[llvm] goldsteinn/non zero overflow (PR #87701)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Apr 4 13:37:12 PDT 2024
https://github.com/goldsteinn created https://github.com/llvm/llvm-project/pull/87701
- **[ValueTracking] Add tests for overflow detection functions is `isKnownNonZero`; NFC**
- **[ValueTracking] Split `isNonZero(mul)` logic to a helper; NFC**
- **[ValueTracking] Add support for overflow detection functions is `isKnownNonZero`**
>From e1afbd6785c916b5459012be9e0fd920eed57e5a Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 3 Apr 2024 21:36:35 -0500
Subject: [PATCH 1/3] [ValueTracking] Add tests for overflow detection
functions is `isKnownNonZero`; NFC
---
.../Transforms/InstSimplify/known-non-zero.ll | 307 +++++++++++++++++-
1 file changed, 302 insertions(+), 5 deletions(-)
diff --git a/llvm/test/Transforms/InstSimplify/known-non-zero.ll b/llvm/test/Transforms/InstSimplify/known-non-zero.ll
index b647f11af4461d..2c34d25373e0a5 100644
--- a/llvm/test/Transforms/InstSimplify/known-non-zero.ll
+++ b/llvm/test/Transforms/InstSimplify/known-non-zero.ll
@@ -99,11 +99,10 @@ exit:
ret i1 %res
}
-
; The code below exposed a bug similar to the one exposed by D60846, see the commit 6ea477590085.
; In a nutshell, we should not replace %result.0 with 0 here.
-define zeroext i8 @update_phi_query_loc_in_recursive_call(ptr nocapture readonly %p){
+define zeroext i8 @update_phi_query_loc_in_recursive_call(ptr nocapture readonly %p) {
; CHECK-LABEL: @update_phi_query_loc_in_recursive_call(
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[FOR_COND:%.*]]
@@ -126,16 +125,16 @@ define zeroext i8 @update_phi_query_loc_in_recursive_call(ptr nocapture readonly
entry:
br label %for.cond
-for.cond: ; preds = %for.body, %entry
+for.cond: ; preds = %for.body, %entry
%result.0 = phi i8 [ 0, %entry ], [ %conv2, %for.body ]
%shift.0 = phi i32 [ 0, %entry ], [ 1, %for.body ]
%cmp = icmp eq i32 %shift.0, 0
br i1 %cmp, label %for.body, label %for.cond.cleanup
-for.cond.cleanup: ; preds = %for.cond
+for.cond.cleanup: ; preds = %for.cond
ret i8 %result.0
-for.body: ; preds = %for.cond
+for.body: ; preds = %for.cond
%0 = load i8, ptr %p, align 1
%conv = zext i8 %0 to i32
%mul = shl nuw nsw i32 %shift.0, 3
@@ -166,3 +165,301 @@ A:
B:
ret i1 0
}
+
+declare void @use.i1(i1)
+declare void @use.i8(i1)
+define i1 @extract_value_uadd(i8 %xx, i8 %yy) {
+; CHECK-LABEL: @extract_value_uadd(
+; CHECK-NEXT: [[X:%.*]] = add nuw i8 [[XX:%.*]], 1
+; CHECK-NEXT: [[Y:%.*]] = add nuw i8 [[YY:%.*]], 1
+; CHECK-NEXT: [[X_LEMMA:%.*]] = icmp ult i8 [[X]], -128
+; CHECK-NEXT: [[Y_LEMMA:%.*]] = icmp ult i8 [[Y]], -128
+; CHECK-NEXT: call void @llvm.assume(i1 [[X_LEMMA]])
+; CHECK-NEXT: call void @llvm.assume(i1 [[Y_LEMMA]])
+; CHECK-NEXT: [[ADD_UOV:%.*]] = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 [[X]], i8 [[Y]])
+; CHECK-NEXT: [[ADD:%.*]] = extractvalue { i8, i1 } [[ADD_UOV]], 0
+; CHECK-NEXT: [[UOV:%.*]] = extractvalue { i8, i1 } [[ADD_UOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[UOV]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[ADD]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = add nuw i8 %xx, 1
+ %y = add nuw i8 %yy, 1
+ %x_lemma = icmp ult i8 %x, 128
+ %y_lemma = icmp ult i8 %y, 128
+ call void @llvm.assume(i1 %x_lemma)
+ call void @llvm.assume(i1 %y_lemma)
+
+ %add_uov = call { i8, i1} @llvm.uadd.with.overflow(i8 %x, i8 %y)
+ %add = extractvalue { i8, i1} %add_uov, 0
+ %uov = extractvalue { i8, i1} %add_uov, 1
+ call void @use.i1(i1 %uov)
+ %r = icmp eq i8 %add, 0
+ ret i1 %r
+}
+
+define i1 @extract_value_uadd_fail(i8 %xx, i8 %yy) {
+; CHECK-LABEL: @extract_value_uadd_fail(
+; CHECK-NEXT: [[X:%.*]] = add i8 [[XX:%.*]], 1
+; CHECK-NEXT: [[Y:%.*]] = add i8 [[YY:%.*]], 1
+; CHECK-NEXT: [[X_LEMMA:%.*]] = icmp ult i8 [[X]], -128
+; CHECK-NEXT: [[Y_LEMMA:%.*]] = icmp ult i8 [[Y]], -128
+; CHECK-NEXT: call void @llvm.assume(i1 [[X_LEMMA]])
+; CHECK-NEXT: call void @llvm.assume(i1 [[Y_LEMMA]])
+; CHECK-NEXT: [[ADD_UOV:%.*]] = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 [[X]], i8 [[Y]])
+; CHECK-NEXT: [[ADD:%.*]] = extractvalue { i8, i1 } [[ADD_UOV]], 0
+; CHECK-NEXT: [[UOV:%.*]] = extractvalue { i8, i1 } [[ADD_UOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[UOV]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[ADD]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = add i8 %xx, 1
+ %y = add i8 %yy, 1
+ %x_lemma = icmp ult i8 %x, 128
+ %y_lemma = icmp ult i8 %y, 128
+ call void @llvm.assume(i1 %x_lemma)
+ call void @llvm.assume(i1 %y_lemma)
+
+ %add_uov = call { i8, i1} @llvm.uadd.with.overflow(i8 %x, i8 %y)
+ %add = extractvalue { i8, i1} %add_uov, 0
+ %uov = extractvalue { i8, i1} %add_uov, 1
+ call void @use.i1(i1 %uov)
+ %r = icmp eq i8 %add, 0
+ ret i1 %r
+}
+
+define i1 @extract_value_sadd(i8 %xx, i8 %yy) {
+; CHECK-LABEL: @extract_value_sadd(
+; CHECK-NEXT: [[X:%.*]] = add nuw i8 [[XX:%.*]], 1
+; CHECK-NEXT: [[Y:%.*]] = add nuw i8 [[YY:%.*]], 1
+; CHECK-NEXT: [[X_LEMMA:%.*]] = icmp ult i8 [[X]], -128
+; CHECK-NEXT: [[Y_LEMMA:%.*]] = icmp ult i8 [[Y]], -128
+; CHECK-NEXT: call void @llvm.assume(i1 [[X_LEMMA]])
+; CHECK-NEXT: call void @llvm.assume(i1 [[Y_LEMMA]])
+; CHECK-NEXT: [[ADD_SOV:%.*]] = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X]], i8 [[Y]])
+; CHECK-NEXT: [[ADD:%.*]] = extractvalue { i8, i1 } [[ADD_SOV]], 0
+; CHECK-NEXT: [[SOV:%.*]] = extractvalue { i8, i1 } [[ADD_SOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[SOV]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[ADD]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = add nuw i8 %xx, 1
+ %y = add nuw i8 %yy, 1
+ %x_lemma = icmp ult i8 %x, 128
+ %y_lemma = icmp ult i8 %y, 128
+ call void @llvm.assume(i1 %x_lemma)
+ call void @llvm.assume(i1 %y_lemma)
+
+ %add_sov = call { i8, i1} @llvm.sadd.with.overflow(i8 %x, i8 %y)
+ %add = extractvalue { i8, i1} %add_sov, 0
+ %sov = extractvalue { i8, i1} %add_sov, 1
+ call void @use.i1(i1 %sov)
+ %r = icmp eq i8 %add, 0
+ ret i1 %r
+}
+
+define i1 @extract_value_sadd_fail(i8 %xx, i8 %yy) {
+; CHECK-LABEL: @extract_value_sadd_fail(
+; CHECK-NEXT: [[X:%.*]] = add i8 [[XX:%.*]], 1
+; CHECK-NEXT: [[Y:%.*]] = add i8 [[YY:%.*]], 1
+; CHECK-NEXT: [[ADD_SOV:%.*]] = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X]], i8 [[Y]])
+; CHECK-NEXT: [[ADD:%.*]] = extractvalue { i8, i1 } [[ADD_SOV]], 0
+; CHECK-NEXT: [[SOV:%.*]] = extractvalue { i8, i1 } [[ADD_SOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[SOV]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[ADD]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = add i8 %xx, 1
+ %y = add i8 %yy, 1
+
+ %add_sov = call { i8, i1} @llvm.sadd.with.overflow(i8 %x, i8 %y)
+ %add = extractvalue { i8, i1} %add_sov, 0
+ %sov = extractvalue { i8, i1} %add_sov, 1
+ call void @use.i1(i1 %sov)
+ %r = icmp eq i8 %add, 0
+ ret i1 %r
+}
+
+define i1 @extract_value_usub(i8 %x, i8 %zz) {
+; CHECK-LABEL: @extract_value_usub(
+; CHECK-NEXT: [[Z:%.*]] = add nuw i8 [[ZZ:%.*]], 1
+; CHECK-NEXT: [[Y:%.*]] = add i8 [[X:%.*]], [[Z]]
+; CHECK-NEXT: [[SUB_UOV:%.*]] = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 [[X]], i8 [[Y]])
+; CHECK-NEXT: [[SUB:%.*]] = extractvalue { i8, i1 } [[SUB_UOV]], 0
+; CHECK-NEXT: [[UOV:%.*]] = extractvalue { i8, i1 } [[SUB_UOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[UOV]])
+; CHECK-NEXT: call void @use.i8(i8 [[SUB]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[SUB]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %z = add nuw i8 %zz, 1
+ %y = add i8 %x, %z
+
+ %sub_uov = call { i8, i1} @llvm.usub.with.overflow(i8 %x, i8 %y)
+ %sub = extractvalue { i8, i1} %sub_uov, 0
+ %uov = extractvalue { i8, i1} %sub_uov, 1
+ call void @use.i1(i1 %uov)
+ call void @use.i8(i8 %sub)
+ %r = icmp eq i8 %sub, 0
+ ret i1 %r
+}
+
+define i1 @extract_value_usub_fail(i8 %x, i8 %z) {
+; CHECK-LABEL: @extract_value_usub_fail(
+; CHECK-NEXT: [[Y:%.*]] = add i8 [[X:%.*]], [[Z:%.*]]
+; CHECK-NEXT: [[SUB_UOV:%.*]] = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 [[X]], i8 [[Y]])
+; CHECK-NEXT: [[SUB:%.*]] = extractvalue { i8, i1 } [[SUB_UOV]], 0
+; CHECK-NEXT: [[UOV:%.*]] = extractvalue { i8, i1 } [[SUB_UOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[UOV]])
+; CHECK-NEXT: call void @use.i8(i8 [[SUB]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[SUB]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %y = add i8 %x, %z
+ %sub_uov = call { i8, i1} @llvm.usub.with.overflow(i8 %x, i8 %y)
+ %sub = extractvalue { i8, i1} %sub_uov, 0
+ %uov = extractvalue { i8, i1} %sub_uov, 1
+ call void @use.i1(i1 %uov)
+ call void @use.i8(i8 %sub)
+ %r = icmp eq i8 %sub, 0
+ ret i1 %r
+}
+
+define i1 @extract_value_ssub(i8 %x, i8 %zz) {
+; CHECK-LABEL: @extract_value_ssub(
+; CHECK-NEXT: [[Z:%.*]] = add nuw i8 [[ZZ:%.*]], 1
+; CHECK-NEXT: [[Y:%.*]] = add i8 [[X:%.*]], [[Z]]
+; CHECK-NEXT: [[SUB_SOV:%.*]] = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 [[Y]], i8 [[X]])
+; CHECK-NEXT: [[SUB:%.*]] = extractvalue { i8, i1 } [[SUB_SOV]], 0
+; CHECK-NEXT: [[SOV:%.*]] = extractvalue { i8, i1 } [[SUB_SOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[SOV]])
+; CHECK-NEXT: call void @use.i8(i8 [[SUB]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[SUB]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %z = add nuw i8 %zz, 1
+ %y = add i8 %x, %z
+
+ %sub_sov = call { i8, i1} @llvm.ssub.with.overflow(i8 %y, i8 %x)
+ %sub = extractvalue { i8, i1} %sub_sov, 0
+ %sov = extractvalue { i8, i1} %sub_sov, 1
+ call void @use.i1(i1 %sov)
+ call void @use.i8(i8 %sub)
+ %r = icmp eq i8 %sub, 0
+ ret i1 %r
+}
+
+define i1 @extract_value_ssub_fail(i8 %x) {
+; CHECK-LABEL: @extract_value_ssub_fail(
+; CHECK-NEXT: [[SUB_SOV:%.*]] = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 10, i8 [[X:%.*]])
+; CHECK-NEXT: [[SUB:%.*]] = extractvalue { i8, i1 } [[SUB_SOV]], 0
+; CHECK-NEXT: [[SOV:%.*]] = extractvalue { i8, i1 } [[SUB_SOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[SOV]])
+; CHECK-NEXT: call void @use.i8(i8 [[SUB]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[SUB]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %sub_sov = call { i8, i1} @llvm.ssub.with.overflow(i8 10, i8 %x)
+ %sub = extractvalue { i8, i1} %sub_sov, 0
+ %sov = extractvalue { i8, i1} %sub_sov, 1
+ call void @use.i1(i1 %sov)
+ call void @use.i8(i8 %sub)
+ %r = icmp eq i8 %sub, 0
+ ret i1 %r
+}
+
+define i1 @extract_value_umul(i8 %xx, i8 %yy) {
+; CHECK-LABEL: @extract_value_umul(
+; CHECK-NEXT: [[X:%.*]] = or i8 [[XX:%.*]], 1
+; CHECK-NEXT: [[Y:%.*]] = add nuw i8 [[YY:%.*]], 1
+; CHECK-NEXT: [[MUL_UOV:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 [[X]], i8 [[Y]])
+; CHECK-NEXT: [[MUL:%.*]] = extractvalue { i8, i1 } [[MUL_UOV]], 0
+; CHECK-NEXT: [[UOV:%.*]] = extractvalue { i8, i1 } [[MUL_UOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[UOV]])
+; CHECK-NEXT: call void @use.i8(i8 [[MUL]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[MUL]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = or i8 %xx, 1
+ %y = add nuw i8 %yy, 1
+
+ %mul_uov = call { i8, i1} @llvm.umul.with.overflow(i8 %x, i8 %y)
+ %mul = extractvalue { i8, i1} %mul_uov, 0
+ %uov = extractvalue { i8, i1} %mul_uov, 1
+ call void @use.i1(i1 %uov)
+ call void @use.i8(i8 %mul)
+ %r = icmp eq i8 %mul, 0
+ ret i1 %r
+}
+
+define i1 @extract_value_umul_fail(i8 %xx, i8 %yy) {
+; CHECK-LABEL: @extract_value_umul_fail(
+; CHECK-NEXT: [[X:%.*]] = or i8 [[XX:%.*]], 2
+; CHECK-NEXT: [[Y:%.*]] = add nuw i8 [[YY:%.*]], 1
+; CHECK-NEXT: [[MUL_UOV:%.*]] = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 [[X]], i8 [[Y]])
+; CHECK-NEXT: [[MUL:%.*]] = extractvalue { i8, i1 } [[MUL_UOV]], 0
+; CHECK-NEXT: [[UOV:%.*]] = extractvalue { i8, i1 } [[MUL_UOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[UOV]])
+; CHECK-NEXT: call void @use.i8(i8 [[MUL]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[MUL]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = or i8 %xx, 2
+ %y = add nuw i8 %yy, 1
+
+ %mul_uov = call { i8, i1} @llvm.umul.with.overflow(i8 %x, i8 %y)
+ %mul = extractvalue { i8, i1} %mul_uov, 0
+ %uov = extractvalue { i8, i1} %mul_uov, 1
+ call void @use.i1(i1 %uov)
+ call void @use.i8(i8 %mul)
+ %r = icmp eq i8 %mul, 0
+ ret i1 %r
+}
+
+define i1 @extract_value_smul(i8 %xx, i8 %yy) {
+; CHECK-LABEL: @extract_value_smul(
+; CHECK-NEXT: [[X:%.*]] = or i8 [[XX:%.*]], 1
+; CHECK-NEXT: [[Y:%.*]] = add nuw i8 [[YY:%.*]], 1
+; CHECK-NEXT: [[MUL_SOV:%.*]] = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 [[Y]], i8 [[X]])
+; CHECK-NEXT: [[MUL:%.*]] = extractvalue { i8, i1 } [[MUL_SOV]], 0
+; CHECK-NEXT: [[SOV:%.*]] = extractvalue { i8, i1 } [[MUL_SOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[SOV]])
+; CHECK-NEXT: call void @use.i8(i8 [[MUL]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[MUL]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = or i8 %xx, 1
+ %y = add nuw i8 %yy, 1
+
+ %mul_sov = call { i8, i1} @llvm.smul.with.overflow(i8 %y, i8 %x)
+ %mul = extractvalue { i8, i1} %mul_sov, 0
+ %sov = extractvalue { i8, i1} %mul_sov, 1
+ call void @use.i1(i1 %sov)
+ call void @use.i8(i8 %mul)
+ %r = icmp eq i8 %mul, 0
+ ret i1 %r
+}
+
+define i1 @extract_value_smul_fail(i8 %xx, i8 %yy) {
+; CHECK-LABEL: @extract_value_smul_fail(
+; CHECK-NEXT: [[X:%.*]] = or i8 [[XX:%.*]], 1
+; CHECK-NEXT: [[Y:%.*]] = add i8 [[YY:%.*]], 1
+; CHECK-NEXT: [[MUL_SOV:%.*]] = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 [[Y]], i8 [[X]])
+; CHECK-NEXT: [[MUL:%.*]] = extractvalue { i8, i1 } [[MUL_SOV]], 0
+; CHECK-NEXT: [[SOV:%.*]] = extractvalue { i8, i1 } [[MUL_SOV]], 1
+; CHECK-NEXT: call void @use.i1(i1 [[SOV]])
+; CHECK-NEXT: call void @use.i8(i8 [[MUL]])
+; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[MUL]], 0
+; CHECK-NEXT: ret i1 [[R]]
+;
+ %x = or i8 %xx, 1
+ %y = add i8 %yy, 1
+
+ %mul_sov = call { i8, i1} @llvm.smul.with.overflow(i8 %y, i8 %x)
+ %mul = extractvalue { i8, i1} %mul_sov, 0
+ %sov = extractvalue { i8, i1} %mul_sov, 1
+ call void @use.i1(i1 %sov)
+ call void @use.i8(i8 %mul)
+ %r = icmp eq i8 %mul, 0
+ ret i1 %r
+}
>From 08d3ac3145bf6c9a4860d6592005f9e6bcfd5f88 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 3 Apr 2024 15:33:18 -0500
Subject: [PATCH 2/3] [ValueTracking] Split `isNonZero(mul)` logic to a helper;
NFC
---
llvm/lib/Analysis/ValueTracking.cpp | 57 ++++++++++++++++-------------
1 file changed, 31 insertions(+), 26 deletions(-)
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 5ad4da43bca7db..61dfe775b2d4b0 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -2464,6 +2464,34 @@ static bool isNonZeroSub(const APInt &DemandedElts, unsigned Depth,
return ::isKnownNonEqual(X, Y, Depth, Q);
}
+static bool isNonZeroMul(const APInt &DemandedElts, unsigned Depth,
+ const SimplifyQuery &Q, unsigned BitWidth, Value *X,
+ Value *Y, bool NSW, bool NUW) {
+ // If X and Y are non-zero then so is X * Y as long as the multiplication
+ // does not overflow.
+ if (NSW || NUW)
+ return isKnownNonZero(X, DemandedElts, Depth, Q) &&
+ isKnownNonZero(Y, DemandedElts, Depth, Q);
+
+ // If either X or Y is odd, then if the other is non-zero the result can't
+ // be zero.
+ KnownBits XKnown = computeKnownBits(X, DemandedElts, Depth, Q);
+ if (XKnown.One[0])
+ return isKnownNonZero(Y, DemandedElts, Depth, Q);
+
+ KnownBits YKnown = computeKnownBits(Y, DemandedElts, Depth, Q);
+ if (YKnown.One[0])
+ return XKnown.isNonZero() || isKnownNonZero(X, DemandedElts, Depth, Q);
+
+ // If there exists any subset of X (sX) and subset of Y (sY) s.t sX * sY is
+ // non-zero, then X * Y is non-zero. We can find sX and sY by just taking
+ // the lowest known One of X and Y. If they are non-zero, the result
+ // must be non-zero. We can check if LSB(X) * LSB(Y) != 0 by doing
+ // X.CountLeadingZeros + Y.CountLeadingZeros < BitWidth.
+ return (XKnown.countMaxTrailingZeros() + YKnown.countMaxTrailingZeros()) <
+ BitWidth;
+}
+
static bool isNonZeroShift(const Operator *I, const APInt &DemandedElts,
unsigned Depth, const SimplifyQuery &Q,
const KnownBits &KnownVal) {
@@ -2659,33 +2687,10 @@ static bool isKnownNonZeroFromOperator(const Operator *I,
Q.IIQ.hasNoUnsignedWrap(BO));
}
case Instruction::Mul: {
- // If X and Y are non-zero then so is X * Y as long as the multiplication
- // does not overflow.
const OverflowingBinaryOperator *BO = cast<OverflowingBinaryOperator>(I);
- if (Q.IIQ.hasNoSignedWrap(BO) || Q.IIQ.hasNoUnsignedWrap(BO))
- return isKnownNonZero(I->getOperand(0), DemandedElts, Depth, Q) &&
- isKnownNonZero(I->getOperand(1), DemandedElts, Depth, Q);
-
- // If either X or Y is odd, then if the other is non-zero the result can't
- // be zero.
- KnownBits XKnown =
- computeKnownBits(I->getOperand(0), DemandedElts, Depth, Q);
- if (XKnown.One[0])
- return isKnownNonZero(I->getOperand(1), DemandedElts, Depth, Q);
-
- KnownBits YKnown =
- computeKnownBits(I->getOperand(1), DemandedElts, Depth, Q);
- if (YKnown.One[0])
- return XKnown.isNonZero() ||
- isKnownNonZero(I->getOperand(0), DemandedElts, Depth, Q);
-
- // If there exists any subset of X (sX) and subset of Y (sY) s.t sX * sY is
- // non-zero, then X * Y is non-zero. We can find sX and sY by just taking
- // the lowest known One of X and Y. If they are non-zero, the result
- // must be non-zero. We can check if LSB(X) * LSB(Y) != 0 by doing
- // X.CountLeadingZeros + Y.CountLeadingZeros < BitWidth.
- return (XKnown.countMaxTrailingZeros() + YKnown.countMaxTrailingZeros()) <
- BitWidth;
+ return isNonZeroMul(DemandedElts, Depth, Q, BitWidth, I->getOperand(0),
+ I->getOperand(1), Q.IIQ.hasNoSignedWrap(BO),
+ Q.IIQ.hasNoUnsignedWrap(BO));
}
case Instruction::Select: {
// (C ? X : Y) != 0 if X != 0 and Y != 0.
>From 90ef6b5eda287101cdec62f9ba4845c6fb48dfd8 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Wed, 3 Apr 2024 15:35:17 -0500
Subject: [PATCH 3/3] [ValueTracking] Add support for overflow detection
functions is `isKnownNonZero`
Adds support for: `{s,u}{add,sub,mul}.with.overflow`
The logic is identical to the the non-overflow binops, we where just
missing the cases.
---
llvm/lib/Analysis/ValueTracking.cpp | 29 +++++++++++++++++++
.../Transforms/InstSimplify/known-non-zero.ll | 20 ++++---------
2 files changed, 35 insertions(+), 14 deletions(-)
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 61dfe775b2d4b0..954a9052e2dce0 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -2791,6 +2791,35 @@ static bool isKnownNonZeroFromOperator(const Operator *I,
// handled in isKnownNonZero.
return false;
}
+ case Instruction::ExtractValue:
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I->getOperand(0))) {
+ const ExtractValueInst *EVI = cast<ExtractValueInst>(I);
+ if (EVI->getNumIndices() != 1 || EVI->getIndices()[0] != 0)
+ break;
+ switch (II->getIntrinsicID()) {
+ default:
+ break;
+ case Intrinsic::uadd_with_overflow:
+ case Intrinsic::sadd_with_overflow:
+ return isNonZeroAdd(APInt::getAllOnes(DemandedElts.getBitWidth()),
+ Depth, Q, BitWidth, II->getArgOperand(0),
+ II->getArgOperand(1),
+ /*NSW=*/false,
+ /*NUW=*/false);
+ case Intrinsic::usub_with_overflow:
+ case Intrinsic::ssub_with_overflow:
+ return isNonZeroSub(APInt::getAllOnes(DemandedElts.getBitWidth()),
+ Depth, Q, BitWidth, II->getArgOperand(0),
+ II->getArgOperand(1));
+ case Intrinsic::umul_with_overflow:
+ case Intrinsic::smul_with_overflow:
+ return isNonZeroMul(APInt::getAllOnes(DemandedElts.getBitWidth()),
+ Depth, Q, BitWidth, II->getArgOperand(0),
+ II->getArgOperand(1), /*NSW=*/false, /*NUW=*/false);
+ break;
+ }
+ }
+ break;
case Instruction::Call:
case Instruction::Invoke: {
const auto *Call = cast<CallBase>(I);
diff --git a/llvm/test/Transforms/InstSimplify/known-non-zero.ll b/llvm/test/Transforms/InstSimplify/known-non-zero.ll
index 2c34d25373e0a5..c34920ce25b6fc 100644
--- a/llvm/test/Transforms/InstSimplify/known-non-zero.ll
+++ b/llvm/test/Transforms/InstSimplify/known-non-zero.ll
@@ -177,11 +177,9 @@ define i1 @extract_value_uadd(i8 %xx, i8 %yy) {
; CHECK-NEXT: call void @llvm.assume(i1 [[X_LEMMA]])
; CHECK-NEXT: call void @llvm.assume(i1 [[Y_LEMMA]])
; CHECK-NEXT: [[ADD_UOV:%.*]] = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 [[X]], i8 [[Y]])
-; CHECK-NEXT: [[ADD:%.*]] = extractvalue { i8, i1 } [[ADD_UOV]], 0
; CHECK-NEXT: [[UOV:%.*]] = extractvalue { i8, i1 } [[ADD_UOV]], 1
; CHECK-NEXT: call void @use.i1(i1 [[UOV]])
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[ADD]], 0
-; CHECK-NEXT: ret i1 [[R]]
+; CHECK-NEXT: ret i1 false
;
%x = add nuw i8 %xx, 1
%y = add nuw i8 %yy, 1
@@ -237,11 +235,9 @@ define i1 @extract_value_sadd(i8 %xx, i8 %yy) {
; CHECK-NEXT: call void @llvm.assume(i1 [[X_LEMMA]])
; CHECK-NEXT: call void @llvm.assume(i1 [[Y_LEMMA]])
; CHECK-NEXT: [[ADD_SOV:%.*]] = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 [[X]], i8 [[Y]])
-; CHECK-NEXT: [[ADD:%.*]] = extractvalue { i8, i1 } [[ADD_SOV]], 0
; CHECK-NEXT: [[SOV:%.*]] = extractvalue { i8, i1 } [[ADD_SOV]], 1
; CHECK-NEXT: call void @use.i1(i1 [[SOV]])
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[ADD]], 0
-; CHECK-NEXT: ret i1 [[R]]
+; CHECK-NEXT: ret i1 false
;
%x = add nuw i8 %xx, 1
%y = add nuw i8 %yy, 1
@@ -289,8 +285,7 @@ define i1 @extract_value_usub(i8 %x, i8 %zz) {
; CHECK-NEXT: [[UOV:%.*]] = extractvalue { i8, i1 } [[SUB_UOV]], 1
; CHECK-NEXT: call void @use.i1(i1 [[UOV]])
; CHECK-NEXT: call void @use.i8(i8 [[SUB]])
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[SUB]], 0
-; CHECK-NEXT: ret i1 [[R]]
+; CHECK-NEXT: ret i1 false
;
%z = add nuw i8 %zz, 1
%y = add i8 %x, %z
@@ -334,8 +329,7 @@ define i1 @extract_value_ssub(i8 %x, i8 %zz) {
; CHECK-NEXT: [[SOV:%.*]] = extractvalue { i8, i1 } [[SUB_SOV]], 1
; CHECK-NEXT: call void @use.i1(i1 [[SOV]])
; CHECK-NEXT: call void @use.i8(i8 [[SUB]])
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[SUB]], 0
-; CHECK-NEXT: ret i1 [[R]]
+; CHECK-NEXT: ret i1 false
;
%z = add nuw i8 %zz, 1
%y = add i8 %x, %z
@@ -377,8 +371,7 @@ define i1 @extract_value_umul(i8 %xx, i8 %yy) {
; CHECK-NEXT: [[UOV:%.*]] = extractvalue { i8, i1 } [[MUL_UOV]], 1
; CHECK-NEXT: call void @use.i1(i1 [[UOV]])
; CHECK-NEXT: call void @use.i8(i8 [[MUL]])
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[MUL]], 0
-; CHECK-NEXT: ret i1 [[R]]
+; CHECK-NEXT: ret i1 false
;
%x = or i8 %xx, 1
%y = add nuw i8 %yy, 1
@@ -425,8 +418,7 @@ define i1 @extract_value_smul(i8 %xx, i8 %yy) {
; CHECK-NEXT: [[SOV:%.*]] = extractvalue { i8, i1 } [[MUL_SOV]], 1
; CHECK-NEXT: call void @use.i1(i1 [[SOV]])
; CHECK-NEXT: call void @use.i8(i8 [[MUL]])
-; CHECK-NEXT: [[R:%.*]] = icmp eq i8 [[MUL]], 0
-; CHECK-NEXT: ret i1 [[R]]
+; CHECK-NEXT: ret i1 false
;
%x = or i8 %xx, 1
%y = add nuw i8 %yy, 1
More information about the llvm-commits
mailing list