llvm::PointerType::getUnqual(Passthru->getType()));
654 unsigned Align =
655 Aligned ? cast<VectorType>(Passthru->getType())->getBitWidth() / 8 : 1;
656
657 // If the mask is all ones just emit a regular store.
658 if (const auto *C = dyn_cast<Constant>(Mask))
659 if (C->isAllOnesValue())
660 return Builder.CreateAlignedLoad(Ptr, Align);
661
662 // Convert the mask from an integer type to a vector of i1.
663 unsigned NumElts = Passthru->getType()->getVectorNumElements();
Craig Topperea703ae2016-06-13 02:36:42 +0000664 Mask = getX86MaskVec(Builder, Mask, NumElts);
Craig Topperf10fbfa2016-06-02 04:19:36 +0000665 return Builder.CreateMaskedLoad(Ptr, Align, Mask, Passthru);
666}
667
Sanjay Patel51ab7572016-06-16 15:48:30 +0000668static Value *upgradeIntMinMax(IRBuilder<> &Builder, CallInst &CI,
669 ICmpInst::Predicate Pred) {
670 Value *Op0 = CI.getArgOperand(0);
671 Value *Op1 = CI.getArgOperand(1);
672 Value *Cmp = Builder.CreateICmp(Pred, Op0, Op1);
Craig Topper8ec5c732016-10-24 04:04:16 +0000673 Value *Res = Builder.CreateSelect(Cmp, Op0, Op1);
674
675 if (CI.getNumArgOperands() == 4)
676 Res = EmitX86Select(Builder, CI.getArgOperand(3), Res, CI.getArgOperand(2));
677
678 return Res;
Sanjay Patel51ab7572016-06-16 15:48:30 +0000679}
680
Craig Topper0a0fb0f2016-06-21 03:53:24 +0000681static Value *upgradeMaskedCompare(IRBuilder<> &Builder, CallInst &CI,
682 ICmpInst::Predicate Pred) {
683 Value *Op0 = CI.getArgOperand(0);
684 unsigned NumElts = Op0->getType()->getVectorNumElements();
685 Value *Cmp = Builder.CreateICmp(Pred, Op0, CI.getArgOperand(1));
686
687 Value *Mask = CI.getArgOperand(2);
688 const auto *C = dyn_cast<Constant>(Mask);
689 if (!C || !C->isAllOnesValue())
690 Cmp = Builder.CreateAnd(Cmp, getX86MaskVec(Builder, Mask, NumElts));
691
692 if (NumElts < 8) {
693 uint32_t Indices[8];
694 for (unsigned i = 0; i != NumElts; ++i)
695 Indices[i] = i;
696 for (unsigned i = NumElts; i != 8; ++i)
Craig Topperd5d2a352016-07-07 06:11:07 +0000697 Indices[i] = NumElts + i % NumElts;
698 Cmp = Builder.CreateShuffleVector(Cmp,
699 Constant::getNullValue(Cmp->getType()),
Craig Topper0a0fb0f2016-06-21 03:53:24 +0000700 Indices);
701 }
702 return Builder.CreateBitCast(Cmp, IntegerType::get(CI.getContext(),
703 std::max(NumElts, 8U)));
704}
705
Craig Topperc9467ed2016-11-06 16:29:08 +0000706// Replace a masked intrinsic with an older unmasked intrinsic.
707static Value *UpgradeX86MaskedShift(IRBuilder<> &Builder, CallInst &CI,
708 Intrinsic::ID IID) {
709 Function *F = CI.getCalledFunction();
710 Function *Intrin = Intrinsic::getDeclaration(F->getParent(), IID);
711 Value *Rep = Builder.CreateCall(Intrin,
712 { CI.getArgOperand(0), CI.getArgOperand(1) });
713 return EmitX86Select(Builder, CI.getArgOperand(3), Rep, CI.getArgOperand(2));
714}
715
Ayman Musa4d602432016-11-16 09:00:28 +0000716static Value* upgradeMaskedMove(IRBuilder<> &Builder, CallInst &CI) {
717 Value* A = CI.getArgOperand(0);
718 Value* B = CI.getArgOperand(1);
719 Value* Src = CI.getArgOperand(2);
720 Value* Mask = CI.getArgOperand(3);
721
722 Value* AndNode = Builder.CreateAnd(Mask, APInt(8, 1));
723 Value* Cmp = Builder.CreateIsNotNull(AndNode);
724 Value* Extract1 = Builder.CreateExtractElement(B, (uint64_t)0);
725 Value* Extract2 = Builder.CreateExtractElement(Src, (uint64_t)0);
726 Value* Select = Builder.CreateSelect(Cmp, Extract1, Extract2);
727 return Builder.CreateInsertElement(A, Select, (uint64_t)0);
728}
Craig Topperc9467ed2016-11-06 16:29:08 +0000729
Sanjay Patel595098f2016-06-15 22:01:28 +0000730/// Upgrade a call to an old intrinsic. All argument and return casting must be
731/// provided to seamlessly integrate with existing context.
Chandler Carruth7132e002007-08-04 01:51:18 +0000732void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
Craig Topper3b1817d2012-02-03 06:10:55 +0000733 Function *F = CI->getCalledFunction();
Nick Lewycky2eb3ade2011-12-12 22:59:34 +0000734 LLVMContext &C = CI->getContext();
Chandler Carruth58a71ed2011-12-12 04:26:04 +0000735 IRBuilder<> Builder(C);
Duncan P. N. Exon Smith52888a62015-10-08 23:49:46 +0000736 Builder.SetInsertPoint(CI->getParent(), CI->getIterator());
Chandler Carruth58a71ed2011-12-12 04:26:04 +0000737
Craig Topper3b1817d2012-02-03 06:10:55 +0000738 assert(F && "Intrinsic call is not direct?");
739
740 if (!NewFn) {
741 // Get the Function's name.
742 StringRef Name = F->getName();
743
Craig Topper5aebb862016-07-04 20:56:38 +0000744 assert(Name.startswith("llvm.") && "Intrinsic doesn't start with 'llvm.'");
745 Name = Name.substr(5);
746
747 bool IsX86 = Name.startswith("x86.");
748 if (IsX86)
749 Name = Name.substr(4);
750
Craig Toppera4744d12016-12-10 21:15:48 +0000751 if (IsX86 && Name.startswith("sse4a.movnt.")) {
752 Module *M = F->getParent();
753 SmallVector<Metadata *, 1> Elts;
754 Elts.push_back(
755 ConstantAsMetadata::get(ConstantInt::get(Type::getInt32Ty(C), 1)));
756 MDNode *Node = MDNode::get(C, Elts);
757
758 Value *Arg0 = CI->getArgOperand(0);
759 Value *Arg1 = CI->getArgOperand(1);
760
761 // Nontemporal (unaligned) store of the 0'th element of the float/double
762 // vector.
763 Type *SrcEltTy = cast<VectorType>(Arg1->getType())->getElementType();
764 PointerType *EltPtrTy = PointerType::getUnqual(SrcEltTy);
765 Value *Addr = Builder.CreateBitCast(Arg0, EltPtrTy, "cast");
766 Value *Extract =
767 Builder.CreateExtractElement(Arg1, (uint64_t)0, "extractelement");
768
769 StoreInst *SI = Builder.CreateAlignedStore(Extract, Addr, 1);
770 SI->setMetadata(M->getMDKindID("nontemporal"), Node);
771
772 // Remove intrinsic.
773 CI->eraseFromParent();
774 return;
775 }
776
777 if (IsX86 && (Name.startswith("avx.movnt.") ||
778 Name.startswith("avx512.storent."))) {
779 Module *M = F->getParent();
780 SmallVector<Metadata *, 1> Elts;
781 Elts.push_back(
782 ConstantAsMetadata::get(ConstantInt::get(Type::getInt32Ty(C), 1)));
783 MDNode *Node = MDNode::get(C, Elts);
784
785 Value *Arg0 = CI->getArgOperand(0);
786 Value *Arg1 = CI->getArgOperand(1);
787
788 // Convert the type of the pointer to a pointer to the stored type.
789 Value *BC = Builder.CreateBitCast(Arg0,
790 PointerType::getUnqual(Arg1->getType()),
791 "cast");
792 VectorType *VTy = cast<VectorType>(Arg1->getType());
793 StoreInst *SI = Builder.CreateAlignedStore(Arg1, BC,
794 VTy->getBitWidth() / 8);
795 SI->setMetadata(M->getMDKindID("nontemporal"), Node);
796
797 // Remove intrinsic.
798 CI->eraseFromParent();
799 return;
800 }
801
802 if (IsX86 && Name == "sse2.storel.dq") {
803 Value *Arg0 = CI->getArgOperand(0);
804 Value *Arg1 = CI->getArgOperand(1);
805
806 Type *NewVecTy = VectorType::get(Type::getInt64Ty(C), 2);
807 Value *BC0 = Builder.CreateBitCast(Arg1, NewVecTy, "cast");
808 Value *Elt = Builder.CreateExtractElement(BC0, (uint64_t)0);
809 Value *BC = Builder.CreateBitCast(Arg0,
810 PointerType::getUnqual(Elt->getType()),
811 "cast");
812 Builder.CreateAlignedStore(Elt, BC, 1);
813
814 // Remove intrinsic.
815 CI->eraseFromParent();
816 return;
817 }
818
819 if (IsX86 && (Name.startswith("sse.storeu.") ||
820 Name.startswith("sse2.storeu.") ||
821 Name.startswith("avx.storeu."))) {
822 Value *Arg0 = CI->getArgOperand(0);
823 Value *Arg1 = CI->getArgOperand(1);
824
825 Arg0 = Builder.CreateBitCast(Arg0,
826 PointerType::getUnqual(Arg1->getType()),
827 "cast");
828 Builder.CreateAlignedStore(Arg1, Arg0, 1);
829
830 // Remove intrinsic.
831 CI->eraseFromParent();
832 return;
833 }
834
835 if (IsX86 && (Name.startswith("avx512.mask.storeu."))) {
836 UpgradeMaskedStore(Builder, CI->getArgOperand(0), CI->getArgOperand(1),
837 CI->getArgOperand(2), /*Aligned*/false);
838
839 // Remove intrinsic.
840 CI->eraseFromParent();
841 return;
842 }
843
844 if (IsX86 && (Name.startswith("avx512.mask.store."))) {
845 UpgradeMaskedStore(Builder, CI->getArgOperand(0), CI->getArgOperand(1),
846 CI->getArgOperand(2), /*Aligned*/true);
847
848 // Remove intrinsic.
849 CI->eraseFromParent();
850 return;
851 }
852
Craig Topper3b1817d2012-02-03 06:10:55 +0000853 Value *Rep;
Sanjay Patel595098f2016-06-15 22:01:28 +0000854 // Upgrade packed integer vector compare intrinsics to compare instructions.
Craig Topper5aebb862016-07-04 20:56:38 +0000855 if (IsX86 && (Name.startswith("sse2.pcmpeq.") ||
856 Name.startswith("avx2.pcmpeq."))) {
Craig Topper3b1817d2012-02-03 06:10:55 +0000857 Rep = Builder.CreateICmpEQ(CI->getArgOperand(0), CI->getArgOperand(1),
858 "pcmpeq");
Craig Topper3b1817d2012-02-03 06:10:55 +0000859 Rep = Builder.CreateSExt(Rep, CI->getType(), "");
Craig Topper5aebb862016-07-04 20:56:38 +0000860 } else if (IsX86 && (Name.startswith("sse2.pcmpgt.") ||
861 Name.startswith("avx2.pcmpgt."))) {
Craig Topper3b1817d2012-02-03 06:10:55 +0000862 Rep = Builder.CreateICmpSGT(CI->getArgOperand(0), CI->getArgOperand(1),
863 "pcmpgt");
Craig Topper3b1817d2012-02-03 06:10:55 +0000864 Rep = Builder.CreateSExt(Rep, CI->getType(), "");
Craig Topper6910fa02016-11-16 05:24:10 +0000865 } else if (IsX86 && (Name == "sse.add.ss" || Name == "sse2.add.sd")) {
866 Type *I32Ty = Type::getInt32Ty(C);
867 Value *Elt0 = Builder.CreateExtractElement(CI->getArgOperand(0),
868 ConstantInt::get(I32Ty, 0));
869 Value *Elt1 = Builder.CreateExtractElement(CI->getArgOperand(1),
870 ConstantInt::get(I32Ty, 0));
871 Rep = Builder.CreateInsertElement(CI->getArgOperand(0),
872 Builder.CreateFAdd(Elt0, Elt1),
873 ConstantInt::get(I32Ty, 0));
874 } else if (IsX86 && (Name == "sse.sub.ss" || Name == "sse2.sub.sd")) {
875 Type *I32Ty = Type::getInt32Ty(C);
876 Value *Elt0 = Builder.CreateExtractElement(CI->getArgOperand(0),
877 ConstantInt::get(I32Ty, 0));
878 Value *Elt1 = Builder.CreateExtractElement(CI->getArgOperand(1),
879 ConstantInt::get(I32Ty, 0));
880 Rep = Builder.CreateInsertElement(CI->getArgOperand(0),
881 Builder.CreateFSub(Elt0, Elt1),
882 ConstantInt::get(I32Ty, 0));
883 } else if (IsX86 && (Name == "sse.mul.ss" || Name == "sse2.mul.sd")) {
884 Type *I32Ty = Type::getInt32Ty(C);
885 Value *Elt0 = Builder.CreateExtractElement(CI->getArgOperand(0),
886 ConstantInt::get(I32Ty, 0));
887 Value *Elt1 = Builder.CreateExtractElement(CI->getArgOperand(1),
888 ConstantInt::get(I32Ty, 0));
889 Rep = Builder.CreateInsertElement(CI->getArgOperand(0),
890 Builder.CreateFMul(Elt0, Elt1),
891 ConstantInt::get(I32Ty, 0));
892 } else if (IsX86 && (Name == "sse.div.ss" || Name == "sse2.div.sd")) {
893 Type *I32Ty = Type::getInt32Ty(C);
894 Value *Elt0 = Builder.CreateExtractElement(CI->getArgOperand(0),
895 ConstantInt::get(I32Ty, 0));
896 Value *Elt1 = Builder.CreateExtractElement(CI->getArgOperand(1),
897 ConstantInt::get(I32Ty, 0));
898 Rep = Builder.CreateInsertElement(CI->getArgOperand(0),
899 Builder.CreateFDiv(Elt0, Elt1),
900 ConstantInt::get(I32Ty, 0));
Craig Topper5aebb862016-07-04 20:56:38 +0000901 } else if (IsX86 && Name.startswith("avx512.mask.pcmpeq.")) {
Craig Topper0a0fb0f2016-06-21 03:53:24 +0000902 Rep = upgradeMaskedCompare(Builder, *CI, ICmpInst::ICMP_EQ);
Craig Topper5aebb862016-07-04 20:56:38 +0000903 } else if (IsX86 && Name.startswith("avx512.mask.pcmpgt.")) {
Craig Topper0a0fb0f2016-06-21 03:53:24 +0000904 Rep = upgradeMaskedCompare(Builder, *CI, ICmpInst::ICMP_SGT);
Craig Topper5aebb862016-07-04 20:56:38 +0000905 } else if (IsX86 && (Name == "sse41.pmaxsb" ||
906 Name == "sse2.pmaxs.w" ||
907 Name == "sse41.pmaxsd" ||
Craig Topper8ec5c732016-10-24 04:04:16 +0000908 Name.startswith("avx2.pmaxs") ||
909 Name.startswith("avx512.mask.pmaxs"))) {
Sanjay Patel51ab7572016-06-16 15:48:30 +0000910 Rep = upgradeIntMinMax(Builder, *CI, ICmpInst::ICMP_SGT);
Craig Topper5aebb862016-07-04 20:56:38 +0000911 } else if (IsX86 && (Name == "sse2.pmaxu.b" ||
912 Name == "sse41.pmaxuw" ||
913 Name == "sse41.pmaxud" ||
Craig Topper8ec5c732016-10-24 04:04:16 +0000914 Name.startswith("avx2.pmaxu") ||
915 Name.startswith("avx512.mask.pmaxu"))) {
Sanjay Patel51ab7572016-06-16 15:48:30 +0000916 Rep = upgradeIntMinMax(Builder, *CI, ICmpInst::ICMP_UGT);
Craig Topper5aebb862016-07-04 20:56:38 +0000917 } else if (IsX86 && (Name == "sse41.pminsb" ||
918 Name == "sse2.pmins.w" ||
919 Name == "sse41.pminsd" ||
Craig Topper8ec5c732016-10-24 04:04:16 +0000920 Name.startswith("avx2.pmins") ||
921 Name.startswith("avx512.mask.pmins"))) {
Sanjay Patel51ab7572016-06-16 15:48:30 +0000922 Rep = upgradeIntMinMax(Builder, *CI, ICmpInst::ICMP_SLT);
Craig Topper5aebb862016-07-04 20:56:38 +0000923 } else if (IsX86 && (Name == "sse2.pminu.b" ||
924 Name == "sse41.pminuw" ||
925 Name == "sse41.pminud" ||
Craig Topper8ec5c732016-10-24 04:04:16 +0000926 Name.startswith("avx2.pminu") ||
927 Name.startswith("avx512.mask.pminu"))) {
Sanjay Patel51ab7572016-06-16 15:48:30 +0000928 Rep = upgradeIntMinMax(Builder, *CI, ICmpInst::ICMP_ULT);
Craig Topper5aebb862016-07-04 20:56:38 +0000929 } else if (IsX86 && (Name == "sse2.cvtdq2pd" ||
930 Name == "sse2.cvtps2pd" ||
931 Name == "avx.cvtdq2.pd.256" ||
Simon Pilgrimb57dd172016-11-16 14:48:32 +0000932 Name == "avx.cvt.ps2.pd.256" ||
933 Name.startswith("avx512.mask.cvtdq2pd.") ||
934 Name.startswith("avx512.mask.cvtudq2pd."))) {
Simon Pilgrim4298d062016-05-25 08:59:18 +0000935 // Lossless i32/float to double conversion.
936 // Extract the bottom elements if necessary and convert to double vector.
937 Value *Src = CI->getArgOperand(0);
938 VectorType *SrcTy = cast<VectorType>(Src->getType());
939 VectorType *DstTy = cast<VectorType>(CI->getType());
940 Rep = CI->getArgOperand(0);
941
942 unsigned NumDstElts = DstTy->getNumElements();
943 if (NumDstElts < SrcTy->getNumElements()) {
944 assert(NumDstElts == 2 && "Unexpected vector size");
Craig Topper99d1eab2016-06-12 00:41:19 +0000945 uint32_t ShuffleMask[2] = { 0, 1 };
946 Rep = Builder.CreateShuffleVector(Rep, UndefValue::get(SrcTy),
947 ShuffleMask);
Simon Pilgrim4298d062016-05-25 08:59:18 +0000948 }
949
Simon Pilgrimb57dd172016-11-16 14:48:32 +0000950 bool SInt2Double = (StringRef::npos != Name.find("cvtdq2"));
951 bool UInt2Double = (StringRef::npos != Name.find("cvtudq2"));
952 if (SInt2Double)
Simon Pilgrim4298d062016-05-25 08:59:18 +0000953 Rep = Builder.CreateSIToFP(Rep, DstTy, "cvtdq2pd");
Simon Pilgrimb57dd172016-11-16 14:48:32 +0000954 else if (UInt2Double)
955 Rep = Builder.CreateUIToFP(Rep, DstTy, "cvtudq2pd");
Simon Pilgrim4298d062016-05-25 08:59:18 +0000956 else
957 Rep = Builder.CreateFPExt(Rep, DstTy, "cvtps2pd");
Simon Pilgrimb57dd172016-11-16 14:48:32 +0000958
959 if (CI->getNumArgOperands() == 3)
960 Rep = EmitX86Select(Builder, CI->getArgOperand(2), Rep,
961 CI->getArgOperand(1));
Craig Toppera57d2ca2016-09-03 23:55:13 +0000962 } else if (IsX86 && (Name.startswith("avx512.mask.loadu."))) {
Craig Topper46b34fe2016-07-12 01:42:33 +0000963 Rep = UpgradeMaskedLoad(Builder, CI->getArgOperand(0),
Craig Topperf10fbfa2016-06-02 04:19:36 +0000964 CI->getArgOperand(1), CI->getArgOperand(2),
965 /*Aligned*/false);
Craig Toppera57d2ca2016-09-03 23:55:13 +0000966 } else if (IsX86 && (Name.startswith("avx512.mask.load."))) {
Craig Topper46b34fe2016-07-12 01:42:33 +0000967 Rep = UpgradeMaskedLoad(Builder, CI->getArgOperand(0),
Craig Topperf10fbfa2016-06-02 04:19:36 +0000968 CI->getArgOperand(1),CI->getArgOperand(2),
969 /*Aligned*/true);
Craig Topper5aebb862016-07-04 20:56:38 +0000970 } else if (IsX86 && Name.startswith("xop.vpcom")) {
Craig Topper3352ba52012-06-09 16:46:13 +0000971 Intrinsic::ID intID;
972 if (Name.endswith("ub"))
973 intID = Intrinsic::x86_xop_vpcomub;
974 else if (Name.endswith("uw"))
975 intID = Intrinsic::x86_xop_vpcomuw;
976 else if (Name.endswith("ud"))
977 intID = Intrinsic::x86_xop_vpcomud;
978 else if (Name.endswith("uq"))
979 intID = Intrinsic::x86_xop_vpcomuq;
980 else if (Name.endswith("b"))
981 intID = Intrinsic::x86_xop_vpcomb;
982 else if (Name.endswith("w"))
983 intID = Intrinsic::x86_xop_vpcomw;
984 else if (Name.endswith("d"))
985 intID = Intrinsic::x86_xop_vpcomd;
986 else if (Name.endswith("q"))
987 intID = Intrinsic::x86_xop_vpcomq;
988 else
989 llvm_unreachable("Unknown suffix");
990
Craig Topper5aebb862016-07-04 20:56:38 +0000991 Name = Name.substr(9); // strip off "xop.vpcom"
Craig Topper3352ba52012-06-09 16:46:13 +0000992 unsigned Imm;
993 if (Name.startswith("lt"))
994 Imm = 0;
995 else if (Name.startswith("le"))
996 Imm = 1;
997 else if (Name.startswith("gt"))
998 Imm = 2;
999 else if (Name.startswith("ge"))
1000 Imm = 3;
1001 else if (Name.startswith("eq"))
1002 Imm = 4;
1003 else if (Name.startswith("ne"))
1004 Imm = 5;
Craig Topper3352ba52012-06-09 16:46:13 +00001005 else if (Name.startswith("false"))
Craig Toppere32546d2015-02-13 07:42:15 +00001006 Imm = 6;
1007 else if (Name.startswith("true"))
Craig Topper3352ba52012-06-09 16:46:13 +00001008 Imm = 7;
1009 else
1010 llvm_unreachable("Unknown condition");
1011
1012 Function *VPCOM = Intrinsic::getDeclaration(F->getParent(), intID);
David Blaikieff6409d2015-05-18 22:13:54 +00001013 Rep =
1014 Builder.CreateCall(VPCOM, {CI->getArgOperand(0), CI->getArgOperand(1),
1015 Builder.getInt8(Imm)});
Craig Topper5aebb862016-07-04 20:56:38 +00001016 } else if (IsX86 && Name == "xop.vpcmov") {
Simon Pilgrime88dc042015-11-03 20:27:01 +00001017 Value *Arg0 = CI->getArgOperand(0);
1018 Value *Arg1 = CI->getArgOperand(1);
1019 Value *Sel = CI->getArgOperand(2);
1020 unsigned NumElts = CI->getType()->getVectorNumElements();
1021 Constant *MinusOne = ConstantVector::getSplat(NumElts, Builder.getInt64(-1));
1022 Value *NotSel = Builder.CreateXor(Sel, MinusOne);
1023 Value *Sel0 = Builder.CreateAnd(Arg0, Sel);
1024 Value *Sel1 = Builder.CreateAnd(Arg1, NotSel);
1025 Rep = Builder.CreateOr(Sel0, Sel1);
Craig Topper5aebb862016-07-04 20:56:38 +00001026 } else if (IsX86 && Name == "sse42.crc32.64.8") {
Craig Topperef9e9932013-10-15 05:20:47 +00001027 Function *CRC32 = Intrinsic::getDeclaration(F->getParent(),
1028 Intrinsic::x86_sse42_crc32_32_8);
1029 Value *Trunc0 = Builder.CreateTrunc(CI->getArgOperand(0), Type::getInt32Ty(C));
David Blaikieff6409d2015-05-18 22:13:54 +00001030 Rep = Builder.CreateCall(CRC32, {Trunc0, CI->getArgOperand(1)});
Craig Topperef9e9932013-10-15 05:20:47 +00001031 Rep = Builder.CreateZExt(Rep, CI->getType(), "");
Simon Pilgrimea0d4f92016-07-22 13:58:44 +00001032 } else if (IsX86 && Name.startswith("avx.vbroadcast.s")) {
Adam Nemet39066802014-05-29 23:35:33 +00001033 // Replace broadcasts with a series of insertelements.
1034 Type *VecTy = CI->getType();
1035 Type *EltTy = VecTy->getVectorElementType();
1036 unsigned EltNum = VecTy->getVectorNumElements();
1037 Value *Cast = Builder.CreateBitCast(CI->getArgOperand(0),
1038 EltTy->getPointerTo());
David Blaikie0c28fd72015-05-20 21:46:30 +00001039 Value *Load = Builder.CreateLoad(EltTy, Cast);
Adam Nemet39066802014-05-29 23:35:33 +00001040 Type *I32Ty = Type::getInt32Ty(C);
1041 Rep = UndefValue::get(VecTy);
1042 for (unsigned I = 0; I < EltNum; ++I)
1043 Rep = Builder.CreateInsertElement(Rep, Load,
1044 ConstantInt::get(I32Ty, I));
Craig Topper5aebb862016-07-04 20:56:38 +00001045 } else if (IsX86 && (Name.startswith("sse41.pmovsx") ||
1046 Name.startswith("sse41.pmovzx") ||
1047 Name.startswith("avx2.pmovsx") ||
Craig Topperb110e042016-11-07 02:12:57 +00001048 Name.startswith("avx2.pmovzx") ||
1049 Name.startswith("avx512.mask.pmovsx") ||
1050 Name.startswith("avx512.mask.pmovzx"))) {
Simon Pilgrim9cb018b2015-09-23 08:48:33 +00001051 VectorType *SrcTy = cast<VectorType>(CI->getArgOperand(0)->getType());
1052 VectorType *DstTy = cast<VectorType>(CI->getType());
1053 unsigned NumDstElts = DstTy->getNumElements();
1054
Simon Pilgrim9602d672016-05-28 18:03:41 +00001055 // Extract a subvector of the first NumDstElts lanes and sign/zero extend.
Craig Topperc0a5fa02016-06-12 04:48:00 +00001056 SmallVector<uint32_t, 8> ShuffleMask(NumDstElts);
Craig Topper99d1eab2016-06-12 00:41:19 +00001057 for (unsigned i = 0; i != NumDstElts; ++i)
Craig Topperc0a5fa02016-06-12 04:48:00 +00001058 ShuffleMask[i] = i;
Simon Pilgrim9cb018b2015-09-23 08:48:33 +00001059
1060 Value *SV = Builder.CreateShuffleVector(
1061 CI->getArgOperand(0), UndefValue::get(SrcTy), ShuffleMask);
Simon Pilgrim9602d672016-05-28 18:03:41 +00001062
1063 bool DoSext = (StringRef::npos != Name.find("pmovsx"));
1064 Rep = DoSext ? Builder.CreateSExt(SV, DstTy)
1065 : Builder.CreateZExt(SV, DstTy);
Craig Topperb110e042016-11-07 02:12:57 +00001066 // If there are 3 arguments, it's a masked intrinsic so we need a select.
1067 if (CI->getNumArgOperands() == 3)
1068 Rep = EmitX86Select(Builder, CI->getArgOperand(2), Rep,
1069 CI->getArgOperand(1));
Simon Pilgrimea0d4f92016-07-22 13:58:44 +00001070 } else if (IsX86 && (Name.startswith("avx.vbroadcastf128") ||
1071 Name == "avx2.vbroadcasti128")) {
1072 // Replace vbroadcastf128/vbroadcasti128 with a vector load+shuffle.
1073 Type *EltTy = CI->getType()->getVectorElementType();
1074 unsigned NumSrcElts = 128 / EltTy->getPrimitiveSizeInBits();
1075 Type *VT = VectorType::get(EltTy, NumSrcElts);
David Blaikie0c28fd72015-05-20 21:46:30 +00001076 Value *Op = Builder.CreatePointerCast(CI->getArgOperand(0),
1077 PointerType::getUnqual(VT));
Chandler Carruth0215e762016-08-10 07:41:26 +00001078 Value *Load = Builder.CreateAlignedLoad(Op, 1);
Simon Pilgrimea0d4f92016-07-22 13:58:44 +00001079 if (NumSrcElts == 2)
1080 Rep = Builder.CreateShuffleVector(Load, UndefValue::get(Load->getType()),
1081 { 0, 1, 0, 1 });
1082 else
1083 Rep = Builder.CreateShuffleVector(Load, UndefValue::get(Load->getType()),
1084 { 0, 1, 2, 3, 0, 1, 2, 3 });
Craig Topper5aebb862016-07-04 20:56:38 +00001085 } else if (IsX86 && (Name.startswith("avx2.pbroadcast") ||
Simon Pilgrim4e96fbf2016-07-05 13:58:47 +00001086 Name.startswith("avx2.vbroadcast") ||
1087 Name.startswith("avx512.pbroadcast") ||
1088 Name.startswith("avx512.mask.broadcast.s"))) {
Ahmed Bougacha1a4987052015-08-20 20:36:19 +00001089 // Replace vp?broadcasts with a vector shuffle.
1090 Value *Op = CI->getArgOperand(0);
1091 unsigned NumElts = CI->getType()->getVectorNumElements();
1092 Type *MaskTy = VectorType::get(Type::getInt32Ty(C), NumElts);
1093 Rep = Builder.CreateShuffleVector(Op, UndefValue::get(Op->getType()),
1094 Constant::getNullValue(MaskTy));
Simon Pilgrim4e96fbf2016-07-05 13:58:47 +00001095
1096 if (CI->getNumArgOperands() == 3)
1097 Rep = EmitX86Select(Builder, CI->getArgOperand(2), Rep,
1098 CI->getArgOperand(1));
Craig Topper5aebb862016-07-04 20:56:38 +00001099 } else if (IsX86 && Name.startswith("avx512.mask.palignr.")) {
Craig Topperf57e17d2016-11-23 06:54:55 +00001100 Rep = UpgradeX86ALIGNIntrinsics(Builder, CI->getArgOperand(0),
1101 CI->getArgOperand(1),
1102 CI->getArgOperand(2),
1103 CI->getArgOperand(3),
1104 CI->getArgOperand(4),
1105 false);
1106 } else if (IsX86 && Name.startswith("avx512.mask.valign.")) {
1107 Rep = UpgradeX86ALIGNIntrinsics(Builder, CI->getArgOperand(0),
1108 CI->getArgOperand(1),
1109 CI->getArgOperand(2),
1110 CI->getArgOperand(3),
1111 CI->getArgOperand(4),
1112 true);
Craig Topper5aebb862016-07-04 20:56:38 +00001113 } else if (IsX86 && (Name == "sse2.psll.dq" ||
1114 Name == "avx2.psll.dq")) {
Craig Topper7355ac32016-05-29 06:37:33 +00001115 // 128/256-bit shift left specified in bits.
Craig Topperb324e432015-02-18 06:24:44 +00001116 unsigned Shift = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
Craig Topper46b34fe2016-07-12 01:42:33 +00001117 Rep = UpgradeX86PSLLDQIntrinsics(Builder, CI->getArgOperand(0),
Craig Topperb324e432015-02-18 06:24:44 +00001118 Shift / 8); // Shift is in bits.
Craig Topper5aebb862016-07-04 20:56:38 +00001119 } else if (IsX86 && (Name == "sse2.psrl.dq" ||
1120 Name == "avx2.psrl.dq")) {
Craig Topper7355ac32016-05-29 06:37:33 +00001121 // 128/256-bit shift right specified in bits.
Craig Topperb324e432015-02-18 06:24:44 +00001122 unsigned Shift = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
Craig Topper46b34fe2016-07-12 01:42:33 +00001123 Rep = UpgradeX86PSRLDQIntrinsics(Builder, CI->getArgOperand(0),
Craig Topperb324e432015-02-18 06:24:44 +00001124 Shift / 8); // Shift is in bits.
Craig Topper5aebb862016-07-04 20:56:38 +00001125 } else if (IsX86 && (Name == "sse2.psll.dq.bs" ||
1126 Name == "avx2.psll.dq.bs" ||
1127 Name == "avx512.psll.dq.512")) {
Simon Pilgrimf7186822016-06-09 21:09:03 +00001128 // 128/256/512-bit shift left specified in bytes.
Craig Topperb324e432015-02-18 06:24:44 +00001129 unsigned Shift = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
Craig Topper46b34fe2016-07-12 01:42:33 +00001130 Rep = UpgradeX86PSLLDQIntrinsics(Builder, CI->getArgOperand(0), Shift);
Craig Topper5aebb862016-07-04 20:56:38 +00001131 } else if (IsX86 && (Name == "sse2.psrl.dq.bs" ||
1132 Name == "avx2.psrl.dq.bs" ||
1133 Name == "avx512.psrl.dq.512")) {
Simon Pilgrimf7186822016-06-09 21:09:03 +00001134 // 128/256/512-bit shift right specified in bytes.
Craig Topperb324e432015-02-18 06:24:44 +00001135 unsigned Shift = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
Craig Topper46b34fe2016-07-12 01:42:33 +00001136 Rep = UpgradeX86PSRLDQIntrinsics(Builder, CI->getArgOperand(0), Shift);
Craig Topper5aebb862016-07-04 20:56:38 +00001137 } else if (IsX86 && (Name == "sse41.pblendw" ||
1138 Name.startswith("sse41.blendp") ||
1139 Name.startswith("avx.blend.p") ||
1140 Name == "avx2.pblendw" ||
1141 Name.startswith("avx2.pblendd."))) {
Craig Topper782d6202015-02-28 19:33:17 +00001142 Value *Op0 = CI->getArgOperand(0);
1143 Value *Op1 = CI->getArgOperand(1);
1144 unsigned Imm = cast <ConstantInt>(CI->getArgOperand(2))->getZExtValue();
1145 VectorType *VecTy = cast<VectorType>(CI->getType());
1146 unsigned NumElts = VecTy->getNumElements();
1147
Craig Topperc0a5fa02016-06-12 04:48:00 +00001148 SmallVector<uint32_t, 16> Idxs(NumElts);
1149 for (unsigned i = 0; i != NumElts; ++i)
1150 Idxs[i] = ((Imm >> (i%8)) & 1) ? i + NumElts : i;
Craig Topper782d6202015-02-28 19:33:17 +00001151
Craig Topper2f561822016-06-12 01:05:59 +00001152 Rep = Builder.CreateShuffleVector(Op0, Op1, Idxs);
Craig Topper5aebb862016-07-04 20:56:38 +00001153 } else if (IsX86 && (Name.startswith("avx.vinsertf128.") ||
1154 Name == "avx2.vinserti128")) {
Sanjay Patel19792fb2015-03-10 16:08:36 +00001155 Value *Op0 = CI->getArgOperand(0);
1156 Value *Op1 = CI->getArgOperand(1);
1157 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(2))->getZExtValue();
1158 VectorType *VecTy = cast<VectorType>(CI->getType());
1159 unsigned NumElts = VecTy->getNumElements();
Simon Pilgrim9cb018b2015-09-23 08:48:33 +00001160
Sanjay Patel19792fb2015-03-10 16:08:36 +00001161 // Mask off the high bits of the immediate value; hardware ignores those.
1162 Imm = Imm & 1;
Simon Pilgrim9cb018b2015-09-23 08:48:33 +00001163
Sanjay Patel19792fb2015-03-10 16:08:36 +00001164 // Extend the second operand into a vector that is twice as big.
1165 Value *UndefV = UndefValue::get(Op1->getType());
Craig Topperc0a5fa02016-06-12 04:48:00 +00001166 SmallVector<uint32_t, 8> Idxs(NumElts);
1167 for (unsigned i = 0; i != NumElts; ++i)
1168 Idxs[i] = i;
Craig Topper2f561822016-06-12 01:05:59 +00001169 Rep = Builder.CreateShuffleVector(Op1, UndefV, Idxs);
Sanjay Patel19792fb2015-03-10 16:08:36 +00001170
1171 // Insert the second operand into the first operand.
1172
1173 // Note that there is no guarantee that instruction lowering will actually
1174 // produce a vinsertf128 instruction for the created shuffles. In
1175 // particular, the 0 immediate case involves no lane changes, so it can
1176 // be handled as a blend.
1177
1178 // Example of shuffle mask for 32-bit elements:
1179 // Imm = 1 <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
1180 // Imm = 0 <i32 8, i32 9, i32 10, i32 11, i32 4, i32 5, i32 6, i32 7 >
1181
Sanjay Patel19792fb2015-03-10 16:08:36 +00001182 // The low half of the result is either the low half of the 1st operand
1183 // or the low half of the 2nd operand (the inserted vector).
Craig Topperc0a5fa02016-06-12 04:48:00 +00001184 for (unsigned i = 0; i != NumElts / 2; ++i)
1185 Idxs[i] = Imm ? i : (i + NumElts);
Sanjay Patel19792fb2015-03-10 16:08:36 +00001186 // The high half of the result is either the low half of the 2nd operand
1187 // (the inserted vector) or the high half of the 1st operand.
Craig Topperc0a5fa02016-06-12 04:48:00 +00001188 for (unsigned i = NumElts / 2; i != NumElts; ++i)
1189 Idxs[i] = Imm ? (i + NumElts / 2) : i;
Craig Topper2f561822016-06-12 01:05:59 +00001190 Rep = Builder.CreateShuffleVector(Op0, Rep, Idxs);
Craig Topper5aebb862016-07-04 20:56:38 +00001191 } else if (IsX86 && (Name.startswith("avx.vextractf128.") ||
1192 Name == "avx2.vextracti128")) {
Sanjay Patelaf1846c2015-03-12 15:15:19 +00001193 Value *Op0 = CI->getArgOperand(0);
1194 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
1195 VectorType *VecTy = cast<VectorType>(CI->getType());
1196 unsigned NumElts = VecTy->getNumElements();
Simon Pilgrim9cb018b2015-09-23 08:48:33 +00001197
Sanjay Patelaf1846c2015-03-12 15:15:19 +00001198 // Mask off the high bits of the immediate value; hardware ignores those.
1199 Imm = Imm & 1;
1200
1201 // Get indexes for either the high half or low half of the input vector.
Craig Topper2f561822016-06-12 01:05:59 +00001202 SmallVector<uint32_t, 4> Idxs(NumElts);
Sanjay Patelaf1846c2015-03-12 15:15:19 +00001203 for (unsigned i = 0; i != NumElts; ++i) {
Craig Topper2f561822016-06-12 01:05:59 +00001204 Idxs[i] = Imm ? (i + NumElts) : i;
Sanjay Patelaf1846c2015-03-12 15:15:19 +00001205 }
1206
1207 Value *UndefV = UndefValue::get(Op0->getType());
Craig Topper2f561822016-06-12 01:05:59 +00001208 Rep = Builder.CreateShuffleVector(Op0, UndefV, Idxs);
Craig Topper5aebb862016-07-04 20:56:38 +00001209 } else if (!IsX86 && Name == "stackprotectorcheck") {
Tim Shen00127562016-04-08 21:26:31 +00001210 Rep = nullptr;
Craig Topper5aebb862016-07-04 20:56:38 +00001211 } else if (IsX86 && (Name.startswith("avx512.mask.perm.df.") ||
1212 Name.startswith("avx512.mask.perm.di."))) {
Simon Pilgrim02d435d2016-07-04 14:19:05 +00001213 Value *Op0 = CI->getArgOperand(0);
1214 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
1215 VectorType *VecTy = cast<VectorType>(CI->getType());
1216 unsigned NumElts = VecTy->getNumElements();
1217
1218 SmallVector<uint32_t, 8> Idxs(NumElts);
1219 for (unsigned i = 0; i != NumElts; ++i)
1220 Idxs[i] = (i & ~0x3) + ((Imm >> (2 * (i & 0x3))) & 3);
1221
1222 Rep = Builder.CreateShuffleVector(Op0, Op0, Idxs);
1223
1224 if (CI->getNumArgOperands() == 4)
1225 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1226 CI->getArgOperand(2));
Craig Topper5aebb862016-07-04 20:56:38 +00001227 } else if (IsX86 && (Name.startswith("avx.vpermil.") ||
1228 Name == "sse2.pshuf.d" ||
1229 Name.startswith("avx512.mask.vpermil.p") ||
1230 Name.startswith("avx512.mask.pshuf.d."))) {
Craig Topper8a105052016-06-12 03:10:47 +00001231 Value *Op0 = CI->getArgOperand(0);
1232 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
1233 VectorType *VecTy = cast<VectorType>(CI->getType());
1234 unsigned NumElts = VecTy->getNumElements();
Simon Pilgrim9fca3002016-07-04 12:40:54 +00001235 // Calculate the size of each index in the immediate.
Craig Topper8a105052016-06-12 03:10:47 +00001236 unsigned IdxSize = 64 / VecTy->getScalarSizeInBits();
1237 unsigned IdxMask = ((1 << IdxSize) - 1);
1238
1239 SmallVector<uint32_t, 8> Idxs(NumElts);
1240 // Lookup the bits for this element, wrapping around the immediate every
1241 // 8-bits. Elements are grouped into sets of 2 or 4 elements so we need
1242 // to offset by the first index of each group.
1243 for (unsigned i = 0; i != NumElts; ++i)
1244 Idxs[i] = ((Imm >> ((i * IdxSize) % 8)) & IdxMask) | (i & ~IdxMask);
1245
1246 Rep = Builder.CreateShuffleVector(Op0, Op0, Idxs);
Craig Topper13cf7ca2016-06-13 02:36:48 +00001247
1248 if (CI->getNumArgOperands() == 4)
1249 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1250 CI->getArgOperand(2));
Craig Topper5aebb862016-07-04 20:56:38 +00001251 } else if (IsX86 && (Name == "sse2.pshufl.w" ||
1252 Name.startswith("avx512.mask.pshufl.w."))) {
Craig Topper10679862016-06-12 14:11:32 +00001253 Value *Op0 = CI->getArgOperand(0);
1254 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
1255 unsigned NumElts = CI->getType()->getVectorNumElements();
1256
1257 SmallVector<uint32_t, 16> Idxs(NumElts);
1258 for (unsigned l = 0; l != NumElts; l += 8) {
1259 for (unsigned i = 0; i != 4; ++i)
1260 Idxs[i + l] = ((Imm >> (2 * i)) & 0x3) + l;
1261 for (unsigned i = 4; i != 8; ++i)
1262 Idxs[i + l] = i + l;
1263 }
1264
1265 Rep = Builder.CreateShuffleVector(Op0, Op0, Idxs);
Craig Topper13cf7ca2016-06-13 02:36:48 +00001266
1267 if (CI->getNumArgOperands() == 4)
1268 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1269 CI->getArgOperand(2));
Craig Topper5aebb862016-07-04 20:56:38 +00001270 } else if (IsX86 && (Name == "sse2.pshufh.w" ||
1271 Name.startswith("avx512.mask.pshufh.w."))) {
Craig Topper10679862016-06-12 14:11:32 +00001272 Value *Op0 = CI->getArgOperand(0);
1273 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
1274 unsigned NumElts = CI->getType()->getVectorNumElements();
1275
1276 SmallVector<uint32_t, 16> Idxs(NumElts);
1277 for (unsigned l = 0; l != NumElts; l += 8) {
1278 for (unsigned i = 0; i != 4; ++i)
1279 Idxs[i + l] = i + l;
1280 for (unsigned i = 0; i != 4; ++i)
1281 Idxs[i + l + 4] = ((Imm >> (2 * i)) & 0x3) + 4 + l;
1282 }
1283
1284 Rep = Builder.CreateShuffleVector(Op0, Op0, Idxs);
Craig Topper13cf7ca2016-06-13 02:36:48 +00001285
1286 if (CI->getNumArgOperands() == 4)
1287 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1288 CI->getArgOperand(2));
Craig Topper4619c9e2016-09-13 07:40:53 +00001289 } else if (IsX86 && Name.startswith("avx512.mask.shuf.p")) {
1290 Value *Op0 = CI->getArgOperand(0);
1291 Value *Op1 = CI->getArgOperand(1);
1292 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(2))->getZExtValue();
1293 unsigned NumElts = CI->getType()->getVectorNumElements();
1294
1295 unsigned NumLaneElts = 128/CI->getType()->getScalarSizeInBits();
1296 unsigned HalfLaneElts = NumLaneElts / 2;
1297
1298 SmallVector<uint32_t, 16> Idxs(NumElts);
1299 for (unsigned i = 0; i != NumElts; ++i) {
1300 // Base index is the starting element of the lane.
1301 Idxs[i] = i - (i % NumLaneElts);
1302 // If we are half way through the lane switch to the other source.
1303 if ((i % NumLaneElts) >= HalfLaneElts)
1304 Idxs[i] += NumElts;
1305 // Now select the specific element. By adding HalfLaneElts bits from
1306 // the immediate. Wrapping around the immediate every 8-bits.
1307 Idxs[i] += (Imm >> ((i * HalfLaneElts) % 8)) & ((1 << HalfLaneElts) - 1);
1308 }
1309
1310 Rep = Builder.CreateShuffleVector(Op0, Op1, Idxs);
1311
1312 Rep = EmitX86Select(Builder, CI->getArgOperand(4), Rep,
1313 CI->getArgOperand(3));
Craig Topper5aebb862016-07-04 20:56:38 +00001314 } else if (IsX86 && (Name.startswith("avx512.mask.movddup") ||
1315 Name.startswith("avx512.mask.movshdup") ||
1316 Name.startswith("avx512.mask.movsldup"))) {
Simon Pilgrim19adee92016-07-02 14:42:35 +00001317 Value *Op0 = CI->getArgOperand(0);
1318 unsigned NumElts = CI->getType()->getVectorNumElements();
1319 unsigned NumLaneElts = 128/CI->getType()->getScalarSizeInBits();
1320
1321 unsigned Offset = 0;
Craig Topper5aebb862016-07-04 20:56:38 +00001322 if (Name.startswith("avx512.mask.movshdup."))
Simon Pilgrim19adee92016-07-02 14:42:35 +00001323 Offset = 1;
1324
1325 SmallVector<uint32_t, 16> Idxs(NumElts);
1326 for (unsigned l = 0; l != NumElts; l += NumLaneElts)
1327 for (unsigned i = 0; i != NumLaneElts; i += 2) {
1328 Idxs[i + l + 0] = i + l + Offset;
1329 Idxs[i + l + 1] = i + l + Offset;
1330 }
1331
1332 Rep = Builder.CreateShuffleVector(Op0, Op0, Idxs);
1333
1334 Rep = EmitX86Select(Builder, CI->getArgOperand(2), Rep,
1335 CI->getArgOperand(1));
Craig Topper5aebb862016-07-04 20:56:38 +00001336 } else if (IsX86 && (Name.startswith("avx512.mask.punpckl") ||
1337 Name.startswith("avx512.mask.unpckl."))) {
Craig Topper597aa422016-06-23 07:37:33 +00001338 Value *Op0 = CI->getArgOperand(0);
1339 Value *Op1 = CI->getArgOperand(1);
1340 int NumElts = CI->getType()->getVectorNumElements();
1341 int NumLaneElts = 128/CI->getType()->getScalarSizeInBits();
1342
1343 SmallVector<uint32_t, 64> Idxs(NumElts);
1344 for (int l = 0; l != NumElts; l += NumLaneElts)
1345 for (int i = 0; i != NumLaneElts; ++i)
1346 Idxs[i + l] = l + (i / 2) + NumElts * (i % 2);
1347
1348 Rep = Builder.CreateShuffleVector(Op0, Op1, Idxs);
1349
1350 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1351 CI->getArgOperand(2));
Craig Topper5aebb862016-07-04 20:56:38 +00001352 } else if (IsX86 && (Name.startswith("avx512.mask.punpckh") ||
1353 Name.startswith("avx512.mask.unpckh."))) {
Craig Topper597aa422016-06-23 07:37:33 +00001354 Value *Op0 = CI->getArgOperand(0);
1355 Value *Op1 = CI->getArgOperand(1);
1356 int NumElts = CI->getType()->getVectorNumElements();
1357 int NumLaneElts = 128/CI->getType()->getScalarSizeInBits();
1358
1359 SmallVector<uint32_t, 64> Idxs(NumElts);
1360 for (int l = 0; l != NumElts; l += NumLaneElts)
1361 for (int i = 0; i != NumLaneElts; ++i)
1362 Idxs[i + l] = (NumLaneElts / 2) + l + (i / 2) + NumElts * (i % 2);
1363
1364 Rep = Builder.CreateShuffleVector(Op0, Op1, Idxs);
1365
1366 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1367 CI->getArgOperand(2));
Craig Toppera6e6feb2016-07-12 05:27:53 +00001368 } else if (IsX86 && Name.startswith("avx512.mask.pand.")) {
1369 Rep = Builder.CreateAnd(CI->getArgOperand(0), CI->getArgOperand(1));
1370 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1371 CI->getArgOperand(2));
1372 } else if (IsX86 && Name.startswith("avx512.mask.pandn.")) {
1373 Rep = Builder.CreateAnd(Builder.CreateNot(CI->getArgOperand(0)),
1374 CI->getArgOperand(1));
1375 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1376 CI->getArgOperand(2));
1377 } else if (IsX86 && Name.startswith("avx512.mask.por.")) {
1378 Rep = Builder.CreateOr(CI->getArgOperand(0), CI->getArgOperand(1));
1379 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1380 CI->getArgOperand(2));
1381 } else if (IsX86 && Name.startswith("avx512.mask.pxor.")) {
1382 Rep = Builder.CreateXor(CI->getArgOperand(0), CI->getArgOperand(1));
1383 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1384 CI->getArgOperand(2));
Craig Toppere75c4952016-09-02 05:29:17 +00001385 } else if (IsX86 && Name.startswith("avx512.mask.and.")) {
1386 VectorType *FTy = cast<VectorType>(CI->getType());
1387 VectorType *ITy = VectorType::getInteger(FTy);
1388 Rep = Builder.CreateAnd(Builder.CreateBitCast(CI->getArgOperand(0), ITy),
1389 Builder.CreateBitCast(CI->getArgOperand(1), ITy));
1390 Rep = Builder.CreateBitCast(Rep, FTy);
1391 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1392 CI->getArgOperand(2));
1393 } else if (IsX86 && Name.startswith("avx512.mask.andn.")) {
1394 VectorType *FTy = cast<VectorType>(CI->getType());
1395 VectorType *ITy = VectorType::getInteger(FTy);
1396 Rep = Builder.CreateNot(Builder.CreateBitCast(CI->getArgOperand(0), ITy));
1397 Rep = Builder.CreateAnd(Rep,
1398 Builder.CreateBitCast(CI->getArgOperand(1), ITy));
1399 Rep = Builder.CreateBitCast(Rep, FTy);
1400 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1401 CI->getArgOperand(2));
1402 } else if (IsX86 && Name.startswith("avx512.mask.or.")) {
1403 VectorType *FTy = cast<VectorType>(CI->getType());
1404 VectorType *ITy = VectorType::getInteger(FTy);
1405 Rep = Builder.CreateOr(Builder.CreateBitCast(CI->getArgOperand(0), ITy),
1406 Builder.CreateBitCast(CI->getArgOperand(1), ITy));
1407 Rep = Builder.CreateBitCast(Rep, FTy);
1408 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1409 CI->getArgOperand(2));
1410 } else if (IsX86 && Name.startswith("avx512.mask.xor.")) {
1411 VectorType *FTy = cast<VectorType>(CI->getType());
1412 VectorType *ITy = VectorType::getInteger(FTy);
1413 Rep = Builder.CreateXor(Builder.CreateBitCast(CI->getArgOperand(0), ITy),
1414 Builder.CreateBitCast(CI->getArgOperand(1), ITy));
1415 Rep = Builder.CreateBitCast(Rep, FTy);
1416 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1417 CI->getArgOperand(2));
Craig Topperaf0d63d2016-09-04 02:09:53 +00001418 } else if (IsX86 && Name.startswith("avx512.mask.padd.")) {
1419 Rep = Builder.CreateAdd(CI->getArgOperand(0), CI->getArgOperand(1));
1420 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1421 CI->getArgOperand(2));
1422 } else if (IsX86 && Name.startswith("avx512.mask.psub.")) {
1423 Rep = Builder.CreateSub(CI->getArgOperand(0), CI->getArgOperand(1));
1424 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1425 CI->getArgOperand(2));
1426 } else if (IsX86 && Name.startswith("avx512.mask.pmull.")) {
1427 Rep = Builder.CreateMul(CI->getArgOperand(0), CI->getArgOperand(1));
1428 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1429 CI->getArgOperand(2));
Craig Topper8f354a72016-11-05 05:35:23 +00001430 } else if (IsX86 && (Name.startswith("avx512.mask.add.p"))) {
Craig Topper41773452016-09-04 18:13:33 +00001431 Rep = Builder.CreateFAdd(CI->getArgOperand(0), CI->getArgOperand(1));
1432 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1433 CI->getArgOperand(2));
Craig Topper96041c62016-11-07 00:13:42 +00001434 } else if (IsX86 && Name.startswith("avx512.mask.div.p")) {
Craig Topper41773452016-09-04 18:13:33 +00001435 Rep = Builder.CreateFDiv(CI->getArgOperand(0), CI->getArgOperand(1));
1436 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1437 CI->getArgOperand(2));
Craig Topper96041c62016-11-07 00:13:42 +00001438 } else if (IsX86 && Name.startswith("avx512.mask.mul.p")) {
Craig Topper41773452016-09-04 18:13:33 +00001439 Rep = Builder.CreateFMul(CI->getArgOperand(0), CI->getArgOperand(1));
1440 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1441 CI->getArgOperand(2));
Craig Topper96041c62016-11-07 00:13:42 +00001442 } else if (IsX86 && Name.startswith("avx512.mask.sub.p")) {
Craig Topper41773452016-09-04 18:13:33 +00001443 Rep = Builder.CreateFSub(CI->getArgOperand(0), CI->getArgOperand(1));
1444 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1445 CI->getArgOperand(2));
Craig Topper7e545332016-11-07 00:13:39 +00001446 } else if (IsX86 && Name.startswith("avx512.mask.pshuf.b.")) {
1447 VectorType *VecTy = cast<VectorType>(CI->getType());
1448 Intrinsic::ID IID;
1449 if (VecTy->getPrimitiveSizeInBits() == 128)
1450 IID = Intrinsic::x86_ssse3_pshuf_b_128;
1451 else if (VecTy->getPrimitiveSizeInBits() == 256)
1452 IID = Intrinsic::x86_avx2_pshuf_b;
Craig Topperedab02b2016-12-10 23:09:43 +00001453 else if (VecTy->getPrimitiveSizeInBits() == 512)
1454 IID = Intrinsic::x86_avx512_pshuf_b_512;
Craig Topper7e545332016-11-07 00:13:39 +00001455 else
1456 llvm_unreachable("Unexpected intrinsic");
1457
1458 Rep = Builder.CreateCall(Intrinsic::getDeclaration(F->getParent(), IID),
1459 { CI->getArgOperand(0), CI->getArgOperand(1) });
1460 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1461 CI->getArgOperand(2));
Craig Topper987dad22016-11-13 19:09:56 +00001462 } else if (IsX86 && Name.startswith("avx512.mask.psll")) {
1463 bool IsImmediate = Name[16] == 'i' ||
1464 (Name.size() > 18 && Name[18] == 'i');
1465 bool IsVariable = Name[16] == 'v';
1466 char Size = Name[16] == '.' ? Name[17] :
1467 Name[17] == '.' ? Name[18] :
Craig Topper02b5a1b2016-11-18 05:04:44 +00001468 Name[18] == '.' ? Name[19] :
1469 Name[20];
Craig Topper987dad22016-11-13 19:09:56 +00001470
1471 Intrinsic::ID IID;
Craig Topper353e59b2016-11-14 01:53:22 +00001472 if (IsVariable && Name[17] != '.') {
Craig Topper987dad22016-11-13 19:09:56 +00001473 if (Size == 'd' && Name[17] == '2') // avx512.mask.psllv2.di
1474 IID = Intrinsic::x86_avx2_psllv_q;
1475 else if (Size == 'd' && Name[17] == '4') // avx512.mask.psllv4.di
1476 IID = Intrinsic::x86_avx2_psllv_q_256;
1477 else if (Size == 's' && Name[17] == '4') // avx512.mask.psllv4.si
1478 IID = Intrinsic::x86_avx2_psllv_d;
1479 else if (Size == 's' && Name[17] == '8') // avx512.mask.psllv8.si
1480 IID = Intrinsic::x86_avx2_psllv_d_256;
Craig Topper02b5a1b2016-11-18 05:04:44 +00001481 else if (Size == 'h' && Name[17] == '8') // avx512.mask.psllv8.hi
1482 IID = Intrinsic::x86_avx512_psllv_w_128;
1483 else if (Size == 'h' && Name[17] == '1') // avx512.mask.psllv16.hi
1484 IID = Intrinsic::x86_avx512_psllv_w_256;
1485 else if (Name[17] == '3' && Name[18] == '2') // avx512.mask.psllv32hi
1486 IID = Intrinsic::x86_avx512_psllv_w_512;
Craig Topper987dad22016-11-13 19:09:56 +00001487 else
1488 llvm_unreachable("Unexpected size");
1489 } else if (Name.endswith(".128")) {
1490 if (Size == 'd') // avx512.mask.psll.d.128, avx512.mask.psll.di.128
1491 IID = IsImmediate ? Intrinsic::x86_sse2_pslli_d
1492 : Intrinsic::x86_sse2_psll_d;
1493 else if (Size == 'q') // avx512.mask.psll.q.128, avx512.mask.psll.qi.128
1494 IID = IsImmediate ? Intrinsic::x86_sse2_pslli_q
1495 : Intrinsic::x86_sse2_psll_q;
1496 else if (Size == 'w') // avx512.mask.psll.w.128, avx512.mask.psll.wi.128
1497 IID = IsImmediate ? Intrinsic::x86_sse2_pslli_w
1498 : Intrinsic::x86_sse2_psll_w;
1499 else
1500 llvm_unreachable("Unexpected size");
1501 } else if (Name.endswith(".256")) {
1502 if (Size == 'd') // avx512.mask.psll.d.256, avx512.mask.psll.di.256
1503 IID = IsImmediate ? Intrinsic::x86_avx2_pslli_d
1504 : Intrinsic::x86_avx2_psll_d;
1505 else if (Size == 'q') // avx512.mask.psll.q.256, avx512.mask.psll.qi.256
1506 IID = IsImmediate ? Intrinsic::x86_avx2_pslli_q
1507 : Intrinsic::x86_avx2_psll_q;
1508 else if (Size == 'w') // avx512.mask.psll.w.256, avx512.mask.psll.wi.256
1509 IID = IsImmediate ? Intrinsic::x86_avx2_pslli_w
1510 : Intrinsic::x86_avx2_psll_w;
1511 else
1512 llvm_unreachable("Unexpected size");
1513 } else {
Craig Topper353e59b2016-11-14 01:53:22 +00001514 if (Size == 'd') // psll.di.512, pslli.d, psll.d, psllv.d.512
1515 IID = IsImmediate ? Intrinsic::x86_avx512_pslli_d_512 :
1516 IsVariable ? Intrinsic::x86_avx512_psllv_d_512 :
1517 Intrinsic::x86_avx512_psll_d_512;
1518 else if (Size == 'q') // psll.qi.512, pslli.q, psll.q, psllv.q.512
1519 IID = IsImmediate ? Intrinsic::x86_avx512_pslli_q_512 :
1520 IsVariable ? Intrinsic::x86_avx512_psllv_q_512 :
1521 Intrinsic::x86_avx512_psll_q_512;
Craig Topper987dad22016-11-13 19:09:56 +00001522 else if (Size == 'w') // psll.wi.512, pslli.w, psll.w
1523 IID = IsImmediate ? Intrinsic::x86_avx512_pslli_w_512
1524 : Intrinsic::x86_avx512_psll_w_512;
1525 else
1526 llvm_unreachable("Unexpected size");
1527 }
1528
1529 Rep = UpgradeX86MaskedShift(Builder, *CI, IID);
1530 } else if (IsX86 && Name.startswith("avx512.mask.psrl")) {
1531 bool IsImmediate = Name[16] == 'i' ||
1532 (Name.size() > 18 && Name[18] == 'i');
1533 bool IsVariable = Name[16] == 'v';
1534 char Size = Name[16] == '.' ? Name[17] :
1535 Name[17] == '.' ? Name[18] :
Craig Topper02b5a1b2016-11-18 05:04:44 +00001536 Name[18] == '.' ? Name[19] :
1537 Name[20];
Craig Topper987dad22016-11-13 19:09:56 +00001538
1539 Intrinsic::ID IID;
Craig Topper353e59b2016-11-14 01:53:22 +00001540 if (IsVariable && Name[17] != '.') {
Craig Topper987dad22016-11-13 19:09:56 +00001541 if (Size == 'd' && Name[17] == '2') // avx512.mask.psrlv2.di
1542 IID = Intrinsic::x86_avx2_psrlv_q;
1543 else if (Size == 'd' && Name[17] == '4') // avx512.mask.psrlv4.di
1544 IID = Intrinsic::x86_avx2_psrlv_q_256;
1545 else if (Size == 's' && Name[17] == '4') // avx512.mask.psrlv4.si
1546 IID = Intrinsic::x86_avx2_psrlv_d;
1547 else if (Size == 's' && Name[17] == '8') // avx512.mask.psrlv8.si
1548 IID = Intrinsic::x86_avx2_psrlv_d_256;
Craig Topper02b5a1b2016-11-18 05:04:44 +00001549 else if (Size == 'h' && Name[17] == '8') // avx512.mask.psrlv8.hi
1550 IID = Intrinsic::x86_avx512_psrlv_w_128;
1551 else if (Size == 'h' && Name[17] == '1') // avx512.mask.psrlv16.hi
1552 IID = Intrinsic::x86_avx512_psrlv_w_256;
1553 else if (Name[17] == '3' && Name[18] == '2') // avx512.mask.psrlv32hi
1554 IID = Intrinsic::x86_avx512_psrlv_w_512;
Craig Topper987dad22016-11-13 19:09:56 +00001555 else
1556 llvm_unreachable("Unexpected size");
1557 } else if (Name.endswith(".128")) {
1558 if (Size == 'd') // avx512.mask.psrl.d.128, avx512.mask.psrl.di.128
1559 IID = IsImmediate ? Intrinsic::x86_sse2_psrli_d
1560 : Intrinsic::x86_sse2_psrl_d;
1561 else if (Size == 'q') // avx512.mask.psrl.q.128, avx512.mask.psrl.qi.128
1562 IID = IsImmediate ? Intrinsic::x86_sse2_psrli_q
1563 : Intrinsic::x86_sse2_psrl_q;
1564 else if (Size == 'w') // avx512.mask.psrl.w.128, avx512.mask.psrl.wi.128
1565 IID = IsImmediate ? Intrinsic::x86_sse2_psrli_w
1566 : Intrinsic::x86_sse2_psrl_w;
1567 else
1568 llvm_unreachable("Unexpected size");
1569 } else if (Name.endswith(".256")) {
1570 if (Size == 'd') // avx512.mask.psrl.d.256, avx512.mask.psrl.di.256
1571 IID = IsImmediate ? Intrinsic::x86_avx2_psrli_d
1572 : Intrinsic::x86_avx2_psrl_d;
1573 else if (Size == 'q') // avx512.mask.psrl.q.256, avx512.mask.psrl.qi.256
1574 IID = IsImmediate ? Intrinsic::x86_avx2_psrli_q
1575 : Intrinsic::x86_avx2_psrl_q;
1576 else if (Size == 'w') // avx512.mask.psrl.w.256, avx512.mask.psrl.wi.256
1577 IID = IsImmediate ? Intrinsic::x86_avx2_psrli_w
1578 : Intrinsic::x86_avx2_psrl_w;
1579 else
1580 llvm_unreachable("Unexpected size");
1581 } else {
Craig Topper353e59b2016-11-14 01:53:22 +00001582 if (Size == 'd') // psrl.di.512, psrli.d, psrl.d, psrl.d.512
1583 IID = IsImmediate ? Intrinsic::x86_avx512_psrli_d_512 :
1584 IsVariable ? Intrinsic::x86_avx512_psrlv_d_512 :
1585 Intrinsic::x86_avx512_psrl_d_512;
1586 else if (Size == 'q') // psrl.qi.512, psrli.q, psrl.q, psrl.q.512
1587 IID = IsImmediate ? Intrinsic::x86_avx512_psrli_q_512 :
1588 IsVariable ? Intrinsic::x86_avx512_psrlv_q_512 :
1589 Intrinsic::x86_avx512_psrl_q_512;
Craig Topper987dad22016-11-13 19:09:56 +00001590 else if (Size == 'w') // psrl.wi.512, psrli.w, psrl.w)
1591 IID = IsImmediate ? Intrinsic::x86_avx512_psrli_w_512
1592 : Intrinsic::x86_avx512_psrl_w_512;
1593 else
1594 llvm_unreachable("Unexpected size");
1595 }
1596
1597 Rep = UpgradeX86MaskedShift(Builder, *CI, IID);
1598 } else if (IsX86 && Name.startswith("avx512.mask.psra")) {
1599 bool IsImmediate = Name[16] == 'i' ||
1600 (Name.size() > 18 && Name[18] == 'i');
1601 bool IsVariable = Name[16] == 'v';
1602 char Size = Name[16] == '.' ? Name[17] :
1603 Name[17] == '.' ? Name[18] :
Craig Topper02b5a1b2016-11-18 05:04:44 +00001604 Name[18] == '.' ? Name[19] :
1605 Name[20];
Craig Topper987dad22016-11-13 19:09:56 +00001606
1607 Intrinsic::ID IID;
Craig Topper353e59b2016-11-14 01:53:22 +00001608 if (IsVariable && Name[17] != '.') {
Craig Topper987dad22016-11-13 19:09:56 +00001609 if (Size == 's' && Name[17] == '4') // avx512.mask.psrav4.si
1610 IID = Intrinsic::x86_avx2_psrav_d;
1611 else if (Size == 's' && Name[17] == '8') // avx512.mask.psrav8.si
1612 IID = Intrinsic::x86_avx2_psrav_d_256;
Craig Topper02b5a1b2016-11-18 05:04:44 +00001613 else if (Size == 'h' && Name[17] == '8') // avx512.mask.psrav8.hi
1614 IID = Intrinsic::x86_avx512_psrav_w_128;
1615 else if (Size == 'h' && Name[17] == '1') // avx512.mask.psrav16.hi
1616 IID = Intrinsic::x86_avx512_psrav_w_256;
1617 else if (Name[17] == '3' && Name[18] == '2') // avx512.mask.psrav32hi
1618 IID = Intrinsic::x86_avx512_psrav_w_512;
Craig Topper987dad22016-11-13 19:09:56 +00001619 else
1620 llvm_unreachable("Unexpected size");
1621 } else if (Name.endswith(".128")) {
1622 if (Size == 'd') // avx512.mask.psra.d.128, avx512.mask.psra.di.128
1623 IID = IsImmediate ? Intrinsic::x86_sse2_psrai_d
1624 : Intrinsic::x86_sse2_psra_d;
1625 else if (Size == 'q') // avx512.mask.psra.q.128, avx512.mask.psra.qi.128
Craig Topper353e59b2016-11-14 01:53:22 +00001626 IID = IsImmediate ? Intrinsic::x86_avx512_psrai_q_128 :
1627 IsVariable ? Intrinsic::x86_avx512_psrav_q_128 :
1628 Intrinsic::x86_avx512_psra_q_128;
Craig Topper987dad22016-11-13 19:09:56 +00001629 else if (Size == 'w') // avx512.mask.psra.w.128, avx512.mask.psra.wi.128
1630 IID = IsImmediate ? Intrinsic::x86_sse2_psrai_w
1631 : Intrinsic::x86_sse2_psra_w;
1632 else
1633 llvm_unreachable("Unexpected size");
1634 } else if (Name.endswith(".256")) {
1635 if (Size == 'd') // avx512.mask.psra.d.256, avx512.mask.psra.di.256
1636 IID = IsImmediate ? Intrinsic::x86_avx2_psrai_d
1637 : Intrinsic::x86_avx2_psra_d;
1638 else if (Size == 'q') // avx512.mask.psra.q.256, avx512.mask.psra.qi.256
Craig Topper353e59b2016-11-14 01:53:22 +00001639 IID = IsImmediate ? Intrinsic::x86_avx512_psrai_q_256 :
1640 IsVariable ? Intrinsic::x86_avx512_psrav_q_256 :
1641 Intrinsic::x86_avx512_psra_q_256;
Craig Topper987dad22016-11-13 19:09:56 +00001642 else if (Size == 'w') // avx512.mask.psra.w.256, avx512.mask.psra.wi.256
1643 IID = IsImmediate ? Intrinsic::x86_avx2_psrai_w
1644 : Intrinsic::x86_avx2_psra_w;
1645 else
1646 llvm_unreachable("Unexpected size");
1647 } else {
Craig Topper353e59b2016-11-14 01:53:22 +00001648 if (Size == 'd') // psra.di.512, psrai.d, psra.d, psrav.d.512
1649 IID = IsImmediate ? Intrinsic::x86_avx512_psrai_d_512 :
1650 IsVariable ? Intrinsic::x86_avx512_psrav_d_512 :
1651 Intrinsic::x86_avx512_psra_d_512;
Craig Topper987dad22016-11-13 19:09:56 +00001652 else if (Size == 'q') // psra.qi.512, psrai.q, psra.q
Craig Topper353e59b2016-11-14 01:53:22 +00001653 IID = IsImmediate ? Intrinsic::x86_avx512_psrai_q_512 :
1654 IsVariable ? Intrinsic::x86_avx512_psrav_q_512 :
1655 Intrinsic::x86_avx512_psra_q_512;
Craig Topper987dad22016-11-13 19:09:56 +00001656 else if (Size == 'w') // psra.wi.512, psrai.w, psra.w
1657 IID = IsImmediate ? Intrinsic::x86_avx512_psrai_w_512
1658 : Intrinsic::x86_avx512_psra_w_512;
1659 else
1660 llvm_unreachable("Unexpected size");
1661 }
1662
1663 Rep = UpgradeX86MaskedShift(Builder, *CI, IID);
Ayman Musa4d602432016-11-16 09:00:28 +00001664 } else if (IsX86 && Name.startswith("avx512.mask.move.s")) {
1665 Rep = upgradeMaskedMove(Builder, *CI);
Craig Topperabe7c5b2016-12-10 21:15:52 +00001666 } else if (IsX86 && Name.startswith("avx512.mask.vpermilvar.")) {
1667 Intrinsic::ID IID;
1668 if (Name.endswith("ps.128"))
1669 IID = Intrinsic::x86_avx_vpermilvar_ps;
1670 else if (Name.endswith("pd.128"))
1671 IID = Intrinsic::x86_avx_vpermilvar_pd;
1672 else if (Name.endswith("ps.256"))
1673 IID = Intrinsic::x86_avx_vpermilvar_ps_256;
1674 else if (Name.endswith("pd.256"))
1675 IID = Intrinsic::x86_avx_vpermilvar_pd_256;
1676 else
1677 llvm_unreachable("Unexpected vpermilvar intrinsic");
1678
1679 Function *Intrin = Intrinsic::getDeclaration(F->getParent(), IID);
1680 Rep = Builder.CreateCall(Intrin,
1681 { CI->getArgOperand(0), CI->getArgOperand(1) });
1682 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1683 CI->getArgOperand(2));
Craig Topper3b1817d2012-02-03 06:10:55 +00001684 } else {
Craig Topper8a105052016-06-12 03:10:47 +00001685 llvm_unreachable("Unknown function for CallInst upgrade.");
Craig Topper3b1817d2012-02-03 06:10:55 +00001686 }
1687
Tim Shen00127562016-04-08 21:26:31 +00001688 if (Rep)
1689 CI->replaceAllUsesWith(Rep);
Craig Topper3b1817d2012-02-03 06:10:55 +00001690 CI->eraseFromParent();
1691 return;
1692 }
1693
Yaron Kerend1fdbe72015-03-30 16:10:39 +00001694 std::string Name = CI->getName();
Adrian Prantl87b7eb92014-10-01 18:55:02 +00001695 if (!Name.empty())
1696 CI->setName(Name + ".old");
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001697
Chandler Carruth58a71ed2011-12-12 04:26:04 +00001698 switch (NewFn->getIntrinsicID()) {
1699 default:
Chris Lattner0bcbde42011-11-27 08:42:07 +00001700 llvm_unreachable("Unknown function for CallInst upgrade.");
Chandler Carruth58a71ed2011-12-12 04:26:04 +00001701
Jeroen Ketemaab99b592015-09-30 10:56:37 +00001702 case Intrinsic::arm_neon_vld1:
1703 case Intrinsic::arm_neon_vld2:
1704 case Intrinsic::arm_neon_vld3:
1705 case Intrinsic::arm_neon_vld4:
1706 case Intrinsic::arm_neon_vld2lane:
1707 case Intrinsic::arm_neon_vld3lane:
1708 case Intrinsic::arm_neon_vld4lane:
1709 case Intrinsic::arm_neon_vst1:
1710 case Intrinsic::arm_neon_vst2:
1711 case Intrinsic::arm_neon_vst3:
1712 case Intrinsic::arm_neon_vst4:
1713 case Intrinsic::arm_neon_vst2lane:
1714 case Intrinsic::arm_neon_vst3lane:
1715 case Intrinsic::arm_neon_vst4lane: {
1716 SmallVector<Value *, 4> Args(CI->arg_operands().begin(),
1717 CI->arg_operands().end());
1718 CI->replaceAllUsesWith(Builder.CreateCall(NewFn, Args));
1719 CI->eraseFromParent();
1720 return;
1721 }
1722
Chandler Carruth58a71ed2011-12-12 04:26:04 +00001723 case Intrinsic::ctlz:
Nuno Lopesad40c0a2012-05-22 15:25:31 +00001724 case Intrinsic::cttz:
Chandler Carruth58a71ed2011-12-12 04:26:04 +00001725 assert(CI->getNumArgOperands() == 1 &&
1726 "Mismatch between function args and call args");
David Blaikieff6409d2015-05-18 22:13:54 +00001727 CI->replaceAllUsesWith(Builder.CreateCall(
1728 NewFn, {CI->getArgOperand(0), Builder.getFalse()}, Name));
Chandler Carruth58a71ed2011-12-12 04:26:04 +00001729 CI->eraseFromParent();
1730 return;
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001731
Matt Arsenaultfbcbce42013-10-07 18:06:48 +00001732 case Intrinsic::objectsize:
David Blaikieff6409d2015-05-18 22:13:54 +00001733 CI->replaceAllUsesWith(Builder.CreateCall(
1734 NewFn, {CI->getArgOperand(0), CI->getArgOperand(1)}, Name));
Matt Arsenaultfbcbce42013-10-07 18:06:48 +00001735 CI->eraseFromParent();
1736 return;
1737
Joel Jonesb84f7be2012-07-18 00:02:16 +00001738 case Intrinsic::ctpop: {
David Blaikieff6409d2015-05-18 22:13:54 +00001739 CI->replaceAllUsesWith(Builder.CreateCall(NewFn, {CI->getArgOperand(0)}));
Joel Jonesb84f7be2012-07-18 00:02:16 +00001740 CI->eraseFromParent();
1741 return;
1742 }
Joel Jones43cb8782012-07-13 23:25:25 +00001743
Craig Topper71dc02d2012-06-13 07:18:53 +00001744 case Intrinsic::x86_xop_vfrcz_ss:
1745 case Intrinsic::x86_xop_vfrcz_sd:
David Blaikieff6409d2015-05-18 22:13:54 +00001746 CI->replaceAllUsesWith(
1747 Builder.CreateCall(NewFn, {CI->getArgOperand(1)}, Name));
Craig Topper71dc02d2012-06-13 07:18:53 +00001748 CI->eraseFromParent();
1749 return;
1750
Simon Pilgrime85506b2016-06-03 08:06:03 +00001751 case Intrinsic::x86_xop_vpermil2pd:
1752 case Intrinsic::x86_xop_vpermil2ps:
1753 case Intrinsic::x86_xop_vpermil2pd_256:
1754 case Intrinsic::x86_xop_vpermil2ps_256: {
1755 SmallVector<Value *, 4> Args(CI->arg_operands().begin(),
1756 CI->arg_operands().end());
1757 VectorType *FltIdxTy = cast<VectorType>(Args[2]->getType());
1758 VectorType *IntIdxTy = VectorType::getInteger(FltIdxTy);
1759 Args[2] = Builder.CreateBitCast(Args[2], IntIdxTy);
1760 CI->replaceAllUsesWith(Builder.CreateCall(NewFn, Args, Name));
1761 CI->eraseFromParent();
1762 return;
1763 }
1764
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001765 case Intrinsic::x86_sse41_ptestc:
1766 case Intrinsic::x86_sse41_ptestz:
Craig Topper71dc02d2012-06-13 07:18:53 +00001767 case Intrinsic::x86_sse41_ptestnzc: {
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001768 // The arguments for these intrinsics used to be v4f32, and changed
1769 // to v2i64. This is purely a nop, since those are bitwise intrinsics.
1770 // So, the only thing required is a bitcast for both arguments.
1771 // First, check the arguments have the old type.
1772 Value *Arg0 = CI->getArgOperand(0);
1773 if (Arg0->getType() != VectorType::get(Type::getFloatTy(C), 4))
1774 return;
1775
1776 // Old intrinsic, add bitcasts
1777 Value *Arg1 = CI->getArgOperand(1);
1778
David Blaikie5bacf372015-04-24 21:16:07 +00001779 Type *NewVecTy = VectorType::get(Type::getInt64Ty(C), 2);
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001780
David Blaikie5bacf372015-04-24 21:16:07 +00001781 Value *BC0 = Builder.CreateBitCast(Arg0, NewVecTy, "cast");
1782 Value *BC1 = Builder.CreateBitCast(Arg1, NewVecTy, "cast");
1783
David Blaikieff6409d2015-05-18 22:13:54 +00001784 CallInst *NewCall = Builder.CreateCall(NewFn, {BC0, BC1}, Name);
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001785 CI->replaceAllUsesWith(NewCall);
1786 CI->eraseFromParent();
1787 return;
Evan Cheng0e179d02007-12-17 22:33:23 +00001788 }
Chandler Carruth373b2b12014-09-06 10:00:01 +00001789
Chandler Carruth373b2b12014-09-06 10:00:01 +00001790 case Intrinsic::x86_sse41_insertps:
1791 case Intrinsic::x86_sse41_dppd:
1792 case Intrinsic::x86_sse41_dpps:
1793 case Intrinsic::x86_sse41_mpsadbw:
Chandler Carruth373b2b12014-09-06 10:00:01 +00001794 case Intrinsic::x86_avx_dp_ps_256:
Chandler Carruth373b2b12014-09-06 10:00:01 +00001795 case Intrinsic::x86_avx2_mpsadbw: {
1796 // Need to truncate the last argument from i32 to i8 -- this argument models
1797 // an inherently 8-bit immediate operand to these x86 instructions.
1798 SmallVector<Value *, 4> Args(CI->arg_operands().begin(),
1799 CI->arg_operands().end());
1800
1801 // Replace the last argument with a trunc.
1802 Args.back() = Builder.CreateTrunc(Args.back(), Type::getInt8Ty(C), "trunc");
1803
1804 CallInst *NewCall = Builder.CreateCall(NewFn, Args);
1805 CI->replaceAllUsesWith(NewCall);
1806 CI->eraseFromParent();
1807 return;
1808 }
Marcin Koscielnicki3fdc2572016-04-19 20:51:05 +00001809
1810 case Intrinsic::thread_pointer: {
1811 CI->replaceAllUsesWith(Builder.CreateCall(NewFn, {}));
1812 CI->eraseFromParent();
1813 return;
1814 }
Artur Pilipenko7ad95ec2016-06-28 18:27:25 +00001815
Mehdi Amini8c629ec2016-08-13 23:31:24 +00001816 case Intrinsic::invariant_start:
1817 case Intrinsic::invariant_end:
Artur Pilipenko7ad95ec2016-06-28 18:27:25 +00001818 case Intrinsic::masked_load:
1819 case Intrinsic::masked_store: {
1820 SmallVector<Value *, 4> Args(CI->arg_operands().begin(),
1821 CI->arg_operands().end());
1822 CI->replaceAllUsesWith(Builder.CreateCall(NewFn, Args));
1823 CI->eraseFromParent();
1824 return;
1825 }
Craig Topper71dc02d2012-06-13 07:18:53 +00001826 }
Chandler Carruth7132e002007-08-04 01:51:18 +00001827}
1828
Sanjay Patelfdf0d5f2016-04-18 19:11:57 +00001829void llvm::UpgradeCallsToIntrinsic(Function *F) {
Chandler Carruth7132e002007-08-04 01:51:18 +00001830 assert(F && "Illegal attempt to upgrade a non-existent intrinsic.");
1831
Sanjay Patelfdf0d5f2016-04-18 19:11:57 +00001832 // Check if this function should be upgraded and get the replacement function
1833 // if there is one.
Chris Lattner80ed9dc2011-06-18 06:05:24 +00001834 Function *NewFn;
Evan Cheng0e179d02007-12-17 22:33:23 +00001835 if (UpgradeIntrinsicFunction(F, NewFn)) {
Sanjay Patelfdf0d5f2016-04-18 19:11:57 +00001836 // Replace all users of the old function with the new function or new
1837 // instructions. This is not a range loop because the call is deleted.
1838 for (auto UI = F->user_begin(), UE = F->user_end(); UI != UE; )
Duncan P. N. Exon Smith93f53c42016-04-17 03:59:37 +00001839 if (CallInst *CI = dyn_cast<CallInst>(*UI++))
Filipe Cabecinhas0011c582015-07-03 20:12:01 +00001840 UpgradeIntrinsicCall(CI, NewFn);
Sanjay Patelfdf0d5f2016-04-18 19:11:57 +00001841
Filipe Cabecinhas0011c582015-07-03 20:12:01 +00001842 // Remove old function, no longer used, from the module.
1843 F->eraseFromParent();
Chandler Carruth7132e002007-08-04 01:51:18 +00001844 }
1845}
Devang Patel80ae3492009-08-28 23:24:31 +00001846
Mehdi Aminie4709272016-09-14 22:29:59 +00001847MDNode *llvm::UpgradeTBAANode(MDNode &MD) {
Manman Ren209b17c2013-09-28 00:22:27 +00001848 // Check if the tag uses struct-path aware TBAA format.
Mehdi Aminie4709272016-09-14 22:29:59 +00001849 if (isa<MDNode>(MD.getOperand(0)) && MD.getNumOperands() >= 3)
1850 return &MD;
Manman Ren209b17c2013-09-28 00:22:27 +00001851
Mehdi Aminie4709272016-09-14 22:29:59 +00001852 auto &Context = MD.getContext();
1853 if (MD.getNumOperands() == 3) {
1854 Metadata *Elts[] = {MD.getOperand(0), MD.getOperand(1)};
1855 MDNode *ScalarType = MDNode::get(Context, Elts);
Manman Ren209b17c2013-09-28 00:22:27 +00001856 // Create a MDNode <ScalarType, ScalarType, offset 0, const>
Duncan P. N. Exon Smith5bf8fef2014-12-09 18:38:53 +00001857 Metadata *Elts2[] = {ScalarType, ScalarType,
Mehdi Aminie4709272016-09-14 22:29:59 +00001858 ConstantAsMetadata::get(
1859 Constant::getNullValue(Type::getInt64Ty(Context))),
1860 MD.getOperand(2)};
1861 return MDNode::get(Context, Elts2);
Manman Ren209b17c2013-09-28 00:22:27 +00001862 }
Mehdi Aminie4709272016-09-14 22:29:59 +00001863 // Create a MDNode <MD, MD, offset 0>
1864 Metadata *Elts[] = {&MD, &MD, ConstantAsMetadata::get(Constant::getNullValue(
1865 Type::getInt64Ty(Context)))};
1866 return MDNode::get(Context, Elts);
Manman Ren209b17c2013-09-28 00:22:27 +00001867}
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001868
1869Instruction *llvm::UpgradeBitCastInst(unsigned Opc, Value *V, Type *DestTy,
1870 Instruction *&Temp) {
1871 if (Opc != Instruction::BitCast)
Craig Topperc6207612014-04-09 06:08:46 +00001872 return nullptr;
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001873
Craig Topperc6207612014-04-09 06:08:46 +00001874 Temp = nullptr;
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001875 Type *SrcTy = V->getType();
1876 if (SrcTy->isPtrOrPtrVectorTy() && DestTy->isPtrOrPtrVectorTy() &&
1877 SrcTy->getPointerAddressSpace() != DestTy->getPointerAddressSpace()) {
1878 LLVMContext &Context = V->getContext();
1879
1880 // We have no information about target data layout, so we assume that
1881 // the maximum pointer size is 64bit.
1882 Type *MidTy = Type::getInt64Ty(Context);
1883 Temp = CastInst::Create(Instruction::PtrToInt, V, MidTy);
1884
1885 return CastInst::Create(Instruction::IntToPtr, Temp, DestTy);
1886 }
1887
Craig Topperc6207612014-04-09 06:08:46 +00001888 return nullptr;
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001889}
1890
1891Value *llvm::UpgradeBitCastExpr(unsigned Opc, Constant *C, Type *DestTy) {
1892 if (Opc != Instruction::BitCast)
Craig Topperc6207612014-04-09 06:08:46 +00001893 return nullptr;
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001894
1895 Type *SrcTy = C->getType();
1896 if (SrcTy->isPtrOrPtrVectorTy() && DestTy->isPtrOrPtrVectorTy() &&
1897 SrcTy->getPointerAddressSpace() != DestTy->getPointerAddressSpace()) {
1898 LLVMContext &Context = C->getContext();
1899
1900 // We have no information about target data layout, so we assume that
1901 // the maximum pointer size is 64bit.
1902 Type *MidTy = Type::getInt64Ty(Context);
1903
1904 return ConstantExpr::getIntToPtr(ConstantExpr::getPtrToInt(C, MidTy),
1905 DestTy);
1906 }
1907
Craig Topperc6207612014-04-09 06:08:46 +00001908 return nullptr;
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001909}
Manman Ren8b4306c2013-12-02 21:29:56 +00001910
1911/// Check the debug info version number, if it is out-dated, drop the debug
1912/// info. Return true if module is modified.
1913bool llvm::UpgradeDebugInfo(Module &M) {
Manman Ren2ebfb422014-01-16 01:51:12 +00001914 unsigned Version = getDebugMetadataVersionFromModule(M);
1915 if (Version == DEBUG_METADATA_VERSION)
Manman Ren8b4306c2013-12-02 21:29:56 +00001916 return false;
1917
Manman Ren2ebfb422014-01-16 01:51:12 +00001918 bool RetCode = StripDebugInfo(M);
1919 if (RetCode) {
1920 DiagnosticInfoDebugMetadataVersion DiagVersion(M, Version);
1921 M.getContext().diagnose(DiagVersion);
1922 }
1923 return RetCode;
Manman Ren8b4306c2013-12-02 21:29:56 +00001924}
Eli Bendersky5d5e18d2014-06-25 15:41:00 +00001925
Manman Renb5d7ff42016-05-25 23:14:48 +00001926bool llvm::UpgradeModuleFlags(Module &M) {
1927 const NamedMDNode *ModFlags = M.getModuleFlagsMetadata();
1928 if (!ModFlags)
1929 return false;
1930
1931 bool HasObjCFlag = false, HasClassProperties = false;
1932 for (unsigned I = 0, E = ModFlags->getNumOperands(); I != E; ++I) {
1933 MDNode *Op = ModFlags->getOperand(I);
1934 if (Op->getNumOperands() < 2)
1935 continue;
1936 MDString *ID = dyn_cast_or_null<MDString>(Op->getOperand(1));
1937 if (!ID)
1938 continue;
1939 if (ID->getString() == "Objective-C Image Info Version")
1940 HasObjCFlag = true;
1941 if (ID->getString() == "Objective-C Class Properties")
1942 HasClassProperties = true;
1943 }
1944 // "Objective-C Class Properties" is recently added for Objective-C. We
1945 // upgrade ObjC bitcodes to contain a "Objective-C Class Properties" module
Mehdi Aminib53b62e2016-09-16 00:38:18 +00001946 // flag of value 0, so we can correclty downgrade this flag when trying to
1947 // link an ObjC bitcode without this module flag with an ObjC bitcode with
1948 // this module flag.
Manman Renb5d7ff42016-05-25 23:14:48 +00001949 if (HasObjCFlag && !HasClassProperties) {
Mehdi Aminib53b62e2016-09-16 00:38:18 +00001950 M.addModuleFlag(llvm::Module::Override, "Objective-C Class Properties",
Manman Renb5d7ff42016-05-25 23:14:48 +00001951 (uint32_t)0);
1952 return true;
1953 }
1954 return false;
1955}
1956
Duncan P. N. Exon Smithefe16c82016-03-25 00:56:13 +00001957static bool isOldLoopArgument(Metadata *MD) {
1958 auto *T = dyn_cast_or_null<MDTuple>(MD);
1959 if (!T)
1960 return false;
1961 if (T->getNumOperands() < 1)
1962 return false;
1963 auto *S = dyn_cast_or_null<MDString>(T->getOperand(0));
1964 if (!S)
1965 return false;
1966 return S->getString().startswith("llvm.vectorizer.");
1967}
1968
1969static MDString *upgradeLoopTag(LLVMContext &C, StringRef OldTag) {
1970 StringRef OldPrefix = "llvm.vectorizer.";
1971 assert(OldTag.startswith(OldPrefix) && "Expected old prefix");
1972
1973 if (OldTag == "llvm.vectorizer.unroll")
1974 return MDString::get(C, "llvm.loop.interleave.count");
1975
1976 return MDString::get(
1977 C, (Twine("llvm.loop.vectorize.") + OldTag.drop_front(OldPrefix.size()))
1978 .str());
1979}
1980
1981static Metadata *upgradeLoopArgument(Metadata *MD) {
1982 auto *T = dyn_cast_or_null<MDTuple>(MD);
1983 if (!T)
1984 return MD;
1985 if (T->getNumOperands() < 1)
1986 return MD;
1987 auto *OldTag = dyn_cast_or_null<MDString>(T->getOperand(0));
1988 if (!OldTag)
1989 return MD;
1990 if (!OldTag->getString().startswith("llvm.vectorizer."))
1991 return MD;
1992
1993 // This has an old tag. Upgrade it.
1994 SmallVector<Metadata *, 8> Ops;
1995 Ops.reserve(T->getNumOperands());
1996 Ops.push_back(upgradeLoopTag(T->getContext(), OldTag->getString()));
1997 for (unsigned I = 1, E = T->getNumOperands(); I != E; ++I)
1998 Ops.push_back(T->getOperand(I));
1999
2000 return MDTuple::get(T->getContext(), Ops);
2001}
2002
2003MDNode *llvm::upgradeInstructionLoopAttachment(MDNode &N) {
2004 auto *T = dyn_cast<MDTuple>(&N);
2005 if (!T)
2006 return &N;
2007
David Majnemer0a16c222016-08-11 21:15:00 +00002008 if (none_of(T->operands(), isOldLoopArgument))
Duncan P. N. Exon Smithefe16c82016-03-25 00:56:13 +00002009 return &N;
2010
2011 SmallVector<Metadata *, 8> Ops;
2012 Ops.reserve(T->getNumOperands());
2013 for (Metadata *MD : T->operands())
2014 Ops.push_back(upgradeLoopArgument(MD));
2015
2016 return MDTuple::get(T->getContext(), Ops);
Eli Bendersky5d5e18d2014-06-25 15:41:00 +00002017}