Remove ARM isel hacks that fold large immediates into a pair of add, sub, and,
and xor. The 32-bit move immediates can be hoisted out of loops by machine
LICM but the isel hacks were preventing them.
Instead, let peephole optimization pass recognize registers that are defined by
immediates and the ARM target hook will fold the immediates in.
Other changes include 1) do not fold and / xor into cmp to isel TST / TEQ
instructions if there are multiple uses. This happens when the 'and' is live
out, machine sink would have sinked the computation and that ends up pessimizing
code. The peephole pass would recognize situations where the 'and' can be
toggled to define CPSR and eliminate the comparison anyway.
2) Move peephole pass to after machine LICM, sink, and CSE to avoid blocking
important optimizations.
rdar://8663787, rdar://8241368
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@119548 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/test/CodeGen/Thumb2/machine-licm.ll b/test/CodeGen/Thumb2/machine-licm.ll
index 14d04a4..37a15ff 100644
--- a/test/CodeGen/Thumb2/machine-licm.ll
+++ b/test/CodeGen/Thumb2/machine-licm.ll
@@ -77,10 +77,49 @@
}
; CHECK-NOT: LCPI1_0:
-; CHECK: .subsections_via_symbols
declare <4 x float> @llvm.arm.neon.vld1.v4f32(i8*, i32) nounwind readonly
declare void @llvm.arm.neon.vst1.v4f32(i8*, <4 x float>, i32) nounwind
declare <4 x float> @llvm.arm.neon.vmaxs.v4f32(<4 x float>, <4 x float>) nounwind readnone
+
+; rdar://8241368
+; isel should not fold immediate into eor's which would have prevented LICM.
+define zeroext i16 @t3(i8 zeroext %data, i16 zeroext %crc) nounwind readnone {
+; CHECK: t3:
+bb.nph:
+; CHECK: bb.nph
+; CHECK: movw {{(r[0-9])|(lr)}}, #32768
+; CHECK: movs {{(r[0-9])|(lr)}}, #8
+; CHECK: movw [[REGISTER:(r[0-9])|(lr)]], #16386
+; CHECK: movw {{(r[0-9])|(lr)}}, #65534
+; CHECK: movt {{(r[0-9])|(lr)}}, #65535
+ br label %bb
+
+bb: ; preds = %bb, %bb.nph
+; CHECK: bb
+; CHECK: eor.w {{(r[0-9])|(lr)}}, {{(r[0-9])|(lr)}}, [[REGISTER]]
+; CHECK: eor.w
+; CHECK-NOT: eor
+; CHECK: and
+ %data_addr.013 = phi i8 [ %data, %bb.nph ], [ %8, %bb ] ; <i8> [#uses=2]
+ %crc_addr.112 = phi i16 [ %crc, %bb.nph ], [ %crc_addr.2, %bb ] ; <i16> [#uses=3]
+ %i.011 = phi i8 [ 0, %bb.nph ], [ %7, %bb ] ; <i8> [#uses=1]
+ %0 = trunc i16 %crc_addr.112 to i8 ; <i8> [#uses=1]
+ %1 = xor i8 %data_addr.013, %0 ; <i8> [#uses=1]
+ %2 = and i8 %1, 1 ; <i8> [#uses=1]
+ %3 = icmp eq i8 %2, 0 ; <i1> [#uses=2]
+ %4 = xor i16 %crc_addr.112, 16386 ; <i16> [#uses=1]
+ %crc_addr.0 = select i1 %3, i16 %crc_addr.112, i16 %4 ; <i16> [#uses=1]
+ %5 = lshr i16 %crc_addr.0, 1 ; <i16> [#uses=2]
+ %6 = or i16 %5, -32768 ; <i16> [#uses=1]
+ %crc_addr.2 = select i1 %3, i16 %5, i16 %6 ; <i16> [#uses=2]
+ %7 = add i8 %i.011, 1 ; <i8> [#uses=2]
+ %8 = lshr i8 %data_addr.013, 1 ; <i8> [#uses=1]
+ %exitcond = icmp eq i8 %7, 8 ; <i1> [#uses=1]
+ br i1 %exitcond, label %bb8, label %bb
+
+bb8: ; preds = %bb
+ ret i16 %crc_addr.2
+}