This is an archive of the discontinued LLVM Phabricator instance.

Differential D95389

AMDGPU: Revert "[IndirectFunctions] Skip propagating attributes to address taken functions"
ClosedPublic

Authored by kzhuravl on Jan 25 2021, 12:54 PM.

Details

Reviewers
msearles
madhur13490
ronlieb
Commits
rG2cdb34efdac5: Revert "[IndirectFunctions] Skip propagating attributes to address taken…
Summary

This reverts commit dd8ae42674b494e46ec40a22f40068db2b4a8b60.

This commit causes infinite loop when compiling rocThrust and hipCUB.

Diff Detail

Event Timeline

kzhuravl created this revision.Jan 25 2021, 12:54 PM
Herald added subscribers: jdoerfert, kerbowa, JDevlieghere and 8 others. · View Herald TranscriptJan 25 2021, 12:54 PM
kzhuravl requested review of this revision.Jan 25 2021, 12:54 PM
Herald added a project: Restricted Project. · View Herald TranscriptJan 25 2021, 12:54 PM
Herald added a subscriber: wdng. · View Herald Transcript
ronlieb accepted this revision.Jan 25 2021, 12:55 PM

LGTM

This revision is now accepted and ready to land.Jan 25 2021, 12:55 PM
Closed by commit rG2cdb34efdac5: Revert "[IndirectFunctions] Skip propagating attributes to address taken… (authored by kzhuravl). · Explain WhyJan 25 2021, 12:58 PM
This revision was automatically updated to reflect the committed changes.
kzhuravl added a commit: rG2cdb34efdac5: Revert "[IndirectFunctions] Skip propagating attributes to address taken….

Revision Contents

Diff 319097

llvm/lib/Target/AMDGPU/AMDGPUPropagateAttributes.cpp

Show First 20 Lines • Show All 234 Lines • ▼ Show 20 Linesbool AMDGPUPropagateAttributes::process() {
do { do {
Roots.insert(NewRoots.begin(), NewRoots.end()); Roots.insert(NewRoots.begin(), NewRoots.end());
NewRoots.clear(); NewRoots.clear();
for (auto &F : M.functions()) { for (auto &F : M.functions()) {
if (F.isDeclaration()) if (F.isDeclaration())
continue; continue;
// Skip propagating attributes and features to
// address taken functions.
if (F.hasAddressTaken()) {
if (!Roots.count(&F))
NewRoots.insert(&F);
continue;
}
const FnProperties CalleeProps(*TM, F); const FnProperties CalleeProps(*TM, F);
SmallVector<std::pair<CallBase *, Function *>, 32> ToReplace; SmallVector<std::pair<CallBase *, Function *>, 32> ToReplace;
SmallSet<CallBase *, 32> Visited; SmallSet<CallBase *, 32> Visited;
for (User *U : F.users()) { for (User *U : F.users()) {
Instruction *I = dyn_cast<Instruction>(U); Instruction *I = dyn_cast<Instruction>(U);
if (!I) if (!I)
continue; continue;
CallBase *CI = dyn_cast<CallBase>(I); CallBase *CI = dyn_cast<CallBase>(I);
if (!CI) if (!CI)
continue; continue;
Function *Caller = CI->getCaller(); Function *Caller = CI->getCaller();
if (!Caller || !Visited.insert(CI).second) if (!Caller || !Visited.insert(CI).second)
continue; continue;
if (!Roots.count(Caller) && !NewRoots.count(Caller)) if (!Roots.count(Caller) && !NewRoots.count(Caller))
continue; continue;
const FnProperties CallerProps(*TM, *Caller); const FnProperties CallerProps(*TM, *Caller);
// Convergence is allowed if the caller has its if (CalleeProps == CallerProps) {
// address taken because all callee's (attributes + features)
// may not agree as the callee may be the target of
// more than one function (called directly or indirectly).
if (Caller->hasAddressTaken() || CalleeProps == CallerProps) {
if (!Roots.count(&F)) if (!Roots.count(&F))
NewRoots.insert(&F); NewRoots.insert(&F);
continue; continue;
} }
Function *NewF = findFunction(CallerProps, &F); Function *NewF = findFunction(CallerProps, &F);
if (!NewF) { if (!NewF) {
const FnProperties NewProps = CalleeProps.adjustToCaller(CallerProps); const FnProperties NewProps = CalleeProps.adjustToCaller(CallerProps);
▲ Show 20 Lines • Show All 156 Lines • Show Last 20 Lines

llvm/test/CodeGen/AMDGPU/propagate-attributes-common-callees.ll

  • This file was deleted.
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-propagate-attributes-early %s | FileCheck %s
; Complicated call graph where a function is called
; directly from a kernel abd also from a function
; whose address is taken.
; CHECK-LABEL: define float @common_callee.gc() #0 {
define float @common_callee.gc() {
ret float 0.0
}
; CHECK-LABEL: define float @foo() {
define float @foo() {
ret float 0.0
}
; CHECK-LABEL: define float @bar() {
define float @bar() {
ret float 0.0
}
; CHECK-LABEL: define float @baz() {
define float @baz() {
ret float 0.0
}
define amdgpu_kernel void @switch_indirect_kernel(float *%result, i32 %type) #1 {
%fn = alloca float ()*
switch i32 %type, label %sw.default [
i32 1, label %sw.bb
i32 2, label %sw.bb2
i32 3, label %sw.bb3
]
sw.bb:
store float ()* @foo, float ()** %fn
br label %sw.epilog
sw.bb2:
store float ()* @bar, float ()** %fn
br label %sw.epilog
sw.bb3:
store float ()* @baz, float ()** %fn
br label %sw.epilog
sw.default:
br label %sw.epilog
sw.epilog:
%fp = load float ()*, float ()** %fn
%direct_call = call contract float @common_callee.gc()
%indirect_call = call contract float %fp()
store float %indirect_call, float* %result
ret void
}
attributes #0 = { "amdgpu-flat-work-group-size"="1,256" "target-features"="-wavefrontsize16,-wavefrontsize32,+wavefrontsize64" }
attributes #1 = { "amdgpu-flat-work-group-size"="1,256" }

llvm/test/CodeGen/AMDGPU/propagate-attributes-direct-indirect-common-callee.ll

  • This file was deleted.
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-propagate-attributes-early %s | FileCheck %s
; Test to check if we skip propgating attributes even if
; a function is called directly as well as
; indirectly. "baz" is called directly as well indirectly.
; CHECK-LABEL: define float @foo() {
define float @foo() {
ret float 0.0
}
; CHECK-LABEL: define float @bar() {
define float @bar() {
ret float 0.0
}
; CHECK-LABEL: define float @baz() {
define float @baz() {
ret float 0.0
}
define amdgpu_kernel void @switch_indirect_kernel(float *%result, i32 %type) #1 {
%fn = alloca float ()*
switch i32 %type, label %sw.default [
i32 1, label %sw.bb
i32 2, label %sw.bb2
i32 3, label %sw.bb3
]
sw.bb:
store float ()* @foo, float ()** %fn
br label %sw.epilog
sw.bb2:
store float ()* @bar, float ()** %fn
br label %sw.epilog
sw.bb3:
store float ()* @baz, float ()** %fn
br label %sw.epilog
sw.default:
br label %sw.epilog
sw.epilog:
%fp = load float ()*, float ()** %fn
%direct_call = call contract float @baz()
%indirect_call = call contract float %fp()
store float %indirect_call, float* %result
ret void
}
attributes #1 = { "amdgpu-flat-work-group-size"="1,256" }

llvm/test/CodeGen/AMDGPU/propagate-attributes-direct-indirect.ll

  • This file was deleted.
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-propagate-attributes-early %s | FileCheck %s
; Test to check if we skip attributes on address
; taken functions but pass to direct callees.
; CHECK-LABEL: define float @foo() {
define float @foo() {
ret float 0.0
}
; CHECK-LABEL: define float @bar() {
define float @bar() {
ret float 0.0
}
; CHECK-LABEL: define float @baz() {
define float @baz() {
ret float 0.0
}
; CHECK-LABEL: define float @baz2() #0 {
define float @baz2() {
ret float 0.0
}
define amdgpu_kernel void @switch_indirect_kernel(float *%result, i32 %type) #1 {
%fn = alloca float ()*
switch i32 %type, label %sw.default [
i32 1, label %sw.bb
i32 2, label %sw.bb2
i32 3, label %sw.bb3
]
sw.bb:
store float ()* @foo, float ()** %fn
br label %sw.epilog
sw.bb2:
store float ()* @bar, float ()** %fn
br label %sw.epilog
sw.bb3:
store float ()* @baz, float ()** %fn
br label %sw.epilog
sw.default:
br label %sw.epilog
sw.epilog:
%fp = load float ()*, float ()** %fn
%direct_call = call contract float @baz2()
%indirect_call = call contract float %fp()
store float %indirect_call, float* %result
ret void
}
attributes #0 = { "amdgpu-flat-work-group-size"="1,256" "target-features"="-wavefrontsize16,-wavefrontsize32,+wavefrontsize64" }
attributes #1 = { "amdgpu-flat-work-group-size"="1,256"}

llvm/test/CodeGen/AMDGPU/propagate-attributes-indirect.ll

  • This file was deleted.
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-propagate-attributes-early %s | FileCheck %s
; Test to check if we skip attributes on address
; taken functions in a simple call graph.
; CHECK-LABEL: define float @foo() {
define float @foo() {
ret float 0.0
}
; CHECK-LABEL: define float @bar() {
define float @bar() {
ret float 0.0
}
; CHECK-LABEL: define float @baz() {
define float @baz() {
ret float 0.0
}
define amdgpu_kernel void @switch_indirect_kernel(float *%result, i32 %type) #1 {
%fn = alloca float ()*
switch i32 %type, label %sw.default [
i32 1, label %sw.bb
i32 2, label %sw.bb2
i32 3, label %sw.bb3
]
sw.bb:
store float ()* @foo, float ()** %fn
br label %sw.epilog
sw.bb2:
store float ()* @bar, float ()** %fn
br label %sw.epilog
sw.bb3:
store float ()* @baz, float ()** %fn
br label %sw.epilog
sw.default:
br label %sw.epilog
sw.epilog:
%fp = load float ()*, float ()** %fn
%indirect_call = call contract float %fp()
store float %indirect_call, float* %result
ret void
}
attributes #1 = { "amdgpu-flat-work-group-size"="1,256" }
llvm/lib/Target/AMDGPU/AMDGPUPropagateAttributes.cpp