Initialize mem fence passing to barrier in runtime if necessary

If a flag passed to ControlBarrier or MemoryBarrier is being initialized in runtime - don't lose it's initialization. Signed-off-by: Dmitry Sidorov <[email protected]>
KhronosGroup · Aug 1, 2019 · 7a53bfb · 7a53bfb
1 parent 4abd17a
commit 7a53bfb
Show file tree

Hide file tree

Showing 5 changed files with 190 additions and 12 deletions.
diff --git a/lib/SPIRV/OCLUtil.h b/lib/SPIRV/OCLUtil.h
@@ -66,6 +66,18 @@ enum OCLMemFenceKind {
   OCLMF_Image = 4,
 };
 
+// This enum declares extra constants for OpenCL mem_fence flag. It includes
+// combinations of local/global/image flags.
+enum OCLMemFenceExtendedKind {
+  OCLMFEx_Local = OCLMF_Local,
+  OCLMFEx_Global = OCLMF_Global,
+  OCLMFEx_Local_Global = OCLMF_Global | OCLMF_Local,
+  OCLMFEx_Image = OCLMF_Image,
+  OCLMFEx_Image_Local = OCLMF_Image | OCLMF_Local,
+  OCLMFEx_Image_Global = OCLMF_Image | OCLMF_Global,
+  OCLMFEx_Image_Local_Global = OCLMF_Image | OCLMF_Global | OCLMF_Local,
+};
+
 enum OCLScopeKind {
   OCLMS_work_item,
   OCLMS_work_group,
@@ -95,6 +107,9 @@ enum OCLMemOrderKind {
 
 typedef SPIRVMap<OCLMemFenceKind, MemorySemanticsMask> OCLMemFenceMap;
 
+typedef SPIRVMap<OCLMemFenceExtendedKind, MemorySemanticsMask>
+    OCLMemFenceExtendedMap;
+
 typedef SPIRVMap<OCLMemOrderKind, unsigned, MemorySemanticsMask> OCLMemOrderMap;
 
 typedef SPIRVMap<OCLScopeKind, Scope> OCLMemScopeMap;
@@ -425,6 +440,22 @@ template <> inline void SPIRVMap<OCLMemFenceKind, MemorySemanticsMask>::init() {
   add(OCLMF_Image, MemorySemanticsImageMemoryMask);
 }
 
+template <>
+inline void SPIRVMap<OCLMemFenceExtendedKind, MemorySemanticsMask>::init() {
+  add(OCLMFEx_Local, MemorySemanticsWorkgroupMemoryMask);
+  add(OCLMFEx_Global, MemorySemanticsCrossWorkgroupMemoryMask);
+  add(OCLMFEx_Local_Global, MemorySemanticsWorkgroupMemoryMask |
+                                MemorySemanticsCrossWorkgroupMemoryMask);
+  add(OCLMFEx_Image, MemorySemanticsImageMemoryMask);
+  add(OCLMFEx_Image_Local,
+      MemorySemanticsWorkgroupMemoryMask | MemorySemanticsImageMemoryMask);
+  add(OCLMFEx_Image_Global,
+      MemorySemanticsCrossWorkgroupMemoryMask | MemorySemanticsImageMemoryMask);
+  add(OCLMFEx_Image_Local_Global, MemorySemanticsWorkgroupMemoryMask |
+                                      MemorySemanticsCrossWorkgroupMemoryMask |
+                                      MemorySemanticsImageMemoryMask);
+}
+
 template <>
 inline void SPIRVMap<OCLMemOrderKind, unsigned, MemorySemanticsMask>::init() {
   add(OCLMO_relaxed, MemorySemanticsMaskNone);

diff --git a/lib/SPIRV/SPIRVInternal.h b/lib/SPIRV/SPIRVInternal.h
@@ -337,6 +337,7 @@ const static char TranslateOCLMemOrder[] = "__translate_ocl_memory_order";
 const static char TranslateOCLMemScope[] = "__translate_ocl_memory_scope";
 const static char TranslateSPIRVMemOrder[] = "__translate_spirv_memory_order";
 const static char TranslateSPIRVMemScope[] = "__translate_spirv_memory_scope";
+const static char TranslateSPIRVMemFence[] = "__translate_spirv_memory_fence";
 } // namespace kSPIRVName
 
 namespace kSPIRVPostfix {

diff --git a/lib/SPIRV/SPIRVToOCL12.cpp b/lib/SPIRV/SPIRVToOCL12.cpp
@@ -123,12 +123,23 @@ void SPIRVToOCL12::visitCallSPIRVMemoryBarrier(CallInst *CI) {
   mutateCallInstOCL(
       M, CI,
       [=](CallInst *, std::vector<Value *> &Args) {
-        auto GetArg = [=](unsigned I) {
-          return cast<ConstantInt>(Args[I])->getZExtValue();
-        };
-        auto Sema = mapSPIRVMemSemanticToOCL(GetArg(1));
-        Args.resize(1);
-        Args[0] = getInt32(M, Sema.first);
+        if (auto Arg = dyn_cast<ConstantInt>(Args[1])) {
+          auto Sema = mapSPIRVMemSemanticToOCL(Arg->getZExtValue());
+          Args.resize(1);
+          Args[0] = getInt32(M, Sema.first);
+        } else {
+          CallInst *TransCall = dyn_cast<CallInst>(Args[1]);
+          Function *F = TransCall ? TransCall->getCalledFunction() : nullptr;
+          if (F && F->getName().equals(kSPIRVName::TranslateOCLMemScope)) {
+            Args[0] = TransCall->getArgOperand(0);
+          } else {
+            Args[0] = getOrCreateSwitchFunc(kSPIRVName::TranslateSPIRVMemFence,
+                                            Args[1],
+                                            OCLMemFenceExtendedMap::getRMap(),
+                                            true /*IsReverse*/, None, CI, M);
+          }
+          Args.resize(1);
+        }
         return kOCLBuiltinName::MemFence;
       },
       &Attrs);
@@ -141,12 +152,23 @@ void SPIRVToOCL12::visitCallSPIRVControlBarrier(CallInst *CI) {
   mutateCallInstOCL(
       M, CI,
       [=](CallInst *, std::vector<Value *> &Args) {
-        auto GetArg = [=](unsigned I) {
-          return cast<ConstantInt>(Args[I])->getZExtValue();
-        };
-        auto Sema = mapSPIRVMemSemanticToOCL(GetArg(2));
-        Args.resize(1);
-        Args[0] = getInt32(M, Sema.first);
+        if (auto Arg = dyn_cast<ConstantInt>(Args[2])) {
+          auto Sema = mapSPIRVMemSemanticToOCL(Arg->getZExtValue());
+          Args.resize(1);
+          Args[0] = getInt32(M, Sema.first);
+        } else {
+          CallInst *TransCall = dyn_cast<CallInst>(Args[2]);
+          Function *F = TransCall ? TransCall->getCalledFunction() : nullptr;
+          if (F && F->getName().equals(kSPIRVName::TranslateOCLMemScope)) {
+            Args[0] = TransCall->getArgOperand(0);
+          } else {
+            Args[0] = getOrCreateSwitchFunc(kSPIRVName::TranslateSPIRVMemFence,
+                                            Args[2],
+                                            OCLMemFenceExtendedMap::getRMap(),
+                                            true /*IsReverse*/, None, CI, M);
+          }
+          Args.resize(1);
+        }
         return kOCLBuiltinName::Barrier;
       },
       &Attrs);

diff --git a/test/barrier_explicit_arguments.spt b/test/barrier_explicit_arguments.spt
@@ -0,0 +1,62 @@
+119734787 65536 393230 10 0
+2 Capability Addresses
+2 Capability Linkage
+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 TypeFunction 4 2 3
+
+5 Function 2 5 0 4
+3 FunctionParameter 3 6
+
+2 Label 7
+4 ControlBarrier 8 8 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
+
+; 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: define private spir_func i32 @__translate_spirv_memory_fence(i32 %key)
+; CHECK: entry:
+; CHECK:   switch i32 %key, label %default [
+; CHECK:     i32 256, label %case.256
+; CHECK:     i32 512, label %case.512
+; CHECK:     i32 768, label %case.768
+; CHECK:     i32 2048, label %case.2048
+; CHECK:     i32 2304, label %case.2304
+; CHECK:     i32 2560, label %case.2560
+; CHECK:     i32 2816, label %case.2816
+; CHECK:   ]
+; CHECK: default:
+; CHECK:   unreachable
+; CHECK: case.256:
+; CHECK:   ret i32 1
+; CHECK: case.512:
+; CHECK:   ret i32 2
+; CHECK: case.768:
+; CHECK:   ret i32 3
+; CHECK: case.2048:
+; CHECK:   ret i32 4
+; CHECK: case.2304:
+; CHECK:   ret i32 5
+; CHECK: case.2560:
+; CHECK:   ret i32 6
+; CHECK: case.2816:
+; CHECK:   ret i32 7
+; CHECK: }
+; CHECK: declare spir_func void @_Z7barrierj(i32)
diff --git a/test/mem_fence_explicit_arguments.spt b/test/mem_fence_explicit_arguments.spt
@@ -0,0 +1,62 @@
+119734787 65536 393230 10 0
+2 Capability Addresses
+2 Capability Linkage
+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 TypeFunction 4 2 3
+
+5 Function 2 5 0 4
+3 FunctionParameter 3 6
+
+2 Label 7
+3 MemoryBarrier 8 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
+
+; 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 @_Z9mem_fencej(i32 %call)
+; CHECK: define private spir_func i32 @__translate_spirv_memory_fence(i32 %key)
+; CHECK: entry:
+; CHECK:   switch i32 %key, label %default [
+; CHECK:     i32 256, label %case.256
+; CHECK:     i32 512, label %case.512
+; CHECK:     i32 768, label %case.768
+; CHECK:     i32 2048, label %case.2048
+; CHECK:     i32 2304, label %case.2304
+; CHECK:     i32 2560, label %case.2560
+; CHECK:     i32 2816, label %case.2816
+; CHECK:   ]
+; CHECK: default:
+; CHECK:   unreachable
+; CHECK: case.256:
+; CHECK:   ret i32 1
+; CHECK: case.512:
+; CHECK:   ret i32 2
+; CHECK: case.768:
+; CHECK:   ret i32 3
+; CHECK: case.2048:
+; CHECK:   ret i32 4
+; CHECK: case.2304:
+; CHECK:   ret i32 5
+; CHECK: case.2560:
+; CHECK:   ret i32 6
+; CHECK: case.2816:
+; CHECK:   ret i32 7
+; CHECK: }
+; CHECK: declare spir_func void @_Z9mem_fencej(i32)