From: Nikita Popov <nikita.ppv@gmail.com>
Date: Fri, 1 Mar 2019 20:07:04 +0000 (+0000)
Subject: [ValueTracking] Known bits support for unsigned saturating add/sub
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=141f3969d781c08e4e0c7b026d47c4ceda0b246b;p=llvm

[ValueTracking] Known bits support for unsigned saturating add/sub

We have two sources of known bits:

1. For adds leading ones of either operand are preserved. For sub
leading zeros of LHS and leading ones of RHS become leading zeros in
the result.

2. The saturating math is a select between add/sub and an all-ones/
zero value. As such we can carry out the add/sub known bits
calculation, and only preseve the known one/zero bits respectively.

Differential Revision: https://reviews.llvm.org/D58329

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355223 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Analysis/ValueTracking.cpp b/lib/Analysis/ValueTracking.cpp
index 8b3d16a5125..b4473f857e2 100644
--- a/lib/Analysis/ValueTracking.cpp
+++ b/lib/Analysis/ValueTracking.cpp
@@ -1523,6 +1523,37 @@ static void computeKnownBitsFromOperator(const Operator *I, KnownBits &Known,
             Known2.One.shl(ShiftAmt) | Known3.One.lshr(BitWidth - ShiftAmt);
         break;
       }
+      case Intrinsic::uadd_sat:
+      case Intrinsic::usub_sat: {
+        bool IsAdd = II->getIntrinsicID() == Intrinsic::uadd_sat;
+        computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
+        computeKnownBits(I->getOperand(1), Known2, Depth + 1, Q);
+
+        // Add: Leading ones of either operand are preserved.
+        // Sub: Leading zeros of LHS and leading ones of RHS are preserved
+        // as leading zeros in the result.
+        unsigned LeadingKnown;
+        if (IsAdd)
+          LeadingKnown = std::max(Known.countMinLeadingOnes(),
+                                  Known2.countMinLeadingOnes());
+        else
+          LeadingKnown = std::max(Known.countMinLeadingZeros(),
+                                  Known2.countMinLeadingOnes());
+
+        Known = KnownBits::computeForAddSub(
+            IsAdd, /* NSW */ false, Known, Known2);
+
+        // We select between the operation result and all-ones/zero
+        // respectively, so we can preserve known ones/zeros.
+        if (IsAdd) {
+          Known.One.setHighBits(LeadingKnown);
+          Known.Zero.clearAllBits();
+        } else {
+          Known.Zero.setHighBits(LeadingKnown);
+          Known.One.clearAllBits();
+        }
+        break;
+      }
       case Intrinsic::x86_sse42_crc32_64_64:
         Known.Zero.setBitsFrom(32);
         break;
diff --git a/unittests/Analysis/ValueTrackingTest.cpp b/unittests/Analysis/ValueTrackingTest.cpp
index a0f5aed2b6d..df32287dbff 100644
--- a/unittests/Analysis/ValueTrackingTest.cpp
+++ b/unittests/Analysis/ValueTrackingTest.cpp
@@ -615,3 +615,74 @@ TEST_F(ComputeKnownBitsTest, ComputeKnownFshlZero) {
       "declare i16 @llvm.fshl.i16(i16, i16, i16)\n");
   expectKnownBits(/*zero*/ 15u, /*one*/ 3840u);
 }
+
+TEST_F(ComputeKnownBitsTest, ComputeKnownUAddSatLeadingOnes) {
+  // uadd.sat(1111...1, ........)
+  // = 1111....
+  parseAssembly(
+      "define i8 @test(i8 %a, i8 %b) {\n"
+      "  %aa = or i8 %a, 241\n"
+      "  %A = call i8 @llvm.uadd.sat.i8(i8 %aa, i8 %b)\n"
+      "  ret i8 %A\n"
+      "}\n"
+      "declare i8 @llvm.uadd.sat.i8(i8, i8)\n");
+  expectKnownBits(/*zero*/ 0u, /*one*/ 240u);
+}
+
+TEST_F(ComputeKnownBitsTest, ComputeKnownUAddSatOnesPreserved) {
+  // uadd.sat(00...011, .1...110)
+  // = .......1
+  parseAssembly(
+      "define i8 @test(i8 %a, i8 %b) {\n"
+      "  %aa = or i8 %a, 3\n"
+      "  %aaa = and i8 %aa, 59\n"
+      "  %bb = or i8 %b, 70\n"
+      "  %bbb = and i8 %bb, 254\n"
+      "  %A = call i8 @llvm.uadd.sat.i8(i8 %aaa, i8 %bbb)\n"
+      "  ret i8 %A\n"
+      "}\n"
+      "declare i8 @llvm.uadd.sat.i8(i8, i8)\n");
+  expectKnownBits(/*zero*/ 0u, /*one*/ 1u);
+}
+
+TEST_F(ComputeKnownBitsTest, ComputeKnownUSubSatLHSLeadingZeros) {
+  // usub.sat(0000...0, ........)
+  // = 0000....
+  parseAssembly(
+      "define i8 @test(i8 %a, i8 %b) {\n"
+      "  %aa = and i8 %a, 14\n"
+      "  %A = call i8 @llvm.usub.sat.i8(i8 %aa, i8 %b)\n"
+      "  ret i8 %A\n"
+      "}\n"
+      "declare i8 @llvm.usub.sat.i8(i8, i8)\n");
+  expectKnownBits(/*zero*/ 240u, /*one*/ 0u);
+}
+
+TEST_F(ComputeKnownBitsTest, ComputeKnownUSubSatRHSLeadingOnes) {
+  // usub.sat(........, 1111...1)
+  // = 0000....
+  parseAssembly(
+      "define i8 @test(i8 %a, i8 %b) {\n"
+      "  %bb = or i8 %a, 241\n"
+      "  %A = call i8 @llvm.usub.sat.i8(i8 %a, i8 %bb)\n"
+      "  ret i8 %A\n"
+      "}\n"
+      "declare i8 @llvm.usub.sat.i8(i8, i8)\n");
+  expectKnownBits(/*zero*/ 240u, /*one*/ 0u);
+}
+
+TEST_F(ComputeKnownBitsTest, ComputeKnownUSubSatZerosPreserved) {
+  // usub.sat(11...011, .1...110)
+  // = ......0.
+  parseAssembly(
+      "define i8 @test(i8 %a, i8 %b) {\n"
+      "  %aa = or i8 %a, 195\n"
+      "  %aaa = and i8 %aa, 251\n"
+      "  %bb = or i8 %b, 70\n"
+      "  %bbb = and i8 %bb, 254\n"
+      "  %A = call i8 @llvm.usub.sat.i8(i8 %aaa, i8 %bbb)\n"
+      "  ret i8 %A\n"
+      "}\n"
+      "declare i8 @llvm.usub.sat.i8(i8, i8)\n");
+  expectKnownBits(/*zero*/ 2u, /*one*/ 0u);
+}