[clang-tools-extra] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
dtcxzyw wrote: Gentle ping. https://github.com/llvm/llvm-project/pull/65934 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [clang][RISCV] Fix the condition of checking signature in getIndex (PR #67403)
https://github.com/dtcxzyw approved this pull request. LGTM https://github.com/llvm/llvm-project/pull/67403 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw closed https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw closed https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Canonicalize `(X +/- Y) & Y` into `~X & Y` when Y is a power of 2 (PR #67915)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/67915
>From c4ce28c942c172e5646b5922f0b02b4169197840 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sun, 1 Oct 2023 21:52:47 +0800
Subject: [PATCH 1/2] [InstCombine] Canonicalize `(X +/- Y) & Y` into `~X & Y`
when Y is a power of 2
---
.../InstCombine/InstCombineAndOrXor.cpp | 8
llvm/test/Transforms/InstCombine/and.ll | 44 +--
2 files changed, 48 insertions(+), 4 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index cbdab3e9c5fb91d..4322cc96f5a2b6c 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -2250,6 +2250,14 @@ Instruction *InstCombinerImpl::visitAnd(BinaryOperator
) {
return SelectInst::Create(Cmp, ConstantInt::getNullValue(Ty), Y);
}
+ // Canonicalize:
+ // (X +/- Y) & Y --> ~X & Y when Y is a power of 2.
+ if (match(, m_c_And(m_Value(Y), m_OneUse(m_CombineOr(
+m_c_Add(m_Value(X), m_Deferred(Y)),
+m_Sub(m_Value(X), m_Deferred(Y)) &&
+ isKnownToBeAPowerOfTwo(Y, /*OrZero*/ true, /*Depth*/ 0, ))
+return BinaryOperator::CreateAnd(Builder.CreateNot(X), Y);
+
const APInt *C;
if (match(Op1, m_APInt(C))) {
const APInt *XorC;
diff --git a/llvm/test/Transforms/InstCombine/and.ll
b/llvm/test/Transforms/InstCombine/and.ll
index 90f027010e2aea6..eb39ff9014ff3a4 100644
--- a/llvm/test/Transforms/InstCombine/and.ll
+++ b/llvm/test/Transforms/InstCombine/and.ll
@@ -1595,8 +1595,8 @@ define <2 x i8> @flip_masked_bit_uniform(<2 x i8> %A) {
define <2 x i8> @flip_masked_bit_undef(<2 x i8> %A) {
; CHECK-LABEL: @flip_masked_bit_undef(
-; CHECK-NEXT:[[B:%.*]] = add <2 x i8> [[A:%.*]],
-; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[B]],
+; CHECK-NEXT:[[TMP1:%.*]] = xor <2 x i8> [[A:%.*]],
+; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[TMP1]],
; CHECK-NEXT:ret <2 x i8> [[C]]
;
%B = add <2 x i8> %A,
@@ -1606,8 +1606,8 @@ define <2 x i8> @flip_masked_bit_undef(<2 x i8> %A) {
define <2 x i8> @flip_masked_bit_nonuniform(<2 x i8> %A) {
; CHECK-LABEL: @flip_masked_bit_nonuniform(
-; CHECK-NEXT:[[B:%.*]] = add <2 x i8> [[A:%.*]],
-; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[B]],
+; CHECK-NEXT:[[TMP1:%.*]] = xor <2 x i8> [[A:%.*]],
+; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[TMP1]],
; CHECK-NEXT:ret <2 x i8> [[C]]
;
%B = add <2 x i8> %A,
@@ -2546,3 +2546,39 @@ define i32 @and_zext_eq_zero(i32 %A, i32 %C) {
%5 = and i32 %2, %4
ret i32 %5
}
+
+define i32 @canonicalize_and_add_power2_or_zero(i32 %x, i32 %y) {
+; CHECK-LABEL: @canonicalize_and_add_power2_or_zero(
+; CHECK-NEXT:[[NY:%.*]] = sub i32 0, [[Y:%.*]]
+; CHECK-NEXT:[[P2:%.*]] = and i32 [[NY]], [[Y]]
+; CHECK-NEXT:call void @use32(i32 [[P2]])
+; CHECK-NEXT:[[TMP1:%.*]] = xor i32 [[X:%.*]], -1
+; CHECK-NEXT:[[AND:%.*]] = and i32 [[P2]], [[TMP1]]
+; CHECK-NEXT:ret i32 [[AND]]
+;
+ %ny = sub i32 0, %y
+ %p2 = and i32 %y, %ny
+ call void @use32(i32 %p2) ; keep p2
+
+ %val = add i32 %x, %p2
+ %and = and i32 %val, %p2
+ ret i32 %and
+}
+
+define i32 @canonicalize_and_sub_power2_or_zero(i32 %x, i32 %y) {
+; CHECK-LABEL: @canonicalize_and_sub_power2_or_zero(
+; CHECK-NEXT:[[NY:%.*]] = sub i32 0, [[Y:%.*]]
+; CHECK-NEXT:[[P2:%.*]] = and i32 [[NY]], [[Y]]
+; CHECK-NEXT:call void @use32(i32 [[P2]])
+; CHECK-NEXT:[[TMP1:%.*]] = xor i32 [[X:%.*]], -1
+; CHECK-NEXT:[[AND:%.*]] = and i32 [[P2]], [[TMP1]]
+; CHECK-NEXT:ret i32 [[AND]]
+;
+ %ny = sub i32 0, %y
+ %p2 = and i32 %y, %ny
+ call void @use32(i32 %p2) ; keep p2
+
+ %val = sub i32 %x, %p2
+ %and = and i32 %val, %p2
+ ret i32 %and
+}
>From b64ca5b5f743e6a935f4ea09154c3a08c6e65c47 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Fri, 6 Oct 2023 17:32:46 +0800
Subject: [PATCH 2/2] fixup! [InstCombine] Canonicalize `(X +/- Y) & Y` into
`~X & Y` when Y is a power of 2
Add additional tests.
---
llvm/test/Transforms/InstCombine/and.ll | 123
1 file changed, 123 insertions(+)
diff --git a/llvm/test/Transforms/InstCombine/and.ll
b/llvm/test/Transforms/InstCombine/and.ll
index eb39ff9014ff3a4..989640ed41f2d2e 100644
--- a/llvm/test/Transforms/InstCombine/and.ll
+++ b/llvm/test/Transforms/InstCombine/and.ll
@@ -2582,3 +2582,126 @@ define i32 @canonicalize_and_sub_power2_or_zero(i32 %x,
i32 %y) {
%and = and i32 %val, %p2
ret i32 %and
}
+
+define i32 @canonicalize_and_add_power2_or_zero_commuted1(i32 %x, i32 %y) {
+; CHECK-LABEL: @canonicalize_and_add_power2_or_zero_commuted1(
+; CHECK-NEXT:[[NY:%.*]] = sub i32 0, [[Y:%.*]]
+; CHECK-NEXT:[[P2:%.*]] = and i32 [[NY]], [[Y]]
+; CHECK-NEXT:call void @use32(i32 [[P2]])
+; CHECK-NEXT:[[TMP1:%.*]] = xor i32 [[X:%.*]],
[clang-tools-extra] [InstCombine] Canonicalize `(X +/- Y) & Y` into `~X & Y` when Y is a power of 2 (PR #67915)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/67915
>From c4ce28c942c172e5646b5922f0b02b4169197840 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sun, 1 Oct 2023 21:52:47 +0800
Subject: [PATCH 1/2] [InstCombine] Canonicalize `(X +/- Y) & Y` into `~X & Y`
when Y is a power of 2
---
.../InstCombine/InstCombineAndOrXor.cpp | 8
llvm/test/Transforms/InstCombine/and.ll | 44 +--
2 files changed, 48 insertions(+), 4 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index cbdab3e9c5fb91d..4322cc96f5a2b6c 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -2250,6 +2250,14 @@ Instruction *InstCombinerImpl::visitAnd(BinaryOperator
) {
return SelectInst::Create(Cmp, ConstantInt::getNullValue(Ty), Y);
}
+ // Canonicalize:
+ // (X +/- Y) & Y --> ~X & Y when Y is a power of 2.
+ if (match(, m_c_And(m_Value(Y), m_OneUse(m_CombineOr(
+m_c_Add(m_Value(X), m_Deferred(Y)),
+m_Sub(m_Value(X), m_Deferred(Y)) &&
+ isKnownToBeAPowerOfTwo(Y, /*OrZero*/ true, /*Depth*/ 0, ))
+return BinaryOperator::CreateAnd(Builder.CreateNot(X), Y);
+
const APInt *C;
if (match(Op1, m_APInt(C))) {
const APInt *XorC;
diff --git a/llvm/test/Transforms/InstCombine/and.ll
b/llvm/test/Transforms/InstCombine/and.ll
index 90f027010e2aea6..eb39ff9014ff3a4 100644
--- a/llvm/test/Transforms/InstCombine/and.ll
+++ b/llvm/test/Transforms/InstCombine/and.ll
@@ -1595,8 +1595,8 @@ define <2 x i8> @flip_masked_bit_uniform(<2 x i8> %A) {
define <2 x i8> @flip_masked_bit_undef(<2 x i8> %A) {
; CHECK-LABEL: @flip_masked_bit_undef(
-; CHECK-NEXT:[[B:%.*]] = add <2 x i8> [[A:%.*]],
-; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[B]],
+; CHECK-NEXT:[[TMP1:%.*]] = xor <2 x i8> [[A:%.*]],
+; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[TMP1]],
; CHECK-NEXT:ret <2 x i8> [[C]]
;
%B = add <2 x i8> %A,
@@ -1606,8 +1606,8 @@ define <2 x i8> @flip_masked_bit_undef(<2 x i8> %A) {
define <2 x i8> @flip_masked_bit_nonuniform(<2 x i8> %A) {
; CHECK-LABEL: @flip_masked_bit_nonuniform(
-; CHECK-NEXT:[[B:%.*]] = add <2 x i8> [[A:%.*]],
-; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[B]],
+; CHECK-NEXT:[[TMP1:%.*]] = xor <2 x i8> [[A:%.*]],
+; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[TMP1]],
; CHECK-NEXT:ret <2 x i8> [[C]]
;
%B = add <2 x i8> %A,
@@ -2546,3 +2546,39 @@ define i32 @and_zext_eq_zero(i32 %A, i32 %C) {
%5 = and i32 %2, %4
ret i32 %5
}
+
+define i32 @canonicalize_and_add_power2_or_zero(i32 %x, i32 %y) {
+; CHECK-LABEL: @canonicalize_and_add_power2_or_zero(
+; CHECK-NEXT:[[NY:%.*]] = sub i32 0, [[Y:%.*]]
+; CHECK-NEXT:[[P2:%.*]] = and i32 [[NY]], [[Y]]
+; CHECK-NEXT:call void @use32(i32 [[P2]])
+; CHECK-NEXT:[[TMP1:%.*]] = xor i32 [[X:%.*]], -1
+; CHECK-NEXT:[[AND:%.*]] = and i32 [[P2]], [[TMP1]]
+; CHECK-NEXT:ret i32 [[AND]]
+;
+ %ny = sub i32 0, %y
+ %p2 = and i32 %y, %ny
+ call void @use32(i32 %p2) ; keep p2
+
+ %val = add i32 %x, %p2
+ %and = and i32 %val, %p2
+ ret i32 %and
+}
+
+define i32 @canonicalize_and_sub_power2_or_zero(i32 %x, i32 %y) {
+; CHECK-LABEL: @canonicalize_and_sub_power2_or_zero(
+; CHECK-NEXT:[[NY:%.*]] = sub i32 0, [[Y:%.*]]
+; CHECK-NEXT:[[P2:%.*]] = and i32 [[NY]], [[Y]]
+; CHECK-NEXT:call void @use32(i32 [[P2]])
+; CHECK-NEXT:[[TMP1:%.*]] = xor i32 [[X:%.*]], -1
+; CHECK-NEXT:[[AND:%.*]] = and i32 [[P2]], [[TMP1]]
+; CHECK-NEXT:ret i32 [[AND]]
+;
+ %ny = sub i32 0, %y
+ %p2 = and i32 %y, %ny
+ call void @use32(i32 %p2) ; keep p2
+
+ %val = sub i32 %x, %p2
+ %and = and i32 %val, %p2
+ ret i32 %and
+}
>From b64ca5b5f743e6a935f4ea09154c3a08c6e65c47 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Fri, 6 Oct 2023 17:32:46 +0800
Subject: [PATCH 2/2] fixup! [InstCombine] Canonicalize `(X +/- Y) & Y` into
`~X & Y` when Y is a power of 2
Add additional tests.
---
llvm/test/Transforms/InstCombine/and.ll | 123
1 file changed, 123 insertions(+)
diff --git a/llvm/test/Transforms/InstCombine/and.ll
b/llvm/test/Transforms/InstCombine/and.ll
index eb39ff9014ff3a4..989640ed41f2d2e 100644
--- a/llvm/test/Transforms/InstCombine/and.ll
+++ b/llvm/test/Transforms/InstCombine/and.ll
@@ -2582,3 +2582,126 @@ define i32 @canonicalize_and_sub_power2_or_zero(i32 %x,
i32 %y) {
%and = and i32 %val, %p2
ret i32 %and
}
+
+define i32 @canonicalize_and_add_power2_or_zero_commuted1(i32 %x, i32 %y) {
+; CHECK-LABEL: @canonicalize_and_add_power2_or_zero_commuted1(
+; CHECK-NEXT:[[NY:%.*]] = sub i32 0, [[Y:%.*]]
+; CHECK-NEXT:[[P2:%.*]] = and i32 [[NY]], [[Y]]
+; CHECK-NEXT:call void @use32(i32 [[P2]])
+; CHECK-NEXT:[[TMP1:%.*]] = xor i32 [[X:%.*]],
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
dtcxzyw wrote: The conflict resolution should be correct now. https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
dtcxzyw wrote: The conflict resolution should be correct now. https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 01/10] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 01/10] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 01/10] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 01/10] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32
[clang] [InstCombine] Canonicalize `(X +/- Y) & Y` into `~X & Y` when Y is a power of 2 (PR #67915)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/67915
>From c4ce28c942c172e5646b5922f0b02b4169197840 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sun, 1 Oct 2023 21:52:47 +0800
Subject: [PATCH 1/2] [InstCombine] Canonicalize `(X +/- Y) & Y` into `~X & Y`
when Y is a power of 2
---
.../InstCombine/InstCombineAndOrXor.cpp | 8
llvm/test/Transforms/InstCombine/and.ll | 44 +--
2 files changed, 48 insertions(+), 4 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index cbdab3e9c5fb91d..4322cc96f5a2b6c 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -2250,6 +2250,14 @@ Instruction *InstCombinerImpl::visitAnd(BinaryOperator
) {
return SelectInst::Create(Cmp, ConstantInt::getNullValue(Ty), Y);
}
+ // Canonicalize:
+ // (X +/- Y) & Y --> ~X & Y when Y is a power of 2.
+ if (match(, m_c_And(m_Value(Y), m_OneUse(m_CombineOr(
+m_c_Add(m_Value(X), m_Deferred(Y)),
+m_Sub(m_Value(X), m_Deferred(Y)) &&
+ isKnownToBeAPowerOfTwo(Y, /*OrZero*/ true, /*Depth*/ 0, ))
+return BinaryOperator::CreateAnd(Builder.CreateNot(X), Y);
+
const APInt *C;
if (match(Op1, m_APInt(C))) {
const APInt *XorC;
diff --git a/llvm/test/Transforms/InstCombine/and.ll
b/llvm/test/Transforms/InstCombine/and.ll
index 90f027010e2aea6..eb39ff9014ff3a4 100644
--- a/llvm/test/Transforms/InstCombine/and.ll
+++ b/llvm/test/Transforms/InstCombine/and.ll
@@ -1595,8 +1595,8 @@ define <2 x i8> @flip_masked_bit_uniform(<2 x i8> %A) {
define <2 x i8> @flip_masked_bit_undef(<2 x i8> %A) {
; CHECK-LABEL: @flip_masked_bit_undef(
-; CHECK-NEXT:[[B:%.*]] = add <2 x i8> [[A:%.*]],
-; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[B]],
+; CHECK-NEXT:[[TMP1:%.*]] = xor <2 x i8> [[A:%.*]],
+; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[TMP1]],
; CHECK-NEXT:ret <2 x i8> [[C]]
;
%B = add <2 x i8> %A,
@@ -1606,8 +1606,8 @@ define <2 x i8> @flip_masked_bit_undef(<2 x i8> %A) {
define <2 x i8> @flip_masked_bit_nonuniform(<2 x i8> %A) {
; CHECK-LABEL: @flip_masked_bit_nonuniform(
-; CHECK-NEXT:[[B:%.*]] = add <2 x i8> [[A:%.*]],
-; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[B]],
+; CHECK-NEXT:[[TMP1:%.*]] = xor <2 x i8> [[A:%.*]],
+; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[TMP1]],
; CHECK-NEXT:ret <2 x i8> [[C]]
;
%B = add <2 x i8> %A,
@@ -2546,3 +2546,39 @@ define i32 @and_zext_eq_zero(i32 %A, i32 %C) {
%5 = and i32 %2, %4
ret i32 %5
}
+
+define i32 @canonicalize_and_add_power2_or_zero(i32 %x, i32 %y) {
+; CHECK-LABEL: @canonicalize_and_add_power2_or_zero(
+; CHECK-NEXT:[[NY:%.*]] = sub i32 0, [[Y:%.*]]
+; CHECK-NEXT:[[P2:%.*]] = and i32 [[NY]], [[Y]]
+; CHECK-NEXT:call void @use32(i32 [[P2]])
+; CHECK-NEXT:[[TMP1:%.*]] = xor i32 [[X:%.*]], -1
+; CHECK-NEXT:[[AND:%.*]] = and i32 [[P2]], [[TMP1]]
+; CHECK-NEXT:ret i32 [[AND]]
+;
+ %ny = sub i32 0, %y
+ %p2 = and i32 %y, %ny
+ call void @use32(i32 %p2) ; keep p2
+
+ %val = add i32 %x, %p2
+ %and = and i32 %val, %p2
+ ret i32 %and
+}
+
+define i32 @canonicalize_and_sub_power2_or_zero(i32 %x, i32 %y) {
+; CHECK-LABEL: @canonicalize_and_sub_power2_or_zero(
+; CHECK-NEXT:[[NY:%.*]] = sub i32 0, [[Y:%.*]]
+; CHECK-NEXT:[[P2:%.*]] = and i32 [[NY]], [[Y]]
+; CHECK-NEXT:call void @use32(i32 [[P2]])
+; CHECK-NEXT:[[TMP1:%.*]] = xor i32 [[X:%.*]], -1
+; CHECK-NEXT:[[AND:%.*]] = and i32 [[P2]], [[TMP1]]
+; CHECK-NEXT:ret i32 [[AND]]
+;
+ %ny = sub i32 0, %y
+ %p2 = and i32 %y, %ny
+ call void @use32(i32 %p2) ; keep p2
+
+ %val = sub i32 %x, %p2
+ %and = and i32 %val, %p2
+ ret i32 %and
+}
>From b64ca5b5f743e6a935f4ea09154c3a08c6e65c47 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Fri, 6 Oct 2023 17:32:46 +0800
Subject: [PATCH 2/2] fixup! [InstCombine] Canonicalize `(X +/- Y) & Y` into
`~X & Y` when Y is a power of 2
Add additional tests.
---
llvm/test/Transforms/InstCombine/and.ll | 123
1 file changed, 123 insertions(+)
diff --git a/llvm/test/Transforms/InstCombine/and.ll
b/llvm/test/Transforms/InstCombine/and.ll
index eb39ff9014ff3a4..989640ed41f2d2e 100644
--- a/llvm/test/Transforms/InstCombine/and.ll
+++ b/llvm/test/Transforms/InstCombine/and.ll
@@ -2582,3 +2582,126 @@ define i32 @canonicalize_and_sub_power2_or_zero(i32 %x,
i32 %y) {
%and = and i32 %val, %p2
ret i32 %and
}
+
+define i32 @canonicalize_and_add_power2_or_zero_commuted1(i32 %x, i32 %y) {
+; CHECK-LABEL: @canonicalize_and_add_power2_or_zero_commuted1(
+; CHECK-NEXT:[[NY:%.*]] = sub i32 0, [[Y:%.*]]
+; CHECK-NEXT:[[P2:%.*]] = and i32 [[NY]], [[Y]]
+; CHECK-NEXT:call void @use32(i32 [[P2]])
+; CHECK-NEXT:[[TMP1:%.*]] = xor i32 [[X:%.*]],
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 01/10] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 01/10] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32
[clang] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
https://github.com/dtcxzyw edited https://github.com/llvm/llvm-project/pull/67917 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
https://github.com/dtcxzyw edited https://github.com/llvm/llvm-project/pull/67917 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
@@ -1735,6 +1746,122 @@ ConstantRange ConstantRange::ctlz(bool ZeroIsPoison)
const {
return getNonEmpty(APInt(getBitWidth(), getUnsignedMax().countl_zero()),
APInt(getBitWidth(), getUnsignedMin().countl_zero() + 1));
}
+static ConstantRange getUnsignedCountTrailingZerosRange(const APInt ,
+const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.countr_zero()));
+ if (Lower.isZero())
+return ConstantRange(APInt::getZero(BitWidth),
+ APInt(BitWidth, BitWidth + 1));
+
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ (Upper - 1)).countl_zero();
+ // If Lower is {LCP, 000...}, the maximum is Lower.countr_zero().
+ // Otherwise, the maximum is BitWidth - LCPLength - 1 ({LCP, 100...}).
+ return ConstantRange(
+ APInt::getZero(BitWidth),
+ APInt(BitWidth, std::max(BitWidth - LCPLength, Lower.countr_zero() +
1)));
+}
+
+ConstantRange ConstantRange::cttz(bool ZeroIsPoison) const {
+ if (isEmptySet())
+return getEmpty();
+
+ APInt Zero = APInt::getZero(getBitWidth());
+
+ if (ZeroIsPoison && contains(Zero)) {
+// ZeroIsPoison is set, and zero is contained. We discern three cases, in
+// which a zero can appear:
+// 1) Lower is zero, handling cases of kind [0, 1), [0, 2), etc.
+// 2) Upper is zero, wrapped set, handling cases of kind [3, 0], etc.
+// 3) Zero contained in a wrapped set, e.g., [3, 2), [3, 1), etc.
+
+if (getLower().isZero()) {
+ if ((getUpper() - 1).isZero()) {
+// We have in input interval of kind [0, 1). In this case we cannot
+// really help but return empty-set.
+return getEmpty();
+ }
+
+ // Compute the resulting range by excluding zero from Lower.
+ return getUnsignedCountTrailingZerosRange(getLower() + 1, getUpper());
+} else if ((getUpper() - 1).isZero()) {
+ // Compute the resulting range by excluding zero from Upper.
+ return ConstantRange(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
+} else {
+ ConstantRange CR1(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
+ ConstantRange CR2 = getUnsignedCountTrailingZerosRange(
+ APInt(getBitWidth(), 1), getUpper());
+ return CR1.unionWith(CR2);
+}
+ }
+
+ if (isFullSet()) {
+return getNonEmpty(Zero, APInt(getBitWidth(), getBitWidth() + 1));
+ }
+ if (!isUpperWrapped()) {
+return getUnsignedCountTrailingZerosRange(getLower(), getUpper());
+ }
+ ConstantRange CR1(
+ Zero,
+ APInt(getBitWidth(), (getUnsignedMax() - getLower() + 1).logBase2() +
1));
+ ConstantRange CR2 = getUnsignedCountTrailingZerosRange(Zero, getUpper());
+ return CR1.unionWith(CR2);
+}
+
+static ConstantRange getUnsignedPopCountRange(const APInt ,
+ const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.popcount()));
+
+ APInt Max = Upper - 1;
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ Max).countl_zero();
+ unsigned LCPPopCount = Lower.getHiBits(LCPLength).popcount();
+ // If Lower is {LCP, 000...}, the minimum is the popcount of LCP.
+ // Otherwise, the minimum is the popcount of LCP + 1.
+ unsigned MinBits =
+ LCPPopCount + (Lower.countr_zero() < BitWidth - LCPLength ? 1 : 0);
+ // If Max is {LCP, 111...}, the maximum is the popcount of LCP + (BitWidth -
+ // length of LCP).
+ // Otherwise, the minimum is the popcount of LCP + (BitWidth -
+ // length of LCP - 1).
+ unsigned MaxBits = LCPPopCount + (BitWidth - LCPLength) +
+ (Max.countr_one() >= BitWidth - LCPLength ? 1 : 0);
dtcxzyw wrote:
I am sure that the implementation is precise. As nikic commented in
[D153505](https://reviews.llvm.org/D153505), non-constant-time methods were
unacceptable.
Please move to #68310 for further review of the `ctpop` parts.
https://github.com/llvm/llvm-project/pull/67917
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
@@ -1735,6 +1746,122 @@ ConstantRange ConstantRange::ctlz(bool ZeroIsPoison)
const {
return getNonEmpty(APInt(getBitWidth(), getUnsignedMax().countl_zero()),
APInt(getBitWidth(), getUnsignedMin().countl_zero() + 1));
}
+static ConstantRange getUnsignedCountTrailingZerosRange(const APInt ,
+const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.countr_zero()));
+ if (Lower.isZero())
+return ConstantRange(APInt::getZero(BitWidth),
+ APInt(BitWidth, BitWidth + 1));
+
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ (Upper - 1)).countl_zero();
+ // If Lower is {LCP, 000...}, the maximum is Lower.countr_zero().
+ // Otherwise, the maximum is BitWidth - LCPLength - 1 ({LCP, 100...}).
+ return ConstantRange(
+ APInt::getZero(BitWidth),
+ APInt(BitWidth, std::max(BitWidth - LCPLength, Lower.countr_zero() +
1)));
+}
+
+ConstantRange ConstantRange::cttz(bool ZeroIsPoison) const {
+ if (isEmptySet())
+return getEmpty();
+
+ APInt Zero = APInt::getZero(getBitWidth());
+
+ if (ZeroIsPoison && contains(Zero)) {
+// ZeroIsPoison is set, and zero is contained. We discern three cases, in
+// which a zero can appear:
+// 1) Lower is zero, handling cases of kind [0, 1), [0, 2), etc.
+// 2) Upper is zero, wrapped set, handling cases of kind [3, 0], etc.
+// 3) Zero contained in a wrapped set, e.g., [3, 2), [3, 1), etc.
+
+if (getLower().isZero()) {
+ if ((getUpper() - 1).isZero()) {
+// We have in input interval of kind [0, 1). In this case we cannot
+// really help but return empty-set.
+return getEmpty();
+ }
+
+ // Compute the resulting range by excluding zero from Lower.
+ return getUnsignedCountTrailingZerosRange(getLower() + 1, getUpper());
+} else if ((getUpper() - 1).isZero()) {
+ // Compute the resulting range by excluding zero from Upper.
+ return ConstantRange(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
+} else {
+ ConstantRange CR1(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
+ ConstantRange CR2 = getUnsignedCountTrailingZerosRange(
+ APInt(getBitWidth(), 1), getUpper());
+ return CR1.unionWith(CR2);
+}
+ }
+
+ if (isFullSet()) {
+return getNonEmpty(Zero, APInt(getBitWidth(), getBitWidth() + 1));
+ }
+ if (!isUpperWrapped()) {
+return getUnsignedCountTrailingZerosRange(getLower(), getUpper());
+ }
+ ConstantRange CR1(
+ Zero,
+ APInt(getBitWidth(), (getUnsignedMax() - getLower() + 1).logBase2() +
1));
+ ConstantRange CR2 = getUnsignedCountTrailingZerosRange(Zero, getUpper());
+ return CR1.unionWith(CR2);
+}
+
+static ConstantRange getUnsignedPopCountRange(const APInt ,
+ const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.popcount()));
+
+ APInt Max = Upper - 1;
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ Max).countl_zero();
+ unsigned LCPPopCount = Lower.getHiBits(LCPLength).popcount();
+ // If Lower is {LCP, 000...}, the minimum is the popcount of LCP.
+ // Otherwise, the minimum is the popcount of LCP + 1.
+ unsigned MinBits =
+ LCPPopCount + (Lower.countr_zero() < BitWidth - LCPLength ? 1 : 0);
+ // If Max is {LCP, 111...}, the maximum is the popcount of LCP + (BitWidth -
+ // length of LCP).
+ // Otherwise, the minimum is the popcount of LCP + (BitWidth -
+ // length of LCP - 1).
+ unsigned MaxBits = LCPPopCount + (BitWidth - LCPLength) +
+ (Max.countr_one() >= BitWidth - LCPLength ? 1 : 0);
dtcxzyw wrote:
I am sure that the implementation is precise. As nikic commented in
[D153505](https://reviews.llvm.org/D153505), non-constant-time methods were
unacceptable.
Please move to #68310 for further review of the `ctpop` parts.
https://github.com/llvm/llvm-project/pull/67917
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/67917
>From b5d134c88a04c524b1d9120a1c1a5dae3722904c Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sun, 1 Oct 2023 22:17:35 +0800
Subject: [PATCH 1/4] [ConstantRange] Handle `Intrinsic::cttz` and
`Intrinsic::ctpop`
---
llvm/include/llvm/IR/ConstantRange.h | 7 +
llvm/lib/IR/ConstantRange.cpp | 127 ++
.../CorrelatedValuePropagation/range.ll | 54
llvm/unittests/IR/ConstantRangeTest.cpp | 20 +++
4 files changed, 208 insertions(+)
diff --git a/llvm/include/llvm/IR/ConstantRange.h
b/llvm/include/llvm/IR/ConstantRange.h
index ca36732e4e2e8c2..e718e6e7e3403de 100644
--- a/llvm/include/llvm/IR/ConstantRange.h
+++ b/llvm/include/llvm/IR/ConstantRange.h
@@ -530,6 +530,13 @@ class [[nodiscard]] ConstantRange {
/// ignoring a possible zero value contained in the input range.
ConstantRange ctlz(bool ZeroIsPoison = false) const;
+ /// Calculate cttz range. If \p ZeroIsPoison is set, the range is computed
+ /// ignoring a possible zero value contained in the input range.
+ ConstantRange cttz(bool ZeroIsPoison = false) const;
+
+ /// Calculate ctpop range.
+ ConstantRange ctpop() const;
+
/// Represents whether an operation on the given constant range is known to
/// always or never overflow.
enum class OverflowResult {
diff --git a/llvm/lib/IR/ConstantRange.cpp b/llvm/lib/IR/ConstantRange.cpp
index 3d71b20f7e853e0..f34a2749543c321 100644
--- a/llvm/lib/IR/ConstantRange.cpp
+++ b/llvm/lib/IR/ConstantRange.cpp
@@ -949,6 +949,8 @@ bool ConstantRange::isIntrinsicSupported(Intrinsic::ID
IntrinsicID) {
case Intrinsic::smax:
case Intrinsic::abs:
case Intrinsic::ctlz:
+ case Intrinsic::cttz:
+ case Intrinsic::ctpop:
return true;
default:
return false;
@@ -986,6 +988,15 @@ ConstantRange ConstantRange::intrinsic(Intrinsic::ID
IntrinsicID,
assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
return Ops[0].ctlz(ZeroIsPoison->getBoolValue());
}
+ case Intrinsic::cttz: {
+const APInt *ZeroIsPoison = Ops[1].getSingleElement();
+assert(ZeroIsPoison && "Must be known (immarg)");
+assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
+return Ops[0].cttz(ZeroIsPoison->getBoolValue());
+ }
+ case Intrinsic::ctpop: {
+return Ops[0].ctpop();
+ }
default:
assert(!isIntrinsicSupported(IntrinsicID) && "Shouldn't be supported");
llvm_unreachable("Unsupported intrinsic");
@@ -1735,6 +1746,122 @@ ConstantRange ConstantRange::ctlz(bool ZeroIsPoison)
const {
return getNonEmpty(APInt(getBitWidth(), getUnsignedMax().countl_zero()),
APInt(getBitWidth(), getUnsignedMin().countl_zero() + 1));
}
+static ConstantRange getUnsignedCountTrailingZerosRange(const APInt ,
+const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.countr_zero()));
+ if (Lower.isZero())
+return ConstantRange(APInt::getZero(BitWidth),
+ APInt(BitWidth, BitWidth + 1));
+
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ (Upper - 1)).countl_zero();
+ // If Lower is {LCP, 000...}, the maximum is Lower.countr_zero().
+ // Otherwise, the maximum is BitWidth - LCPLength - 1 ({LCP, 100...}).
+ return ConstantRange(
+ APInt::getZero(BitWidth),
+ APInt(BitWidth, std::max(BitWidth - LCPLength, Lower.countr_zero() +
1)));
+}
+
+ConstantRange ConstantRange::cttz(bool ZeroIsPoison) const {
+ if (isEmptySet())
+return getEmpty();
+
+ APInt Zero = APInt::getZero(getBitWidth());
+
+ if (ZeroIsPoison && contains(Zero)) {
+// ZeroIsPoison is set, and zero is contained. We discern three cases, in
+// which a zero can appear:
+// 1) Lower is zero, handling cases of kind [0, 1), [0, 2), etc.
+// 2) Upper is zero, wrapped set, handling cases of kind [3, 0], etc.
+// 3) Zero contained in a wrapped set, e.g., [3, 2), [3, 1), etc.
+
+if (getLower().isZero()) {
+ if ((getUpper() - 1).isZero()) {
+// We have in input interval of kind [0, 1). In this case we cannot
+// really help but return empty-set.
+return getEmpty();
+ }
+
+ // Compute the resulting range by excluding zero from Lower.
+ return getUnsignedCountTrailingZerosRange(getLower() + 1, getUpper());
+} else if ((getUpper() - 1).isZero()) {
+ // Compute the resulting range by excluding zero from Upper.
+ return ConstantRange(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
+} else {
+ ConstantRange CR1(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() +
[clang-tools-extra] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/67917
>From b5d134c88a04c524b1d9120a1c1a5dae3722904c Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sun, 1 Oct 2023 22:17:35 +0800
Subject: [PATCH 1/4] [ConstantRange] Handle `Intrinsic::cttz` and
`Intrinsic::ctpop`
---
llvm/include/llvm/IR/ConstantRange.h | 7 +
llvm/lib/IR/ConstantRange.cpp | 127 ++
.../CorrelatedValuePropagation/range.ll | 54
llvm/unittests/IR/ConstantRangeTest.cpp | 20 +++
4 files changed, 208 insertions(+)
diff --git a/llvm/include/llvm/IR/ConstantRange.h
b/llvm/include/llvm/IR/ConstantRange.h
index ca36732e4e2e8c2..e718e6e7e3403de 100644
--- a/llvm/include/llvm/IR/ConstantRange.h
+++ b/llvm/include/llvm/IR/ConstantRange.h
@@ -530,6 +530,13 @@ class [[nodiscard]] ConstantRange {
/// ignoring a possible zero value contained in the input range.
ConstantRange ctlz(bool ZeroIsPoison = false) const;
+ /// Calculate cttz range. If \p ZeroIsPoison is set, the range is computed
+ /// ignoring a possible zero value contained in the input range.
+ ConstantRange cttz(bool ZeroIsPoison = false) const;
+
+ /// Calculate ctpop range.
+ ConstantRange ctpop() const;
+
/// Represents whether an operation on the given constant range is known to
/// always or never overflow.
enum class OverflowResult {
diff --git a/llvm/lib/IR/ConstantRange.cpp b/llvm/lib/IR/ConstantRange.cpp
index 3d71b20f7e853e0..f34a2749543c321 100644
--- a/llvm/lib/IR/ConstantRange.cpp
+++ b/llvm/lib/IR/ConstantRange.cpp
@@ -949,6 +949,8 @@ bool ConstantRange::isIntrinsicSupported(Intrinsic::ID
IntrinsicID) {
case Intrinsic::smax:
case Intrinsic::abs:
case Intrinsic::ctlz:
+ case Intrinsic::cttz:
+ case Intrinsic::ctpop:
return true;
default:
return false;
@@ -986,6 +988,15 @@ ConstantRange ConstantRange::intrinsic(Intrinsic::ID
IntrinsicID,
assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
return Ops[0].ctlz(ZeroIsPoison->getBoolValue());
}
+ case Intrinsic::cttz: {
+const APInt *ZeroIsPoison = Ops[1].getSingleElement();
+assert(ZeroIsPoison && "Must be known (immarg)");
+assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
+return Ops[0].cttz(ZeroIsPoison->getBoolValue());
+ }
+ case Intrinsic::ctpop: {
+return Ops[0].ctpop();
+ }
default:
assert(!isIntrinsicSupported(IntrinsicID) && "Shouldn't be supported");
llvm_unreachable("Unsupported intrinsic");
@@ -1735,6 +1746,122 @@ ConstantRange ConstantRange::ctlz(bool ZeroIsPoison)
const {
return getNonEmpty(APInt(getBitWidth(), getUnsignedMax().countl_zero()),
APInt(getBitWidth(), getUnsignedMin().countl_zero() + 1));
}
+static ConstantRange getUnsignedCountTrailingZerosRange(const APInt ,
+const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.countr_zero()));
+ if (Lower.isZero())
+return ConstantRange(APInt::getZero(BitWidth),
+ APInt(BitWidth, BitWidth + 1));
+
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ (Upper - 1)).countl_zero();
+ // If Lower is {LCP, 000...}, the maximum is Lower.countr_zero().
+ // Otherwise, the maximum is BitWidth - LCPLength - 1 ({LCP, 100...}).
+ return ConstantRange(
+ APInt::getZero(BitWidth),
+ APInt(BitWidth, std::max(BitWidth - LCPLength, Lower.countr_zero() +
1)));
+}
+
+ConstantRange ConstantRange::cttz(bool ZeroIsPoison) const {
+ if (isEmptySet())
+return getEmpty();
+
+ APInt Zero = APInt::getZero(getBitWidth());
+
+ if (ZeroIsPoison && contains(Zero)) {
+// ZeroIsPoison is set, and zero is contained. We discern three cases, in
+// which a zero can appear:
+// 1) Lower is zero, handling cases of kind [0, 1), [0, 2), etc.
+// 2) Upper is zero, wrapped set, handling cases of kind [3, 0], etc.
+// 3) Zero contained in a wrapped set, e.g., [3, 2), [3, 1), etc.
+
+if (getLower().isZero()) {
+ if ((getUpper() - 1).isZero()) {
+// We have in input interval of kind [0, 1). In this case we cannot
+// really help but return empty-set.
+return getEmpty();
+ }
+
+ // Compute the resulting range by excluding zero from Lower.
+ return getUnsignedCountTrailingZerosRange(getLower() + 1, getUpper());
+} else if ((getUpper() - 1).isZero()) {
+ // Compute the resulting range by excluding zero from Upper.
+ return ConstantRange(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
+} else {
+ ConstantRange CR1(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() +
[clang] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
@@ -1735,6 +1746,122 @@ ConstantRange ConstantRange::ctlz(bool ZeroIsPoison)
const {
return getNonEmpty(APInt(getBitWidth(), getUnsignedMax().countl_zero()),
APInt(getBitWidth(), getUnsignedMin().countl_zero() + 1));
}
+static ConstantRange getUnsignedCountTrailingZerosRange(const APInt ,
+const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.countr_zero()));
+ if (Lower.isZero())
+return ConstantRange(APInt::getZero(BitWidth),
+ APInt(BitWidth, BitWidth + 1));
+
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ (Upper - 1)).countl_zero();
+ // If Lower is {LCP, 000...}, the maximum is Lower.countr_zero().
+ // Otherwise, the maximum is BitWidth - LCPLength - 1 ({LCP, 100...}).
+ return ConstantRange(
+ APInt::getZero(BitWidth),
+ APInt(BitWidth, std::max(BitWidth - LCPLength, Lower.countr_zero() +
1)));
+}
+
+ConstantRange ConstantRange::cttz(bool ZeroIsPoison) const {
+ if (isEmptySet())
+return getEmpty();
+
+ APInt Zero = APInt::getZero(getBitWidth());
+
+ if (ZeroIsPoison && contains(Zero)) {
+// ZeroIsPoison is set, and zero is contained. We discern three cases, in
+// which a zero can appear:
+// 1) Lower is zero, handling cases of kind [0, 1), [0, 2), etc.
+// 2) Upper is zero, wrapped set, handling cases of kind [3, 0], etc.
+// 3) Zero contained in a wrapped set, e.g., [3, 2), [3, 1), etc.
+
+if (getLower().isZero()) {
+ if ((getUpper() - 1).isZero()) {
+// We have in input interval of kind [0, 1). In this case we cannot
+// really help but return empty-set.
+return getEmpty();
+ }
+
+ // Compute the resulting range by excluding zero from Lower.
+ return getUnsignedCountTrailingZerosRange(getLower() + 1, getUpper());
+} else if ((getUpper() - 1).isZero()) {
+ // Compute the resulting range by excluding zero from Upper.
+ return ConstantRange(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
dtcxzyw wrote:
It calculates the range of `cttz` for the input range `[Lower, UnsignedMax]`.
See also https://godbolt.org/z/jMhcs5375
https://github.com/llvm/llvm-project/pull/67917
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
@@ -1735,6 +1746,122 @@ ConstantRange ConstantRange::ctlz(bool ZeroIsPoison)
const {
return getNonEmpty(APInt(getBitWidth(), getUnsignedMax().countl_zero()),
APInt(getBitWidth(), getUnsignedMin().countl_zero() + 1));
}
+static ConstantRange getUnsignedCountTrailingZerosRange(const APInt ,
+const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.countr_zero()));
+ if (Lower.isZero())
+return ConstantRange(APInt::getZero(BitWidth),
+ APInt(BitWidth, BitWidth + 1));
+
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ (Upper - 1)).countl_zero();
+ // If Lower is {LCP, 000...}, the maximum is Lower.countr_zero().
+ // Otherwise, the maximum is BitWidth - LCPLength - 1 ({LCP, 100...}).
+ return ConstantRange(
+ APInt::getZero(BitWidth),
+ APInt(BitWidth, std::max(BitWidth - LCPLength, Lower.countr_zero() +
1)));
+}
+
+ConstantRange ConstantRange::cttz(bool ZeroIsPoison) const {
+ if (isEmptySet())
+return getEmpty();
+
+ APInt Zero = APInt::getZero(getBitWidth());
+
+ if (ZeroIsPoison && contains(Zero)) {
+// ZeroIsPoison is set, and zero is contained. We discern three cases, in
+// which a zero can appear:
+// 1) Lower is zero, handling cases of kind [0, 1), [0, 2), etc.
+// 2) Upper is zero, wrapped set, handling cases of kind [3, 0], etc.
+// 3) Zero contained in a wrapped set, e.g., [3, 2), [3, 1), etc.
+
+if (getLower().isZero()) {
+ if ((getUpper() - 1).isZero()) {
+// We have in input interval of kind [0, 1). In this case we cannot
+// really help but return empty-set.
+return getEmpty();
+ }
+
+ // Compute the resulting range by excluding zero from Lower.
+ return getUnsignedCountTrailingZerosRange(getLower() + 1, getUpper());
+} else if ((getUpper() - 1).isZero()) {
+ // Compute the resulting range by excluding zero from Upper.
+ return ConstantRange(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
dtcxzyw wrote:
It calculates the range of `cttz` for the input range `[Lower, UnsignedMax]`.
See also https://godbolt.org/z/jMhcs5375
https://github.com/llvm/llvm-project/pull/67917
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
https://github.com/dtcxzyw edited https://github.com/llvm/llvm-project/pull/67917 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/67917
>From b5d134c88a04c524b1d9120a1c1a5dae3722904c Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sun, 1 Oct 2023 22:17:35 +0800
Subject: [PATCH 1/3] [ConstantRange] Handle `Intrinsic::cttz` and
`Intrinsic::ctpop`
---
llvm/include/llvm/IR/ConstantRange.h | 7 +
llvm/lib/IR/ConstantRange.cpp | 127 ++
.../CorrelatedValuePropagation/range.ll | 54
llvm/unittests/IR/ConstantRangeTest.cpp | 20 +++
4 files changed, 208 insertions(+)
diff --git a/llvm/include/llvm/IR/ConstantRange.h
b/llvm/include/llvm/IR/ConstantRange.h
index ca36732e4e2e8c2..e718e6e7e3403de 100644
--- a/llvm/include/llvm/IR/ConstantRange.h
+++ b/llvm/include/llvm/IR/ConstantRange.h
@@ -530,6 +530,13 @@ class [[nodiscard]] ConstantRange {
/// ignoring a possible zero value contained in the input range.
ConstantRange ctlz(bool ZeroIsPoison = false) const;
+ /// Calculate cttz range. If \p ZeroIsPoison is set, the range is computed
+ /// ignoring a possible zero value contained in the input range.
+ ConstantRange cttz(bool ZeroIsPoison = false) const;
+
+ /// Calculate ctpop range.
+ ConstantRange ctpop() const;
+
/// Represents whether an operation on the given constant range is known to
/// always or never overflow.
enum class OverflowResult {
diff --git a/llvm/lib/IR/ConstantRange.cpp b/llvm/lib/IR/ConstantRange.cpp
index 3d71b20f7e853e0..f34a2749543c321 100644
--- a/llvm/lib/IR/ConstantRange.cpp
+++ b/llvm/lib/IR/ConstantRange.cpp
@@ -949,6 +949,8 @@ bool ConstantRange::isIntrinsicSupported(Intrinsic::ID
IntrinsicID) {
case Intrinsic::smax:
case Intrinsic::abs:
case Intrinsic::ctlz:
+ case Intrinsic::cttz:
+ case Intrinsic::ctpop:
return true;
default:
return false;
@@ -986,6 +988,15 @@ ConstantRange ConstantRange::intrinsic(Intrinsic::ID
IntrinsicID,
assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
return Ops[0].ctlz(ZeroIsPoison->getBoolValue());
}
+ case Intrinsic::cttz: {
+const APInt *ZeroIsPoison = Ops[1].getSingleElement();
+assert(ZeroIsPoison && "Must be known (immarg)");
+assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
+return Ops[0].cttz(ZeroIsPoison->getBoolValue());
+ }
+ case Intrinsic::ctpop: {
+return Ops[0].ctpop();
+ }
default:
assert(!isIntrinsicSupported(IntrinsicID) && "Shouldn't be supported");
llvm_unreachable("Unsupported intrinsic");
@@ -1735,6 +1746,122 @@ ConstantRange ConstantRange::ctlz(bool ZeroIsPoison)
const {
return getNonEmpty(APInt(getBitWidth(), getUnsignedMax().countl_zero()),
APInt(getBitWidth(), getUnsignedMin().countl_zero() + 1));
}
+static ConstantRange getUnsignedCountTrailingZerosRange(const APInt ,
+const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.countr_zero()));
+ if (Lower.isZero())
+return ConstantRange(APInt::getZero(BitWidth),
+ APInt(BitWidth, BitWidth + 1));
+
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ (Upper - 1)).countl_zero();
+ // If Lower is {LCP, 000...}, the maximum is Lower.countr_zero().
+ // Otherwise, the maximum is BitWidth - LCPLength - 1 ({LCP, 100...}).
+ return ConstantRange(
+ APInt::getZero(BitWidth),
+ APInt(BitWidth, std::max(BitWidth - LCPLength, Lower.countr_zero() +
1)));
+}
+
+ConstantRange ConstantRange::cttz(bool ZeroIsPoison) const {
+ if (isEmptySet())
+return getEmpty();
+
+ APInt Zero = APInt::getZero(getBitWidth());
+
+ if (ZeroIsPoison && contains(Zero)) {
+// ZeroIsPoison is set, and zero is contained. We discern three cases, in
+// which a zero can appear:
+// 1) Lower is zero, handling cases of kind [0, 1), [0, 2), etc.
+// 2) Upper is zero, wrapped set, handling cases of kind [3, 0], etc.
+// 3) Zero contained in a wrapped set, e.g., [3, 2), [3, 1), etc.
+
+if (getLower().isZero()) {
+ if ((getUpper() - 1).isZero()) {
+// We have in input interval of kind [0, 1). In this case we cannot
+// really help but return empty-set.
+return getEmpty();
+ }
+
+ // Compute the resulting range by excluding zero from Lower.
+ return getUnsignedCountTrailingZerosRange(getLower() + 1, getUpper());
+} else if ((getUpper() - 1).isZero()) {
+ // Compute the resulting range by excluding zero from Upper.
+ return ConstantRange(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
+} else {
+ ConstantRange CR1(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() +
[clang-tools-extra] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/67917
>From b5d134c88a04c524b1d9120a1c1a5dae3722904c Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sun, 1 Oct 2023 22:17:35 +0800
Subject: [PATCH 1/3] [ConstantRange] Handle `Intrinsic::cttz` and
`Intrinsic::ctpop`
---
llvm/include/llvm/IR/ConstantRange.h | 7 +
llvm/lib/IR/ConstantRange.cpp | 127 ++
.../CorrelatedValuePropagation/range.ll | 54
llvm/unittests/IR/ConstantRangeTest.cpp | 20 +++
4 files changed, 208 insertions(+)
diff --git a/llvm/include/llvm/IR/ConstantRange.h
b/llvm/include/llvm/IR/ConstantRange.h
index ca36732e4e2e8c2..e718e6e7e3403de 100644
--- a/llvm/include/llvm/IR/ConstantRange.h
+++ b/llvm/include/llvm/IR/ConstantRange.h
@@ -530,6 +530,13 @@ class [[nodiscard]] ConstantRange {
/// ignoring a possible zero value contained in the input range.
ConstantRange ctlz(bool ZeroIsPoison = false) const;
+ /// Calculate cttz range. If \p ZeroIsPoison is set, the range is computed
+ /// ignoring a possible zero value contained in the input range.
+ ConstantRange cttz(bool ZeroIsPoison = false) const;
+
+ /// Calculate ctpop range.
+ ConstantRange ctpop() const;
+
/// Represents whether an operation on the given constant range is known to
/// always or never overflow.
enum class OverflowResult {
diff --git a/llvm/lib/IR/ConstantRange.cpp b/llvm/lib/IR/ConstantRange.cpp
index 3d71b20f7e853e0..f34a2749543c321 100644
--- a/llvm/lib/IR/ConstantRange.cpp
+++ b/llvm/lib/IR/ConstantRange.cpp
@@ -949,6 +949,8 @@ bool ConstantRange::isIntrinsicSupported(Intrinsic::ID
IntrinsicID) {
case Intrinsic::smax:
case Intrinsic::abs:
case Intrinsic::ctlz:
+ case Intrinsic::cttz:
+ case Intrinsic::ctpop:
return true;
default:
return false;
@@ -986,6 +988,15 @@ ConstantRange ConstantRange::intrinsic(Intrinsic::ID
IntrinsicID,
assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
return Ops[0].ctlz(ZeroIsPoison->getBoolValue());
}
+ case Intrinsic::cttz: {
+const APInt *ZeroIsPoison = Ops[1].getSingleElement();
+assert(ZeroIsPoison && "Must be known (immarg)");
+assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
+return Ops[0].cttz(ZeroIsPoison->getBoolValue());
+ }
+ case Intrinsic::ctpop: {
+return Ops[0].ctpop();
+ }
default:
assert(!isIntrinsicSupported(IntrinsicID) && "Shouldn't be supported");
llvm_unreachable("Unsupported intrinsic");
@@ -1735,6 +1746,122 @@ ConstantRange ConstantRange::ctlz(bool ZeroIsPoison)
const {
return getNonEmpty(APInt(getBitWidth(), getUnsignedMax().countl_zero()),
APInt(getBitWidth(), getUnsignedMin().countl_zero() + 1));
}
+static ConstantRange getUnsignedCountTrailingZerosRange(const APInt ,
+const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.countr_zero()));
+ if (Lower.isZero())
+return ConstantRange(APInt::getZero(BitWidth),
+ APInt(BitWidth, BitWidth + 1));
+
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ (Upper - 1)).countl_zero();
+ // If Lower is {LCP, 000...}, the maximum is Lower.countr_zero().
+ // Otherwise, the maximum is BitWidth - LCPLength - 1 ({LCP, 100...}).
+ return ConstantRange(
+ APInt::getZero(BitWidth),
+ APInt(BitWidth, std::max(BitWidth - LCPLength, Lower.countr_zero() +
1)));
+}
+
+ConstantRange ConstantRange::cttz(bool ZeroIsPoison) const {
+ if (isEmptySet())
+return getEmpty();
+
+ APInt Zero = APInt::getZero(getBitWidth());
+
+ if (ZeroIsPoison && contains(Zero)) {
+// ZeroIsPoison is set, and zero is contained. We discern three cases, in
+// which a zero can appear:
+// 1) Lower is zero, handling cases of kind [0, 1), [0, 2), etc.
+// 2) Upper is zero, wrapped set, handling cases of kind [3, 0], etc.
+// 3) Zero contained in a wrapped set, e.g., [3, 2), [3, 1), etc.
+
+if (getLower().isZero()) {
+ if ((getUpper() - 1).isZero()) {
+// We have in input interval of kind [0, 1). In this case we cannot
+// really help but return empty-set.
+return getEmpty();
+ }
+
+ // Compute the resulting range by excluding zero from Lower.
+ return getUnsignedCountTrailingZerosRange(getLower() + 1, getUpper());
+} else if ((getUpper() - 1).isZero()) {
+ // Compute the resulting range by excluding zero from Upper.
+ return ConstantRange(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
+} else {
+ ConstantRange CR1(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() +
[clang-tools-extra] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
https://github.com/dtcxzyw edited https://github.com/llvm/llvm-project/pull/67917 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [ConstantRange] Handle `Intrinsic::cttz` (PR #67917)
https://github.com/dtcxzyw edited https://github.com/llvm/llvm-project/pull/67917 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [ConstantRange] Handle `Intrinsic::cttz` and `Intrinsic::ctpop` (PR #67917)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/67917
>From b5d134c88a04c524b1d9120a1c1a5dae3722904c Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sun, 1 Oct 2023 22:17:35 +0800
Subject: [PATCH 1/2] [ConstantRange] Handle `Intrinsic::cttz` and
`Intrinsic::ctpop`
---
llvm/include/llvm/IR/ConstantRange.h | 7 +
llvm/lib/IR/ConstantRange.cpp | 127 ++
.../CorrelatedValuePropagation/range.ll | 54
llvm/unittests/IR/ConstantRangeTest.cpp | 20 +++
4 files changed, 208 insertions(+)
diff --git a/llvm/include/llvm/IR/ConstantRange.h
b/llvm/include/llvm/IR/ConstantRange.h
index ca36732e4e2e8c2..e718e6e7e3403de 100644
--- a/llvm/include/llvm/IR/ConstantRange.h
+++ b/llvm/include/llvm/IR/ConstantRange.h
@@ -530,6 +530,13 @@ class [[nodiscard]] ConstantRange {
/// ignoring a possible zero value contained in the input range.
ConstantRange ctlz(bool ZeroIsPoison = false) const;
+ /// Calculate cttz range. If \p ZeroIsPoison is set, the range is computed
+ /// ignoring a possible zero value contained in the input range.
+ ConstantRange cttz(bool ZeroIsPoison = false) const;
+
+ /// Calculate ctpop range.
+ ConstantRange ctpop() const;
+
/// Represents whether an operation on the given constant range is known to
/// always or never overflow.
enum class OverflowResult {
diff --git a/llvm/lib/IR/ConstantRange.cpp b/llvm/lib/IR/ConstantRange.cpp
index 3d71b20f7e853e0..f34a2749543c321 100644
--- a/llvm/lib/IR/ConstantRange.cpp
+++ b/llvm/lib/IR/ConstantRange.cpp
@@ -949,6 +949,8 @@ bool ConstantRange::isIntrinsicSupported(Intrinsic::ID
IntrinsicID) {
case Intrinsic::smax:
case Intrinsic::abs:
case Intrinsic::ctlz:
+ case Intrinsic::cttz:
+ case Intrinsic::ctpop:
return true;
default:
return false;
@@ -986,6 +988,15 @@ ConstantRange ConstantRange::intrinsic(Intrinsic::ID
IntrinsicID,
assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
return Ops[0].ctlz(ZeroIsPoison->getBoolValue());
}
+ case Intrinsic::cttz: {
+const APInt *ZeroIsPoison = Ops[1].getSingleElement();
+assert(ZeroIsPoison && "Must be known (immarg)");
+assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
+return Ops[0].cttz(ZeroIsPoison->getBoolValue());
+ }
+ case Intrinsic::ctpop: {
+return Ops[0].ctpop();
+ }
default:
assert(!isIntrinsicSupported(IntrinsicID) && "Shouldn't be supported");
llvm_unreachable("Unsupported intrinsic");
@@ -1735,6 +1746,122 @@ ConstantRange ConstantRange::ctlz(bool ZeroIsPoison)
const {
return getNonEmpty(APInt(getBitWidth(), getUnsignedMax().countl_zero()),
APInt(getBitWidth(), getUnsignedMin().countl_zero() + 1));
}
+static ConstantRange getUnsignedCountTrailingZerosRange(const APInt ,
+const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.countr_zero()));
+ if (Lower.isZero())
+return ConstantRange(APInt::getZero(BitWidth),
+ APInt(BitWidth, BitWidth + 1));
+
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ (Upper - 1)).countl_zero();
+ // If Lower is {LCP, 000...}, the maximum is Lower.countr_zero().
+ // Otherwise, the maximum is BitWidth - LCPLength - 1 ({LCP, 100...}).
+ return ConstantRange(
+ APInt::getZero(BitWidth),
+ APInt(BitWidth, std::max(BitWidth - LCPLength, Lower.countr_zero() +
1)));
+}
+
+ConstantRange ConstantRange::cttz(bool ZeroIsPoison) const {
+ if (isEmptySet())
+return getEmpty();
+
+ APInt Zero = APInt::getZero(getBitWidth());
+
+ if (ZeroIsPoison && contains(Zero)) {
+// ZeroIsPoison is set, and zero is contained. We discern three cases, in
+// which a zero can appear:
+// 1) Lower is zero, handling cases of kind [0, 1), [0, 2), etc.
+// 2) Upper is zero, wrapped set, handling cases of kind [3, 0], etc.
+// 3) Zero contained in a wrapped set, e.g., [3, 2), [3, 1), etc.
+
+if (getLower().isZero()) {
+ if ((getUpper() - 1).isZero()) {
+// We have in input interval of kind [0, 1). In this case we cannot
+// really help but return empty-set.
+return getEmpty();
+ }
+
+ // Compute the resulting range by excluding zero from Lower.
+ return getUnsignedCountTrailingZerosRange(getLower() + 1, getUpper());
+} else if ((getUpper() - 1).isZero()) {
+ // Compute the resulting range by excluding zero from Upper.
+ return ConstantRange(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
+} else {
+ ConstantRange CR1(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() +
[clang] [ConstantRange] Handle `Intrinsic::cttz` and `Intrinsic::ctpop` (PR #67917)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/67917
>From b5d134c88a04c524b1d9120a1c1a5dae3722904c Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sun, 1 Oct 2023 22:17:35 +0800
Subject: [PATCH 1/2] [ConstantRange] Handle `Intrinsic::cttz` and
`Intrinsic::ctpop`
---
llvm/include/llvm/IR/ConstantRange.h | 7 +
llvm/lib/IR/ConstantRange.cpp | 127 ++
.../CorrelatedValuePropagation/range.ll | 54
llvm/unittests/IR/ConstantRangeTest.cpp | 20 +++
4 files changed, 208 insertions(+)
diff --git a/llvm/include/llvm/IR/ConstantRange.h
b/llvm/include/llvm/IR/ConstantRange.h
index ca36732e4e2e8c2..e718e6e7e3403de 100644
--- a/llvm/include/llvm/IR/ConstantRange.h
+++ b/llvm/include/llvm/IR/ConstantRange.h
@@ -530,6 +530,13 @@ class [[nodiscard]] ConstantRange {
/// ignoring a possible zero value contained in the input range.
ConstantRange ctlz(bool ZeroIsPoison = false) const;
+ /// Calculate cttz range. If \p ZeroIsPoison is set, the range is computed
+ /// ignoring a possible zero value contained in the input range.
+ ConstantRange cttz(bool ZeroIsPoison = false) const;
+
+ /// Calculate ctpop range.
+ ConstantRange ctpop() const;
+
/// Represents whether an operation on the given constant range is known to
/// always or never overflow.
enum class OverflowResult {
diff --git a/llvm/lib/IR/ConstantRange.cpp b/llvm/lib/IR/ConstantRange.cpp
index 3d71b20f7e853e0..f34a2749543c321 100644
--- a/llvm/lib/IR/ConstantRange.cpp
+++ b/llvm/lib/IR/ConstantRange.cpp
@@ -949,6 +949,8 @@ bool ConstantRange::isIntrinsicSupported(Intrinsic::ID
IntrinsicID) {
case Intrinsic::smax:
case Intrinsic::abs:
case Intrinsic::ctlz:
+ case Intrinsic::cttz:
+ case Intrinsic::ctpop:
return true;
default:
return false;
@@ -986,6 +988,15 @@ ConstantRange ConstantRange::intrinsic(Intrinsic::ID
IntrinsicID,
assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
return Ops[0].ctlz(ZeroIsPoison->getBoolValue());
}
+ case Intrinsic::cttz: {
+const APInt *ZeroIsPoison = Ops[1].getSingleElement();
+assert(ZeroIsPoison && "Must be known (immarg)");
+assert(ZeroIsPoison->getBitWidth() == 1 && "Must be boolean");
+return Ops[0].cttz(ZeroIsPoison->getBoolValue());
+ }
+ case Intrinsic::ctpop: {
+return Ops[0].ctpop();
+ }
default:
assert(!isIntrinsicSupported(IntrinsicID) && "Shouldn't be supported");
llvm_unreachable("Unsupported intrinsic");
@@ -1735,6 +1746,122 @@ ConstantRange ConstantRange::ctlz(bool ZeroIsPoison)
const {
return getNonEmpty(APInt(getBitWidth(), getUnsignedMax().countl_zero()),
APInt(getBitWidth(), getUnsignedMin().countl_zero() + 1));
}
+static ConstantRange getUnsignedCountTrailingZerosRange(const APInt ,
+const APInt ) {
+ assert(Lower.ule(Upper));
+ unsigned BitWidth = Lower.getBitWidth();
+ if (Lower == Upper)
+return ConstantRange::getEmpty(BitWidth);
+ if (Lower + 1 == Upper)
+return ConstantRange(APInt(BitWidth, Lower.countr_zero()));
+ if (Lower.isZero())
+return ConstantRange(APInt::getZero(BitWidth),
+ APInt(BitWidth, BitWidth + 1));
+
+ // Calculate longest common prefix.
+ unsigned LCPLength = (Lower ^ (Upper - 1)).countl_zero();
+ // If Lower is {LCP, 000...}, the maximum is Lower.countr_zero().
+ // Otherwise, the maximum is BitWidth - LCPLength - 1 ({LCP, 100...}).
+ return ConstantRange(
+ APInt::getZero(BitWidth),
+ APInt(BitWidth, std::max(BitWidth - LCPLength, Lower.countr_zero() +
1)));
+}
+
+ConstantRange ConstantRange::cttz(bool ZeroIsPoison) const {
+ if (isEmptySet())
+return getEmpty();
+
+ APInt Zero = APInt::getZero(getBitWidth());
+
+ if (ZeroIsPoison && contains(Zero)) {
+// ZeroIsPoison is set, and zero is contained. We discern three cases, in
+// which a zero can appear:
+// 1) Lower is zero, handling cases of kind [0, 1), [0, 2), etc.
+// 2) Upper is zero, wrapped set, handling cases of kind [3, 0], etc.
+// 3) Zero contained in a wrapped set, e.g., [3, 2), [3, 1), etc.
+
+if (getLower().isZero()) {
+ if ((getUpper() - 1).isZero()) {
+// We have in input interval of kind [0, 1). In this case we cannot
+// really help but return empty-set.
+return getEmpty();
+ }
+
+ // Compute the resulting range by excluding zero from Lower.
+ return getUnsignedCountTrailingZerosRange(getLower() + 1, getUpper());
+} else if ((getUpper() - 1).isZero()) {
+ // Compute the resulting range by excluding zero from Upper.
+ return ConstantRange(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() + 1).logBase2() + 1));
+} else {
+ ConstantRange CR1(
+ Zero, APInt(getBitWidth(),
+ (getUnsignedMax() - getLower() +
[clang] [InstCombine] Canonicalize `(X +/- Y) & Y` into `~X & Y` when Y is a power of 2 (PR #67915)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/67915
>From c4ce28c942c172e5646b5922f0b02b4169197840 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sun, 1 Oct 2023 21:52:47 +0800
Subject: [PATCH] [InstCombine] Canonicalize `(X +/- Y) & Y` into `~X & Y` when
Y is a power of 2
---
.../InstCombine/InstCombineAndOrXor.cpp | 8
llvm/test/Transforms/InstCombine/and.ll | 44 +--
2 files changed, 48 insertions(+), 4 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index cbdab3e9c5fb91d..4322cc96f5a2b6c 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -2250,6 +2250,14 @@ Instruction *InstCombinerImpl::visitAnd(BinaryOperator
) {
return SelectInst::Create(Cmp, ConstantInt::getNullValue(Ty), Y);
}
+ // Canonicalize:
+ // (X +/- Y) & Y --> ~X & Y when Y is a power of 2.
+ if (match(, m_c_And(m_Value(Y), m_OneUse(m_CombineOr(
+m_c_Add(m_Value(X), m_Deferred(Y)),
+m_Sub(m_Value(X), m_Deferred(Y)) &&
+ isKnownToBeAPowerOfTwo(Y, /*OrZero*/ true, /*Depth*/ 0, ))
+return BinaryOperator::CreateAnd(Builder.CreateNot(X), Y);
+
const APInt *C;
if (match(Op1, m_APInt(C))) {
const APInt *XorC;
diff --git a/llvm/test/Transforms/InstCombine/and.ll
b/llvm/test/Transforms/InstCombine/and.ll
index 90f027010e2aea6..eb39ff9014ff3a4 100644
--- a/llvm/test/Transforms/InstCombine/and.ll
+++ b/llvm/test/Transforms/InstCombine/and.ll
@@ -1595,8 +1595,8 @@ define <2 x i8> @flip_masked_bit_uniform(<2 x i8> %A) {
define <2 x i8> @flip_masked_bit_undef(<2 x i8> %A) {
; CHECK-LABEL: @flip_masked_bit_undef(
-; CHECK-NEXT:[[B:%.*]] = add <2 x i8> [[A:%.*]],
-; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[B]],
+; CHECK-NEXT:[[TMP1:%.*]] = xor <2 x i8> [[A:%.*]],
+; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[TMP1]],
; CHECK-NEXT:ret <2 x i8> [[C]]
;
%B = add <2 x i8> %A,
@@ -1606,8 +1606,8 @@ define <2 x i8> @flip_masked_bit_undef(<2 x i8> %A) {
define <2 x i8> @flip_masked_bit_nonuniform(<2 x i8> %A) {
; CHECK-LABEL: @flip_masked_bit_nonuniform(
-; CHECK-NEXT:[[B:%.*]] = add <2 x i8> [[A:%.*]],
-; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[B]],
+; CHECK-NEXT:[[TMP1:%.*]] = xor <2 x i8> [[A:%.*]],
+; CHECK-NEXT:[[C:%.*]] = and <2 x i8> [[TMP1]],
; CHECK-NEXT:ret <2 x i8> [[C]]
;
%B = add <2 x i8> %A,
@@ -2546,3 +2546,39 @@ define i32 @and_zext_eq_zero(i32 %A, i32 %C) {
%5 = and i32 %2, %4
ret i32 %5
}
+
+define i32 @canonicalize_and_add_power2_or_zero(i32 %x, i32 %y) {
+; CHECK-LABEL: @canonicalize_and_add_power2_or_zero(
+; CHECK-NEXT:[[NY:%.*]] = sub i32 0, [[Y:%.*]]
+; CHECK-NEXT:[[P2:%.*]] = and i32 [[NY]], [[Y]]
+; CHECK-NEXT:call void @use32(i32 [[P2]])
+; CHECK-NEXT:[[TMP1:%.*]] = xor i32 [[X:%.*]], -1
+; CHECK-NEXT:[[AND:%.*]] = and i32 [[P2]], [[TMP1]]
+; CHECK-NEXT:ret i32 [[AND]]
+;
+ %ny = sub i32 0, %y
+ %p2 = and i32 %y, %ny
+ call void @use32(i32 %p2) ; keep p2
+
+ %val = add i32 %x, %p2
+ %and = and i32 %val, %p2
+ ret i32 %and
+}
+
+define i32 @canonicalize_and_sub_power2_or_zero(i32 %x, i32 %y) {
+; CHECK-LABEL: @canonicalize_and_sub_power2_or_zero(
+; CHECK-NEXT:[[NY:%.*]] = sub i32 0, [[Y:%.*]]
+; CHECK-NEXT:[[P2:%.*]] = and i32 [[NY]], [[Y]]
+; CHECK-NEXT:call void @use32(i32 [[P2]])
+; CHECK-NEXT:[[TMP1:%.*]] = xor i32 [[X:%.*]], -1
+; CHECK-NEXT:[[AND:%.*]] = and i32 [[P2]], [[TMP1]]
+; CHECK-NEXT:ret i32 [[AND]]
+;
+ %ny = sub i32 0, %y
+ %p2 = and i32 %y, %ny
+ call void @use32(i32 %p2) ; keep p2
+
+ %val = sub i32 %x, %p2
+ %and = and i32 %val, %p2
+ ret i32 %and
+}
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
dtcxzyw wrote: Ping. https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
dtcxzyw wrote: Ping. https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
dtcxzyw wrote: Ping. https://github.com/llvm/llvm-project/pull/65934 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 1/9] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32 %a)
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 1/9] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32 %a)
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 1/9] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32 %a)
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 1/9] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32 %a)
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
@@ -6380,7 +6380,71 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ // icmp eq/ne X, (zext/sext (icmp eq/ne X, C))
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ Instruction *ExtI;
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_CombineAnd(m_Instruction(ExtI),
+ m_ZExtOrSExt(m_ICmp(Pred2, m_Deferred(X),
+ m_APInt(C))) {
dtcxzyw wrote:
Sorry, I accidentally dropped the checks in
70a70fb44d0e628a1cf485e1767ada3eaaa26b0f.
https://github.com/llvm/llvm-project/pull/65852
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65934
>From 2fe5756dd4d49580d3a23b0ff1b72535f725915e Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Mon, 11 Sep 2023 15:51:46 +0800
Subject: [PATCH 1/4] [RISCV] Eliminate dead li after emitting VSETVLIs
---
llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp | 22 ++---
.../RISCV/rvv/fixed-vectors-masked-gather.ll | 24 ---
2 files changed, 19 insertions(+), 27 deletions(-)
diff --git a/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
b/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
index b42ad269c18de6f..918c96beb29afca 100644
--- a/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
+++ b/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
@@ -757,7 +757,8 @@ class RISCVInsertVSETVLI : public MachineFunctionPass {
bool computeVLVTYPEChanges(const MachineBasicBlock ,
VSETVLIInfo ) const;
void computeIncomingVLVTYPE(const MachineBasicBlock );
- void emitVSETVLIs(MachineBasicBlock );
+ void emitVSETVLIs(MachineBasicBlock ,
+SmallVectorImpl );
void doLocalPostpass(MachineBasicBlock );
void doPRE(MachineBasicBlock );
void insertReadVL(MachineBasicBlock );
@@ -1216,7 +1217,8 @@ bool RISCVInsertVSETVLI::needVSETVLIPHI(const VSETVLIInfo
,
return false;
}
-void RISCVInsertVSETVLI::emitVSETVLIs(MachineBasicBlock ) {
+void RISCVInsertVSETVLI::emitVSETVLIs(
+MachineBasicBlock , SmallVectorImpl ) {
VSETVLIInfo CurInfo = BlockInfo[MBB.getNumber()].Pred;
// Track whether the prefix of the block we've scanned is transparent
// (meaning has not yet changed the abstract state).
@@ -1255,6 +1257,13 @@ void RISCVInsertVSETVLI::emitVSETVLIs(MachineBasicBlock
) {
MachineOperand = MI.getOperand(getVLOpNum(MI));
if (VLOp.isReg()) {
// Erase the AVL operand from the instruction.
+ if (MachineInstr *MI = MRI->getVRegDef(VLOp.getReg());
+ MI && MI->getOpcode() == RISCV::ADDI &&
+ MI->getOperand(1).isReg() && MI->getOperand(2).isImm() &&
+ MI->getOperand(1).getReg() == RISCV::X0 &&
+ MI->getOperand(2).getImm() != 0)
+DeadVLInstrs.push_back(MI);
+
VLOp.setReg(RISCV::NoRegister);
VLOp.setIsKill(false);
}
@@ -1580,8 +1589,9 @@ bool
RISCVInsertVSETVLI::runOnMachineFunction(MachineFunction ) {
// Phase 2 information to avoid adding vsetvlis before the first vector
// instruction in the block if the VL/VTYPE is satisfied by its
// predecessors.
+ SmallVector DeadVLInstrs;
for (MachineBasicBlock : MF)
-emitVSETVLIs(MBB);
+emitVSETVLIs(MBB, DeadVLInstrs);
// Now that all vsetvlis are explicit, go through and do block local
// DSE and peephole based demanded fields based transforms. Note that
@@ -1592,6 +1602,12 @@ bool
RISCVInsertVSETVLI::runOnMachineFunction(MachineFunction ) {
for (MachineBasicBlock : MF)
doLocalPostpass(MBB);
+ // Remove dead LI instructions that set VL.
+ for (MachineInstr *MI : DeadVLInstrs) {
+if (MRI->use_nodbg_empty(MI->getOperand(0).getReg()))
+ MI->eraseFromParent();
+ }
+
// Once we're fully done rewriting all the instructions, do a final pass
// through to check for VSETVLIs which write to an unused destination.
// For the non X0, X0 variant, we can replace the destination register
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
index f7352b4659e5a9b..0d306775528ed86 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
@@ -12393,7 +12393,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8> %idxs, <32 x i1> %m
; RV64ZVE32F-NEXT:vmv.x.s a2, v12
; RV64ZVE32F-NEXT:add a2, a0, a2
; RV64ZVE32F-NEXT:lbu a2, 0(a2)
-; RV64ZVE32F-NEXT:li a3, 32
; RV64ZVE32F-NEXT:vmv.s.x v12, a2
; RV64ZVE32F-NEXT:vsetivli zero, 2, e8, m2, tu, ma
; RV64ZVE32F-NEXT:vslideup.vi v10, v12, 1
@@ -12431,7 +12430,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8> %idxs, <32 x i1> %m
; RV64ZVE32F-NEXT:vmv.x.s a2, v14
; RV64ZVE32F-NEXT:add a2, a0, a2
; RV64ZVE32F-NEXT:lbu a2, 0(a2)
-; RV64ZVE32F-NEXT:li a3, 32
; RV64ZVE32F-NEXT:vmv.s.x v14, a2
; RV64ZVE32F-NEXT:vsetivli zero, 6, e8, m2, tu, ma
; RV64ZVE32F-NEXT:vslideup.vi v10, v14, 5
@@ -12455,7 +12453,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8> %idxs, <32 x i1> %m
; RV64ZVE32F-NEXT:vmv.x.s a2, v13
; RV64ZVE32F-NEXT:add a2, a0, a2
; RV64ZVE32F-NEXT:lbu a2, 0(a2)
-; RV64ZVE32F-NEXT:li a3, 32
; RV64ZVE32F-NEXT:vmv.s.x v14, a2
; RV64ZVE32F-NEXT:vsetivli zero, 10, e8, m2, tu, ma
; RV64ZVE32F-NEXT:vslideup.vi v10, v14, 9
@@ -12470,7 +12467,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8>
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
dtcxzyw wrote: Ping. https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Canonicalize `and(zext(A), B)` into `select A, B & 1, 0` (PR #66740)
https://github.com/dtcxzyw closed https://github.com/llvm/llvm-project/pull/66740 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Canonicalize `and(zext(A), B)` into `select A, B & 1, 0` (PR #66740)
https://github.com/dtcxzyw closed https://github.com/llvm/llvm-project/pull/66740 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [ValueTracking] Simplify uaddo pattern (PR #65910)
https://github.com/dtcxzyw closed https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Canonicalize `and(zext(A), B)` into `select A, B & 1, 0` (PR #66740)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/66740
>From 85f7911dfe0f1e9112881a9f503bcd68edfde580 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Tue, 19 Sep 2023 10:19:46 +0800
Subject: [PATCH 1/3] [InstCombine] Canonicalize `and(zext(A), B)` into `select
A, B & 1, 0`
---
.../InstCombine/InstCombineAndOrXor.cpp | 6
.../Transforms/InstCombine/and-or-icmps.ll| 12 +++
llvm/test/Transforms/InstCombine/icmp.ll | 31 ++-
llvm/test/Transforms/InstCombine/narrow.ll| 3 +-
.../Transforms/InstCombine/zext-or-icmp.ll| 10 +++---
5 files changed, 33 insertions(+), 29 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index ca7dfa82ab5a5d7..7fa432e1fdc1790 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -2654,6 +2654,12 @@ Instruction *InstCombinerImpl::visitAnd(BinaryOperator
) {
A->getType()->isIntOrIntVectorTy(1))
return SelectInst::Create(A, Constant::getNullValue(Ty), B);
+ // and(zext(A), B) -> A ? (B & 1) : 0
+ if (match(, m_c_And(m_OneUse(m_ZExt(m_Value(A))), m_Value(B))) &&
+ A->getType()->isIntOrIntVectorTy(1))
+return SelectInst::Create(A, Builder.CreateAnd(B, ConstantInt::get(Ty, 1)),
+ Constant::getNullValue(Ty));
+
// (iN X s>> (N-1)) & Y --> (X s< 0) ? Y : 0 -- with optional sext
if (match(, m_c_And(m_OneUse(m_SExtOrSelf(
m_AShr(m_Value(X), m_APIntAllowUndef(C,
diff --git a/llvm/test/Transforms/InstCombine/and-or-icmps.ll
b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
index 065dbf261e131bf..881a9b7ff129dbb 100644
--- a/llvm/test/Transforms/InstCombine/and-or-icmps.ll
+++ b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
@@ -2769,9 +2769,9 @@ define i64 @icmp_slt_0_and_icmp_sgt_neg1_i64(i64 %x) {
define i64 @icmp_slt_0_and_icmp_sge_neg1_i64_fail(i64 %x) {
; CHECK-LABEL: @icmp_slt_0_and_icmp_sge_neg1_i64_fail(
; CHECK-NEXT:[[A:%.*]] = icmp sgt i64 [[X:%.*]], -2
-; CHECK-NEXT:[[B:%.*]] = zext i1 [[A]] to i64
; CHECK-NEXT:[[C:%.*]] = lshr i64 [[X]], 62
-; CHECK-NEXT:[[D:%.*]] = and i64 [[C]], [[B]]
+; CHECK-NEXT:[[TMP1:%.*]] = and i64 [[C]], 1
+; CHECK-NEXT:[[D:%.*]] = select i1 [[A]], i64 [[TMP1]], i64 0
; CHECK-NEXT:ret i64 [[D]]
;
%A = icmp sge i64 %x, -1
@@ -2871,9 +2871,8 @@ define i32
@icmp_slt_0_and_icmp_sge_neg2_i32_multiuse1(i32 %x) {
define i32 @icmp_slt_0_and_icmp_sge_neg2_i32_multiuse2(i32 %x) {
; CHECK-LABEL: @icmp_slt_0_and_icmp_sge_neg2_i32_multiuse2(
; CHECK-NEXT:[[A:%.*]] = icmp sgt i32 [[X:%.*]], -3
-; CHECK-NEXT:[[B:%.*]] = zext i1 [[A]] to i32
; CHECK-NEXT:[[C:%.*]] = lshr i32 [[X]], 31
-; CHECK-NEXT:[[D:%.*]] = and i32 [[C]], [[B]]
+; CHECK-NEXT:[[D:%.*]] = select i1 [[A]], i32 [[C]], i32 0
; CHECK-NEXT:call void @use32(i32 [[C]])
; CHECK-NEXT:ret i32 [[D]]
;
@@ -2923,10 +2922,9 @@ define i32
@icmp_slt_0_or_icmp_eq_100_i32_multiuse_fail1(i32 %x) {
define i32 @icmp_x_slt_0_and_icmp_y_ne_neg2_i32_multiuse_fail2(i32 %x, i32 %y)
{
; CHECK-LABEL: @icmp_x_slt_0_and_icmp_y_ne_neg2_i32_multiuse_fail2(
-; CHECK-NEXT:[[A:%.*]] = icmp ne i32 [[X:%.*]], -2
-; CHECK-NEXT:[[B:%.*]] = zext i1 [[A]] to i32
+; CHECK-NEXT:[[A_NOT:%.*]] = icmp eq i32 [[X:%.*]], -2
; CHECK-NEXT:[[C:%.*]] = lshr i32 [[Y:%.*]], 31
-; CHECK-NEXT:[[D:%.*]] = and i32 [[C]], [[B]]
+; CHECK-NEXT:[[D:%.*]] = select i1 [[A_NOT]], i32 0, i32 [[C]]
; CHECK-NEXT:call void @use32(i32 [[C]])
; CHECK-NEXT:ret i32 [[D]]
;
diff --git a/llvm/test/Transforms/InstCombine/icmp.ll
b/llvm/test/Transforms/InstCombine/icmp.ll
index 666d85d6cc32103..a5c05278f4232e8 100644
--- a/llvm/test/Transforms/InstCombine/icmp.ll
+++ b/llvm/test/Transforms/InstCombine/icmp.ll
@@ -4441,9 +4441,9 @@ define i1 @redundant_sign_bit_count_ugt_31_30(i32 %x) {
define i1 @zext_bool_and_eq0(i1 %x, i8 %y) {
; CHECK-LABEL: @zext_bool_and_eq0(
; CHECK-NEXT:[[TMP1:%.*]] = and i8 [[Y:%.*]], 1
-; CHECK-NEXT:[[TMP2:%.*]] = icmp ne i8 [[TMP1]], 0
-; CHECK-NEXT:[[TMP3:%.*]] = and i1 [[TMP2]], [[X:%.*]]
-; CHECK-NEXT:[[R:%.*]] = xor i1 [[TMP3]], true
+; CHECK-NEXT:[[R1:%.*]] = icmp eq i8 [[TMP1]], 0
+; CHECK-NEXT:[[NOT_X:%.*]] = xor i1 [[X:%.*]], true
+; CHECK-NEXT:[[R:%.*]] = select i1 [[NOT_X]], i1 true, i1 [[R1]]
; CHECK-NEXT:ret i1 [[R]]
;
%zx = zext i1 %x to i8
@@ -4454,9 +4454,10 @@ define i1 @zext_bool_and_eq0(i1 %x, i8 %y) {
define <2 x i1> @zext_bool_and_eq0_commute(<2 x i1> %x, <2 x i8> %p) {
; CHECK-LABEL: @zext_bool_and_eq0_commute(
-; CHECK-NEXT:[[TMP1:%.*]] = trunc <2 x i8> [[P:%.*]] to <2 x i1>
-; CHECK-NEXT:[[TMP2:%.*]] = and <2 x i1> [[TMP1]], [[X:%.*]]
-; CHECK-NEXT:[[R:%.*]] = xor <2 x i1> [[TMP2]],
+; CHECK-NEXT:
[clang] [ValueTracking] Simplify uaddo pattern (PR #65910)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [ValueTracking] Simplify uaddo pattern (PR #65910)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Add combines/simplifications for `llvm.ptrmask` (PR #67166)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/67166 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Add combines/simplifications for `llvm.ptrmask` (PR #67166)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/67166 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65934 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65934 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65934 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65934 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
@@ -757,7 +757,8 @@ class RISCVInsertVSETVLI : public MachineFunctionPass {
bool computeVLVTYPEChanges(const MachineBasicBlock ,
VSETVLIInfo ) const;
void computeIncomingVLVTYPE(const MachineBasicBlock );
- void emitVSETVLIs(MachineBasicBlock );
+ void emitVSETVLIs(MachineBasicBlock ,
+SmallVectorImpl );
dtcxzyw wrote:
I don't think it is a typo.
https://github.com/llvm/llvm-project/pull/65934
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65934 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
@@ -757,7 +757,8 @@ class RISCVInsertVSETVLI : public MachineFunctionPass {
bool computeVLVTYPEChanges(const MachineBasicBlock ,
VSETVLIInfo ) const;
void computeIncomingVLVTYPE(const MachineBasicBlock );
- void emitVSETVLIs(MachineBasicBlock );
+ void emitVSETVLIs(MachineBasicBlock ,
+SmallVectorImpl );
dtcxzyw wrote:
I don't think it is a typo.
https://github.com/llvm/llvm-project/pull/65934
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65934 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65934 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65934
>From 2fe5756dd4d49580d3a23b0ff1b72535f725915e Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Mon, 11 Sep 2023 15:51:46 +0800
Subject: [PATCH 1/4] [RISCV] Eliminate dead li after emitting VSETVLIs
---
llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp | 22 ++---
.../RISCV/rvv/fixed-vectors-masked-gather.ll | 24 ---
2 files changed, 19 insertions(+), 27 deletions(-)
diff --git a/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
b/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
index b42ad269c18de6f..918c96beb29afca 100644
--- a/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
+++ b/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
@@ -757,7 +757,8 @@ class RISCVInsertVSETVLI : public MachineFunctionPass {
bool computeVLVTYPEChanges(const MachineBasicBlock ,
VSETVLIInfo ) const;
void computeIncomingVLVTYPE(const MachineBasicBlock );
- void emitVSETVLIs(MachineBasicBlock );
+ void emitVSETVLIs(MachineBasicBlock ,
+SmallVectorImpl );
void doLocalPostpass(MachineBasicBlock );
void doPRE(MachineBasicBlock );
void insertReadVL(MachineBasicBlock );
@@ -1216,7 +1217,8 @@ bool RISCVInsertVSETVLI::needVSETVLIPHI(const VSETVLIInfo
,
return false;
}
-void RISCVInsertVSETVLI::emitVSETVLIs(MachineBasicBlock ) {
+void RISCVInsertVSETVLI::emitVSETVLIs(
+MachineBasicBlock , SmallVectorImpl ) {
VSETVLIInfo CurInfo = BlockInfo[MBB.getNumber()].Pred;
// Track whether the prefix of the block we've scanned is transparent
// (meaning has not yet changed the abstract state).
@@ -1255,6 +1257,13 @@ void RISCVInsertVSETVLI::emitVSETVLIs(MachineBasicBlock
) {
MachineOperand = MI.getOperand(getVLOpNum(MI));
if (VLOp.isReg()) {
// Erase the AVL operand from the instruction.
+ if (MachineInstr *MI = MRI->getVRegDef(VLOp.getReg());
+ MI && MI->getOpcode() == RISCV::ADDI &&
+ MI->getOperand(1).isReg() && MI->getOperand(2).isImm() &&
+ MI->getOperand(1).getReg() == RISCV::X0 &&
+ MI->getOperand(2).getImm() != 0)
+DeadVLInstrs.push_back(MI);
+
VLOp.setReg(RISCV::NoRegister);
VLOp.setIsKill(false);
}
@@ -1580,8 +1589,9 @@ bool
RISCVInsertVSETVLI::runOnMachineFunction(MachineFunction ) {
// Phase 2 information to avoid adding vsetvlis before the first vector
// instruction in the block if the VL/VTYPE is satisfied by its
// predecessors.
+ SmallVector DeadVLInstrs;
for (MachineBasicBlock : MF)
-emitVSETVLIs(MBB);
+emitVSETVLIs(MBB, DeadVLInstrs);
// Now that all vsetvlis are explicit, go through and do block local
// DSE and peephole based demanded fields based transforms. Note that
@@ -1592,6 +1602,12 @@ bool
RISCVInsertVSETVLI::runOnMachineFunction(MachineFunction ) {
for (MachineBasicBlock : MF)
doLocalPostpass(MBB);
+ // Remove dead LI instructions that set VL.
+ for (MachineInstr *MI : DeadVLInstrs) {
+if (MRI->use_nodbg_empty(MI->getOperand(0).getReg()))
+ MI->eraseFromParent();
+ }
+
// Once we're fully done rewriting all the instructions, do a final pass
// through to check for VSETVLIs which write to an unused destination.
// For the non X0, X0 variant, we can replace the destination register
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
index f7352b4659e5a9b..0d306775528ed86 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
@@ -12393,7 +12393,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8> %idxs, <32 x i1> %m
; RV64ZVE32F-NEXT:vmv.x.s a2, v12
; RV64ZVE32F-NEXT:add a2, a0, a2
; RV64ZVE32F-NEXT:lbu a2, 0(a2)
-; RV64ZVE32F-NEXT:li a3, 32
; RV64ZVE32F-NEXT:vmv.s.x v12, a2
; RV64ZVE32F-NEXT:vsetivli zero, 2, e8, m2, tu, ma
; RV64ZVE32F-NEXT:vslideup.vi v10, v12, 1
@@ -12431,7 +12430,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8> %idxs, <32 x i1> %m
; RV64ZVE32F-NEXT:vmv.x.s a2, v14
; RV64ZVE32F-NEXT:add a2, a0, a2
; RV64ZVE32F-NEXT:lbu a2, 0(a2)
-; RV64ZVE32F-NEXT:li a3, 32
; RV64ZVE32F-NEXT:vmv.s.x v14, a2
; RV64ZVE32F-NEXT:vsetivli zero, 6, e8, m2, tu, ma
; RV64ZVE32F-NEXT:vslideup.vi v10, v14, 5
@@ -12455,7 +12453,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8> %idxs, <32 x i1> %m
; RV64ZVE32F-NEXT:vmv.x.s a2, v13
; RV64ZVE32F-NEXT:add a2, a0, a2
; RV64ZVE32F-NEXT:lbu a2, 0(a2)
-; RV64ZVE32F-NEXT:li a3, 32
; RV64ZVE32F-NEXT:vmv.s.x v14, a2
; RV64ZVE32F-NEXT:vsetivli zero, 10, e8, m2, tu, ma
; RV64ZVE32F-NEXT:vslideup.vi v10, v14, 9
@@ -12470,7 +12467,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8>
[clang-tools-extra] [RISCV] Eliminate dead li after emitting VSETVLIs (PR #65934)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65934
>From 2fe5756dd4d49580d3a23b0ff1b72535f725915e Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Mon, 11 Sep 2023 15:51:46 +0800
Subject: [PATCH 1/4] [RISCV] Eliminate dead li after emitting VSETVLIs
---
llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp | 22 ++---
.../RISCV/rvv/fixed-vectors-masked-gather.ll | 24 ---
2 files changed, 19 insertions(+), 27 deletions(-)
diff --git a/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
b/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
index b42ad269c18de6f..918c96beb29afca 100644
--- a/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
+++ b/llvm/lib/Target/RISCV/RISCVInsertVSETVLI.cpp
@@ -757,7 +757,8 @@ class RISCVInsertVSETVLI : public MachineFunctionPass {
bool computeVLVTYPEChanges(const MachineBasicBlock ,
VSETVLIInfo ) const;
void computeIncomingVLVTYPE(const MachineBasicBlock );
- void emitVSETVLIs(MachineBasicBlock );
+ void emitVSETVLIs(MachineBasicBlock ,
+SmallVectorImpl );
void doLocalPostpass(MachineBasicBlock );
void doPRE(MachineBasicBlock );
void insertReadVL(MachineBasicBlock );
@@ -1216,7 +1217,8 @@ bool RISCVInsertVSETVLI::needVSETVLIPHI(const VSETVLIInfo
,
return false;
}
-void RISCVInsertVSETVLI::emitVSETVLIs(MachineBasicBlock ) {
+void RISCVInsertVSETVLI::emitVSETVLIs(
+MachineBasicBlock , SmallVectorImpl ) {
VSETVLIInfo CurInfo = BlockInfo[MBB.getNumber()].Pred;
// Track whether the prefix of the block we've scanned is transparent
// (meaning has not yet changed the abstract state).
@@ -1255,6 +1257,13 @@ void RISCVInsertVSETVLI::emitVSETVLIs(MachineBasicBlock
) {
MachineOperand = MI.getOperand(getVLOpNum(MI));
if (VLOp.isReg()) {
// Erase the AVL operand from the instruction.
+ if (MachineInstr *MI = MRI->getVRegDef(VLOp.getReg());
+ MI && MI->getOpcode() == RISCV::ADDI &&
+ MI->getOperand(1).isReg() && MI->getOperand(2).isImm() &&
+ MI->getOperand(1).getReg() == RISCV::X0 &&
+ MI->getOperand(2).getImm() != 0)
+DeadVLInstrs.push_back(MI);
+
VLOp.setReg(RISCV::NoRegister);
VLOp.setIsKill(false);
}
@@ -1580,8 +1589,9 @@ bool
RISCVInsertVSETVLI::runOnMachineFunction(MachineFunction ) {
// Phase 2 information to avoid adding vsetvlis before the first vector
// instruction in the block if the VL/VTYPE is satisfied by its
// predecessors.
+ SmallVector DeadVLInstrs;
for (MachineBasicBlock : MF)
-emitVSETVLIs(MBB);
+emitVSETVLIs(MBB, DeadVLInstrs);
// Now that all vsetvlis are explicit, go through and do block local
// DSE and peephole based demanded fields based transforms. Note that
@@ -1592,6 +1602,12 @@ bool
RISCVInsertVSETVLI::runOnMachineFunction(MachineFunction ) {
for (MachineBasicBlock : MF)
doLocalPostpass(MBB);
+ // Remove dead LI instructions that set VL.
+ for (MachineInstr *MI : DeadVLInstrs) {
+if (MRI->use_nodbg_empty(MI->getOperand(0).getReg()))
+ MI->eraseFromParent();
+ }
+
// Once we're fully done rewriting all the instructions, do a final pass
// through to check for VSETVLIs which write to an unused destination.
// For the non X0, X0 variant, we can replace the destination register
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
index f7352b4659e5a9b..0d306775528ed86 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-masked-gather.ll
@@ -12393,7 +12393,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8> %idxs, <32 x i1> %m
; RV64ZVE32F-NEXT:vmv.x.s a2, v12
; RV64ZVE32F-NEXT:add a2, a0, a2
; RV64ZVE32F-NEXT:lbu a2, 0(a2)
-; RV64ZVE32F-NEXT:li a3, 32
; RV64ZVE32F-NEXT:vmv.s.x v12, a2
; RV64ZVE32F-NEXT:vsetivli zero, 2, e8, m2, tu, ma
; RV64ZVE32F-NEXT:vslideup.vi v10, v12, 1
@@ -12431,7 +12430,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8> %idxs, <32 x i1> %m
; RV64ZVE32F-NEXT:vmv.x.s a2, v14
; RV64ZVE32F-NEXT:add a2, a0, a2
; RV64ZVE32F-NEXT:lbu a2, 0(a2)
-; RV64ZVE32F-NEXT:li a3, 32
; RV64ZVE32F-NEXT:vmv.s.x v14, a2
; RV64ZVE32F-NEXT:vsetivli zero, 6, e8, m2, tu, ma
; RV64ZVE32F-NEXT:vslideup.vi v10, v14, 5
@@ -12455,7 +12453,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8> %idxs, <32 x i1> %m
; RV64ZVE32F-NEXT:vmv.x.s a2, v13
; RV64ZVE32F-NEXT:add a2, a0, a2
; RV64ZVE32F-NEXT:lbu a2, 0(a2)
-; RV64ZVE32F-NEXT:li a3, 32
; RV64ZVE32F-NEXT:vmv.s.x v14, a2
; RV64ZVE32F-NEXT:vsetivli zero, 10, e8, m2, tu, ma
; RV64ZVE32F-NEXT:vslideup.vi v10, v14, 9
@@ -12470,7 +12467,6 @@ define <32 x i8> @mgather_baseidx_v32i8(ptr %base, <32
x i8>
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 1/8] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32 %a)
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 1/8] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32 %a)
[clang] [ValueTracking] Simplify uaddo pattern (PR #65910)
https://github.com/dtcxzyw unresolved https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [ValueTracking] Simplify uaddo pattern (PR #65910)
@@ -8291,6 +8291,29 @@ static std::optional isImpliedCondICmps(const ICmpInst *LHS, if (L0 == R0 && match(L1, m_APInt(LC)) && match(R1, m_APInt(RC))) return isImpliedCondCommonOperandWithConstants(LPred, *LC, RPred, *RC); + // L0 = R0 = L1 + R1, L0 >=u L1 implies R0 >=u R1, L0 https://alive2.llvm.org/ce/z/eWsJH2. https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 1/7] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32 %a)
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 1/7] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32 %a)
[clang-tools-extra] [ValueTracking] Simplify uaddo pattern (PR #65910)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [ValueTracking] Simplify uaddo pattern (PR #65910)
@@ -8291,6 +8291,29 @@ static std::optional isImpliedCondICmps(const ICmpInst *LHS, if (L0 == R0 && match(L1, m_APInt(LC)) && match(R1, m_APInt(RC))) return isImpliedCondCommonOperandWithConstants(LPred, *LC, RPred, *RC); + // L0 = R0 = L1 + R1, L0 >=u L1 implies R0 >=u R1, L0 u L1 -> R0 >u R1` and `L0 <=u L1 -> R0 <=u R1`. Alive2: https://alive2.llvm.org/ce/z/VKtFkD https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [ValueTracking] Simplify uaddo pattern (PR #65910)
@@ -8291,6 +8291,29 @@ static std::optional isImpliedCondICmps(const ICmpInst *LHS, if (L0 == R0 && match(L1, m_APInt(LC)) && match(R1, m_APInt(RC))) return isImpliedCondCommonOperandWithConstants(LPred, *LC, RPred, *RC); + // L0 = R0 = L1 + R1, L0 >=u L1 implies R0 >=u R1, L0 u L1 -> R0 >u R1` and `L0 <=u L1 -> R0 <=u R1`. Alive2: https://alive2.llvm.org/ce/z/VKtFkD https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [ValueTracking] Simplify uaddo pattern (PR #65910)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [ValueTracking] Simplify uaddo pattern (PR #65910)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [ValueTracking] Simplify uaddo pattern (PR #65910)
https://github.com/dtcxzyw updated https://github.com/llvm/llvm-project/pull/65910 >From 9bb817aa40dd9bc1bbb18b4cf4bc079145c8ecaa Mon Sep 17 00:00:00 2001 From: Yingwei Zheng Date: Mon, 11 Sep 2023 03:58:02 +0800 Subject: [PATCH 1/2] [ValueTracking] Simplify uaddo pattern --- llvm/lib/Analysis/ValueTracking.cpp | 23 .../InstSimplify/and-or-implied-cond.ll | 104 ++ 2 files changed, 127 insertions(+) diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp index c4153b824c37e0a..5aaff4ee81be969 100644 --- a/llvm/lib/Analysis/ValueTracking.cpp +++ b/llvm/lib/Analysis/ValueTracking.cpp @@ -8291,6 +8291,29 @@ static std::optional isImpliedCondICmps(const ICmpInst *LHS, if (L0 == R0 && match(L1, m_APInt(LC)) && match(R1, m_APInt(RC))) return isImpliedCondCommonOperandWithConstants(LPred, *LC, RPred, *RC); + // L0 = R0 = L1 + R1, L0 >=u L1 implies R0 >=u R1, L0 From 83fa03f89c69a0a13336f6761023bb1415f92505 Mon Sep 17 00:00:00 2001 From: Yingwei Zheng Date: Sun, 24 Sep 2023 17:04:32 +0800 Subject: [PATCH 2/2] fixup! [ValueTracking] Simplify uaddo pattern --- llvm/lib/Analysis/ValueTracking.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp index 5aaff4ee81be969..f2278f3b7097dac 100644 --- a/llvm/lib/Analysis/ValueTracking.cpp +++ b/llvm/lib/Analysis/ValueTracking.cpp @@ -8308,7 +8308,7 @@ static std::optional isImpliedCondICmps(const ICmpInst *LHS, RPred = ICmpInst::getSwappedPredicate(RPred); } if (L0 == R0 && -(LPred == ICmpInst::ICMP_UGE || LPred == ICmpInst::ICMP_ULT) && +(LPred == ICmpInst::ICMP_ULT || LPred == ICmpInst::ICMP_UGE) && (RPred == ICmpInst::ICMP_ULT || RPred == ICmpInst::ICMP_UGE) && match(L0, m_c_Add(m_Specific(L1), m_Specific(R1 return LPred == RPred; ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [ValueTracking] Simplify uaddo pattern (PR #65910)
https://github.com/dtcxzyw updated https://github.com/llvm/llvm-project/pull/65910 >From 9bb817aa40dd9bc1bbb18b4cf4bc079145c8ecaa Mon Sep 17 00:00:00 2001 From: Yingwei Zheng Date: Mon, 11 Sep 2023 03:58:02 +0800 Subject: [PATCH 1/2] [ValueTracking] Simplify uaddo pattern --- llvm/lib/Analysis/ValueTracking.cpp | 23 .../InstSimplify/and-or-implied-cond.ll | 104 ++ 2 files changed, 127 insertions(+) diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp index c4153b824c37e0a..5aaff4ee81be969 100644 --- a/llvm/lib/Analysis/ValueTracking.cpp +++ b/llvm/lib/Analysis/ValueTracking.cpp @@ -8291,6 +8291,29 @@ static std::optional isImpliedCondICmps(const ICmpInst *LHS, if (L0 == R0 && match(L1, m_APInt(LC)) && match(R1, m_APInt(RC))) return isImpliedCondCommonOperandWithConstants(LPred, *LC, RPred, *RC); + // L0 = R0 = L1 + R1, L0 >=u L1 implies R0 >=u R1, L0 From 83fa03f89c69a0a13336f6761023bb1415f92505 Mon Sep 17 00:00:00 2001 From: Yingwei Zheng Date: Sun, 24 Sep 2023 17:04:32 +0800 Subject: [PATCH 2/2] fixup! [ValueTracking] Simplify uaddo pattern --- llvm/lib/Analysis/ValueTracking.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp index 5aaff4ee81be969..f2278f3b7097dac 100644 --- a/llvm/lib/Analysis/ValueTracking.cpp +++ b/llvm/lib/Analysis/ValueTracking.cpp @@ -8308,7 +8308,7 @@ static std::optional isImpliedCondICmps(const ICmpInst *LHS, RPred = ICmpInst::getSwappedPredicate(RPred); } if (L0 == R0 && -(LPred == ICmpInst::ICMP_UGE || LPred == ICmpInst::ICMP_ULT) && +(LPred == ICmpInst::ICMP_ULT || LPred == ICmpInst::ICMP_UGE) && (RPred == ICmpInst::ICMP_ULT || RPred == ICmpInst::ICMP_UGE) && match(L0, m_c_Add(m_Specific(L1), m_Specific(R1 return LPred == RPred; ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Add combines/simplifications for `llvm.ptrmask` (PR #67166)
@@ -1964,18 +1964,78 @@ Instruction *InstCombinerImpl::visitCallInst(CallInst
) {
break;
}
case Intrinsic::ptrmask: {
+KnownBits Known(DL.getPointerTypeSizeInBits(II->getType()));
+if (SimplifyDemandedInstructionBits(*II, Known))
+ return II;
+
+Value *Op0 = II->getArgOperand(0);
+Value *Op1 = II->getArgOperand(1);
+// Fail loudly in case this is ever changed.
+// TODO: If vector types are supported the merging of (ptrmask (ptrmask))
+// need to ensure we don't merge a vectype with non-vec type.
+assert(!Op0->getType()->isVectorTy() && !Op1->getType()->isVectorTy() &&
+ "These combines where written at a time when ptrmask did not "
+ "support vector types and may not work for vectors");
+
Value *InnerPtr, *InnerMask;
-if (match(II->getArgOperand(0),
- m_OneUse(m_Intrinsic(m_Value(InnerPtr),
- m_Value(InnerMask) {
- if (II->getArgOperand(1)->getType() == InnerMask->getType()) {
-Value *NewMask = Builder.CreateAnd(II->getArgOperand(1), InnerMask);
-return replaceInstUsesWith(
-*II,
-Builder.CreateIntrinsic(InnerPtr->getType(), Intrinsic::ptrmask,
-{InnerPtr, NewMask}));
+bool Changed = false;
+// Combine:
+// (ptrmask (ptrmask p, A), B)
+//-> (ptrmask p, (and A, B))
+if (match(Op0, m_OneUse(m_Intrinsic(
+ m_Value(InnerPtr), m_Value(InnerMask) {
+ // See if combining the two masks is free.
+ bool OkayToMerge = InnerMask->getType() == Op1->getType();
+ bool NeedsNew = false;
+ if (!OkayToMerge) {
+if (match(InnerMask, m_ImmConstant())) {
+ InnerMask = Builder.CreateZExtOrTrunc(InnerMask, Op1->getType());
+ OkayToMerge = true;
+} else if (match(Op1, m_ImmConstant())) {
+ Op1 = Builder.CreateZExtOrTrunc(Op1, InnerMask->getType());
+ OkayToMerge = true;
+ // Need to create a new one here, as the intrinsic id needs to
change.
+ NeedsNew = true;
+}
+ }
+ if (InnerMask->getType() == Op1->getType()) {
dtcxzyw wrote:
`OkayToMerge` is unused?
https://github.com/llvm/llvm-project/pull/67166
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Canonicalize `and(zext(A), B)` into `select A, B & 1, 0` (PR #66740)
dtcxzyw wrote: To avoid backend regressions, I will merge this PR after #66793. https://github.com/llvm/llvm-project/pull/66740 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Canonicalize `and(zext(A), B)` into `select A, B & 1, 0` (PR #66740)
dtcxzyw wrote: > You might want to check whether this also fixes #66740 and if so add a test. Confirmed. I have added a pre-commit test for 66733 in 1a73a6b80b54cd23ee8380c6fa3304847e5cb5d1. https://github.com/llvm/llvm-project/pull/66740 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Canonicalize `and(zext(A), B)` into `select A, B & 1, 0` (PR #66740)
dtcxzyw wrote: > You might want to check whether this also fixes #66740 and if so add a test. Confirmed. I have added a pre-commit test for 66733 in 1a73a6b80b54cd23ee8380c6fa3304847e5cb5d1. https://github.com/llvm/llvm-project/pull/66740 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Canonicalize `and(zext(A), B)` into `select A, B & 1, 0` (PR #66740)
https://github.com/dtcxzyw edited https://github.com/llvm/llvm-project/pull/66740 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Canonicalize `and(zext(A), B)` into `select A, B & 1, 0` (PR #66740)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/66740
>From 85f7911dfe0f1e9112881a9f503bcd68edfde580 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Tue, 19 Sep 2023 10:19:46 +0800
Subject: [PATCH 1/3] [InstCombine] Canonicalize `and(zext(A), B)` into `select
A, B & 1, 0`
---
.../InstCombine/InstCombineAndOrXor.cpp | 6
.../Transforms/InstCombine/and-or-icmps.ll| 12 +++
llvm/test/Transforms/InstCombine/icmp.ll | 31 ++-
llvm/test/Transforms/InstCombine/narrow.ll| 3 +-
.../Transforms/InstCombine/zext-or-icmp.ll| 10 +++---
5 files changed, 33 insertions(+), 29 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index ca7dfa82ab5a5d7..7fa432e1fdc1790 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -2654,6 +2654,12 @@ Instruction *InstCombinerImpl::visitAnd(BinaryOperator
) {
A->getType()->isIntOrIntVectorTy(1))
return SelectInst::Create(A, Constant::getNullValue(Ty), B);
+ // and(zext(A), B) -> A ? (B & 1) : 0
+ if (match(, m_c_And(m_OneUse(m_ZExt(m_Value(A))), m_Value(B))) &&
+ A->getType()->isIntOrIntVectorTy(1))
+return SelectInst::Create(A, Builder.CreateAnd(B, ConstantInt::get(Ty, 1)),
+ Constant::getNullValue(Ty));
+
// (iN X s>> (N-1)) & Y --> (X s< 0) ? Y : 0 -- with optional sext
if (match(, m_c_And(m_OneUse(m_SExtOrSelf(
m_AShr(m_Value(X), m_APIntAllowUndef(C,
diff --git a/llvm/test/Transforms/InstCombine/and-or-icmps.ll
b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
index 065dbf261e131bf..881a9b7ff129dbb 100644
--- a/llvm/test/Transforms/InstCombine/and-or-icmps.ll
+++ b/llvm/test/Transforms/InstCombine/and-or-icmps.ll
@@ -2769,9 +2769,9 @@ define i64 @icmp_slt_0_and_icmp_sgt_neg1_i64(i64 %x) {
define i64 @icmp_slt_0_and_icmp_sge_neg1_i64_fail(i64 %x) {
; CHECK-LABEL: @icmp_slt_0_and_icmp_sge_neg1_i64_fail(
; CHECK-NEXT:[[A:%.*]] = icmp sgt i64 [[X:%.*]], -2
-; CHECK-NEXT:[[B:%.*]] = zext i1 [[A]] to i64
; CHECK-NEXT:[[C:%.*]] = lshr i64 [[X]], 62
-; CHECK-NEXT:[[D:%.*]] = and i64 [[C]], [[B]]
+; CHECK-NEXT:[[TMP1:%.*]] = and i64 [[C]], 1
+; CHECK-NEXT:[[D:%.*]] = select i1 [[A]], i64 [[TMP1]], i64 0
; CHECK-NEXT:ret i64 [[D]]
;
%A = icmp sge i64 %x, -1
@@ -2871,9 +2871,8 @@ define i32
@icmp_slt_0_and_icmp_sge_neg2_i32_multiuse1(i32 %x) {
define i32 @icmp_slt_0_and_icmp_sge_neg2_i32_multiuse2(i32 %x) {
; CHECK-LABEL: @icmp_slt_0_and_icmp_sge_neg2_i32_multiuse2(
; CHECK-NEXT:[[A:%.*]] = icmp sgt i32 [[X:%.*]], -3
-; CHECK-NEXT:[[B:%.*]] = zext i1 [[A]] to i32
; CHECK-NEXT:[[C:%.*]] = lshr i32 [[X]], 31
-; CHECK-NEXT:[[D:%.*]] = and i32 [[C]], [[B]]
+; CHECK-NEXT:[[D:%.*]] = select i1 [[A]], i32 [[C]], i32 0
; CHECK-NEXT:call void @use32(i32 [[C]])
; CHECK-NEXT:ret i32 [[D]]
;
@@ -2923,10 +2922,9 @@ define i32
@icmp_slt_0_or_icmp_eq_100_i32_multiuse_fail1(i32 %x) {
define i32 @icmp_x_slt_0_and_icmp_y_ne_neg2_i32_multiuse_fail2(i32 %x, i32 %y)
{
; CHECK-LABEL: @icmp_x_slt_0_and_icmp_y_ne_neg2_i32_multiuse_fail2(
-; CHECK-NEXT:[[A:%.*]] = icmp ne i32 [[X:%.*]], -2
-; CHECK-NEXT:[[B:%.*]] = zext i1 [[A]] to i32
+; CHECK-NEXT:[[A_NOT:%.*]] = icmp eq i32 [[X:%.*]], -2
; CHECK-NEXT:[[C:%.*]] = lshr i32 [[Y:%.*]], 31
-; CHECK-NEXT:[[D:%.*]] = and i32 [[C]], [[B]]
+; CHECK-NEXT:[[D:%.*]] = select i1 [[A_NOT]], i32 0, i32 [[C]]
; CHECK-NEXT:call void @use32(i32 [[C]])
; CHECK-NEXT:ret i32 [[D]]
;
diff --git a/llvm/test/Transforms/InstCombine/icmp.ll
b/llvm/test/Transforms/InstCombine/icmp.ll
index 666d85d6cc32103..a5c05278f4232e8 100644
--- a/llvm/test/Transforms/InstCombine/icmp.ll
+++ b/llvm/test/Transforms/InstCombine/icmp.ll
@@ -4441,9 +4441,9 @@ define i1 @redundant_sign_bit_count_ugt_31_30(i32 %x) {
define i1 @zext_bool_and_eq0(i1 %x, i8 %y) {
; CHECK-LABEL: @zext_bool_and_eq0(
; CHECK-NEXT:[[TMP1:%.*]] = and i8 [[Y:%.*]], 1
-; CHECK-NEXT:[[TMP2:%.*]] = icmp ne i8 [[TMP1]], 0
-; CHECK-NEXT:[[TMP3:%.*]] = and i1 [[TMP2]], [[X:%.*]]
-; CHECK-NEXT:[[R:%.*]] = xor i1 [[TMP3]], true
+; CHECK-NEXT:[[R1:%.*]] = icmp eq i8 [[TMP1]], 0
+; CHECK-NEXT:[[NOT_X:%.*]] = xor i1 [[X:%.*]], true
+; CHECK-NEXT:[[R:%.*]] = select i1 [[NOT_X]], i1 true, i1 [[R1]]
; CHECK-NEXT:ret i1 [[R]]
;
%zx = zext i1 %x to i8
@@ -4454,9 +4454,10 @@ define i1 @zext_bool_and_eq0(i1 %x, i8 %y) {
define <2 x i1> @zext_bool_and_eq0_commute(<2 x i1> %x, <2 x i8> %p) {
; CHECK-LABEL: @zext_bool_and_eq0_commute(
-; CHECK-NEXT:[[TMP1:%.*]] = trunc <2 x i8> [[P:%.*]] to <2 x i1>
-; CHECK-NEXT:[[TMP2:%.*]] = and <2 x i1> [[TMP1]], [[X:%.*]]
-; CHECK-NEXT:[[R:%.*]] = xor <2 x i1> [[TMP2]],
+; CHECK-NEXT:
[clang] [InstCombine] Canonicalize `and(zext(A), B)` into `select A, B & 1, 0` (PR #66740)
https://github.com/dtcxzyw edited https://github.com/llvm/llvm-project/pull/66740 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Canonicalize `and(zext(A), B)` into `select A, B & 1, 0` (PR #66740)
dtcxzyw wrote: > We should use rebase instead of merge. And I recommend force push to change > the code before the review starts. I am confused with the LLVM GitHub user guide. I witnessed someone mess up by doing rebase + force push. Examples: #65853 #65543 #65535 https://github.com/llvm/llvm-project/pull/66740 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [ValueTracking] Simplify uaddo pattern (PR #65910)
dtcxzyw wrote: Ping. https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [ValueTracking] Simplify uaddo pattern (PR #65910)
dtcxzyw wrote: Ping. https://github.com/llvm/llvm-project/pull/65910 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 1/6] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32 %a)
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw updated
https://github.com/llvm/llvm-project/pull/65852
>From d9d8bcbb98e8f5aecb9733329389d61a489bd731 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng
Date: Sat, 9 Sep 2023 23:07:29 +0800
Subject: [PATCH 1/6] [InstCombine] Simplify the pattern `a ne/eq (zext (a
ne/eq c))`
---
.../InstCombine/InstCombineCompares.cpp | 62 ++
.../test/Transforms/InstCombine/icmp-range.ll | 181 ++
2 files changed, 243 insertions(+)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 9fdc46fec631679..837b8e6d2619989 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -6309,7 +6309,69 @@ Instruction
*InstCombinerImpl::foldICmpUsingBoolRange(ICmpInst ) {
Y->getType()->isIntOrIntVectorTy(1) && Pred == ICmpInst::ICMP_ULE)
return BinaryOperator::CreateOr(Builder.CreateIsNull(X), Y);
+ ICmpInst::Predicate Pred1, Pred2;
const APInt *C;
+ // icmp eq/ne X, (zext (icmp eq/ne X, C))
+ if (match(, m_c_ICmp(Pred1, m_Value(X),
+ m_ZExt(m_ICmp(Pred2, m_Deferred(X), m_APInt(C) &&
+ ICmpInst::isEquality(Pred1) && ICmpInst::isEquality(Pred2)) {
+if (C->isZero()) {
+ if (Pred2 == ICmpInst::ICMP_EQ) {
+// icmp eq X, (zext (icmp eq X, 0)) --> false
+// icmp ne X, (zext (icmp eq X, 0)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp ne X, 0)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp ne X, 0)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else if (C->isOne()) {
+ if (Pred2 == ICmpInst::ICMP_NE) {
+// icmp eq X, (zext (icmp ne X, 1)) --> false
+// icmp ne X, (zext (icmp ne X, 1)) --> true
+return replaceInstUsesWith(
+I,
+Constant::getIntegerValue(
+I.getType(),
+APInt(1U, static_cast(Pred1 == ICmpInst::ICMP_NE;
+ } else {
+// icmp eq X, (zext (icmp eq X, 1)) --> icmp ult X, 2
+// icmp ne X, (zext (icmp eq X, 1)) --> icmp ugt X, 1
+return ICmpInst::Create(
+Instruction::ICmp,
+Pred1 == ICmpInst::ICMP_NE ? ICmpInst::ICMP_UGT
+ : ICmpInst::ICMP_ULT,
+X,
+Constant::getIntegerValue(
+X->getType(), APInt(X->getType()->getScalarSizeInBits(),
+Pred1 == ICmpInst::ICMP_NE ? 1 : 2)));
+ }
+} else {
+ // C != 0 && C != 1
+ // icmp eq X, (zext (icmp eq X, C)) --> icmp eq X, 0
+ // icmp eq X, (zext (icmp ne X, C)) --> icmp eq X, 1
+ // icmp ne X, (zext (icmp eq X, C)) --> icmp ne X, 0
+ // icmp ne X, (zext (icmp ne X, C)) --> icmp ne X, 1
+ return ICmpInst::Create(
+ Instruction::ICmp, Pred1, X,
+ Constant::getIntegerValue(
+ X->getType(),
+ APInt(X->getType()->getScalarSizeInBits(),
+static_cast(Pred2 == ICmpInst::ICMP_NE;
+}
+ }
+
if (match(I.getOperand(0), m_c_Add(m_ZExt(m_Value(X)), m_SExt(m_Value(Y
&&
match(I.getOperand(1), m_APInt(C)) &&
X->getType()->isIntOrIntVectorTy(1) &&
diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll
b/llvm/test/Transforms/InstCombine/icmp-range.ll
index 4281e09cb0309c8..15424fce33fdeea 100644
--- a/llvm/test/Transforms/InstCombine/icmp-range.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1034,6 +1034,187 @@ define i1 @icmp_ne_bool_1(ptr %ptr) {
ret i1 %cmp
}
+define i1 @icmp_ne_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_eq_zero(
+; CHECK-NEXT:ret i1 true
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_ne_zext_ne_zero(i32 %a) {
+; CHECK-LABEL: @icmp_ne_zext_ne_zero(
+; CHECK-NEXT:[[CMP1:%.*]] = icmp ugt i32 [[A:%.*]], 1
+; CHECK-NEXT:ret i1 [[CMP1]]
+;
+ %cmp = icmp ne i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp ne i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_eq_zero(i32 %a) {
+; CHECK-LABEL: @icmp_eq_zext_eq_zero(
+; CHECK-NEXT:ret i1 false
+;
+ %cmp = icmp eq i32 %a, 0
+ %conv = zext i1 %cmp to i32
+ %cmp1 = icmp eq i32 %conv, %a
+ ret i1 %cmp1
+}
+
+define i1 @icmp_eq_zext_ne_zero(i32 %a)
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw resolved https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
@@ -366,19 +366,10 @@ define void @simplify_before_foldAndOfICmps(ptr %p) {
; CHECK-LABEL: @simplify_before_foldAndOfICmps(
; CHECK-NEXT:[[A8:%.*]] = alloca i16, align 2
; CHECK-NEXT:[[L7:%.*]] = load i16, ptr [[A8]], align 2
-; CHECK-NEXT:[[TMP1:%.*]] = icmp eq i16 [[L7]], -1
-; CHECK-NEXT:[[B11:%.*]] = zext i1 [[TMP1]] to i16
-; CHECK-NEXT:[[C10:%.*]] = icmp ugt i16 [[L7]], [[B11]]
-; CHECK-NEXT:[[C5:%.*]] = icmp slt i16 [[L7]], 1
-; CHECK-NEXT:[[C7:%.*]] = icmp slt i16 [[L7]], 0
-; CHECK-NEXT:[[B15:%.*]] = xor i1 [[C7]], [[C10]]
-; CHECK-NEXT:[[C6:%.*]] = xor i1 [[B15]], true
-; CHECK-NEXT:[[TMP2:%.*]] = and i1 [[C5]], [[C6]]
-; CHECK-NEXT:[[C3:%.*]] = and i1 [[TMP2]], [[C10]]
-; CHECK-NEXT:[[TMP3:%.*]] = xor i1 [[C10]], true
-; CHECK-NEXT:[[C18:%.*]] = or i1 [[C7]], [[TMP3]]
-; CHECK-NEXT:[[TMP4:%.*]] = sext i1 [[C3]] to i64
-; CHECK-NEXT:[[G26:%.*]] = getelementptr i1, ptr null, i64 [[TMP4]]
+; CHECK-NEXT:[[C18:%.*]] = icmp slt i16 [[L7]], 1
+; CHECK-NEXT:[[L7_LOBIT:%.*]] = ashr i16 [[L7]], 15
+; CHECK-NEXT:[[TMP1:%.*]] = sext i16 [[L7_LOBIT]] to i64
+; CHECK-NEXT:[[G26:%.*]] = getelementptr i1, ptr null, i64 [[TMP1]]
; CHECK-NEXT:store i16 [[L7]], ptr [[P:%.*]], align 2
dtcxzyw wrote:
Proof: https://alive2.llvm.org/ce/z/wBGCFz
https://github.com/llvm/llvm-project/pull/65852
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw edited https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [InstCombine] Simplify the pattern `a ne/eq (zext/sext (a ne/eq c))` (PR #65852)
https://github.com/dtcxzyw edited https://github.com/llvm/llvm-project/pull/65852 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
