[ConstantFold] Fold ilogb and ilogbf when the input parameter is a constant value.
#113014
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
c8ef
changed the title
ilogb and ilogbf when the input parameter is a constant value.
|
@llvm/pr-subscribers-llvm-transforms @llvm/pr-subscribers-llvm-analysis Author: None (c8ef) ChangesThis patch adds support for constant folding for the Full diff: https://github.com/llvm/llvm-project/pull/113014.diff 2 Files Affected:
diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index 74df67a4ff9b43..d9eea2cc2915bb 100644
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -1677,6 +1677,8 @@ bool llvm::canConstantFoldCallTo(const CallBase *Call, const Function *F) {
return Name == "fabs" || Name == "fabsf" ||
Name == "floor" || Name == "floorf" ||
Name == "fmod" || Name == "fmodf";
+ case 'i':
+ return Name == "ilogb" || Name == "ilogbf";
case 'l':
return Name == "log" || Name == "logf" || Name == "logl" ||
Name == "log2" || Name == "log2f" || Name == "log10" ||
@@ -1772,6 +1774,17 @@ inline bool llvm_fenv_testexcept() {
return false;
}
+Constant *ConstantFoldInt(int (*NativeFP)(double), const APFloat &V, Type *Ty) {
+ llvm_fenv_clearexcept();
+ int Result = NativeFP(V.convertToDouble());
+ if (llvm_fenv_testexcept()) {
+ llvm_fenv_clearexcept();
+ return nullptr;
+ }
+
+ return ConstantInt::get(Ty, Result, true);
+}
+
Constant *ConstantFoldFP(double (*NativeFP)(double), const APFloat &V,
Type *Ty) {
llvm_fenv_clearexcept();
@@ -2131,7 +2144,8 @@ static Constant *ConstantFoldScalarCall1(StringRef Name,
}
#endif
- if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy())
+ if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy() &&
+ !Ty->isIntegerTy())
return nullptr;
// Use internal versions of these intrinsics.
@@ -2391,6 +2405,11 @@ static Constant *ConstantFoldScalarCall1(StringRef Name,
// TODO: What about hosts that lack a C99 library?
return ConstantFoldFP(log10, APF, Ty);
break;
+ case LibFunc_ilogb:
+ case LibFunc_ilogbf:
+ if (!APF.isZero() && TLI->has(Func))
+ return ConstantFoldInt(ilogb, APF, Ty);
+ break;
case LibFunc_logb:
case LibFunc_logbf:
if (!APF.isZero() && TLI->has(Func))
diff --git a/llvm/test/Transforms/InstCombine/ilogb.ll b/llvm/test/Transforms/InstCombine/ilogb.ll
new file mode 100644
index 00000000000000..e30791fe68e7b2
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/ilogb.ll
@@ -0,0 +1,203 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -passes=instcombine -S | FileCheck %s
+
+define i32 @ilogbf_const1() {
+; CHECK-LABEL: define i32 @ilogbf_const1() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogbf(float 7.000000e+00)
+; CHECK-NEXT: ret i32 2
+;
+ %r = call i32 @ilogbf(float 7.000000e+00)
+ ret i32 %r
+}
+
+define i32 @ilogb_const1() {
+; CHECK-LABEL: define i32 @ilogb_const1() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogb(double -7.000000e+00)
+; CHECK-NEXT: ret i32 2
+;
+ %r = call i32 @ilogb(double -7.000000e+00)
+ ret i32 %r
+}
+
+define i32 @ilogbf_const2() {
+; CHECK-LABEL: define i32 @ilogbf_const2() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogbf(float 5.000000e-01)
+; CHECK-NEXT: ret i32 -1
+;
+ %r = call i32 @ilogbf(float 5.000000e-01)
+ ret i32 %r
+}
+
+define i32 @ilogb_const2() {
+; CHECK-LABEL: define i32 @ilogb_const2() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogb(double -5.000000e-01)
+; CHECK-NEXT: ret i32 -1
+;
+ %r = call i32 @ilogb(double -5.000000e-01)
+ ret i32 %r
+}
+
+define i32 @ilogbf_zero() {
+; CHECK-LABEL: define i32 @ilogbf_zero() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogbf(float 0.000000e+00)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogbf(float 0.000000e+00)
+ ret i32 %r
+}
+
+define i32 @ilogb_zero() {
+; CHECK-LABEL: define i32 @ilogb_zero() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogb(double 0.000000e+00)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogb(double 0.000000e+00)
+ ret i32 %r
+}
+
+define i32 @ilogbf_neg_zero() {
+; CHECK-LABEL: define i32 @ilogbf_neg_zero() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogbf(float -0.000000e+00)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogbf(float -0.000000e+00)
+ ret i32 %r
+}
+
+define i32 @ilogb_neg_zero() {
+; CHECK-LABEL: define i32 @ilogb_neg_zero() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogb(double -0.000000e+00)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogb(double -0.000000e+00)
+ ret i32 %r
+}
+
+define i32 @ilogbf_inf() {
+; CHECK-LABEL: define i32 @ilogbf_inf() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogbf(float 0x7FF0000000000000)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogbf(float 0x7FF0000000000000)
+ ret i32 %r
+}
+
+define i32 @ilogb_inf() {
+; CHECK-LABEL: define i32 @ilogb_inf() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogb(double 0x7FF0000000000000)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogb(double 0x7FF0000000000000)
+ ret i32 %r
+}
+
+define i32 @ilogbf_nan() {
+; CHECK-LABEL: define i32 @ilogbf_nan() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogbf(float 0x7FF8000000000000)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogbf(float 0x7FF8000000000000)
+ ret i32 %r
+}
+
+define i32 @ilogb_nan() {
+; CHECK-LABEL: define i32 @ilogb_nan() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogb(double 0x7FF8000000000000)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogb(double 0x7FF8000000000000)
+ ret i32 %r
+}
+
+define i32 @ilogbf_zero_readnone() {
+; CHECK-LABEL: define i32 @ilogbf_zero_readnone() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogbf(float 0.000000e+00) #[[ATTR0:[0-9]+]]
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogbf(float 0.000000e+00) readnone
+ ret i32 %r
+}
+
+define i32 @ilogb_zero_readnone() {
+; CHECK-LABEL: define i32 @ilogb_zero_readnone() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogb(double 0.000000e+00) #[[ATTR0]]
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogb(double 0.000000e+00) readnone
+ ret i32 %r
+}
+
+define i32 @ilogbf_neg_zero_readnone() {
+; CHECK-LABEL: define i32 @ilogbf_neg_zero_readnone() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogbf(float -0.000000e+00) #[[ATTR0]]
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogbf(float -0.000000e+00) readnone
+ ret i32 %r
+}
+
+define i32 @ilogb_neg_zero_readnone() {
+; CHECK-LABEL: define i32 @ilogb_neg_zero_readnone() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogb(double -0.000000e+00) #[[ATTR0]]
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogb(double -0.000000e+00) readnone
+ ret i32 %r
+}
+
+define i32 @ilogbf_inf_readnone() {
+; CHECK-LABEL: define i32 @ilogbf_inf_readnone() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogbf(float 0x7FF0000000000000) #[[ATTR0]]
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogbf(float 0x7FF0000000000000) readnone
+ ret i32 %r
+}
+
+define i32 @ilogb_inf_readnone() {
+; CHECK-LABEL: define i32 @ilogb_inf_readnone() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogb(double 0x7FF0000000000000) #[[ATTR0]]
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogb(double 0x7FF0000000000000) readnone
+ ret i32 %r
+}
+
+define i32 @ilogbf_nan_readnone() {
+; CHECK-LABEL: define i32 @ilogbf_nan_readnone() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogbf(float 0x7FF8000000000000) #[[ATTR0]]
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogbf(float 0x7FF8000000000000) readnone
+ ret i32 %r
+}
+
+define i32 @ilogb_nan_readnone() {
+; CHECK-LABEL: define i32 @ilogb_nan_readnone() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogb(double 0x7FF8000000000000) #[[ATTR0]]
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogb(double 0x7FF8000000000000) readnone
+ ret i32 %r
+}
+
+define i32 @ilogbf_poison() {
+; CHECK-LABEL: define i32 @ilogbf_poison() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogbf(float poison)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogbf(float poison)
+ ret i32 %r
+}
+
+define i32 @ilogb_poison() {
+; CHECK-LABEL: define i32 @ilogb_poison() {
+; CHECK-NEXT: [[R:%.*]] = call i32 @ilogb(double poison)
+; CHECK-NEXT: ret i32 [[R]]
+;
+ %r = call i32 @ilogb(double poison)
+ ret i32 %r
+}
+
+declare i32 @ilogbf(float)
+declare i32 @ilogb(double)
|
dtcxzyw
reviewed
Oct 19, 2024
| case LibFunc_ilogb: | ||
| case LibFunc_ilogbf: | ||
| if (!APF.isZero() && TLI->has(Func)) | ||
| return ConstantFoldInt(ilogb, APF, Ty); |
There was a problem hiding this comment.
Use int ilogb(const APFloat &Arg).
arsenm
reviewed
Oct 19, 2024
| @@ -2131,7 +2133,8 @@ static Constant *ConstantFoldScalarCall1(StringRef Name, | |||
| } | |||
| #endif | |||
|
|
|||
| if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy()) | |||
| if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy() && | |||
There was a problem hiding this comment.
Probably should just handle this case before here, or just delete the whole check.
This function could use some cleanup. Why isn't this using a switch over the intrinsic ID?
There was a problem hiding this comment.
Probably should just handle this case before here, or just delete the whole check.
It appears that the check is necessary to guard certain types, such as vector or fp128 types.
arsenm
approved these changes
Oct 19, 2024
There was a problem hiding this comment.
LGTM but please consider cleaning up this function. There should be a switch over intrinsic IDs.
The repeated TLI->has(func) in each case could be done once for all the libcall cases.
It would also be good to avoid explicitly listing the edge cases in the pre-check of the ConstantFoldFP call
This reverts commit 5f1f693.
For libcalls like
This could be partially achieved by the solution proposed in #112113 (comment). |
Yes, there needs to be duplicated handling for intrinsics and libcalls (part of why I think most of the math intrinsics should be removed) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This patch adds support for constant folding for the
ilogbandilogbflibc functions.