Skip to content

Commit 4393ffb

Browse files
haonanya1haonanya1
authored
Merge pull request #279 from haonanya/atomic_half_100
Update cl_ext_float_atomics patch
2 parents 0eec88e + 6c9040c commit 4393ffb

File tree

1 file changed

+33
-21
lines changed

1 file changed

+33
-21
lines changed

‎patches/spirv/0001-Add-support-for-cl_ext_float_atomics-in-SPIRVWriter.patch

Lines changed: 33 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -1,34 +1,33 @@
1-
From 794c7d32e521060190bbfca7c52f0bb4fa49f2f2 Mon Sep 17 00:00:00 2001
1+
From 89e3be862e9a389ba858faa1270f0f8856b37bcf Mon Sep 17 00:00:00 2001
22
From: haonanya <[email protected]>
33
Date: Mon, 19 Jul 2021 10:14:20 +0800
44
Subject: [PATCH] Add support for cl_ext_float_atomics in SPIRVWriter
55

66
Signed-off-by: haonanya <[email protected]>
7-
Signed-off-by: Haonan Yang <[email protected]>
87
---
98
lib/SPIRV/OCL20ToSPIRV.cpp | 26 +++++-
10-
lib/SPIRV/OCLUtil.cpp | 19 +++--
9+
lib/SPIRV/OCLUtil.cpp | 15 ++--
1110
lib/SPIRV/SPIRVToOCL.h | 3 +
1211
lib/SPIRV/SPIRVToOCL12.cpp | 21 +++++
1312
lib/SPIRV/SPIRVToOCL20.cpp | 28 ++++++-
1413
lib/SPIRV/libSPIRV/SPIRVNameMapEnum.h | 1 -
1514
lib/SPIRV/libSPIRV/SPIRVOpCode.h | 8 +-
16-
test/AtomicBuiltinsFloat.ll | 79 ++++++++++++++++++
15+
test/AtomicBuiltinsFloat.ll | 94 +++++++++++++++++++++
1716
test/AtomicFAddEXTForOCL.ll | 88 ++++++++++++++++++++
1817
test/AtomicFAddExt.ll | 111 ++++++++-----------------
1918
test/AtomicFMaxEXT.ll | 113 +++++++-------------------
2019
test/AtomicFMaxEXTForOCL.ll | 88 ++++++++++++++++++++
2120
test/AtomicFMinEXT.ll | 113 +++++++-------------------
2221
test/AtomicFMinEXTForOCL.ll | 85 +++++++++++++++++++
2322
test/InvalidAtomicBuiltins.cl | 16 ----
24-
15 files changed, 526 insertions(+), 273 deletions(-)
23+
15 files changed, 539 insertions(+), 271 deletions(-)
2524
create mode 100644 test/AtomicBuiltinsFloat.ll
2625
create mode 100644 test/AtomicFAddEXTForOCL.ll
2726
create mode 100644 test/AtomicFMaxEXTForOCL.ll
2827
create mode 100644 test/AtomicFMinEXTForOCL.ll
2928

3029
diff --git a/lib/SPIRV/OCL20ToSPIRV.cpp b/lib/SPIRV/OCL20ToSPIRV.cpp
31-
index e30aa5be..98d4289e 100644
30+
index e30aa5be..79323de2 100644
3231
--- a/lib/SPIRV/OCL20ToSPIRV.cpp
3332
+++ b/lib/SPIRV/OCL20ToSPIRV.cpp
3433
@@ -407,7 +407,6 @@ void OCL20ToSPIRV::visitCallInst(CallInst &CI) {
@@ -53,7 +52,7 @@ index e30aa5be..98d4289e 100644
5352
Args.end() - Offset);
5453
}
5554
- return getSPIRVFuncName(OCLSPIRVBuiltinMap::map(Info.UniqName));
56-
+ llvm::Type* AtomicBuiltinsReturnType =
55+
+ llvm::Type* AtomicBuiltinsReturnType =
5756
+ CI->getCalledFunction()->getReturnType();
5857
+ auto IsFPType = [](llvm::Type *ReturnType) {
5958
+ return ReturnType->isHalfTy() || ReturnType->isFloatTy() ||
@@ -79,30 +78,28 @@ index e30aa5be..98d4289e 100644
7978
&Attrs);
8079
}
8180
diff --git a/lib/SPIRV/OCLUtil.cpp b/lib/SPIRV/OCLUtil.cpp
82-
index 992f173f..539c196c 100644
81+
index 992f173f..8ae882c6 100644
8382
--- a/lib/SPIRV/OCLUtil.cpp
8483
+++ b/lib/SPIRV/OCLUtil.cpp
8584
@@ -120,29 +120,32 @@ size_t getSPIRVAtomicBuiltinNumMemoryOrderArgs(Op OC) {
8685
return 1;
8786
}
8887

89-
+// atomic_fetch_[add, sub, min, max] and atomic_fetch_[add, sub, min,
90-
+// max]_explicit functions are defined on OpenCL headers, they are not
91-
+// translated to function call
88+
+// atomic_fetch_[add, min, max] and atomic_fetch_[add, min, max]_explicit
89+
+// functions declared in clang headers should be translated to corresponding
90+
+// FP-typed Atomic Instructions
9291
bool isComputeAtomicOCLBuiltin(StringRef DemangledName) {
9392
if (!DemangledName.startswith(kOCLBuiltinName::AtomicPrefix) &&
9493
!DemangledName.startswith(kOCLBuiltinName::AtomPrefix))
9594
return false;
9695

9796
return llvm::StringSwitch<bool>(DemangledName)
9897
- .EndsWith("add", true)
99-
- .EndsWith("sub", true)
98+
.EndsWith("sub", true)
10099
+ .EndsWith("atomic_add", true)
101-
+ .EndsWith("atomic_sub", true)
102100
+ .EndsWith("atomic_min", true)
103101
+ .EndsWith("atomic_max", true)
104102
+ .EndsWith("atom_add", true)
105-
+ .EndsWith("atom_sub", true)
106103
+ .EndsWith("atom_min", true)
107104
+ .EndsWith("atom_max", true)
108105
.EndsWith("inc", true)
@@ -114,7 +111,7 @@ index 992f173f..539c196c 100644
114111
.EndsWith("or", true)
115112
.EndsWith("xor", true)
116113
- .EndsWith("add_explicit", true)
117-
- .EndsWith("sub_explicit", true)
114+
.EndsWith("sub_explicit", true)
118115
.EndsWith("or_explicit", true)
119116
.EndsWith("xor_explicit", true)
120117
.EndsWith("and_explicit", true)
@@ -184,7 +181,7 @@ index 1a62c6b8..dc0ba9cc 100644
184181
}
185182

186183
diff --git a/lib/SPIRV/SPIRVToOCL20.cpp b/lib/SPIRV/SPIRVToOCL20.cpp
187-
index 8c437858..0033b853 100644
184+
index 8c437858..07612ce6 100644
188185
--- a/lib/SPIRV/SPIRVToOCL20.cpp
189186
+++ b/lib/SPIRV/SPIRVToOCL20.cpp
190187
@@ -82,6 +82,9 @@ public:
@@ -232,8 +229,8 @@ index 8c437858..0033b853 100644
232229
}
233230
auto Ptr = findFirstPtr(Args);
234231
- auto Name = OCLSPIRVBuiltinMap::rmap(OC);
235-
+ std::string Name;
236-
+ // Map fp atomic instructions to regular OpenCL built-ins.
232+
+ std::string Name;
233+
+ // Map fp atomic instructions to regular OpenCL built-ins.
237234
+ if (isFPAtomicOpCode(OC))
238235
+ Name = mapFPAtomicName(OC);
239236
+ else
@@ -278,11 +275,13 @@ index feec70f6..8e595e83 100644
278275
return ((unsigned)OpCode >= OpIAdd && (unsigned)OpCode <= OpFMod) ||
279276
diff --git a/test/AtomicBuiltinsFloat.ll b/test/AtomicBuiltinsFloat.ll
280277
new file mode 100644
281-
index 00000000..18ee1c86
278+
index 00000000..b688cb2a
282279
--- /dev/null
283280
+++ b/test/AtomicBuiltinsFloat.ll
284-
@@ -0,0 +1,79 @@
281+
@@ -0,0 +1,94 @@
285282
+; Check that translator generate atomic instructions for atomic builtins
283+
+; FP-typed atomic_fetch_sub and atomic_fetch_sub_explicit should be translated
284+
+; to FunctionCall
286285
+; RUN: llvm-as %s -o %t.bc
287286
+; RUN: llvm-spirv %t.bc -spirv-text -o - | FileCheck %s
288287
+; RUN: llvm-spirv %t.bc -o %t.spv
@@ -293,6 +292,7 @@ index 00000000..18ee1c86
293292
+; CHECK-COUNT-3: AtomicStore
294293
+; CHECK-COUNT-3: AtomicLoad
295294
+; CHECK-COUNT-3: AtomicExchange
295+
+; CHECK-COUNT-3: FunctionCall
296296
+
297297
+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"
298298
+target triple = "spir-unknown-unknown"
@@ -311,6 +311,9 @@ index 00000000..18ee1c86
311311
+ %call3 = tail call spir_func float @_Z15atomic_exchangePU3AS4VU7_Atomicff(float addrspace(4)* %0, float 1.000000e+00) #2
312312
+ %call4 = tail call spir_func float @_Z24atomic_exchange_explicitPU3AS4VU7_Atomicff12memory_order(float addrspace(4)* %0, float 1.000000e+00, i32 0) #2
313313
+ %call5 = tail call spir_func float @_Z24atomic_exchange_explicitPU3AS4VU7_Atomicff12memory_order12memory_scope(float addrspace(4)* %0, float 1.000000e+00, i32 0, i32 1) #2
314+
+ %call6 = tail call spir_func float @_Z16atomic_fetch_subPU3AS3VU7_Atomicff(float addrspace(3)* %ff, float 1.000000e+00) #2
315+
+ %call7 = tail call spir_func float @_Z25atomic_fetch_sub_explicitPU3AS3VU7_Atomicff12memory_order(float addrspace(3)* %ff, float 1.000000e+00, i32 0) #2
316+
+ %call8 = tail call spir_func float @_Z25atomic_fetch_sub_explicitPU3AS3VU7_Atomicff12memory_order12memory_scope(float addrspace(3)* %ff, float 1.000000e+00, i32 0, i32 1) #2
314317
+ ret void
315318
+}
316319
+
@@ -344,6 +347,15 @@ index 00000000..18ee1c86
344347
+; Function Attrs: convergent
345348
+declare spir_func float @_Z24atomic_exchange_explicitPU3AS4VU7_Atomicff12memory_order12memory_scope(float addrspace(4)*, float, i32, i32) local_unnamed_addr #1
346349
+
350+
+; Function Attrs: convergent
351+
+declare spir_func float @_Z16atomic_fetch_subPU3AS3VU7_Atomicff(float addrspace(3)*, float) local_unnamed_addr #1
352+
+
353+
+; Function Attrs: convergent
354+
+declare spir_func float @_Z25atomic_fetch_sub_explicitPU3AS3VU7_Atomicff12memory_order(float addrspace(3)*, float, i32) local_unnamed_addr #1
355+
+
356+
+; Function Attrs: convergent
357+
+declare spir_func float @_Z25atomic_fetch_sub_explicitPU3AS3VU7_Atomicff12memory_order12memory_scope(float addrspace(3)*, float, i32, i32) local_unnamed_addr #1
358+
+
347359
+attributes #0 = { convergent nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "denorms-are-zero"="false" "disable-tail-calls"="false" "frame-pointer"="none" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "uniform-work-group-size"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
348360
+attributes #1 = { convergent "correctly-rounded-divide-sqrt-fp-math"="false" "denorms-are-zero"="false" "disable-tail-calls"="false" "frame-pointer"="none" "less-precise-fpmad"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
349361
+attributes #2 = { convergent nounwind }
@@ -355,7 +367,7 @@ index 00000000..18ee1c86
355367
+
356368
+!0 = !{i32 1, !"wchar_size", i32 4}
357369
+!1 = !{i32 2, i32 0}
358-
+!2 = !{!"clang version 10.0.1 (8560093eba963fba2edd47ca85404cdaff22f174)"}
370+
+!2 = !{!"clang version 10.0.1 (0d3ce3267dd78b77ab8f302347cc547afaf43ede)"}
359371
+!3 = !{i32 3}
360372
+!4 = !{!"none"}
361373
+!5 = !{!"atomic_float*"}

0 commit comments

Comments
 (0)