blob: fe7ba8b16077a493dc8b7c9dddd093ec30fffaaf [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
261 Name == "avx512.mask.pshuf.b.128" || // Added in 4.0
262 Name == "avx512.mask.pshuf.b.256" || // Added in 4.0
263 Name.startswith("avx2.pmax") || // Added in 3.9
264 Name.startswith("avx2.pmin") || // Added in 3.9
265 Name.startswith("avx512.mask.pmax") || // Added in 4.0
266 Name.startswith("avx512.mask.pmin") || // Added in 4.0
267 Name.startswith("avx2.vbroadcast") || // Added in 3.8
268 Name.startswith("avx2.pbroadcast") || // Added in 3.8
269 Name.startswith("avx.vpermil.") || // Added in 3.1
270 Name.startswith("sse2.pshuf") || // Added in 3.9
271 Name.startswith("avx512.pbroadcast") || // Added in 3.9
272 Name.startswith("avx512.mask.broadcast.s") || // Added in 3.9
273 Name.startswith("avx512.mask.movddup") || // Added in 3.9
274 Name.startswith("avx512.mask.movshdup") || // Added in 3.9
275 Name.startswith("avx512.mask.movsldup") || // Added in 3.9
276 Name.startswith("avx512.mask.pshuf.d.") || // Added in 3.9
277 Name.startswith("avx512.mask.pshufl.w.") || // Added in 3.9
278 Name.startswith("avx512.mask.pshufh.w.") || // Added in 3.9
279 Name.startswith("avx512.mask.shuf.p") || // Added in 4.0
280 Name.startswith("avx512.mask.vpermil.p") || // Added in 3.9
281 Name.startswith("avx512.mask.perm.df.") || // Added in 3.9
282 Name.startswith("avx512.mask.perm.di.") || // Added in 3.9
283 Name.startswith("avx512.mask.punpckl") || // Added in 3.9
284 Name.startswith("avx512.mask.punpckh") || // Added in 3.9
285 Name.startswith("avx512.mask.unpckl.") || // Added in 3.9
286 Name.startswith("avx512.mask.unpckh.") || // Added in 3.9
287 Name.startswith("avx512.mask.pand.") || // Added in 3.9
288 Name.startswith("avx512.mask.pandn.") || // Added in 3.9
289 Name.startswith("avx512.mask.por.") || // Added in 3.9
290 Name.startswith("avx512.mask.pxor.") || // Added in 3.9
291 Name.startswith("avx512.mask.and.") || // Added in 3.9
292 Name.startswith("avx512.mask.andn.") || // Added in 3.9
293 Name.startswith("avx512.mask.or.") || // Added in 3.9
294 Name.startswith("avx512.mask.xor.") || // Added in 3.9
295 Name.startswith("avx512.mask.padd.") || // Added in 4.0
296 Name.startswith("avx512.mask.psub.") || // Added in 4.0
297 Name.startswith("avx512.mask.pmull.") || // Added in 4.0
Simon Pilgrimb57dd172016-11-16 14:48:32 +0000298 Name.startswith("avx512.mask.cvtdq2pd.") || // Added in 4.0
299 Name.startswith("avx512.mask.cvtudq2pd.") || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000300 Name == "avx512.mask.add.pd.128" || // Added in 4.0
301 Name == "avx512.mask.add.pd.256" || // Added in 4.0
302 Name == "avx512.mask.add.ps.128" || // Added in 4.0
303 Name == "avx512.mask.add.ps.256" || // Added in 4.0
304 Name == "avx512.mask.div.pd.128" || // Added in 4.0
305 Name == "avx512.mask.div.pd.256" || // Added in 4.0
306 Name == "avx512.mask.div.ps.128" || // Added in 4.0
307 Name == "avx512.mask.div.ps.256" || // Added in 4.0
308 Name == "avx512.mask.mul.pd.128" || // Added in 4.0
309 Name == "avx512.mask.mul.pd.256" || // Added in 4.0
310 Name == "avx512.mask.mul.ps.128" || // Added in 4.0
311 Name == "avx512.mask.mul.ps.256" || // Added in 4.0
312 Name == "avx512.mask.sub.pd.128" || // Added in 4.0
313 Name == "avx512.mask.sub.pd.256" || // Added in 4.0
314 Name == "avx512.mask.sub.ps.128" || // Added in 4.0
315 Name == "avx512.mask.sub.ps.256" || // Added in 4.0
Craig Topperabe7c5b2016-12-10 21:15:52 +0000316 Name == "avx512.mask.vpermilvar.ps.128" || // Added in 4.0
317 Name == "avx512.mask.vpermilvar.ps.256" || // Added in 4.0
318 Name == "avx512.mask.vpermilvar.pd.128" || // Added in 4.0
319 Name == "avx512.mask.vpermilvar.pd.256" || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000320 Name.startswith("avx512.mask.psll.d") || // Added in 4.0
321 Name.startswith("avx512.mask.psll.q") || // Added in 4.0
322 Name.startswith("avx512.mask.psll.w") || // Added in 4.0
323 Name.startswith("avx512.mask.psra.d") || // Added in 4.0
324 Name.startswith("avx512.mask.psra.q") || // Added in 4.0
325 Name.startswith("avx512.mask.psra.w") || // Added in 4.0
326 Name.startswith("avx512.mask.psrl.d") || // Added in 4.0
327 Name.startswith("avx512.mask.psrl.q") || // Added in 4.0
328 Name.startswith("avx512.mask.psrl.w") || // Added in 4.0
329 Name.startswith("avx512.mask.pslli") || // Added in 4.0
330 Name.startswith("avx512.mask.psrai") || // Added in 4.0
331 Name.startswith("avx512.mask.psrli") || // Added in 4.0
Craig Topper02b5a1b2016-11-18 05:04:44 +0000332 Name.startswith("avx512.mask.psllv") || // Added in 4.0
333 Name.startswith("avx512.mask.psrav") || // Added in 4.0
334 Name.startswith("avx512.mask.psrlv") || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000335 Name.startswith("sse41.pmovsx") || // Added in 3.8
336 Name.startswith("sse41.pmovzx") || // Added in 3.9
337 Name.startswith("avx2.pmovsx") || // Added in 3.9
338 Name.startswith("avx2.pmovzx") || // Added in 3.9
339 Name.startswith("avx512.mask.pmovsx") || // Added in 4.0
340 Name.startswith("avx512.mask.pmovzx") || // Added in 4.0
341 Name == "sse2.cvtdq2pd" || // Added in 3.9
342 Name == "sse2.cvtps2pd" || // Added in 3.9
343 Name == "avx.cvtdq2.pd.256" || // Added in 3.9
344 Name == "avx.cvt.ps2.pd.256" || // Added in 3.9
345 Name.startswith("avx.vinsertf128.") || // Added in 3.7
346 Name == "avx2.vinserti128" || // Added in 3.7
347 Name.startswith("avx.vextractf128.") || // Added in 3.7
348 Name == "avx2.vextracti128" || // Added in 3.7
349 Name.startswith("sse4a.movnt.") || // Added in 3.9
350 Name.startswith("avx.movnt.") || // Added in 3.2
351 Name.startswith("avx512.storent.") || // Added in 3.9
352 Name == "sse2.storel.dq" || // Added in 3.9
353 Name.startswith("sse.storeu.") || // Added in 3.9
354 Name.startswith("sse2.storeu.") || // Added in 3.9
355 Name.startswith("avx.storeu.") || // Added in 3.9
356 Name.startswith("avx512.mask.storeu.") || // Added in 3.9
357 Name.startswith("avx512.mask.store.p") || // Added in 3.9
358 Name.startswith("avx512.mask.store.b.") || // Added in 3.9
359 Name.startswith("avx512.mask.store.w.") || // Added in 3.9
360 Name.startswith("avx512.mask.store.d.") || // Added in 3.9
361 Name.startswith("avx512.mask.store.q.") || // Added in 3.9
362 Name.startswith("avx512.mask.loadu.") || // Added in 3.9
363 Name.startswith("avx512.mask.load.") || // Added in 3.9
364 Name == "sse42.crc32.64.8" || // Added in 3.4
365 Name.startswith("avx.vbroadcast.s") || // Added in 3.5
366 Name.startswith("avx512.mask.palignr.") || // Added in 3.9
Craig Topperf57e17d2016-11-23 06:54:55 +0000367 Name.startswith("avx512.mask.valign.") || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000368 Name.startswith("sse2.psll.dq") || // Added in 3.7
369 Name.startswith("sse2.psrl.dq") || // Added in 3.7
370 Name.startswith("avx2.psll.dq") || // Added in 3.7
371 Name.startswith("avx2.psrl.dq") || // Added in 3.7
372 Name.startswith("avx512.psll.dq") || // Added in 3.9
373 Name.startswith("avx512.psrl.dq") || // Added in 3.9
374 Name == "sse41.pblendw" || // Added in 3.7
375 Name.startswith("sse41.blendp") || // Added in 3.7
376 Name.startswith("avx.blend.p") || // Added in 3.7
377 Name == "avx2.pblendw" || // Added in 3.7
378 Name.startswith("avx2.pblendd.") || // Added in 3.7
379 Name.startswith("avx.vbroadcastf128") || // Added in 4.0
380 Name == "avx2.vbroadcasti128" || // Added in 3.7
381 Name == "xop.vpcmov" || // Added in 3.8
Ayman Musa4d602432016-11-16 09:00:28 +0000382 Name.startswith("avx512.mask.move.s") || // Added in 4.0
Craig Toppere6915b82016-11-15 05:04:51 +0000383 (Name.startswith("xop.vpcom") && // Added in 3.2
384 F->arg_size() == 2))) {
Craig Topperc6207612014-04-09 06:08:46 +0000385 NewFn = nullptr;
Craig Topper3b1817d2012-02-03 06:10:55 +0000386 return true;
387 }
Nadav Rotem17ee58a2012-06-10 18:42:51 +0000388 // SSE4.1 ptest functions may have an old signature.
Craig Toppere6915b82016-11-15 05:04:51 +0000389 if (IsX86 && Name.startswith("sse41.ptest")) { // Added in 3.2
Craig Topper5aebb862016-07-04 20:56:38 +0000390 if (Name.substr(11) == "c")
Nadav Rotem17ee58a2012-06-10 18:42:51 +0000391 return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestc, NewFn);
Craig Topper5aebb862016-07-04 20:56:38 +0000392 if (Name.substr(11) == "z")
Nadav Rotem17ee58a2012-06-10 18:42:51 +0000393 return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestz, NewFn);
Craig Topper5aebb862016-07-04 20:56:38 +0000394 if (Name.substr(11) == "nzc")
Nadav Rotem17ee58a2012-06-10 18:42:51 +0000395 return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestnzc, NewFn);
396 }
Sanjay Patel1c3eaec2015-02-28 22:25:06 +0000397 // Several blend and other instructions with masks used the wrong number of
Chandler Carruth373b2b12014-09-06 10:00:01 +0000398 // bits.
Craig Toppere6915b82016-11-15 05:04:51 +0000399 if (IsX86 && Name == "sse41.insertps") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000400 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_insertps,
401 NewFn);
Craig Toppere6915b82016-11-15 05:04:51 +0000402 if (IsX86 && Name == "sse41.dppd") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000403 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_dppd,
404 NewFn);
Craig Toppere6915b82016-11-15 05:04:51 +0000405 if (IsX86 && Name == "sse41.dpps") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000406 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_dpps,
407 NewFn);
Craig Toppere6915b82016-11-15 05:04:51 +0000408 if (IsX86 && Name == "sse41.mpsadbw") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000409 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_mpsadbw,
410 NewFn);
Craig Toppere6915b82016-11-15 05:04:51 +0000411 if (IsX86 && Name == "avx.dp.ps.256") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000412 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_avx_dp_ps_256,
413 NewFn);
Craig Toppere6915b82016-11-15 05:04:51 +0000414 if (IsX86 && Name == "avx2.mpsadbw") // Added in 3.6
Chandler Carruth373b2b12014-09-06 10:00:01 +0000415 return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_avx2_mpsadbw,
416 NewFn);
Craig Topper29f2e952015-01-25 23:26:02 +0000417
Craig Toppere6915b82016-11-15 05:04:51 +0000418 // frcz.ss/sd may need to have an argument dropped. Added in 3.2
Craig Topper5aebb862016-07-04 20:56:38 +0000419 if (IsX86 && Name.startswith("xop.vfrcz.ss") && F->arg_size() == 2) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000420 rename(F);
Craig Topper71dc02d2012-06-13 07:18:53 +0000421 NewFn = Intrinsic::getDeclaration(F->getParent(),
422 Intrinsic::x86_xop_vfrcz_ss);
423 return true;
424 }
Craig Topper5aebb862016-07-04 20:56:38 +0000425 if (IsX86 && Name.startswith("xop.vfrcz.sd") && F->arg_size() == 2) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000426 rename(F);
Craig Topper71dc02d2012-06-13 07:18:53 +0000427 NewFn = Intrinsic::getDeclaration(F->getParent(),
428 Intrinsic::x86_xop_vfrcz_sd);
429 return true;
430 }
Simon Pilgrime85506b2016-06-03 08:06:03 +0000431 // Upgrade any XOP PERMIL2 index operand still using a float/double vector.
Craig Toppere6915b82016-11-15 05:04:51 +0000432 if (IsX86 && Name.startswith("xop.vpermil2")) { // Added in 3.9
Simon Pilgrime85506b2016-06-03 08:06:03 +0000433 auto Params = F->getFunctionType()->params();
434 auto Idx = Params[2];
435 if (Idx->getScalarType()->isFloatingPointTy()) {
Rafael Espindolad7325ee2016-10-03 15:51:42 +0000436 rename(F);
Simon Pilgrime85506b2016-06-03 08:06:03 +0000437 unsigned IdxSize = Idx->getPrimitiveSizeInBits();
438 unsigned EltSize = Idx->getScalarSizeInBits();
439 Intrinsic::ID Permil2ID;
440 if (EltSize == 64 && IdxSize == 128)
441 Permil2ID = Intrinsic::x86_xop_vpermil2pd;
442 else if (EltSize == 32 && IdxSize == 128)
443 Permil2ID = Intrinsic::x86_xop_vpermil2ps;
444 else if (EltSize == 64 && IdxSize == 256)
445 Permil2ID = Intrinsic::x86_xop_vpermil2pd_256;
446 else
447 Permil2ID = Intrinsic::x86_xop_vpermil2ps_256;
448 NewFn = Intrinsic::getDeclaration(F->getParent(), Permil2ID);
449 return true;
450 }
451 }
Craig Topper3b1817d2012-02-03 06:10:55 +0000452 break;
453 }
Chris Lattnerb372f662011-06-18 18:56:39 +0000454 }
Chandler Carruth7132e002007-08-04 01:51:18 +0000455
Nadav Rotem17ee58a2012-06-10 18:42:51 +0000456 // This may not belong here. This function is effectively being overloaded
457 // to both detect an intrinsic which needs upgrading, and to provide the
458 // upgraded form of the intrinsic. We should perhaps have two separate
Chandler Carruth7132e002007-08-04 01:51:18 +0000459 // functions for this.
Evan Cheng0e179d02007-12-17 22:33:23 +0000460 return false;
Chandler Carruth7132e002007-08-04 01:51:18 +0000461}
462
Evan Cheng0e179d02007-12-17 22:33:23 +0000463bool llvm::UpgradeIntrinsicFunction(Function *F, Function *&NewFn) {
Craig Topperc6207612014-04-09 06:08:46 +0000464 NewFn = nullptr;
Evan Cheng0e179d02007-12-17 22:33:23 +0000465 bool Upgraded = UpgradeIntrinsicFunction1(F, NewFn);
Filipe Cabecinhas0011c582015-07-03 20:12:01 +0000466 assert(F != NewFn && "Intrinsic function upgraded to the same function");
Duncan Sands38ef3a82007-12-03 20:06:50 +0000467
468 // Upgrade intrinsic attributes. This does not change the function.
Evan Cheng0e179d02007-12-17 22:33:23 +0000469 if (NewFn)
470 F = NewFn;
Pete Cooper9e1d3352015-05-20 17:16:39 +0000471 if (Intrinsic::ID id = F->getIntrinsicID())
472 F->setAttributes(Intrinsic::getAttributes(F->getContext(), id));
Duncan Sands38ef3a82007-12-03 20:06:50 +0000473 return Upgraded;
474}
475
Bill Wendlinge26fffc2010-09-10 18:51:56 +0000476bool llvm::UpgradeGlobalVariable(GlobalVariable *GV) {
Chris Lattner80ed9dc2011-06-18 06:05:24 +0000477 // Nothing to do yet.
Bill Wendlinge26fffc2010-09-10 18:51:56 +0000478 return false;
479}
480
Simon Pilgrimf7186822016-06-09 21:09:03 +0000481// Handles upgrading SSE2/AVX2/AVX512BW PSLLDQ intrinsics by converting them
Craig Topperb324e432015-02-18 06:24:44 +0000482// to byte shuffles.
Craig Topper46b34fe2016-07-12 01:42:33 +0000483static Value *UpgradeX86PSLLDQIntrinsics(IRBuilder<> &Builder,
Craig Topper7355ac32016-05-29 06:37:33 +0000484 Value *Op, unsigned Shift) {
485 Type *ResultTy = Op->getType();
486 unsigned NumElts = ResultTy->getVectorNumElements() * 8;
Craig Topperb324e432015-02-18 06:24:44 +0000487
488 // Bitcast from a 64-bit element type to a byte element type.
Craig Topper46b34fe2016-07-12 01:42:33 +0000489 Type *VecTy = VectorType::get(Builder.getInt8Ty(), NumElts);
Craig Topper7355ac32016-05-29 06:37:33 +0000490 Op = Builder.CreateBitCast(Op, VecTy, "cast");
491
Craig Topperb324e432015-02-18 06:24:44 +0000492 // We'll be shuffling in zeroes.
Craig Topper7355ac32016-05-29 06:37:33 +0000493 Value *Res = Constant::getNullValue(VecTy);
Craig Topperb324e432015-02-18 06:24:44 +0000494
495 // If shift is less than 16, emit a shuffle to move the bytes. Otherwise,
496 // we'll just return the zero vector.
497 if (Shift < 16) {
Craig Topper99d1eab2016-06-12 00:41:19 +0000498 uint32_t Idxs[64];
Simon Pilgrimf7186822016-06-09 21:09:03 +0000499 // 256/512-bit version is split into 2/4 16-byte lanes.
Craig Topperb324e432015-02-18 06:24:44 +0000500 for (unsigned l = 0; l != NumElts; l += 16)
501 for (unsigned i = 0; i != 16; ++i) {
502 unsigned Idx = NumElts + i - Shift;
503 if (Idx < NumElts)
504 Idx -= NumElts - 16; // end of lane, switch operand.
Craig Topper7355ac32016-05-29 06:37:33 +0000505 Idxs[l + i] = Idx + l;
Craig Topperb324e432015-02-18 06:24:44 +0000506 }
507
Craig Topper7355ac32016-05-29 06:37:33 +0000508 Res = Builder.CreateShuffleVector(Res, Op, makeArrayRef(Idxs, NumElts));
Craig Topperb324e432015-02-18 06:24:44 +0000509 }
510
511 // Bitcast back to a 64-bit element type.
Craig Topper7355ac32016-05-29 06:37:33 +0000512 return Builder.CreateBitCast(Res, ResultTy, "cast");
Craig Topperb324e432015-02-18 06:24:44 +0000513}
514
Craig Topperea703ae2016-06-13 02:36:42 +0000515// Handles upgrading SSE2/AVX2/AVX512BW PSRLDQ intrinsics by converting them
516// to byte shuffles.
Craig Topper46b34fe2016-07-12 01:42:33 +0000517static Value *UpgradeX86PSRLDQIntrinsics(IRBuilder<> &Builder, Value *Op,
Craig Topperea703ae2016-06-13 02:36:42 +0000518 unsigned Shift) {
519 Type *ResultTy = Op->getType();
520 unsigned NumElts = ResultTy->getVectorNumElements() * 8;
521
522 // Bitcast from a 64-bit element type to a byte element type.
Craig Topper46b34fe2016-07-12 01:42:33 +0000523 Type *VecTy = VectorType::get(Builder.getInt8Ty(), NumElts);
Craig Topperea703ae2016-06-13 02:36:42 +0000524 Op = Builder.CreateBitCast(Op, VecTy, "cast");
525
526 // We'll be shuffling in zeroes.
527 Value *Res = Constant::getNullValue(VecTy);
528
529 // If shift is less than 16, emit a shuffle to move the bytes. Otherwise,
530 // we'll just return the zero vector.
531 if (Shift < 16) {
532 uint32_t Idxs[64];
533 // 256/512-bit version is split into 2/4 16-byte lanes.
534 for (unsigned l = 0; l != NumElts; l += 16)
535 for (unsigned i = 0; i != 16; ++i) {
536 unsigned Idx = i + Shift;
537 if (Idx >= 16)
538 Idx += NumElts - 16; // end of lane, switch operand.
539 Idxs[l + i] = Idx + l;
540 }
541
542 Res = Builder.CreateShuffleVector(Op, Res, makeArrayRef(Idxs, NumElts));
543 }
544
545 // Bitcast back to a 64-bit element type.
546 return Builder.CreateBitCast(Res, ResultTy, "cast");
547}
548
549static Value *getX86MaskVec(IRBuilder<> &Builder, Value *Mask,
550 unsigned NumElts) {
551 llvm::VectorType *MaskTy = llvm::VectorType::get(Builder.getInt1Ty(),
552 cast<IntegerType>(Mask->getType())->getBitWidth());
553 Mask = Builder.CreateBitCast(Mask, MaskTy);
554
555 // If we have less than 8 elements, then the starting mask was an i8 and
556 // we need to extract down to the right number of elements.
557 if (NumElts < 8) {
558 uint32_t Indices[4];
559 for (unsigned i = 0; i != NumElts; ++i)
560 Indices[i] = i;
561 Mask = Builder.CreateShuffleVector(Mask, Mask,
562 makeArrayRef(Indices, NumElts),
563 "extract");
564 }
565
566 return Mask;
567}
568
569static Value *EmitX86Select(IRBuilder<> &Builder, Value *Mask,
570 Value *Op0, Value *Op1) {
571 // If the mask is all ones just emit the align operation.
572 if (const auto *C = dyn_cast<Constant>(Mask))
573 if (C->isAllOnesValue())
574 return Op0;
575
576 Mask = getX86MaskVec(Builder, Mask, Op0->getType()->getVectorNumElements());
577 return Builder.CreateSelect(Mask, Op0, Op1);
578}
579
Craig Topperf57e17d2016-11-23 06:54:55 +0000580// Handle autoupgrade for masked PALIGNR and VALIGND/Q intrinsics.
581// PALIGNR handles large immediates by shifting while VALIGN masks the immediate
582// so we need to handle both cases. VALIGN also doesn't have 128-bit lanes.
583static Value *UpgradeX86ALIGNIntrinsics(IRBuilder<> &Builder, Value *Op0,
584 Value *Op1, Value *Shift,
585 Value *Passthru, Value *Mask,
586 bool IsVALIGN) {
Craig Topper33350cc2016-06-06 06:12:54 +0000587 unsigned ShiftVal = cast<llvm::ConstantInt>(Shift)->getZExtValue();
588
589 unsigned NumElts = Op0->getType()->getVectorNumElements();
Craig Topperf57e17d2016-11-23 06:54:55 +0000590 assert((IsVALIGN || NumElts % 16 == 0) && "Illegal NumElts for PALIGNR!");
591 assert((!IsVALIGN || NumElts <= 16) && "NumElts too large for VALIGN!");
592 assert(isPowerOf2_32(NumElts) && "NumElts not a power of 2!");
593
594 // Mask the immediate for VALIGN.
595 if (IsVALIGN)
596 ShiftVal &= (NumElts - 1);
Craig Topper33350cc2016-06-06 06:12:54 +0000597
598 // If palignr is shifting the pair of vectors more than the size of two
599 // lanes, emit zero.
600 if (ShiftVal >= 32)
601 return llvm::Constant::getNullValue(Op0->getType());
602
603 // If palignr is shifting the pair of input vectors more than one lane,
604 // but less than two lanes, convert to shifting in zeroes.
605 if (ShiftVal > 16) {
606 ShiftVal -= 16;
607 Op1 = Op0;
608 Op0 = llvm::Constant::getNullValue(Op0->getType());
609 }
610
Craig Topper99d1eab2016-06-12 00:41:19 +0000611 uint32_t Indices[64];
Craig Topper33350cc2016-06-06 06:12:54 +0000612 // 256-bit palignr operates on 128-bit lanes so we need to handle that
Craig Topperf57e17d2016-11-23 06:54:55 +0000613 for (unsigned l = 0; l < NumElts; l += 16) {
Craig Topper33350cc2016-06-06 06:12:54 +0000614 for (unsigned i = 0; i != 16; ++i) {
615 unsigned Idx = ShiftVal + i;
Craig Topperf57e17d2016-11-23 06:54:55 +0000616 if (!IsVALIGN && Idx >= 16) // Disable wrap for VALIGN.
Craig Topper33350cc2016-06-06 06:12:54 +0000617 Idx += NumElts - 16; // End of lane, switch operand.
618 Indices[l + i] = Idx + l;
619 }
620 }
621
622 Value *Align = Builder.CreateShuffleVector(Op1, Op0,
623 makeArrayRef(Indices, NumElts),
624 "palignr");
625
Craig Topperea703ae2016-06-13 02:36:42 +0000626 return EmitX86Select(Builder, Mask, Align, Passthru);
Craig Topperb324e432015-02-18 06:24:44 +0000627}
628
Craig Topper46b34fe2016-07-12 01:42:33 +0000629static Value *UpgradeMaskedStore(IRBuilder<> &Builder,
Craig Topper50f85c22016-05-31 01:50:02 +0000630 Value *Ptr, Value *Data, Value *Mask,
631 bool Aligned) {
632 // Cast the pointer to the right type.
633 Ptr = Builder.CreateBitCast(Ptr,
634 llvm::PointerType::getUnqual(Data->getType()));
635 unsigned Align =
636 Aligned ? cast<VectorType>(Data->getType())->getBitWidth() / 8 : 1;
637
638 // If the mask is all ones just emit a regular store.
639 if (const auto *C = dyn_cast<Constant>(Mask))
640 if (C->isAllOnesValue())
641 return Builder.CreateAlignedStore(Data, Ptr, Align);
642
643 // Convert the mask from an integer type to a vector of i1.
644 unsigned NumElts = Data->getType()->getVectorNumElements();
Craig Topperea703ae2016-06-13 02:36:42 +0000645 Mask = getX86MaskVec(Builder, Mask, NumElts);
Craig Topper50f85c22016-05-31 01:50:02 +0000646 return Builder.CreateMaskedStore(Data, Ptr, Align, Mask);
647}
648
Craig Topper46b34fe2016-07-12 01:42:33 +0000649static Value *UpgradeMaskedLoad(IRBuilder<> &Builder,
Craig Topperf10fbfa2016-06-02 04:19:36 +0000650 Value *Ptr, Value *Passthru, Value *Mask,
651 bool Aligned) {
652 // Cast the pointer to the right type.
653 Ptr = Builder.CreateBitCast(Ptr,
654 llvm::PointerType::getUnqual(Passthru->getType()));
655 unsigned Align =
656 Aligned ? cast<VectorType>(Passthru->getType())->getBitWidth() / 8 : 1;
657
658 // If the mask is all ones just emit a regular store.
659 if (const auto *C = dyn_cast<Constant>(Mask))
660 if (C->isAllOnesValue())
661 return Builder.CreateAlignedLoad(Ptr, Align);
662
663 // Convert the mask from an integer type to a vector of i1.
664 unsigned NumElts = Passthru->getType()->getVectorNumElements();
Craig Topperea703ae2016-06-13 02:36:42 +0000665 Mask = getX86MaskVec(Builder, Mask, NumElts);
Craig Topperf10fbfa2016-06-02 04:19:36 +0000666 return Builder.CreateMaskedLoad(Ptr, Align, Mask, Passthru);
667}
668
Sanjay Patel51ab7572016-06-16 15:48:30 +0000669static Value *upgradeIntMinMax(IRBuilder<> &Builder, CallInst &CI,
670 ICmpInst::Predicate Pred) {
671 Value *Op0 = CI.getArgOperand(0);
672 Value *Op1 = CI.getArgOperand(1);
673 Value *Cmp = Builder.CreateICmp(Pred, Op0, Op1);
Craig Topper8ec5c732016-10-24 04:04:16 +0000674 Value *Res = Builder.CreateSelect(Cmp, Op0, Op1);
675
676 if (CI.getNumArgOperands() == 4)
677 Res = EmitX86Select(Builder, CI.getArgOperand(3), Res, CI.getArgOperand(2));
678
679 return Res;
Sanjay Patel51ab7572016-06-16 15:48:30 +0000680}
681
Craig Topper0a0fb0f2016-06-21 03:53:24 +0000682static Value *upgradeMaskedCompare(IRBuilder<> &Builder, CallInst &CI,
683 ICmpInst::Predicate Pred) {
684 Value *Op0 = CI.getArgOperand(0);
685 unsigned NumElts = Op0->getType()->getVectorNumElements();
686 Value *Cmp = Builder.CreateICmp(Pred, Op0, CI.getArgOperand(1));
687
688 Value *Mask = CI.getArgOperand(2);
689 const auto *C = dyn_cast<Constant>(Mask);
690 if (!C || !C->isAllOnesValue())
691 Cmp = Builder.CreateAnd(Cmp, getX86MaskVec(Builder, Mask, NumElts));
692
693 if (NumElts < 8) {
694 uint32_t Indices[8];
695 for (unsigned i = 0; i != NumElts; ++i)
696 Indices[i] = i;
697 for (unsigned i = NumElts; i != 8; ++i)
Craig Topperd5d2a352016-07-07 06:11:07 +0000698 Indices[i] = NumElts + i % NumElts;
699 Cmp = Builder.CreateShuffleVector(Cmp,
700 Constant::getNullValue(Cmp->getType()),
Craig Topper0a0fb0f2016-06-21 03:53:24 +0000701 Indices);
702 }
703 return Builder.CreateBitCast(Cmp, IntegerType::get(CI.getContext(),
704 std::max(NumElts, 8U)));
705}
706
Craig Topperc9467ed2016-11-06 16:29:08 +0000707// Replace a masked intrinsic with an older unmasked intrinsic.
708static Value *UpgradeX86MaskedShift(IRBuilder<> &Builder, CallInst &CI,
709 Intrinsic::ID IID) {
710 Function *F = CI.getCalledFunction();
711 Function *Intrin = Intrinsic::getDeclaration(F->getParent(), IID);
712 Value *Rep = Builder.CreateCall(Intrin,
713 { CI.getArgOperand(0), CI.getArgOperand(1) });
714 return EmitX86Select(Builder, CI.getArgOperand(3), Rep, CI.getArgOperand(2));
715}
716
Ayman Musa4d602432016-11-16 09:00:28 +0000717static Value* upgradeMaskedMove(IRBuilder<> &Builder, CallInst &CI) {
718 Value* A = CI.getArgOperand(0);
719 Value* B = CI.getArgOperand(1);
720 Value* Src = CI.getArgOperand(2);
721 Value* Mask = CI.getArgOperand(3);
722
723 Value* AndNode = Builder.CreateAnd(Mask, APInt(8, 1));
724 Value* Cmp = Builder.CreateIsNotNull(AndNode);
725 Value* Extract1 = Builder.CreateExtractElement(B, (uint64_t)0);
726 Value* Extract2 = Builder.CreateExtractElement(Src, (uint64_t)0);
727 Value* Select = Builder.CreateSelect(Cmp, Extract1, Extract2);
728 return Builder.CreateInsertElement(A, Select, (uint64_t)0);
729}
Craig Topperc9467ed2016-11-06 16:29:08 +0000730
Sanjay Patel595098f2016-06-15 22:01:28 +0000731/// Upgrade a call to an old intrinsic. All argument and return casting must be
732/// provided to seamlessly integrate with existing context.
Chandler Carruth7132e002007-08-04 01:51:18 +0000733void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
Craig Topper3b1817d2012-02-03 06:10:55 +0000734 Function *F = CI->getCalledFunction();
Nick Lewycky2eb3ade2011-12-12 22:59:34 +0000735 LLVMContext &C = CI->getContext();
Chandler Carruth58a71ed2011-12-12 04:26:04 +0000736 IRBuilder<> Builder(C);
Duncan P. N. Exon Smith52888a62015-10-08 23:49:46 +0000737 Builder.SetInsertPoint(CI->getParent(), CI->getIterator());
Chandler Carruth58a71ed2011-12-12 04:26:04 +0000738
Craig Topper3b1817d2012-02-03 06:10:55 +0000739 assert(F && "Intrinsic call is not direct?");
740
741 if (!NewFn) {
742 // Get the Function's name.
743 StringRef Name = F->getName();
744
Craig Topper5aebb862016-07-04 20:56:38 +0000745 assert(Name.startswith("llvm.") && "Intrinsic doesn't start with 'llvm.'");
746 Name = Name.substr(5);
747
748 bool IsX86 = Name.startswith("x86.");
749 if (IsX86)
750 Name = Name.substr(4);
751
Craig Toppera4744d12016-12-10 21:15:48 +0000752 if (IsX86 && Name.startswith("sse4a.movnt.")) {
753 Module *M = F->getParent();
754 SmallVector<Metadata *, 1> Elts;
755 Elts.push_back(
756 ConstantAsMetadata::get(ConstantInt::get(Type::getInt32Ty(C), 1)));
757 MDNode *Node = MDNode::get(C, Elts);
758
759 Value *Arg0 = CI->getArgOperand(0);
760 Value *Arg1 = CI->getArgOperand(1);
761
762 // Nontemporal (unaligned) store of the 0'th element of the float/double
763 // vector.
764 Type *SrcEltTy = cast<VectorType>(Arg1->getType())->getElementType();
765 PointerType *EltPtrTy = PointerType::getUnqual(SrcEltTy);
766 Value *Addr = Builder.CreateBitCast(Arg0, EltPtrTy, "cast");
767 Value *Extract =
768 Builder.CreateExtractElement(Arg1, (uint64_t)0, "extractelement");
769
770 StoreInst *SI = Builder.CreateAlignedStore(Extract, Addr, 1);
771 SI->setMetadata(M->getMDKindID("nontemporal"), Node);
772
773 // Remove intrinsic.
774 CI->eraseFromParent();
775 return;
776 }
777
778 if (IsX86 && (Name.startswith("avx.movnt.") ||
779 Name.startswith("avx512.storent."))) {
780 Module *M = F->getParent();
781 SmallVector<Metadata *, 1> Elts;
782 Elts.push_back(
783 ConstantAsMetadata::get(ConstantInt::get(Type::getInt32Ty(C), 1)));
784 MDNode *Node = MDNode::get(C, Elts);
785
786 Value *Arg0 = CI->getArgOperand(0);
787 Value *Arg1 = CI->getArgOperand(1);
788
789 // Convert the type of the pointer to a pointer to the stored type.
790 Value *BC = Builder.CreateBitCast(Arg0,
791 PointerType::getUnqual(Arg1->getType()),
792 "cast");
793 VectorType *VTy = cast<VectorType>(Arg1->getType());
794 StoreInst *SI = Builder.CreateAlignedStore(Arg1, BC,
795 VTy->getBitWidth() / 8);
796 SI->setMetadata(M->getMDKindID("nontemporal"), Node);
797
798 // Remove intrinsic.
799 CI->eraseFromParent();
800 return;
801 }
802
803 if (IsX86 && Name == "sse2.storel.dq") {
804 Value *Arg0 = CI->getArgOperand(0);
805 Value *Arg1 = CI->getArgOperand(1);
806
807 Type *NewVecTy = VectorType::get(Type::getInt64Ty(C), 2);
808 Value *BC0 = Builder.CreateBitCast(Arg1, NewVecTy, "cast");
809 Value *Elt = Builder.CreateExtractElement(BC0, (uint64_t)0);
810 Value *BC = Builder.CreateBitCast(Arg0,
811 PointerType::getUnqual(Elt->getType()),
812 "cast");
813 Builder.CreateAlignedStore(Elt, BC, 1);
814
815 // Remove intrinsic.
816 CI->eraseFromParent();
817 return;
818 }
819
820 if (IsX86 && (Name.startswith("sse.storeu.") ||
821 Name.startswith("sse2.storeu.") ||
822 Name.startswith("avx.storeu."))) {
823 Value *Arg0 = CI->getArgOperand(0);
824 Value *Arg1 = CI->getArgOperand(1);
825
826 Arg0 = Builder.CreateBitCast(Arg0,
827 PointerType::getUnqual(Arg1->getType()),
828 "cast");
829 Builder.CreateAlignedStore(Arg1, Arg0, 1);
830
831 // Remove intrinsic.
832 CI->eraseFromParent();
833 return;
834 }
835
836 if (IsX86 && (Name.startswith("avx512.mask.storeu."))) {
837 UpgradeMaskedStore(Builder, CI->getArgOperand(0), CI->getArgOperand(1),
838 CI->getArgOperand(2), /*Aligned*/false);
839
840 // Remove intrinsic.
841 CI->eraseFromParent();
842 return;
843 }
844
845 if (IsX86 && (Name.startswith("avx512.mask.store."))) {
846 UpgradeMaskedStore(Builder, CI->getArgOperand(0), CI->getArgOperand(1),
847 CI->getArgOperand(2), /*Aligned*/true);
848
849 // Remove intrinsic.
850 CI->eraseFromParent();
851 return;
852 }
853
Craig Topper3b1817d2012-02-03 06:10:55 +0000854 Value *Rep;
Sanjay Patel595098f2016-06-15 22:01:28 +0000855 // Upgrade packed integer vector compare intrinsics to compare instructions.
Craig Topper5aebb862016-07-04 20:56:38 +0000856 if (IsX86 && (Name.startswith("sse2.pcmpeq.") ||
857 Name.startswith("avx2.pcmpeq."))) {
Craig Topper3b1817d2012-02-03 06:10:55 +0000858 Rep = Builder.CreateICmpEQ(CI->getArgOperand(0), CI->getArgOperand(1),
859 "pcmpeq");
Craig Topper3b1817d2012-02-03 06:10:55 +0000860 Rep = Builder.CreateSExt(Rep, CI->getType(), "");
Craig Topper5aebb862016-07-04 20:56:38 +0000861 } else if (IsX86 && (Name.startswith("sse2.pcmpgt.") ||
862 Name.startswith("avx2.pcmpgt."))) {
Craig Topper3b1817d2012-02-03 06:10:55 +0000863 Rep = Builder.CreateICmpSGT(CI->getArgOperand(0), CI->getArgOperand(1),
864 "pcmpgt");
Craig Topper3b1817d2012-02-03 06:10:55 +0000865 Rep = Builder.CreateSExt(Rep, CI->getType(), "");
Craig Topper6910fa02016-11-16 05:24:10 +0000866 } else if (IsX86 && (Name == "sse.add.ss" || Name == "sse2.add.sd")) {
867 Type *I32Ty = Type::getInt32Ty(C);
868 Value *Elt0 = Builder.CreateExtractElement(CI->getArgOperand(0),
869 ConstantInt::get(I32Ty, 0));
870 Value *Elt1 = Builder.CreateExtractElement(CI->getArgOperand(1),
871 ConstantInt::get(I32Ty, 0));
872 Rep = Builder.CreateInsertElement(CI->getArgOperand(0),
873 Builder.CreateFAdd(Elt0, Elt1),
874 ConstantInt::get(I32Ty, 0));
875 } else if (IsX86 && (Name == "sse.sub.ss" || Name == "sse2.sub.sd")) {
876 Type *I32Ty = Type::getInt32Ty(C);
877 Value *Elt0 = Builder.CreateExtractElement(CI->getArgOperand(0),
878 ConstantInt::get(I32Ty, 0));
879 Value *Elt1 = Builder.CreateExtractElement(CI->getArgOperand(1),
880 ConstantInt::get(I32Ty, 0));
881 Rep = Builder.CreateInsertElement(CI->getArgOperand(0),
882 Builder.CreateFSub(Elt0, Elt1),
883 ConstantInt::get(I32Ty, 0));
884 } else if (IsX86 && (Name == "sse.mul.ss" || Name == "sse2.mul.sd")) {
885 Type *I32Ty = Type::getInt32Ty(C);
886 Value *Elt0 = Builder.CreateExtractElement(CI->getArgOperand(0),
887 ConstantInt::get(I32Ty, 0));
888 Value *Elt1 = Builder.CreateExtractElement(CI->getArgOperand(1),
889 ConstantInt::get(I32Ty, 0));
890 Rep = Builder.CreateInsertElement(CI->getArgOperand(0),
891 Builder.CreateFMul(Elt0, Elt1),
892 ConstantInt::get(I32Ty, 0));
893 } else if (IsX86 && (Name == "sse.div.ss" || Name == "sse2.div.sd")) {
894 Type *I32Ty = Type::getInt32Ty(C);
895 Value *Elt0 = Builder.CreateExtractElement(CI->getArgOperand(0),
896 ConstantInt::get(I32Ty, 0));
897 Value *Elt1 = Builder.CreateExtractElement(CI->getArgOperand(1),
898 ConstantInt::get(I32Ty, 0));
899 Rep = Builder.CreateInsertElement(CI->getArgOperand(0),
900 Builder.CreateFDiv(Elt0, Elt1),
901 ConstantInt::get(I32Ty, 0));
Craig Topper5aebb862016-07-04 20:56:38 +0000902 } else if (IsX86 && Name.startswith("avx512.mask.pcmpeq.")) {
Craig Topper0a0fb0f2016-06-21 03:53:24 +0000903 Rep = upgradeMaskedCompare(Builder, *CI, ICmpInst::ICMP_EQ);
Craig Topper5aebb862016-07-04 20:56:38 +0000904 } else if (IsX86 && Name.startswith("avx512.mask.pcmpgt.")) {
Craig Topper0a0fb0f2016-06-21 03:53:24 +0000905 Rep = upgradeMaskedCompare(Builder, *CI, ICmpInst::ICMP_SGT);
Craig Topper5aebb862016-07-04 20:56:38 +0000906 } else if (IsX86 && (Name == "sse41.pmaxsb" ||
907 Name == "sse2.pmaxs.w" ||
908 Name == "sse41.pmaxsd" ||
Craig Topper8ec5c732016-10-24 04:04:16 +0000909 Name.startswith("avx2.pmaxs") ||
910 Name.startswith("avx512.mask.pmaxs"))) {
Sanjay Patel51ab7572016-06-16 15:48:30 +0000911 Rep = upgradeIntMinMax(Builder, *CI, ICmpInst::ICMP_SGT);
Craig Topper5aebb862016-07-04 20:56:38 +0000912 } else if (IsX86 && (Name == "sse2.pmaxu.b" ||
913 Name == "sse41.pmaxuw" ||
914 Name == "sse41.pmaxud" ||
Craig Topper8ec5c732016-10-24 04:04:16 +0000915 Name.startswith("avx2.pmaxu") ||
916 Name.startswith("avx512.mask.pmaxu"))) {
Sanjay Patel51ab7572016-06-16 15:48:30 +0000917 Rep = upgradeIntMinMax(Builder, *CI, ICmpInst::ICMP_UGT);
Craig Topper5aebb862016-07-04 20:56:38 +0000918 } else if (IsX86 && (Name == "sse41.pminsb" ||
919 Name == "sse2.pmins.w" ||
920 Name == "sse41.pminsd" ||
Craig Topper8ec5c732016-10-24 04:04:16 +0000921 Name.startswith("avx2.pmins") ||
922 Name.startswith("avx512.mask.pmins"))) {
Sanjay Patel51ab7572016-06-16 15:48:30 +0000923 Rep = upgradeIntMinMax(Builder, *CI, ICmpInst::ICMP_SLT);
Craig Topper5aebb862016-07-04 20:56:38 +0000924 } else if (IsX86 && (Name == "sse2.pminu.b" ||
925 Name == "sse41.pminuw" ||
926 Name == "sse41.pminud" ||
Craig Topper8ec5c732016-10-24 04:04:16 +0000927 Name.startswith("avx2.pminu") ||
928 Name.startswith("avx512.mask.pminu"))) {
Sanjay Patel51ab7572016-06-16 15:48:30 +0000929 Rep = upgradeIntMinMax(Builder, *CI, ICmpInst::ICMP_ULT);
Craig Topper5aebb862016-07-04 20:56:38 +0000930 } else if (IsX86 && (Name == "sse2.cvtdq2pd" ||
931 Name == "sse2.cvtps2pd" ||
932 Name == "avx.cvtdq2.pd.256" ||
Simon Pilgrimb57dd172016-11-16 14:48:32 +0000933 Name == "avx.cvt.ps2.pd.256" ||
934 Name.startswith("avx512.mask.cvtdq2pd.") ||
935 Name.startswith("avx512.mask.cvtudq2pd."))) {
Simon Pilgrim4298d062016-05-25 08:59:18 +0000936 // Lossless i32/float to double conversion.
937 // Extract the bottom elements if necessary and convert to double vector.
938 Value *Src = CI->getArgOperand(0);
939 VectorType *SrcTy = cast<VectorType>(Src->getType());
940 VectorType *DstTy = cast<VectorType>(CI->getType());
941 Rep = CI->getArgOperand(0);
942
943 unsigned NumDstElts = DstTy->getNumElements();
944 if (NumDstElts < SrcTy->getNumElements()) {
945 assert(NumDstElts == 2 && "Unexpected vector size");
Craig Topper99d1eab2016-06-12 00:41:19 +0000946 uint32_t ShuffleMask[2] = { 0, 1 };
947 Rep = Builder.CreateShuffleVector(Rep, UndefValue::get(SrcTy),
948 ShuffleMask);
Simon Pilgrim4298d062016-05-25 08:59:18 +0000949 }
950
Simon Pilgrimb57dd172016-11-16 14:48:32 +0000951 bool SInt2Double = (StringRef::npos != Name.find("cvtdq2"));
952 bool UInt2Double = (StringRef::npos != Name.find("cvtudq2"));
953 if (SInt2Double)
Simon Pilgrim4298d062016-05-25 08:59:18 +0000954 Rep = Builder.CreateSIToFP(Rep, DstTy, "cvtdq2pd");
Simon Pilgrimb57dd172016-11-16 14:48:32 +0000955 else if (UInt2Double)
956 Rep = Builder.CreateUIToFP(Rep, DstTy, "cvtudq2pd");
Simon Pilgrim4298d062016-05-25 08:59:18 +0000957 else
958 Rep = Builder.CreateFPExt(Rep, DstTy, "cvtps2pd");
Simon Pilgrimb57dd172016-11-16 14:48:32 +0000959
960 if (CI->getNumArgOperands() == 3)
961 Rep = EmitX86Select(Builder, CI->getArgOperand(2), Rep,
962 CI->getArgOperand(1));
Craig Toppera57d2ca2016-09-03 23:55:13 +0000963 } else if (IsX86 && (Name.startswith("avx512.mask.loadu."))) {
Craig Topper46b34fe2016-07-12 01:42:33 +0000964 Rep = UpgradeMaskedLoad(Builder, CI->getArgOperand(0),
Craig Topperf10fbfa2016-06-02 04:19:36 +0000965 CI->getArgOperand(1), CI->getArgOperand(2),
966 /*Aligned*/false);
Craig Toppera57d2ca2016-09-03 23:55:13 +0000967 } else if (IsX86 && (Name.startswith("avx512.mask.load."))) {
Craig Topper46b34fe2016-07-12 01:42:33 +0000968 Rep = UpgradeMaskedLoad(Builder, CI->getArgOperand(0),
Craig Topperf10fbfa2016-06-02 04:19:36 +0000969 CI->getArgOperand(1),CI->getArgOperand(2),
970 /*Aligned*/true);
Craig Topper5aebb862016-07-04 20:56:38 +0000971 } else if (IsX86 && Name.startswith("xop.vpcom")) {
Craig Topper3352ba52012-06-09 16:46:13 +0000972 Intrinsic::ID intID;
973 if (Name.endswith("ub"))
974 intID = Intrinsic::x86_xop_vpcomub;
975 else if (Name.endswith("uw"))
976 intID = Intrinsic::x86_xop_vpcomuw;
977 else if (Name.endswith("ud"))
978 intID = Intrinsic::x86_xop_vpcomud;
979 else if (Name.endswith("uq"))
980 intID = Intrinsic::x86_xop_vpcomuq;
981 else if (Name.endswith("b"))
982 intID = Intrinsic::x86_xop_vpcomb;
983 else if (Name.endswith("w"))
984 intID = Intrinsic::x86_xop_vpcomw;
985 else if (Name.endswith("d"))
986 intID = Intrinsic::x86_xop_vpcomd;
987 else if (Name.endswith("q"))
988 intID = Intrinsic::x86_xop_vpcomq;
989 else
990 llvm_unreachable("Unknown suffix");
991
Craig Topper5aebb862016-07-04 20:56:38 +0000992 Name = Name.substr(9); // strip off "xop.vpcom"
Craig Topper3352ba52012-06-09 16:46:13 +0000993 unsigned Imm;
994 if (Name.startswith("lt"))
995 Imm = 0;
996 else if (Name.startswith("le"))
997 Imm = 1;
998 else if (Name.startswith("gt"))
999 Imm = 2;
1000 else if (Name.startswith("ge"))
1001 Imm = 3;
1002 else if (Name.startswith("eq"))
1003 Imm = 4;
1004 else if (Name.startswith("ne"))
1005 Imm = 5;
Craig Topper3352ba52012-06-09 16:46:13 +00001006 else if (Name.startswith("false"))
Craig Toppere32546d2015-02-13 07:42:15 +00001007 Imm = 6;
1008 else if (Name.startswith("true"))
Craig Topper3352ba52012-06-09 16:46:13 +00001009 Imm = 7;
1010 else
1011 llvm_unreachable("Unknown condition");
1012
1013 Function *VPCOM = Intrinsic::getDeclaration(F->getParent(), intID);
David Blaikieff6409d2015-05-18 22:13:54 +00001014 Rep =
1015 Builder.CreateCall(VPCOM, {CI->getArgOperand(0), CI->getArgOperand(1),
1016 Builder.getInt8(Imm)});
Craig Topper5aebb862016-07-04 20:56:38 +00001017 } else if (IsX86 && Name == "xop.vpcmov") {
Simon Pilgrime88dc042015-11-03 20:27:01 +00001018 Value *Arg0 = CI->getArgOperand(0);
1019 Value *Arg1 = CI->getArgOperand(1);
1020 Value *Sel = CI->getArgOperand(2);
1021 unsigned NumElts = CI->getType()->getVectorNumElements();
1022 Constant *MinusOne = ConstantVector::getSplat(NumElts, Builder.getInt64(-1));
1023 Value *NotSel = Builder.CreateXor(Sel, MinusOne);
1024 Value *Sel0 = Builder.CreateAnd(Arg0, Sel);
1025 Value *Sel1 = Builder.CreateAnd(Arg1, NotSel);
1026 Rep = Builder.CreateOr(Sel0, Sel1);
Craig Topper5aebb862016-07-04 20:56:38 +00001027 } else if (IsX86 && Name == "sse42.crc32.64.8") {
Craig Topperef9e9932013-10-15 05:20:47 +00001028 Function *CRC32 = Intrinsic::getDeclaration(F->getParent(),
1029 Intrinsic::x86_sse42_crc32_32_8);
1030 Value *Trunc0 = Builder.CreateTrunc(CI->getArgOperand(0), Type::getInt32Ty(C));
David Blaikieff6409d2015-05-18 22:13:54 +00001031 Rep = Builder.CreateCall(CRC32, {Trunc0, CI->getArgOperand(1)});
Craig Topperef9e9932013-10-15 05:20:47 +00001032 Rep = Builder.CreateZExt(Rep, CI->getType(), "");
Simon Pilgrimea0d4f92016-07-22 13:58:44 +00001033 } else if (IsX86 && Name.startswith("avx.vbroadcast.s")) {
Adam Nemet39066802014-05-29 23:35:33 +00001034 // Replace broadcasts with a series of insertelements.
1035 Type *VecTy = CI->getType();
1036 Type *EltTy = VecTy->getVectorElementType();
1037 unsigned EltNum = VecTy->getVectorNumElements();
1038 Value *Cast = Builder.CreateBitCast(CI->getArgOperand(0),
1039 EltTy->getPointerTo());
David Blaikie0c28fd72015-05-20 21:46:30 +00001040 Value *Load = Builder.CreateLoad(EltTy, Cast);
Adam Nemet39066802014-05-29 23:35:33 +00001041 Type *I32Ty = Type::getInt32Ty(C);
1042 Rep = UndefValue::get(VecTy);
1043 for (unsigned I = 0; I < EltNum; ++I)
1044 Rep = Builder.CreateInsertElement(Rep, Load,
1045 ConstantInt::get(I32Ty, I));
Craig Topper5aebb862016-07-04 20:56:38 +00001046 } else if (IsX86 && (Name.startswith("sse41.pmovsx") ||
1047 Name.startswith("sse41.pmovzx") ||
1048 Name.startswith("avx2.pmovsx") ||
Craig Topperb110e042016-11-07 02:12:57 +00001049 Name.startswith("avx2.pmovzx") ||
1050 Name.startswith("avx512.mask.pmovsx") ||
1051 Name.startswith("avx512.mask.pmovzx"))) {
Simon Pilgrim9cb018b2015-09-23 08:48:33 +00001052 VectorType *SrcTy = cast<VectorType>(CI->getArgOperand(0)->getType());
1053 VectorType *DstTy = cast<VectorType>(CI->getType());
1054 unsigned NumDstElts = DstTy->getNumElements();
1055
Simon Pilgrim9602d672016-05-28 18:03:41 +00001056 // Extract a subvector of the first NumDstElts lanes and sign/zero extend.
Craig Topperc0a5fa02016-06-12 04:48:00 +00001057 SmallVector<uint32_t, 8> ShuffleMask(NumDstElts);
Craig Topper99d1eab2016-06-12 00:41:19 +00001058 for (unsigned i = 0; i != NumDstElts; ++i)
Craig Topperc0a5fa02016-06-12 04:48:00 +00001059 ShuffleMask[i] = i;
Simon Pilgrim9cb018b2015-09-23 08:48:33 +00001060
1061 Value *SV = Builder.CreateShuffleVector(
1062 CI->getArgOperand(0), UndefValue::get(SrcTy), ShuffleMask);
Simon Pilgrim9602d672016-05-28 18:03:41 +00001063
1064 bool DoSext = (StringRef::npos != Name.find("pmovsx"));
1065 Rep = DoSext ? Builder.CreateSExt(SV, DstTy)
1066 : Builder.CreateZExt(SV, DstTy);
Craig Topperb110e042016-11-07 02:12:57 +00001067 // If there are 3 arguments, it's a masked intrinsic so we need a select.
1068 if (CI->getNumArgOperands() == 3)
1069 Rep = EmitX86Select(Builder, CI->getArgOperand(2), Rep,
1070 CI->getArgOperand(1));
Simon Pilgrimea0d4f92016-07-22 13:58:44 +00001071 } else if (IsX86 && (Name.startswith("avx.vbroadcastf128") ||
1072 Name == "avx2.vbroadcasti128")) {
1073 // Replace vbroadcastf128/vbroadcasti128 with a vector load+shuffle.
1074 Type *EltTy = CI->getType()->getVectorElementType();
1075 unsigned NumSrcElts = 128 / EltTy->getPrimitiveSizeInBits();
1076 Type *VT = VectorType::get(EltTy, NumSrcElts);
David Blaikie0c28fd72015-05-20 21:46:30 +00001077 Value *Op = Builder.CreatePointerCast(CI->getArgOperand(0),
1078 PointerType::getUnqual(VT));
Chandler Carruth0215e762016-08-10 07:41:26 +00001079 Value *Load = Builder.CreateAlignedLoad(Op, 1);
Simon Pilgrimea0d4f92016-07-22 13:58:44 +00001080 if (NumSrcElts == 2)
1081 Rep = Builder.CreateShuffleVector(Load, UndefValue::get(Load->getType()),
1082 { 0, 1, 0, 1 });
1083 else
1084 Rep = Builder.CreateShuffleVector(Load, UndefValue::get(Load->getType()),
1085 { 0, 1, 2, 3, 0, 1, 2, 3 });
Craig Topper5aebb862016-07-04 20:56:38 +00001086 } else if (IsX86 && (Name.startswith("avx2.pbroadcast") ||
Simon Pilgrim4e96fbf2016-07-05 13:58:47 +00001087 Name.startswith("avx2.vbroadcast") ||
1088 Name.startswith("avx512.pbroadcast") ||
1089 Name.startswith("avx512.mask.broadcast.s"))) {
Ahmed Bougacha1a4987052015-08-20 20:36:19 +00001090 // Replace vp?broadcasts with a vector shuffle.
1091 Value *Op = CI->getArgOperand(0);
1092 unsigned NumElts = CI->getType()->getVectorNumElements();
1093 Type *MaskTy = VectorType::get(Type::getInt32Ty(C), NumElts);
1094 Rep = Builder.CreateShuffleVector(Op, UndefValue::get(Op->getType()),
1095 Constant::getNullValue(MaskTy));
Simon Pilgrim4e96fbf2016-07-05 13:58:47 +00001096
1097 if (CI->getNumArgOperands() == 3)
1098 Rep = EmitX86Select(Builder, CI->getArgOperand(2), Rep,
1099 CI->getArgOperand(1));
Craig Topper5aebb862016-07-04 20:56:38 +00001100 } else if (IsX86 && Name.startswith("avx512.mask.palignr.")) {
Craig Topperf57e17d2016-11-23 06:54:55 +00001101 Rep = UpgradeX86ALIGNIntrinsics(Builder, CI->getArgOperand(0),
1102 CI->getArgOperand(1),
1103 CI->getArgOperand(2),
1104 CI->getArgOperand(3),
1105 CI->getArgOperand(4),
1106 false);
1107 } else if (IsX86 && Name.startswith("avx512.mask.valign.")) {
1108 Rep = UpgradeX86ALIGNIntrinsics(Builder, CI->getArgOperand(0),
1109 CI->getArgOperand(1),
1110 CI->getArgOperand(2),
1111 CI->getArgOperand(3),
1112 CI->getArgOperand(4),
1113 true);
Craig Topper5aebb862016-07-04 20:56:38 +00001114 } else if (IsX86 && (Name == "sse2.psll.dq" ||
1115 Name == "avx2.psll.dq")) {
Craig Topper7355ac32016-05-29 06:37:33 +00001116 // 128/256-bit shift left specified in bits.
Craig Topperb324e432015-02-18 06:24:44 +00001117 unsigned Shift = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
Craig Topper46b34fe2016-07-12 01:42:33 +00001118 Rep = UpgradeX86PSLLDQIntrinsics(Builder, CI->getArgOperand(0),
Craig Topperb324e432015-02-18 06:24:44 +00001119 Shift / 8); // Shift is in bits.
Craig Topper5aebb862016-07-04 20:56:38 +00001120 } else if (IsX86 && (Name == "sse2.psrl.dq" ||
1121 Name == "avx2.psrl.dq")) {
Craig Topper7355ac32016-05-29 06:37:33 +00001122 // 128/256-bit shift right specified in bits.
Craig Topperb324e432015-02-18 06:24:44 +00001123 unsigned Shift = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
Craig Topper46b34fe2016-07-12 01:42:33 +00001124 Rep = UpgradeX86PSRLDQIntrinsics(Builder, CI->getArgOperand(0),
Craig Topperb324e432015-02-18 06:24:44 +00001125 Shift / 8); // Shift is in bits.
Craig Topper5aebb862016-07-04 20:56:38 +00001126 } else if (IsX86 && (Name == "sse2.psll.dq.bs" ||
1127 Name == "avx2.psll.dq.bs" ||
1128 Name == "avx512.psll.dq.512")) {
Simon Pilgrimf7186822016-06-09 21:09:03 +00001129 // 128/256/512-bit shift left specified in bytes.
Craig Topperb324e432015-02-18 06:24:44 +00001130 unsigned Shift = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
Craig Topper46b34fe2016-07-12 01:42:33 +00001131 Rep = UpgradeX86PSLLDQIntrinsics(Builder, CI->getArgOperand(0), Shift);
Craig Topper5aebb862016-07-04 20:56:38 +00001132 } else if (IsX86 && (Name == "sse2.psrl.dq.bs" ||
1133 Name == "avx2.psrl.dq.bs" ||
1134 Name == "avx512.psrl.dq.512")) {
Simon Pilgrimf7186822016-06-09 21:09:03 +00001135 // 128/256/512-bit shift right specified in bytes.
Craig Topperb324e432015-02-18 06:24:44 +00001136 unsigned Shift = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
Craig Topper46b34fe2016-07-12 01:42:33 +00001137 Rep = UpgradeX86PSRLDQIntrinsics(Builder, CI->getArgOperand(0), Shift);
Craig Topper5aebb862016-07-04 20:56:38 +00001138 } else if (IsX86 && (Name == "sse41.pblendw" ||
1139 Name.startswith("sse41.blendp") ||
1140 Name.startswith("avx.blend.p") ||
1141 Name == "avx2.pblendw" ||
1142 Name.startswith("avx2.pblendd."))) {
Craig Topper782d6202015-02-28 19:33:17 +00001143 Value *Op0 = CI->getArgOperand(0);
1144 Value *Op1 = CI->getArgOperand(1);
1145 unsigned Imm = cast <ConstantInt>(CI->getArgOperand(2))->getZExtValue();
1146 VectorType *VecTy = cast<VectorType>(CI->getType());
1147 unsigned NumElts = VecTy->getNumElements();
1148
Craig Topperc0a5fa02016-06-12 04:48:00 +00001149 SmallVector<uint32_t, 16> Idxs(NumElts);
1150 for (unsigned i = 0; i != NumElts; ++i)
1151 Idxs[i] = ((Imm >> (i%8)) & 1) ? i + NumElts : i;
Craig Topper782d6202015-02-28 19:33:17 +00001152
Craig Topper2f561822016-06-12 01:05:59 +00001153 Rep = Builder.CreateShuffleVector(Op0, Op1, Idxs);
Craig Topper5aebb862016-07-04 20:56:38 +00001154 } else if (IsX86 && (Name.startswith("avx.vinsertf128.") ||
1155 Name == "avx2.vinserti128")) {
Sanjay Patel19792fb2015-03-10 16:08:36 +00001156 Value *Op0 = CI->getArgOperand(0);
1157 Value *Op1 = CI->getArgOperand(1);
1158 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(2))->getZExtValue();
1159 VectorType *VecTy = cast<VectorType>(CI->getType());
1160 unsigned NumElts = VecTy->getNumElements();
Simon Pilgrim9cb018b2015-09-23 08:48:33 +00001161
Sanjay Patel19792fb2015-03-10 16:08:36 +00001162 // Mask off the high bits of the immediate value; hardware ignores those.
1163 Imm = Imm & 1;
Simon Pilgrim9cb018b2015-09-23 08:48:33 +00001164
Sanjay Patel19792fb2015-03-10 16:08:36 +00001165 // Extend the second operand into a vector that is twice as big.
1166 Value *UndefV = UndefValue::get(Op1->getType());
Craig Topperc0a5fa02016-06-12 04:48:00 +00001167 SmallVector<uint32_t, 8> Idxs(NumElts);
1168 for (unsigned i = 0; i != NumElts; ++i)
1169 Idxs[i] = i;
Craig Topper2f561822016-06-12 01:05:59 +00001170 Rep = Builder.CreateShuffleVector(Op1, UndefV, Idxs);
Sanjay Patel19792fb2015-03-10 16:08:36 +00001171
1172 // Insert the second operand into the first operand.
1173
1174 // Note that there is no guarantee that instruction lowering will actually
1175 // produce a vinsertf128 instruction for the created shuffles. In
1176 // particular, the 0 immediate case involves no lane changes, so it can
1177 // be handled as a blend.
1178
1179 // Example of shuffle mask for 32-bit elements:
1180 // Imm = 1 <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
1181 // Imm = 0 <i32 8, i32 9, i32 10, i32 11, i32 4, i32 5, i32 6, i32 7 >
1182
Sanjay Patel19792fb2015-03-10 16:08:36 +00001183 // The low half of the result is either the low half of the 1st operand
1184 // or the low half of the 2nd operand (the inserted vector).
Craig Topperc0a5fa02016-06-12 04:48:00 +00001185 for (unsigned i = 0; i != NumElts / 2; ++i)
1186 Idxs[i] = Imm ? i : (i + NumElts);
Sanjay Patel19792fb2015-03-10 16:08:36 +00001187 // The high half of the result is either the low half of the 2nd operand
1188 // (the inserted vector) or the high half of the 1st operand.
Craig Topperc0a5fa02016-06-12 04:48:00 +00001189 for (unsigned i = NumElts / 2; i != NumElts; ++i)
1190 Idxs[i] = Imm ? (i + NumElts / 2) : i;
Craig Topper2f561822016-06-12 01:05:59 +00001191 Rep = Builder.CreateShuffleVector(Op0, Rep, Idxs);
Craig Topper5aebb862016-07-04 20:56:38 +00001192 } else if (IsX86 && (Name.startswith("avx.vextractf128.") ||
1193 Name == "avx2.vextracti128")) {
Sanjay Patelaf1846c2015-03-12 15:15:19 +00001194 Value *Op0 = CI->getArgOperand(0);
1195 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
1196 VectorType *VecTy = cast<VectorType>(CI->getType());
1197 unsigned NumElts = VecTy->getNumElements();
Simon Pilgrim9cb018b2015-09-23 08:48:33 +00001198
Sanjay Patelaf1846c2015-03-12 15:15:19 +00001199 // Mask off the high bits of the immediate value; hardware ignores those.
1200 Imm = Imm & 1;
1201
1202 // Get indexes for either the high half or low half of the input vector.
Craig Topper2f561822016-06-12 01:05:59 +00001203 SmallVector<uint32_t, 4> Idxs(NumElts);
Sanjay Patelaf1846c2015-03-12 15:15:19 +00001204 for (unsigned i = 0; i != NumElts; ++i) {
Craig Topper2f561822016-06-12 01:05:59 +00001205 Idxs[i] = Imm ? (i + NumElts) : i;
Sanjay Patelaf1846c2015-03-12 15:15:19 +00001206 }
1207
1208 Value *UndefV = UndefValue::get(Op0->getType());
Craig Topper2f561822016-06-12 01:05:59 +00001209 Rep = Builder.CreateShuffleVector(Op0, UndefV, Idxs);
Craig Topper5aebb862016-07-04 20:56:38 +00001210 } else if (!IsX86 && Name == "stackprotectorcheck") {
Tim Shen00127562016-04-08 21:26:31 +00001211 Rep = nullptr;
Craig Topper5aebb862016-07-04 20:56:38 +00001212 } else if (IsX86 && (Name.startswith("avx512.mask.perm.df.") ||
1213 Name.startswith("avx512.mask.perm.di."))) {
Simon Pilgrim02d435d2016-07-04 14:19:05 +00001214 Value *Op0 = CI->getArgOperand(0);
1215 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
1216 VectorType *VecTy = cast<VectorType>(CI->getType());
1217 unsigned NumElts = VecTy->getNumElements();
1218
1219 SmallVector<uint32_t, 8> Idxs(NumElts);
1220 for (unsigned i = 0; i != NumElts; ++i)
1221 Idxs[i] = (i & ~0x3) + ((Imm >> (2 * (i & 0x3))) & 3);
1222
1223 Rep = Builder.CreateShuffleVector(Op0, Op0, Idxs);
1224
1225 if (CI->getNumArgOperands() == 4)
1226 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1227 CI->getArgOperand(2));
Craig Topper5aebb862016-07-04 20:56:38 +00001228 } else if (IsX86 && (Name.startswith("avx.vpermil.") ||
1229 Name == "sse2.pshuf.d" ||
1230 Name.startswith("avx512.mask.vpermil.p") ||
1231 Name.startswith("avx512.mask.pshuf.d."))) {
Craig Topper8a105052016-06-12 03:10:47 +00001232 Value *Op0 = CI->getArgOperand(0);
1233 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
1234 VectorType *VecTy = cast<VectorType>(CI->getType());
1235 unsigned NumElts = VecTy->getNumElements();
Simon Pilgrim9fca3002016-07-04 12:40:54 +00001236 // Calculate the size of each index in the immediate.
Craig Topper8a105052016-06-12 03:10:47 +00001237 unsigned IdxSize = 64 / VecTy->getScalarSizeInBits();
1238 unsigned IdxMask = ((1 << IdxSize) - 1);
1239
1240 SmallVector<uint32_t, 8> Idxs(NumElts);
1241 // Lookup the bits for this element, wrapping around the immediate every
1242 // 8-bits. Elements are grouped into sets of 2 or 4 elements so we need
1243 // to offset by the first index of each group.
1244 for (unsigned i = 0; i != NumElts; ++i)
1245 Idxs[i] = ((Imm >> ((i * IdxSize) % 8)) & IdxMask) | (i & ~IdxMask);
1246
1247 Rep = Builder.CreateShuffleVector(Op0, Op0, Idxs);
Craig Topper13cf7ca2016-06-13 02:36:48 +00001248
1249 if (CI->getNumArgOperands() == 4)
1250 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1251 CI->getArgOperand(2));
Craig Topper5aebb862016-07-04 20:56:38 +00001252 } else if (IsX86 && (Name == "sse2.pshufl.w" ||
1253 Name.startswith("avx512.mask.pshufl.w."))) {
Craig Topper10679862016-06-12 14:11:32 +00001254 Value *Op0 = CI->getArgOperand(0);
1255 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
1256 unsigned NumElts = CI->getType()->getVectorNumElements();
1257
1258 SmallVector<uint32_t, 16> Idxs(NumElts);
1259 for (unsigned l = 0; l != NumElts; l += 8) {
1260 for (unsigned i = 0; i != 4; ++i)
1261 Idxs[i + l] = ((Imm >> (2 * i)) & 0x3) + l;
1262 for (unsigned i = 4; i != 8; ++i)
1263 Idxs[i + l] = i + l;
1264 }
1265
1266 Rep = Builder.CreateShuffleVector(Op0, Op0, Idxs);
Craig Topper13cf7ca2016-06-13 02:36:48 +00001267
1268 if (CI->getNumArgOperands() == 4)
1269 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1270 CI->getArgOperand(2));
Craig Topper5aebb862016-07-04 20:56:38 +00001271 } else if (IsX86 && (Name == "sse2.pshufh.w" ||
1272 Name.startswith("avx512.mask.pshufh.w."))) {
Craig Topper10679862016-06-12 14:11:32 +00001273 Value *Op0 = CI->getArgOperand(0);
1274 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
1275 unsigned NumElts = CI->getType()->getVectorNumElements();
1276
1277 SmallVector<uint32_t, 16> Idxs(NumElts);
1278 for (unsigned l = 0; l != NumElts; l += 8) {
1279 for (unsigned i = 0; i != 4; ++i)
1280 Idxs[i + l] = i + l;
1281 for (unsigned i = 0; i != 4; ++i)
1282 Idxs[i + l + 4] = ((Imm >> (2 * i)) & 0x3) + 4 + l;
1283 }
1284
1285 Rep = Builder.CreateShuffleVector(Op0, Op0, Idxs);
Craig Topper13cf7ca2016-06-13 02:36:48 +00001286
1287 if (CI->getNumArgOperands() == 4)
1288 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1289 CI->getArgOperand(2));
Craig Topper4619c9e2016-09-13 07:40:53 +00001290 } else if (IsX86 && Name.startswith("avx512.mask.shuf.p")) {
1291 Value *Op0 = CI->getArgOperand(0);
1292 Value *Op1 = CI->getArgOperand(1);
1293 unsigned Imm = cast<ConstantInt>(CI->getArgOperand(2))->getZExtValue();
1294 unsigned NumElts = CI->getType()->getVectorNumElements();
1295
1296 unsigned NumLaneElts = 128/CI->getType()->getScalarSizeInBits();
1297 unsigned HalfLaneElts = NumLaneElts / 2;
1298
1299 SmallVector<uint32_t, 16> Idxs(NumElts);
1300 for (unsigned i = 0; i != NumElts; ++i) {
1301 // Base index is the starting element of the lane.
1302 Idxs[i] = i - (i % NumLaneElts);
1303 // If we are half way through the lane switch to the other source.
1304 if ((i % NumLaneElts) >= HalfLaneElts)
1305 Idxs[i] += NumElts;
1306 // Now select the specific element. By adding HalfLaneElts bits from
1307 // the immediate. Wrapping around the immediate every 8-bits.
1308 Idxs[i] += (Imm >> ((i * HalfLaneElts) % 8)) & ((1 << HalfLaneElts) - 1);
1309 }
1310
1311 Rep = Builder.CreateShuffleVector(Op0, Op1, Idxs);
1312
1313 Rep = EmitX86Select(Builder, CI->getArgOperand(4), Rep,
1314 CI->getArgOperand(3));
Craig Topper5aebb862016-07-04 20:56:38 +00001315 } else if (IsX86 && (Name.startswith("avx512.mask.movddup") ||
1316 Name.startswith("avx512.mask.movshdup") ||
1317 Name.startswith("avx512.mask.movsldup"))) {
Simon Pilgrim19adee92016-07-02 14:42:35 +00001318 Value *Op0 = CI->getArgOperand(0);
1319 unsigned NumElts = CI->getType()->getVectorNumElements();
1320 unsigned NumLaneElts = 128/CI->getType()->getScalarSizeInBits();
1321
1322 unsigned Offset = 0;
Craig Topper5aebb862016-07-04 20:56:38 +00001323 if (Name.startswith("avx512.mask.movshdup."))
Simon Pilgrim19adee92016-07-02 14:42:35 +00001324 Offset = 1;
1325
1326 SmallVector<uint32_t, 16> Idxs(NumElts);
1327 for (unsigned l = 0; l != NumElts; l += NumLaneElts)
1328 for (unsigned i = 0; i != NumLaneElts; i += 2) {
1329 Idxs[i + l + 0] = i + l + Offset;
1330 Idxs[i + l + 1] = i + l + Offset;
1331 }
1332
1333 Rep = Builder.CreateShuffleVector(Op0, Op0, Idxs);
1334
1335 Rep = EmitX86Select(Builder, CI->getArgOperand(2), Rep,
1336 CI->getArgOperand(1));
Craig Topper5aebb862016-07-04 20:56:38 +00001337 } else if (IsX86 && (Name.startswith("avx512.mask.punpckl") ||
1338 Name.startswith("avx512.mask.unpckl."))) {
Craig Topper597aa422016-06-23 07:37:33 +00001339 Value *Op0 = CI->getArgOperand(0);
1340 Value *Op1 = CI->getArgOperand(1);
1341 int NumElts = CI->getType()->getVectorNumElements();
1342 int NumLaneElts = 128/CI->getType()->getScalarSizeInBits();
1343
1344 SmallVector<uint32_t, 64> Idxs(NumElts);
1345 for (int l = 0; l != NumElts; l += NumLaneElts)
1346 for (int i = 0; i != NumLaneElts; ++i)
1347 Idxs[i + l] = l + (i / 2) + NumElts * (i % 2);
1348
1349 Rep = Builder.CreateShuffleVector(Op0, Op1, Idxs);
1350
1351 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1352 CI->getArgOperand(2));
Craig Topper5aebb862016-07-04 20:56:38 +00001353 } else if (IsX86 && (Name.startswith("avx512.mask.punpckh") ||
1354 Name.startswith("avx512.mask.unpckh."))) {
Craig Topper597aa422016-06-23 07:37:33 +00001355 Value *Op0 = CI->getArgOperand(0);
1356 Value *Op1 = CI->getArgOperand(1);
1357 int NumElts = CI->getType()->getVectorNumElements();
1358 int NumLaneElts = 128/CI->getType()->getScalarSizeInBits();
1359
1360 SmallVector<uint32_t, 64> Idxs(NumElts);
1361 for (int l = 0; l != NumElts; l += NumLaneElts)
1362 for (int i = 0; i != NumLaneElts; ++i)
1363 Idxs[i + l] = (NumLaneElts / 2) + l + (i / 2) + NumElts * (i % 2);
1364
1365 Rep = Builder.CreateShuffleVector(Op0, Op1, Idxs);
1366
1367 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1368 CI->getArgOperand(2));
Craig Toppera6e6feb2016-07-12 05:27:53 +00001369 } else if (IsX86 && Name.startswith("avx512.mask.pand.")) {
1370 Rep = Builder.CreateAnd(CI->getArgOperand(0), CI->getArgOperand(1));
1371 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1372 CI->getArgOperand(2));
1373 } else if (IsX86 && Name.startswith("avx512.mask.pandn.")) {
1374 Rep = Builder.CreateAnd(Builder.CreateNot(CI->getArgOperand(0)),
1375 CI->getArgOperand(1));
1376 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1377 CI->getArgOperand(2));
1378 } else if (IsX86 && Name.startswith("avx512.mask.por.")) {
1379 Rep = Builder.CreateOr(CI->getArgOperand(0), CI->getArgOperand(1));
1380 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1381 CI->getArgOperand(2));
1382 } else if (IsX86 && Name.startswith("avx512.mask.pxor.")) {
1383 Rep = Builder.CreateXor(CI->getArgOperand(0), CI->getArgOperand(1));
1384 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1385 CI->getArgOperand(2));
Craig Toppere75c4952016-09-02 05:29:17 +00001386 } else if (IsX86 && Name.startswith("avx512.mask.and.")) {
1387 VectorType *FTy = cast<VectorType>(CI->getType());
1388 VectorType *ITy = VectorType::getInteger(FTy);
1389 Rep = Builder.CreateAnd(Builder.CreateBitCast(CI->getArgOperand(0), ITy),
1390 Builder.CreateBitCast(CI->getArgOperand(1), ITy));
1391 Rep = Builder.CreateBitCast(Rep, FTy);
1392 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1393 CI->getArgOperand(2));
1394 } else if (IsX86 && Name.startswith("avx512.mask.andn.")) {
1395 VectorType *FTy = cast<VectorType>(CI->getType());
1396 VectorType *ITy = VectorType::getInteger(FTy);
1397 Rep = Builder.CreateNot(Builder.CreateBitCast(CI->getArgOperand(0), ITy));
1398 Rep = Builder.CreateAnd(Rep,
1399 Builder.CreateBitCast(CI->getArgOperand(1), ITy));
1400 Rep = Builder.CreateBitCast(Rep, FTy);
1401 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1402 CI->getArgOperand(2));
1403 } else if (IsX86 && Name.startswith("avx512.mask.or.")) {
1404 VectorType *FTy = cast<VectorType>(CI->getType());
1405 VectorType *ITy = VectorType::getInteger(FTy);
1406 Rep = Builder.CreateOr(Builder.CreateBitCast(CI->getArgOperand(0), ITy),
1407 Builder.CreateBitCast(CI->getArgOperand(1), ITy));
1408 Rep = Builder.CreateBitCast(Rep, FTy);
1409 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1410 CI->getArgOperand(2));
1411 } else if (IsX86 && Name.startswith("avx512.mask.xor.")) {
1412 VectorType *FTy = cast<VectorType>(CI->getType());
1413 VectorType *ITy = VectorType::getInteger(FTy);
1414 Rep = Builder.CreateXor(Builder.CreateBitCast(CI->getArgOperand(0), ITy),
1415 Builder.CreateBitCast(CI->getArgOperand(1), ITy));
1416 Rep = Builder.CreateBitCast(Rep, FTy);
1417 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1418 CI->getArgOperand(2));
Craig Topperaf0d63d2016-09-04 02:09:53 +00001419 } else if (IsX86 && Name.startswith("avx512.mask.padd.")) {
1420 Rep = Builder.CreateAdd(CI->getArgOperand(0), CI->getArgOperand(1));
1421 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1422 CI->getArgOperand(2));
1423 } else if (IsX86 && Name.startswith("avx512.mask.psub.")) {
1424 Rep = Builder.CreateSub(CI->getArgOperand(0), CI->getArgOperand(1));
1425 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1426 CI->getArgOperand(2));
1427 } else if (IsX86 && Name.startswith("avx512.mask.pmull.")) {
1428 Rep = Builder.CreateMul(CI->getArgOperand(0), CI->getArgOperand(1));
1429 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1430 CI->getArgOperand(2));
Craig Topper8f354a72016-11-05 05:35:23 +00001431 } else if (IsX86 && (Name.startswith("avx512.mask.add.p"))) {
Craig Topper41773452016-09-04 18:13:33 +00001432 Rep = Builder.CreateFAdd(CI->getArgOperand(0), CI->getArgOperand(1));
1433 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1434 CI->getArgOperand(2));
Craig Topper96041c62016-11-07 00:13:42 +00001435 } else if (IsX86 && Name.startswith("avx512.mask.div.p")) {
Craig Topper41773452016-09-04 18:13:33 +00001436 Rep = Builder.CreateFDiv(CI->getArgOperand(0), CI->getArgOperand(1));
1437 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1438 CI->getArgOperand(2));
Craig Topper96041c62016-11-07 00:13:42 +00001439 } else if (IsX86 && Name.startswith("avx512.mask.mul.p")) {
Craig Topper41773452016-09-04 18:13:33 +00001440 Rep = Builder.CreateFMul(CI->getArgOperand(0), CI->getArgOperand(1));
1441 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1442 CI->getArgOperand(2));
Craig Topper96041c62016-11-07 00:13:42 +00001443 } else if (IsX86 && Name.startswith("avx512.mask.sub.p")) {
Craig Topper41773452016-09-04 18:13:33 +00001444 Rep = Builder.CreateFSub(CI->getArgOperand(0), CI->getArgOperand(1));
1445 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1446 CI->getArgOperand(2));
Craig Topper7e545332016-11-07 00:13:39 +00001447 } else if (IsX86 && Name.startswith("avx512.mask.pshuf.b.")) {
1448 VectorType *VecTy = cast<VectorType>(CI->getType());
1449 Intrinsic::ID IID;
1450 if (VecTy->getPrimitiveSizeInBits() == 128)
1451 IID = Intrinsic::x86_ssse3_pshuf_b_128;
1452 else if (VecTy->getPrimitiveSizeInBits() == 256)
1453 IID = Intrinsic::x86_avx2_pshuf_b;
1454 else
1455 llvm_unreachable("Unexpected intrinsic");
1456
1457 Rep = Builder.CreateCall(Intrinsic::getDeclaration(F->getParent(), IID),
1458 { CI->getArgOperand(0), CI->getArgOperand(1) });
1459 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1460 CI->getArgOperand(2));
Craig Topper987dad22016-11-13 19:09:56 +00001461 } else if (IsX86 && Name.startswith("avx512.mask.psll")) {
1462 bool IsImmediate = Name[16] == 'i' ||
1463 (Name.size() > 18 && Name[18] == 'i');
1464 bool IsVariable = Name[16] == 'v';
1465 char Size = Name[16] == '.' ? Name[17] :
1466 Name[17] == '.' ? Name[18] :
Craig Topper02b5a1b2016-11-18 05:04:44 +00001467 Name[18] == '.' ? Name[19] :
1468 Name[20];
Craig Topper987dad22016-11-13 19:09:56 +00001469
1470 Intrinsic::ID IID;
Craig Topper353e59b2016-11-14 01:53:22 +00001471 if (IsVariable && Name[17] != '.') {
Craig Topper987dad22016-11-13 19:09:56 +00001472 if (Size == 'd' && Name[17] == '2') // avx512.mask.psllv2.di
1473 IID = Intrinsic::x86_avx2_psllv_q;
1474 else if (Size == 'd' && Name[17] == '4') // avx512.mask.psllv4.di
1475 IID = Intrinsic::x86_avx2_psllv_q_256;
1476 else if (Size == 's' && Name[17] == '4') // avx512.mask.psllv4.si
1477 IID = Intrinsic::x86_avx2_psllv_d;
1478 else if (Size == 's' && Name[17] == '8') // avx512.mask.psllv8.si
1479 IID = Intrinsic::x86_avx2_psllv_d_256;
Craig Topper02b5a1b2016-11-18 05:04:44 +00001480 else if (Size == 'h' && Name[17] == '8') // avx512.mask.psllv8.hi
1481 IID = Intrinsic::x86_avx512_psllv_w_128;
1482 else if (Size == 'h' && Name[17] == '1') // avx512.mask.psllv16.hi
1483 IID = Intrinsic::x86_avx512_psllv_w_256;
1484 else if (Name[17] == '3' && Name[18] == '2') // avx512.mask.psllv32hi
1485 IID = Intrinsic::x86_avx512_psllv_w_512;
Craig Topper987dad22016-11-13 19:09:56 +00001486 else
1487 llvm_unreachable("Unexpected size");
1488 } else if (Name.endswith(".128")) {
1489 if (Size == 'd') // avx512.mask.psll.d.128, avx512.mask.psll.di.128
1490 IID = IsImmediate ? Intrinsic::x86_sse2_pslli_d
1491 : Intrinsic::x86_sse2_psll_d;
1492 else if (Size == 'q') // avx512.mask.psll.q.128, avx512.mask.psll.qi.128
1493 IID = IsImmediate ? Intrinsic::x86_sse2_pslli_q
1494 : Intrinsic::x86_sse2_psll_q;
1495 else if (Size == 'w') // avx512.mask.psll.w.128, avx512.mask.psll.wi.128
1496 IID = IsImmediate ? Intrinsic::x86_sse2_pslli_w
1497 : Intrinsic::x86_sse2_psll_w;
1498 else
1499 llvm_unreachable("Unexpected size");
1500 } else if (Name.endswith(".256")) {
1501 if (Size == 'd') // avx512.mask.psll.d.256, avx512.mask.psll.di.256
1502 IID = IsImmediate ? Intrinsic::x86_avx2_pslli_d
1503 : Intrinsic::x86_avx2_psll_d;
1504 else if (Size == 'q') // avx512.mask.psll.q.256, avx512.mask.psll.qi.256
1505 IID = IsImmediate ? Intrinsic::x86_avx2_pslli_q
1506 : Intrinsic::x86_avx2_psll_q;
1507 else if (Size == 'w') // avx512.mask.psll.w.256, avx512.mask.psll.wi.256
1508 IID = IsImmediate ? Intrinsic::x86_avx2_pslli_w
1509 : Intrinsic::x86_avx2_psll_w;
1510 else
1511 llvm_unreachable("Unexpected size");
1512 } else {
Craig Topper353e59b2016-11-14 01:53:22 +00001513 if (Size == 'd') // psll.di.512, pslli.d, psll.d, psllv.d.512
1514 IID = IsImmediate ? Intrinsic::x86_avx512_pslli_d_512 :
1515 IsVariable ? Intrinsic::x86_avx512_psllv_d_512 :
1516 Intrinsic::x86_avx512_psll_d_512;
1517 else if (Size == 'q') // psll.qi.512, pslli.q, psll.q, psllv.q.512
1518 IID = IsImmediate ? Intrinsic::x86_avx512_pslli_q_512 :
1519 IsVariable ? Intrinsic::x86_avx512_psllv_q_512 :
1520 Intrinsic::x86_avx512_psll_q_512;
Craig Topper987dad22016-11-13 19:09:56 +00001521 else if (Size == 'w') // psll.wi.512, pslli.w, psll.w
1522 IID = IsImmediate ? Intrinsic::x86_avx512_pslli_w_512
1523 : Intrinsic::x86_avx512_psll_w_512;
1524 else
1525 llvm_unreachable("Unexpected size");
1526 }
1527
1528 Rep = UpgradeX86MaskedShift(Builder, *CI, IID);
1529 } else if (IsX86 && Name.startswith("avx512.mask.psrl")) {
1530 bool IsImmediate = Name[16] == 'i' ||
1531 (Name.size() > 18 && Name[18] == 'i');
1532 bool IsVariable = Name[16] == 'v';
1533 char Size = Name[16] == '.' ? Name[17] :
1534 Name[17] == '.' ? Name[18] :
Craig Topper02b5a1b2016-11-18 05:04:44 +00001535 Name[18] == '.' ? Name[19] :
1536 Name[20];
Craig Topper987dad22016-11-13 19:09:56 +00001537
1538 Intrinsic::ID IID;
Craig Topper353e59b2016-11-14 01:53:22 +00001539 if (IsVariable && Name[17] != '.') {
Craig Topper987dad22016-11-13 19:09:56 +00001540 if (Size == 'd' && Name[17] == '2') // avx512.mask.psrlv2.di
1541 IID = Intrinsic::x86_avx2_psrlv_q;
1542 else if (Size == 'd' && Name[17] == '4') // avx512.mask.psrlv4.di
1543 IID = Intrinsic::x86_avx2_psrlv_q_256;
1544 else if (Size == 's' && Name[17] == '4') // avx512.mask.psrlv4.si
1545 IID = Intrinsic::x86_avx2_psrlv_d;
1546 else if (Size == 's' && Name[17] == '8') // avx512.mask.psrlv8.si
1547 IID = Intrinsic::x86_avx2_psrlv_d_256;
Craig Topper02b5a1b2016-11-18 05:04:44 +00001548 else if (Size == 'h' && Name[17] == '8') // avx512.mask.psrlv8.hi
1549 IID = Intrinsic::x86_avx512_psrlv_w_128;
1550 else if (Size == 'h' && Name[17] == '1') // avx512.mask.psrlv16.hi
1551 IID = Intrinsic::x86_avx512_psrlv_w_256;
1552 else if (Name[17] == '3' && Name[18] == '2') // avx512.mask.psrlv32hi
1553 IID = Intrinsic::x86_avx512_psrlv_w_512;
Craig Topper987dad22016-11-13 19:09:56 +00001554 else
1555 llvm_unreachable("Unexpected size");
1556 } else if (Name.endswith(".128")) {
1557 if (Size == 'd') // avx512.mask.psrl.d.128, avx512.mask.psrl.di.128
1558 IID = IsImmediate ? Intrinsic::x86_sse2_psrli_d
1559 : Intrinsic::x86_sse2_psrl_d;
1560 else if (Size == 'q') // avx512.mask.psrl.q.128, avx512.mask.psrl.qi.128
1561 IID = IsImmediate ? Intrinsic::x86_sse2_psrli_q
1562 : Intrinsic::x86_sse2_psrl_q;
1563 else if (Size == 'w') // avx512.mask.psrl.w.128, avx512.mask.psrl.wi.128
1564 IID = IsImmediate ? Intrinsic::x86_sse2_psrli_w
1565 : Intrinsic::x86_sse2_psrl_w;
1566 else
1567 llvm_unreachable("Unexpected size");
1568 } else if (Name.endswith(".256")) {
1569 if (Size == 'd') // avx512.mask.psrl.d.256, avx512.mask.psrl.di.256
1570 IID = IsImmediate ? Intrinsic::x86_avx2_psrli_d
1571 : Intrinsic::x86_avx2_psrl_d;
1572 else if (Size == 'q') // avx512.mask.psrl.q.256, avx512.mask.psrl.qi.256
1573 IID = IsImmediate ? Intrinsic::x86_avx2_psrli_q
1574 : Intrinsic::x86_avx2_psrl_q;
1575 else if (Size == 'w') // avx512.mask.psrl.w.256, avx512.mask.psrl.wi.256
1576 IID = IsImmediate ? Intrinsic::x86_avx2_psrli_w
1577 : Intrinsic::x86_avx2_psrl_w;
1578 else
1579 llvm_unreachable("Unexpected size");
1580 } else {
Craig Topper353e59b2016-11-14 01:53:22 +00001581 if (Size == 'd') // psrl.di.512, psrli.d, psrl.d, psrl.d.512
1582 IID = IsImmediate ? Intrinsic::x86_avx512_psrli_d_512 :
1583 IsVariable ? Intrinsic::x86_avx512_psrlv_d_512 :
1584 Intrinsic::x86_avx512_psrl_d_512;
1585 else if (Size == 'q') // psrl.qi.512, psrli.q, psrl.q, psrl.q.512
1586 IID = IsImmediate ? Intrinsic::x86_avx512_psrli_q_512 :
1587 IsVariable ? Intrinsic::x86_avx512_psrlv_q_512 :
1588 Intrinsic::x86_avx512_psrl_q_512;
Craig Topper987dad22016-11-13 19:09:56 +00001589 else if (Size == 'w') // psrl.wi.512, psrli.w, psrl.w)
1590 IID = IsImmediate ? Intrinsic::x86_avx512_psrli_w_512
1591 : Intrinsic::x86_avx512_psrl_w_512;
1592 else
1593 llvm_unreachable("Unexpected size");
1594 }
1595
1596 Rep = UpgradeX86MaskedShift(Builder, *CI, IID);
1597 } else if (IsX86 && Name.startswith("avx512.mask.psra")) {
1598 bool IsImmediate = Name[16] == 'i' ||
1599 (Name.size() > 18 && Name[18] == 'i');
1600 bool IsVariable = Name[16] == 'v';
1601 char Size = Name[16] == '.' ? Name[17] :
1602 Name[17] == '.' ? Name[18] :
Craig Topper02b5a1b2016-11-18 05:04:44 +00001603 Name[18] == '.' ? Name[19] :
1604 Name[20];
Craig Topper987dad22016-11-13 19:09:56 +00001605
1606 Intrinsic::ID IID;
Craig Topper353e59b2016-11-14 01:53:22 +00001607 if (IsVariable && Name[17] != '.') {
Craig Topper987dad22016-11-13 19:09:56 +00001608 if (Size == 's' && Name[17] == '4') // avx512.mask.psrav4.si
1609 IID = Intrinsic::x86_avx2_psrav_d;
1610 else if (Size == 's' && Name[17] == '8') // avx512.mask.psrav8.si
1611 IID = Intrinsic::x86_avx2_psrav_d_256;
Craig Topper02b5a1b2016-11-18 05:04:44 +00001612 else if (Size == 'h' && Name[17] == '8') // avx512.mask.psrav8.hi
1613 IID = Intrinsic::x86_avx512_psrav_w_128;
1614 else if (Size == 'h' && Name[17] == '1') // avx512.mask.psrav16.hi
1615 IID = Intrinsic::x86_avx512_psrav_w_256;
1616 else if (Name[17] == '3' && Name[18] == '2') // avx512.mask.psrav32hi
1617 IID = Intrinsic::x86_avx512_psrav_w_512;
Craig Topper987dad22016-11-13 19:09:56 +00001618 else
1619 llvm_unreachable("Unexpected size");
1620 } else if (Name.endswith(".128")) {
1621 if (Size == 'd') // avx512.mask.psra.d.128, avx512.mask.psra.di.128
1622 IID = IsImmediate ? Intrinsic::x86_sse2_psrai_d
1623 : Intrinsic::x86_sse2_psra_d;
1624 else if (Size == 'q') // avx512.mask.psra.q.128, avx512.mask.psra.qi.128
Craig Topper353e59b2016-11-14 01:53:22 +00001625 IID = IsImmediate ? Intrinsic::x86_avx512_psrai_q_128 :
1626 IsVariable ? Intrinsic::x86_avx512_psrav_q_128 :
1627 Intrinsic::x86_avx512_psra_q_128;
Craig Topper987dad22016-11-13 19:09:56 +00001628 else if (Size == 'w') // avx512.mask.psra.w.128, avx512.mask.psra.wi.128
1629 IID = IsImmediate ? Intrinsic::x86_sse2_psrai_w
1630 : Intrinsic::x86_sse2_psra_w;
1631 else
1632 llvm_unreachable("Unexpected size");
1633 } else if (Name.endswith(".256")) {
1634 if (Size == 'd') // avx512.mask.psra.d.256, avx512.mask.psra.di.256
1635 IID = IsImmediate ? Intrinsic::x86_avx2_psrai_d
1636 : Intrinsic::x86_avx2_psra_d;
1637 else if (Size == 'q') // avx512.mask.psra.q.256, avx512.mask.psra.qi.256
Craig Topper353e59b2016-11-14 01:53:22 +00001638 IID = IsImmediate ? Intrinsic::x86_avx512_psrai_q_256 :
1639 IsVariable ? Intrinsic::x86_avx512_psrav_q_256 :
1640 Intrinsic::x86_avx512_psra_q_256;
Craig Topper987dad22016-11-13 19:09:56 +00001641 else if (Size == 'w') // avx512.mask.psra.w.256, avx512.mask.psra.wi.256
1642 IID = IsImmediate ? Intrinsic::x86_avx2_psrai_w
1643 : Intrinsic::x86_avx2_psra_w;
1644 else
1645 llvm_unreachable("Unexpected size");
1646 } else {
Craig Topper353e59b2016-11-14 01:53:22 +00001647 if (Size == 'd') // psra.di.512, psrai.d, psra.d, psrav.d.512
1648 IID = IsImmediate ? Intrinsic::x86_avx512_psrai_d_512 :
1649 IsVariable ? Intrinsic::x86_avx512_psrav_d_512 :
1650 Intrinsic::x86_avx512_psra_d_512;
Craig Topper987dad22016-11-13 19:09:56 +00001651 else if (Size == 'q') // psra.qi.512, psrai.q, psra.q
Craig Topper353e59b2016-11-14 01:53:22 +00001652 IID = IsImmediate ? Intrinsic::x86_avx512_psrai_q_512 :
1653 IsVariable ? Intrinsic::x86_avx512_psrav_q_512 :
1654 Intrinsic::x86_avx512_psra_q_512;
Craig Topper987dad22016-11-13 19:09:56 +00001655 else if (Size == 'w') // psra.wi.512, psrai.w, psra.w
1656 IID = IsImmediate ? Intrinsic::x86_avx512_psrai_w_512
1657 : Intrinsic::x86_avx512_psra_w_512;
1658 else
1659 llvm_unreachable("Unexpected size");
1660 }
1661
1662 Rep = UpgradeX86MaskedShift(Builder, *CI, IID);
Ayman Musa4d602432016-11-16 09:00:28 +00001663 } else if (IsX86 && Name.startswith("avx512.mask.move.s")) {
1664 Rep = upgradeMaskedMove(Builder, *CI);
Craig Topperabe7c5b2016-12-10 21:15:52 +00001665 } else if (IsX86 && Name.startswith("avx512.mask.vpermilvar.")) {
1666 Intrinsic::ID IID;
1667 if (Name.endswith("ps.128"))
1668 IID = Intrinsic::x86_avx_vpermilvar_ps;
1669 else if (Name.endswith("pd.128"))
1670 IID = Intrinsic::x86_avx_vpermilvar_pd;
1671 else if (Name.endswith("ps.256"))
1672 IID = Intrinsic::x86_avx_vpermilvar_ps_256;
1673 else if (Name.endswith("pd.256"))
1674 IID = Intrinsic::x86_avx_vpermilvar_pd_256;
1675 else
1676 llvm_unreachable("Unexpected vpermilvar intrinsic");
1677
1678 Function *Intrin = Intrinsic::getDeclaration(F->getParent(), IID);
1679 Rep = Builder.CreateCall(Intrin,
1680 { CI->getArgOperand(0), CI->getArgOperand(1) });
1681 Rep = EmitX86Select(Builder, CI->getArgOperand(3), Rep,
1682 CI->getArgOperand(2));
Craig Topper3b1817d2012-02-03 06:10:55 +00001683 } else {
Craig Topper8a105052016-06-12 03:10:47 +00001684 llvm_unreachable("Unknown function for CallInst upgrade.");
Craig Topper3b1817d2012-02-03 06:10:55 +00001685 }
1686
Tim Shen00127562016-04-08 21:26:31 +00001687 if (Rep)
1688 CI->replaceAllUsesWith(Rep);
Craig Topper3b1817d2012-02-03 06:10:55 +00001689 CI->eraseFromParent();
1690 return;
1691 }
1692
Yaron Kerend1fdbe72015-03-30 16:10:39 +00001693 std::string Name = CI->getName();
Adrian Prantl87b7eb92014-10-01 18:55:02 +00001694 if (!Name.empty())
1695 CI->setName(Name + ".old");
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001696
Chandler Carruth58a71ed2011-12-12 04:26:04 +00001697 switch (NewFn->getIntrinsicID()) {
1698 default:
Chris Lattner0bcbde42011-11-27 08:42:07 +00001699 llvm_unreachable("Unknown function for CallInst upgrade.");
Chandler Carruth58a71ed2011-12-12 04:26:04 +00001700
Jeroen Ketemaab99b592015-09-30 10:56:37 +00001701 case Intrinsic::arm_neon_vld1:
1702 case Intrinsic::arm_neon_vld2:
1703 case Intrinsic::arm_neon_vld3:
1704 case Intrinsic::arm_neon_vld4:
1705 case Intrinsic::arm_neon_vld2lane:
1706 case Intrinsic::arm_neon_vld3lane:
1707 case Intrinsic::arm_neon_vld4lane:
1708 case Intrinsic::arm_neon_vst1:
1709 case Intrinsic::arm_neon_vst2:
1710 case Intrinsic::arm_neon_vst3:
1711 case Intrinsic::arm_neon_vst4:
1712 case Intrinsic::arm_neon_vst2lane:
1713 case Intrinsic::arm_neon_vst3lane:
1714 case Intrinsic::arm_neon_vst4lane: {
1715 SmallVector<Value *, 4> Args(CI->arg_operands().begin(),
1716 CI->arg_operands().end());
1717 CI->replaceAllUsesWith(Builder.CreateCall(NewFn, Args));
1718 CI->eraseFromParent();
1719 return;
1720 }
1721
Chandler Carruth58a71ed2011-12-12 04:26:04 +00001722 case Intrinsic::ctlz:
Nuno Lopesad40c0a2012-05-22 15:25:31 +00001723 case Intrinsic::cttz:
Chandler Carruth58a71ed2011-12-12 04:26:04 +00001724 assert(CI->getNumArgOperands() == 1 &&
1725 "Mismatch between function args and call args");
David Blaikieff6409d2015-05-18 22:13:54 +00001726 CI->replaceAllUsesWith(Builder.CreateCall(
1727 NewFn, {CI->getArgOperand(0), Builder.getFalse()}, Name));
Chandler Carruth58a71ed2011-12-12 04:26:04 +00001728 CI->eraseFromParent();
1729 return;
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001730
Matt Arsenaultfbcbce42013-10-07 18:06:48 +00001731 case Intrinsic::objectsize:
David Blaikieff6409d2015-05-18 22:13:54 +00001732 CI->replaceAllUsesWith(Builder.CreateCall(
1733 NewFn, {CI->getArgOperand(0), CI->getArgOperand(1)}, Name));
Matt Arsenaultfbcbce42013-10-07 18:06:48 +00001734 CI->eraseFromParent();
1735 return;
1736
Joel Jonesb84f7be2012-07-18 00:02:16 +00001737 case Intrinsic::ctpop: {
David Blaikieff6409d2015-05-18 22:13:54 +00001738 CI->replaceAllUsesWith(Builder.CreateCall(NewFn, {CI->getArgOperand(0)}));
Joel Jonesb84f7be2012-07-18 00:02:16 +00001739 CI->eraseFromParent();
1740 return;
1741 }
Joel Jones43cb8782012-07-13 23:25:25 +00001742
Craig Topper71dc02d2012-06-13 07:18:53 +00001743 case Intrinsic::x86_xop_vfrcz_ss:
1744 case Intrinsic::x86_xop_vfrcz_sd:
David Blaikieff6409d2015-05-18 22:13:54 +00001745 CI->replaceAllUsesWith(
1746 Builder.CreateCall(NewFn, {CI->getArgOperand(1)}, Name));
Craig Topper71dc02d2012-06-13 07:18:53 +00001747 CI->eraseFromParent();
1748 return;
1749
Simon Pilgrime85506b2016-06-03 08:06:03 +00001750 case Intrinsic::x86_xop_vpermil2pd:
1751 case Intrinsic::x86_xop_vpermil2ps:
1752 case Intrinsic::x86_xop_vpermil2pd_256:
1753 case Intrinsic::x86_xop_vpermil2ps_256: {
1754 SmallVector<Value *, 4> Args(CI->arg_operands().begin(),
1755 CI->arg_operands().end());
1756 VectorType *FltIdxTy = cast<VectorType>(Args[2]->getType());
1757 VectorType *IntIdxTy = VectorType::getInteger(FltIdxTy);
1758 Args[2] = Builder.CreateBitCast(Args[2], IntIdxTy);
1759 CI->replaceAllUsesWith(Builder.CreateCall(NewFn, Args, Name));
1760 CI->eraseFromParent();
1761 return;
1762 }
1763
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001764 case Intrinsic::x86_sse41_ptestc:
1765 case Intrinsic::x86_sse41_ptestz:
Craig Topper71dc02d2012-06-13 07:18:53 +00001766 case Intrinsic::x86_sse41_ptestnzc: {
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001767 // The arguments for these intrinsics used to be v4f32, and changed
1768 // to v2i64. This is purely a nop, since those are bitwise intrinsics.
1769 // So, the only thing required is a bitcast for both arguments.
1770 // First, check the arguments have the old type.
1771 Value *Arg0 = CI->getArgOperand(0);
1772 if (Arg0->getType() != VectorType::get(Type::getFloatTy(C), 4))
1773 return;
1774
1775 // Old intrinsic, add bitcasts
1776 Value *Arg1 = CI->getArgOperand(1);
1777
David Blaikie5bacf372015-04-24 21:16:07 +00001778 Type *NewVecTy = VectorType::get(Type::getInt64Ty(C), 2);
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001779
David Blaikie5bacf372015-04-24 21:16:07 +00001780 Value *BC0 = Builder.CreateBitCast(Arg0, NewVecTy, "cast");
1781 Value *BC1 = Builder.CreateBitCast(Arg1, NewVecTy, "cast");
1782
David Blaikieff6409d2015-05-18 22:13:54 +00001783 CallInst *NewCall = Builder.CreateCall(NewFn, {BC0, BC1}, Name);
Nadav Rotem17ee58a2012-06-10 18:42:51 +00001784 CI->replaceAllUsesWith(NewCall);
1785 CI->eraseFromParent();
1786 return;
Evan Cheng0e179d02007-12-17 22:33:23 +00001787 }
Chandler Carruth373b2b12014-09-06 10:00:01 +00001788
Chandler Carruth373b2b12014-09-06 10:00:01 +00001789 case Intrinsic::x86_sse41_insertps:
1790 case Intrinsic::x86_sse41_dppd:
1791 case Intrinsic::x86_sse41_dpps:
1792 case Intrinsic::x86_sse41_mpsadbw:
Chandler Carruth373b2b12014-09-06 10:00:01 +00001793 case Intrinsic::x86_avx_dp_ps_256:
Chandler Carruth373b2b12014-09-06 10:00:01 +00001794 case Intrinsic::x86_avx2_mpsadbw: {
1795 // Need to truncate the last argument from i32 to i8 -- this argument models
1796 // an inherently 8-bit immediate operand to these x86 instructions.
1797 SmallVector<Value *, 4> Args(CI->arg_operands().begin(),
1798 CI->arg_operands().end());
1799
1800 // Replace the last argument with a trunc.
1801 Args.back() = Builder.CreateTrunc(Args.back(), Type::getInt8Ty(C), "trunc");
1802
1803 CallInst *NewCall = Builder.CreateCall(NewFn, Args);
1804 CI->replaceAllUsesWith(NewCall);
1805 CI->eraseFromParent();
1806 return;
1807 }
Marcin Koscielnicki3fdc2572016-04-19 20:51:05 +00001808
1809 case Intrinsic::thread_pointer: {
1810 CI->replaceAllUsesWith(Builder.CreateCall(NewFn, {}));
1811 CI->eraseFromParent();
1812 return;
1813 }
Artur Pilipenko7ad95ec2016-06-28 18:27:25 +00001814
Mehdi Amini8c629ec2016-08-13 23:31:24 +00001815 case Intrinsic::invariant_start:
1816 case Intrinsic::invariant_end:
Artur Pilipenko7ad95ec2016-06-28 18:27:25 +00001817 case Intrinsic::masked_load:
1818 case Intrinsic::masked_store: {
1819 SmallVector<Value *, 4> Args(CI->arg_operands().begin(),
1820 CI->arg_operands().end());
1821 CI->replaceAllUsesWith(Builder.CreateCall(NewFn, Args));
1822 CI->eraseFromParent();
1823 return;
1824 }
Craig Topper71dc02d2012-06-13 07:18:53 +00001825 }
Chandler Carruth7132e002007-08-04 01:51:18 +00001826}
1827
Sanjay Patelfdf0d5f2016-04-18 19:11:57 +00001828void llvm::UpgradeCallsToIntrinsic(Function *F) {
Chandler Carruth7132e002007-08-04 01:51:18 +00001829 assert(F && "Illegal attempt to upgrade a non-existent intrinsic.");
1830
Sanjay Patelfdf0d5f2016-04-18 19:11:57 +00001831 // Check if this function should be upgraded and get the replacement function
1832 // if there is one.
Chris Lattner80ed9dc2011-06-18 06:05:24 +00001833 Function *NewFn;
Evan Cheng0e179d02007-12-17 22:33:23 +00001834 if (UpgradeIntrinsicFunction(F, NewFn)) {
Sanjay Patelfdf0d5f2016-04-18 19:11:57 +00001835 // Replace all users of the old function with the new function or new
1836 // instructions. This is not a range loop because the call is deleted.
1837 for (auto UI = F->user_begin(), UE = F->user_end(); UI != UE; )
Duncan P. N. Exon Smith93f53c42016-04-17 03:59:37 +00001838 if (CallInst *CI = dyn_cast<CallInst>(*UI++))
Filipe Cabecinhas0011c582015-07-03 20:12:01 +00001839 UpgradeIntrinsicCall(CI, NewFn);
Sanjay Patelfdf0d5f2016-04-18 19:11:57 +00001840
Filipe Cabecinhas0011c582015-07-03 20:12:01 +00001841 // Remove old function, no longer used, from the module.
1842 F->eraseFromParent();
Chandler Carruth7132e002007-08-04 01:51:18 +00001843 }
1844}
Devang Patel80ae3492009-08-28 23:24:31 +00001845
Mehdi Aminie4709272016-09-14 22:29:59 +00001846MDNode *llvm::UpgradeTBAANode(MDNode &MD) {
Manman Ren209b17c2013-09-28 00:22:27 +00001847 // Check if the tag uses struct-path aware TBAA format.
Mehdi Aminie4709272016-09-14 22:29:59 +00001848 if (isa<MDNode>(MD.getOperand(0)) && MD.getNumOperands() >= 3)
1849 return &MD;
Manman Ren209b17c2013-09-28 00:22:27 +00001850
Mehdi Aminie4709272016-09-14 22:29:59 +00001851 auto &Context = MD.getContext();
1852 if (MD.getNumOperands() == 3) {
1853 Metadata *Elts[] = {MD.getOperand(0), MD.getOperand(1)};
1854 MDNode *ScalarType = MDNode::get(Context, Elts);
Manman Ren209b17c2013-09-28 00:22:27 +00001855 // Create a MDNode <ScalarType, ScalarType, offset 0, const>
Duncan P. N. Exon Smith5bf8fef2014-12-09 18:38:53 +00001856 Metadata *Elts2[] = {ScalarType, ScalarType,
Mehdi Aminie4709272016-09-14 22:29:59 +00001857 ConstantAsMetadata::get(
1858 Constant::getNullValue(Type::getInt64Ty(Context))),
1859 MD.getOperand(2)};
1860 return MDNode::get(Context, Elts2);
Manman Ren209b17c2013-09-28 00:22:27 +00001861 }
Mehdi Aminie4709272016-09-14 22:29:59 +00001862 // Create a MDNode <MD, MD, offset 0>
1863 Metadata *Elts[] = {&MD, &MD, ConstantAsMetadata::get(Constant::getNullValue(
1864 Type::getInt64Ty(Context)))};
1865 return MDNode::get(Context, Elts);
Manman Ren209b17c2013-09-28 00:22:27 +00001866}
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001867
1868Instruction *llvm::UpgradeBitCastInst(unsigned Opc, Value *V, Type *DestTy,
1869 Instruction *&Temp) {
1870 if (Opc != Instruction::BitCast)
Craig Topperc6207612014-04-09 06:08:46 +00001871 return nullptr;
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001872
Craig Topperc6207612014-04-09 06:08:46 +00001873 Temp = nullptr;
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001874 Type *SrcTy = V->getType();
1875 if (SrcTy->isPtrOrPtrVectorTy() && DestTy->isPtrOrPtrVectorTy() &&
1876 SrcTy->getPointerAddressSpace() != DestTy->getPointerAddressSpace()) {
1877 LLVMContext &Context = V->getContext();
1878
1879 // We have no information about target data layout, so we assume that
1880 // the maximum pointer size is 64bit.
1881 Type *MidTy = Type::getInt64Ty(Context);
1882 Temp = CastInst::Create(Instruction::PtrToInt, V, MidTy);
1883
1884 return CastInst::Create(Instruction::IntToPtr, Temp, DestTy);
1885 }
1886
Craig Topperc6207612014-04-09 06:08:46 +00001887 return nullptr;
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001888}
1889
1890Value *llvm::UpgradeBitCastExpr(unsigned Opc, Constant *C, Type *DestTy) {
1891 if (Opc != Instruction::BitCast)
Craig Topperc6207612014-04-09 06:08:46 +00001892 return nullptr;
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001893
1894 Type *SrcTy = C->getType();
1895 if (SrcTy->isPtrOrPtrVectorTy() && DestTy->isPtrOrPtrVectorTy() &&
1896 SrcTy->getPointerAddressSpace() != DestTy->getPointerAddressSpace()) {
1897 LLVMContext &Context = C->getContext();
1898
1899 // We have no information about target data layout, so we assume that
1900 // the maximum pointer size is 64bit.
1901 Type *MidTy = Type::getInt64Ty(Context);
1902
1903 return ConstantExpr::getIntToPtr(ConstantExpr::getPtrToInt(C, MidTy),
1904 DestTy);
1905 }
1906
Craig Topperc6207612014-04-09 06:08:46 +00001907 return nullptr;
Matt Arsenaultb03bd4d2013-11-15 01:34:59 +00001908}
Manman Ren8b4306c2013-12-02 21:29:56 +00001909
1910/// Check the debug info version number, if it is out-dated, drop the debug
1911/// info. Return true if module is modified.
1912bool llvm::UpgradeDebugInfo(Module &M) {
Manman Ren2ebfb422014-01-16 01:51:12 +00001913 unsigned Version = getDebugMetadataVersionFromModule(M);
1914 if (Version == DEBUG_METADATA_VERSION)
Manman Ren8b4306c2013-12-02 21:29:56 +00001915 return false;
1916
Manman Ren2ebfb422014-01-16 01:51:12 +00001917 bool RetCode = StripDebugInfo(M);
1918 if (RetCode) {
1919 DiagnosticInfoDebugMetadataVersion DiagVersion(M, Version);
1920 M.getContext().diagnose(DiagVersion);
1921 }
1922 return RetCode;
Manman Ren8b4306c2013-12-02 21:29:56 +00001923}
Eli Bendersky5d5e18d2014-06-25 15:41:00 +00001924
Manman Renb5d7ff42016-05-25 23:14:48 +00001925bool llvm::UpgradeModuleFlags(Module &M) {
1926 const NamedMDNode *ModFlags = M.getModuleFlagsMetadata();
1927 if (!ModFlags)
1928 return false;
1929
1930 bool HasObjCFlag = false, HasClassProperties = false;
1931 for (unsigned I = 0, E = ModFlags->getNumOperands(); I != E; ++I) {
1932 MDNode *Op = ModFlags->getOperand(I);
1933 if (Op->getNumOperands() < 2)
1934 continue;
1935 MDString *ID = dyn_cast_or_null<MDString>(Op->getOperand(1));
1936 if (!ID)
1937 continue;
1938 if (ID->getString() == "Objective-C Image Info Version")
1939 HasObjCFlag = true;
1940 if (ID->getString() == "Objective-C Class Properties")
1941 HasClassProperties = true;
1942 }
1943 // "Objective-C Class Properties" is recently added for Objective-C. We
1944 // upgrade ObjC bitcodes to contain a "Objective-C Class Properties" module
Mehdi Aminib53b62e2016-09-16 00:38:18 +00001945 // flag of value 0, so we can correclty downgrade this flag when trying to
1946 // link an ObjC bitcode without this module flag with an ObjC bitcode with
1947 // this module flag.
Manman Renb5d7ff42016-05-25 23:14:48 +00001948 if (HasObjCFlag && !HasClassProperties) {
Mehdi Aminib53b62e2016-09-16 00:38:18 +00001949 M.addModuleFlag(llvm::Module::Override, "Objective-C Class Properties",
Manman Renb5d7ff42016-05-25 23:14:48 +00001950 (uint32_t)0);
1951 return true;
1952 }
1953 return false;
1954}
1955
Duncan P. N. Exon Smithefe16c82016-03-25 00:56:13 +00001956static bool isOldLoopArgument(Metadata *MD) {
1957 auto *T = dyn_cast_or_null<MDTuple>(MD);
1958 if (!T)
1959 return false;
1960 if (T->getNumOperands() < 1)
1961 return false;
1962 auto *S = dyn_cast_or_null<MDString>(T->getOperand(0));
1963 if (!S)
1964 return false;
1965 return S->getString().startswith("llvm.vectorizer.");
1966}
1967
1968static MDString *upgradeLoopTag(LLVMContext &C, StringRef OldTag) {
1969 StringRef OldPrefix = "llvm.vectorizer.";
1970 assert(OldTag.startswith(OldPrefix) && "Expected old prefix");
1971
1972 if (OldTag == "llvm.vectorizer.unroll")
1973 return MDString::get(C, "llvm.loop.interleave.count");
1974
1975 return MDString::get(
1976 C, (Twine("llvm.loop.vectorize.") + OldTag.drop_front(OldPrefix.size()))
1977 .str());
1978}
1979
1980static Metadata *upgradeLoopArgument(Metadata *MD) {
1981 auto *T = dyn_cast_or_null<MDTuple>(MD);
1982 if (!T)
1983 return MD;
1984 if (T->getNumOperands() < 1)
1985 return MD;
1986 auto *OldTag = dyn_cast_or_null<MDString>(T->getOperand(0));
1987 if (!OldTag)
1988 return MD;
1989 if (!OldTag->getString().startswith("llvm.vectorizer."))
1990 return MD;
1991
1992 // This has an old tag. Upgrade it.
1993 SmallVector<Metadata *, 8> Ops;
1994 Ops.reserve(T->getNumOperands());
1995 Ops.push_back(upgradeLoopTag(T->getContext(), OldTag->getString()));
1996 for (unsigned I = 1, E = T->getNumOperands(); I != E; ++I)
1997 Ops.push_back(T->getOperand(I));
1998
1999 return MDTuple::get(T->getContext(), Ops);
2000}
2001
2002MDNode *llvm::upgradeInstructionLoopAttachment(MDNode &N) {
2003 auto *T = dyn_cast<MDTuple>(&N);
2004 if (!T)
2005 return &N;
2006
David Majnemer0a16c222016-08-11 21:15:00 +00002007 if (none_of(T->operands(), isOldLoopArgument))
Duncan P. N. Exon Smithefe16c82016-03-25 00:56:13 +00002008 return &N;
2009
2010 SmallVector<Metadata *, 8> Ops;
2011 Ops.reserve(T->getNumOperands());
2012 for (Metadata *MD : T->operands())
2013 Ops.push_back(upgradeLoopArgument(MD));
2014
2015 return MDTuple::get(T->getContext(), Ops);
Eli Bendersky5d5e18d2014-06-25 15:41:00 +00002016}