Modify names of SYCL/ESIMD shared functions before linkage, add test.

kbobrovs · kbobrovs · commit 44cbb8ea27d2 · 2022-05-19T15:43:16.000-07:00
Signed-off-by: Konstantin S Bobrovsky &lt;konstantin.s.bobrovsky@intel.com&gt;
diff --git a/llvm/test/tools/sycl-post-link/sycl-esimd/no-sycl-esimd-split-shared-func.ll b/llvm/test/tools/sycl-post-link/sycl-esimd/no-sycl-esimd-split-shared-func.ll
@@ -0,0 +1,107 @@
+; This test checks "integrated" invoke_simd support by the sycl-post-link tool:
+; - data flow analysis is able to determine actual target of
+;   _Z33__regcall3____builtin_invoke_simd*
+; - functions shared by SYCL and ESIMD callgraphs get cloned and renamed, thus
+;   making sure no functions are shared by the callgraphs (currently required by
+;   IGC)
+
+; RUN: sycl-post-link -lower-esimd -symbols -split=auto -S %s -o %t.table
+; RUN: FileCheck %s -input-file=%t.table
+; RUN: FileCheck %s -input-file=%t_0.ll --check-prefixes CHECK-IR
+; RUN: FileCheck %s -input-file=%t_0.sym --check-prefixes CHECK-SYM
+
+;------------- Verify generated table file.
+; CHECK: [Code|Properties|Symbols]
+; CHECK: {{.*}}_0.ll|{{.*}}_0.prop|{{.*}}_0.sym
+; CHECK-EMPTY:
+
+;------------- Verify generated symbol file.
+; CHECK-SYM: SYCL_kernel
+; CHECK-SYM: _SIMD_CALLEE
+; CHECK-SYM: ESIMD_kernel
+; CHECK-SYM-EMPTY:
+
+target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64"
+target triple = "spir64-unknown-unknown"
+
+; Function Attrs: convergent
+
+declare dso_local spir_func <4 x float> @__intrin(i64)
+
+define dso_local spir_func <4 x float> @block_read(i64 %addr) noinline {
+;------------- This function is participates in both SYCL and ESIMD callgraphs -
+;------------- check that it is duplicated:
+; CHECK-IR-DAG: define dso_local spir_func <4 x float> @block_read.esimd
+  %res = call spir_func <4 x float> @__intrin(i64 %addr)
+  ret <4 x float> %res
+}
+
+; Function Attrs: convergent
+declare dso_local spir_func float @_Z33__regcall3____builtin_invoke_simdXX(<4 x float> (float addrspace(4)*, <4 x float>, i32)*, float addrspace(4)*, float, i32) local_unnamed_addr
+
+; Function Attrs: convergent mustprogress noinline norecurse optnone
+define dso_local x86_regcallcc <4 x float> @_SIMD_CALLEE(float addrspace(4)* %A, <4 x float> %non_uni_val, i32 %uni_val) #0 !sycl_explicit_simd !0 !intel_reqd_sub_group_size !0 {
+; Verify that correct attributes are attached to the function:
+; CHECK-IR-DAG: {{.*}} @_SIMD_CALLEE(float addrspace(4)* %A, <4 x float> %non_uni_val, i32 %uni_val) #[[ATTR1:[0-9]+]]
+entry:
+  %AA = ptrtoint float addrspace(4)* %A to i64
+  %ii = zext i32 %uni_val to i64
+  %addr = add nuw nsw i64 %ii, %AA
+  %data = call spir_func <4 x float> @block_read(i64 %addr)
+  %add = fadd <4 x float> %non_uni_val, %data
+  ret <4 x float> %add
+}
+
+define internal spir_func float @foo(float addrspace(1)* %ptr) align 2 {
+entry:
+;------------- Typical data flow of the @_SIMD_CALLEE function address in worst
+;------------- case (-O0), when invoke_simd uses function name:
+;------------- float res = invoke_simd(sg, SIMD_CALLEE, uniform{ A }, x, uniform{ y });
+  %f.addr.i = alloca <4 x float> (float addrspace(4)*, <4 x float>, i32)*, align 8
+  %f.addr.ascast.i = addrspacecast <4 x float> (float addrspace(4)*, <4 x float>, i32)** %f.addr.i to <4 x float> (float addrspace(4)*, <4 x float>, i32)* addrspace(4)*
+  store <4 x float> (float addrspace(4)*, <4 x float>, i32)* @_SIMD_CALLEE, <4 x float> (float addrspace(4)*, <4 x float>, i32)* addrspace(4)* %f.addr.ascast.i, align 8
+  %FUNC_PTR = load <4 x float> (float addrspace(4)*, <4 x float>, i32)*, <4 x float> (float addrspace(4)*, <4 x float>, i32)* addrspace(4)* %f.addr.ascast.i, align 8
+
+;------------- Data flow for the parameters of SIMD_CALLEE
+  %param_A = addrspacecast float addrspace(1)* %ptr to float addrspace(4)*
+  %param_non_uni_val = load float, float addrspace(4)* %param_A, align 4
+
+;------------- The invoke_simd calls.
+  %res1 = call spir_func float @_Z33__regcall3____builtin_invoke_simdXX(<4 x float> (float addrspace(4)*, <4 x float>, i32)* %FUNC_PTR, float addrspace(4)* %param_A, float %param_non_uni_val, i32 10)
+; Verify that %FUNC_PTR is replaced with @_SIMD_CALLEE:
+;  CHECK-IR-DAG: %{{.*}} = call spir_func float @_Z33__regcall3____builtin_invoke_simdXX(<4 x float> (float addrspace(4)*, <4 x float>, i32)* @_SIMD_CALLEE, float addrspace(4)* %param_A, float %param_non_uni_val, i32 10)
+  ret float %res1
+}
+
+;------------- SYCL kernel, an entry point
+define dso_local spir_kernel void @SYCL_kernel(float addrspace(1)* %ptr) #2 {
+entry:
+  %res1 = call spir_func float @foo(float addrspace(1)* %ptr)
+  %ptri = ptrtoint float addrspace(1)* %ptr to i64
+  %vec = call spir_func <4 x float> @block_read(i64 %ptri)
+  %res2 = extractelement <4 x float> %vec, i32 0
+  %res = fadd float %res1, %res2
+  store float %res, float addrspace(1)* %ptr
+  ret void
+}
+
+;------------- ESIMD kernel, an entry point
+define dso_local spir_kernel void @ESIMD_kernel(float addrspace(1)* %ptr) #3 !sycl_explicit_simd !0 {
+entry:
+  %ptr_as4 = addrspacecast float addrspace(1)* %ptr to float addrspace(4)*
+  %res = call x86_regcallcc <4 x float> @_SIMD_CALLEE(float addrspace(4)* %ptr_as4, <4 x float> <float 1.0, float 1.0, float 1.0, float 1.0>, i32 0)
+  %ptr_x16 = bitcast float addrspace(4)* %ptr_as4 to <4 x float> addrspace(4)*
+  store <4 x float> %res, <4 x float> addrspace(4)* %ptr_x16
+  ret void
+}
+
+
+; Check that VCStackCall attribute is added to the invoke_simd target functions:
+attributes #0 = { convergent mustprogress norecurse "sycl-module-id"="invoke_simd.cpp" }
+; CHECK-IR-DAG: attributes #[[ATTR1]] = { {{.*}} "VCStackCall" {{.*}}"sycl-module-id"="invoke_simd.cpp" }
+
+attributes #2 = { "sycl-module-id"="a.cpp" }
+attributes #3 = { "sycl-module-id"="b.cpp" }
+
+
+!0 = !{}
diff --git a/llvm/test/tools/sycl-post-link/sycl-esimd/sycl-esimd-split-per-source.ll b/llvm/test/tools/sycl-post-link/sycl-esimd/sycl-esimd-split-per-source.ll
@@ -1,9 +1,9 @@
 ; RUN: sycl-post-link -split-esimd -split=source -S %s -o %t.table
 ; RUN: FileCheck %s -input-file=%t.table
-; RxUN: FileCheck %s -input-file=%t_0.ll --check-prefixes CHECK-SYCL-IR-0
-; RxUN: FileCheck %s -input-file=%t_1.ll --check-prefixes CHECK-SYCL-IR-1
-; RxUN: FileCheck %s -input-file=%t_esimd_0.ll --check-prefixes CHECK-ESIMD-IR-0
-; RxUN: FileCheck %s -input-file=%t_esimd_1.ll --check-prefixes CHECK-ESIMD-IR-1
+; RUN: FileCheck %s -input-file=%t_0.ll --check-prefixes CHECK-SYCL-IR-0
+; RUN: FileCheck %s -input-file=%t_1.ll --check-prefixes CHECK-SYCL-IR-1
+; RUN: FileCheck %s -input-file=%t_esimd_0.ll --check-prefixes CHECK-ESIMD-IR-0
+; RUN: FileCheck %s -input-file=%t_esimd_1.ll --check-prefixes CHECK-ESIMD-IR-1
 
 ; This test checks that after we split SYCL and ESIMD kernels into
 ; separate modules, we split those two modules further according to
@@ -71,16 +71,16 @@ attributes #1 = { "sycl-module-id"="b.cpp" }
 ; CHECK-DAG: {{.*}}tmp_esimd_0.ll|{{.*}}_esimd_0.prop
 ; CHECK-DAG: {{.*}}tmp_esimd_1.ll|{{.*}}_esimd_1.prop
 
-; CHECK-SYCL-IR-0-DAG: define dso_local spir_kernel void @SYCL_kernel1()
-; CHECK-SYCL-IR-0-DAG: define dso_local spir_kernel void @SYCL_kernel2()
-; CHECK-SYCL-IR-0-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
-
-; CHECK-SYCL-IR-1-DAG: define dso_local spir_kernel void @SYCL_kernel3()
+; CHECK-SYCL-IR-1-DAG: define dso_local spir_kernel void @SYCL_kernel1()
+; CHECK-SYCL-IR-1-DAG: define dso_local spir_kernel void @SYCL_kernel2()
 ; CHECK-SYCL-IR-1-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
 
-; CHECK-ESIMD-IR-0-DAG: define dso_local spir_kernel void @ESIMD_kernel1()
-; CHECK-ESIMD-IR-0-DAG: define dso_local spir_kernel void @ESIMD_kernel2()
-; CHECK-ESIMD-IR-0-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+; CHECK-SYCL-IR-0-DAG: define dso_local spir_kernel void @SYCL_kernel3()
+; CHECK-SYCL-IR-0-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
 
-; CHECK-ESIMD-IR-1-DAG: define dso_local spir_kernel void @ESIMD_kernel3()
+; CHECK-ESIMD-IR-1-DAG: define dso_local spir_kernel void @ESIMD_kernel1()
+; CHECK-ESIMD-IR-1-DAG: define dso_local spir_kernel void @ESIMD_kernel2()
 ; CHECK-ESIMD-IR-1-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+
+; CHECK-ESIMD-IR-0-DAG: define dso_local spir_kernel void @ESIMD_kernel3()
+; CHECK-ESIMD-IR-0-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
diff --git a/llvm/tools/sycl-post-link/ModuleSplitter.cpp b/llvm/tools/sycl-post-link/ModuleSplitter.cpp
@@ -472,9 +472,39 @@ void dumpEntryPoints(const Module &M, bool OnlyKernelsAreEntryPoints,
   llvm::errs() << "}\n";
 }
 
-// Updates Props.HasEsimd to ESIMDStatus::ESIMD_ONLY/SYCL_ONLY if this module
-// descriptor is a ESIMD/SYCL part of the ESIMD/SYCL module split. Otherwise
-// assumes the module has both SYCL and ESIMD.
+void ModuleDesc::renameDuplicatesOf(const Module &MA, StringRef Suff) {
+  Module &MB = getModule();
+#ifndef _NDEBUG
+  DenseSet<StringRef> EntryNamesB;
+  auto It0 = entries().cbegin();
+  auto It1 = entries().cend();
+  std::for_each(It0, It1,
+                [&](const Function *F) { EntryNamesB.insert(F->getName()); });
+#endif // _NDEBUG
+  for (const GlobalObject &GoA : MA.global_objects()) {
+    if (GoA.isDeclaration()) {
+      continue;
+    }
+    StringRef Name = GoA.getName();
+    auto *F = dyn_cast<Function>(&GoA);
+    GlobalObject *GoB =
+        F ? cast_or_null<GlobalObject>(MB.getFunction(Name))
+          : cast_or_null<GlobalObject>(MB.getGlobalVariable(Name));
+
+    if (!GoB || GoB->isDeclaration()) {
+      // function or variable is not shared or is a declaration in MB
+      continue;
+    }
+#ifndef _NDEBUG
+    if (F) {
+      // this is a shared function, must not be an entry point:
+      assert(!EntryNamesB.contains(Name));
+    }
+#endif // _NDEBUG
+    // rename the global object in MB:
+    GoB->setName(Name + Suff);
+  }
+}
 
 void ModuleDesc::assignESIMDProperty() {
   if (EntryPoints.isEsimd()) {
diff --git a/llvm/tools/sycl-post-link/ModuleSplitter.h b/llvm/tools/sycl-post-link/ModuleSplitter.h
@@ -80,6 +80,8 @@ class ModuleDesc {
     rebuildEntryPoints(Names);
   }
 
+  bool isESIMD() const { return Props.HasEsimd == ESIMDStatus::ESIMD_ONLY; }
+  bool isSYCL() const { return Props.HasEsimd == ESIMDStatus::SYCL_ONLY; }
   const EntryPointVec &entries() const { return EntryPoints.Functions; }
   EntryPointVec &entries() { return EntryPoints.Functions; }
   Module &getModule() { return *M; }
@@ -101,6 +103,8 @@ class ModuleDesc {
 
   void rebuildEntryPoints(const Module &M) { EntryPoints.rebuild(M); }
 
+  void renameDuplicatesOf(const Module &M, StringRef Suff);
+
   // Updates Props.HasEsimd to ESIMDStatus::ESIMD_ONLY/SYCL_ONLY if this module
   // descriptor is a ESIMD/SYCL part of the ESIMD/SYCL module split. Otherwise
   // assumes the module has both SYCL and ESIMD.
diff --git a/llvm/tools/sycl-post-link/sycl-post-link.cpp b/llvm/tools/sycl-post-link/sycl-post-link.cpp
@@ -722,7 +722,8 @@ processInputModule(std::unique_ptr<Module> M) {
     // undergo different set of LLVMIR passes. After this they are linked back
     // together to form single module with disjoint SYCL and ESIMD call graphs
     // unless --split-esimd option is specified. The graphs become disjoint when
-    // linked back because functions shared between graphs are cloned.
+    // linked back because functions shared between graphs are cloned and
+    // renamed.
     std::unique_ptr<module_split::ModuleSplitterBase> ESIMDSplitter =
         module_split::getSplitterByKernelType(
             std::move(MDesc.releaseModulePtr()), EmitOnlyKernelsAsEntryPoints,
@@ -750,8 +751,14 @@ processInputModule(std::unique_ptr<Module> M) {
 #endif // _NDEBUG
 
     if (!SplitEsimd && (MMs.size() > 1)) {
-      // SYCL/ESIMD splitting is not requested, link back into single module
+      // SYCL/ESIMD splitting is not requested, link back into single module.
       assert(MMs.size() == 2);
+      assert(MMs[0].isESIMD() && MMs[1].isSYCL() ||
+             MMs[1].isESIMD() && MMs[0].isSYCL());
+      int ESIMDInd = MMs[0].isESIMD() ? 0 : 1;
+      int SYCLInd = MMs[0].isESIMD() ? 1 : 0;
+      // ... but before that, make sure no link conflicts will occur.
+      MMs[ESIMDInd].renameDuplicatesOf(MMs[SYCLInd].getModule(), ".esimd");
       module_split::ModuleDesc M2 = link(std::move(MMs[0]), std::move(MMs[1]));
       MMs.clear();
       MMs.emplace_back(std::move(M2));
