[FuncSpec] Enable SpecializeLiteralConstant by default (#113442)

hazzlim · labrinea · web-flow · commit 06664fdc7680 · 2024-10-29T11:41:25.000Z
Enable specialization on literal constant arguments by default in
Function Specialization.

---------

Co-authored-by: Alexandros Lamprineas &lt;alexandros.lamprineas@arm.com&gt;
diff --git a/llvm/include/llvm/Transforms/IPO/FunctionSpecialization.h b/llvm/include/llvm/Transforms/IPO/FunctionSpecialization.h
@@ -64,6 +64,13 @@
 // - Perhaps a post-inlining function specialization pass could be more
 //   aggressive on literal constants.
 //
+// Limitations:
+// ------------
+// - We are unable to consider specializations of functions called from indirect
+//   callsites whose pointer operand has a lattice value that is known to be
+//   constant, either from IPSCCP or previous iterations of FuncSpec. This is
+//   because SCCP has not yet replaced the uses of the known constant.
+//
 // References:
 // -----------
 // 2021 LLVM Dev Mtg “Introducing function specialisation, and can we enable
diff --git a/llvm/lib/Transforms/IPO/FunctionSpecialization.cpp b/llvm/lib/Transforms/IPO/FunctionSpecialization.cpp
@@ -84,14 +84,11 @@ static cl::opt<bool> SpecializeOnAddress(
     "funcspec-on-address", cl::init(false), cl::Hidden, cl::desc(
     "Enable function specialization on the address of global values"));
 
-// Disabled by default as it can significantly increase compilation times.
-//
-// https://llvm-compile-time-tracker.com
-// https://github.com/nikic/llvm-compile-time-tracker
 static cl::opt<bool> SpecializeLiteralConstant(
-    "funcspec-for-literal-constant", cl::init(false), cl::Hidden, cl::desc(
-    "Enable specialization of functions that take a literal constant as an "
-    "argument"));
+    "funcspec-for-literal-constant", cl::init(true), cl::Hidden,
+    cl::desc(
+        "Enable specialization of functions that take a literal constant as an "
+        "argument"));
 
 bool InstCostVisitor::canEliminateSuccessor(BasicBlock *BB, BasicBlock *Succ,
                                          DenseSet<BasicBlock *> &DeadBlocks) {
@@ -682,10 +679,11 @@ bool FunctionSpecializer::run() {
         (RequireMinSize && Metrics.NumInsts < MinFunctionSize))
       continue;
 
-    // TODO: For now only consider recursive functions when running multiple
-    // times. This should change if specialization on literal constants gets
-    // enabled.
-    if (!Inserted && !Metrics.isRecursive && !SpecializeLiteralConstant)
+    // When specialization on literal constants is disabled, only consider
+    // recursive functions when running multiple times to save wasted analysis,
+    // as we will not be able to specialize on any newly found literal constant
+    // return values.
+    if (!SpecializeLiteralConstant && !Inserted && !Metrics.isRecursive)
       continue;
 
     int64_t Sz = *Metrics.NumInsts.getValue();
diff --git a/llvm/test/DebugInfo/Generic/ipsccp-remap-assign-id.ll b/llvm/test/DebugInfo/Generic/ipsccp-remap-assign-id.ll
@@ -1,5 +1,5 @@
-; RUN: opt -passes=ipsccp %s -S -o - | FileCheck %s
-; RUN: opt --try-experimental-debuginfo-iterators -passes=ipsccp %s -S -o - | FileCheck %s
+; RUN: opt -passes=ipsccp -funcspec-for-literal-constant=false %s -S -o - | FileCheck %s
+; RUN: opt --try-experimental-debuginfo-iterators -passes=ipsccp -funcspec-for-literal-constant=false %s -S -o - | FileCheck %s
 
 ;; Check the dbg.assign DIAssignID operand gets remapped after cloning.
 
diff --git a/llvm/test/Transforms/FunctionSpecialization/compiler-crash-58759.ll b/llvm/test/Transforms/FunctionSpecialization/compiler-crash-58759.ll
@@ -1,4 +1,4 @@
-; RUN: opt -S --passes="default<O3>" < %s | FileCheck %s
+; RUN: opt -S --passes="default<O3>" -funcspec-for-literal-constant=false < %s | FileCheck %s
 
 define dso_local i32 @g0(i32 noundef %x) local_unnamed_addr {
 entry:
diff --git a/llvm/test/Transforms/FunctionSpecialization/function-specialization-constant-expression.ll b/llvm/test/Transforms/FunctionSpecialization/function-specialization-constant-expression.ll
@@ -4,7 +4,7 @@
 ; Note that this test case shows that function specialization pass would
 ; transform the function even if no specialization happened.
 
-; RUN: opt -passes="ipsccp<func-spec>" -force-specialization -S < %s | FileCheck %s
+; RUN: opt -passes="ipsccp<func-spec>" -force-specialization -funcspec-for-literal-constant=false -S < %s | FileCheck %s
 
 %struct = type { i8, i16, i32, i64, i64}
 @Global = internal constant %struct {i8 0, i16 1, i32 2, i64 3, i64 4}
diff --git a/llvm/test/Transforms/FunctionSpecialization/function-specialization2.ll b/llvm/test/Transforms/FunctionSpecialization/function-specialization2.ll
@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --include-generated-funcs --version 5
-; RUN: opt -passes="ipsccp<func-spec>,deadargelim" -force-specialization -S < %s | FileCheck %s
-; RUN: opt -passes="ipsccp<func-spec>,deadargelim" -funcspec-max-iters=1 -force-specialization -S < %s | FileCheck %s --check-prefix=ONE-ITER
-; RUN: opt -passes="ipsccp<func-spec>,deadargelim" -funcspec-max-iters=0 -force-specialization -S < %s | FileCheck %s --check-prefix=DISABLED
+; RUN: opt -passes="ipsccp<func-spec>,deadargelim" -funcspec-for-literal-constant=false -force-specialization -S < %s | FileCheck %s
+; RUN: opt -passes="ipsccp<func-spec>,deadargelim" -funcspec-for-literal-constant=false -funcspec-max-iters=1 -force-specialization -S < %s | FileCheck %s --check-prefix=ONE-ITER
+; RUN: opt -passes="ipsccp<func-spec>,deadargelim" -funcspec-for-literal-constant=false -funcspec-max-iters=0 -force-specialization -S < %s | FileCheck %s --check-prefix=DISABLED
 
 
 define internal i32 @func(ptr %0, i32 %1, ptr nocapture %2) {
diff --git a/llvm/test/Transforms/FunctionSpecialization/function-specialization4.ll b/llvm/test/Transforms/FunctionSpecialization/function-specialization4.ll
@@ -1,8 +1,10 @@
 ; RUN: opt -passes="ipsccp<func-spec>" -force-specialization \
-; RUN:   -funcspec-max-clones=2 -S < %s | FileCheck %s
+; RUN:   -funcspec-for-literal-constant=false -funcspec-max-clones=2 \
+; RUN:   -S < %s | FileCheck %s
 
 ; RUN: opt -passes="ipsccp<func-spec>" -force-specialization \
-; RUN:   -funcspec-max-clones=1 -S < %s | FileCheck %s --check-prefix=CONST1
+; RUN:   -funcspec-for-literal-constant=false -funcspec-max-clones=1 \
+; RUN:   -S < %s | FileCheck %s --check-prefix=CONST1
 
 target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
 
diff --git a/llvm/test/Transforms/FunctionSpecialization/get-possible-constants.ll b/llvm/test/Transforms/FunctionSpecialization/get-possible-constants.ll
@@ -1,4 +1,4 @@
-; RUN: opt -S --passes="ipsccp<func-spec>" < %s | FileCheck %s
+; RUN: opt -S --passes="ipsccp<func-spec>" -funcspec-for-literal-constant=false < %s | FileCheck %s
 define dso_local i32 @p0(i32 noundef %x) {
 entry:
   %add = add nsw i32 %x, 1
diff --git a/llvm/test/Transforms/FunctionSpecialization/global-rank.ll b/llvm/test/Transforms/FunctionSpecialization/global-rank.ll
@@ -1,4 +1,4 @@
-; RUN: opt -S --passes="ipsccp<func-spec>" -funcspec-max-clones=1 < %s | FileCheck %s
+; RUN: opt -S --passes="ipsccp<func-spec>" -funcspec-for-literal-constant=false -funcspec-max-clones=1 < %s | FileCheck %s
 define internal i32 @f(i32 noundef %x, ptr nocapture noundef readonly %p, ptr nocapture noundef readonly %q) noinline {
 entry:
   %call = tail call i32 %p(i32 noundef %x)
diff --git a/llvm/test/Transforms/FunctionSpecialization/identical-specializations.ll b/llvm/test/Transforms/FunctionSpecialization/identical-specializations.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --include-generated-funcs --version 5
-; RUN: opt -passes="ipsccp<func-spec>" -force-specialization -S < %s | FileCheck %s
+; RUN: opt -passes="ipsccp<func-spec>" -force-specialization -funcspec-for-literal-constant=false -S < %s | FileCheck %s
 
 define i64 @main(i64 %x, i64 %y, i1 %flag) {
 entry:
diff --git a/llvm/test/Transforms/FunctionSpecialization/literal-const.ll b/llvm/test/Transforms/FunctionSpecialization/literal-const.ll
@@ -1,7 +1,7 @@
 ; RUN: opt -S --passes="ipsccp<func-spec>" \
+; RUN:        -funcspec-for-literal-constant=false \
 ; RUN:        -force-specialization < %s | FileCheck %s -check-prefix CHECK-NOLIT
 ; RUN: opt -S --passes="ipsccp<func-spec>" \
-; RUN:        -funcspec-for-literal-constant \
 ; RUN:        -force-specialization < %s | FileCheck %s -check-prefix CHECK-LIT
 
 define i32 @f0(i32 noundef %x) {
diff --git a/llvm/test/Transforms/FunctionSpecialization/specialize-multiple-arguments.ll b/llvm/test/Transforms/FunctionSpecialization/specialize-multiple-arguments.ll
@@ -1,8 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
-; RUN: opt -passes="ipsccp<func-spec>" -funcspec-max-clones=0 -funcspec-min-function-size=14 -S < %s | FileCheck %s --check-prefix=NONE
-; RUN: opt -passes="ipsccp<func-spec>" -funcspec-max-clones=1 -funcspec-min-function-size=14 -S < %s | FileCheck %s --check-prefix=ONE
-; RUN: opt -passes="ipsccp<func-spec>" -funcspec-max-clones=2 -funcspec-min-function-size=14 -S < %s | FileCheck %s --check-prefix=TWO
-; RUN: opt -passes="ipsccp<func-spec>" -funcspec-max-clones=3 -funcspec-min-function-size=14 -S < %s | FileCheck %s --check-prefix=THREE
+; RUN: opt -passes="ipsccp<func-spec>" -funcspec-for-literal-constant=false -funcspec-max-clones=0 -funcspec-min-function-size=14 -S < %s | FileCheck %s --check-prefix=NONE
+; RUN: opt -passes="ipsccp<func-spec>" -funcspec-for-literal-constant=false -funcspec-max-clones=1 -funcspec-min-function-size=14 -S < %s | FileCheck %s --check-prefix=ONE
+; RUN: opt -passes="ipsccp<func-spec>" -funcspec-for-literal-constant=false -funcspec-max-clones=2 -funcspec-min-function-size=14 -S < %s | FileCheck %s --check-prefix=TWO
+; RUN: opt -passes="ipsccp<func-spec>" -funcspec-for-literal-constant=false -funcspec-max-clones=3 -funcspec-min-function-size=14 -S < %s | FileCheck %s --check-prefix=THREE
 
 ; Make sure that we iterate correctly after sorting the specializations:
 ; FnSpecialization: Specializations for function compute
diff --git a/llvm/test/Transforms/FunctionSpecialization/track-ptr-return.ll b/llvm/test/Transforms/FunctionSpecialization/track-ptr-return.ll
@@ -0,0 +1,147 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --include-generated-funcs --version 5
+; RUN: opt -passes="ipsccp<func-spec>" -force-specialization \
+; RUN:     -funcspec-max-iters=3 -S < %s | FileCheck %s
+
+; RUN: opt -passes="ipsccp<func-spec>" -force-specialization          \
+; RUN:     -funcspec-for-literal-constant=false -funcspec-max-iters=3 \
+; RUN:     -S < %s | FileCheck %s --check-prefix=NOLIT
+
+@global_true = constant i1 true
+@global_false = constant i1 false
+
+define i64 @main() {
+entry:
+  %op1 = call ptr @select_op(ptr @global_true)
+  %op2 = call ptr @select_op(ptr @global_false)
+
+  %c1 = call i64 @compute(ptr %op1)
+  %c2 = call i64 @compute(ptr %op2)
+  %add = add i64 %c1, %c2
+  ret i64 %add
+}
+
+define ptr @select_op(ptr %flag) {
+  %flag.val = load i1, ptr %flag
+  %op = select i1 %flag.val, ptr @plus, ptr @minus
+  ret ptr %op
+}
+
+define internal i64 @compute(ptr %op) {
+entry:
+  %res = call i64 %op(i64 1)
+  ret i64 %res
+}
+
+define internal i64 @plus(i64 %x) {
+entry:
+  %sum = add i64 %x, 1
+  ret i64 %sum
+}
+
+define internal i64 @minus(i64 %x) {
+entry:
+  %diff = sub i64 %x, 1
+  ret i64 %diff
+}
+; CHECK-LABEL: define i64 @main() {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[OP1:%.*]] = call ptr @select_op.specialized.1(ptr @global_true)
+; CHECK-NEXT:    [[OP2:%.*]] = call ptr @select_op.specialized.2(ptr @global_false)
+; CHECK-NEXT:    [[C1:%.*]] = call i64 @compute.specialized.3(ptr @plus)
+; CHECK-NEXT:    [[C2:%.*]] = call i64 @compute.specialized.4(ptr @minus)
+; CHECK-NEXT:    [[ADD:%.*]] = add i64 [[C1]], [[C2]]
+; CHECK-NEXT:    ret i64 [[ADD]]
+;
+;
+; CHECK-LABEL: define ptr @select_op(
+; CHECK-SAME: ptr [[FLAG:%.*]]) {
+; CHECK-NEXT:    [[FLAG_VAL:%.*]] = load i1, ptr [[FLAG]], align 1
+; CHECK-NEXT:    [[OP:%.*]] = select i1 [[FLAG_VAL]], ptr @plus, ptr @minus
+; CHECK-NEXT:    ret ptr [[OP]]
+;
+;
+; CHECK-LABEL: define internal i64 @plus(
+; CHECK-SAME: i64 [[X:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[SUM:%.*]] = add i64 [[X]], 1
+; CHECK-NEXT:    ret i64 [[SUM]]
+;
+;
+; CHECK-LABEL: define internal i64 @minus(
+; CHECK-SAME: i64 [[X:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[DIFF:%.*]] = sub i64 [[X]], 1
+; CHECK-NEXT:    ret i64 [[DIFF]]
+;
+;
+; CHECK-LABEL: define internal ptr @select_op.specialized.1(
+; CHECK-SAME: ptr [[FLAG:%.*]]) {
+; CHECK-NEXT:    ret ptr poison
+;
+;
+; CHECK-LABEL: define internal ptr @select_op.specialized.2(
+; CHECK-SAME: ptr [[FLAG:%.*]]) {
+; CHECK-NEXT:    ret ptr poison
+;
+;
+; CHECK-LABEL: define internal i64 @compute.specialized.3(
+; CHECK-SAME: ptr [[OP:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[RES:%.*]] = call i64 @plus(i64 1)
+; CHECK-NEXT:    ret i64 [[RES]]
+;
+;
+; CHECK-LABEL: define internal i64 @compute.specialized.4(
+; CHECK-SAME: ptr [[OP:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[RES:%.*]] = call i64 @minus(i64 1)
+; CHECK-NEXT:    ret i64 [[RES]]
+;
+;
+; NOLIT-LABEL: define i64 @main() {
+; NOLIT-NEXT:  [[ENTRY:.*:]]
+; NOLIT-NEXT:    [[OP1:%.*]] = call ptr @select_op.specialized.1(ptr @global_true)
+; NOLIT-NEXT:    [[OP2:%.*]] = call ptr @select_op.specialized.2(ptr @global_false)
+; NOLIT-NEXT:    [[C1:%.*]] = call i64 @compute(ptr @plus)
+; NOLIT-NEXT:    [[C2:%.*]] = call i64 @compute(ptr @minus)
+; NOLIT-NEXT:    [[ADD:%.*]] = add i64 [[C1]], [[C2]]
+; NOLIT-NEXT:    ret i64 [[ADD]]
+;
+;
+; NOLIT-LABEL: define ptr @select_op(
+; NOLIT-SAME: ptr [[FLAG:%.*]]) {
+; NOLIT-NEXT:    [[FLAG_VAL:%.*]] = load i1, ptr [[FLAG]], align 1
+; NOLIT-NEXT:    [[OP:%.*]] = select i1 [[FLAG_VAL]], ptr @plus, ptr @minus
+; NOLIT-NEXT:    ret ptr [[OP]]
+;
+;
+; NOLIT-LABEL: define internal i64 @compute(
+; NOLIT-SAME: ptr [[OP:%.*]]) {
+; NOLIT-NEXT:  [[ENTRY:.*:]]
+; NOLIT-NEXT:    [[RES:%.*]] = call i64 [[OP]](i64 1)
+; NOLIT-NEXT:    ret i64 [[RES]]
+;
+;
+; NOLIT-LABEL: define internal i64 @plus(
+; NOLIT-SAME: i64 [[X:%.*]]) {
+; NOLIT-NEXT:  [[ENTRY:.*:]]
+; NOLIT-NEXT:    [[SUM:%.*]] = add i64 [[X]], 1
+; NOLIT-NEXT:    ret i64 [[SUM]]
+;
+;
+; NOLIT-LABEL: define internal i64 @minus(
+; NOLIT-SAME: i64 [[X:%.*]]) {
+; NOLIT-NEXT:  [[ENTRY:.*:]]
+; NOLIT-NEXT:    [[DIFF:%.*]] = sub i64 [[X]], 1
+; NOLIT-NEXT:    ret i64 [[DIFF]]
+;
+;
+; NOLIT-LABEL: define internal ptr @select_op.specialized.1(
+; NOLIT-SAME: ptr [[FLAG:%.*]]) {
+; NOLIT-NEXT:    ret ptr poison
+;
+;
+; NOLIT-LABEL: define internal ptr @select_op.specialized.2(
+; NOLIT-SAME: ptr [[FLAG:%.*]]) {
+; NOLIT-NEXT:    ret ptr poison
+;

Original file line number	Diff line number	Diff line change
`@@ -1,4 +1,4 @@`
`1`		`-; RUN: opt -S --passes="default<O3>" < %s \| FileCheck %s`
	`1`	`+; RUN: opt -S --passes="default<O3>" -funcspec-for-literal-constant=false < %s \| FileCheck %s`
`2`	`2`
`3`	`3`	`define dso_local i32 @g0(i32 noundef %x) local_unnamed_addr {`
`4`	`4`	`entry:`
Original file line number	Diff line number	Diff line change
`@@ -1,4 +1,4 @@`
`1`		`-; RUN: opt -S --passes="ipsccp<func-spec>" < %s \| FileCheck %s`
	`1`	`+; RUN: opt -S --passes="ipsccp<func-spec>" -funcspec-for-literal-constant=false < %s \| FileCheck %s`
`2`	`2`	`define dso_local i32 @p0(i32 noundef %x) {`
`3`	`3`	`entry:`
`4`	`4`	`%add = add nsw i32 %x, 1`