Skip to content

Commit 1f6b16c

Browse files
haonanya1haonanya1
authored
Merge pull request #282 from haonanya/atomic_half_120
Update cl_ext_float_atomics patch
2 parents 6d0c9e3 + 7a7abff commit 1f6b16c

File tree

1 file changed

+33
-20
lines changed

1 file changed

+33
-20
lines changed

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

Lines changed: 33 additions & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
From 8a7885884d8b7074f716332768ed957d849b9a72 Mon Sep 17 00:00:00 2001
1+
From cc687f3c2399b83da0156ff24d09c24bab31e74e Mon Sep 17 00:00:00 2001
22
From: haonanya <[email protected]>
33
Date: Wed, 28 Jul 2021 14:24:23 +0800
44
Subject: [PATCH] Add support for cl_ext_float_atomics in SPIRVWriter
@@ -7,7 +7,7 @@ Signed-off-by: haonanya <[email protected]>
77
---
88
include/LLVMSPIRVExtensions.inc | 1 +
99
lib/SPIRV/OCLToSPIRV.cpp | 27 +++++-
10-
lib/SPIRV/OCLUtil.cpp | 19 ++--
10+
lib/SPIRV/OCLUtil.cpp | 15 ++--
1111
lib/SPIRV/SPIRVToOCL.h | 3 +
1212
lib/SPIRV/SPIRVToOCL12.cpp | 21 +++++
1313
lib/SPIRV/SPIRVToOCL20.cpp | 28 +++++-
@@ -16,7 +16,7 @@ Signed-off-by: haonanya <[email protected]>
1616
lib/SPIRV/libSPIRV/SPIRVOpCode.h | 8 +-
1717
lib/SPIRV/libSPIRV/SPIRVOpCodeEnum.h | 2 +
1818
lib/SPIRV/libSPIRV/spirv.hpp | 7 ++
19-
test/AtomicBuiltinsFloat.ll | 79 ++++++++++++++++
19+
test/AtomicBuiltinsFloat.ll | 94 +++++++++++++++++++
2020
test/AtomicFAddEXT.ll | 72 +++++++++++++++
2121
test/AtomicFAddEXTForOCL.ll | 84 +++++++++++++++++
2222
test/AtomicFAddExt.ll | 119 -------------------------
@@ -25,7 +25,7 @@ Signed-off-by: haonanya <[email protected]>
2525
test/AtomicFMinEXT.ll | 73 +++++++++++++++
2626
test/AtomicFMinEXTForOCL.ll | 81 +++++++++++++++++
2727
test/negative/InvalidAtomicBuiltins.cl | 18 +---
28-
20 files changed, 675 insertions(+), 148 deletions(-)
28+
20 files changed, 688 insertions(+), 146 deletions(-)
2929
create mode 100644 test/AtomicBuiltinsFloat.ll
3030
create mode 100644 test/AtomicFAddEXT.ll
3131
create mode 100644 test/AtomicFAddEXTForOCL.ll
@@ -100,30 +100,28 @@ index 7c65b9e8..7ea350ff 100644
100100
&Attrs);
101101
}
102102
diff --git a/lib/SPIRV/OCLUtil.cpp b/lib/SPIRV/OCLUtil.cpp
103-
index 2cc5d815..e4d7a7cf 100644
103+
index 2cc5d815..89ae7fe7 100644
104104
--- a/lib/SPIRV/OCLUtil.cpp
105105
+++ b/lib/SPIRV/OCLUtil.cpp
106106
@@ -655,29 +655,32 @@ size_t getSPIRVAtomicBuiltinNumMemoryOrderArgs(Op OC) {
107107
return 1;
108108
}
109109

110-
+// atomic_fetch_[add, sub, min, max] and atomic_fetch_[add, sub, min,
111-
+// max]_explicit functions are defined on OpenCL headers, they are not
112-
+// translated to function call
110+
+// atomic_fetch_[add, min, max] and atomic_fetch_[add, min, max]_explicit
111+
+// functions declared in clang headers should be translated to corresponding
112+
+// FP-typed Atomic Instructions
113113
bool isComputeAtomicOCLBuiltin(StringRef DemangledName) {
114114
if (!DemangledName.startswith(kOCLBuiltinName::AtomicPrefix) &&
115115
!DemangledName.startswith(kOCLBuiltinName::AtomPrefix))
116116
return false;
117117

118118
return llvm::StringSwitch<bool>(DemangledName)
119119
- .EndsWith("add", true)
120-
- .EndsWith("sub", true)
120+
.EndsWith("sub", true)
121121
+ .EndsWith("atomic_add", true)
122-
+ .EndsWith("atomic_sub", true)
123122
+ .EndsWith("atomic_min", true)
124123
+ .EndsWith("atomic_max", true)
125124
+ .EndsWith("atom_add", true)
126-
+ .EndsWith("atom_sub", true)
127125
+ .EndsWith("atom_min", true)
128126
+ .EndsWith("atom_max", true)
129127
.EndsWith("inc", true)
@@ -135,7 +133,7 @@ index 2cc5d815..e4d7a7cf 100644
135133
.EndsWith("or", true)
136134
.EndsWith("xor", true)
137135
- .EndsWith("add_explicit", true)
138-
- .EndsWith("sub_explicit", true)
136+
.EndsWith("sub_explicit", true)
139137
.EndsWith("or_explicit", true)
140138
.EndsWith("xor_explicit", true)
141139
.EndsWith("and_explicit", true)
@@ -385,11 +383,13 @@ index f0e311c6..2a86f32e 100644
385383
case OpVmeImageINTEL: *hasResult = true; *hasResultType = true; break;
386384
diff --git a/test/AtomicBuiltinsFloat.ll b/test/AtomicBuiltinsFloat.ll
387385
new file mode 100644
388-
index 00000000..d9300558
386+
index 00000000..778c0cb0
389387
--- /dev/null
390388
+++ b/test/AtomicBuiltinsFloat.ll
391-
@@ -0,0 +1,79 @@
389+
@@ -0,0 +1,94 @@
392390
+; Check that translator generate atomic instructions for atomic builtins
391+
+; FP-typed atomic_fetch_sub and atomic_fetch_sub_explicit should be translated
392+
+; to FunctionCall
393393
+; RUN: llvm-as %s -o %t.bc
394394
+; RUN: llvm-spirv %t.bc -spirv-text -o - | FileCheck %s
395395
+; RUN: llvm-spirv %t.bc -o %t.spv
@@ -400,12 +400,13 @@ index 00000000..d9300558
400400
+; CHECK-COUNT-3: AtomicStore
401401
+; CHECK-COUNT-3: AtomicLoad
402402
+; CHECK-COUNT-3: AtomicExchange
403+
+; CHECK-COUNT-3: FunctionCall
403404
+
404405
+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"
405406
+target triple = "spir-unknown-unknown"
406407
+
407408
+; Function Attrs: convergent norecurse nounwind
408-
+define dso_local spir_kernel void @test_atomic_kernel(float addrspace(3)* %ff, float addrspace(3)* nocapture readnone %a) local_unnamed_addr #0 !kernel_arg_addr_space !3 !kernel_arg_access_qual !4 !kernel_arg_type !5 !kernel_arg_base_type !6 !kernel_arg_type_qual !7 {
409+
+define dso_local spir_kernel void @test_atomic_kernel(float addrspace(3)* %ff) local_unnamed_addr #0 !kernel_arg_addr_space !3 !kernel_arg_access_qual !4 !kernel_arg_type !5 !kernel_arg_base_type !6 !kernel_arg_type_qual !7 {
409410
+entry:
410411
+ %0 = addrspacecast float addrspace(3)* %ff to float addrspace(4)*
411412
+ tail call spir_func void @_Z11atomic_initPU3AS4VU7_Atomicff(float addrspace(4)* %0, float 1.000000e+00) #2
@@ -418,6 +419,9 @@ index 00000000..d9300558
418419
+ %call3 = tail call spir_func float @_Z15atomic_exchangePU3AS4VU7_Atomicff(float addrspace(4)* %0, float 1.000000e+00) #2
419420
+ %call4 = tail call spir_func float @_Z24atomic_exchange_explicitPU3AS4VU7_Atomicff12memory_order(float addrspace(4)* %0, float 1.000000e+00, i32 0) #2
420421
+ %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
422+
+ %call6 = tail call spir_func float @_Z16atomic_fetch_subPU3AS3VU7_Atomicff(float addrspace(3)* %ff, float 1.000000e+00) #2
423+
+ %call7 = tail call spir_func float @_Z25atomic_fetch_sub_explicitPU3AS3VU7_Atomicff12memory_order(float addrspace(3)* %ff, float 1.000000e+00, i32 0) #2
424+
+ %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
421425
+ ret void
422426
+}
423427
+
@@ -451,6 +455,15 @@ index 00000000..d9300558
451455
+; Function Attrs: convergent
452456
+declare spir_func float @_Z24atomic_exchange_explicitPU3AS4VU7_Atomicff12memory_order12memory_scope(float addrspace(4)*, float, i32, i32) local_unnamed_addr #1
453457
+
458+
+; Function Attrs: convergent
459+
+declare spir_func float @_Z16atomic_fetch_subPU3AS3VU7_Atomicff(float addrspace(3)*, float) local_unnamed_addr #1
460+
+
461+
+; Function Attrs: convergent
462+
+declare spir_func float @_Z25atomic_fetch_sub_explicitPU3AS3VU7_Atomicff12memory_order(float addrspace(3)*, float, i32) local_unnamed_addr #1
463+
+
464+
+; Function Attrs: convergent
465+
+declare spir_func float @_Z25atomic_fetch_sub_explicitPU3AS3VU7_Atomicff12memory_order12memory_scope(float addrspace(3)*, float, i32, i32) local_unnamed_addr #1
466+
+
454467
+attributes #0 = { convergent norecurse nounwind "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"="true" "stack-protector-buffer-size"="8" "uniform-work-group-size"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
455468
+attributes #1 = { convergent "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"="true" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
456469
+attributes #2 = { convergent nounwind }
@@ -463,11 +476,11 @@ index 00000000..d9300558
463476
+!0 = !{i32 1, !"wchar_size", i32 4}
464477
+!1 = !{i32 2, i32 0}
465478
+!2 = !{!"clang version 12.0.1 (https://github.com/llvm/llvm-project.git 23fe7b104a0adaaaecd52108105f49297c420c9b)"}
466-
+!3 = !{i32 3, i32 3}
467-
+!4 = !{!"none", !"none"}
468-
+!5 = !{!"atomic_float*", !"float*"}
469-
+!6 = !{!"_Atomic(float)*", !"float*"}
470-
+!7 = !{!"volatile", !""}
479+
+!3 = !{i32 3}
480+
+!4 = !{!"none"}
481+
+!5 = !{!"atomic_float*"}
482+
+!6 = !{!"_Atomic(float)*"}
483+
+!7 = !{!"volatile"}
471484
diff --git a/test/AtomicFAddEXT.ll b/test/AtomicFAddEXT.ll
472485
new file mode 100644
473486
index 00000000..b012c904

0 commit comments

Comments
 (0)