Extend known bits to understand @llvm.bswap

[reames]

Summary

This is a cleaned up patch from the one written by John Regehr based on the findings of the Souper superoptimizer.

When writing tests, I was surprised to find that instsimplify apparently doesn't know how to collapse bit test sequences based purely on known bits. This required me to split my tests across both instsimplify and instcombine.

[spatel]

test/Transforms/InstCombine/bswap-known-bits.ll

3 ↗

(On Diff #35983)

Interesting...I didn't consider -instsimplify for the similar tests in D13222. I will look again.

Typo: "apparently"

28 ↗

(On Diff #35983)

Do we need this test? Seems like test4 is a superset.
Use different bswap sizes for more test coverage?

llvm/trunk/lib/Analysis/ValueTracking.cpp

case Instruction::Invoke:
     if (MDNode *MD = cast<Instruction>(I)->getMetadata(LLVMContext::MD_range))
       computeKnownBitsFromRangeMetadata(*MD, KnownZero);
     // If a range metadata is attached to this IntrinsicInst, intersect the
     // explicit range specified by the metadata and the implicit range of
     // the intrinsic.
     if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
       switch (II->getIntrinsicID()) {
       default: break;
+      case Intrinsic::bswap:
+        computeKnownBits(I->getOperand(0), KnownZero2, KnownOne2, DL,
+                         Depth + 1, Q);
+        KnownZero |= KnownZero2.byteSwap();
+        KnownOne |= KnownOne2.byteSwap();
+        break;
       case Intrinsic::ctlz:
       case Intrinsic::cttz: {
         unsigned LowBits = Log2_32(BitWidth)+1;
         // If this call is undefined for 0, the result will be less than 2^n.
         if (II->getArgOperand(1) == ConstantInt::getTrue(II->getContext()))
           LowBits -= 1;
         KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - LowBits);
         break;

llvm/trunk/test/Transforms/InstCombine/bswap-known-bits.ll

+; RUN: opt < %s -S -instcombine | FileCheck %s
+; Note: This is testing functionality in computeKnownBits.  I'd have rather
+; used instsimplify, but the bit test folding is apparently only in instcombine.
+
+declare i16 @llvm.bswap.i16(i16)
+declare i32 @llvm.bswap.i32(i32)
+
+define i1 @test1(i16 %arg) {
+; CHECK-LABEL: @test1
+; CHECK: ret i1 true
+  %a = or i16 %arg, 511
+  %b = call i16 @llvm.bswap.i16(i16 %a)
+  %and = and i16 %b, 256
+  %res = icmp eq i16 %and, 256
+  ret i1 %res
+}
+
+define i1 @test2(i16 %arg) {
+; CHECK-LABEL: @test2
+; CHECK: ret i1 true
+  %a = or i16 %arg, 1
+  %b = call i16 @llvm.bswap.i16(i16 %a)
+  %and = and i16 %b, 256
+  %res = icmp eq i16 %and, 256
+  ret i1 %res
+}
+
+
+define i1 @test3(i16 %arg) {
+; CHECK-LABEL: @test3
+; CHECK: ret i1 true
+  %a = or i16 %arg, 256
+  %b = call i16 @llvm.bswap.i16(i16 %a)
+  %and = and i16 %b, 1
+  %res = icmp eq i16 %and, 1
+  ret i1 %res
+}
+
+define i1 @test4(i32 %arg) {
+; CHECK-LABEL: @test4
+; CHECK: ret i1 true
+  %a = or i32 %arg, 2147483647  ; i32_MAX
+  %b = call i32 @llvm.bswap.i32(i32 %a)
+  %and = and i32 %b, 127
+  %res = icmp eq i32 %and, 127
+  ret i1 %res
+}

llvm/trunk/test/Transforms/InstSimplify/bswap.ll

+; RUN: opt < %s -S -instsimplify | FileCheck %s
+
+declare i16 @llvm.bswap.i16(i16)
+
+define i1 @test1(i16 %arg) {
+; CHECK-LABEL: @test1
+; CHECK: ret i1 false
+  %a = or i16 %arg, 1
+  %b = call i16 @llvm.bswap.i16(i16 %a)
+  %res = icmp eq i16 %b, 0
+  ret i1 %res
+}
+
+define i1 @test2(i16 %arg) {
+; CHECK-LABEL: @test2
+; CHECK: ret i1 false
+  %a = or i16 %arg, 1024
+  %b = call i16 @llvm.bswap.i16(i16 %a)
+  %res = icmp eq i16 %b, 0
+  ret i1 %res
+}
+
+define i1 @test3(i16 %arg) {
+; CHECK-LABEL: @test3
+; CHECK: ret i1 false
+  %a = and i16 %arg, 1
+  %b = call i16 @llvm.bswap.i16(i16 %a)
+  %and = and i16 %b, 1
+  %res = icmp eq i16 %and, 1
+  ret i1 %res
+}
+
+define i1 @test4(i16 %arg) {
+; CHECK-LABEL: @test4
+; CHECK: ret i1 false
+  %a = and i16 %arg, 511
+  %b = call i16 @llvm.bswap.i16(i16 %a)
+  %and = and i16 %b, 256
+  %res = icmp eq i16 %and, 1
+  ret i1 %res
+}