Skip to content

Commit bdd7652

Browse files
jcranmer-inteljcranmer-intel
authored
[OpaquePointers] Handle llvm.memset intrinsic mangling mismatches. (#1845)
Non-constant @llvm.memset calls are presently lowered by generating synthetic functions with the mangled name of memset. However, the reader tries to use this name to generate the intrinsic call again. This causes verification issues if the SPIRVWriter and SPIRVReader do not agree on whether or not to use opaque pointers. This change uses the actual type of the function (which will take into account whether or not it is in opaque pointer mode) to generate the LLVM intrinsic name, fixing the mismatch issues.
1 parent 9633373 commit bdd7652

File tree

3 files changed

+150
-0
lines changed

3 files changed

+150
-0
lines changed

‎lib/SPIRV/SPIRVReader.cpp‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2885,6 +2885,14 @@ Function *SPIRVToLLVM::transFunction(SPIRVFunction *BF) {
28852885
// assuming llvm.memset is supported by the device compiler. If this
28862886
// assumption is not safe, we should have a command line option to control
28872887
// this behavior.
2888+
if (FuncNameRef.startswith("spirv.llvm_memset_p")) {
2889+
// We can't guarantee that the name is correctly mangled due to opaque
2890+
// pointers. Derive the correct name from the function type.
2891+
FuncName =
2892+
Intrinsic::getDeclaration(M, Intrinsic::memset,
2893+
{FT->getParamType(0), FT->getParamType(2)})
2894+
->getName();
2895+
}
28882896
if (FuncNameRef.consume_front("spirv.")) {
28892897
FuncNameRef.consume_back(".volatile");
28902898
FuncName = FuncNameRef.str();

‎test/llvm-intrinsics/memset-opaque.ll‎

Lines changed: 134 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,134 @@
1+
; RUN: llvm-as %s -o %t.bc
2+
; RUN: llvm-spirv %t.bc -spirv-text -o %t.spt
3+
; RUN: FileCheck < %t.spt %s --check-prefix=CHECK-SPIRV
4+
; RUN: llvm-spirv %t.bc -o %t.spv
5+
; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
6+
; RUN: spirv-val %t.spv
7+
; RUN: llvm-dis -opaque-pointers=0 < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
8+
; RUN: llvm-spirv -r %t.spv -o - -emit-opaque-pointers | llvm-dis | FileCheck %s --check-prefix=CHECK-LLVM-OPAQUE
9+
10+
; CHECK-SPIRV: Decorate [[#NonConstMemset:]] LinkageAttributes "spirv.llvm_memset_p3_i32"
11+
; CHECK-SPIRV: TypeInt [[Int8:[0-9]+]] 8 0
12+
; CHECK-SPIRV: Constant {{[0-9]+}} [[Lenmemset21:[0-9]+]] 4
13+
; CHECK-SPIRV: Constant {{[0-9]+}} [[Lenmemset0:[0-9]+]] 12
14+
; CHECK-SPIRV: Constant {{[0-9]+}} [[Const21:[0-9]+]] 21
15+
; CHECK-SPIRV: TypeArray [[Int8x4:[0-9]+]] [[Int8]] [[Lenmemset21]]
16+
; CHECK-SPIRV: TypePointer [[Int8Ptr:[0-9]+]] 8 [[Int8]]
17+
; CHECK-SPIRV: TypeArray [[Int8x12:[0-9]+]] [[Int8]] [[Lenmemset0]]
18+
; CHECK-SPIRV: TypePointer [[Int8PtrConst:[0-9]+]] 0 [[Int8]]
19+
20+
; CHECK-SPIRV: ConstantNull [[Int8x12]] [[Init:[0-9]+]]
21+
; CHECK-SPIRV: Variable {{[0-9]+}} [[Val:[0-9]+]] 0 [[Init]]
22+
; CHECK-SPIRV: 7 ConstantComposite [[Int8x4]] [[InitComp:[0-9]+]] [[Const21]] [[Const21]] [[Const21]] [[Const21]]
23+
; CHECK-SPIRV: Variable {{[0-9]+}} [[ValComp:[0-9]+]] 0 [[InitComp]]
24+
; CHECK-SPIRV: ConstantFalse [[#]] [[#False:]]
25+
26+
; CHECK-SPIRV: Bitcast [[Int8Ptr]] [[Target:[0-9]+]] {{[0-9]+}}
27+
; CHECK-SPIRV: Bitcast [[Int8PtrConst]] [[Source:[0-9]+]] [[Val]]
28+
; CHECK-SPIRV: CopyMemorySized [[Target]] [[Source]] [[Lenmemset0]] 2 4
29+
30+
; CHECK-SPIRV: Bitcast [[Int8PtrConst]] [[SourceComp:[0-9]+]] [[ValComp]]
31+
; CHECK-SPIRV: CopyMemorySized {{[0-9]+}} [[SourceComp]] [[Lenmemset21]] 2 4
32+
33+
; CHECK-SPIRV: FunctionCall [[#]] [[#]] [[#NonConstMemset]] [[#]] [[#]] [[#]] [[#False]]
34+
35+
; CHECK-SPIRV: Function [[#]] [[#NonConstMemset]]
36+
; CHECK-SPIRV: FunctionParameter [[#]] [[#Dest:]]
37+
; CHECK-SPIRV: FunctionParameter [[#]] [[#Value:]]
38+
; CHECK-SPIRV: FunctionParameter [[#]] [[#Len:]]
39+
; CHECK-SPIRV: FunctionParameter [[#]] [[#Volatile:]]
40+
41+
; CHECK-SPIRV: Label [[#Entry:]]
42+
; CHECK-SPIRV: IEqual [[#]] [[#IsZeroLen:]] [[#Zero:]] [[#Len]]
43+
; CHECK-SPIRV: BranchConditional [[#IsZeroLen]] [[#End:]] [[#WhileBody:]]
44+
45+
; CHECK-SPIRV: Label [[#WhileBody]]
46+
; CHECK-SPIRV: Phi [[#]] [[#Offset:]] [[#Zero]] [[#Entry]] [[#OffsetInc:]] [[#WhileBody]]
47+
; CHECK-SPIRV: Bitcast [[#]] [[#DestU8:]] [[#Dest]]
48+
; CHECK-SPIRV: InBoundsPtrAccessChain [[#]] [[#Ptr:]] [[#DestU8]] [[#Offset]]
49+
; CHECK-SPIRV: Store [[#Ptr]] [[#Value]] 2 1
50+
; CHECK-SPIRV: IAdd [[#]] [[#OffsetInc]] [[#Offset]] [[#One:]]
51+
; CHECK-SPIRV: ULessThan [[#]] [[#NotEnd:]] [[#OffsetInc]] [[#Len]]
52+
; CHECK-SPIRV: BranchConditional [[#NotEnd]] [[#WhileBody]] [[#End]]
53+
54+
; CHECK-SPIRV: Label [[#End]]
55+
; CHECK-SPIRV: Return
56+
57+
; CHECK-SPIRV: FunctionEnd
58+
59+
target datalayout = "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
60+
target triple = "spir"
61+
62+
%struct.S1 = type { i32, i32, i32 }
63+
64+
; CHECK-LLVM: internal unnamed_addr addrspace(2) constant [12 x i8] zeroinitializer
65+
; CHECK-LLVM: internal unnamed_addr addrspace(2) constant [4 x i8] c"\15\15\15\15"
66+
67+
; Function Attrs: nounwind
68+
define spir_func void @_Z5foo11v(ptr addrspace(4) noalias nocapture sret(%struct.S1) %agg.result, i32 %s1, i64 %s2, i8 %v) #0 {
69+
%x = alloca [4 x i8]
70+
tail call void @llvm.memset.p4.i32(ptr addrspace(4) align 4 %agg.result, i8 0, i32 12, i1 false)
71+
; CHECK-LLVM: call void @llvm.memcpy.p4i8.p2i8.i32(i8 addrspace(4)* align 4 %1, i8 addrspace(2)* align 4 %2, i32 12, i1 false)
72+
; CHECK-LLVM-OPAQUE: call void @llvm.memcpy.p4.p2.i32(ptr addrspace(4) align 4 %1, ptr addrspace(2) align 4 %2, i32 12, i1 false)
73+
tail call void @llvm.memset.p0.i32(ptr align 4 %x, i8 21, i32 4, i1 false)
74+
; CHECK-LLVM: call void @llvm.memcpy.p0i8.p2i8.i32(i8* align 4 %3, i8 addrspace(2)* align 4 %4, i32 4, i1 false)
75+
; CHECK-LLVM-OPAQUE: call void @llvm.memcpy.p0.p2.i32(ptr align 4 %3, ptr addrspace(2) align 4 %4, i32 4, i1 false)
76+
77+
; non-const value
78+
tail call void @llvm.memset.p0.i32(ptr align 4 %x, i8 %v, i32 3, i1 false)
79+
; CHECK-LLVM: call void @llvm.memset.p0a4i8.i32([4 x i8]* %x, i8 %v, i32 3, i1 false)
80+
; CHECK-LLVM-OPAQUE: call void @llvm.memset.p0.i32(ptr %x, i8 %v, i32 3, i1 false)
81+
82+
; non-const value and size
83+
tail call void @llvm.memset.p0.i32(ptr align 4 %x, i8 %v, i32 %s1, i1 false)
84+
; CHECK-LLVM: call void @llvm.memset.p0a4i8.i32([4 x i8]* %x, i8 %v, i32 %s1, i1 false)
85+
; CHECK-LLVM-OPAQUE: call void @llvm.memset.p0.i32(ptr %x, i8 %v, i32 %s1, i1 false)
86+
87+
; Address spaces, non-const value and size
88+
%a = addrspacecast ptr addrspace(4) %agg.result to ptr addrspace(3)
89+
tail call void @llvm.memset.p3.i32(ptr addrspace(3) align 4 %a, i8 %v, i32 %s1, i1 false)
90+
; CHECK-LLVM: call void @llvm.memset.p3s_struct.S1s.i32(%struct.S1 addrspace(3)* %a, i8 %v, i32 %s1, i1 false)
91+
; CHECK-LLVM-OPAQUE: call void @llvm.memset.p3.i32(ptr addrspace(3) %a, i8 %v, i32 %s1, i1 false)
92+
%b = addrspacecast ptr addrspace(4) %agg.result to ptr addrspace(1)
93+
tail call void @llvm.memset.p1.i64(ptr addrspace(1) align 4 %b, i8 %v, i64 %s2, i1 false)
94+
; CHECK-LLVM: call void @llvm.memset.p1s_struct.S1s.i64(%struct.S1 addrspace(1)* %b, i8 %v, i64 %s2, i1 false)
95+
; CHECK-LLVM-OPAQUE: call void @llvm.memset.p1.i64(ptr addrspace(1) %b, i8 %v, i64 %s2, i1 false)
96+
97+
; Volatile
98+
tail call void @llvm.memset.p1.i64(ptr addrspace(1) align 4 %b, i8 %v, i64 %s2, i1 true)
99+
; CHECK-LLVM: call void @llvm.memset.p1s_struct.S1s.i64(%struct.S1 addrspace(1)* %b, i8 %v, i64 %s2, i1 true)
100+
; CHECK-LLVM-OPAQUE: call void @llvm.memset.p1.i64(ptr addrspace(1) %b, i8 %v, i64 %s2, i1 true)
101+
ret void
102+
}
103+
104+
; Function Attrs: nounwind
105+
declare void @llvm.memset.p4.i32(ptr addrspace(4) nocapture, i8, i32, i1) #1
106+
107+
; Function Attrs: nounwind
108+
declare void @llvm.memset.p0.i32(ptr nocapture, i8, i32, i1) #1
109+
110+
; Function Attrs: nounwind
111+
declare void @llvm.memset.p3.i32(ptr addrspace(3), i8, i32, i1) #1
112+
113+
; Function Attrs: nounwind
114+
declare void @llvm.memset.p1.i64(ptr addrspace(1), i8, i64, i1) #1
115+
116+
attributes #0 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-realign-stack" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
117+
attributes #1 = { nounwind }
118+
119+
!opencl.enable.FP_CONTRACT = !{}
120+
!opencl.spir.version = !{!0}
121+
!opencl.ocl.version = !{!1}
122+
!opencl.used.extensions = !{!2}
123+
!opencl.used.optional.core.features = !{!2}
124+
!opencl.compiler.options = !{!2}
125+
!llvm.ident = !{!3}
126+
!spirv.Source = !{!4}
127+
!spirv.String = !{!5}
128+
129+
!0 = !{i32 1, i32 2}
130+
!1 = !{i32 2, i32 2}
131+
!2 = !{}
132+
!3 = !{!"clang version 3.6.1 "}
133+
!4 = !{i32 4, i32 202000, !5}
134+
!5 = !{!"llvm.memset.cl"}

‎test/llvm-intrinsics/memset.ll‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5,6 +5,7 @@
55
; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
66
; RUN: spirv-val %t.spv
77
; RUN: llvm-dis -opaque-pointers=0 < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
8+
; RUN: llvm-spirv -r %t.spv -o - -emit-opaque-pointers | llvm-dis | FileCheck %s --check-prefix=CHECK-LLVM-OPAQUE
89

910
; CHECK-SPIRV: Decorate [[#NonConstMemset:]] LinkageAttributes "spirv.llvm_memset_p3i8_i32"
1011
; CHECK-SPIRV: TypeInt [[Int8:[0-9]+]] 8 0
@@ -69,28 +70,35 @@ define spir_func void @_Z5foo11v(%struct.S1 addrspace(4)* noalias nocapture sret
6970
%1 = bitcast %struct.S1 addrspace(4)* %agg.result to i8 addrspace(4)*
7071
tail call void @llvm.memset.p4i8.i32(i8 addrspace(4)* align 4 %1, i8 0, i32 12, i1 false)
7172
; CHECK-LLVM: call void @llvm.memcpy.p4i8.p2i8.i32(i8 addrspace(4)* align 4 %1, i8 addrspace(2)* align 4 %2, i32 12, i1 false)
73+
; CHECK-LLVM-OPAQUE: call void @llvm.memcpy.p4.p2.i32(ptr addrspace(4) align 4 %1, ptr addrspace(2) align 4 %2, i32 12, i1 false)
7274
tail call void @llvm.memset.p0i8.i32(i8* align 4 %x.bc, i8 21, i32 4, i1 false)
7375
; CHECK-LLVM: call void @llvm.memcpy.p0i8.p2i8.i32(i8* align 4 %x.bc, i8 addrspace(2)* align 4 %3, i32 4, i1 false)
76+
; CHECK-LLVM-OPAQUE: call void @llvm.memcpy.p0.p2.i32(ptr align 4 %x.bc, ptr addrspace(2) align 4 %3, i32 4, i1 false)
7477

7578
; non-const value
7679
tail call void @llvm.memset.p0i8.i32(i8* align 4 %x.bc, i8 %v, i32 3, i1 false)
7780
; CHECK-LLVM: call void @llvm.memset.p0i8.i32(i8* %x.bc, i8 %v, i32 3, i1 false)
81+
; CHECK-LLVM-OPAQUE: call void @llvm.memset.p0.i32(ptr %x.bc, i8 %v, i32 3, i1 false)
7882

7983
; non-const value and size
8084
tail call void @llvm.memset.p0i8.i32(i8* align 4 %x.bc, i8 %v, i32 %s1, i1 false)
8185
; CHECK-LLVM: call void @llvm.memset.p0i8.i32(i8* %x.bc, i8 %v, i32 %s1, i1 false)
86+
; CHECK-LLVM-OPAQUE: call void @llvm.memset.p0.i32(ptr %x.bc, i8 %v, i32 %s1, i1 false)
8287

8388
; Address spaces, non-const value and size
8489
%a = addrspacecast i8 addrspace(4)* %1 to i8 addrspace(3)*
8590
tail call void @llvm.memset.p3i8.i32(i8 addrspace(3)* align 4 %a, i8 %v, i32 %s1, i1 false)
8691
; CHECK-LLVM: call void @llvm.memset.p3i8.i32(i8 addrspace(3)* %a, i8 %v, i32 %s1, i1 false)
92+
; CHECK-LLVM-OPAQUE: call void @llvm.memset.p3.i32(ptr addrspace(3) %a, i8 %v, i32 %s1, i1 false)
8793
%b = addrspacecast i8 addrspace(4)* %1 to i8 addrspace(1)*
8894
tail call void @llvm.memset.p1i8.i64(i8 addrspace(1)* align 4 %b, i8 %v, i64 %s2, i1 false)
8995
; CHECK-LLVM: call void @llvm.memset.p1i8.i64(i8 addrspace(1)* %b, i8 %v, i64 %s2, i1 false)
96+
; CHECK-LLVM-OPAQUE: call void @llvm.memset.p1.i64(ptr addrspace(1) %b, i8 %v, i64 %s2, i1 false)
9097

9198
; Volatile
9299
tail call void @llvm.memset.p1i8.i64(i8 addrspace(1)* align 4 %b, i8 %v, i64 %s2, i1 true)
93100
; CHECK-LLVM: call void @llvm.memset.p1i8.i64(i8 addrspace(1)* %b, i8 %v, i64 %s2, i1 true)
101+
; CHECK-LLVM-OPAQUE: call void @llvm.memset.p1.i64(ptr addrspace(1) %b, i8 %v, i64 %s2, i1 true)
94102
ret void
95103
}
96104

0 commit comments

Comments
 (0)