mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-23 11:13:28 +01:00
remove autoupgrade support for old forms of llvm.prefetch and the old
trampoline forms. Both of these were correct in LLVM 3.0, and we don't need to support LLVM 2.9 and earlier in mainline. llvm-svn: 145174
This commit is contained in:
Chris Lattner
parent
bdef0b7b87
commit
84bf52737a
@ -37,9 +37,6 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
|
||||
if (Name.size() <= 8 || !Name.startswith("llvm."))
|
||||
return false;
|
||||
Name = Name.substr(5); // Strip off "llvm."
|
||||
|
||||
FunctionType *FTy = F->getFunctionType();
|
||||
Module *M = F->getParent();
|
||||
|
||||
switch (Name[0]) {
|
||||
default: break;
|
||||
@ -57,55 +54,10 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
|
||||
Name.startswith("atomic.load.umax") ||
|
||||
Name.startswith("atomic.load.umin"))
|
||||
return true;
|
||||
case 'i':
|
||||
// This upgrades the old llvm.init.trampoline to the new
|
||||
// llvm.init.trampoline and llvm.adjust.trampoline pair.
|
||||
if (Name == "init.trampoline") {
|
||||
// The new llvm.init.trampoline returns nothing.
|
||||
if (FTy->getReturnType()->isVoidTy())
|
||||
break;
|
||||
|
||||
assert(FTy->getNumParams() == 3 && "old init.trampoline takes 3 args!");
|
||||
|
||||
// Change the name of the old intrinsic so that we can play with its type.
|
||||
std::string NameTmp = F->getName();
|
||||
F->setName("");
|
||||
NewFn = cast<Function>(M->getOrInsertFunction(
|
||||
NameTmp,
|
||||
Type::getVoidTy(M->getContext()),
|
||||
FTy->getParamType(0), FTy->getParamType(1),
|
||||
FTy->getParamType(2), (Type *)0));
|
||||
return true;
|
||||
}
|
||||
break;
|
||||
case 'm':
|
||||
if (Name == "memory.barrier")
|
||||
return true;
|
||||
case 'p':
|
||||
// This upgrades the llvm.prefetch intrinsic to accept one more parameter,
|
||||
// which is a instruction / data cache identifier. The old version only
|
||||
// implicitly accepted the data version.
|
||||
if (Name == "prefetch") {
|
||||
// Don't do anything if it has the correct number of arguments already
|
||||
if (FTy->getNumParams() == 4)
|
||||
break;
|
||||
|
||||
assert(FTy->getNumParams() == 3 && "old prefetch takes 3 args!");
|
||||
// We first need to change the name of the old (bad) intrinsic, because
|
||||
// its type is incorrect, but we cannot overload that name. We
|
||||
// arbitrarily unique it here allowing us to construct a correctly named
|
||||
// and typed function below.
|
||||
std::string NameTmp = F->getName();
|
||||
F->setName("");
|
||||
NewFn = cast<Function>(M->getOrInsertFunction(NameTmp,
|
||||
FTy->getReturnType(),
|
||||
FTy->getParamType(0),
|
||||
FTy->getParamType(1),
|
||||
FTy->getParamType(2),
|
||||
FTy->getParamType(2),
|
||||
(Type*)0));
|
||||
return true;
|
||||
}
|
||||
|
||||
break;
|
||||
}
|
||||
|
||||
@ -223,58 +175,6 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
|
||||
}
|
||||
return;
|
||||
}
|
||||
|
||||
switch (NewFn->getIntrinsicID()) {
|
||||
case Intrinsic::prefetch: {
|
||||
IRBuilder<> Builder(C);
|
||||
Builder.SetInsertPoint(CI->getParent(), CI);
|
||||
llvm::Type *I32Ty = llvm::Type::getInt32Ty(CI->getContext());
|
||||
|
||||
// Add the extra "data cache" argument
|
||||
Value *Operands[4] = { CI->getArgOperand(0), CI->getArgOperand(1),
|
||||
CI->getArgOperand(2),
|
||||
llvm::ConstantInt::get(I32Ty, 1) };
|
||||
CallInst *NewCI = CallInst::Create(NewFn, Operands,
|
||||
CI->getName(), CI);
|
||||
NewCI->setTailCall(CI->isTailCall());
|
||||
NewCI->setCallingConv(CI->getCallingConv());
|
||||
// Handle any uses of the old CallInst.
|
||||
if (!CI->use_empty())
|
||||
// Replace all uses of the old call with the new cast which has the
|
||||
// correct type.
|
||||
CI->replaceAllUsesWith(NewCI);
|
||||
|
||||
// Clean up the old call now that it has been completely upgraded.
|
||||
CI->eraseFromParent();
|
||||
break;
|
||||
}
|
||||
case Intrinsic::init_trampoline: {
|
||||
|
||||
// Transform
|
||||
// %tramp = call i8* llvm.init.trampoline (i8* x, i8* y, i8* z)
|
||||
// to
|
||||
// call void llvm.init.trampoline (i8* %x, i8* %y, i8* %z)
|
||||
// %tramp = call i8* llvm.adjust.trampoline (i8* %x)
|
||||
|
||||
Function *AdjustTrampolineFn =
|
||||
cast<Function>(Intrinsic::getDeclaration(F->getParent(),
|
||||
Intrinsic::adjust_trampoline));
|
||||
|
||||
IRBuilder<> Builder(C);
|
||||
Builder.SetInsertPoint(CI);
|
||||
|
||||
Builder.CreateCall3(NewFn, CI->getArgOperand(0), CI->getArgOperand(1),
|
||||
CI->getArgOperand(2));
|
||||
|
||||
CallInst *AdjustCall = Builder.CreateCall(AdjustTrampolineFn,
|
||||
CI->getArgOperand(0),
|
||||
CI->getName());
|
||||
if (!CI->use_empty())
|
||||
CI->replaceAllUsesWith(AdjustCall);
|
||||
CI->eraseFromParent();
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// This tests each Function to determine if it needs upgrading. When we find
|
||||
|
||||
@ -1,24 +0,0 @@
|
||||
; Tests to make sure intrinsics are automatically upgraded.
|
||||
; RUN: llvm-as < %s | llvm-dis | FileCheck %s
|
||||
|
||||
|
||||
declare void @llvm.prefetch(i8*, i32, i32) nounwind
|
||||
|
||||
define void @p(i8* %ptr) {
|
||||
; CHECK: llvm.prefetch(i8* %ptr, i32 0, i32 1, i32 1)
|
||||
tail call void @llvm.prefetch(i8* %ptr, i32 0, i32 1)
|
||||
ret void
|
||||
}
|
||||
|
||||
declare i32 @nest_f(i8* nest, i32)
|
||||
declare i8* @llvm.init.trampoline(i8*, i8*, i8*)
|
||||
|
||||
define void @test_trampolines() {
|
||||
; CHECK: call void @llvm.init.trampoline(i8* null, i8* bitcast (i32 (i8*, i32)* @nest_f to i8*), i8* null)
|
||||
; CHECK: call i8* @llvm.adjust.trampoline(i8* null)
|
||||
|
||||
call i8* @llvm.init.trampoline(i8* null,
|
||||
i8* bitcast (i32 (i8*, i32)* @nest_f to i8*),
|
||||
i8* null)
|
||||
ret void
|
||||
}
|
||||
@ -1,14 +0,0 @@
|
||||
; RUN: llc < %s -march=x86
|
||||
; RUN: llc < %s -march=x86-64
|
||||
|
||||
%struct.FRAME.gnat__perfect_hash_generators__select_char_position__build_identical_keys_sets = type { i32, i32, void (i32, i32)*, i8 (i32, i32)* }
|
||||
|
||||
define fastcc i32 @gnat__perfect_hash_generators__select_char_position__build_identical_keys_sets.5146(i64 %table.0.0, i64 %table.0.1, i32 %last, i32 %pos) {
|
||||
entry:
|
||||
%tramp22 = call i8* @llvm.init.trampoline( i8* null, i8* bitcast (void (%struct.FRAME.gnat__perfect_hash_generators__select_char_position__build_identical_keys_sets*, i32, i32)* @gnat__perfect_hash_generators__select_char_position__build_identical_keys_sets__move.5177 to i8*), i8* null ) ; <i8*> [#uses=0]
|
||||
unreachable
|
||||
}
|
||||
|
||||
declare void @gnat__perfect_hash_generators__select_char_position__build_identical_keys_sets__move.5177(%struct.FRAME.gnat__perfect_hash_generators__select_char_position__build_identical_keys_sets* nest , i32, i32) nounwind
|
||||
|
||||
declare i8* @llvm.init.trampoline(i8*, i8*, i8*) nounwind
|
||||
@ -6,7 +6,6 @@ declare i1 @llvm.isunordered.f32(float, float)
|
||||
|
||||
declare i1 @llvm.isunordered.f64(double, double)
|
||||
|
||||
declare void @llvm.prefetch(i8*, i32, i32)
|
||||
|
||||
declare i8 @llvm.ctpop.i8(i8)
|
||||
|
||||
@ -41,7 +40,6 @@ declare double @llvm.sqrt.f64(double)
|
||||
define void @libm() {
|
||||
fcmp uno float 1.000000e+00, 2.000000e+00 ; <i1>:1 [#uses=0]
|
||||
fcmp uno double 3.000000e+00, 4.000000e+00 ; <i1>:2 [#uses=0]
|
||||
call void @llvm.prefetch( i8* null, i32 1, i32 3 )
|
||||
call float @llvm.sqrt.f32( float 5.000000e+00 ) ; <float>:3 [#uses=0]
|
||||
call double @llvm.sqrt.f64( double 6.000000e+00 ) ; <double>:4 [#uses=0]
|
||||
call i8 @llvm.ctpop.i8( i8 10 ) ; <i32>:5 [#uses=0]
|
||||
|
||||
@ -1,57 +0,0 @@
|
||||
; RUN: opt < %s -globalopt -S | grep { nest } | count 1
|
||||
%struct.FRAME.nest = type { i32, i32 (i32)* }
|
||||
%struct.__builtin_trampoline = type { [10 x i8] }
|
||||
@.str = internal constant [7 x i8] c"%d %d\0A\00" ; <[7 x i8]*> [#uses=1]
|
||||
|
||||
define i32 @process(i32 (i32)* %func) nounwind {
|
||||
entry:
|
||||
%tmp2 = tail call i32 %func( i32 1 ) nounwind ; <i32> [#uses=1]
|
||||
ret i32 %tmp2
|
||||
}
|
||||
|
||||
define internal fastcc i32 @g.1478(%struct.FRAME.nest* nest %CHAIN.1, i32 %m) nounwind {
|
||||
entry:
|
||||
%tmp3 = getelementptr %struct.FRAME.nest* %CHAIN.1, i32 0, i32 0 ; <i32*> [#uses=1]
|
||||
%tmp4 = load i32* %tmp3, align 4 ; <i32> [#uses=1]
|
||||
%tmp7 = icmp eq i32 %tmp4, %m ; <i1> [#uses=1]
|
||||
%tmp78 = zext i1 %tmp7 to i32 ; <i32> [#uses=1]
|
||||
ret i32 %tmp78
|
||||
}
|
||||
|
||||
define internal i32 @f.1481(%struct.FRAME.nest* nest %CHAIN.2, i32 %m) nounwind {
|
||||
entry:
|
||||
%tmp4 = tail call fastcc i32 @g.1478( %struct.FRAME.nest* nest %CHAIN.2, i32 %m ) nounwind ; <i32> [#uses=1]
|
||||
%tmp6 = getelementptr %struct.FRAME.nest* %CHAIN.2, i32 0, i32 0 ; <i32*> [#uses=1]
|
||||
%tmp7 = load i32* %tmp6, align 4 ; <i32> [#uses=1]
|
||||
%tmp9 = icmp eq i32 %tmp4, %tmp7 ; <i1> [#uses=1]
|
||||
%tmp910 = zext i1 %tmp9 to i32 ; <i32> [#uses=1]
|
||||
ret i32 %tmp910
|
||||
}
|
||||
|
||||
define i32 @nest(i32 %n) nounwind {
|
||||
entry:
|
||||
%TRAMP.316 = alloca [10 x i8] ; <[10 x i8]*> [#uses=1]
|
||||
%FRAME.0 = alloca %struct.FRAME.nest ; <%struct.FRAME.nest*> [#uses=3]
|
||||
%TRAMP.316.sub = getelementptr [10 x i8]* %TRAMP.316, i32 0, i32 0 ; <i8*> [#uses=1]
|
||||
%tmp3 = getelementptr %struct.FRAME.nest* %FRAME.0, i32 0, i32 0 ; <i32*> [#uses=1]
|
||||
store i32 %n, i32* %tmp3, align 8
|
||||
%FRAME.06 = bitcast %struct.FRAME.nest* %FRAME.0 to i8* ; <i8*> [#uses=1]
|
||||
%tramp = call i8* @llvm.init.trampoline( i8* %TRAMP.316.sub, i8* bitcast (i32 (%struct.FRAME.nest*, i32)* @f.1481 to i8*), i8* %FRAME.06 ) ; <i8*> [#uses=1]
|
||||
%tmp7 = getelementptr %struct.FRAME.nest* %FRAME.0, i32 0, i32 1 ; <i32 (i32)**> [#uses=1]
|
||||
%tmp89 = bitcast i8* %tramp to i32 (i32)* ; <i32 (i32)*> [#uses=2]
|
||||
store i32 (i32)* %tmp89, i32 (i32)** %tmp7, align 4
|
||||
%tmp13 = call i32 @process( i32 (i32)* %tmp89 ) nounwind ; <i32> [#uses=1]
|
||||
ret i32 %tmp13
|
||||
}
|
||||
|
||||
declare i8* @llvm.init.trampoline(i8*, i8*, i8*) nounwind
|
||||
|
||||
define i32 @main() nounwind {
|
||||
entry:
|
||||
%tmp = tail call i32 @nest( i32 2 ) nounwind ; <i32> [#uses=1]
|
||||
%tmp1 = tail call i32 @nest( i32 1 ) nounwind ; <i32> [#uses=1]
|
||||
%tmp3 = tail call i32 (i8*, ...)* @printf( i8* noalias getelementptr ([7 x i8]* @.str, i32 0, i32 0), i32 %tmp1, i32 %tmp ) nounwind ; <i32> [#uses=0]
|
||||
ret i32 undef
|
||||
}
|
||||
|
||||
declare i32 @printf(i8*, ...) nounwind
|
||||
@ -1,24 +0,0 @@
|
||||
; RUN: opt < %s -instcombine -S | grep {call i32 @f}
|
||||
|
||||
%struct.FRAME.nest = type { i32, i32 (i32)* }
|
||||
%struct.__builtin_trampoline = type { [10 x i8] }
|
||||
|
||||
declare i8* @llvm.init.trampoline(i8*, i8*, i8*)
|
||||
|
||||
declare i32 @f(%struct.FRAME.nest* nest , i32 )
|
||||
|
||||
define i32 @nest(i32 %n) {
|
||||
entry:
|
||||
%FRAME.0 = alloca %struct.FRAME.nest, align 8 ; <%struct.FRAME.nest*> [#uses=3]
|
||||
%TRAMP.216 = alloca [10 x i8], align 16 ; <[10 x i8]*> [#uses=1]
|
||||
%TRAMP.216.sub = getelementptr [10 x i8]* %TRAMP.216, i32 0, i32 0 ; <i8*> [#uses=1]
|
||||
%tmp3 = getelementptr %struct.FRAME.nest* %FRAME.0, i32 0, i32 0 ; <i32*> [#uses=1]
|
||||
store i32 %n, i32* %tmp3, align 8
|
||||
%FRAME.06 = bitcast %struct.FRAME.nest* %FRAME.0 to i8* ; <i8*> [#uses=1]
|
||||
%tramp = call i8* @llvm.init.trampoline( i8* %TRAMP.216.sub, i8* bitcast (i32 (%struct.FRAME.nest* , i32)* @f to i8*), i8* %FRAME.06 ) ; <i8*> [#uses=1]
|
||||
%tmp7 = getelementptr %struct.FRAME.nest* %FRAME.0, i32 0, i32 1 ; <i32 (i32)**> [#uses=1]
|
||||
%tmp89 = bitcast i8* %tramp to i32 (i32)* ; <i32 (i32)*> [#uses=2]
|
||||
store i32 (i32)* %tmp89, i32 (i32)** %tmp7, align 8
|
||||
%tmp2.i = call i32 %tmp89( i32 1 ) ; <i32> [#uses=1]
|
||||
ret i32 %tmp2.i
|
||||
}
|
||||
@ -1,24 +0,0 @@
|
||||
; RUN: opt < %s -instcombine -disable-output
|
||||
|
||||
%struct.FRAME.nest = type { i32, i32 (i32*)* }
|
||||
%struct.__builtin_trampoline = type { [10 x i8] }
|
||||
|
||||
declare i8* @llvm.init.trampoline(i8*, i8*, i8*) nounwind
|
||||
|
||||
declare i32 @f(%struct.FRAME.nest* nest , i32*)
|
||||
|
||||
define i32 @nest(i32 %n) {
|
||||
entry:
|
||||
%FRAME.0 = alloca %struct.FRAME.nest, align 8 ; <%struct.FRAME.nest*> [#uses=3]
|
||||
%TRAMP.216 = alloca [10 x i8], align 16 ; <[10 x i8]*> [#uses=1]
|
||||
%TRAMP.216.sub = getelementptr [10 x i8]* %TRAMP.216, i32 0, i32 0 ; <i8*> [#uses=1]
|
||||
%tmp3 = getelementptr %struct.FRAME.nest* %FRAME.0, i32 0, i32 0 ; <i32*> [#uses=1]
|
||||
store i32 %n, i32* %tmp3, align 8
|
||||
%FRAME.06 = bitcast %struct.FRAME.nest* %FRAME.0 to i8* ; <i8*> [#uses=1]
|
||||
%tramp = call i8* @llvm.init.trampoline( i8* %TRAMP.216.sub, i8* bitcast (i32 (%struct.FRAME.nest*, i32*)* @f to i8*), i8* %FRAME.06 ) ; <i8*> [#uses=1]
|
||||
%tmp7 = getelementptr %struct.FRAME.nest* %FRAME.0, i32 0, i32 1 ; <i32 (i32*)**> [#uses=1]
|
||||
%tmp89 = bitcast i8* %tramp to i32 (i32*)* ; <i32 (i32*)*> [#uses=2]
|
||||
store i32 (i32*)* %tmp89, i32 (i32*)** %tmp7, align 8
|
||||
%tmp2.i = call i32 %tmp89( i32* nest null ) ; <i32> [#uses=1]
|
||||
ret i32 %tmp2.i
|
||||
}
|
||||
Loading…
Reference in New Issue
Block a user