RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 19:23:23 +01:00
Code Issues Projects Releases Wiki Activity

AArch64: mark small types (i1, i8, i16) as promoted

Browse Source 
This means the output of LowerFormalArguments returns a lowered
SDValue with the correct type (expected in SelectionDAGBuilder).
Without this, an assertion under a DEBUG macro triggers when those
types are passed on the stack.

llvm-svn: 210102
This commit is contained in:
Tim Northover 2014-06-03 13:54:53 +00:00
parent 863c5719fd
commit d56609ce6c
4 changed files with 27 additions and 34 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
lib/Target/AArch64/AArch64CallingConvention.td
View File

@ -18,9 +18,6 @@ class CCIfAlign<string Align, CCAction A> :
class CCIfBigEndian<CCAction A> : class CCIfBigEndian<CCAction A> :
CCIf<"State.getTarget().getDataLayout()->isBigEndian()", A>; CCIf<"State.getTarget().getDataLayout()->isBigEndian()", A>;
class CCIfUnallocated<string Reg, CCAction A> :
CCIf<"!State.isAllocated(AArch64::" # Reg # ")", A>;
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
// ARM AAPCS64 Calling Convention // ARM AAPCS64 Calling Convention
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
@ -45,7 +42,7 @@ def CC_AArch64_AAPCS : CallingConv<[
// Handle i1, i8, i16, i32, i64, f32, f64 and v2f64 by passing in registers, // Handle i1, i8, i16, i32, i64, f32, f64 and v2f64 by passing in registers,
// up to eight each of GPR and FPR. // up to eight each of GPR and FPR.
CCIfType<[i1, i8, i16], CCIfUnallocated<"X7", CCPromoteToType<i32>>>, CCIfType<[i1, i8, i16], CCPromoteToType<i32>>,
CCIfType<[i32], CCAssignToRegWithShadow<[W0, W1, W2, W3, W4, W5, W6, W7], CCIfType<[i32], CCAssignToRegWithShadow<[W0, W1, W2, W3, W4, W5, W6, W7],
[X0, X1, X2, X3, X4, X5, X6, X7]>>, [X0, X1, X2, X3, X4, X5, X6, X7]>>,
// i128 is split to two i64s, we can't fit half to register X7. // i128 is split to two i64s, we can't fit half to register X7.
@ -120,7 +117,7 @@ def CC_AArch64_DarwinPCS : CallingConv<[
// Handle i1, i8, i16, i32, i64, f32, f64 and v2f64 by passing in registers, // Handle i1, i8, i16, i32, i64, f32, f64 and v2f64 by passing in registers,
// up to eight each of GPR and FPR. // up to eight each of GPR and FPR.
CCIfType<[i1, i8, i16], CCIfUnallocated<"X7", CCPromoteToType<i32>>>, CCIfType<[i1, i8, i16], CCPromoteToType<i32>>,
CCIfType<[i32], CCAssignToRegWithShadow<[W0, W1, W2, W3, W4, W5, W6, W7], CCIfType<[i32], CCAssignToRegWithShadow<[W0, W1, W2, W3, W4, W5, W6, W7],
[X0, X1, X2, X3, X4, X5, X6, X7]>>, [X0, X1, X2, X3, X4, X5, X6, X7]>>,
// i128 is split to two i64s, we can't fit half to register X7. // i128 is split to two i64s, we can't fit half to register X7.
@ -143,8 +140,8 @@ def CC_AArch64_DarwinPCS : CallingConv<[
CCAssignToReg<[Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7]>>, CCAssignToReg<[Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7]>>,
// If more than will fit in registers, pass them on the stack instead. // If more than will fit in registers, pass them on the stack instead.
CCIfType<[i1, i8], CCAssignToStack<1, 1>>, CCIf<"ValVT == MVT::i1 || ValVT == MVT::i8", CCAssignToStack<1, 1>>,
CCIfType<[i16], CCAssignToStack<2, 2>>, CCIf<"ValVT == MVT::i16", CCAssignToStack<2, 2>>,
CCIfType<[i32, f32], CCAssignToStack<4, 4>>, CCIfType<[i32, f32], CCAssignToStack<4, 4>>,
CCIfType<[i64, f64, v1f64, v2f32, v1i64, v2i32, v4i16, v8i8], CCIfType<[i64, f64, v1f64, v2f32, v1i64, v2i32, v4i16, v8i8],
CCAssignToStack<8, 8>>, CCAssignToStack<8, 8>>,
@ -172,12 +169,11 @@ def CC_AArch64_DarwinPCS_VarArg : CallingConv<[
// 32bit quantity as undef. // 32bit quantity as undef.
def CC_AArch64_WebKit_JS : CallingConv<[ def CC_AArch64_WebKit_JS : CallingConv<[
// Handle i1, i8, i16, i32, and i64 passing in register X0 (W0). // Handle i1, i8, i16, i32, and i64 passing in register X0 (W0).
CCIfType<[i1, i8, i16], CCIfUnallocated<"X0", CCPromoteToType<i32>>>, CCIfType<[i1, i8, i16], CCPromoteToType<i32>>,
CCIfType<[i32], CCAssignToRegWithShadow<[W0], [X0]>>, CCIfType<[i32], CCAssignToRegWithShadow<[W0], [X0]>>,
CCIfType<[i64], CCAssignToRegWithShadow<[X0], [W0]>>, CCIfType<[i64], CCAssignToRegWithShadow<[X0], [W0]>>,
// Pass the remaining arguments on the stack instead. // Pass the remaining arguments on the stack instead.
CCIfType<[i1, i8, i16], CCAssignToStack<4, 4>>,
CCIfType<[i32, f32], CCAssignToStack<4, 4>>, CCIfType<[i32, f32], CCAssignToStack<4, 4>>,
CCIfType<[i64, f64], CCAssignToStack<8, 8>> CCIfType<[i64, f64], CCAssignToStack<8, 8>>
]>; ]>;

4
lib/Target/AArch64/AArch64FastISel.cpp
View File

@ -1218,7 +1218,6 @@ bool AArch64FastISel::ProcessCallArgs(
Arg = EmitIntExt(SrcVT, Arg, DestVT, /*isZExt*/ false); Arg = EmitIntExt(SrcVT, Arg, DestVT, /*isZExt*/ false);
if (Arg == 0) if (Arg == 0)
return false; return false;
ArgVT = DestVT;
break; break;
} }
case CCValAssign::AExt: case CCValAssign::AExt:
@ -1229,7 +1228,6 @@ bool AArch64FastISel::ProcessCallArgs(
Arg = EmitIntExt(SrcVT, Arg, DestVT, /*isZExt*/ true); Arg = EmitIntExt(SrcVT, Arg, DestVT, /*isZExt*/ true);
if (Arg == 0) if (Arg == 0)
return false; return false;
ArgVT = DestVT;
break; break;
} }
default: default:
@ -1248,7 +1246,7 @@ bool AArch64FastISel::ProcessCallArgs(
assert(VA.isMemLoc() && "Assuming store on stack."); assert(VA.isMemLoc() && "Assuming store on stack.");
// Need to store on the stack. // Need to store on the stack.
unsigned ArgSize = VA.getLocVT().getSizeInBits() / 8; unsigned ArgSize = (ArgVT.getSizeInBits() + 7) / 8;
unsigned BEAlign = 0; unsigned BEAlign = 0;
if (ArgSize < 8 && !Subtarget->isLittleEndian()) if (ArgSize < 8 && !Subtarget->isLittleEndian())

16
lib/Target/AArch64/AArch64ISelLowering.cpp
View File

@ -1781,21 +1781,21 @@ SDValue AArch64TargetLowering::LowerFormalArguments(
switch (VA.getLocInfo()) { switch (VA.getLocInfo()) {
default: default:
break; break;
case CCValAssign::BCvt:
MemVT = VA.getLocVT();
break;
case CCValAssign::SExt: case CCValAssign::SExt:
ExtType = ISD::SEXTLOAD; ExtType = ISD::SEXTLOAD;
MemVT = VA.getLocVT();
break; break;
case CCValAssign::ZExt: case CCValAssign::ZExt:
ExtType = ISD::ZEXTLOAD; ExtType = ISD::ZEXTLOAD;
MemVT = VA.getLocVT();
break; break;
case CCValAssign::AExt: case CCValAssign::AExt:
ExtType = ISD::EXTLOAD; ExtType = ISD::EXTLOAD;
MemVT = VA.getLocVT();
break; break;
} }
ArgValue = DAG.getExtLoad(ExtType, DL, VA.getValVT(), Chain, FIN, ArgValue = DAG.getExtLoad(ExtType, DL, VA.getLocVT(), Chain, FIN,
MachinePointerInfo::getFixedStack(FI), MachinePointerInfo::getFixedStack(FI),
MemVT, false, false, false, 0); MemVT, false, false, false, 0);
@ -2346,11 +2346,9 @@ AArch64TargetLowering::LowerCall(CallLoweringInfo &CLI,
// Since we pass i1/i8/i16 as i1/i8/i16 on stack and Arg is already // Since we pass i1/i8/i16 as i1/i8/i16 on stack and Arg is already
// promoted to a legal register type i32, we should truncate Arg back to // promoted to a legal register type i32, we should truncate Arg back to
// i1/i8/i16. // i1/i8/i16.
if (Arg.getValueType().isSimple() && if (VA.getValVT() == MVT::i1 || VA.getValVT() == MVT::i8 ||
Arg.getValueType().getSimpleVT() == MVT::i32 && VA.getValVT() == MVT::i16)
(VA.getLocVT() == MVT::i1 || VA.getLocVT() == MVT::i8 || Arg = DAG.getNode(ISD::TRUNCATE, DL, VA.getValVT(), Arg);
VA.getLocVT() == MVT::i16))
Arg = DAG.getNode(ISD::TRUNCATE, DL, VA.getLocVT(), Arg);
SDValue Store = SDValue Store =
DAG.getStore(Chain, DL, Arg, DstAddr, DstInfo, false, false, 0); DAG.getStore(Chain, DL, Arg, DstAddr, DstInfo, false, false, 0);

27
test/CodeGen/AArch64/arm64-abi.ll
View File

@ -1,5 +1,6 @@
; RUN: llc < %s -march=arm64 -mcpu=cyclone -enable-misched=false | FileCheck %s ; RUN: llc < %s -debug -march=arm64 -mcpu=cyclone -enable-misched=false | FileCheck %s
; RUN: llc < %s -O0 | FileCheck -check-prefix=FAST %s ; RUN: llc < %s -O0 | FileCheck -check-prefix=FAST %s
; REQUIRES: asserts
target triple = "arm64-apple-darwin" target triple = "arm64-apple-darwin"
; rdar://9932559 ; rdar://9932559
@ -8,15 +9,15 @@ entry:
; CHECK-LABEL: i8i16callee: ; CHECK-LABEL: i8i16callee:
; The 8th, 9th, 10th and 11th arguments are passed at sp, sp+2, sp+4, sp+5. ; The 8th, 9th, 10th and 11th arguments are passed at sp, sp+2, sp+4, sp+5.
; They are i8, i16, i8 and i8. ; They are i8, i16, i8 and i8.
; CHECK: ldrsb {{w[0-9]+}}, [sp, #5] ; CHECK-DAG: ldrsb {{w[0-9]+}}, [sp, #5]
; CHECK: ldrsh {{w[0-9]+}}, [sp, #2] ; CHECK-DAG: ldrsb {{w[0-9]+}}, [sp, #4]
; CHECK: ldrsb {{w[0-9]+}}, [sp] ; CHECK-DAG: ldrsh {{w[0-9]+}}, [sp, #2]
; CHECK: ldrsb {{w[0-9]+}}, [sp, #4] ; CHECK-DAG: ldrsb {{w[0-9]+}}, [sp]
; FAST-LABEL: i8i16callee: ; FAST-LABEL: i8i16callee:
; FAST: ldrb {{w[0-9]+}}, [sp, #5] ; FAST-DAG: ldrsb {{w[0-9]+}}, [sp, #5]
; FAST: ldrb {{w[0-9]+}}, [sp, #4] ; FAST-DAG: ldrsb {{w[0-9]+}}, [sp, #4]
; FAST: ldrh {{w[0-9]+}}, [sp, #2] ; FAST-DAG: ldrsh {{w[0-9]+}}, [sp, #2]
; FAST: ldrb {{w[0-9]+}}, [sp] ; FAST-DAG: ldrsb {{w[0-9]+}}, [sp]
%conv = sext i8 %a4 to i64 %conv = sext i8 %a4 to i64
%conv3 = sext i16 %a5 to i64 %conv3 = sext i16 %a5 to i64
%conv8 = sext i8 %b1 to i64 %conv8 = sext i8 %b1 to i64
@ -44,10 +45,10 @@ entry:
; CHECK: i8i16caller ; CHECK: i8i16caller
; The 8th, 9th, 10th and 11th arguments are passed at sp, sp+2, sp+4, sp+5. ; The 8th, 9th, 10th and 11th arguments are passed at sp, sp+2, sp+4, sp+5.
; They are i8, i16, i8 and i8. ; They are i8, i16, i8 and i8.
; CHECK: strb {{w[0-9]+}}, [sp, #5] ; CHECK-DAG: strb {{w[0-9]+}}, [sp, #5]
; CHECK: strb {{w[0-9]+}}, [sp, #4] ; CHECK-DAG: strb {{w[0-9]+}}, [sp, #4]
; CHECK: strh {{w[0-9]+}}, [sp, #2] ; CHECK-DAG: strh {{w[0-9]+}}, [sp, #2]
; CHECK: strb {{w[0-9]+}}, [sp] ; CHECK-DAG: strb {{w[0-9]+}}, [sp]
; CHECK: bl ; CHECK: bl
; FAST: i8i16caller ; FAST: i8i16caller
; FAST: strb {{w[0-9]+}}, [sp] ; FAST: strb {{w[0-9]+}}, [sp]