[llvm] ea093f6 - Handle cases for subregisters.

[Brian Cain via llvm-commits]

Author: Suyog Sarda
Date: 2020-04-30T20:32:33-05:00
New Revision: ea093f64813dbaaff2e08186060f6388a4173562

URL: https://github.com/llvm/llvm-project/commit/ea093f64813dbaaff2e08186060f6388a4173562
DIFF: https://github.com/llvm/llvm-project/commit/ea093f64813dbaaff2e08186060f6388a4173562.diff

LOG: Handle cases for subregisters.

While restoring latency, check if any of the registers of
source instruction is a subregister of the successor instructions
apart from being same register.

Added: 
    llvm/test/CodeGen/Hexagon/check-subregister-for-latency.ll

Modified: 
    llvm/lib/Target/Hexagon/HexagonSubtarget.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/Hexagon/HexagonSubtarget.cpp b/llvm/lib/Target/Hexagon/HexagonSubtarget.cpp
index 4dcda3b10f4d..5f18767c76cc 100644
--- a/llvm/lib/Target/Hexagon/HexagonSubtarget.cpp
+++ b/llvm/lib/Target/Hexagon/HexagonSubtarget.cpp
@@ -424,8 +424,17 @@ void HexagonSubtarget::restoreLatency(SUnit *Src, SUnit *Dst) const {
     int DefIdx = -1;
     for (unsigned OpNum = 0; OpNum < SrcI->getNumOperands(); OpNum++) {
       const MachineOperand &MO = SrcI->getOperand(OpNum);
-      if (MO.isReg() && MO.isDef() && MO.getReg() == DepR)
-        DefIdx = OpNum;
+      bool IsSameOrSubReg = false;
+      if (MO.isReg()) {
+        unsigned MOReg = MO.getReg();
+        if (Register::isVirtualRegister(DepR)) {
+          IsSameOrSubReg = (MOReg == DepR);
+        } else {
+          IsSameOrSubReg = getRegisterInfo()->isSubRegisterEq(DepR, MOReg);
+        }
+        if (MO.isDef() && IsSameOrSubReg)
+          DefIdx = OpNum;
+      }
     }
     assert(DefIdx >= 0 && "Def Reg not found in Src MI");
     MachineInstr *DstI = Dst->getInstr();

diff  --git a/llvm/test/CodeGen/Hexagon/check-subregister-for-latency.ll b/llvm/test/CodeGen/Hexagon/check-subregister-for-latency.ll
new file mode 100644
index 000000000000..8290f768585d
--- /dev/null
+++ b/llvm/test/CodeGen/Hexagon/check-subregister-for-latency.ll
@@ -0,0 +1,66 @@
+; RUN: llc -march=hexagon -mcpu=hexagonv67t < %s
+; REQUIRES: asserts
+
+target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048"
+target triple = "hexagon"
+
+%s.0 = type { double, double, double, double, double, double, i32, double, double, double, double, i8*, i8, [9 x i8], double, double, double, double, double, double, double, double, double, double, double, double, double, double, double, double, double, double, double, double, double, double, double, double, double, [200 x i8*], [32 x i8*], [32 x i8], i32 }
+
+define hidden fastcc void @f0() unnamed_addr #0 {
+b0:
+  %v0 = getelementptr inbounds %s.0, %s.0* null, i32 0, i32 33
+  %v1 = getelementptr inbounds %s.0, %s.0* null, i32 0, i32 34
+  br label %b1
+
+b1:                                               ; preds = %b1, %b0
+  %v2 = phi i32* [ undef, %b0 ], [ %v27, %b1 ]
+  %v3 = load i32, i32* %v2, align 1, !tbaa !1
+  %v4 = getelementptr inbounds [0 x %s.0*], [0 x %s.0*]* null, i32 0, i32 %v3
+  %v5 = load %s.0*, %s.0** %v4, align 1, !tbaa !5
+  %v6 = load double, double* undef, align 1, !tbaa !7
+  %v7 = fdiv double 1.000000e+00, %v6
+  %v8 = fmul double %v7, 0.000000e+00
+  %v9 = fmul double %v7, 0.000000e+00
+  %v10 = fmul double %v8, -4.800000e+01
+  %v11 = fmul double %v9, 1.680000e+02
+  %v12 = fmul double %v7, 0.000000e+00
+  %v13 = load double, double* null, align 1, !tbaa !7
+  %v14 = fmul double %v7, %v13
+  %v15 = fmul double %v12, 0.000000e+00
+  %v16 = getelementptr inbounds %s.0, %s.0* %v5, i32 0, i32 30
+  %v17 = fsub double 0.000000e+00, %v15
+  store double %v17, double* %v16, align 8, !tbaa !9
+  %v18 = fmul double %v14, 0.000000e+00
+  %v19 = getelementptr inbounds %s.0, %s.0* %v5, i32 0, i32 32
+  %v20 = load double, double* %v19, align 8, !tbaa !11
+  %v21 = fsub double %v20, %v18
+  store double %v21, double* %v19, align 8, !tbaa !11
+  %v22 = fmul double %v10, 0.000000e+00
+  %v23 = fadd double 0.000000e+00, %v22
+  %v24 = fmul double 0.000000e+00, %v11
+  %v25 = fadd double %v23, %v24
+  %v26 = fsub double 0.000000e+00, %v25
+  store double %v26, double* %v0, align 8, !tbaa !12
+  store double 0.000000e+00, double* %v1, align 8, !tbaa !13
+  %v27 = getelementptr i32, i32* %v2, i32 1
+  br label %b1
+}
+
+attributes #0 = { "use-soft-float"="false" }
+
+!llvm.module.flags = !{!0}
+
+!0 = !{i32 1, !"EnableSplitLTOUnit", i32 0}
+!1 = !{!2, !2, i64 0}
+!2 = !{!"int", !3, i64 0}
+!3 = !{!"omnipotent char", !4, i64 0}
+!4 = !{!"Simple C/C++ TBAA"}
+!5 = !{!6, !6, i64 0}
+!6 = !{!"any pointer", !3, i64 0}
+!7 = !{!8, !8, i64 0}
+!8 = !{!"double", !3, i64 0}
+!9 = !{!10, !8, i64 232}
+!10 = !{!"", !8, i64 0, !8, i64 8, !8, i64 16, !8, i64 24, !8, i64 32, !8, i64 40, !2, i64 48, !8, i64 56, !8, i64 64, !8, i64 72, !8, i64 80, !6, i64 88, !3, i64 92, !3, i64 93, !8, i64 104, !8, i64 112, !8, i64 120, !8, i64 128, !8, i64 136, !8, i64 144, !8, i64 152, !8, i64 160, !8, i64 168, !8, i64 176, !8, i64 184, !8, i64 192, !8, i64 200, !8, i64 208, !8, i64 216, !8, i64 224, !8, i64 232, !8, i64 240, !8, i64 248, !8, i64 256, !8, i64 264, !8, i64 272, !8, i64 280, !8, i64 288, !8, i64 296, !3, i64 304, !3, i64 1104, !3, i64 1232, !2, i64 1264}
+!11 = !{!10, !8, i64 248}
+!12 = !{!10, !8, i64 256}
+!13 = !{!10, !8, i64 264}