[llvm] 4447603 - [InstCombine] Fold strtoul and strtoull and avoid PR #56293
Martin Sebor via llvm-commits
llvm-commits at lists.llvm.org
Tue Jul 26 13:12:44 PDT 2022
Author: Martin Sebor
Date: 2022-07-26T14:11:40-06:00
New Revision: 4447603616bb43af58af89a8a6f880853f546a39
URL: https://github.com/llvm/llvm-project/commit/4447603616bb43af58af89a8a6f880853f546a39
DIFF: https://github.com/llvm/llvm-project/commit/4447603616bb43af58af89a8a6f880853f546a39.diff
LOG: [InstCombine] Fold strtoul and strtoull and avoid PR #56293
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D129224
Added:
llvm/test/Transforms/InstCombine/str-int-4.ll
llvm/test/Transforms/InstCombine/str-int-5.ll
Modified:
llvm/include/llvm/Transforms/Utils/SimplifyLibCalls.h
llvm/lib/Analysis/ValueTracking.cpp
llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp
llvm/test/Transforms/InstCombine/str-int-2.ll
llvm/test/Transforms/InstCombine/str-int.ll
llvm/test/Transforms/InstCombine/strcall-no-nul.ll
Removed:
################################################################################
diff --git a/llvm/include/llvm/Transforms/Utils/SimplifyLibCalls.h b/llvm/include/llvm/Transforms/Utils/SimplifyLibCalls.h
index 79a44b6674452..1b2482a2363de 100644
--- a/llvm/include/llvm/Transforms/Utils/SimplifyLibCalls.h
+++ b/llvm/include/llvm/Transforms/Utils/SimplifyLibCalls.h
@@ -208,7 +208,7 @@ class LibCallSimplifier {
Value *optimizeIsAscii(CallInst *CI, IRBuilderBase &B);
Value *optimizeToAscii(CallInst *CI, IRBuilderBase &B);
Value *optimizeAtoi(CallInst *CI, IRBuilderBase &B);
- Value *optimizeStrtol(CallInst *CI, IRBuilderBase &B);
+ Value *optimizeStrToInt(CallInst *CI, IRBuilderBase &B, bool AsSigned);
// Formatting and IO Library Call Optimizations
Value *optimizeErrorReporting(CallInst *CI, IRBuilderBase &B,
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 89450c6c764e9..2dd671b4ab9e3 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -4266,9 +4266,10 @@ bool llvm::getConstantDataArrayInfo(const Value *V,
return true;
}
-/// This function computes the length of a null-terminated C string pointed to
-/// by V. If successful, it returns true and returns the string in Str.
-/// If unsuccessful, it returns false.
+/// Extract bytes from the initializer of the constant array V, which need
+/// not be a nul-terminated string. On success, store the bytes in Str and
+/// return true. When TrimAtNul is set, Str will contain only the bytes up
+/// to but not including the first nul. Return false on failure.
bool llvm::getConstantStringInfo(const Value *V, StringRef &Str,
uint64_t Offset, bool TrimAtNul) {
ConstantDataArraySlice Slice;
diff --git a/llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp b/llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp
index bca3b0538c5d8..03087d8370d59 100644
--- a/llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp
@@ -75,39 +75,109 @@ static bool callHasFP128Argument(const CallInst *CI) {
});
}
-static Value *convertStrToNumber(CallInst *CI, StringRef &Str, Value *EndPtr,
- int64_t Base, IRBuilderBase &B) {
+// Convert the entire string Str representing an integer in Base, up to
+// the terminating nul if present, to a constant according to the rules
+// of strtoul[l] or, when AsSigned is set, of strtol[l]. On success
+// return the result, otherwise null.
+// The function assumes the string is encoded in ASCII and carefully
+// avoids converting sequences (including "") that the corresponding
+// library call might fail and set errno for.
+static Value *convertStrToInt(CallInst *CI, StringRef &Str, Value *EndPtr,
+ uint64_t Base, bool AsSigned, IRBuilderBase &B) {
if (Base < 2 || Base > 36)
- // handle special zero base
if (Base != 0)
+ // Fail for an invalid base (required by POSIX).
return nullptr;
- char *End;
- std::string nptr = Str.str();
- errno = 0;
- long long int Result = strtoll(nptr.c_str(), &End, Base);
- if (errno)
- return nullptr;
+ // Strip leading whitespace.
+ for (unsigned i = 0; i != Str.size(); ++i)
+ if (!isSpace((unsigned char)Str[i])) {
+ Str = Str.substr(i);
+ break;
+ }
- // if we assume all possible target locales are ASCII supersets,
- // then if strtoll successfully parses a number on the host,
- // it will also successfully parse the same way on the target
- if (*End != '\0')
+ if (Str.empty())
+ // Fail for empty subject sequences (POSIX allows but doesn't require
+ // strtol[l]/strtoul[l] to fail with EINVAL).
return nullptr;
- if (!isIntN(CI->getType()->getPrimitiveSizeInBits(), Result))
- return nullptr;
+ // Strip but remember the sign.
+ bool Negate = Str[0] == '-';
+ if (Str[0] == '-' || Str[0] == '+') {
+ Str = Str.drop_front();
+ if (Str.empty())
+ // Fail for a sign with nothing after it.
+ return nullptr;
+ }
+
+ // Set Max to the absolute value of the minimum (for signed), or
+ // to the maximum (for unsigned) value representable in the type.
+ Type *RetTy = CI->getType();
+ unsigned NBits = RetTy->getPrimitiveSizeInBits();
+ uint64_t Max = AsSigned && Negate ? 1 : 0;
+ Max += AsSigned ? maxIntN(NBits) : maxUIntN(NBits);
+
+ // Autodetect Base if it's zero and consume the "0x" prefix.
+ if (Str.size() > 1) {
+ if (Str[0] == '0') {
+ if (toUpper((unsigned char)Str[1]) == 'X') {
+ if (Str.size() == 2 || (Base && Base != 16))
+ // Fail if Base doesn't allow the "0x" prefix or for the prefix
+ // alone that implementations like BSD set errno to EINVAL for.
+ return nullptr;
+
+ Str = Str.drop_front(2);
+ Base = 16;
+ }
+ else if (Base == 0)
+ Base = 8;
+ } else if (Base == 0)
+ Base = 10;
+ }
+ else if (Base == 0)
+ Base = 10;
+
+ // Convert the rest of the subject sequence, not including the sign,
+ // to its uint64_t representation (this assumes the source character
+ // set is ASCII).
+ uint64_t Result = 0;
+ for (unsigned i = 0; i != Str.size(); ++i) {
+ unsigned char DigVal = Str[i];
+ if (isDigit(DigVal))
+ DigVal = DigVal - '0';
+ else {
+ DigVal = toUpper(DigVal);
+ if (isAlpha(DigVal))
+ DigVal = DigVal - 'A' + 10;
+ else
+ return nullptr;
+ }
+
+ if (DigVal >= Base)
+ // Fail if the digit is not valid in the Base.
+ return nullptr;
+
+ // Add the digit and fail if the result is not representable in
+ // the (unsigned form of the) destination type.
+ bool VFlow;
+ Result = SaturatingMultiplyAdd(Result, Base, (uint64_t)DigVal, &VFlow);
+ if (VFlow || Result > Max)
+ return nullptr;
+ }
if (EndPtr) {
// Store the pointer to the end.
- uint64_t ILen = End - nptr.c_str();
- Value *Off = B.getInt64(ILen);
+ Value *Off = B.getInt64(Str.size());
Value *StrBeg = CI->getArgOperand(0);
Value *StrEnd = B.CreateInBoundsGEP(B.getInt8Ty(), StrBeg, Off, "endptr");
B.CreateStore(StrEnd, EndPtr);
}
- return ConstantInt::get(CI->getType(), Result);
+ if (Negate)
+ // Unsigned negation doesn't overflow.
+ Result = -Result;
+
+ return ConstantInt::get(RetTy, Result);
}
static bool isOnlyUsedInComparisonWithZero(Value *V) {
@@ -2531,27 +2601,35 @@ Value *LibCallSimplifier::optimizeToAscii(CallInst *CI, IRBuilderBase &B) {
ConstantInt::get(CI->getType(), 0x7F));
}
+// Fold calls to atoi, atol, and atoll.
Value *LibCallSimplifier::optimizeAtoi(CallInst *CI, IRBuilderBase &B) {
+ CI->addParamAttr(0, Attribute::NoCapture);
+
StringRef Str;
if (!getConstantStringInfo(CI->getArgOperand(0), Str))
return nullptr;
- return convertStrToNumber(CI, Str, nullptr, 10, B);
+ return convertStrToInt(CI, Str, nullptr, 10, /*AsSigned=*/true, B);
}
-Value *LibCallSimplifier::optimizeStrtol(CallInst *CI, IRBuilderBase &B) {
- StringRef Str;
- if (!getConstantStringInfo(CI->getArgOperand(0), Str))
- return nullptr;
-
+// Fold calls to strtol, strtoll, strtoul, and strtoull.
+Value *LibCallSimplifier::optimizeStrToInt(CallInst *CI, IRBuilderBase &B,
+ bool AsSigned) {
Value *EndPtr = CI->getArgOperand(1);
- if (isa<ConstantPointerNull>(EndPtr))
+ if (isa<ConstantPointerNull>(EndPtr)) {
+ // With a null EndPtr, this function won't capture the main argument.
+ // It would be readonly too, except that it still may write to errno.
+ CI->addParamAttr(0, Attribute::NoCapture);
EndPtr = nullptr;
- else if (!isKnownNonZero(EndPtr, DL))
+ } else if (!isKnownNonZero(EndPtr, DL))
+ return nullptr;
+
+ StringRef Str;
+ if (!getConstantStringInfo(CI->getArgOperand(0), Str))
return nullptr;
if (ConstantInt *CInt = dyn_cast<ConstantInt>(CI->getArgOperand(2))) {
- return convertStrToNumber(CI, Str, EndPtr, CInt->getSExtValue(), B);
+ return convertStrToInt(CI, Str, EndPtr, CInt->getSExtValue(), AsSigned, B);
}
return nullptr;
@@ -3390,7 +3468,10 @@ Value *LibCallSimplifier::optimizeCall(CallInst *CI, IRBuilderBase &Builder) {
return optimizeAtoi(CI, Builder);
case LibFunc_strtol:
case LibFunc_strtoll:
- return optimizeStrtol(CI, Builder);
+ return optimizeStrToInt(CI, Builder, /*AsSigned=*/true);
+ case LibFunc_strtoul:
+ case LibFunc_strtoull:
+ return optimizeStrToInt(CI, Builder, /*AsSigned=*/false);
case LibFunc_printf:
return optimizePrintF(CI, Builder);
case LibFunc_sprintf:
diff --git a/llvm/test/Transforms/InstCombine/str-int-2.ll b/llvm/test/Transforms/InstCombine/str-int-2.ll
index d81ecbb774029..5ccf39500b836 100644
--- a/llvm/test/Transforms/InstCombine/str-int-2.ll
+++ b/llvm/test/Transforms/InstCombine/str-int-2.ll
@@ -81,7 +81,7 @@ define i64 @strtol_not_const_str(i8* %s) #0 {
define i32 @atoi_not_const_str(i8* %s) #0 {
; CHECK-LABEL: @atoi_not_const_str(
-; CHECK-NEXT: [[CALL:%.*]] = call i32 @atoi(i8* [[S:%.*]])
+; CHECK-NEXT: [[CALL:%.*]] = call i32 @atoi(i8* nocapture [[S:%.*]])
; CHECK-NEXT: ret i32 [[CALL]]
;
%call = call i32 @atoi(i8* %s) #4
diff --git a/llvm/test/Transforms/InstCombine/str-int-4.ll b/llvm/test/Transforms/InstCombine/str-int-4.ll
new file mode 100644
index 0000000000000..0de3808b3a4f5
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/str-int-4.ll
@@ -0,0 +1,421 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; Verify that calls to strtol and strtoll are interpreted correctly even
+; in corner cases (or not folded).
+;
+; RUN: opt < %s -passes=instcombine -S | FileCheck %s
+
+declare i32 @strtol(i8*, i8**, i32)
+declare i64 @strtoll(i8*, i8**, i32)
+
+
+; All POSIX whitespace characters.
+ at ws = constant [7 x i8] c"\09\0d\0a\0b\0c \00"
+
+; A negative and positive number preceded by all POSIX whitespace.
+ at ws_im123 = constant [11 x i8] c"\09\0d\0a\0b\0c -123\00"
+ at ws_ip234 = constant [11 x i8] c"\09\0d\0a\0b\0c +234\00"
+
+ at wsplus = constant [3 x i8] c" +\00"
+ at i0 = constant [3 x i8] c" 0\00"
+ at i8 = constant [3 x i8] c" 8\00"
+ at i9 = constant [3 x i8] c" 9\00"
+ at ia = constant [3 x i8] c" a\00"
+ at i08 = constant [3 x i8] c"08\00"
+ at x0x = constant [3 x i8] c"0x\00"
+ at wsplusws0 = constant [5 x i8] c" + 0\00"
+ at i19azAZ = constant [7 x i8] c"19azAZ\00"
+ at i32min = constant [13 x i8] c" -2147483648\00"
+ at i32min_m1 = constant [13 x i8] c" -2147483649\00"
+ at o32min = constant [15 x i8] c" +020000000000\00"
+ at mo32min = constant [15 x i8] c" -020000000000\00"
+ at x32min = constant [13 x i8] c" +0x80000000\00"
+ at mx32min = constant [13 x i8] c" -0x80000000\00"
+
+ at i32max = constant [12 x i8] c" 2147483647\00"
+ at x32max = constant [12 x i8] c" 0x7fffffff\00"
+ at i32max_p1 = constant [12 x i8] c" 2147483648\00"
+
+ at ui32max = constant [12 x i8] c" 4294967295\00"
+ at ui32max_p1 = constant [12 x i8] c" 4294967296\00"
+
+
+; Exercise folding calls to 32-bit strtol.
+
+define void @fold_strtol(i32* %ps) {
+; CHECK-LABEL: @fold_strtol(
+; CHECK-NEXT: store i32 -123, i32* [[PS:%.*]], align 4
+; CHECK-NEXT: [[PS1:%.*]] = getelementptr i32, i32* [[PS]], i64 1
+; CHECK-NEXT: store i32 234, i32* [[PS1]], align 4
+; CHECK-NEXT: [[PS2:%.*]] = getelementptr i32, i32* [[PS]], i64 2
+; CHECK-NEXT: store i32 0, i32* [[PS2]], align 4
+; CHECK-NEXT: [[PS3:%.*]] = getelementptr i32, i32* [[PS]], i64 3
+; CHECK-NEXT: store i32 9, i32* [[PS3]], align 4
+; CHECK-NEXT: [[PS4:%.*]] = getelementptr i32, i32* [[PS]], i64 4
+; CHECK-NEXT: store i32 10, i32* [[PS4]], align 4
+; CHECK-NEXT: [[PS5:%.*]] = getelementptr i32, i32* [[PS]], i64 5
+; CHECK-NEXT: store i32 76095035, i32* [[PS5]], align 4
+; CHECK-NEXT: [[PS6:%.*]] = getelementptr i32, i32* [[PS]], i64 6
+; CHECK-NEXT: store i32 -2147483648, i32* [[PS6]], align 4
+; CHECK-NEXT: [[PS7:%.*]] = getelementptr i32, i32* [[PS]], i64 7
+; CHECK-NEXT: store i32 -2147483648, i32* [[PS7]], align 4
+; CHECK-NEXT: [[PS8:%.*]] = getelementptr i32, i32* [[PS]], i64 8
+; CHECK-NEXT: store i32 -2147483648, i32* [[PS8]], align 4
+; CHECK-NEXT: [[PS9:%.*]] = getelementptr i32, i32* [[PS]], i64 9
+; CHECK-NEXT: store i32 -2147483648, i32* [[PS9]], align 4
+; CHECK-NEXT: [[PS10:%.*]] = getelementptr i32, i32* [[PS]], i64 10
+; CHECK-NEXT: store i32 2147483647, i32* [[PS10]], align 4
+; CHECK-NEXT: [[PS11:%.*]] = getelementptr i32, i32* [[PS]], i64 11
+; CHECK-NEXT: store i32 2147483647, i32* [[PS11]], align 4
+; CHECK-NEXT: ret void
+;
+; Fold a valid sequence with leading POSIX whitespace and a minus to -123.
+ %pwsm123 = getelementptr [11 x i8], [11 x i8]* @ws_im123, i32 0, i32 0
+ %im123 = call i32 @strtol(i8* %pwsm123, i8** null, i32 10)
+ %ps0 = getelementptr i32, i32* %ps, i32 0
+ store i32 %im123, i32* %ps0
+
+; Fold a valid sequence with leading POSIX whitespace and a plus to +234.
+ %pwsp234 = getelementptr [11 x i8], [11 x i8]* @ws_ip234, i32 0, i32 0
+ %ip234 = call i32 @strtol(i8* %pwsp234, i8** null, i32 10)
+ %ps1 = getelementptr i32, i32* %ps, i32 1
+ store i32 %ip234, i32* %ps1
+
+; Fold " 0" in base 0 to verify correct base autodetection.
+ %psi0 = getelementptr [3 x i8], [3 x i8]* @i0, i32 0, i32 0
+ %i0 = call i32 @strtol(i8* %psi0, i8** null, i32 0)
+ %ps2 = getelementptr i32, i32* %ps, i32 2
+ store i32 %i0, i32* %ps2
+
+; Fold " 9" in base 0 to verify correct base autodetection.
+ %psi9 = getelementptr [3 x i8], [3 x i8]* @i9, i32 0, i32 0
+ %i9 = call i32 @strtol(i8* %psi9, i8** null, i32 0)
+ %ps3 = getelementptr i32, i32* %ps, i32 3
+ store i32 %i9, i32* %ps3
+
+; Fold " a" in base 16 to 10.
+ %psia = getelementptr [3 x i8], [3 x i8]* @ia, i32 0, i32 0
+ %ia = call i32 @strtol(i8* %psia, i8** null, i32 16)
+ %ps4 = getelementptr i32, i32* %ps, i32 4
+ store i32 %ia, i32* %ps4
+
+; Fold "19azAZ" in base 36 to 76095035.
+ %psi19azAZ = getelementptr [7 x i8], [7 x i8]* @i19azAZ, i32 0, i32 0
+ %i19azAZ = call i32 @strtol(i8* %psi19azAZ, i8** null, i32 36)
+ %ps5 = getelementptr i32, i32* %ps, i32 5
+ store i32 %i19azAZ, i32* %ps5
+
+; Fold INT32_MIN.
+ %psmin = getelementptr [13 x i8], [13 x i8]* @i32min, i32 0, i32 0
+ %min = call i32 @strtol(i8* %psmin, i8** null, i32 10)
+ %ps6 = getelementptr i32, i32* %ps, i32 6
+ store i32 %min, i32* %ps6
+
+; Fold -INT32_MIN in octal.
+ %psmo32min = getelementptr [15 x i8], [15 x i8]* @mo32min, i32 0, i32 0
+ %mo32min = call i32 @strtol(i8* %psmo32min, i8** null, i32 0)
+ %ps7 = getelementptr i32, i32* %ps, i32 7
+ store i32 %mo32min, i32* %ps7
+
+; Fold -INT32_MIN in hex and base 0.
+ %psmx32min = getelementptr [13 x i8], [13 x i8]* @mx32min, i32 0, i32 0
+ %mx32min_0 = call i32 @strtol(i8* %psmx32min, i8** null, i32 0)
+ %ps8 = getelementptr i32, i32* %ps, i32 8
+ store i32 %mx32min_0, i32* %ps8
+
+; Fold -INT32_MIN in hex and base 16.
+ %mx32min_16 = call i32 @strtol(i8* %psmx32min, i8** null, i32 16)
+ %ps9 = getelementptr i32, i32* %ps, i32 9
+ store i32 %mx32min_16, i32* %ps9
+
+; Fold INT32_MAX.
+ %psmax = getelementptr [12 x i8], [12 x i8]* @i32max, i32 0, i32 0
+ %max = call i32 @strtol(i8* %psmax, i8** null, i32 10)
+ %ps10 = getelementptr i32, i32* %ps, i32 10
+ store i32 %max, i32* %ps10
+
+; Fold INT32_MAX in hex.
+ %psxmax = getelementptr [12 x i8], [12 x i8]* @x32max, i32 0, i32 0
+ %xmax = call i32 @strtol(i8* %psxmax, i8** null, i32 0)
+ %ps11 = getelementptr i32, i32* %ps, i32 11
+ store i32 %xmax, i32* %ps11
+
+ ret void
+}
+
+
+; Exercise not folding calls to 32-bit strtol.
+
+define void @call_strtol(i32* %ps) {
+; CHECK-LABEL: @call_strtol(
+; CHECK-NEXT: [[MINM1:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([13 x i8], [13 x i8]* @i32min_m1, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: store i32 [[MINM1]], i32* [[PS:%.*]], align 4
+; CHECK-NEXT: [[MAXP1:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([12 x i8], [12 x i8]* @i32max_p1, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: [[PS1:%.*]] = getelementptr i32, i32* [[PS]], i64 1
+; CHECK-NEXT: store i32 [[MAXP1]], i32* [[PS1]], align 4
+; CHECK-NEXT: [[IPLUS:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([3 x i8], [3 x i8]* @wsplus, i64 0, i64 0), i8** null, i32 0)
+; CHECK-NEXT: [[PS2:%.*]] = getelementptr i32, i32* [[PS]], i64 2
+; CHECK-NEXT: store i32 [[IPLUS]], i32* [[PS2]], align 4
+; CHECK-NEXT: [[IA:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([3 x i8], [3 x i8]* @ia, i64 0, i64 0), i8** null, i32 0)
+; CHECK-NEXT: [[PS3:%.*]] = getelementptr i32, i32* [[PS]], i64 3
+; CHECK-NEXT: store i32 [[IA]], i32* [[PS3]], align 4
+; CHECK-NEXT: [[I8:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([3 x i8], [3 x i8]* @i8, i64 0, i64 0), i8** null, i32 8)
+; CHECK-NEXT: [[PS4:%.*]] = getelementptr i32, i32* [[PS]], i64 4
+; CHECK-NEXT: store i32 [[I8]], i32* [[PS4]], align 4
+; CHECK-NEXT: [[I0X:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([3 x i8], [3 x i8]* @x0x, i64 0, i64 0), i8** null, i32 0)
+; CHECK-NEXT: [[PS5:%.*]] = getelementptr i32, i32* [[PS]], i64 5
+; CHECK-NEXT: store i32 [[I0X]], i32* [[PS5]], align 4
+; CHECK-NEXT: [[IWSPWS0:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([5 x i8], [5 x i8]* @wsplusws0, i64 0, i64 0), i8** null, i32 0)
+; CHECK-NEXT: [[PS6:%.*]] = getelementptr i32, i32* [[PS]], i64 6
+; CHECK-NEXT: store i32 [[IWSPWS0]], i32* [[PS6]], align 4
+; CHECK-NEXT: [[I19AZAZ:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([7 x i8], [7 x i8]* @i19azAZ, i64 0, i64 0), i8** null, i32 35)
+; CHECK-NEXT: [[PS7:%.*]] = getelementptr i32, i32* [[PS]], i64 7
+; CHECK-NEXT: store i32 [[I19AZAZ]], i32* [[PS7]], align 4
+; CHECK-NEXT: [[O32MIN:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([15 x i8], [15 x i8]* @o32min, i64 0, i64 0), i8** null, i32 0)
+; CHECK-NEXT: [[PS8:%.*]] = getelementptr i32, i32* [[PS]], i64 8
+; CHECK-NEXT: store i32 [[O32MIN]], i32* [[PS8]], align 4
+; CHECK-NEXT: [[X32MIN:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([13 x i8], [13 x i8]* @x32min, i64 0, i64 0), i8** null, i32 0)
+; CHECK-NEXT: [[PS9:%.*]] = getelementptr i32, i32* [[PS]], i64 9
+; CHECK-NEXT: store i32 [[X32MIN]], i32* [[PS9]], align 4
+; CHECK-NEXT: [[X32MIN_10:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([13 x i8], [13 x i8]* @x32min, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: [[PS10:%.*]] = getelementptr i32, i32* [[PS]], i64 10
+; CHECK-NEXT: store i32 [[X32MIN_10]], i32* [[PS10]], align 4
+; CHECK-NEXT: [[NWS:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([7 x i8], [7 x i8]* @ws, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: [[PS11:%.*]] = getelementptr i32, i32* [[PS]], i64 11
+; CHECK-NEXT: store i32 [[NWS]], i32* [[PS11]], align 4
+; CHECK-NEXT: [[NWSP6:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([7 x i8], [7 x i8]* @ws, i64 0, i64 6), i8** null, i32 10)
+; CHECK-NEXT: [[PS12:%.*]] = getelementptr i32, i32* [[PS]], i64 12
+; CHECK-NEXT: store i32 [[NWSP6]], i32* [[PS12]], align 4
+; CHECK-NEXT: [[I0B1:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([3 x i8], [3 x i8]* @i0, i64 0, i64 0), i8** null, i32 1)
+; CHECK-NEXT: [[PS13:%.*]] = getelementptr i32, i32* [[PS]], i64 13
+; CHECK-NEXT: store i32 [[I0B1]], i32* [[PS13]], align 4
+; CHECK-NEXT: [[I0B256:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([3 x i8], [3 x i8]* @i0, i64 0, i64 0), i8** null, i32 256)
+; CHECK-NEXT: [[PS14:%.*]] = getelementptr i32, i32* [[PS]], i64 14
+; CHECK-NEXT: store i32 [[I0B256]], i32* [[PS14]], align 4
+; CHECK-NEXT: ret void
+;
+
+; Do not fold the result of conversion that's less than INT32_MIN.
+ %psminm1 = getelementptr [13 x i8], [13 x i8]* @i32min_m1, i32 0, i32 0
+ %minm1 = call i32 @strtol(i8* %psminm1, i8** null, i32 10)
+ %ps0 = getelementptr i32, i32* %ps, i32 0
+ store i32 %minm1, i32* %ps0
+
+; Do not fold the result of conversion that's greater than INT32_MAX.
+ %psmaxp1 = getelementptr [12 x i8], [12 x i8]* @i32max_p1, i32 0, i32 0
+ %maxp1 = call i32 @strtol(i8* %psmaxp1, i8** null, i32 10)
+ %ps1 = getelementptr i32, i32* %ps, i32 1
+ store i32 %maxp1, i32* %ps1
+
+; Do not fold " +".
+ %psplus = getelementptr [3 x i8], [3 x i8]* @wsplus, i32 0, i32 0
+ %iplus = call i32 @strtol(i8* %psplus, i8** null, i32 0)
+ %ps2 = getelementptr i32, i32* %ps, i32 2
+ store i32 %iplus, i32* %ps2
+
+; Do not fold " a" in base 0.
+ %psia = getelementptr [3 x i8], [3 x i8]* @ia, i32 0, i32 0
+ %ia = call i32 @strtol(i8* %psia, i8** null, i32 0)
+ %ps3 = getelementptr i32, i32* %ps, i32 3
+ store i32 %ia, i32* %ps3
+
+; Do not fold " 8" in base 8.
+ %psi8 = getelementptr [3 x i8], [3 x i8]* @i8, i32 0, i32 0
+ %i8 = call i32 @strtol(i8* %psi8, i8** null, i32 8)
+ %ps4 = getelementptr i32, i32* %ps, i32 4
+ store i32 %i8, i32* %ps4
+
+; Do not fold the "0x" alone in base 0 that some implementations (e.g.,
+; BSD and Darwin) set errno to EINVAL for.
+ %psx0x = getelementptr [3 x i8], [3 x i8]* @x0x, i32 0, i32 0
+ %i0x = call i32 @strtol(i8* %psx0x, i8** null, i32 0)
+ %ps5 = getelementptr i32, i32* %ps, i32 5
+ store i32 %i0x, i32* %ps5
+
+; Do not fold " + 0".
+ %pwspws0 = getelementptr [5 x i8], [5 x i8]* @wsplusws0, i32 0, i32 0
+ %iwspws0 = call i32 @strtol(i8* %pwspws0, i8** null, i32 0)
+ %ps6 = getelementptr i32, i32* %ps, i32 6
+ store i32 %iwspws0, i32* %ps6
+
+; Do not fold "19azAZ" in base 35.
+ %psi19azAZ = getelementptr [7 x i8], [7 x i8]* @i19azAZ, i32 0, i32 0
+ %i19azAZ = call i32 @strtol(i8* %psi19azAZ, i8** null, i32 35)
+ %ps7 = getelementptr i32, i32* %ps, i32 7
+ store i32 %i19azAZ, i32* %ps7
+
+; Do not fold INT32_MIN in octal.
+ %pso32min = getelementptr [15 x i8], [15 x i8]* @o32min, i32 0, i32 0
+ %o32min = call i32 @strtol(i8* %pso32min, i8** null, i32 0)
+ %ps8 = getelementptr i32, i32* %ps, i32 8
+ store i32 %o32min, i32* %ps8
+
+; Do not fold INT32_MIN in hex.
+ %psx32min = getelementptr [13 x i8], [13 x i8]* @x32min, i32 0, i32 0
+ %x32min = call i32 @strtol(i8* %psx32min, i8** null, i32 0)
+ %ps9 = getelementptr i32, i32* %ps, i32 9
+ store i32 %x32min, i32* %ps9
+
+; Do not fold INT32_MIN in hex in base 10.
+ %x32min_10 = call i32 @strtol(i8* %psx32min, i8** null, i32 10)
+ %ps10 = getelementptr i32, i32* %ps, i32 10
+ store i32 %x32min_10, i32* %ps10
+
+; Do not fold a sequence consisting of just whitespace characters.
+ %psws = getelementptr [7 x i8], [7 x i8]* @ws, i32 0, i32 0
+ %nws = call i32 @strtol(i8* %psws, i8** null, i32 10)
+ %ps11 = getelementptr i32, i32* %ps, i32 11
+ store i32 %nws, i32* %ps11
+
+; Do not fold an empty sequence.
+ %pswsp6 = getelementptr [7 x i8], [7 x i8]* @ws, i32 0, i32 6
+ %nwsp6 = call i32 @strtol(i8* %pswsp6, i8** null, i32 10)
+ %ps12 = getelementptr i32, i32* %ps, i32 12
+ store i32 %nwsp6, i32* %ps12
+
+; Do not fold the invalid base 1.
+ %psi0 = getelementptr [3 x i8], [3 x i8]* @i0, i32 0, i32 0
+ %i0b1 = call i32 @strtol(i8* %psi0, i8** null, i32 1)
+ %ps13 = getelementptr i32, i32* %ps, i32 13
+ store i32 %i0b1, i32* %ps13
+
+; Do not fold the invalid base 256.
+ %i0b256 = call i32 @strtol(i8* %psi0, i8** null, i32 256)
+ %ps14 = getelementptr i32, i32* %ps, i32 14
+ store i32 %i0b256, i32* %ps14
+
+ ret void
+}
+
+
+ at i64min = constant [22 x i8] c" -9223372036854775808\00"
+ at i64min_m1 = constant [22 x i8] c" -9223372036854775809\00"
+
+ at i64max = constant [21 x i8] c" 9223372036854775807\00"
+ at i64max_p1 = constant [21 x i8] c" 9223372036854775808\00"
+
+ at ui64max = constant [22 x i8] c" 18446744073709551615\00"
+ at ui64max_p1 = constant [22 x i8] c" 18446744073709551616\00"
+
+
+; Exercise folding calls to the 64-bit strtoll.
+
+define void @fold_strtoll(i64* %ps) {
+; CHECK-LABEL: @fold_strtoll(
+; CHECK-NEXT: store i64 -123, i64* [[PS:%.*]], align 4
+; CHECK-NEXT: [[PS1:%.*]] = getelementptr i64, i64* [[PS]], i64 1
+; CHECK-NEXT: store i64 234, i64* [[PS1]], align 4
+; CHECK-NEXT: [[PS2:%.*]] = getelementptr i64, i64* [[PS]], i64 2
+; CHECK-NEXT: store i64 -9223372036854775808, i64* [[PS2]], align 4
+; CHECK-NEXT: [[PS3:%.*]] = getelementptr i64, i64* [[PS]], i64 3
+; CHECK-NEXT: store i64 9223372036854775807, i64* [[PS3]], align 4
+; CHECK-NEXT: ret void
+;
+; Fold a valid sequence with leading POSIX whitespace and a minus to -123.
+ %pwsm123 = getelementptr [11 x i8], [11 x i8]* @ws_im123, i32 0, i32 0
+ %im123 = call i64 @strtoll(i8* %pwsm123, i8** null, i32 10)
+ %ps0 = getelementptr i64, i64* %ps, i32 0
+ store i64 %im123, i64* %ps0
+
+; Fold a valid sequence with leading POSIX whitespace and a plus to +234.
+ %pwsp234 = getelementptr [11 x i8], [11 x i8]* @ws_ip234, i32 0, i32 0
+ %ip234 = call i64 @strtoll(i8* %pwsp234, i8** null, i32 10)
+ %ps1 = getelementptr i64, i64* %ps, i32 1
+ store i64 %ip234, i64* %ps1
+
+; Fold INT64_MIN.
+ %psmin = getelementptr [22 x i8], [22 x i8]* @i64min, i32 0, i32 0
+ %min = call i64 @strtoll(i8* %psmin, i8** null, i32 10)
+ %ps2 = getelementptr i64, i64* %ps, i32 2
+ store i64 %min, i64* %ps2
+
+; Fold INT64_MAX.
+ %psmax = getelementptr [21 x i8], [21 x i8]* @i64max, i32 0, i32 0
+ %max = call i64 @strtoll(i8* %psmax, i8** null, i32 10)
+ %ps3 = getelementptr i64, i64* %ps, i32 3
+ store i64 %max, i64* %ps3
+
+ ret void
+}
+
+
+; Exercise not folding calls to the 64-bit strtoll.
+
+define void @call_strtoll(i64* %ps) {
+; CHECK-LABEL: @call_strtoll(
+; CHECK-NEXT: [[MINM1:%.*]] = call i64 @strtoll(i8* nocapture getelementptr inbounds ([22 x i8], [22 x i8]* @i64min_m1, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: store i64 [[MINM1]], i64* [[PS:%.*]], align 4
+; CHECK-NEXT: [[MAXP1:%.*]] = call i64 @strtoll(i8* nocapture getelementptr inbounds ([21 x i8], [21 x i8]* @i64max_p1, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: [[PS1:%.*]] = getelementptr i64, i64* [[PS]], i64 1
+; CHECK-NEXT: store i64 [[MAXP1]], i64* [[PS1]], align 4
+; CHECK-NEXT: [[NWS:%.*]] = call i64 @strtoll(i8* nocapture getelementptr inbounds ([7 x i8], [7 x i8]* @ws, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: [[PS2:%.*]] = getelementptr i64, i64* [[PS]], i64 2
+; CHECK-NEXT: store i64 [[NWS]], i64* [[PS2]], align 4
+; CHECK-NEXT: [[NWSP6:%.*]] = call i64 @strtoll(i8* nocapture getelementptr inbounds ([7 x i8], [7 x i8]* @ws, i64 0, i64 6), i8** null, i32 10)
+; CHECK-NEXT: [[PS3:%.*]] = getelementptr i64, i64* [[PS]], i64 3
+; CHECK-NEXT: store i64 [[NWSP6]], i64* [[PS3]], align 4
+; CHECK-NEXT: ret void
+;
+; Do not fold the result of conversion that's less than INT64_MIN.
+ %psminm1 = getelementptr [22 x i8], [22 x i8]* @i64min_m1, i32 0, i32 0
+ %minm1 = call i64 @strtoll(i8* %psminm1, i8** null, i32 10)
+ %ps0 = getelementptr i64, i64* %ps, i32 0
+ store i64 %minm1, i64* %ps0
+
+; Do not fold the result of conversion that's greater than INT64_MAX.
+ %psmaxp1 = getelementptr [21 x i8], [21 x i8]* @i64max_p1, i32 0, i32 0
+ %maxp1 = call i64 @strtoll(i8* %psmaxp1, i8** null, i32 10)
+ %ps1 = getelementptr i64, i64* %ps, i32 1
+ store i64 %maxp1, i64* %ps1
+
+; Do not fold a sequence consisting of just whitespace characters.
+ %psws = getelementptr [7 x i8], [7 x i8]* @ws, i32 0, i32 0
+ %nws = call i64 @strtoll(i8* %psws, i8** null, i32 10)
+ %ps2 = getelementptr i64, i64* %ps, i32 2
+ store i64 %nws, i64* %ps2
+
+; Do not fold an empty sequence.
+ %pswsp6 = getelementptr [7 x i8], [7 x i8]* @ws, i32 0, i32 6
+ %nwsp6 = call i64 @strtoll(i8* %pswsp6, i8** null, i32 10)
+ %ps3 = getelementptr i64, i64* %ps, i32 3
+ store i64 %nwsp6, i64* %ps3
+
+ ret void
+}
+
+ at i_1_2_3_ = constant [9 x i8] c" 1 2\09\3\0a\00";
+
+; Verify that strings of digits that are followed by whitespace are not
+; folded (the whitespace could be interpreted in locales other than C
+; as part of the leading digits, such as "123 456" is interepreted as
+; 123456 in the French locale).
+
+define void @call_strtol_trailing_space(i32* %ps) {
+; CHECK-LABEL: @call_strtol_trailing_space(
+; CHECK-NEXT: [[N1:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([9 x i8], [9 x i8]* @i_1_2_3_, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: [[PS1:%.*]] = getelementptr i32, i32* [[PS:%.*]], i64 1
+; CHECK-NEXT: store i32 [[N1]], i32* [[PS1]], align 4
+; CHECK-NEXT: [[N2:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([9 x i8], [9 x i8]* @i_1_2_3_, i64 0, i64 2), i8** null, i32 10)
+; CHECK-NEXT: [[PS2:%.*]] = getelementptr i32, i32* [[PS]], i64 2
+; CHECK-NEXT: store i32 [[N2]], i32* [[PS2]], align 4
+; CHECK-NEXT: [[N3:%.*]] = call i32 @strtol(i8* nocapture getelementptr inbounds ([9 x i8], [9 x i8]* @i_1_2_3_, i64 0, i64 4), i8** null, i32 10)
+; CHECK-NEXT: [[PS3:%.*]] = getelementptr i32, i32* [[PS]], i64 3
+; CHECK-NEXT: store i32 [[N3]], i32* [[PS3]], align 4
+; CHECK-NEXT: ret void
+;
+ %p1 = getelementptr [9 x i8], [9 x i8]* @i_1_2_3_, i32 0, i32 0
+ %n1 = call i32 @strtol(i8* %p1, i8** null, i32 10)
+ %ps1 = getelementptr i32, i32* %ps, i32 1
+ store i32 %n1, i32* %ps1
+
+ %p2 = getelementptr [9 x i8], [9 x i8]* @i_1_2_3_, i32 0, i32 2
+ %n2 = call i32 @strtol(i8* %p2, i8** null, i32 10)
+ %ps2 = getelementptr i32, i32* %ps, i32 2
+ store i32 %n2, i32* %ps2
+
+ %p3 = getelementptr [9 x i8], [9 x i8]* @i_1_2_3_, i32 0, i32 4
+ %n3 = call i32 @strtol(i8* %p3, i8** null, i32 10)
+ %ps3 = getelementptr i32, i32* %ps, i32 3
+ store i32 %n3, i32* %ps3
+
+ ret void
+}
diff --git a/llvm/test/Transforms/InstCombine/str-int-5.ll b/llvm/test/Transforms/InstCombine/str-int-5.ll
new file mode 100644
index 0000000000000..87e3455a79717
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/str-int-5.ll
@@ -0,0 +1,331 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; Verify that calls to strtoul and strtoull are interpreted correctly even
+; in corner cases (or not folded).
+;
+; RUN: opt < %s -passes=instcombine -S | FileCheck %s
+
+declare i32 @strtoul(i8*, i8**, i32)
+declare i64 @strtoull(i8*, i8**, i32)
+
+
+; All POSIX whitespace characters.
+ at ws = constant [7 x i8] c"\09\0d\0a\0b\0c \00"
+
+; A negative and positive number preceded by all POSIX whitespace.
+ at ws_im123 = constant [11 x i8] c"\09\0d\0a\0b\0c -123\00"
+ at ws_ip234 = constant [11 x i8] c"\09\0d\0a\0b\0c +234\00"
+
+ at i32min = constant [13 x i8] c" -2147483648\00"
+ at i32min_m1 = constant [13 x i8] c" -2147483649\00"
+ at o32min = constant [15 x i8] c" +020000000000\00"
+ at mo32min = constant [15 x i8] c" -020000000000\00"
+ at x32min = constant [13 x i8] c" +0x80000000\00"
+ at mx32min = constant [13 x i8] c" +0x80000000\00"
+
+ at i32max = constant [12 x i8] c" 2147483647\00"
+ at i32max_p1 = constant [12 x i8] c" 2147483648\00"
+ at mX01 = constant [6 x i8] c" -0X1\00"
+
+ at ui32max = constant [12 x i8] c" 4294967295\00"
+ at ui32max_p1 = constant [12 x i8] c" 4294967296\00"
+
+ at i64min = constant [22 x i8] c" -9223372036854775808\00"
+ at i64min_m1 = constant [22 x i8] c" -9223372036854775809\00"
+
+ at i64max = constant [21 x i8] c" 9223372036854775807\00"
+ at i64max_p1 = constant [21 x i8] c" 9223372036854775808\00"
+
+ at ui64max = constant [22 x i8] c" 18446744073709551615\00"
+ at x64max = constant [20 x i8] c" 0xffffffffffffffff\00"
+ at ui64max_p1 = constant [22 x i8] c" 18446744073709551616\00"
+
+
+; Exercise folding calls to 32-bit strtoul.
+
+define void @fold_strtoul(i32* %ps) {
+; CHECK-LABEL: @fold_strtoul(
+; CHECK-NEXT: store i32 -123, i32* [[PS:%.*]], align 4
+; CHECK-NEXT: [[PS1:%.*]] = getelementptr i32, i32* [[PS]], i64 1
+; CHECK-NEXT: store i32 234, i32* [[PS1]], align 4
+; CHECK-NEXT: [[PS2:%.*]] = getelementptr i32, i32* [[PS]], i64 2
+; CHECK-NEXT: store i32 2147483647, i32* [[PS2]], align 4
+; CHECK-NEXT: [[PS3:%.*]] = getelementptr i32, i32* [[PS]], i64 3
+; CHECK-NEXT: store i32 -2147483648, i32* [[PS3]], align 4
+; CHECK-NEXT: [[PS4:%.*]] = getelementptr i32, i32* [[PS]], i64 4
+; CHECK-NEXT: store i32 -2147483648, i32* [[PS4]], align 4
+; CHECK-NEXT: [[PS5:%.*]] = getelementptr i32, i32* [[PS]], i64 5
+; CHECK-NEXT: store i32 -2147483648, i32* [[PS5]], align 4
+; CHECK-NEXT: [[PS6:%.*]] = getelementptr i32, i32* [[PS]], i64 6
+; CHECK-NEXT: store i32 -2147483648, i32* [[PS6]], align 4
+; CHECK-NEXT: [[PS7:%.*]] = getelementptr i32, i32* [[PS]], i64 7
+; CHECK-NEXT: store i32 -2147483648, i32* [[PS7]], align 4
+; CHECK-NEXT: [[PS8:%.*]] = getelementptr i32, i32* [[PS]], i64 8
+; CHECK-NEXT: store i32 2147483647, i32* [[PS8]], align 4
+; CHECK-NEXT: [[PS9:%.*]] = getelementptr i32, i32* [[PS]], i64 9
+; CHECK-NEXT: store i32 -1, i32* [[PS9]], align 4
+; CHECK-NEXT: [[PS10:%.*]] = getelementptr i32, i32* [[PS]], i64 10
+; CHECK-NEXT: store i32 -2147483648, i32* [[PS10]], align 4
+; CHECK-NEXT: [[PS11:%.*]] = getelementptr i32, i32* [[PS]], i64 11
+; CHECK-NEXT: store i32 -1, i32* [[PS11]], align 4
+; CHECK-NEXT: ret void
+;
+; Fold a valid sequence with leading POSIX whitespace and a minus to
+; (uint32_t)-123.
+ %pwsm123 = getelementptr [11 x i8], [11 x i8]* @ws_im123, i32 0, i32 0
+ %im123 = call i32 @strtoul(i8* %pwsm123, i8** null, i32 10)
+ %ps0 = getelementptr i32, i32* %ps, i32 0
+ store i32 %im123, i32* %ps0
+
+; Fold a valid sequence with leading POSIX whitespace and a plus to +234.
+ %pwsp234 = getelementptr [11 x i8], [11 x i8]* @ws_ip234, i32 0, i32 0
+ %ip234 = call i32 @strtoul(i8* %pwsp234, i8** null, i32 10)
+ %ps1 = getelementptr i32, i32* %ps, i32 1
+ store i32 %ip234, i32* %ps1
+
+; Fold the result of conversion that's equal to INT32_MIN - 1.
+ %psi32minm1 = getelementptr [13 x i8], [13 x i8]* @i32min_m1, i32 0, i32 0
+ %i32min32m1 = call i32 @strtoul(i8* %psi32minm1, i8** null, i32 10)
+ %ps2 = getelementptr i32, i32* %ps, i32 2
+ store i32 %i32min32m1, i32* %ps2
+
+; Fold INT32_MIN.
+ %psi32min = getelementptr [13 x i8], [13 x i8]* @i32min, i32 0, i32 0
+ %i32min = call i32 @strtoul(i8* %psi32min, i8** null, i32 10)
+ %ps3 = getelementptr i32, i32* %ps, i32 3
+ store i32 %i32min, i32* %ps3
+
+; Fold INT32_MIN in octal.
+ %pso32min = getelementptr [15 x i8], [15 x i8]* @o32min, i32 0, i32 0
+ %o32min = call i32 @strtoul(i8* %pso32min, i8** null, i32 0)
+ %ps4 = getelementptr i32, i32* %ps, i32 4
+ store i32 %o32min, i32* %ps4
+
+; Fold -INT32_MIN in octal.
+ %psmo32min = getelementptr [15 x i8], [15 x i8]* @mo32min, i32 0, i32 0
+ %mo32min = call i32 @strtoul(i8* %psmo32min, i8** null, i32 0)
+ %ps5 = getelementptr i32, i32* %ps, i32 5
+ store i32 %mo32min, i32* %ps5
+
+; Fold INT32_MIN in hex.
+ %psx32min = getelementptr [13 x i8], [13 x i8]* @x32min, i32 0, i32 0
+ %x32min = call i32 @strtoul(i8* %psx32min, i8** null, i32 0)
+ %ps6 = getelementptr i32, i32* %ps, i32 6
+ store i32 %x32min, i32* %ps6
+
+; Fold -INT32_MIN in hex.
+ %psmx32min = getelementptr [13 x i8], [13 x i8]* @mx32min, i32 0, i32 0
+ %mx32min = call i32 @strtoul(i8* %psmx32min, i8** null, i32 0)
+ %ps7 = getelementptr i32, i32* %ps, i32 7
+ store i32 %x32min, i32* %ps7
+
+; Fold INT32_MAX.
+ %psi32max = getelementptr [12 x i8], [12 x i8]* @i32max, i32 0, i32 0
+ %i32max = call i32 @strtoul(i8* %psi32max, i8** null, i32 10)
+ %ps8 = getelementptr i32, i32* %ps, i32 8
+ store i32 %i32max, i32* %ps8
+
+; Fold -0x01.
+ %psmX01 = getelementptr [6 x i8], [6 x i8]* @mX01, i32 0, i32 0
+ %mX01 = call i32 @strtoul(i8* %psmX01, i8** null, i32 0)
+ %ps9 = getelementptr i32, i32* %ps, i32 9
+ store i32 %mX01, i32* %ps9
+
+; Fold the result of conversion that's equal to INT32_MAX + 1.
+ %psmax32p1 = getelementptr [12 x i8], [12 x i8]* @i32max_p1, i32 0, i32 0
+ %i32max32p1 = call i32 @strtoul(i8* %psmax32p1, i8** null, i32 10)
+ %ps10 = getelementptr i32, i32* %ps, i32 10
+ store i32 %i32max32p1, i32* %ps10
+
+; Fold UINT32_MAX.
+ %psmax = getelementptr [12 x i8], [12 x i8]* @ui32max, i32 0, i32 0
+ %ui32max = call i32 @strtoul(i8* %psmax, i8** null, i32 10)
+ %ps11 = getelementptr i32, i32* %ps, i32 11
+ store i32 %ui32max, i32* %ps11
+
+ ret void
+}
+
+
+; Exercise not folding calls to 32-bit strtoul.
+
+define void @call_strtoul(i32* %ps) {
+; CHECK-LABEL: @call_strtoul(
+; CHECK-NEXT: [[MINM1:%.*]] = call i32 @strtoul(i8* nocapture getelementptr inbounds ([22 x i8], [22 x i8]* @i64min_m1, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: store i32 [[MINM1]], i32* [[PS:%.*]], align 4
+; CHECK-NEXT: [[MAXP1:%.*]] = call i32 @strtoul(i8* nocapture getelementptr inbounds ([12 x i8], [12 x i8]* @ui32max_p1, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: [[PS1:%.*]] = getelementptr i32, i32* [[PS]], i64 1
+; CHECK-NEXT: store i32 [[MAXP1]], i32* [[PS1]], align 4
+; CHECK-NEXT: [[NWS:%.*]] = call i32 @strtoul(i8* nocapture getelementptr inbounds ([7 x i8], [7 x i8]* @ws, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: [[PS2:%.*]] = getelementptr i32, i32* [[PS]], i64 2
+; CHECK-NEXT: store i32 [[NWS]], i32* [[PS2]], align 4
+; CHECK-NEXT: [[NWSP6:%.*]] = call i32 @strtoul(i8* nocapture getelementptr inbounds ([7 x i8], [7 x i8]* @ws, i64 0, i64 6), i8** null, i32 10)
+; CHECK-NEXT: [[PS3:%.*]] = getelementptr i32, i32* [[PS]], i64 3
+; CHECK-NEXT: store i32 [[NWSP6]], i32* [[PS3]], align 4
+; CHECK-NEXT: ret void
+;
+; Do not fold the result of conversion that overflows uint32_t. This
+; could be folded into a constant provided errnor were set to ERANGE.
+ %psminm1 = getelementptr [22 x i8], [22 x i8]* @i64min_m1, i32 0, i32 0
+ %minm1 = call i32 @strtoul(i8* %psminm1, i8** null, i32 10)
+ %ps0 = getelementptr i32, i32* %ps, i32 0
+ store i32 %minm1, i32* %ps0
+
+; Do not fold the result of conversion that's greater than UINT32_MAX
+; (same logic as above applies here).
+ %psui32maxp1 = getelementptr [12 x i8], [12 x i8]* @ui32max_p1, i32 0, i32 0
+ %maxp1 = call i32 @strtoul(i8* %psui32maxp1, i8** null, i32 10)
+ %ps1 = getelementptr i32, i32* %ps, i32 1
+ store i32 %maxp1, i32* %ps1
+
+; Do not fold a sequence consisting of just whitespace characters.
+ %psws = getelementptr [7 x i8], [7 x i8]* @ws, i32 0, i32 0
+ %nws = call i32 @strtoul(i8* %psws, i8** null, i32 10)
+ %ps2 = getelementptr i32, i32* %ps, i32 2
+ store i32 %nws, i32* %ps2
+
+; Do not fold an empty sequence. The library call may or may not end up
+; storing EINVAL in errno.
+ %pswsp6 = getelementptr [7 x i8], [7 x i8]* @ws, i32 0, i32 6
+ %nwsp6 = call i32 @strtoul(i8* %pswsp6, i8** null, i32 10)
+ %ps3 = getelementptr i32, i32* %ps, i32 3
+ store i32 %nwsp6, i32* %ps3
+
+ ret void
+}
+
+
+; Exercise folding calls to 64-bit strtoull.
+
+define void @fold_strtoull(i64* %ps) {
+; CHECK-LABEL: @fold_strtoull(
+; CHECK-NEXT: store i64 -123, i64* [[PS:%.*]], align 4
+; CHECK-NEXT: [[PS1:%.*]] = getelementptr i64, i64* [[PS]], i64 1
+; CHECK-NEXT: store i64 234, i64* [[PS1]], align 4
+; CHECK-NEXT: [[PS2:%.*]] = getelementptr i64, i64* [[PS]], i64 2
+; CHECK-NEXT: store i64 9223372036854775807, i64* [[PS2]], align 4
+; CHECK-NEXT: [[PS3:%.*]] = getelementptr i64, i64* [[PS]], i64 3
+; CHECK-NEXT: store i64 -2147483648, i64* [[PS3]], align 4
+; CHECK-NEXT: [[PS4:%.*]] = getelementptr i64, i64* [[PS]], i64 4
+; CHECK-NEXT: store i64 2147483648, i64* [[PS4]], align 4
+; CHECK-NEXT: [[PS5:%.*]] = getelementptr i64, i64* [[PS]], i64 5
+; CHECK-NEXT: store i64 2147483648, i64* [[PS5]], align 4
+; CHECK-NEXT: [[PS6:%.*]] = getelementptr i64, i64* [[PS]], i64 6
+; CHECK-NEXT: store i64 -9223372036854775808, i64* [[PS6]], align 4
+; CHECK-NEXT: [[PS7:%.*]] = getelementptr i64, i64* [[PS]], i64 7
+; CHECK-NEXT: store i64 9223372036854775807, i64* [[PS7]], align 4
+; CHECK-NEXT: [[PS8:%.*]] = getelementptr i64, i64* [[PS]], i64 8
+; CHECK-NEXT: store i64 -9223372036854775808, i64* [[PS8]], align 4
+; CHECK-NEXT: [[PS9:%.*]] = getelementptr i64, i64* [[PS]], i64 9
+; CHECK-NEXT: store i64 -1, i64* [[PS9]], align 4
+; CHECK-NEXT: [[PS10:%.*]] = getelementptr i64, i64* [[PS]], i64 10
+; CHECK-NEXT: store i64 -1, i64* [[PS10]], align 4
+; CHECK-NEXT: ret void
+;
+; Fold a valid sequence with leading POSIX whitespace and a minus to
+; (uint64_t)-123.
+ %pwsm123 = getelementptr [11 x i8], [11 x i8]* @ws_im123, i32 0, i32 0
+ %im123 = call i64 @strtoull(i8* %pwsm123, i8** null, i32 10)
+ %ps0 = getelementptr i64, i64* %ps, i32 0
+ store i64 %im123, i64* %ps0
+
+; Fold a valid sequence with leading POSIX whitespace and a plus to +234.
+ %pwsp234 = getelementptr [11 x i8], [11 x i8]* @ws_ip234, i32 0, i32 0
+ %ip234 = call i64 @strtoull(i8* %pwsp234, i8** null, i32 10)
+ %ps1 = getelementptr i64, i64* %ps, i32 1
+ store i64 %ip234, i64* %ps1
+
+; Fold the result of conversion that's equal to INT64_MIN - 1.
+ %psi64minm1 = getelementptr [22 x i8], [22 x i8]* @i64min_m1, i32 0, i32 0
+ %i64min32m1 = call i64 @strtoull(i8* %psi64minm1, i8** null, i32 10)
+ %ps2 = getelementptr i64, i64* %ps, i32 2
+ store i64 %i64min32m1, i64* %ps2
+
+; Fold INT32_MIN.
+ %psi32min = getelementptr [13 x i8], [13 x i8]* @i32min, i32 0, i32 0
+ %i32min = call i64 @strtoull(i8* %psi32min, i8** null, i32 10)
+ %ps3 = getelementptr i64, i64* %ps, i32 3
+ store i64 %i32min, i64* %ps3
+
+; Fold INT32_MIN in octal.
+ %pso32min = getelementptr [15 x i8], [15 x i8]* @o32min, i32 0, i32 0
+ %o32min = call i64 @strtoull(i8* %pso32min, i8** null, i32 0)
+ %ps4 = getelementptr i64, i64* %ps, i32 4
+ store i64 %o32min, i64* %ps4
+
+; Fold INT32_MIN in hex.
+ %psx32min = getelementptr [13 x i8], [13 x i8]* @x32min, i32 0, i32 0
+ %x32min = call i64 @strtoull(i8* %psx32min, i8** null, i32 0)
+ %ps5 = getelementptr i64, i64* %ps, i32 5
+ store i64 %x32min, i64* %ps5
+
+; Fold INT64_MIN.
+ %psi64min = getelementptr [22 x i8], [22 x i8]* @i64min, i32 0, i32 0
+ %i64min = call i64 @strtoull(i8* %psi64min, i8** null, i32 10)
+ %ps6 = getelementptr i64, i64* %ps, i32 6
+ store i64 %i64min, i64* %ps6
+
+; Fold INT64_MAX.
+ %psi64max = getelementptr [21 x i8], [21 x i8]* @i64max, i32 0, i32 0
+ %i64max = call i64 @strtoull(i8* %psi64max, i8** null, i32 10)
+ %ps7 = getelementptr i64, i64* %ps, i32 7
+ store i64 %i64max, i64* %ps7
+
+; Fold the result of conversion that's equal to INT64_MAX + 1 to INT64_MIN.
+ %psmax32p1 = getelementptr [21 x i8], [21 x i8]* @i64max_p1, i32 0, i32 0
+ %i64max32p1 = call i64 @strtoull(i8* %psmax32p1, i8** null, i32 10)
+ %ps8 = getelementptr i64, i64* %ps, i32 8
+ store i64 %i64max32p1, i64* %ps8
+
+; Fold UINT64_MAX.
+ %psmax = getelementptr [22 x i8], [22 x i8]* @ui64max, i32 0, i32 0
+ %ui64max = call i64 @strtoull(i8* %psmax, i8** null, i32 10)
+ %ps9 = getelementptr i64, i64* %ps, i32 9
+ store i64 %ui64max, i64* %ps9
+
+; Fold UINT64_MAX in hex.
+ %psxmax = getelementptr [20 x i8], [20 x i8]* @x64max, i32 0, i32 0
+ %x64max = call i64 @strtoull(i8* %psxmax, i8** null, i32 0)
+ %ps10 = getelementptr i64, i64* %ps, i32 10
+ store i64 %x64max, i64* %ps10
+
+ ret void
+}
+
+
+; Exercise not folding calls to 64-bit strtoull.
+
+define void @call_strtoull(i64* %ps) {
+; CHECK-LABEL: @call_strtoull(
+; CHECK-NEXT: [[MAXP1:%.*]] = call i64 @strtoull(i8* nocapture getelementptr inbounds ([22 x i8], [22 x i8]* @ui64max_p1, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: [[PS1:%.*]] = getelementptr i64, i64* [[PS:%.*]], i64 1
+; CHECK-NEXT: store i64 [[MAXP1]], i64* [[PS1]], align 4
+; CHECK-NEXT: [[NWS:%.*]] = call i64 @strtoull(i8* nocapture getelementptr inbounds ([7 x i8], [7 x i8]* @ws, i64 0, i64 0), i8** null, i32 10)
+; CHECK-NEXT: [[PS2:%.*]] = getelementptr i64, i64* [[PS]], i64 2
+; CHECK-NEXT: store i64 [[NWS]], i64* [[PS2]], align 4
+; CHECK-NEXT: [[NWSP6:%.*]] = call i64 @strtoull(i8* nocapture getelementptr inbounds ([7 x i8], [7 x i8]* @ws, i64 0, i64 6), i8** null, i32 10)
+; CHECK-NEXT: [[PS3:%.*]] = getelementptr i64, i64* [[PS]], i64 3
+; CHECK-NEXT: store i64 [[NWSP6]], i64* [[PS3]], align 4
+; CHECK-NEXT: ret void
+;
+; Do not fold the result of conversion that overflows uint64_t. This
+; could be folded into a constant provided errnor were set to ERANGE.
+ %psui64maxp1 = getelementptr [22 x i8], [22 x i8]* @ui64max_p1, i32 0, i32 0
+ %maxp1 = call i64 @strtoull(i8* %psui64maxp1, i8** null, i32 10)
+ %ps1 = getelementptr i64, i64* %ps, i32 1
+ store i64 %maxp1, i64* %ps1
+
+; Do not fold a sequence consisting of just whitespace characters.
+ %psws = getelementptr [7 x i8], [7 x i8]* @ws, i32 0, i32 0
+ %nws = call i64 @strtoull(i8* %psws, i8** null, i32 10)
+ %ps2 = getelementptr i64, i64* %ps, i32 2
+ store i64 %nws, i64* %ps2
+
+; Do not fold an empty sequence. The library call may or may not end up
+; storing EINVAL in errno.
+ %pswsp6 = getelementptr [7 x i8], [7 x i8]* @ws, i32 0, i32 6
+ %nwsp6 = call i64 @strtoull(i8* %pswsp6, i8** null, i32 10)
+ %ps3 = getelementptr i64, i64* %ps, i32 3
+ store i64 %nwsp6, i64* %ps3
+
+ ret void
+}
diff --git a/llvm/test/Transforms/InstCombine/str-int.ll b/llvm/test/Transforms/InstCombine/str-int.ll
index f74d774b42cbf..fb30ccc73e5fc 100644
--- a/llvm/test/Transforms/InstCombine/str-int.ll
+++ b/llvm/test/Transforms/InstCombine/str-int.ll
@@ -84,7 +84,7 @@ define i32 @strtol_not_const_str(i8* %s) #0 {
define i32 @atoi_not_const_str(i8* %s) #0 {
; CHECK-LABEL: @atoi_not_const_str(
-; CHECK-NEXT: [[CALL:%.*]] = call i32 @atoi(i8* [[S:%.*]])
+; CHECK-NEXT: [[CALL:%.*]] = call i32 @atoi(i8* nocapture [[S:%.*]])
; CHECK-NEXT: ret i32 [[CALL]]
;
%call = call i32 @atoi(i8* %s) #4
@@ -130,7 +130,7 @@ define i32 @atol_test() #0 {
define i32 @atoll_test() #0 {
; CHECK-LABEL: @atoll_test(
-; CHECK-NEXT: [[CALL:%.*]] = call i32 @atoll(i8* getelementptr inbounds ([11 x i8], [11 x i8]* @.str.5, i64 0, i64 0))
+; CHECK-NEXT: [[CALL:%.*]] = call i32 @atoll(i8* nocapture getelementptr inbounds ([11 x i8], [11 x i8]* @.str.5, i64 0, i64 0))
; CHECK-NEXT: ret i32 [[CALL]]
;
%call = call i32 @atoll(i8* getelementptr inbounds ([11 x i8], [11 x i8]* @.str.5, i32 0, i32 0)) #3
diff --git a/llvm/test/Transforms/InstCombine/strcall-no-nul.ll b/llvm/test/Transforms/InstCombine/strcall-no-nul.ll
index e46ea737bb4da..991f88cabfe31 100644
--- a/llvm/test/Transforms/InstCombine/strcall-no-nul.ll
+++ b/llvm/test/Transforms/InstCombine/strcall-no-nul.ll
@@ -31,6 +31,14 @@ declare i8* @strpbrk(i8*, i8*)
declare i64 @strspn(i8*, i8*)
declare i64 @strcspn(i8*, i8*)
+declare i32 @atoi(i8*)
+declare i64 @atol(i8*)
+declare i64 @atoll(i8*)
+declare i64 @strtol(i8*, i8**, i32)
+declare i64 @strtoll(i8*, i8**, i32)
+declare i64 @strtoul(i8*, i8**, i32)
+declare i64 @strtoull(i8*, i8**, i32)
+
declare i32 @sprintf(i8*, i8*, ...)
declare i32 @snprintf(i8*, i64, i8*, ...)
@@ -267,6 +275,74 @@ define void @fold_strcspn_past_end(i64* %poff) {
}
+; TODO: Fold the 32-bit atoi(a5 + 5) to zero.
+; Verify that processing the invalid call doesn't run into trouble.
+
+define i32 @fold_atoi_past_end() {
+; CHECK-LABEL: @fold_atoi_past_end(
+; CHECK-NEXT: [[I:%.*]] = call i32 @atoi(i8* nocapture getelementptr inbounds ([5 x i8], [5 x i8]* @a5, i64 1, i64 0))
+; CHECK-NEXT: ret i32 [[I]]
+;
+ %p5 = getelementptr [5 x i8], [5 x i8]* @a5, i32 0, i32 5
+ %i = call i32 @atoi(i8* %p5)
+ ret i32 %i
+}
+
+; TODO: Likewise, fold the 64-bit atol(a5 + 5) to zero, and similarly
+; for atoll and strtrol and similar.
+; Verify that processing the invalid call doesn't run into trouble.
+
+define void @fold_atol_strtol_past_end(i64* %ps) {
+; CHECK-LABEL: @fold_atol_strtol_past_end(
+; CHECK-NEXT: [[I0:%.*]] = call i64 @atol(i8* nocapture getelementptr inbounds ([5 x i8], [5 x i8]* @a5, i64 1, i64 0))
+; CHECK-NEXT: store i64 [[I0]], i64* [[PS:%.*]], align 4
+; CHECK-NEXT: [[I1:%.*]] = call i64 @atoll(i8* nocapture getelementptr inbounds ([5 x i8], [5 x i8]* @a5, i64 1, i64 0))
+; CHECK-NEXT: [[P1:%.*]] = getelementptr i64, i64* [[PS]], i64 1
+; CHECK-NEXT: store i64 [[I1]], i64* [[P1]], align 4
+; CHECK-NEXT: [[I2:%.*]] = call i64 @strtol(i8* nocapture getelementptr inbounds ([5 x i8], [5 x i8]* @a5, i64 1, i64 0), i8** null, i32 0)
+; CHECK-NEXT: [[P2:%.*]] = getelementptr i64, i64* [[PS]], i64 2
+; CHECK-NEXT: store i64 [[I2]], i64* [[P2]], align 4
+; CHECK-NEXT: [[I3:%.*]] = call i64 @strtoul(i8* nocapture getelementptr inbounds ([5 x i8], [5 x i8]* @a5, i64 1, i64 0), i8** null, i32 8)
+; CHECK-NEXT: [[P3:%.*]] = getelementptr i64, i64* [[PS]], i64 3
+; CHECK-NEXT: store i64 [[I3]], i64* [[P3]], align 4
+; CHECK-NEXT: [[I4:%.*]] = call i64 @strtoll(i8* nocapture getelementptr inbounds ([5 x i8], [5 x i8]* @a5, i64 1, i64 0), i8** null, i32 10)
+; CHECK-NEXT: [[P4:%.*]] = getelementptr i64, i64* [[PS]], i64 4
+; CHECK-NEXT: store i64 [[I4]], i64* [[P4]], align 4
+; CHECK-NEXT: [[I5:%.*]] = call i64 @strtoul(i8* nocapture getelementptr inbounds ([5 x i8], [5 x i8]* @a5, i64 1, i64 0), i8** null, i32 16)
+; CHECK-NEXT: [[P5:%.*]] = getelementptr i64, i64* [[PS]], i64 5
+; CHECK-NEXT: store i64 [[I5]], i64* [[P5]], align 4
+; CHECK-NEXT: ret void
+;
+ %pa5 = getelementptr [5 x i8], [5 x i8]* @a5, i32 0, i32 5
+
+ %i0 = call i64 @atol(i8* %pa5)
+ %p0 = getelementptr i64, i64* %ps, i32 0
+ store i64 %i0, i64* %p0
+
+ %i1 = call i64 @atoll(i8* %pa5)
+ %p1 = getelementptr i64, i64* %ps, i32 1
+ store i64 %i1, i64* %p1
+
+ %i2 = call i64 @strtol(i8* %pa5, i8** null, i32 0)
+ %p2 = getelementptr i64, i64* %ps, i32 2
+ store i64 %i2, i64* %p2
+
+ %i3 = call i64 @strtoul(i8* %pa5, i8** null, i32 8)
+ %p3 = getelementptr i64, i64* %ps, i32 3
+ store i64 %i3, i64* %p3
+
+ %i4 = call i64 @strtoll(i8* %pa5, i8** null, i32 10)
+ %p4 = getelementptr i64, i64* %ps, i32 4
+ store i64 %i4, i64* %p4
+
+ %i5 = call i64 @strtoul(i8* %pa5, i8** null, i32 16)
+ %p5 = getelementptr i64, i64* %ps, i32 5
+ store i64 %i5, i64* %p5
+
+ ret void
+}
+
+
; Fold sprintf(dst, a5 + 5) to zero, and also
; TODO: fold sprintf(dst, "%s", a5 + 5) to zero.
More information about the llvm-commits
mailing list