[ValueTracking] Improve ComputeNumSignBits for SRem.

[craig.topper]

Summary

The result will have the same sign as the dividend unless the
result is 0. The magnitude of the result will always be less
than or equal to the dividend. So the result will have at least
as many sign bits as the dividend.

Previously we would do this if the divisor was a positive constant,
but that isn't required.

[craig.topper]

Rebase

[RKSimon]

LGTM

llvm/lib/Analysis/ValueTracking.cpp

 //===- ValueTracking.cpp - Walk computations to compute properties --------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
case Instruction::SDiv: {
 
         // Add floor(log(C)) bits to the numerator bits.
         return std::min(TyBits, NumBits + Denominator->logBase2());
       }
       break;
     }
 
     case Instruction::SRem: {
+      Tmp = ComputeNumSignBits(U->getOperand(0), Depth + 1, Q);
+
       const APInt *Denominator;
       // srem X, C -> we know that the result is within [-C+1,C) when C is a
       // positive constant.  This let us put a lower bound on the number of sign
       // bits.
       if (match(U->getOperand(1), m_APInt(Denominator))) {
 
         // Ignore non-positive denominator.
-        if (!Denominator->isStrictlyPositive())
-          break;
-
-        // Calculate the incoming numerator bits. SRem by a positive constant
-        // can't lower the number of sign bits.
-        unsigned NumrBits = ComputeNumSignBits(U->getOperand(0), Depth + 1, Q);
-
+        if (Denominator->isStrictlyPositive()) {
           // Calculate the leading sign bit constraints by examining the
           // denominator.  Given that the denominator is positive, there are two
           // cases:
           //
-        //  1. the numerator is positive. The result range is [0,C) and [0,C) u<
-        //     (1 << ceilLogBase2(C)).
+          //  1. The numerator is positive. The result range is [0,C) and
+          //     [0,C) u< (1 << ceilLogBase2(C)).
           //
-        //  2. the numerator is negative. Then the result range is (-C,0] and
+          //  2. The numerator is negative. Then the result range is (-C,0] and
           //     integers in (-C,0] are either 0 or >u (-1 << ceilLogBase2(C)).
           //
           // Thus a lower bound on the number of sign bits is `TyBits -
           // ceilLogBase2(C)`.
 
           unsigned ResBits = TyBits - Denominator->ceilLogBase2();
-        return std::max(NumrBits, ResBits);
+          Tmp = std::max(Tmp, ResBits);
         }
-      break;
+      }
+      return Tmp;
     }
 
     case Instruction::AShr: {
       Tmp = ComputeNumSignBits(U->getOperand(0), Depth + 1, Q);
       // ashr X, C   -> adds C sign bits.  Vectors too.
       const APInt *ShAmt;
       if (match(U->getOperand(1), m_APInt(ShAmt))) {
         if (ShAmt->uge(TyBits))

llvm/test/Transforms/InstCombine/with_overflow.ll

;
   ; and the result greater than 32.
   %t = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 %rem, i32 %rem)
   %obit = extractvalue { i32, i1 } %t, 1
   ret i1 %obit
 }
 
 define i1 @overflow_mod_mul2(i16 %v1, i32 %v2) nounwind {
 ; CHECK-LABEL: @overflow_mod_mul2(
-; CHECK-NEXT:    [[A:%.*]] = sext i16 [[V1:%.*]] to i32
-; CHECK-NEXT:    [[REM:%.*]] = srem i32 [[A]], [[V2:%.*]]
-; CHECK-NEXT:    [[T:%.*]] = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 [[REM]], i32 [[REM]])
-; CHECK-NEXT:    [[OBIT:%.*]] = extractvalue { i32, i1 } [[T]], 1
-; CHECK-NEXT:    ret i1 [[OBIT]]
+; CHECK-NEXT:    ret i1 false
 ;
   %a = sext i16 %v1 to i32
   %rem = srem i32 %a, %v2
   %t = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 %rem, i32 %rem)
   %obit = extractvalue { i32, i1 } %t, 1
   ret i1 %obit
 }