LLVM Bug Fix 13709: Remove needless lsr(Rp, #32) instruction access the

subreg_hireg of register pair Rp.

	* lib/Target/Hexagon/HexagonPeephole.cpp(PeepholeDoubleRegsMap): New
	 DenseMap similar to PeepholeMap that additionally records subreg info
	 too.
        (runOnMachineFunction): Record information in PeepholeDoubleRegsMap
        and copy propagate the high sub-reg of Rp0 in Rp1 = lsr(Rp0, #32) to
	the instruction Rx = COPY Rp1:logreg_subreg.
	* test/CodeGen/Hexagon/remove_lsr.ll: New test.
	



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@163214 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 80 insertions, 0 deletions

diff --git a/test/CodeGen/Hexagon/remove_lsr.ll b/test/CodeGen/Hexagon/remove_lsr.ll
new file mode 100644
index 0000000000..79b5f4ae7c
--- /dev/null
+++ b/test/CodeGen/Hexagon/remove_lsr.ll
@@ -0,0 +1,80 @@
+; Test fix for PR-13709.
+; RUN: llc -march=hexagon -mcpu=hexagonv4 < %s | FileCheck %s
+; CHECK: foo
+; CHECK-NOT: lsr(r{{[0-9]+}}:{{[0-9]+}}, #32)
+; CHECK-NOT: lsr(r{{[0-9]+}}:{{[0-9]+}}, #32)
+
+; Convert the sequence
+; r17:16 = lsr(r11:10, #32)
+; .. = r16
+; into
+; r17:16 = lsr(r11:10, #32)
+; .. = r11
+; This makes the lsr instruction dead and it gets removed subsequently
+; by a dead code removal pass.
+
+%union.vect64 = type { i64 }
+%union.vect32 = type { i32 }
+
+define void @foo(%union.vect64* nocapture %sss_extracted_bit_rx_data_ptr,
+ %union.vect32* nocapture %s_even, %union.vect32* nocapture %s_odd,
+ i8* nocapture %scr_s_even_code_ptr, i8* nocapture %scr_s_odd_code_ptr)
+ nounwind {
+entry:
+  %scevgep = getelementptr %union.vect64* %sss_extracted_bit_rx_data_ptr, i32 1
+  %scevgep28 = getelementptr %union.vect32* %s_odd, i32 1
+  %scevgep32 = getelementptr %union.vect32* %s_even, i32 1
+  %scevgep36 = getelementptr i8* %scr_s_odd_code_ptr, i32 1
+  %scevgep39 = getelementptr i8* %scr_s_even_code_ptr, i32 1
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %lsr.iv42 = phi i32 [ %lsr.iv.next, %for.body ], [ 2, %entry ]
+  %lsr.iv40 = phi i8* [ %scevgep41, %for.body ], [ %scevgep39, %entry ]
+  %lsr.iv37 = phi i8* [ %scevgep38, %for.body ], [ %scevgep36, %entry ]
+  %lsr.iv33 = phi %union.vect32* [ %scevgep34, %for.body ], [ %scevgep32, %entry ]
+  %lsr.iv29 = phi %union.vect32* [ %scevgep30, %for.body ], [ %scevgep28, %entry ]
+  %lsr.iv = phi %union.vect64* [ %scevgep26, %for.body ], [ %scevgep, %entry ]
+  %predicate_1.023 = phi i8 [ undef, %entry ], [ %10, %for.body ]
+  %predicate.022 = phi i8 [ undef, %entry ], [ %9, %for.body ]
+  %val.021 = phi i64 [ undef, %entry ], [ %srcval, %for.body ]
+  %lsr.iv3335 = bitcast %union.vect32* %lsr.iv33 to i32*
+  %lsr.iv2931 = bitcast %union.vect32* %lsr.iv29 to i32*
+  %lsr.iv27 = bitcast %union.vect64* %lsr.iv to i64*
+  %0 = tail call i64 @llvm.hexagon.A2.vsubhs(i64 0, i64 %val.021)
+  %conv3 = sext i8 %predicate.022 to i32
+  %1 = trunc i64 %val.021 to i32
+  %2 = trunc i64 %0 to i32
+  %3 = tail call i32 @llvm.hexagon.C2.mux(i32 %conv3, i32 %1, i32 %2)
+  store i32 %3, i32* %lsr.iv3335, align 4, !tbaa !0
+  %conv8 = sext i8 %predicate_1.023 to i32
+  %4 = lshr i64 %val.021, 32
+  %5 = trunc i64 %4 to i32
+  %6 = lshr i64 %0, 32
+  %7 = trunc i64 %6 to i32
+  %8 = tail call i32 @llvm.hexagon.C2.mux(i32 %conv8, i32 %5, i32 %7)
+  store i32 %8, i32* %lsr.iv2931, align 4, !tbaa !0
+  %srcval = load i64* %lsr.iv27, align 8
+  %9 = load i8* %lsr.iv40, align 1, !tbaa !1
+  %10 = load i8* %lsr.iv37, align 1, !tbaa !1
+  %lftr.wideiv = trunc i32 %lsr.iv42 to i8
+  %exitcond = icmp eq i8 %lftr.wideiv, 32
+  %scevgep26 = getelementptr %union.vect64* %lsr.iv, i32 1
+  %scevgep30 = getelementptr %union.vect32* %lsr.iv29, i32 1
+  %scevgep34 = getelementptr %union.vect32* %lsr.iv33, i32 1
+  %scevgep38 = getelementptr i8* %lsr.iv37, i32 1
+  %scevgep41 = getelementptr i8* %lsr.iv40, i32 1
+  %lsr.iv.next = add i32 %lsr.iv42, 1
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+declare i64 @llvm.hexagon.A2.vsubhs(i64, i64) nounwind readnone
+
+declare i32 @llvm.hexagon.C2.mux(i32, i32, i32) nounwind readnone
+
+!0 = metadata !{metadata !"long", metadata !1}
+!1 = metadata !{metadata !"omnipotent char", metadata !2}
+!2 = metadata !{metadata !"Simple C/C++ TBAA", null}