release/21.x: [IR] Handle fabs LHS in fcmpImpliesClass (#152913)
#152921
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
|
@arsenm What do you think about merging this PR to the release branch? |
llvmbot
added
llvm:instcombine
|
@llvm/pr-subscribers-llvm-transforms Author: None (llvmbot) ChangesBackport d8b1b46 Requested by: @dtcxzyw Full diff: https://github.com/llvm/llvm-project/pull/152921.diff 3 Files Affected:
diff --git a/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h b/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h
index 8aac9d5b49dbb..448a6e913eb86 100644
--- a/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h
+++ b/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h
@@ -135,6 +135,12 @@ template <typename ContextT> class GenericFloatingPointPredicateUtils {
if (Mode.Input != DenormalMode::IEEE)
return {Invalid, fcAllFlags, fcAllFlags};
+ auto ExactClass = [IsFabs, Src](FPClassTest Mask) {
+ if (IsFabs)
+ Mask = llvm::inverse_fabs(Mask);
+ return exactClass(Src, Mask);
+ };
+
switch (Pred) {
case FCmpInst::FCMP_OEQ: // Match x == 0.0
return exactClass(Src, fcZero);
@@ -151,26 +157,24 @@ template <typename ContextT> class GenericFloatingPointPredicateUtils {
case FCmpInst::FCMP_UNO:
return exactClass(Src, fcNan);
case FCmpInst::FCMP_OGT: // x > 0
- return exactClass(Src, fcPosSubnormal | fcPosNormal | fcPosInf);
+ return ExactClass(fcPosSubnormal | fcPosNormal | fcPosInf);
case FCmpInst::FCMP_UGT: // isnan(x) || x > 0
- return exactClass(Src, fcPosSubnormal | fcPosNormal | fcPosInf | fcNan);
+ return ExactClass(fcPosSubnormal | fcPosNormal | fcPosInf | fcNan);
case FCmpInst::FCMP_OGE: // x >= 0
- return exactClass(Src, fcPositive | fcNegZero);
+ return ExactClass(fcPositive | fcNegZero);
case FCmpInst::FCMP_UGE: // isnan(x) || x >= 0
- return exactClass(Src, fcPositive | fcNegZero | fcNan);
+ return ExactClass(fcPositive | fcNegZero | fcNan);
case FCmpInst::FCMP_OLT: // x < 0
- return exactClass(Src, fcNegSubnormal | fcNegNormal | fcNegInf);
+ return ExactClass(fcNegSubnormal | fcNegNormal | fcNegInf);
case FCmpInst::FCMP_ULT: // isnan(x) || x < 0
- return exactClass(Src, fcNegSubnormal | fcNegNormal | fcNegInf | fcNan);
+ return ExactClass(fcNegSubnormal | fcNegNormal | fcNegInf | fcNan);
case FCmpInst::FCMP_OLE: // x <= 0
- return exactClass(Src, fcNegative | fcPosZero);
+ return ExactClass(fcNegative | fcPosZero);
case FCmpInst::FCMP_ULE: // isnan(x) || x <= 0
- return exactClass(Src, fcNegative | fcPosZero | fcNan);
+ return ExactClass(fcNegative | fcPosZero | fcNan);
default:
llvm_unreachable("all compare types are handled");
}
-
- return {Invalid, fcAllFlags, fcAllFlags};
}
const bool IsDenormalRHS = (OrigClass & fcSubnormal) == OrigClass;
diff --git a/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll b/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll
index ab4448b460bfc..820fff433e9e0 100644
--- a/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll
+++ b/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll
@@ -213,7 +213,7 @@ define double @fmul_nnan_ninf_nneg_n0.0_commute(i127 %x) {
define float @fmul_ninf_nnan_mul_zero_nsz(float nofpclass(inf nan) %f) {
; CHECK-LABEL: @fmul_ninf_nnan_mul_zero_nsz(
-; CHECK-NEXT: ret float 0.000000e+00
+; CHECK-NEXT: ret float 0.000000e+00
;
%r = fmul nsz float %f, 0.0
ret float %r
@@ -221,7 +221,7 @@ define float @fmul_ninf_nnan_mul_zero_nsz(float nofpclass(inf nan) %f) {
define float @fmul_ninf_nnan_mul_nzero_nsz(float nofpclass(inf nan) %f) {
; CHECK-LABEL: @fmul_ninf_nnan_mul_nzero_nsz(
-; CHECK-NEXT: ret float 0.000000e+00
+; CHECK-NEXT: ret float 0.000000e+00
;
%r = fmul nsz float %f, -0.0
ret float %r
@@ -1255,3 +1255,20 @@ define i1 @fptrunc_round_unknown_positive(double %unknown) {
%cmp = fcmp nnan oge float %op, 0.0
ret i1 %cmp
}
+
+define half @fabs_select_fabs(half noundef %x) {
+; CHECK-LABEL: @fabs_select_fabs(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[ABS1:%.*]] = call half @llvm.fabs.f16(half [[X:%.*]])
+; CHECK-NEXT: [[CMP:%.*]] = fcmp ogt half [[ABS1]], 0xH0000
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[CMP]], half [[X]], half 0xH0000
+; CHECK-NEXT: [[ABS2:%.*]] = call half @llvm.fabs.f16(half [[SEL]])
+; CHECK-NEXT: ret half [[ABS2]]
+;
+entry:
+ %abs1 = call half @llvm.fabs.f16(half %x)
+ %cmp = fcmp ogt half %abs1, 0xH0000
+ %sel = select i1 %cmp, half %x, half 0xH0000
+ %abs2 = call half @llvm.fabs.f16(half %sel)
+ ret half %abs2
+}
diff --git a/llvm/unittests/Analysis/ValueTrackingTest.cpp b/llvm/unittests/Analysis/ValueTrackingTest.cpp
index 7a48105a1dc99..bf396499e35ca 100644
--- a/llvm/unittests/Analysis/ValueTrackingTest.cpp
+++ b/llvm/unittests/Analysis/ValueTrackingTest.cpp
@@ -7,6 +7,7 @@
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/ADT/FloatingPointMode.h"
#include "llvm/Analysis/AssumptionCache.h"
#include "llvm/Analysis/FloatingPointPredicateUtils.h"
#include "llvm/AsmParser/Parser.h"
@@ -2208,6 +2209,41 @@ TEST_F(ComputeKnownFPClassTest, Constants) {
}
}
+TEST_F(ComputeKnownFPClassTest, fcmpImpliesClass_fabs_zero) {
+ parseAssembly("define float @test(float %x) {\n"
+ " %A = call float @llvm.fabs.f32(float %x)\n"
+ " ret float %A\n"
+ "}\n");
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OEQ, *F, A, fcZero)),
+ fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UEQ, *F, A, fcZero)),
+ fcZero | fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UNE, *F, A, fcZero)),
+ ~fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ONE, *F, A, fcZero)),
+ ~fcNan & ~fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ORD, *F, A, fcZero)),
+ ~fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UNO, *F, A, fcZero)),
+ fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OGT, *F, A, fcZero)),
+ fcSubnormal | fcNormal | fcInf);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UGT, *F, A, fcZero)),
+ fcSubnormal | fcNormal | fcInf | fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OGE, *F, A, fcZero)),
+ ~fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UGE, *F, A, fcZero)),
+ fcAllFlags);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OLT, *F, A, fcZero)),
+ fcNone);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ULT, *F, A, fcZero)),
+ fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OLE, *F, A, fcZero)),
+ fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ULE, *F, A, fcZero)),
+ fcZero | fcNan);
+}
+
TEST_F(ValueTrackingTest, isNonZeroRecurrence) {
parseAssembly(R"(
define i1 @test(i8 %n, i8 %r) {
|
|
@llvm/pr-subscribers-llvm-analysis Author: None (llvmbot) ChangesBackport d8b1b46 Requested by: @dtcxzyw Full diff: https://github.com/llvm/llvm-project/pull/152921.diff 3 Files Affected:
diff --git a/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h b/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h
index 8aac9d5b49dbb..448a6e913eb86 100644
--- a/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h
+++ b/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h
@@ -135,6 +135,12 @@ template <typename ContextT> class GenericFloatingPointPredicateUtils {
if (Mode.Input != DenormalMode::IEEE)
return {Invalid, fcAllFlags, fcAllFlags};
+ auto ExactClass = [IsFabs, Src](FPClassTest Mask) {
+ if (IsFabs)
+ Mask = llvm::inverse_fabs(Mask);
+ return exactClass(Src, Mask);
+ };
+
switch (Pred) {
case FCmpInst::FCMP_OEQ: // Match x == 0.0
return exactClass(Src, fcZero);
@@ -151,26 +157,24 @@ template <typename ContextT> class GenericFloatingPointPredicateUtils {
case FCmpInst::FCMP_UNO:
return exactClass(Src, fcNan);
case FCmpInst::FCMP_OGT: // x > 0
- return exactClass(Src, fcPosSubnormal | fcPosNormal | fcPosInf);
+ return ExactClass(fcPosSubnormal | fcPosNormal | fcPosInf);
case FCmpInst::FCMP_UGT: // isnan(x) || x > 0
- return exactClass(Src, fcPosSubnormal | fcPosNormal | fcPosInf | fcNan);
+ return ExactClass(fcPosSubnormal | fcPosNormal | fcPosInf | fcNan);
case FCmpInst::FCMP_OGE: // x >= 0
- return exactClass(Src, fcPositive | fcNegZero);
+ return ExactClass(fcPositive | fcNegZero);
case FCmpInst::FCMP_UGE: // isnan(x) || x >= 0
- return exactClass(Src, fcPositive | fcNegZero | fcNan);
+ return ExactClass(fcPositive | fcNegZero | fcNan);
case FCmpInst::FCMP_OLT: // x < 0
- return exactClass(Src, fcNegSubnormal | fcNegNormal | fcNegInf);
+ return ExactClass(fcNegSubnormal | fcNegNormal | fcNegInf);
case FCmpInst::FCMP_ULT: // isnan(x) || x < 0
- return exactClass(Src, fcNegSubnormal | fcNegNormal | fcNegInf | fcNan);
+ return ExactClass(fcNegSubnormal | fcNegNormal | fcNegInf | fcNan);
case FCmpInst::FCMP_OLE: // x <= 0
- return exactClass(Src, fcNegative | fcPosZero);
+ return ExactClass(fcNegative | fcPosZero);
case FCmpInst::FCMP_ULE: // isnan(x) || x <= 0
- return exactClass(Src, fcNegative | fcPosZero | fcNan);
+ return ExactClass(fcNegative | fcPosZero | fcNan);
default:
llvm_unreachable("all compare types are handled");
}
-
- return {Invalid, fcAllFlags, fcAllFlags};
}
const bool IsDenormalRHS = (OrigClass & fcSubnormal) == OrigClass;
diff --git a/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll b/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll
index ab4448b460bfc..820fff433e9e0 100644
--- a/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll
+++ b/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll
@@ -213,7 +213,7 @@ define double @fmul_nnan_ninf_nneg_n0.0_commute(i127 %x) {
define float @fmul_ninf_nnan_mul_zero_nsz(float nofpclass(inf nan) %f) {
; CHECK-LABEL: @fmul_ninf_nnan_mul_zero_nsz(
-; CHECK-NEXT: ret float 0.000000e+00
+; CHECK-NEXT: ret float 0.000000e+00
;
%r = fmul nsz float %f, 0.0
ret float %r
@@ -221,7 +221,7 @@ define float @fmul_ninf_nnan_mul_zero_nsz(float nofpclass(inf nan) %f) {
define float @fmul_ninf_nnan_mul_nzero_nsz(float nofpclass(inf nan) %f) {
; CHECK-LABEL: @fmul_ninf_nnan_mul_nzero_nsz(
-; CHECK-NEXT: ret float 0.000000e+00
+; CHECK-NEXT: ret float 0.000000e+00
;
%r = fmul nsz float %f, -0.0
ret float %r
@@ -1255,3 +1255,20 @@ define i1 @fptrunc_round_unknown_positive(double %unknown) {
%cmp = fcmp nnan oge float %op, 0.0
ret i1 %cmp
}
+
+define half @fabs_select_fabs(half noundef %x) {
+; CHECK-LABEL: @fabs_select_fabs(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[ABS1:%.*]] = call half @llvm.fabs.f16(half [[X:%.*]])
+; CHECK-NEXT: [[CMP:%.*]] = fcmp ogt half [[ABS1]], 0xH0000
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[CMP]], half [[X]], half 0xH0000
+; CHECK-NEXT: [[ABS2:%.*]] = call half @llvm.fabs.f16(half [[SEL]])
+; CHECK-NEXT: ret half [[ABS2]]
+;
+entry:
+ %abs1 = call half @llvm.fabs.f16(half %x)
+ %cmp = fcmp ogt half %abs1, 0xH0000
+ %sel = select i1 %cmp, half %x, half 0xH0000
+ %abs2 = call half @llvm.fabs.f16(half %sel)
+ ret half %abs2
+}
diff --git a/llvm/unittests/Analysis/ValueTrackingTest.cpp b/llvm/unittests/Analysis/ValueTrackingTest.cpp
index 7a48105a1dc99..bf396499e35ca 100644
--- a/llvm/unittests/Analysis/ValueTrackingTest.cpp
+++ b/llvm/unittests/Analysis/ValueTrackingTest.cpp
@@ -7,6 +7,7 @@
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/ADT/FloatingPointMode.h"
#include "llvm/Analysis/AssumptionCache.h"
#include "llvm/Analysis/FloatingPointPredicateUtils.h"
#include "llvm/AsmParser/Parser.h"
@@ -2208,6 +2209,41 @@ TEST_F(ComputeKnownFPClassTest, Constants) {
}
}
+TEST_F(ComputeKnownFPClassTest, fcmpImpliesClass_fabs_zero) {
+ parseAssembly("define float @test(float %x) {\n"
+ " %A = call float @llvm.fabs.f32(float %x)\n"
+ " ret float %A\n"
+ "}\n");
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OEQ, *F, A, fcZero)),
+ fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UEQ, *F, A, fcZero)),
+ fcZero | fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UNE, *F, A, fcZero)),
+ ~fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ONE, *F, A, fcZero)),
+ ~fcNan & ~fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ORD, *F, A, fcZero)),
+ ~fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UNO, *F, A, fcZero)),
+ fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OGT, *F, A, fcZero)),
+ fcSubnormal | fcNormal | fcInf);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UGT, *F, A, fcZero)),
+ fcSubnormal | fcNormal | fcInf | fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OGE, *F, A, fcZero)),
+ ~fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UGE, *F, A, fcZero)),
+ fcAllFlags);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OLT, *F, A, fcZero)),
+ fcNone);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ULT, *F, A, fcZero)),
+ fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OLE, *F, A, fcZero)),
+ fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ULE, *F, A, fcZero)),
+ fcZero | fcNan);
+}
+
TEST_F(ValueTrackingTest, isNonZeroRecurrence) {
parseAssembly(R"(
define i1 @test(i8 %n, i8 %r) {
|
|
@llvm/pr-subscribers-llvm-ir Author: None (llvmbot) ChangesBackport d8b1b46 Requested by: @dtcxzyw Full diff: https://github.com/llvm/llvm-project/pull/152921.diff 3 Files Affected:
diff --git a/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h b/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h
index 8aac9d5b49dbb..448a6e913eb86 100644
--- a/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h
+++ b/llvm/include/llvm/IR/GenericFloatingPointPredicateUtils.h
@@ -135,6 +135,12 @@ template <typename ContextT> class GenericFloatingPointPredicateUtils {
if (Mode.Input != DenormalMode::IEEE)
return {Invalid, fcAllFlags, fcAllFlags};
+ auto ExactClass = [IsFabs, Src](FPClassTest Mask) {
+ if (IsFabs)
+ Mask = llvm::inverse_fabs(Mask);
+ return exactClass(Src, Mask);
+ };
+
switch (Pred) {
case FCmpInst::FCMP_OEQ: // Match x == 0.0
return exactClass(Src, fcZero);
@@ -151,26 +157,24 @@ template <typename ContextT> class GenericFloatingPointPredicateUtils {
case FCmpInst::FCMP_UNO:
return exactClass(Src, fcNan);
case FCmpInst::FCMP_OGT: // x > 0
- return exactClass(Src, fcPosSubnormal | fcPosNormal | fcPosInf);
+ return ExactClass(fcPosSubnormal | fcPosNormal | fcPosInf);
case FCmpInst::FCMP_UGT: // isnan(x) || x > 0
- return exactClass(Src, fcPosSubnormal | fcPosNormal | fcPosInf | fcNan);
+ return ExactClass(fcPosSubnormal | fcPosNormal | fcPosInf | fcNan);
case FCmpInst::FCMP_OGE: // x >= 0
- return exactClass(Src, fcPositive | fcNegZero);
+ return ExactClass(fcPositive | fcNegZero);
case FCmpInst::FCMP_UGE: // isnan(x) || x >= 0
- return exactClass(Src, fcPositive | fcNegZero | fcNan);
+ return ExactClass(fcPositive | fcNegZero | fcNan);
case FCmpInst::FCMP_OLT: // x < 0
- return exactClass(Src, fcNegSubnormal | fcNegNormal | fcNegInf);
+ return ExactClass(fcNegSubnormal | fcNegNormal | fcNegInf);
case FCmpInst::FCMP_ULT: // isnan(x) || x < 0
- return exactClass(Src, fcNegSubnormal | fcNegNormal | fcNegInf | fcNan);
+ return ExactClass(fcNegSubnormal | fcNegNormal | fcNegInf | fcNan);
case FCmpInst::FCMP_OLE: // x <= 0
- return exactClass(Src, fcNegative | fcPosZero);
+ return ExactClass(fcNegative | fcPosZero);
case FCmpInst::FCMP_ULE: // isnan(x) || x <= 0
- return exactClass(Src, fcNegative | fcPosZero | fcNan);
+ return ExactClass(fcNegative | fcPosZero | fcNan);
default:
llvm_unreachable("all compare types are handled");
}
-
- return {Invalid, fcAllFlags, fcAllFlags};
}
const bool IsDenormalRHS = (OrigClass & fcSubnormal) == OrigClass;
diff --git a/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll b/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll
index ab4448b460bfc..820fff433e9e0 100644
--- a/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll
+++ b/llvm/test/Transforms/InstSimplify/floating-point-arithmetic.ll
@@ -213,7 +213,7 @@ define double @fmul_nnan_ninf_nneg_n0.0_commute(i127 %x) {
define float @fmul_ninf_nnan_mul_zero_nsz(float nofpclass(inf nan) %f) {
; CHECK-LABEL: @fmul_ninf_nnan_mul_zero_nsz(
-; CHECK-NEXT: ret float 0.000000e+00
+; CHECK-NEXT: ret float 0.000000e+00
;
%r = fmul nsz float %f, 0.0
ret float %r
@@ -221,7 +221,7 @@ define float @fmul_ninf_nnan_mul_zero_nsz(float nofpclass(inf nan) %f) {
define float @fmul_ninf_nnan_mul_nzero_nsz(float nofpclass(inf nan) %f) {
; CHECK-LABEL: @fmul_ninf_nnan_mul_nzero_nsz(
-; CHECK-NEXT: ret float 0.000000e+00
+; CHECK-NEXT: ret float 0.000000e+00
;
%r = fmul nsz float %f, -0.0
ret float %r
@@ -1255,3 +1255,20 @@ define i1 @fptrunc_round_unknown_positive(double %unknown) {
%cmp = fcmp nnan oge float %op, 0.0
ret i1 %cmp
}
+
+define half @fabs_select_fabs(half noundef %x) {
+; CHECK-LABEL: @fabs_select_fabs(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[ABS1:%.*]] = call half @llvm.fabs.f16(half [[X:%.*]])
+; CHECK-NEXT: [[CMP:%.*]] = fcmp ogt half [[ABS1]], 0xH0000
+; CHECK-NEXT: [[SEL:%.*]] = select i1 [[CMP]], half [[X]], half 0xH0000
+; CHECK-NEXT: [[ABS2:%.*]] = call half @llvm.fabs.f16(half [[SEL]])
+; CHECK-NEXT: ret half [[ABS2]]
+;
+entry:
+ %abs1 = call half @llvm.fabs.f16(half %x)
+ %cmp = fcmp ogt half %abs1, 0xH0000
+ %sel = select i1 %cmp, half %x, half 0xH0000
+ %abs2 = call half @llvm.fabs.f16(half %sel)
+ ret half %abs2
+}
diff --git a/llvm/unittests/Analysis/ValueTrackingTest.cpp b/llvm/unittests/Analysis/ValueTrackingTest.cpp
index 7a48105a1dc99..bf396499e35ca 100644
--- a/llvm/unittests/Analysis/ValueTrackingTest.cpp
+++ b/llvm/unittests/Analysis/ValueTrackingTest.cpp
@@ -7,6 +7,7 @@
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/ADT/FloatingPointMode.h"
#include "llvm/Analysis/AssumptionCache.h"
#include "llvm/Analysis/FloatingPointPredicateUtils.h"
#include "llvm/AsmParser/Parser.h"
@@ -2208,6 +2209,41 @@ TEST_F(ComputeKnownFPClassTest, Constants) {
}
}
+TEST_F(ComputeKnownFPClassTest, fcmpImpliesClass_fabs_zero) {
+ parseAssembly("define float @test(float %x) {\n"
+ " %A = call float @llvm.fabs.f32(float %x)\n"
+ " ret float %A\n"
+ "}\n");
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OEQ, *F, A, fcZero)),
+ fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UEQ, *F, A, fcZero)),
+ fcZero | fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UNE, *F, A, fcZero)),
+ ~fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ONE, *F, A, fcZero)),
+ ~fcNan & ~fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ORD, *F, A, fcZero)),
+ ~fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UNO, *F, A, fcZero)),
+ fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OGT, *F, A, fcZero)),
+ fcSubnormal | fcNormal | fcInf);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UGT, *F, A, fcZero)),
+ fcSubnormal | fcNormal | fcInf | fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OGE, *F, A, fcZero)),
+ ~fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_UGE, *F, A, fcZero)),
+ fcAllFlags);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OLT, *F, A, fcZero)),
+ fcNone);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ULT, *F, A, fcZero)),
+ fcNan);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_OLE, *F, A, fcZero)),
+ fcZero);
+ EXPECT_EQ(std::get<1>(fcmpImpliesClass(FCmpInst::FCMP_ULE, *F, A, fcZero)),
+ fcZero | fcNan);
+}
+
TEST_F(ValueTrackingTest, isNonZeroRecurrence) {
parseAssembly(R"(
define i1 @test(i8 %n, i8 %r) {
|
arsenm
approved these changes
Aug 10, 2025
github-project-automation
bot
moved this from Needs Triage
to Needs Merge
in LLVM Release Status
github-project-automation
bot
moved this from Needs Merge
to Done
in LLVM Release Status
Closes llvm#152824. (cherry picked from commit d8b1b46)
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.
Backport d8b1b46
Requested by: @dtcxzyw