1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
|
Pull in r198145 from upstream llvm trunk (by Venkatraman Govindaraju):
[SparcV9]: Implement lowering of long double (fp128) arguments in Sparc64 ABI.
Also, pass fp128 arguments to varargs through integer registers if necessary.
Introduced here: http://svnweb.freebsd.org/changeset/base/262261
Index: test/CodeGen/SPARC/64abi.ll
===================================================================
--- test/CodeGen/SPARC/64abi.ll
+++ test/CodeGen/SPARC/64abi.ll
@@ -411,3 +411,33 @@ entry:
}
declare i32 @use_buf(i32, i8*)
+
+; CHECK-LABEL: test_fp128_args
+; CHECK-DAG: std %f0, [%fp+{{.+}}]
+; CHECK-DAG: std %f2, [%fp+{{.+}}]
+; CHECK-DAG: std %f6, [%fp+{{.+}}]
+; CHECK-DAG: std %f4, [%fp+{{.+}}]
+; CHECK: add %fp, [[Offset:[0-9]+]], %o0
+; CHECK: call _Qp_add
+; CHECK: ldd [%fp+[[Offset]]], %f0
+define fp128 @test_fp128_args(fp128 %a, fp128 %b) {
+entry:
+ %0 = fadd fp128 %a, %b
+ ret fp128 %0
+}
+
+declare i64 @receive_fp128(i64 %a, ...)
+
+; CHECK-LABEL: test_fp128_variable_args
+; CHECK-DAG: std %f4, [%sp+[[Offset0:[0-9]+]]]
+; CHECK-DAG: std %f6, [%sp+[[Offset1:[0-9]+]]]
+; CHECK-DAG: ldx [%sp+[[Offset0]]], %o2
+; CHECK-DAG: ldx [%sp+[[Offset1]]], %o3
+; CHECK: call receive_fp128
+define i64 @test_fp128_variable_args(i64 %a, fp128 %b) {
+entry:
+ %0 = call i64 (i64, ...)* @receive_fp128(i64 %a, fp128 %b)
+ ret i64 %0
+}
+
+
Index: lib/Target/Sparc/SparcISelLowering.cpp
===================================================================
--- lib/Target/Sparc/SparcISelLowering.cpp
+++ lib/Target/Sparc/SparcISelLowering.cpp
@@ -80,11 +80,14 @@ static bool CC_Sparc_Assign_f64(unsigned &ValNo, M
static bool CC_Sparc64_Full(unsigned &ValNo, MVT &ValVT,
MVT &LocVT, CCValAssign::LocInfo &LocInfo,
ISD::ArgFlagsTy &ArgFlags, CCState &State) {
- assert((LocVT == MVT::f32 || LocVT.getSizeInBits() == 64) &&
+ assert((LocVT == MVT::f32 || LocVT == MVT::f128
+ || LocVT.getSizeInBits() == 64) &&
"Can't handle non-64 bits locations");
// Stack space is allocated for all arguments starting from [%fp+BIAS+128].
- unsigned Offset = State.AllocateStack(8, 8);
+ unsigned size = (LocVT == MVT::f128) ? 16 : 8;
+ unsigned alignment = (LocVT == MVT::f128) ? 16 : 8;
+ unsigned Offset = State.AllocateStack(size, alignment);
unsigned Reg = 0;
if (LocVT == MVT::i64 && Offset < 6*8)
@@ -96,6 +99,9 @@ static bool CC_Sparc64_Full(unsigned &ValNo, MVT &
else if (LocVT == MVT::f32 && Offset < 16*8)
// Promote floats to %f1, %f3, ...
Reg = SP::F1 + Offset/4;
+ else if (LocVT == MVT::f128 && Offset < 16*8)
+ // Promote long doubles to %q0-%q28. (Which LLVM calls Q0-Q7).
+ Reg = SP::Q0 + Offset/16;
// Promote to register when possible, otherwise use the stack slot.
if (Reg) {
@@ -998,9 +1004,10 @@ static void fixupVariableFloatArgs(SmallVectorImpl
ArrayRef<ISD::OutputArg> Outs) {
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
const CCValAssign &VA = ArgLocs[i];
+ MVT ValTy = VA.getLocVT();
// FIXME: What about f32 arguments? C promotes them to f64 when calling
// varargs functions.
- if (!VA.isRegLoc() || VA.getLocVT() != MVT::f64)
+ if (!VA.isRegLoc() || (ValTy != MVT::f64 && ValTy != MVT::f128))
continue;
// The fixed arguments to a varargs function still go in FP registers.
if (Outs[VA.getValNo()].IsFixed)
@@ -1010,15 +1017,25 @@ static void fixupVariableFloatArgs(SmallVectorImpl
CCValAssign NewVA;
// Determine the offset into the argument array.
- unsigned Offset = 8 * (VA.getLocReg() - SP::D0);
+ unsigned firstReg = (ValTy == MVT::f64) ? SP::D0 : SP::Q0;
+ unsigned argSize = (ValTy == MVT::f64) ? 8 : 16;
+ unsigned Offset = argSize * (VA.getLocReg() - firstReg);
assert(Offset < 16*8 && "Offset out of range, bad register enum?");
if (Offset < 6*8) {
// This argument should go in %i0-%i5.
unsigned IReg = SP::I0 + Offset/8;
- // Full register, just bitconvert into i64.
- NewVA = CCValAssign::getReg(VA.getValNo(), VA.getValVT(),
- IReg, MVT::i64, CCValAssign::BCvt);
+ if (ValTy == MVT::f64)
+ // Full register, just bitconvert into i64.
+ NewVA = CCValAssign::getReg(VA.getValNo(), VA.getValVT(),
+ IReg, MVT::i64, CCValAssign::BCvt);
+ else {
+ assert(ValTy == MVT::f128 && "Unexpected type!");
+ // Full register, just bitconvert into i128 -- We will lower this into
+ // two i64s in LowerCall_64.
+ NewVA = CCValAssign::getCustomReg(VA.getValNo(), VA.getValVT(),
+ IReg, MVT::i128, CCValAssign::BCvt);
+ }
} else {
// This needs to go to memory, we're out of integer registers.
NewVA = CCValAssign::getMem(VA.getValNo(), VA.getValVT(),
@@ -1094,11 +1111,46 @@ SparcTargetLowering::LowerCall_64(TargetLowering::
Arg = DAG.getNode(ISD::ANY_EXTEND, DL, VA.getLocVT(), Arg);
break;
case CCValAssign::BCvt:
- Arg = DAG.getNode(ISD::BITCAST, DL, VA.getLocVT(), Arg);
+ // fixupVariableFloatArgs() may create bitcasts from f128 to i128. But
+ // SPARC does not support i128 natively. Lower it into two i64, see below.
+ if (!VA.needsCustom() || VA.getValVT() != MVT::f128
+ || VA.getLocVT() != MVT::i128)
+ Arg = DAG.getNode(ISD::BITCAST, DL, VA.getLocVT(), Arg);
break;
}
if (VA.isRegLoc()) {
+ if (VA.needsCustom() && VA.getValVT() == MVT::f128
+ && VA.getLocVT() == MVT::i128) {
+ // Store and reload into the interger register reg and reg+1.
+ unsigned Offset = 8 * (VA.getLocReg() - SP::I0);
+ unsigned StackOffset = Offset + Subtarget->getStackPointerBias() + 128;
+ SDValue StackPtr = DAG.getRegister(SP::O6, getPointerTy());
+ SDValue HiPtrOff = DAG.getIntPtrConstant(StackOffset);
+ HiPtrOff = DAG.getNode(ISD::ADD, DL, getPointerTy(), StackPtr,
+ HiPtrOff);
+ SDValue LoPtrOff = DAG.getIntPtrConstant(StackOffset + 8);
+ LoPtrOff = DAG.getNode(ISD::ADD, DL, getPointerTy(), StackPtr,
+ LoPtrOff);
+
+ // Store to %sp+BIAS+128+Offset
+ SDValue Store = DAG.getStore(Chain, DL, Arg, HiPtrOff,
+ MachinePointerInfo(),
+ false, false, 0);
+ // Load into Reg and Reg+1
+ SDValue Hi64 = DAG.getLoad(MVT::i64, DL, Store, HiPtrOff,
+ MachinePointerInfo(),
+ false, false, false, 0);
+ SDValue Lo64 = DAG.getLoad(MVT::i64, DL, Store, LoPtrOff,
+ MachinePointerInfo(),
+ false, false, false, 0);
+ RegsToPass.push_back(std::make_pair(toCallerWindow(VA.getLocReg()),
+ Hi64));
+ RegsToPass.push_back(std::make_pair(toCallerWindow(VA.getLocReg()+1),
+ Lo64));
+ continue;
+ }
+
// The custom bit on an i32 return value indicates that it should be
// passed in the high bits of the register.
if (VA.getValVT() == MVT::i32 && VA.needsCustom()) {
|
