From 98992d305f493d40da23cfe8841fbffb1a3ce267 Mon Sep 17 00:00:00 2001 From: Yingwei Zheng Date: Tue, 22 Apr 2025 11:18:43 +0800 Subject: [PATCH 1/3] [InstCombine] Preserve signbit semantics of NaN with fold to fabs --- .../InstCombine/InstCombineSelect.cpp | 6 +++- llvm/test/Transforms/InstCombine/fabs.ll | 35 +++++++++++++------ 2 files changed, 29 insertions(+), 12 deletions(-) diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp index 4bba2f406b4c1..06792e32de3e1 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp @@ -2793,7 +2793,11 @@ static Instruction *foldSelectWithFCmpToFabs(SelectInst &SI, // fold (X <= +/-0.0) ? (0.0 - X) : X to fabs(X), when 'Swap' is false // fold (X > +/-0.0) ? X : (0.0 - X) to fabs(X), when 'Swap' is true - if (match(TrueVal, m_FSub(m_PosZeroFP(), m_Specific(X)))) { + // Note: We require "nnan" for this fold because fcmp ignores the signbit + // of NAN, but IEEE-754 specifies the signbit of NAN values with + // fneg/fabs operations. + if (cast(CondVal)->hasNoNaNs() && + match(TrueVal, m_FSub(m_PosZeroFP(), m_Specific(X)))) { if (!Swap && (Pred == FCmpInst::FCMP_OLE || Pred == FCmpInst::FCMP_ULE)) { Value *Fabs = IC.Builder.CreateUnaryIntrinsic(Intrinsic::fabs, X, &SI); return IC.replaceInstUsesWith(SI, Fabs); diff --git a/llvm/test/Transforms/InstCombine/fabs.ll b/llvm/test/Transforms/InstCombine/fabs.ll index 7b9a672f188ca..827315b184085 100644 --- a/llvm/test/Transforms/InstCombine/fabs.ll +++ b/llvm/test/Transforms/InstCombine/fabs.ll @@ -256,6 +256,19 @@ define double @select_fcmp_ole_zero(double %x) { ; CHECK-LABEL: @select_fcmp_ole_zero( ; CHECK-NEXT: [[FABS:%.*]] = call double @llvm.fabs.f64(double [[X:%.*]]) ; CHECK-NEXT: ret double [[FABS]] +; + %lezero = fcmp nnan ole double %x, 0.0 + %negx = fsub double 0.0, %x + %fabs = select i1 %lezero, double %negx, double %x + ret double %fabs +} + +define double @select_fcmp_ole_zero_no_nnan(double %x) { +; CHECK-LABEL: @select_fcmp_ole_zero_no_nnan( +; CHECK-NEXT: [[LEZERO:%.*]] = fcmp ole double [[X:%.*]], 0.000000e+00 +; CHECK-NEXT: [[NEGX:%.*]] = fsub double 0.000000e+00, [[X]] +; CHECK-NEXT: [[FABS:%.*]] = select i1 [[LEZERO]], double [[NEGX]], double [[X]] +; CHECK-NEXT: ret double [[FABS]] ; %lezero = fcmp ole double %x, 0.0 %negx = fsub double 0.0, %x @@ -268,7 +281,7 @@ define double @select_fcmp_nnan_ole_zero(double %x) { ; CHECK-NEXT: [[FABS:%.*]] = call double @llvm.fabs.f64(double [[X:%.*]]) ; CHECK-NEXT: ret double [[FABS]] ; - %lezero = fcmp ole double %x, 0.0 + %lezero = fcmp nnan ole double %x, 0.0 %negx = fsub nnan double 0.0, %x %fabs = select i1 %lezero, double %negx, double %x ret double %fabs @@ -279,7 +292,7 @@ define double @select_nnan_fcmp_nnan_ole_zero(double %x) { ; CHECK-NEXT: [[FABS:%.*]] = call nnan double @llvm.fabs.f64(double [[X:%.*]]) ; CHECK-NEXT: ret double [[FABS]] ; - %lezero = fcmp ole double %x, 0.0 + %lezero = fcmp nnan ole double %x, 0.0 %negx = fsub nnan double 0.0, %x %fabs = select nnan i1 %lezero, double %negx, double %x ret double %fabs @@ -292,7 +305,7 @@ define double @select_fcmp_nnan_ule_zero(double %x) { ; CHECK-NEXT: [[FABS:%.*]] = call double @llvm.fabs.f64(double [[X:%.*]]) ; CHECK-NEXT: ret double [[FABS]] ; - %lezero = fcmp ule double %x, 0.0 + %lezero = fcmp nnan ule double %x, 0.0 %negx = fsub nnan double 0.0, %x %fabs = select i1 %lezero, double %negx, double %x ret double %fabs @@ -320,7 +333,7 @@ define <2 x float> @select_fcmp_nnan_ole_negzero(<2 x float> %x) { ; CHECK-NEXT: [[FABS:%.*]] = call <2 x float> @llvm.fabs.v2f32(<2 x float> [[X:%.*]]) ; CHECK-NEXT: ret <2 x float> [[FABS]] ; - %lezero = fcmp ole <2 x float> %x, + %lezero = fcmp nnan ole <2 x float> %x, %negx = fsub nnan <2 x float> , %x %fabs = select <2 x i1> %lezero, <2 x float> %negx, <2 x float> %x ret <2 x float> %fabs @@ -331,7 +344,7 @@ define <2 x float> @select_nnan_fcmp_nnan_ole_negzero(<2 x float> %x) { ; CHECK-NEXT: [[FABS:%.*]] = call nnan <2 x float> @llvm.fabs.v2f32(<2 x float> [[X:%.*]]) ; CHECK-NEXT: ret <2 x float> [[FABS]] ; - %lezero = fcmp ole <2 x float> %x, + %lezero = fcmp nnan ole <2 x float> %x, %negx = fsub nnan <2 x float> , %x %fabs = select nnan <2 x i1> %lezero, <2 x float> %negx, <2 x float> %x ret <2 x float> %fabs @@ -344,7 +357,7 @@ define fp128 @select_fcmp_ogt_zero(fp128 %x) { ; CHECK-NEXT: [[FABS:%.*]] = call fp128 @llvm.fabs.f128(fp128 [[X:%.*]]) ; CHECK-NEXT: ret fp128 [[FABS]] ; - %gtzero = fcmp ogt fp128 %x, zeroinitializer + %gtzero = fcmp nnan ogt fp128 %x, zeroinitializer %negx = fsub fp128 zeroinitializer, %x %fabs = select i1 %gtzero, fp128 %x, fp128 %negx ret fp128 %fabs @@ -382,7 +395,7 @@ define fp128 @select_fcmp_nnan_ogt_zero(fp128 %x) { ; CHECK-NEXT: [[FABS:%.*]] = call fp128 @llvm.fabs.f128(fp128 [[X:%.*]]) ; CHECK-NEXT: ret fp128 [[FABS]] ; - %gtzero = fcmp ogt fp128 %x, zeroinitializer + %gtzero = fcmp nnan ogt fp128 %x, zeroinitializer %negx = fsub nnan fp128 zeroinitializer, %x %fabs = select i1 %gtzero, fp128 %x, fp128 %negx ret fp128 %fabs @@ -393,7 +406,7 @@ define fp128 @select_nnan_fcmp_nnan_ogt_zero(fp128 %x) { ; CHECK-NEXT: [[FABS:%.*]] = call nnan fp128 @llvm.fabs.f128(fp128 [[X:%.*]]) ; CHECK-NEXT: ret fp128 [[FABS]] ; - %gtzero = fcmp ogt fp128 %x, zeroinitializer + %gtzero = fcmp nnan ogt fp128 %x, zeroinitializer %negx = fsub nnan fp128 zeroinitializer, %x %fabs = select nnan i1 %gtzero, fp128 %x, fp128 %negx ret fp128 %fabs @@ -406,7 +419,7 @@ define half @select_fcmp_nnan_ogt_negzero(half %x) { ; CHECK-NEXT: [[FABS:%.*]] = call half @llvm.fabs.f16(half [[X:%.*]]) ; CHECK-NEXT: ret half [[FABS]] ; - %gtzero = fcmp ogt half %x, -0.0 + %gtzero = fcmp nnan ogt half %x, -0.0 %negx = fsub nnan half 0.0, %x %fabs = select i1 %gtzero, half %x, half %negx ret half %fabs @@ -417,7 +430,7 @@ define half @select_nnan_fcmp_nnan_ogt_negzero(half %x) { ; CHECK-NEXT: [[FABS:%.*]] = call nnan half @llvm.fabs.f16(half [[X:%.*]]) ; CHECK-NEXT: ret half [[FABS]] ; - %gtzero = fcmp ogt half %x, -0.0 + %gtzero = fcmp nnan ogt half %x, -0.0 %negx = fsub nnan half 0.0, %x %fabs = select nnan i1 %gtzero, half %x, half %negx ret half %fabs @@ -430,7 +443,7 @@ define half @select_fcmp_nnan_ugt_negzero(half %x) { ; CHECK-NEXT: [[FABS:%.*]] = call half @llvm.fabs.f16(half [[X:%.*]]) ; CHECK-NEXT: ret half [[FABS]] ; - %gtzero = fcmp ugt half %x, -0.0 + %gtzero = fcmp nnan ugt half %x, -0.0 %negx = fsub nnan half 0.0, %x %fabs = select i1 %gtzero, half %x, half %negx ret half %fabs From fc2ed05d4f98fc238a4d718df4e9f9d90ecd8116 Mon Sep 17 00:00:00 2001 From: Yingwei Zheng Date: Wed, 23 Apr 2025 20:08:18 +0800 Subject: [PATCH 2/3] [InstCombine] Use `isKnownNeverNaN` --- llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp | 7 +++++-- llvm/test/Transforms/InstCombine/fabs.ll | 11 +++++++++++ 2 files changed, 16 insertions(+), 2 deletions(-) diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp index 06792e32de3e1..0454089d78c96 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp @@ -2796,8 +2796,11 @@ static Instruction *foldSelectWithFCmpToFabs(SelectInst &SI, // Note: We require "nnan" for this fold because fcmp ignores the signbit // of NAN, but IEEE-754 specifies the signbit of NAN values with // fneg/fabs operations. - if (cast(CondVal)->hasNoNaNs() && - match(TrueVal, m_FSub(m_PosZeroFP(), m_Specific(X)))) { + if (match(TrueVal, m_FSub(m_PosZeroFP(), m_Specific(X))) && + (cast(CondVal)->hasNoNaNs() || + isKnownNeverNaN(X, /*Depth=*/0, + IC.getSimplifyQuery().getWithInstruction( + cast(CondVal))))) { if (!Swap && (Pred == FCmpInst::FCMP_OLE || Pred == FCmpInst::FCMP_ULE)) { Value *Fabs = IC.Builder.CreateUnaryIntrinsic(Intrinsic::fabs, X, &SI); return IC.replaceInstUsesWith(SI, Fabs); diff --git a/llvm/test/Transforms/InstCombine/fabs.ll b/llvm/test/Transforms/InstCombine/fabs.ll index 827315b184085..96f01b7bfe4e2 100644 --- a/llvm/test/Transforms/InstCombine/fabs.ll +++ b/llvm/test/Transforms/InstCombine/fabs.ll @@ -276,6 +276,17 @@ define double @select_fcmp_ole_zero_no_nnan(double %x) { ret double %fabs } +define double @select_fcmp_ole_zero_no_nnan_input_nofpclass_nan(double nofpclass(nan) %x) { +; CHECK-LABEL: @select_fcmp_ole_zero_no_nnan_input_nofpclass_nan( +; CHECK-NEXT: [[FABS:%.*]] = call double @llvm.fabs.f64(double [[X:%.*]]) +; CHECK-NEXT: ret double [[FABS]] +; + %lezero = fcmp ole double %x, 0.0 + %negx = fsub double 0.0, %x + %fabs = select i1 %lezero, double %negx, double %x + ret double %fabs +} + define double @select_fcmp_nnan_ole_zero(double %x) { ; CHECK-LABEL: @select_fcmp_nnan_ole_zero( ; CHECK-NEXT: [[FABS:%.*]] = call double @llvm.fabs.f64(double [[X:%.*]]) From 322bb136240559fc6597fd254e77b4ac08cba91d Mon Sep 17 00:00:00 2001 From: Yingwei Zheng Date: Wed, 23 Apr 2025 20:26:05 +0800 Subject: [PATCH 3/3] [InstCombine] Check nnan on select --- llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp | 2 +- llvm/test/Transforms/InstCombine/fabs.ll | 11 +++++++++++ 2 files changed, 12 insertions(+), 1 deletion(-) diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp index 0454089d78c96..5a42a5c996da5 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp @@ -2797,7 +2797,7 @@ static Instruction *foldSelectWithFCmpToFabs(SelectInst &SI, // of NAN, but IEEE-754 specifies the signbit of NAN values with // fneg/fabs operations. if (match(TrueVal, m_FSub(m_PosZeroFP(), m_Specific(X))) && - (cast(CondVal)->hasNoNaNs() || + (cast(CondVal)->hasNoNaNs() || SI.hasNoNaNs() || isKnownNeverNaN(X, /*Depth=*/0, IC.getSimplifyQuery().getWithInstruction( cast(CondVal))))) { diff --git a/llvm/test/Transforms/InstCombine/fabs.ll b/llvm/test/Transforms/InstCombine/fabs.ll index 96f01b7bfe4e2..f449d4b8e6b37 100644 --- a/llvm/test/Transforms/InstCombine/fabs.ll +++ b/llvm/test/Transforms/InstCombine/fabs.ll @@ -287,6 +287,17 @@ define double @select_fcmp_ole_zero_no_nnan_input_nofpclass_nan(double nofpclass ret double %fabs } +define double @select_fcmp_ole_zero_select_nnan(double %x) { +; CHECK-LABEL: @select_fcmp_ole_zero_select_nnan( +; CHECK-NEXT: [[FABS:%.*]] = call nnan double @llvm.fabs.f64(double [[X:%.*]]) +; CHECK-NEXT: ret double [[FABS]] +; + %lezero = fcmp ole double %x, 0.0 + %negx = fsub double 0.0, %x + %fabs = select nnan i1 %lezero, double %negx, double %x + ret double %fabs +} + define double @select_fcmp_nnan_ole_zero(double %x) { ; CHECK-LABEL: @select_fcmp_nnan_ole_zero( ; CHECK-NEXT: [[FABS:%.*]] = call double @llvm.fabs.f64(double [[X:%.*]])