2007-08-04 01:51:18 +00:00
|
|
|
//===-- AutoUpgrade.cpp - Implement auto-upgrade helper functions ---------===//
|
|
|
|
//
|
|
|
|
// The LLVM Compiler Infrastructure
|
|
|
|
//
|
2007-12-29 20:36:04 +00:00
|
|
|
// This file is distributed under the University of Illinois Open Source
|
|
|
|
// License. See LICENSE.TXT for details.
|
2007-08-04 01:51:18 +00:00
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
//
|
2013-07-20 17:46:00 +00:00
|
|
|
// This file implements the auto-upgrade helper functions
|
2007-08-04 01:51:18 +00:00
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
2014-03-05 10:34:14 +00:00
|
|
|
#include "llvm/IR/AutoUpgrade.h"
|
2014-03-04 11:45:46 +00:00
|
|
|
#include "llvm/IR/CFG.h"
|
2014-03-04 11:01:28 +00:00
|
|
|
#include "llvm/IR/CallSite.h"
|
2013-01-02 11:36:10 +00:00
|
|
|
#include "llvm/IR/Constants.h"
|
2014-03-06 00:46:21 +00:00
|
|
|
#include "llvm/IR/DebugInfo.h"
|
2014-01-16 01:51:12 +00:00
|
|
|
#include "llvm/IR/DiagnosticInfo.h"
|
Move the complex address expression out of DIVariable and into an extra
argument of the llvm.dbg.declare/llvm.dbg.value intrinsics.
Previously, DIVariable was a variable-length field that has an optional
reference to a Metadata array consisting of a variable number of
complex address expressions. In the case of OpPiece expressions this is
wasting a lot of storage in IR, because when an aggregate type is, e.g.,
SROA'd into all of its n individual members, the IR will contain n copies
of the DIVariable, all alike, only differing in the complex address
reference at the end.
By making the complex address into an extra argument of the
dbg.value/dbg.declare intrinsics, all of the pieces can reference the
same variable and the complex address expressions can be uniqued across
the CU, too.
Down the road, this will allow us to move other flags, such as
"indirection" out of the DIVariable, too.
The new intrinsics look like this:
declare void @llvm.dbg.declare(metadata %storage, metadata %var, metadata %expr)
declare void @llvm.dbg.value(metadata %storage, i64 %offset, metadata %var, metadata %expr)
This patch adds a new LLVM-local tag to DIExpressions, so we can detect
and pretty-print DIExpression metadata nodes.
What this patch doesn't do:
This patch does not touch the "Indirect" field in DIVariable; but moving
that into the expression would be a natural next step.
http://reviews.llvm.org/D4919
rdar://problem/17994491
Thanks to dblaikie and dexonsmith for reviewing this patch!
Note: I accidentally committed a bogus older version of this patch previously.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218787 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-01 18:55:02 +00:00
|
|
|
#include "llvm/IR/DIBuilder.h"
|
2013-01-02 11:36:10 +00:00
|
|
|
#include "llvm/IR/Function.h"
|
|
|
|
#include "llvm/IR/IRBuilder.h"
|
|
|
|
#include "llvm/IR/Instruction.h"
|
|
|
|
#include "llvm/IR/IntrinsicInst.h"
|
|
|
|
#include "llvm/IR/LLVMContext.h"
|
|
|
|
#include "llvm/IR/Module.h"
|
2009-07-11 20:10:48 +00:00
|
|
|
#include "llvm/Support/ErrorHandling.h"
|
2008-02-20 11:08:44 +00:00
|
|
|
#include <cstring>
|
2007-08-04 01:51:18 +00:00
|
|
|
using namespace llvm;
|
|
|
|
|
2012-06-10 18:42:51 +00:00
|
|
|
// Upgrade the declarations of the SSE4.1 functions whose arguments have
|
|
|
|
// changed their type from v4f32 to v2i64.
|
|
|
|
static bool UpgradeSSE41Function(Function* F, Intrinsic::ID IID,
|
|
|
|
Function *&NewFn) {
|
|
|
|
// Check whether this is an old version of the function, which received
|
|
|
|
// v4f32 arguments.
|
|
|
|
Type *Arg0Type = F->getFunctionType()->getParamType(0);
|
|
|
|
if (Arg0Type != VectorType::get(Type::getFloatTy(F->getContext()), 4))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
// Yes, it's old, replace it with new version.
|
|
|
|
F->setName(F->getName() + ".old");
|
|
|
|
NewFn = Intrinsic::getDeclaration(F->getParent(), IID);
|
|
|
|
return true;
|
|
|
|
}
|
2007-08-04 01:51:18 +00:00
|
|
|
|
[x86] Fix a pretty horrible bug and inconsistency in the x86 asm
parsing (and latent bug in the instruction definitions).
This is effectively a revert of r136287 which tried to address
a specific and narrow case of immediate operands failing to be accepted
by x86 instructions with a pretty heavy hammer: it introduced a new kind
of operand that behaved differently. All of that is removed with this
commit, but the test cases are both preserved and enhanced.
The core problem that r136287 and this commit are trying to handle is
that gas accepts both of the following instructions:
insertps $192, %xmm0, %xmm1
insertps $-64, %xmm0, %xmm1
These will encode to the same byte sequence, with the immediate
occupying an 8-bit entry. The first form was fixed by r136287 but that
broke the prior handling of the second form! =[ Ironically, we would
still emit the second form in some cases and then be unable to
re-assemble the output.
The reason why the first instruction failed to be handled is because
prior to r136287 the operands ere marked 'i32i8imm' which forces them to
be sign-extenable. Clearly, that won't work for 192 in a single byte.
However, making thim zero-extended or "unsigned" doesn't really address
the core issue either because it breaks negative immediates. The correct
fix is to make these operands 'i8imm' reflecting that they can be either
signed or unsigned but must be 8-bit immediates. This patch backs out
r136287 and then changes those places as well as some others to use
'i8imm' rather than one of the extended variants.
Naturally, this broke something else. The custom DAG nodes had to be
updated to have a much more accurate type constraint of an i8 node, and
a bunch of Pat immediates needed to be specified as i8 values.
The fallout didn't end there though. We also then ceased to be able to
match the instruction-specific intrinsics to the instructions so
modified. Digging, this is because they too used i32 rather than i8 in
their signature. So I've also switched those intrinsics to i8 arguments
in line with the instructions.
In order to make the intrinsic adjustments of course, I also had to add
auto upgrading for the intrinsics.
I suspect that the intrinsic argument types may have led everything down
this rabbit hole. Pretty happy with the result.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@217310 91177308-0d34-0410-b5e6-96231b3b80d8
2014-09-06 10:00:01 +00:00
|
|
|
// Upgrade the declarations of intrinsic functions whose 8-bit immediate mask
|
|
|
|
// arguments have changed their type from i32 to i8.
|
|
|
|
static bool UpgradeX86IntrinsicsWith8BitMask(Function *F, Intrinsic::ID IID,
|
|
|
|
Function *&NewFn) {
|
|
|
|
// Check that the last argument is an i32.
|
|
|
|
Type *LastArgType = F->getFunctionType()->getParamType(
|
|
|
|
F->getFunctionType()->getNumParams() - 1);
|
|
|
|
if (!LastArgType->isIntegerTy(32))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
// Move this function aside and map down.
|
|
|
|
F->setName(F->getName() + ".old");
|
|
|
|
NewFn = Intrinsic::getDeclaration(F->getParent(), IID);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2007-12-17 22:33:23 +00:00
|
|
|
static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
|
2007-08-04 01:51:18 +00:00
|
|
|
assert(F && "Illegal to upgrade a non-existent Function.");
|
|
|
|
|
|
|
|
// Quickly eliminate it, if it's not a candidate.
|
2011-06-18 18:56:39 +00:00
|
|
|
StringRef Name = F->getName();
|
|
|
|
if (Name.size() <= 8 || !Name.startswith("llvm."))
|
2007-12-17 22:33:23 +00:00
|
|
|
return false;
|
2011-06-18 18:56:39 +00:00
|
|
|
Name = Name.substr(5); // Strip off "llvm."
|
2011-11-27 08:42:07 +00:00
|
|
|
|
2011-06-18 18:56:39 +00:00
|
|
|
switch (Name[0]) {
|
2007-08-04 01:51:18 +00:00
|
|
|
default: break;
|
2012-07-13 23:25:25 +00:00
|
|
|
case 'a': {
|
|
|
|
if (Name.startswith("arm.neon.vclz")) {
|
|
|
|
Type* args[2] = {
|
2013-07-20 17:46:00 +00:00
|
|
|
F->arg_begin()->getType(),
|
2012-07-13 23:25:25 +00:00
|
|
|
Type::getInt1Ty(F->getContext())
|
|
|
|
};
|
|
|
|
// Can't use Intrinsic::getDeclaration here as it adds a ".i1" to
|
|
|
|
// the end of the name. Change name from llvm.arm.neon.vclz.* to
|
|
|
|
// llvm.ctlz.*
|
|
|
|
FunctionType* fType = FunctionType::get(F->getReturnType(), args, false);
|
2013-07-20 17:46:00 +00:00
|
|
|
NewFn = Function::Create(fType, F->getLinkage(),
|
2012-07-13 23:25:25 +00:00
|
|
|
"llvm.ctlz." + Name.substr(14), F->getParent());
|
|
|
|
return true;
|
|
|
|
}
|
2012-07-18 00:02:16 +00:00
|
|
|
if (Name.startswith("arm.neon.vcnt")) {
|
|
|
|
NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::ctpop,
|
|
|
|
F->arg_begin()->getType());
|
|
|
|
return true;
|
|
|
|
}
|
2012-07-13 23:25:25 +00:00
|
|
|
break;
|
|
|
|
}
|
2011-12-12 04:26:04 +00:00
|
|
|
case 'c': {
|
|
|
|
if (Name.startswith("ctlz.") && F->arg_size() == 1) {
|
|
|
|
F->setName(Name + ".old");
|
2011-12-12 10:57:20 +00:00
|
|
|
NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::ctlz,
|
|
|
|
F->arg_begin()->getType());
|
2011-12-12 04:26:04 +00:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
if (Name.startswith("cttz.") && F->arg_size() == 1) {
|
|
|
|
F->setName(Name + ".old");
|
2011-12-12 10:57:20 +00:00
|
|
|
NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::cttz,
|
|
|
|
F->arg_begin()->getType());
|
2011-12-12 04:26:04 +00:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
Move the complex address expression out of DIVariable and into an extra
argument of the llvm.dbg.declare/llvm.dbg.value intrinsics.
Previously, DIVariable was a variable-length field that has an optional
reference to a Metadata array consisting of a variable number of
complex address expressions. In the case of OpPiece expressions this is
wasting a lot of storage in IR, because when an aggregate type is, e.g.,
SROA'd into all of its n individual members, the IR will contain n copies
of the DIVariable, all alike, only differing in the complex address
reference at the end.
By making the complex address into an extra argument of the
dbg.value/dbg.declare intrinsics, all of the pieces can reference the
same variable and the complex address expressions can be uniqued across
the CU, too.
Down the road, this will allow us to move other flags, such as
"indirection" out of the DIVariable, too.
The new intrinsics look like this:
declare void @llvm.dbg.declare(metadata %storage, metadata %var, metadata %expr)
declare void @llvm.dbg.value(metadata %storage, i64 %offset, metadata %var, metadata %expr)
This patch adds a new LLVM-local tag to DIExpressions, so we can detect
and pretty-print DIExpression metadata nodes.
What this patch doesn't do:
This patch does not touch the "Indirect" field in DIVariable; but moving
that into the expression would be a natural next step.
http://reviews.llvm.org/D4919
rdar://problem/17994491
Thanks to dblaikie and dexonsmith for reviewing this patch!
Note: I accidentally committed a bogus older version of this patch previously.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218787 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-01 18:55:02 +00:00
|
|
|
case 'd': {
|
|
|
|
if (Name.startswith("dbg.declare") && F->arg_size() == 2) {
|
|
|
|
F->setName(Name + ".old");
|
|
|
|
NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::dbg_declare);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
if (Name.startswith("dbg.value") && F->arg_size() == 3) {
|
|
|
|
F->setName(Name + ".old");
|
|
|
|
NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::dbg_value);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2013-10-07 18:06:48 +00:00
|
|
|
case 'o':
|
|
|
|
// We only need to change the name to match the mangling including the
|
|
|
|
// address space.
|
|
|
|
if (F->arg_size() == 2 && Name.startswith("objectsize.")) {
|
|
|
|
Type *Tys[2] = { F->getReturnType(), F->arg_begin()->getType() };
|
|
|
|
if (F->getName() != Intrinsic::getName(Intrinsic::objectsize, Tys)) {
|
|
|
|
F->setName(Name + ".old");
|
|
|
|
NewFn = Intrinsic::getDeclaration(F->getParent(),
|
|
|
|
Intrinsic::objectsize, Tys);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
|
2012-02-03 06:10:55 +00:00
|
|
|
case 'x': {
|
|
|
|
if (Name.startswith("x86.sse2.pcmpeq.") ||
|
|
|
|
Name.startswith("x86.sse2.pcmpgt.") ||
|
|
|
|
Name.startswith("x86.avx2.pcmpeq.") ||
|
2012-04-18 05:24:00 +00:00
|
|
|
Name.startswith("x86.avx2.pcmpgt.") ||
|
2012-05-08 06:58:15 +00:00
|
|
|
Name.startswith("x86.avx.vpermil.") ||
|
|
|
|
Name == "x86.avx.movnt.dq.256" ||
|
|
|
|
Name == "x86.avx.movnt.pd.256" ||
|
2012-06-09 16:46:13 +00:00
|
|
|
Name == "x86.avx.movnt.ps.256" ||
|
2013-10-15 05:20:47 +00:00
|
|
|
Name == "x86.sse42.crc32.64.8" ||
|
2014-05-29 23:35:33 +00:00
|
|
|
Name == "x86.avx.vbroadcast.ss" ||
|
|
|
|
Name == "x86.avx.vbroadcast.ss.256" ||
|
|
|
|
Name == "x86.avx.vbroadcast.sd.256" ||
|
2012-06-09 16:46:13 +00:00
|
|
|
(Name.startswith("x86.xop.vpcom") && F->arg_size() == 2)) {
|
2014-04-09 06:08:46 +00:00
|
|
|
NewFn = nullptr;
|
2012-02-03 06:10:55 +00:00
|
|
|
return true;
|
|
|
|
}
|
2012-06-10 18:42:51 +00:00
|
|
|
// SSE4.1 ptest functions may have an old signature.
|
|
|
|
if (Name.startswith("x86.sse41.ptest")) {
|
|
|
|
if (Name == "x86.sse41.ptestc")
|
|
|
|
return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestc, NewFn);
|
|
|
|
if (Name == "x86.sse41.ptestz")
|
|
|
|
return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestz, NewFn);
|
|
|
|
if (Name == "x86.sse41.ptestnzc")
|
|
|
|
return UpgradeSSE41Function(F, Intrinsic::x86_sse41_ptestnzc, NewFn);
|
|
|
|
}
|
[x86] Fix a pretty horrible bug and inconsistency in the x86 asm
parsing (and latent bug in the instruction definitions).
This is effectively a revert of r136287 which tried to address
a specific and narrow case of immediate operands failing to be accepted
by x86 instructions with a pretty heavy hammer: it introduced a new kind
of operand that behaved differently. All of that is removed with this
commit, but the test cases are both preserved and enhanced.
The core problem that r136287 and this commit are trying to handle is
that gas accepts both of the following instructions:
insertps $192, %xmm0, %xmm1
insertps $-64, %xmm0, %xmm1
These will encode to the same byte sequence, with the immediate
occupying an 8-bit entry. The first form was fixed by r136287 but that
broke the prior handling of the second form! =[ Ironically, we would
still emit the second form in some cases and then be unable to
re-assemble the output.
The reason why the first instruction failed to be handled is because
prior to r136287 the operands ere marked 'i32i8imm' which forces them to
be sign-extenable. Clearly, that won't work for 192 in a single byte.
However, making thim zero-extended or "unsigned" doesn't really address
the core issue either because it breaks negative immediates. The correct
fix is to make these operands 'i8imm' reflecting that they can be either
signed or unsigned but must be 8-bit immediates. This patch backs out
r136287 and then changes those places as well as some others to use
'i8imm' rather than one of the extended variants.
Naturally, this broke something else. The custom DAG nodes had to be
updated to have a much more accurate type constraint of an i8 node, and
a bunch of Pat immediates needed to be specified as i8 values.
The fallout didn't end there though. We also then ceased to be able to
match the instruction-specific intrinsics to the instructions so
modified. Digging, this is because they too used i32 rather than i8 in
their signature. So I've also switched those intrinsics to i8 arguments
in line with the instructions.
In order to make the intrinsic adjustments of course, I also had to add
auto upgrading for the intrinsics.
I suspect that the intrinsic argument types may have led everything down
this rabbit hole. Pretty happy with the result.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@217310 91177308-0d34-0410-b5e6-96231b3b80d8
2014-09-06 10:00:01 +00:00
|
|
|
// Several blend and other instructions with maskes used the wrong number of
|
|
|
|
// bits.
|
|
|
|
if (Name == "x86.sse41.pblendw")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_pblendw,
|
|
|
|
NewFn);
|
|
|
|
if (Name == "x86.sse41.blendpd")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_blendpd,
|
|
|
|
NewFn);
|
|
|
|
if (Name == "x86.sse41.blendps")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_blendps,
|
|
|
|
NewFn);
|
|
|
|
if (Name == "x86.sse41.insertps")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_insertps,
|
|
|
|
NewFn);
|
|
|
|
if (Name == "x86.sse41.dppd")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_dppd,
|
|
|
|
NewFn);
|
|
|
|
if (Name == "x86.sse41.dpps")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_dpps,
|
|
|
|
NewFn);
|
|
|
|
if (Name == "x86.sse41.mpsadbw")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_sse41_mpsadbw,
|
|
|
|
NewFn);
|
|
|
|
if (Name == "x86.avx.blend.pd.256")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(
|
|
|
|
F, Intrinsic::x86_avx_blend_pd_256, NewFn);
|
|
|
|
if (Name == "x86.avx.blend.ps.256")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(
|
|
|
|
F, Intrinsic::x86_avx_blend_ps_256, NewFn);
|
|
|
|
if (Name == "x86.avx.dp.ps.256")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_avx_dp_ps_256,
|
|
|
|
NewFn);
|
|
|
|
if (Name == "x86.avx2.pblendw")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_avx2_pblendw,
|
|
|
|
NewFn);
|
|
|
|
if (Name == "x86.avx2.pblendd.128")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(
|
|
|
|
F, Intrinsic::x86_avx2_pblendd_128, NewFn);
|
|
|
|
if (Name == "x86.avx2.pblendd.256")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(
|
|
|
|
F, Intrinsic::x86_avx2_pblendd_256, NewFn);
|
|
|
|
if (Name == "x86.avx2.mpsadbw")
|
|
|
|
return UpgradeX86IntrinsicsWith8BitMask(F, Intrinsic::x86_avx2_mpsadbw,
|
|
|
|
NewFn);
|
|
|
|
|
2012-06-13 07:18:53 +00:00
|
|
|
// frcz.ss/sd may need to have an argument dropped
|
|
|
|
if (Name.startswith("x86.xop.vfrcz.ss") && F->arg_size() == 2) {
|
|
|
|
F->setName(Name + ".old");
|
|
|
|
NewFn = Intrinsic::getDeclaration(F->getParent(),
|
|
|
|
Intrinsic::x86_xop_vfrcz_ss);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
if (Name.startswith("x86.xop.vfrcz.sd") && F->arg_size() == 2) {
|
|
|
|
F->setName(Name + ".old");
|
|
|
|
NewFn = Intrinsic::getDeclaration(F->getParent(),
|
|
|
|
Intrinsic::x86_xop_vfrcz_sd);
|
|
|
|
return true;
|
|
|
|
}
|
2012-06-03 08:07:25 +00:00
|
|
|
// Fix the FMA4 intrinsics to remove the 4
|
|
|
|
if (Name.startswith("x86.fma4.")) {
|
2012-06-03 16:48:52 +00:00
|
|
|
F->setName("llvm.x86.fma" + Name.substr(8));
|
|
|
|
NewFn = F;
|
|
|
|
return true;
|
2012-06-03 08:07:25 +00:00
|
|
|
}
|
2012-02-03 06:10:55 +00:00
|
|
|
break;
|
|
|
|
}
|
2007-08-04 01:51:18 +00:00
|
|
|
}
|
|
|
|
|
2012-06-10 18:42:51 +00:00
|
|
|
// This may not belong here. This function is effectively being overloaded
|
|
|
|
// to both detect an intrinsic which needs upgrading, and to provide the
|
|
|
|
// upgraded form of the intrinsic. We should perhaps have two separate
|
2007-08-04 01:51:18 +00:00
|
|
|
// functions for this.
|
2007-12-17 22:33:23 +00:00
|
|
|
return false;
|
2007-08-04 01:51:18 +00:00
|
|
|
}
|
|
|
|
|
2007-12-17 22:33:23 +00:00
|
|
|
bool llvm::UpgradeIntrinsicFunction(Function *F, Function *&NewFn) {
|
2014-04-09 06:08:46 +00:00
|
|
|
NewFn = nullptr;
|
2007-12-17 22:33:23 +00:00
|
|
|
bool Upgraded = UpgradeIntrinsicFunction1(F, NewFn);
|
2007-12-03 20:06:50 +00:00
|
|
|
|
|
|
|
// Upgrade intrinsic attributes. This does not change the function.
|
2007-12-17 22:33:23 +00:00
|
|
|
if (NewFn)
|
|
|
|
F = NewFn;
|
2009-02-05 01:49:45 +00:00
|
|
|
if (unsigned id = F->getIntrinsicID())
|
2012-10-15 04:46:55 +00:00
|
|
|
F->setAttributes(Intrinsic::getAttributes(F->getContext(),
|
|
|
|
(Intrinsic::ID)id));
|
2007-12-03 20:06:50 +00:00
|
|
|
return Upgraded;
|
|
|
|
}
|
|
|
|
|
2010-09-10 18:51:56 +00:00
|
|
|
bool llvm::UpgradeGlobalVariable(GlobalVariable *GV) {
|
2011-06-18 06:05:24 +00:00
|
|
|
// Nothing to do yet.
|
2010-09-10 18:51:56 +00:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
Move the complex address expression out of DIVariable and into an extra
argument of the llvm.dbg.declare/llvm.dbg.value intrinsics.
Previously, DIVariable was a variable-length field that has an optional
reference to a Metadata array consisting of a variable number of
complex address expressions. In the case of OpPiece expressions this is
wasting a lot of storage in IR, because when an aggregate type is, e.g.,
SROA'd into all of its n individual members, the IR will contain n copies
of the DIVariable, all alike, only differing in the complex address
reference at the end.
By making the complex address into an extra argument of the
dbg.value/dbg.declare intrinsics, all of the pieces can reference the
same variable and the complex address expressions can be uniqued across
the CU, too.
Down the road, this will allow us to move other flags, such as
"indirection" out of the DIVariable, too.
The new intrinsics look like this:
declare void @llvm.dbg.declare(metadata %storage, metadata %var, metadata %expr)
declare void @llvm.dbg.value(metadata %storage, i64 %offset, metadata %var, metadata %expr)
This patch adds a new LLVM-local tag to DIExpressions, so we can detect
and pretty-print DIExpression metadata nodes.
What this patch doesn't do:
This patch does not touch the "Indirect" field in DIVariable; but moving
that into the expression would be a natural next step.
http://reviews.llvm.org/D4919
rdar://problem/17994491
Thanks to dblaikie and dexonsmith for reviewing this patch!
Note: I accidentally committed a bogus older version of this patch previously.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218787 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-01 18:55:02 +00:00
|
|
|
static MDNode *getNodeField(const MDNode *DbgNode, unsigned Elt) {
|
|
|
|
if (!DbgNode || Elt >= DbgNode->getNumOperands())
|
|
|
|
return nullptr;
|
|
|
|
return dyn_cast_or_null<MDNode>(DbgNode->getOperand(Elt));
|
|
|
|
}
|
|
|
|
|
|
|
|
static DIExpression getExpression(Value *VarOperand, Function *F) {
|
|
|
|
// Old-style DIVariables have an optional expression as the 8th element.
|
|
|
|
DIExpression Expr(getNodeField(cast<MDNode>(VarOperand), 8));
|
|
|
|
if (!Expr) {
|
|
|
|
DIBuilder DIB(*F->getParent());
|
|
|
|
Expr = DIB.createExpression();
|
|
|
|
}
|
|
|
|
return Expr;
|
|
|
|
}
|
|
|
|
|
2012-06-10 18:42:51 +00:00
|
|
|
// UpgradeIntrinsicCall - Upgrade a call to an old intrinsic to be a call the
|
|
|
|
// upgraded intrinsic. All argument and return casting must be provided in
|
2007-08-04 01:51:18 +00:00
|
|
|
// order to seamlessly integrate with existing context.
|
|
|
|
void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
|
2012-02-03 06:10:55 +00:00
|
|
|
Function *F = CI->getCalledFunction();
|
2011-12-12 22:59:34 +00:00
|
|
|
LLVMContext &C = CI->getContext();
|
2011-12-12 04:26:04 +00:00
|
|
|
IRBuilder<> Builder(C);
|
|
|
|
Builder.SetInsertPoint(CI->getParent(), CI);
|
|
|
|
|
2012-02-03 06:10:55 +00:00
|
|
|
assert(F && "Intrinsic call is not direct?");
|
|
|
|
|
|
|
|
if (!NewFn) {
|
|
|
|
// Get the Function's name.
|
|
|
|
StringRef Name = F->getName();
|
|
|
|
|
|
|
|
Value *Rep;
|
|
|
|
// Upgrade packed integer vector compares intrinsics to compare instructions
|
|
|
|
if (Name.startswith("llvm.x86.sse2.pcmpeq.") ||
|
|
|
|
Name.startswith("llvm.x86.avx2.pcmpeq.")) {
|
|
|
|
Rep = Builder.CreateICmpEQ(CI->getArgOperand(0), CI->getArgOperand(1),
|
|
|
|
"pcmpeq");
|
|
|
|
// need to sign extend since icmp returns vector of i1
|
|
|
|
Rep = Builder.CreateSExt(Rep, CI->getType(), "");
|
|
|
|
} else if (Name.startswith("llvm.x86.sse2.pcmpgt.") ||
|
|
|
|
Name.startswith("llvm.x86.avx2.pcmpgt.")) {
|
|
|
|
Rep = Builder.CreateICmpSGT(CI->getArgOperand(0), CI->getArgOperand(1),
|
|
|
|
"pcmpgt");
|
|
|
|
// need to sign extend since icmp returns vector of i1
|
|
|
|
Rep = Builder.CreateSExt(Rep, CI->getType(), "");
|
2012-05-08 06:58:15 +00:00
|
|
|
} else if (Name == "llvm.x86.avx.movnt.dq.256" ||
|
|
|
|
Name == "llvm.x86.avx.movnt.ps.256" ||
|
|
|
|
Name == "llvm.x86.avx.movnt.pd.256") {
|
|
|
|
IRBuilder<> Builder(C);
|
|
|
|
Builder.SetInsertPoint(CI->getParent(), CI);
|
|
|
|
|
|
|
|
Module *M = F->getParent();
|
|
|
|
SmallVector<Value *, 1> Elts;
|
|
|
|
Elts.push_back(ConstantInt::get(Type::getInt32Ty(C), 1));
|
|
|
|
MDNode *Node = MDNode::get(C, Elts);
|
|
|
|
|
|
|
|
Value *Arg0 = CI->getArgOperand(0);
|
|
|
|
Value *Arg1 = CI->getArgOperand(1);
|
|
|
|
|
|
|
|
// Convert the type of the pointer to a pointer to the stored type.
|
|
|
|
Value *BC = Builder.CreateBitCast(Arg0,
|
|
|
|
PointerType::getUnqual(Arg1->getType()),
|
|
|
|
"cast");
|
|
|
|
StoreInst *SI = Builder.CreateStore(Arg1, BC);
|
|
|
|
SI->setMetadata(M->getMDKindID("nontemporal"), Node);
|
|
|
|
SI->setAlignment(16);
|
|
|
|
|
|
|
|
// Remove intrinsic.
|
|
|
|
CI->eraseFromParent();
|
|
|
|
return;
|
2012-06-09 16:46:13 +00:00
|
|
|
} else if (Name.startswith("llvm.x86.xop.vpcom")) {
|
|
|
|
Intrinsic::ID intID;
|
|
|
|
if (Name.endswith("ub"))
|
|
|
|
intID = Intrinsic::x86_xop_vpcomub;
|
|
|
|
else if (Name.endswith("uw"))
|
|
|
|
intID = Intrinsic::x86_xop_vpcomuw;
|
|
|
|
else if (Name.endswith("ud"))
|
|
|
|
intID = Intrinsic::x86_xop_vpcomud;
|
|
|
|
else if (Name.endswith("uq"))
|
|
|
|
intID = Intrinsic::x86_xop_vpcomuq;
|
|
|
|
else if (Name.endswith("b"))
|
|
|
|
intID = Intrinsic::x86_xop_vpcomb;
|
|
|
|
else if (Name.endswith("w"))
|
|
|
|
intID = Intrinsic::x86_xop_vpcomw;
|
|
|
|
else if (Name.endswith("d"))
|
|
|
|
intID = Intrinsic::x86_xop_vpcomd;
|
|
|
|
else if (Name.endswith("q"))
|
|
|
|
intID = Intrinsic::x86_xop_vpcomq;
|
|
|
|
else
|
|
|
|
llvm_unreachable("Unknown suffix");
|
|
|
|
|
|
|
|
Name = Name.substr(18); // strip off "llvm.x86.xop.vpcom"
|
|
|
|
unsigned Imm;
|
|
|
|
if (Name.startswith("lt"))
|
|
|
|
Imm = 0;
|
|
|
|
else if (Name.startswith("le"))
|
|
|
|
Imm = 1;
|
|
|
|
else if (Name.startswith("gt"))
|
|
|
|
Imm = 2;
|
|
|
|
else if (Name.startswith("ge"))
|
|
|
|
Imm = 3;
|
|
|
|
else if (Name.startswith("eq"))
|
|
|
|
Imm = 4;
|
|
|
|
else if (Name.startswith("ne"))
|
|
|
|
Imm = 5;
|
|
|
|
else if (Name.startswith("true"))
|
|
|
|
Imm = 6;
|
|
|
|
else if (Name.startswith("false"))
|
|
|
|
Imm = 7;
|
|
|
|
else
|
|
|
|
llvm_unreachable("Unknown condition");
|
|
|
|
|
|
|
|
Function *VPCOM = Intrinsic::getDeclaration(F->getParent(), intID);
|
|
|
|
Rep = Builder.CreateCall3(VPCOM, CI->getArgOperand(0),
|
|
|
|
CI->getArgOperand(1), Builder.getInt8(Imm));
|
2013-10-15 05:20:47 +00:00
|
|
|
} else if (Name == "llvm.x86.sse42.crc32.64.8") {
|
|
|
|
Function *CRC32 = Intrinsic::getDeclaration(F->getParent(),
|
|
|
|
Intrinsic::x86_sse42_crc32_32_8);
|
|
|
|
Value *Trunc0 = Builder.CreateTrunc(CI->getArgOperand(0), Type::getInt32Ty(C));
|
|
|
|
Rep = Builder.CreateCall2(CRC32, Trunc0, CI->getArgOperand(1));
|
|
|
|
Rep = Builder.CreateZExt(Rep, CI->getType(), "");
|
2014-05-29 23:35:33 +00:00
|
|
|
} else if (Name.startswith("llvm.x86.avx.vbroadcast")) {
|
|
|
|
// Replace broadcasts with a series of insertelements.
|
|
|
|
Type *VecTy = CI->getType();
|
|
|
|
Type *EltTy = VecTy->getVectorElementType();
|
|
|
|
unsigned EltNum = VecTy->getVectorNumElements();
|
|
|
|
Value *Cast = Builder.CreateBitCast(CI->getArgOperand(0),
|
|
|
|
EltTy->getPointerTo());
|
|
|
|
Value *Load = Builder.CreateLoad(Cast);
|
|
|
|
Type *I32Ty = Type::getInt32Ty(C);
|
|
|
|
Rep = UndefValue::get(VecTy);
|
|
|
|
for (unsigned I = 0; I < EltNum; ++I)
|
|
|
|
Rep = Builder.CreateInsertElement(Rep, Load,
|
|
|
|
ConstantInt::get(I32Ty, I));
|
2012-02-03 06:10:55 +00:00
|
|
|
} else {
|
2012-04-18 05:24:00 +00:00
|
|
|
bool PD128 = false, PD256 = false, PS128 = false, PS256 = false;
|
2012-05-08 06:58:15 +00:00
|
|
|
if (Name == "llvm.x86.avx.vpermil.pd.256")
|
2012-04-18 05:24:00 +00:00
|
|
|
PD256 = true;
|
2012-05-08 06:58:15 +00:00
|
|
|
else if (Name == "llvm.x86.avx.vpermil.pd")
|
2012-04-18 05:24:00 +00:00
|
|
|
PD128 = true;
|
2012-05-08 06:58:15 +00:00
|
|
|
else if (Name == "llvm.x86.avx.vpermil.ps.256")
|
2012-04-18 05:24:00 +00:00
|
|
|
PS256 = true;
|
2012-05-08 06:58:15 +00:00
|
|
|
else if (Name == "llvm.x86.avx.vpermil.ps")
|
2012-04-18 05:24:00 +00:00
|
|
|
PS128 = true;
|
|
|
|
|
|
|
|
if (PD256 || PD128 || PS256 || PS128) {
|
|
|
|
Value *Op0 = CI->getArgOperand(0);
|
|
|
|
unsigned Imm = cast<ConstantInt>(CI->getArgOperand(1))->getZExtValue();
|
|
|
|
SmallVector<Constant*, 8> Idxs;
|
|
|
|
|
|
|
|
if (PD128)
|
|
|
|
for (unsigned i = 0; i != 2; ++i)
|
|
|
|
Idxs.push_back(Builder.getInt32((Imm >> i) & 0x1));
|
|
|
|
else if (PD256)
|
|
|
|
for (unsigned l = 0; l != 4; l+=2)
|
|
|
|
for (unsigned i = 0; i != 2; ++i)
|
|
|
|
Idxs.push_back(Builder.getInt32(((Imm >> (l+i)) & 0x1) + l));
|
|
|
|
else if (PS128)
|
|
|
|
for (unsigned i = 0; i != 4; ++i)
|
|
|
|
Idxs.push_back(Builder.getInt32((Imm >> (2 * i)) & 0x3));
|
|
|
|
else if (PS256)
|
|
|
|
for (unsigned l = 0; l != 8; l+=4)
|
|
|
|
for (unsigned i = 0; i != 4; ++i)
|
|
|
|
Idxs.push_back(Builder.getInt32(((Imm >> (2 * i)) & 0x3) + l));
|
|
|
|
else
|
|
|
|
llvm_unreachable("Unexpected function");
|
|
|
|
|
|
|
|
Rep = Builder.CreateShuffleVector(Op0, Op0, ConstantVector::get(Idxs));
|
|
|
|
} else {
|
|
|
|
llvm_unreachable("Unknown function for CallInst upgrade.");
|
|
|
|
}
|
2012-02-03 06:10:55 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
CI->replaceAllUsesWith(Rep);
|
|
|
|
CI->eraseFromParent();
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2012-07-20 21:09:18 +00:00
|
|
|
std::string Name = CI->getName().str();
|
Move the complex address expression out of DIVariable and into an extra
argument of the llvm.dbg.declare/llvm.dbg.value intrinsics.
Previously, DIVariable was a variable-length field that has an optional
reference to a Metadata array consisting of a variable number of
complex address expressions. In the case of OpPiece expressions this is
wasting a lot of storage in IR, because when an aggregate type is, e.g.,
SROA'd into all of its n individual members, the IR will contain n copies
of the DIVariable, all alike, only differing in the complex address
reference at the end.
By making the complex address into an extra argument of the
dbg.value/dbg.declare intrinsics, all of the pieces can reference the
same variable and the complex address expressions can be uniqued across
the CU, too.
Down the road, this will allow us to move other flags, such as
"indirection" out of the DIVariable, too.
The new intrinsics look like this:
declare void @llvm.dbg.declare(metadata %storage, metadata %var, metadata %expr)
declare void @llvm.dbg.value(metadata %storage, i64 %offset, metadata %var, metadata %expr)
This patch adds a new LLVM-local tag to DIExpressions, so we can detect
and pretty-print DIExpression metadata nodes.
What this patch doesn't do:
This patch does not touch the "Indirect" field in DIVariable; but moving
that into the expression would be a natural next step.
http://reviews.llvm.org/D4919
rdar://problem/17994491
Thanks to dblaikie and dexonsmith for reviewing this patch!
Note: I accidentally committed a bogus older version of this patch previously.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218787 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-01 18:55:02 +00:00
|
|
|
if (!Name.empty())
|
|
|
|
CI->setName(Name + ".old");
|
2012-06-10 18:42:51 +00:00
|
|
|
|
2011-12-12 04:26:04 +00:00
|
|
|
switch (NewFn->getIntrinsicID()) {
|
|
|
|
default:
|
2011-11-27 08:42:07 +00:00
|
|
|
llvm_unreachable("Unknown function for CallInst upgrade.");
|
2011-12-12 04:26:04 +00:00
|
|
|
|
Move the complex address expression out of DIVariable and into an extra
argument of the llvm.dbg.declare/llvm.dbg.value intrinsics.
Previously, DIVariable was a variable-length field that has an optional
reference to a Metadata array consisting of a variable number of
complex address expressions. In the case of OpPiece expressions this is
wasting a lot of storage in IR, because when an aggregate type is, e.g.,
SROA'd into all of its n individual members, the IR will contain n copies
of the DIVariable, all alike, only differing in the complex address
reference at the end.
By making the complex address into an extra argument of the
dbg.value/dbg.declare intrinsics, all of the pieces can reference the
same variable and the complex address expressions can be uniqued across
the CU, too.
Down the road, this will allow us to move other flags, such as
"indirection" out of the DIVariable, too.
The new intrinsics look like this:
declare void @llvm.dbg.declare(metadata %storage, metadata %var, metadata %expr)
declare void @llvm.dbg.value(metadata %storage, i64 %offset, metadata %var, metadata %expr)
This patch adds a new LLVM-local tag to DIExpressions, so we can detect
and pretty-print DIExpression metadata nodes.
What this patch doesn't do:
This patch does not touch the "Indirect" field in DIVariable; but moving
that into the expression would be a natural next step.
http://reviews.llvm.org/D4919
rdar://problem/17994491
Thanks to dblaikie and dexonsmith for reviewing this patch!
Note: I accidentally committed a bogus older version of this patch previously.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218787 91177308-0d34-0410-b5e6-96231b3b80d8
2014-10-01 18:55:02 +00:00
|
|
|
// Upgrade debug intrinsics to use an additional DIExpression argument.
|
|
|
|
case Intrinsic::dbg_declare: {
|
|
|
|
auto NewCI =
|
|
|
|
Builder.CreateCall3(NewFn, CI->getArgOperand(0), CI->getArgOperand(1),
|
|
|
|
getExpression(CI->getArgOperand(1), F), Name);
|
|
|
|
NewCI->setDebugLoc(CI->getDebugLoc());
|
|
|
|
CI->replaceAllUsesWith(NewCI);
|
|
|
|
CI->eraseFromParent();
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
case Intrinsic::dbg_value: {
|
|
|
|
auto NewCI = Builder.CreateCall4(
|
|
|
|
NewFn, CI->getArgOperand(0), CI->getArgOperand(1), CI->getArgOperand(2),
|
|
|
|
getExpression(CI->getArgOperand(2), F), Name);
|
|
|
|
NewCI->setDebugLoc(CI->getDebugLoc());
|
|
|
|
CI->replaceAllUsesWith(NewCI);
|
|
|
|
CI->eraseFromParent();
|
|
|
|
return;
|
|
|
|
}
|
2011-12-12 04:26:04 +00:00
|
|
|
case Intrinsic::ctlz:
|
2012-05-22 15:25:31 +00:00
|
|
|
case Intrinsic::cttz:
|
2011-12-12 04:26:04 +00:00
|
|
|
assert(CI->getNumArgOperands() == 1 &&
|
|
|
|
"Mismatch between function args and call args");
|
|
|
|
CI->replaceAllUsesWith(Builder.CreateCall2(NewFn, CI->getArgOperand(0),
|
|
|
|
Builder.getFalse(), Name));
|
|
|
|
CI->eraseFromParent();
|
|
|
|
return;
|
2012-06-10 18:42:51 +00:00
|
|
|
|
2013-10-07 18:06:48 +00:00
|
|
|
case Intrinsic::objectsize:
|
|
|
|
CI->replaceAllUsesWith(Builder.CreateCall2(NewFn,
|
|
|
|
CI->getArgOperand(0),
|
|
|
|
CI->getArgOperand(1),
|
|
|
|
Name));
|
|
|
|
CI->eraseFromParent();
|
|
|
|
return;
|
|
|
|
|
2012-07-18 00:02:16 +00:00
|
|
|
case Intrinsic::ctpop: {
|
|
|
|
CI->replaceAllUsesWith(Builder.CreateCall(NewFn, CI->getArgOperand(0)));
|
|
|
|
CI->eraseFromParent();
|
|
|
|
return;
|
|
|
|
}
|
2012-07-13 23:25:25 +00:00
|
|
|
|
2012-06-13 07:18:53 +00:00
|
|
|
case Intrinsic::x86_xop_vfrcz_ss:
|
|
|
|
case Intrinsic::x86_xop_vfrcz_sd:
|
|
|
|
CI->replaceAllUsesWith(Builder.CreateCall(NewFn, CI->getArgOperand(1),
|
|
|
|
Name));
|
|
|
|
CI->eraseFromParent();
|
|
|
|
return;
|
|
|
|
|
2012-06-10 18:42:51 +00:00
|
|
|
case Intrinsic::x86_sse41_ptestc:
|
|
|
|
case Intrinsic::x86_sse41_ptestz:
|
2012-06-13 07:18:53 +00:00
|
|
|
case Intrinsic::x86_sse41_ptestnzc: {
|
2012-06-10 18:42:51 +00:00
|
|
|
// The arguments for these intrinsics used to be v4f32, and changed
|
|
|
|
// to v2i64. This is purely a nop, since those are bitwise intrinsics.
|
|
|
|
// So, the only thing required is a bitcast for both arguments.
|
|
|
|
// First, check the arguments have the old type.
|
|
|
|
Value *Arg0 = CI->getArgOperand(0);
|
|
|
|
if (Arg0->getType() != VectorType::get(Type::getFloatTy(C), 4))
|
|
|
|
return;
|
|
|
|
|
|
|
|
// Old intrinsic, add bitcasts
|
|
|
|
Value *Arg1 = CI->getArgOperand(1);
|
|
|
|
|
|
|
|
Value *BC0 =
|
|
|
|
Builder.CreateBitCast(Arg0,
|
|
|
|
VectorType::get(Type::getInt64Ty(C), 2),
|
|
|
|
"cast");
|
|
|
|
Value *BC1 =
|
|
|
|
Builder.CreateBitCast(Arg1,
|
|
|
|
VectorType::get(Type::getInt64Ty(C), 2),
|
|
|
|
"cast");
|
|
|
|
|
|
|
|
CallInst* NewCall = Builder.CreateCall2(NewFn, BC0, BC1, Name);
|
|
|
|
CI->replaceAllUsesWith(NewCall);
|
[x86] Fix a pretty horrible bug and inconsistency in the x86 asm
parsing (and latent bug in the instruction definitions).
This is effectively a revert of r136287 which tried to address
a specific and narrow case of immediate operands failing to be accepted
by x86 instructions with a pretty heavy hammer: it introduced a new kind
of operand that behaved differently. All of that is removed with this
commit, but the test cases are both preserved and enhanced.
The core problem that r136287 and this commit are trying to handle is
that gas accepts both of the following instructions:
insertps $192, %xmm0, %xmm1
insertps $-64, %xmm0, %xmm1
These will encode to the same byte sequence, with the immediate
occupying an 8-bit entry. The first form was fixed by r136287 but that
broke the prior handling of the second form! =[ Ironically, we would
still emit the second form in some cases and then be unable to
re-assemble the output.
The reason why the first instruction failed to be handled is because
prior to r136287 the operands ere marked 'i32i8imm' which forces them to
be sign-extenable. Clearly, that won't work for 192 in a single byte.
However, making thim zero-extended or "unsigned" doesn't really address
the core issue either because it breaks negative immediates. The correct
fix is to make these operands 'i8imm' reflecting that they can be either
signed or unsigned but must be 8-bit immediates. This patch backs out
r136287 and then changes those places as well as some others to use
'i8imm' rather than one of the extended variants.
Naturally, this broke something else. The custom DAG nodes had to be
updated to have a much more accurate type constraint of an i8 node, and
a bunch of Pat immediates needed to be specified as i8 values.
The fallout didn't end there though. We also then ceased to be able to
match the instruction-specific intrinsics to the instructions so
modified. Digging, this is because they too used i32 rather than i8 in
their signature. So I've also switched those intrinsics to i8 arguments
in line with the instructions.
In order to make the intrinsic adjustments of course, I also had to add
auto upgrading for the intrinsics.
I suspect that the intrinsic argument types may have led everything down
this rabbit hole. Pretty happy with the result.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@217310 91177308-0d34-0410-b5e6-96231b3b80d8
2014-09-06 10:00:01 +00:00
|
|
|
CI->eraseFromParent();
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
case Intrinsic::x86_sse41_pblendw:
|
|
|
|
case Intrinsic::x86_sse41_blendpd:
|
|
|
|
case Intrinsic::x86_sse41_blendps:
|
|
|
|
case Intrinsic::x86_sse41_insertps:
|
|
|
|
case Intrinsic::x86_sse41_dppd:
|
|
|
|
case Intrinsic::x86_sse41_dpps:
|
|
|
|
case Intrinsic::x86_sse41_mpsadbw:
|
|
|
|
case Intrinsic::x86_avx_blend_pd_256:
|
|
|
|
case Intrinsic::x86_avx_blend_ps_256:
|
|
|
|
case Intrinsic::x86_avx_dp_ps_256:
|
|
|
|
case Intrinsic::x86_avx2_pblendw:
|
|
|
|
case Intrinsic::x86_avx2_pblendd_128:
|
|
|
|
case Intrinsic::x86_avx2_pblendd_256:
|
|
|
|
case Intrinsic::x86_avx2_mpsadbw: {
|
|
|
|
// Need to truncate the last argument from i32 to i8 -- this argument models
|
|
|
|
// an inherently 8-bit immediate operand to these x86 instructions.
|
|
|
|
SmallVector<Value *, 4> Args(CI->arg_operands().begin(),
|
|
|
|
CI->arg_operands().end());
|
|
|
|
|
|
|
|
// Replace the last argument with a trunc.
|
|
|
|
Args.back() = Builder.CreateTrunc(Args.back(), Type::getInt8Ty(C), "trunc");
|
|
|
|
|
|
|
|
CallInst *NewCall = Builder.CreateCall(NewFn, Args);
|
|
|
|
CI->replaceAllUsesWith(NewCall);
|
2012-06-10 18:42:51 +00:00
|
|
|
CI->eraseFromParent();
|
|
|
|
return;
|
2007-12-17 22:33:23 +00:00
|
|
|
}
|
2012-06-13 07:18:53 +00:00
|
|
|
}
|
2007-08-04 01:51:18 +00:00
|
|
|
}
|
|
|
|
|
2013-07-20 17:46:00 +00:00
|
|
|
// This tests each Function to determine if it needs upgrading. When we find
|
|
|
|
// one we are interested in, we then upgrade all calls to reflect the new
|
2007-08-04 01:51:18 +00:00
|
|
|
// function.
|
|
|
|
void llvm::UpgradeCallsToIntrinsic(Function* F) {
|
|
|
|
assert(F && "Illegal attempt to upgrade a non-existent intrinsic.");
|
|
|
|
|
|
|
|
// Upgrade the function and check if it is a totaly new function.
|
2011-06-18 06:05:24 +00:00
|
|
|
Function *NewFn;
|
2007-12-17 22:33:23 +00:00
|
|
|
if (UpgradeIntrinsicFunction(F, NewFn)) {
|
2007-08-04 01:51:18 +00:00
|
|
|
if (NewFn != F) {
|
|
|
|
// Replace all uses to the old function with the new one if necessary.
|
2014-03-09 03:16:01 +00:00
|
|
|
for (Value::user_iterator UI = F->user_begin(), UE = F->user_end();
|
2007-08-04 01:51:18 +00:00
|
|
|
UI != UE; ) {
|
2011-06-18 06:05:24 +00:00
|
|
|
if (CallInst *CI = dyn_cast<CallInst>(*UI++))
|
2007-08-04 01:51:18 +00:00
|
|
|
UpgradeIntrinsicCall(CI, NewFn);
|
|
|
|
}
|
|
|
|
// Remove old function, no longer used, from the module.
|
|
|
|
F->eraseFromParent();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2009-08-28 23:24:31 +00:00
|
|
|
|
2013-09-28 00:22:27 +00:00
|
|
|
void llvm::UpgradeInstWithTBAATag(Instruction *I) {
|
2014-11-01 00:10:31 +00:00
|
|
|
MDNode *MD = I->getMDNode(LLVMContext::MD_tbaa);
|
2013-09-28 00:22:27 +00:00
|
|
|
assert(MD && "UpgradeInstWithTBAATag should have a TBAA tag");
|
|
|
|
// Check if the tag uses struct-path aware TBAA format.
|
|
|
|
if (isa<MDNode>(MD->getOperand(0)) && MD->getNumOperands() >= 3)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (MD->getNumOperands() == 3) {
|
|
|
|
Value *Elts[] = {
|
|
|
|
MD->getOperand(0),
|
|
|
|
MD->getOperand(1)
|
|
|
|
};
|
|
|
|
MDNode *ScalarType = MDNode::get(I->getContext(), Elts);
|
|
|
|
// Create a MDNode <ScalarType, ScalarType, offset 0, const>
|
|
|
|
Value *Elts2[] = {
|
|
|
|
ScalarType, ScalarType,
|
|
|
|
Constant::getNullValue(Type::getInt64Ty(I->getContext())),
|
|
|
|
MD->getOperand(2)
|
|
|
|
};
|
|
|
|
I->setMetadata(LLVMContext::MD_tbaa, MDNode::get(I->getContext(), Elts2));
|
|
|
|
} else {
|
|
|
|
// Create a MDNode <MD, MD, offset 0>
|
|
|
|
Value *Elts[] = {MD, MD,
|
|
|
|
Constant::getNullValue(Type::getInt64Ty(I->getContext()))};
|
|
|
|
I->setMetadata(LLVMContext::MD_tbaa, MDNode::get(I->getContext(), Elts));
|
|
|
|
}
|
|
|
|
}
|
2013-11-15 01:34:59 +00:00
|
|
|
|
|
|
|
Instruction *llvm::UpgradeBitCastInst(unsigned Opc, Value *V, Type *DestTy,
|
|
|
|
Instruction *&Temp) {
|
|
|
|
if (Opc != Instruction::BitCast)
|
2014-04-09 06:08:46 +00:00
|
|
|
return nullptr;
|
2013-11-15 01:34:59 +00:00
|
|
|
|
2014-04-09 06:08:46 +00:00
|
|
|
Temp = nullptr;
|
2013-11-15 01:34:59 +00:00
|
|
|
Type *SrcTy = V->getType();
|
|
|
|
if (SrcTy->isPtrOrPtrVectorTy() && DestTy->isPtrOrPtrVectorTy() &&
|
|
|
|
SrcTy->getPointerAddressSpace() != DestTy->getPointerAddressSpace()) {
|
|
|
|
LLVMContext &Context = V->getContext();
|
|
|
|
|
|
|
|
// We have no information about target data layout, so we assume that
|
|
|
|
// the maximum pointer size is 64bit.
|
|
|
|
Type *MidTy = Type::getInt64Ty(Context);
|
|
|
|
Temp = CastInst::Create(Instruction::PtrToInt, V, MidTy);
|
|
|
|
|
|
|
|
return CastInst::Create(Instruction::IntToPtr, Temp, DestTy);
|
|
|
|
}
|
|
|
|
|
2014-04-09 06:08:46 +00:00
|
|
|
return nullptr;
|
2013-11-15 01:34:59 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
Value *llvm::UpgradeBitCastExpr(unsigned Opc, Constant *C, Type *DestTy) {
|
|
|
|
if (Opc != Instruction::BitCast)
|
2014-04-09 06:08:46 +00:00
|
|
|
return nullptr;
|
2013-11-15 01:34:59 +00:00
|
|
|
|
|
|
|
Type *SrcTy = C->getType();
|
|
|
|
if (SrcTy->isPtrOrPtrVectorTy() && DestTy->isPtrOrPtrVectorTy() &&
|
|
|
|
SrcTy->getPointerAddressSpace() != DestTy->getPointerAddressSpace()) {
|
|
|
|
LLVMContext &Context = C->getContext();
|
|
|
|
|
|
|
|
// We have no information about target data layout, so we assume that
|
|
|
|
// the maximum pointer size is 64bit.
|
|
|
|
Type *MidTy = Type::getInt64Ty(Context);
|
|
|
|
|
|
|
|
return ConstantExpr::getIntToPtr(ConstantExpr::getPtrToInt(C, MidTy),
|
|
|
|
DestTy);
|
|
|
|
}
|
|
|
|
|
2014-04-09 06:08:46 +00:00
|
|
|
return nullptr;
|
2013-11-15 01:34:59 +00:00
|
|
|
}
|
2013-12-02 21:29:56 +00:00
|
|
|
|
|
|
|
/// Check the debug info version number, if it is out-dated, drop the debug
|
|
|
|
/// info. Return true if module is modified.
|
|
|
|
bool llvm::UpgradeDebugInfo(Module &M) {
|
2014-01-16 01:51:12 +00:00
|
|
|
unsigned Version = getDebugMetadataVersionFromModule(M);
|
|
|
|
if (Version == DEBUG_METADATA_VERSION)
|
2013-12-02 21:29:56 +00:00
|
|
|
return false;
|
|
|
|
|
2014-01-16 01:51:12 +00:00
|
|
|
bool RetCode = StripDebugInfo(M);
|
|
|
|
if (RetCode) {
|
|
|
|
DiagnosticInfoDebugMetadataVersion DiagVersion(M, Version);
|
|
|
|
M.getContext().diagnose(DiagVersion);
|
|
|
|
}
|
|
|
|
return RetCode;
|
2013-12-02 21:29:56 +00:00
|
|
|
}
|
2014-06-25 15:41:00 +00:00
|
|
|
|
|
|
|
void llvm::UpgradeMDStringConstant(std::string &String) {
|
|
|
|
const std::string OldPrefix = "llvm.vectorizer.";
|
2014-07-21 23:11:03 +00:00
|
|
|
if (String == "llvm.vectorizer.unroll") {
|
|
|
|
String = "llvm.loop.interleave.count";
|
|
|
|
} else if (String.find(OldPrefix) == 0) {
|
|
|
|
String.replace(0, OldPrefix.size(), "llvm.loop.vectorize.");
|
2014-06-25 15:41:00 +00:00
|
|
|
}
|
|
|
|
}
|