[llvm] eb73af4 - [Attributor] Handle undef and null in AAAlignFloating
Johannes Doerfert via llvm-commits
llvm-commits at lists.llvm.org
Sun Mar 6 21:39:23 PST 2022
Author: Johannes Doerfert
Date: 2022-03-06T23:26:22-06:00
New Revision: eb73af4af4699a6907b32790a0f178302c57d802
URL: https://github.com/llvm/llvm-project/commit/eb73af4af4699a6907b32790a0f178302c57d802
DIFF: https://github.com/llvm/llvm-project/commit/eb73af4af4699a6907b32790a0f178302c57d802.diff
LOG: [Attributor] Handle undef and null in AAAlignFloating
Both `undef` and `nullptr` are maximally aligned. This is especially
important as we often see `undef` until a proper value has been
identified during simplification.
Added:
Modified:
llvm/lib/Transforms/IPO/AttributorAttributes.cpp
llvm/test/Transforms/Attributor/ArgumentPromotion/variadic.ll
llvm/test/Transforms/Attributor/align.ll
llvm/test/Transforms/Attributor/value-simplify-gpu.ll
llvm/test/Transforms/Attributor/value-simplify.ll
llvm/test/Transforms/OpenMP/spmdization.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/IPO/AttributorAttributes.cpp b/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
index 9589a99b238b6..4729fc753b4a6 100644
--- a/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
+++ b/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
@@ -4593,6 +4593,8 @@ struct AAAlignFloating : AAAlignImpl {
auto VisitValueCB = [&](Value &V, const Instruction *,
AAAlign::StateType &T, bool Stripped) -> bool {
+ if (isa<UndefValue>(V) || isa<ConstantPointerNull>(V))
+ return true;
const auto &AA = A.getAAFor<AAAlign>(*this, IRPosition::value(V),
DepClassTy::REQUIRED);
if (!Stripped && this == &AA) {
@@ -4600,6 +4602,7 @@ struct AAAlignFloating : AAAlignImpl {
unsigned Alignment = 1;
if (const Value *Base =
GetPointerBaseWithConstantOffset(&V, Offset, DL)) {
+ // TODO: Use AAAlign for the base too.
Align PA = Base->getPointerAlignment(DL);
// BasePointerAddr + Offset = Alignment * Q for some integer Q.
// So we can say that the maximum power of two which is a divisor of
diff --git a/llvm/test/Transforms/Attributor/ArgumentPromotion/variadic.ll b/llvm/test/Transforms/Attributor/ArgumentPromotion/variadic.ll
index 4c93cf5dec685..6c9d34fa3bbfb 100644
--- a/llvm/test/Transforms/Attributor/ArgumentPromotion/variadic.ll
+++ b/llvm/test/Transforms/Attributor/ArgumentPromotion/variadic.ll
@@ -35,7 +35,7 @@ entry:
; Function Attrs: nounwind uwtable
define internal void @callee_t0f(i8* nocapture readnone %tp13, i8* nocapture readnone %tp14, i8* nocapture readnone %tp15, i8* nocapture readnone %tp16, i8* nocapture readnone %tp17, ...) {
; CHECK-LABEL: define {{[^@]+}}@callee_t0f
-; CHECK-SAME: (i8* noalias nocapture nofree nonnull readnone [[TP13:%.*]], i8* noalias nocapture nofree nonnull readnone [[TP14:%.*]], i8* noalias nocapture nofree nonnull readnone [[TP15:%.*]], i8* noalias nocapture nofree nonnull readnone [[TP16:%.*]], i8* noalias nocapture nofree nonnull readnone [[TP17:%.*]], ...) {
+; CHECK-SAME: (i8* noalias nocapture nofree nonnull readnone align 4294967296 [[TP13:%.*]], i8* noalias nocapture nofree nonnull readnone align 4294967296 [[TP14:%.*]], i8* noalias nocapture nofree nonnull readnone align 4294967296 [[TP15:%.*]], i8* noalias nocapture nofree nonnull readnone align 4294967296 [[TP16:%.*]], i8* noalias nocapture nofree nonnull readnone align 4294967296 [[TP17:%.*]], ...) {
; CHECK-NEXT: entry:
; CHECK-NEXT: call void @sink(i32 noundef 0)
; CHECK-NEXT: ret void
diff --git a/llvm/test/Transforms/Attributor/align.ll b/llvm/test/Transforms/Attributor/align.ll
index 748cc25f9c661..b0ed4239abdd9 100644
--- a/llvm/test/Transforms/Attributor/align.ll
+++ b/llvm/test/Transforms/Attributor/align.ll
@@ -16,6 +16,7 @@ target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
; CHECK: @[[A1:[a-zA-Z0-9_$"\\.-]+]] = common global i8 0, align 8
; CHECK: @[[A2:[a-zA-Z0-9_$"\\.-]+]] = common global i8 0, align 16
; CHECK: @[[CND:[a-zA-Z0-9_$"\\.-]+]] = external global i1
+; CHECK: @[[G:[a-zA-Z0-9_$"\\.-]+]] = global i8 0, align 32
;.
define i32* @test1(i32* align 8 %0) #0 {
; CHECK: Function Attrs: nofree noinline norecurse nosync nounwind readnone willreturn uwtable
@@ -1114,6 +1115,75 @@ define void @align4_caller(i8* %p) {
declare void @align4_callee(i8* align(4) %p)
+ at G = global i8 0, align 32
+
+define internal i8* @aligned_8_return(i8* %a, i1 %c1, i1 %c2) norecurse {
+; NOT_CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
+; NOT_CGSCC_OPM-LABEL: define {{[^@]+}}@aligned_8_return
+; NOT_CGSCC_OPM-SAME: (i8* noalias nofree readnone align 16 "no-capture-maybe-returned" [[A:%.*]], i1 [[C1:%.*]], i1 [[C2:%.*]]) #[[ATTR9]] {
+; NOT_CGSCC_OPM-NEXT: [[STACK:%.*]] = alloca i8*, align 8
+; NOT_CGSCC_OPM-NEXT: br i1 [[C1]], label [[T:%.*]], label [[F:%.*]]
+; NOT_CGSCC_OPM: t:
+; NOT_CGSCC_OPM-NEXT: [[GEP:%.*]] = getelementptr i8, i8* @G, i32 8
+; NOT_CGSCC_OPM-NEXT: [[SEL:%.*]] = select i1 [[C2]], i8* [[A]], i8* [[GEP]]
+; NOT_CGSCC_OPM-NEXT: store i8* [[SEL]], i8** [[STACK]], align 8
+; NOT_CGSCC_OPM-NEXT: br label [[END:%.*]]
+; NOT_CGSCC_OPM: f:
+; NOT_CGSCC_OPM-NEXT: store i8* @G, i8** [[STACK]], align 8
+; NOT_CGSCC_OPM-NEXT: br label [[END]]
+; NOT_CGSCC_OPM: end:
+; NOT_CGSCC_OPM-NEXT: [[L:%.*]] = load i8*, i8** [[STACK]], align 8
+; NOT_CGSCC_OPM-NEXT: ret i8* [[L]]
+;
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@aligned_8_return
+; IS__CGSCC_OPM-SAME: (i8* noalias nofree readnone align 16 "no-capture-maybe-returned" [[A:%.*]], i1 [[C1:%.*]], i1 [[C2:%.*]]) #[[ATTR10]] {
+; IS__CGSCC_OPM-NEXT: [[STACK:%.*]] = alloca i8*, align 8
+; IS__CGSCC_OPM-NEXT: br i1 [[C1]], label [[T:%.*]], label [[F:%.*]]
+; IS__CGSCC_OPM: t:
+; IS__CGSCC_OPM-NEXT: [[GEP:%.*]] = getelementptr i8, i8* @G, i32 8
+; IS__CGSCC_OPM-NEXT: [[SEL:%.*]] = select i1 [[C2]], i8* [[A]], i8* [[GEP]]
+; IS__CGSCC_OPM-NEXT: store i8* [[SEL]], i8** [[STACK]], align 8
+; IS__CGSCC_OPM-NEXT: br label [[END:%.*]]
+; IS__CGSCC_OPM: f:
+; IS__CGSCC_OPM-NEXT: store i8* @G, i8** [[STACK]], align 8
+; IS__CGSCC_OPM-NEXT: br label [[END]]
+; IS__CGSCC_OPM: end:
+; IS__CGSCC_OPM-NEXT: [[L:%.*]] = load i8*, i8** [[STACK]], align 8
+; IS__CGSCC_OPM-NEXT: ret i8* [[L]]
+;
+ %stack = alloca i8*
+ br i1 %c1, label %t, label %f
+t:
+ %gep = getelementptr i8, i8* @G, i32 8
+ %sel = select i1 %c2, i8* %a, i8* %gep
+ store i8* %sel, i8** %stack
+ br label %end
+f:
+ store i8* @G, i8** %stack
+ br label %end
+end:
+ %l = load i8*, i8** %stack
+ ret i8* %l
+}
+
+define i8* @aligned_8_return_caller(i8* align(16) %a, i1 %c1, i1 %c2) {
+; NOT_CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
+; NOT_CGSCC_OPM-LABEL: define {{[^@]+}}@aligned_8_return_caller
+; NOT_CGSCC_OPM-SAME: (i8* nofree readnone align 16 "no-capture-maybe-returned" [[A:%.*]], i1 [[C1:%.*]], i1 [[C2:%.*]]) #[[ATTR9]] {
+; NOT_CGSCC_OPM-NEXT: [[R:%.*]] = call i8* @aligned_8_return(i8* noalias nofree readnone align 16 "no-capture-maybe-returned" [[A]], i1 [[C1]], i1 [[C2]]) #[[ATTR12:[0-9]+]]
+; NOT_CGSCC_OPM-NEXT: ret i8* [[R]]
+;
+; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
+; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@aligned_8_return_caller
+; IS__CGSCC_OPM-SAME: (i8* nofree readnone align 16 "no-capture-maybe-returned" [[A:%.*]], i1 [[C1:%.*]], i1 [[C2:%.*]]) #[[ATTR10]] {
+; IS__CGSCC_OPM-NEXT: [[R:%.*]] = call i8* @aligned_8_return(i8* noalias nofree readnone align 16 "no-capture-maybe-returned" [[A]], i1 [[C1]], i1 [[C2]]) #[[ATTR13:[0-9]+]]
+; IS__CGSCC_OPM-NEXT: ret i8* [[R]]
+;
+ %r = call i8* @aligned_8_return(i8* %a, i1 %c1, i1 %c2)
+ ret i8* %r
+}
+
attributes #0 = { nounwind uwtable noinline }
attributes #1 = { uwtable noinline }
attributes #2 = { null_pointer_is_valid }
@@ -1130,6 +1200,7 @@ attributes #2 = { null_pointer_is_valid }
; IS__TUNIT____: attributes #[[ATTR9]] = { nofree norecurse nosync nounwind readnone willreturn }
; IS__TUNIT____: attributes #[[ATTR10]] = { nofree norecurse nosync nounwind readonly willreturn }
; IS__TUNIT____: attributes #[[ATTR11]] = { nofree nosync nounwind readonly willreturn }
+; IS__TUNIT____: attributes #[[ATTR12]] = { nofree nosync nounwind readnone willreturn }
;.
; IS__CGSCC_OPM: attributes #[[ATTR0]] = { nofree noinline norecurse nosync nounwind readnone willreturn uwtable }
; IS__CGSCC_OPM: attributes #[[ATTR1]] = { nofree noinline nosync nounwind readnone willreturn uwtable }
@@ -1144,6 +1215,7 @@ attributes #2 = { null_pointer_is_valid }
; IS__CGSCC_OPM: attributes #[[ATTR10]] = { nofree norecurse nosync nounwind readnone willreturn }
; IS__CGSCC_OPM: attributes #[[ATTR11]] = { nofree norecurse nosync nounwind readonly willreturn }
; IS__CGSCC_OPM: attributes #[[ATTR12]] = { readonly willreturn }
+; IS__CGSCC_OPM: attributes #[[ATTR13]] = { readnone willreturn }
;.
; IS__CGSCC_NPM: attributes #[[ATTR0]] = { nofree noinline norecurse nosync nounwind readnone willreturn uwtable }
; IS__CGSCC_NPM: attributes #[[ATTR1]] = { noinline norecurse nounwind uwtable }
@@ -1157,4 +1229,5 @@ attributes #2 = { null_pointer_is_valid }
; IS__CGSCC_NPM: attributes #[[ATTR9]] = { nofree norecurse nosync nounwind readnone willreturn }
; IS__CGSCC_NPM: attributes #[[ATTR10]] = { nofree norecurse nosync nounwind readonly willreturn }
; IS__CGSCC_NPM: attributes #[[ATTR11]] = { readonly willreturn }
+; IS__CGSCC_NPM: attributes #[[ATTR12]] = { readnone willreturn }
;.
diff --git a/llvm/test/Transforms/Attributor/value-simplify-gpu.ll b/llvm/test/Transforms/Attributor/value-simplify-gpu.ll
index be3bb318e23e6..24764f12a6475 100644
--- a/llvm/test/Transforms/Attributor/value-simplify-gpu.ll
+++ b/llvm/test/Transforms/Attributor/value-simplify-gpu.ll
@@ -308,11 +308,11 @@ define internal void @level2a(i32* %addr) {
;
; IS__CGSCC_OPM: Function Attrs: norecurse nosync nounwind
; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@level2a
-; IS__CGSCC_OPM-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[ADDR:%.*]]) #[[ATTR1]] {
+; IS__CGSCC_OPM-SAME: (i32* noalias nocapture nofree nonnull readnone align 4294967296 dereferenceable(4) [[ADDR:%.*]]) #[[ATTR1]] {
; IS__CGSCC_OPM-NEXT: entry:
; IS__CGSCC_OPM-NEXT: [[TMP0:%.*]] = load i32, i32* addrspacecast (i32 addrspace(3)* @ReachableNonKernel to i32*), align 4
; IS__CGSCC_OPM-NEXT: [[TMP1:%.*]] = load i32, i32* addrspacecast (i32 addrspace(3)* @UnreachableNonKernel to i32*), align 4
-; IS__CGSCC_OPM-NEXT: [[TMP2:%.*]] = load i32, i32* undef, align 4
+; IS__CGSCC_OPM-NEXT: [[TMP2:%.*]] = load i32, i32* undef, align 4294967296
; IS__CGSCC_OPM-NEXT: call void @use(i32 [[TMP0]], i32 [[TMP1]], i32 17) #[[ATTR4]]
; IS__CGSCC_OPM-NEXT: ret void
;
@@ -347,11 +347,11 @@ define internal void @level2b(i32* %addr) {
;
; IS__CGSCC_OPM: Function Attrs: norecurse nosync nounwind
; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@level2b
-; IS__CGSCC_OPM-SAME: (i32* noalias nocapture nofree nonnull readnone align 4 dereferenceable(4) [[ADDR:%.*]]) #[[ATTR1]] {
+; IS__CGSCC_OPM-SAME: (i32* noalias nocapture nofree nonnull readnone align 4294967296 dereferenceable(4) [[ADDR:%.*]]) #[[ATTR1]] {
; IS__CGSCC_OPM-NEXT: entry:
; IS__CGSCC_OPM-NEXT: [[TMP0:%.*]] = load i32, i32* addrspacecast (i32 addrspace(3)* @ReachableNonKernel to i32*), align 4
; IS__CGSCC_OPM-NEXT: [[TMP1:%.*]] = load i32, i32* addrspacecast (i32 addrspace(3)* @UnreachableNonKernel to i32*), align 4
-; IS__CGSCC_OPM-NEXT: [[TMP2:%.*]] = load i32, i32* undef, align 4
+; IS__CGSCC_OPM-NEXT: [[TMP2:%.*]] = load i32, i32* undef, align 4294967296
; IS__CGSCC_OPM-NEXT: call void @use(i32 [[TMP0]], i32 [[TMP1]], i32 17) #[[ATTR4]]
; IS__CGSCC_OPM-NEXT: ret void
;
diff --git a/llvm/test/Transforms/Attributor/value-simplify.ll b/llvm/test/Transforms/Attributor/value-simplify.ll
index 4bd0e5a66434b..69b4b8ba90d33 100644
--- a/llvm/test/Transforms/Attributor/value-simplify.ll
+++ b/llvm/test/Transforms/Attributor/value-simplify.ll
@@ -916,7 +916,7 @@ define internal void @callee_is_undef(void ()* %fn) {
; IS__TUNIT____-NEXT: ret void
;
; IS__CGSCC____-LABEL: define {{[^@]+}}@callee_is_undef
-; IS__CGSCC____-SAME: (void ()* nocapture nofree noundef nonnull [[FN:%.*]]) {
+; IS__CGSCC____-SAME: (void ()* nocapture nofree noundef nonnull align 4294967296 [[FN:%.*]]) {
; IS__CGSCC____-NEXT: call void [[FN]]()
; IS__CGSCC____-NEXT: ret void
;
diff --git a/llvm/test/Transforms/OpenMP/spmdization.ll b/llvm/test/Transforms/OpenMP/spmdization.ll
index 7ba9349256d61..b6bcd700d197f 100644
--- a/llvm/test/Transforms/OpenMP/spmdization.ll
+++ b/llvm/test/Transforms/OpenMP/spmdization.ll
@@ -2169,25 +2169,25 @@ user_code.entry: ; preds = %entry
; Function Attrs: alwaysinline convergent nounwind
define internal void @.omp_outlined.(i32 %.global_tid., i32* noalias %.part_id., i8* noalias %.privates., void (i8*, ...)* noalias %.copy_fn., i8* %.task_t., %struct.anon* noalias %__context) #9 {
; AMDGPU-LABEL: define {{[^@]+}}@.omp_outlined.
-; AMDGPU-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
+; AMDGPU-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone align 4294967296 [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
; AMDGPU-NEXT: entry:
; AMDGPU-NEXT: call void @spmd_amenable() #[[ATTR5]]
; AMDGPU-NEXT: ret void
;
; NVPTX-LABEL: define {{[^@]+}}@.omp_outlined.
-; NVPTX-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
+; NVPTX-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone align 4294967296 [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
; NVPTX-NEXT: entry:
; NVPTX-NEXT: call void @spmd_amenable() #[[ATTR5]]
; NVPTX-NEXT: ret void
;
; AMDGPU-DISABLED-LABEL: define {{[^@]+}}@.omp_outlined.
-; AMDGPU-DISABLED-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
+; AMDGPU-DISABLED-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone align 4294967296 [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
; AMDGPU-DISABLED-NEXT: entry:
; AMDGPU-DISABLED-NEXT: call void @spmd_amenable() #[[ATTR5]]
; AMDGPU-DISABLED-NEXT: ret void
;
; NVPTX-DISABLED-LABEL: define {{[^@]+}}@.omp_outlined.
-; NVPTX-DISABLED-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
+; NVPTX-DISABLED-SAME: (i32 [[DOTGLOBAL_TID_:%.*]], i32* noalias nocapture nofree nonnull readnone [[DOTPART_ID_:%.*]], i8* noalias nocapture nofree readnone align 4294967296 [[DOTPRIVATES_:%.*]], void (i8*, ...)* noalias nocapture nofree readnone align 4294967296 [[DOTCOPY_FN_:%.*]], i8* noalias nocapture nofree nonnull readnone align 8 dereferenceable(8) [[DOTTASK_T_:%.*]], %struct.anon* noalias nocapture nofree readnone [[__CONTEXT:%.*]]) #[[ATTR0]] {
; NVPTX-DISABLED-NEXT: entry:
; NVPTX-DISABLED-NEXT: call void @spmd_amenable() #[[ATTR5]]
; NVPTX-DISABLED-NEXT: ret void
More information about the llvm-commits
mailing list