Improve/fix translation of SPIR-V OpControlBarier to OCL1.2

This patch features two things: - fixed translation of OpControlBarrier into OpenCL 2.0 built-ins: Scope of the operation is now respected and we can either get work_group_barrier or sub_group_barrier depending on it - improved the translation so it handles non-constant 'Memory Semantic' operand of the instruction
KhronosGroup · Apr 1, 2020 · 33d1092 · 33d1092
1 parent e4e8888
commit 33d1092
Show file tree

Hide file tree

Showing 2 changed files with 28 additions and 9 deletions.
diff --git a/lib/SPIRV/SPIRVToOCL20.cpp b/lib/SPIRV/SPIRVToOCL20.cpp
@@ -135,11 +135,23 @@ void SPIRVToOCL20::visitCallSPIRVControlBarrier(CallInst *CI) {
           return cast<ConstantInt>(Args[I])->getZExtValue();
         };
         auto ExecScope = static_cast<Scope>(GetArg(0));
-        auto MScope = static_cast<Scope>(GetArg(1));
-        auto Sema = mapSPIRVMemSemanticToOCL(GetArg(2));
+        auto MemScope = static_cast<Scope>(GetArg(1));
+
+        if (auto Arg = dyn_cast<ConstantInt>(Args[2])) {
+          auto Sema = mapSPIRVMemSemanticToOCL(Arg->getZExtValue());
+          Args[0] = getInt32(M, Sema.first);
+        } else {
+          int ClMemFenceMask = MemorySemanticsWorkgroupMemoryMask |
+                               MemorySemanticsCrossWorkgroupMemoryMask |
+                               MemorySemanticsImageMemoryMask;
+          Args[0] = getOrCreateSwitchFunc(
+              kSPIRVName::TranslateSPIRVMemFence, Args[2],
+              OCLMemFenceExtendedMap::getRMap(), true /*IsReverse*/, None, CI,
+              M, ClMemFenceMask);
+        }
+        Args[1] = getInt32(M, rmap<OCLScopeKind>(MemScope));
         Args.resize(2);
-        Args[0] = getInt32(M, Sema.first);
-        Args[1] = getInt32(M, rmap<OCLScopeKind>(MScope));
+
         return (ExecScope == ScopeWorkgroup) ? kOCLBuiltinName::WorkGroupBarrier
                                              : kOCLBuiltinName::SubGroupBarrier;
       },

diff --git a/test/barrier_explicit_arguments.spt b/test/barrier_explicit_arguments.spt
@@ -4,33 +4,41 @@
 2 Capability Kernel
 5 ExtInstImport 1 "OpenCL.std"
 3 MemoryModel 1 2
-3 Source 3 102000
 4 Name 5 "test"
 3 Name 6 "val"
 4 Name 7 "entry"
 6 Decorate 5 LinkageAttributes "test" Export
 4 TypeInt 3 32 0
 4 Constant 3 8 2
 2 TypeVoid 2
+4 Constant 3 9 3
 4 TypeFunction 4 2 3
 
 5 Function 2 5 0 4
 3 FunctionParameter 3 6
 
 2 Label 7
 4 ControlBarrier 8 8 6
+4 ControlBarrier 9 9 6
 1 Return
 
 1 FunctionEnd
 
 ; RUN: llvm-spirv %s -to-binary -o %t.spv
 ; RUN: spirv-val %t.spv
-; RUN: llvm-spirv -r %t.spv -o %t.bc
-; RUN: llvm-dis < %t.bc | FileCheck %s
+; RUN: llvm-spirv -r -spirv-ocl-builtins-version=CL2.0 %t.spv -o %t.bc
+; RUN: llvm-dis < %t.bc | FileCheck %s --check-prefixes=CHECK,CHECK-20
+; RUN: llvm-spirv -r -spirv-ocl-builtins-version=CL1.2 %t.spv -o %t.bc
+; RUN: llvm-dis < %t.bc | FileCheck %s --check-prefixes=CHECK,CHECK-12
 
 ; CHECK: define spir_func void @test(i32 %val)
 ; CHECK: %call = call spir_func i32 @__translate_spirv_memory_fence(i32 %val)
-; CHECK: call spir_func void @_Z7barrierj(i32 %call)
+; CHECK-12: call spir_func void @_Z7barrierj(i32 %call)
+; There is no support for sub-groups in OpenCL 1.2, so sub-group barrier is
+; equal to regular barrier in there
+; CHECK-12: call spir_func void @_Z7barrierj(i32 %call1)
+; CHECK-20: call spir_func void @_Z18work_group_barrierj12memory_scope(i32 %call
+; CHECK-20: call spir_func void @_Z17sub_group_barrierj12memory_scope(i32 %call
 ; CHECK: define private spir_func i32 @__translate_spirv_memory_fence(i32 %key)
 ; CHECK: entry:
 ; CHECK:   %key.masked = and i32 2816, %key
@@ -60,4 +68,3 @@
 ; CHECK: case.2816:
 ; CHECK:   ret i32 7
 ; CHECK: }
-; CHECK: declare spir_func void @_Z7barrierj(i32)