Skip to content

[SYCL] Fix sycl-post-link to work when a spec const is referenced twice. #1524

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Apr 16, 2020
Merged

[SYCL] Fix sycl-post-link to work when a spec const is referenced twice. #1524

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 27 additions & 0 deletions llvm/test/tools/sycl-post-link/sc_sym_two_refs.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,27 @@
; This test checks that the tool does not crash and removes the unused spec
; constant global symbol when it is referenced more than once.

; RUN: sycl-post-link -spec-const=rt --ir-output-only %s -S -o - \
; RUN: | FileCheck %s

target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
target triple = "spir64-unknown-unknown-sycldevice"

%"sycl::experimental::spec_constant" = type { i8 }

@SCSymID = private unnamed_addr constant [10 x i8] c"SpecConst\00", align 1
; CHECK-NOT: @SCSymID

declare dso_local spir_func float @_Z27__sycl_getSpecConstantValueIfET_PKc(i8 addrspace(4)*)

; Function Attrs: norecurse
define weak_odr dso_local spir_kernel void @Kernel() {
%1 = call spir_func float @_Z27__sycl_getSpecConstantValueIfET_PKc(i8 addrspace(4)* addrspacecast (i8* getelementptr inbounds ([10 x i8], [10 x i8]* @SCSymID, i64 0, i64 0) to i8 addrspace(4)*))
ret void
}

; Function Attrs: norecurse
define dso_local spir_func float @foo_float(%"sycl::experimental::spec_constant" addrspace(4)* nocapture readnone dereferenceable(1) %0) local_unnamed_addr #3 {
%2 = tail call spir_func float @_Z27__sycl_getSpecConstantValueIfET_PKc(i8 addrspace(4)* addrspacecast (i8* getelementptr inbounds ([10 x i8], [10 x i8]* @SCSymID, i64 0, i64 0) to i8 addrspace(4)*))
ret float %2
}
2 changes: 1 addition & 1 deletion llvm/test/tools/sycl-post-link/spec_const_O0.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,7 @@ declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1

; Function Attrs: norecurse
define linkonce_odr dso_local spir_func zeroext i1 @FOO(%"UserSpecConstIDType" addrspace(4)* %0) comdat align 2 {
define spir_func zeroext i1 @FOO(%"UserSpecConstIDType" addrspace(4)* %0) comdat align 2 {
%2 = alloca %"UserSpecConstIDType" addrspace(4)*, align 8
%3 = alloca i8 addrspace(4)*, align 8
store %"UserSpecConstIDType" addrspace(4)* %0, %"UserSpecConstIDType" addrspace(4)** %2, align 8, !tbaa !8
Expand Down
26 changes: 26 additions & 0 deletions llvm/test/tools/sycl-post-link/spec_const_and_split.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,26 @@
; This test checks that the post-link tool works correctly when both
; device code splitting and specialization constant processing are
; requested.
;
; RUN: sycl-post-link -split=kernel -spec-const=rt -S %s -o %t.files.table
; RUN: FileCheck %s -input-file=%t.files_0.ll --check-prefixes CHECK0,CHECK
; RUN: FileCheck %s -input-file=%t.files_1.ll --check-prefixes CHECK1,CHECK

@SCSymID = private unnamed_addr constant [10 x i8] c"SpecConst\00", align 1
; CHECK-NOT: @SCSymID

declare dso_local spir_func zeroext i1 @_Z27__sycl_getSpecConstantValueIbET_PKc(i8 addrspace(4)*)

define dso_local spir_kernel void @KERNEL_AAA() {
%1 = call spir_func zeroext i1 @_Z27__sycl_getSpecConstantValueIbET_PKc(i8 addrspace(4)* addrspacecast (i8* getelementptr inbounds ([10 x i8], [10 x i8]* @SCSymID, i64 0, i64 0) to i8 addrspace(4)*))
; CHECK0: %{{[0-9]+}} = call i1 @_Z20__spirv_SpecConstantib(i32 0, i1 false), !SYCL_SPEC_CONST_SYM_ID ![[MD_ID:[0-9]+]]
ret void
}

define dso_local spir_kernel void @KERNEL_BBB() {
%1 = call spir_func zeroext i1 @_Z27__sycl_getSpecConstantValueIbET_PKc(i8 addrspace(4)* addrspacecast (i8* getelementptr inbounds ([10 x i8], [10 x i8]* @SCSymID, i64 0, i64 0) to i8 addrspace(4)*))
; CHECK1: %{{[0-9]+}} = call i1 @_Z20__spirv_SpecConstantib(i32 0, i1 false), !SYCL_SPEC_CONST_SYM_ID ![[MD_ID:[0-9]+]]
ret void
}

; CHECK: ![[MD_ID]] = !{!"SpecConst", i32 0}
18 changes: 8 additions & 10 deletions llvm/tools/sycl-post-link/SpecConstants.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,19 +40,18 @@ static void AssertRelease(bool Cond, const char *Msg) {

StringRef getStringLiteralArg(const CallInst *CI, unsigned ArgNo,
SmallVectorImpl<Instruction *> &DelInsts,
GlobalVariable *&DelGlob) {
GlobalVariable *&SymGlob) {
Value *V = CI->getArgOperand(ArgNo)->stripPointerCasts();

if (auto *L = dyn_cast<LoadInst>(V)) {
// Must be a
// vvvvvvvvvvvvvvvvvvvv
// @.str = private unnamed_addr constant[18 x i8]
// c"_ZTS11MyBoolConst\00", align 1
// @.str = private unnamed_addr constant[10 x i8] c"SpecConst\00", align 1
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This change makes the following comment with LLVM IR invalid for types.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ok

// ...
// %TName = alloca i8 addrspace(4)*, align 8
// ...
// store i8 addrspace(4)* addrspacecast(
// i8* getelementptr inbounds([18 x i8], [18 x i8] * @.str, i32 0, i32 0)
// i8* getelementptr inbounds([10 x i8], [10 x i8] * @.str, i32 0, i32 0)
// to i8 addrspace(4)*), i8 addrspace(4)** %TName, align 8, !tbaa !10
// %1 = load i8 addrspace(4)*, i8 addrspace(4)** %TName, align 8, !tbaa !10
// %call = call spir_func zeroext
Expand Down Expand Up @@ -96,7 +95,7 @@ StringRef getStringLiteralArg(const CallInst *CI, unsigned ArgNo,
V = Store->getValueOperand()->stripPointerCasts();
}
const Constant *Init = cast<GlobalVariable>(V)->getInitializer();
DelGlob = cast<GlobalVariable>(V);
SymGlob = cast<GlobalVariable>(V);
StringRef Res = cast<ConstantDataArray>(Init)->getAsString();
if (Res.size() > 0 && Res[Res.size() - 1] == '\0')
Res = Res.substr(0, Res.size() - 1);
Expand Down Expand Up @@ -214,8 +213,8 @@ PreservedAnalyses SpecConstantsPass::run(Module &M,
// code can't use this intrinsic directly.
SmallVector<Instruction *, 3> DelInsts;
DelInsts.push_back(CI);
GlobalVariable *DelGlob = nullptr;
StringRef SymID = getStringLiteralArg(CI, 0, DelInsts, DelGlob);
GlobalVariable *SymGlob = nullptr;
StringRef SymID = getStringLiteralArg(CI, 0, DelInsts, SymGlob);
Type *SCTy = CI->getType();

if (SetValAtRT) {
Expand Down Expand Up @@ -262,9 +261,8 @@ PreservedAnalyses SpecConstantsPass::run(Module &M,
I->removeFromParent();
I->deleteValue();
}
DelGlob->replaceAllUsesWith(ConstantPointerNull::get(DelGlob->getType()));
DelGlob->removeFromParent();
DelGlob->deleteValue();
// Don't delete SymGlob here, as it may be referenced from multiple
// functions if __sycl_getSpecConstantValue is inlined.
}
}
return IRModified ? PreservedAnalyses::none() : PreservedAnalyses::all();
Expand Down
4 changes: 4 additions & 0 deletions llvm/tools/sycl-post-link/sycl-post-link.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,6 +31,7 @@
#include "llvm/Support/SystemUtils.h"
#include "llvm/Support/WithColor.h"
#include "llvm/Transforms/IPO.h"
#include "llvm/Transforms/IPO/GlobalDCE.h"
#include "llvm/Transforms/Utils/Cloning.h"
#include <memory>

Expand Down Expand Up @@ -427,6 +428,9 @@ int main(int argc, char **argv) {
// Register required analysis
MAM.registerPass([&] { return PassInstrumentationAnalysis(); });
RunSpecConst.addPass(SCP);
if (!DoSplit)
// This pass deletes unreachable globals. Code splitter runs it later.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It would be great if we add a test with spec constants handling and split both.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This does not really apply to this fix. But this is important missing test case - agree, so I'll add a test.

RunSpecConst.addPass(GlobalDCEPass());
PreservedAnalyses Res = RunSpecConst.run(*MPtr, MAM);
SpecConstsMet = !Res.areAllPreserved();
}
Expand Down